Graphene for Post‐Moore Silicon Optoelectronics 9783527351817, 9783527840991, 9783527841004, 9783527841011

Table of contents :
Cover
Title Page
Copyright Page
Contents
Preface
Acknowledgments
Biography
Chapter 1 Graphene for Silicon Optoelectronics
1.1 Introduction
1.2 Optical Absorption
1.3 Emergence of Graphene in Silicon Optoelectronics
1.4 Photodetection in Graphene
1.4.1 Performance Metrics
1.4.2 Photovoltaic Effect
1.4.3 Photoemission in Graphene Schottky Junctions
1.4.4 Thermionic Emission in Graphene-based Interfaces
1.4.5 Hot Electron-based Photodetection
1.4.5.1 Photothermoelectric Effect (PTE)
1.4.5.2 Photobolometric Effect (PBE)
1.4.5.3 Photothermionic (PTI) Effect
1.4.5.4 Photogating Effect
1.4.6 Infrared Modulators
1.4.7 Photovoltaic Devices
1.5 Outlook
References
Chapter 2 Growth and Transfer of Graphene for Silicon Optoelectronics
2.1 Introduction
2.2 Growth of Graphene
2.2.1 Growth Dynamics of CVD Gr and Choice of Substrate
2.2.2 Growth on Metallic Substrates
2.2.3 Direct Growth on Dielectric Substrates
2.2.4 Direct Growth on Semiconductor Substrates
2.2.5 Large-scale CVD Growth of Graphene
2.3 Dielectric Deposition on Graphene
2.4 Graphene Transfer Methods
2.5 Fabrication of Solution-processed Graphene and Integration with Silicon
2.6 Graphene Transfer on Flexible Silicon
2.7 Graphene Integration with Silicon in CMOS Process
2.8 Challenges and Future Prospectives
References
Chapter 3 Physics of Graphene/Silicon Junctions
3.1 Introduction
3.2 Physics of Schottky Junction
3.3 Measurement of Schottky Barrier Height
3.3.1 Capacitance Voltage Measurement
3.3.2 Current–Voltage Measurement
3.3.3 Photoelectric Measurement
3.3.4 Thermionic Emission Measurements
3.4 2D Materials and Schottky Junctions
3.5 Challenges and Future Prospective
References
Chapter 4 Graphene/Silicon Junction for High-performance Photodetectors
4.1 Introduction
4.2 Ultraviolet Photodetectors
4.3 Visible to Near-infrared Photodetector
4.4 Broadband Photodetectors
4.5 Hybrid Gr/Si Photodetectors
4.6 Challenges and Perspectives
References
Chapter 5 Graphene/Silicon Solar Energy Harvesting Devices
5.1 Introduction
5.2 Photovoltaic Mechanism and Performance Parameters of Graphene/Silicon Solar Cells
5.3 Theoretical Efficiency Limits of Graphene Silicon Solar Cells
5.4 Optimization of Graphene/Silicon Solar Cells
5.4.1 Doping of Graphene
5.4.2 Light Trapping in Silicon
5.4.3 Antireflection Coating
5.4.4 Interface Engineering
5.4.5 Surface Passivation
5.5 Challenges and Perspectives
References
Chapter 6 Graphene Silicon-integrated Waveguide Devices
6.1 Introduction
6.2 Hybrid Waveguide Photodetector
6.3 Hybrid Waveguide Modulator
6.3.1 Electro-optical Modulator
6.3.2 Thermo-optic Modulator
6.4 Challenges and Prospectives
References
Chapter 7 Graphene for Silicon Image Sensor
7.1 Introduction
7.2 Quantum Dot-based Infrared Graphene Image Sensor
7.3 Graphene Thermopile Image Sensor
7.4 Graphene THz Image Sensor
7.5 Curved Image Sensor Array
7.6 Neural Network Image Sensors
7.7 Graphene Charge-coupled Device Image Sensor
7.8 Graphene-based Position-sensitive Detector
7.9 Challenges and Perspectives
Chapter 8 System Integration with Graphene for Silicon Optoelectronics
8.1 Introduction
8.2 Graphene Silicon Flip Chips
8.3 Graphene Silicon Heterogeneous Integration
8.4 Graphene Silicon Monolithic Integration for Optoelectronics Applications
8.5 Challenges and Prospective
References
Chapter 9 Graphene for Silicon Optoelectronic Synaptic Devices
9.1 Introduction
9.2 Silicon Neurons
9.3 Synaptic Devices
9.4 Silicon Optoelectronic Synaptic Devices
9.5 ORAM Synaptic Devices
9.6 Graphene for Silicon Synaptic Devices
9.7 Synaptic Phototransistor
9.8 Broadband, Low-power Optoelectronic Synaptic Devices
9.9 Challenges and Prospects
References
Chapter 10 Challenges and Prospects of Graphene–Silicon Optoelectronics
10.1 Emergence of Wafer‐scale Systems
10.2 Wafer‐scale Synthesis and Foundry Process
10.3 Scalable Transfer and Quality Metrics
10.4 Scaling Laws and Hot‐electron Effects
10.5 Optical Modulators
10.6 Infrared Photodetectors
10.7 Neuromorphic Optoelectronics
References
Index
EULA

Citation preview

Graphene for Post-­Moore Silicon Optoelectronics

Graphene for Post‐Moore Silicon Optoelectronics

Yang Xu, Khurram Shehzad, Srikrishna Chanakya Bodepudi, Ali Imran, and Bin Yu

Authors Prof. Yang Xu

Zhejiang University School of Micro-­Nano Electronics No. 388, YuHangTang Rd. Xihu District 310027 Hangzhou China Dr. Khurram Shehzad

Zhejiang University School of Micro-­Nano Electronics No. 388, YuHangTang Rd. Xihu District 310027 Hangzhou China Dr. Srikrishna Chanakya Bodepudi

Zhejiang University School of Micro-­Nano Electronics No. 388, YuHangTang Rd. Xihu District 310027 Hangzhou China Dr. Ali Imran

Zhejiang University School of Micro-­Nano Electronics No. 388, YuHangTang Rd. Xihu District 310027 Hangzhou China Prof. Bin Yu

Zhejiang University School of Micro-­Nano Electronics No. 388, YuHangTang Rd. Xihu District 310027 Hangzhou China Cover Images: © GrAl/Shutterstock

All books published by WILEY-­VCH are carefully produced. Nevertheless, authors, editors, and publisher do not warrant the information contained in these books, including this book, to be free of errors. Readers are advised to keep in mind that statements, data, illustrations, procedural details or other items may inadvertently be inaccurate. Library of Congress Card No.: applied for British Library Cataloguing-­in-­Publication Data:

A catalogue record for this book is available from the British Library. Bibliographic information published by the Deutsche Nationalbibliothek

The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http:// dnb.d-­nb.de. © 2023 WILEY-­VCH GmbH, Boschstr. 12, 69469 Weinheim, Germany All rights reserved (including those of translation into other languages). No part of this book may be reproduced in any form – by photoprinting, microfilm, or any other means – nor transmitted or translated into a machine language without written permission from the publishers. Registered names, trademarks, etc. used in this book, even when not specifically marked as such, are not to be considered unprotected by law. Print ISBN  978-­3-­527-­35181-­7 ePDF ISBN  978-­3-­527-­84099-­1 ePub ISBN  978-­3-­527-­84100-­4 oBook ISBN  978-­3-­527-­84101-­1 Typesetting  Straive, Chennai, India

v

Contents Preface  ix Acknowledgments  xi Biography  xii 1 1.1 1.2 1.3 1.4 1.4.1 1.4.2 1.4.3 1.4.4 1.4.5 1.4.5.1 1.4.5.2 1.4.5.3 1.4.5.4 1.4.6 1.4.7 1.5 ­

Graphene for Silicon Optoelectronics  1 ­Introduction  1 ­Optical Absorption  2 ­Emergence of Graphene in Silicon Optoelectronics  3 ­Photodetection in Graphene  4 Performance Metrics  5 Photovoltaic Effect  5 Photoemission in Graphene Schottky Junctions  6 Thermionic Emission in Graphene-based Interfaces  7 Hot Electron-based Photodetection  9 Photothermoelectric Effect (PTE)  10 Photobolometric Effect (PBE)  12 Photothermionic (PTI) Effect  13 Photogating Effect  13 Infrared Modulators  16 Photovoltaic Devices  16 ­Outlook  17 References  18

2 2.1 2.2 2.2.1 2.2.2 2.2.3 2.2.4 2.2.5 2.3 2.4

Growth and Transfer of Graphene for Silicon Optoelectronics  21 ­Introduction  21 ­Growth of Graphene  21 Growth Dynamics of CVD Gr and Choice of Substrate  22 Growth on Metallic Substrates  24 Direct Growth on Dielectric Substrates  26 Direct Growth on Semiconductor Substrates  29 Large-­scale CVD Growth of Graphene  31 ­Dielectric Deposition on Graphene  33 ­Graphene Transfer Methods  35

vi

Contents

2.5 2.6 2.7 2.8

­ abrication of Solution-­processed Graphene and Integration with Silicon  38 F ­Graphene Transfer on Flexible Silicon  39 ­Graphene Integration with Silicon in CMOS Process  40 ­Challenges and Future Prospectives  41 ­References  42

3 3.1 3.2 3.3 3.3.1 3.3.2 3.3.3 3.3.4 3.4 3.5

Physics of Graphene/Silicon Junctions  47 ­Introduction  47 ­Physics of Schottky Junction  48 ­Measurement of Schottky Barrier Height  53 Capacitance Voltage Measurement  53 Current–Voltage Measurement  54 Photoelectric Measurement  55 Thermionic Emission Measurements  55 ­2D Materials and Schottky Junctions  58 ­Challenges and Future Prospective  61 ­References  63

4 4.1 4.2 4.3 4.4 4.5 4.6

Graphene/Silicon Junction for High-performance Photodetectors  65 ­Introduction  65 ­Ultraviolet Photodetectors  65 ­Visible to Near-infrared Photodetector  68 ­Broadband Photodetectors  71 ­Hybrid Gr/Si Photodetectors  75 ­Challenges and Perspectives  80 ­References  81

5 5.1 5.2 5.3 5.4 5.4.1 5.4.2 5.4.3 5.4.4 5.4.5 5.5

Graphene/Silicon Solar Energy Harvesting Devices  85 ­Introduction  85 ­Photovoltaic Mechanism and Performance Parameters of Graphene/ Silicon Solar Cells  86 ­Theoretical Efficiency Limits of Graphene Silicon Solar Cells  88 ­Optimization of Graphene/Silicon Solar Cells  89 Doping of Graphene  89 Light Trapping in Silicon  92 Antireflection Coating  94 Interface Engineering  97 Surface Passivation  100 ­Challenges and Perspectives  101 ­References  102

6 6.1 6.2 6.3 6.3.1 6.3.2

Graphene Silicon-­integrated Waveguide Devices  107 ­Introduction  107 ­Hybrid Waveguide Photodetector  111 ­Hybrid Waveguide Modulator  114 Electro-­optical Modulator  115 Thermo-­optic Modulator  117

Contents

6.4 ­

­ hallenges and Prospectives  117 C References  118

7 ­7.1 ­7.2 ­7.3 ­7.4 ­7.5 ­7.6 ­7.7 ­7.8 ­7.9

Graphene for Silicon Image Sensor  121 Introduction  121 Quantum Dot-­based Infrared Graphene Image Sensor  123 Graphene Thermopile Image Sensor  124 Graphene THz Image Sensor  125 Curved Image Sensor Array  126 Neural Network Image Sensors  127 Graphene Charge-­coupled Device Image Sensor  128 Graphene-­based Position-­sensitive Detector  132 Challenges and Perspectives  136 ­References  137

8 8.1 8.2 8.3 8.4 8.5

System Integration with Graphene for Silicon Optoelectronics  141 ­Introduction  141 ­Graphene Silicon Flip Chips  142 ­Graphene Silicon Heterogeneous Integration  145 ­Graphene Silicon Monolithic Integration for Optoelectronics Applications  147 ­Challenges and Prospective  150 ­References  152

9 9­ .1 ­9.2 ­9.3 ­9.4 ­9.5 ­9.6 ­9.7 ­9.8 ­9.9 ­

Graphene for Silicon Optoelectronic Synaptic Devices  153 Introduction  153 Silicon Neurons  154 Synaptic Devices  156 Silicon Optoelectronic Synaptic Devices  157 ORAM Synaptic Devices  159 Graphene for Silicon Synaptic Devices  159 Synaptic Phototransistor  160 Broadband, Low-power Optoelectronic Synaptic Devices  163 Challenges and Prospects  164 References  167

Challenges and Prospects of Graphene–Silicon Optoelectronics  169 ­Emergence of Wafer‐scale Systems  169 ­Wafer‐scale Synthesis and Foundry Process  169 ­Scalable Transfer and Quality Metrics  171 ­Scaling Laws and Hot‐electron Effects  172 ­Optical Modulators  173 ­Infrared Photodetectors  174 ­Neuromorphic Optoelectronics  176 ­References  176 10 10.1 10.2 10.3 10.4 10.5 10.6 10.7



Index  177

vii

ix

Preface Miniaturization of electronic devices  –  a primary step that drives Moore’s law  – allows the number of digital electronic devices to roughly double by every two years within a fixed cost and area while improving their performance and functionality. However, such progress in power‐efficient, high‐performance, small device footprint, and low‐cost devices has come to a halt as further scaling down leads to difficulty in achieving complex doping profiles and excessive leakage currents. This issue is elevated when devices are scaled down below 3 nm, where bulk semiconductors lose their structural quality and show degrading charge transport and optoelectronic properties. In this context, projecting device performance beyond the scaling limits of Moore’s law requires technologies based on novel materials, circuits, and device architecture. Graphene and two‐dimensional (2D) materials have emerged as alternate candidates with atomically thin structures showing excellent charge transport properties and prototypes in computational and noncomputational applications. Although the domination of Si technology is unlikely to be abandoned in the foreseeable future, the growing benefits of graphene‐based electronics call for hybrid device architectures that incorporate existing remarkable technological evolution and commercial success of Si CMOS technology while adopting the novel features of graphene. “More than Moore” or noncomputational systems, such as photodetectors and modulators for image sensors, light detection and ranging (LiDAR), lasers, biomedical sensors, and neuromorphic and radio‐frequency devices, are swiftly advancing beyond Si electronics when integrated with graphene and other 2D materials by adapting their benefits of low‐power consumption and intrinsic scalability. Fully integrated prototypes of 2D/Si chips, especially graphene, have been realized for diverse applications, including image sensor arrays and optical receivers. Most of these prototypes are developed on the integrated silicon chips where silicon devices provide driver, source, and readout circuitry. This book discusses the basics, applications, challenges, and opportunities regarding integrating graphene with Si technologies, with a special emphasis on graphene–Si (Gr/Si) optoelectronic devices in the post‐Moore era. It might be helpful to summarize the important aspects of Gr/Si‐integrated devices in optoelectronics in the post‐Moore era. Our book also discusses the progress and future challenges from synthesis to device fabrication and related physics

x

Preface

of high‐quality, wafer‐scale Gr/Si‐integrated optoelectronic devices. All these aspects of the Gr/Si devices are relevant to a broad research community in chemistry, materials science, and electronic engineering. This book is arranged to discuss the opportunities and challenges of Gr/Si systems, where each chapter emphasizes selected topics such as high‐performance photodetectors, energy‐harvesting devices, and image sensors and their corresponding progress and challenges. Special emphasis is given to emerging applications like optoelectronic synaptic devices, optical modulators, and infrared image sensors. This book will serve as a good reference for graduate students, postdocs, and scientists from academia and industry. 15 July 2022

Prof. Yang Xu, On behalf of all the authors, Zhejiang University, Hangzhou, China

xi

Acknowledgments We would like to thank Ms. Shaoyu Qian and their publishing team from Wiley for their great support. We also would like to sincerely thank the significant and outstanding contributions from our team members including Dr. Lixiang Liu, Dr. Dajian Liu, Dr. Zhixiang Zhang, Dr. P. Pham, Dr. Jianhang Lv, Dr. Dong Pu, Dr. K. Dianey, M. Ali, Xiaocheng Wang, Xiaoxue Cao, A. Anwar, M. Malik, and Xinyu Liu. Without their great support and remarkable dedication, we could not have finished this book. This book is supported by National Natural Science Foundation of China (NSFC) (Grant Nos. 92164106, 61874094, and 62090034).

xii

Biography Prof. Yang Xu is a Fellow of the Institute of Physics (FInstP, IOP Fellow), IEEE NTC Distinguished Lecturer, and IEEE Senior Member of the Electron Devices Society. He received his B.S. degree from Department of EE, Institute of Microelectronics, Tsinghua University, M.S. and Ph.D. degrees in ECE from the University of Illinois Urbana‐Champaign (UIUC), USA. He is now a full professor at the School of Micro‐Nano Electronics, Zhejiang University, China. He was also a visiting by‐Fellow of Churchill College at the University of Cambridge, UK, and a visiting professor at the University of California, Los Angeles (UCLA). He has published more than 120 papers in Nature Nanotechnology, Nature Electronics, Nature Photonics, Chemical Reviews, Advanced Materials, Chemical Society Reviews, Nature Communications, Nano Letters, ACS Nano, IEEE‐EDL, IEEE‐TED, IEEE‐TNANO, IEDM, etc. He holds over 30 granted patents and has given more than 50 conference talks. He also served as Associate Editor of IEEE Nanotechnology Magazine, Microelectronics Journal, Micro & Nano Letters, and IET Circuits, Devices & Systems, Advisory Panel Member of IOP Nanotechnology, and was TPC committee members of IEEE‐EDTM, IEEE‐IPFA, and IEEE‐EDAPS ­conferences. His research interests include emerging 2D/3D integrated opto­electronic devices for Internet‐of‐Things and Post‐Moore Ubiquitous Electronics.

1

1 Graphene for Silicon Optoelectronics 1.1 ­Introduction Rich electronic, optical, and mechanical properties of graphene, such as high carrier mobilities, optical transparency, flexibility, robustness, and environmental stability, offer solutions to many technological challenges in the communication, health, and energy sectors. Unique characteristics of Dirac fermions in graphene 2 enable specific integer and fractional hall effects, minimum conductivity of ~4 e ,  and near ballistic transport at room temperature [1]. In specific, the true potential of graphene lies in optoelectronics and photonics [2], where intrinsic properties such as linear dispersion of Dirac electrons enabling ultrawideband tunability, transmittance expressed in terms of fine-structure constant, saturable absorption due to Pauli blocking, and hot luminescence originating from nonequilibrium carriers enable highly efficient physical mechanisms at the interest of condensed matter physics and the optoelectronic applications. The present image sensor market is dominated by the Si detectors integrated with CMOS readout circuits, offering high resolution, compact footprint, low manufacturing cost, and high performance [3]. As silicon, in general, is the active material in these systems, the absorption bandwidth is limited to visible to infrared light (300–900 nm). Some crucial applications of image sensors, such as biomedical imaging, gas sensing, spectroscopy, environmental monitoring, and night vision, require a broad operating range in the infrared region that can even extend to the terahertz region [4]. In addition, many of these applications need the detector’s operation in both visible and infrared, where emerging applications such as 3D imaging for autonomous vehicles and augmented virtual reality systems benefit. Specifically, the extension of 3D imaging into the infrared range reduces interference with the strong visible background light and thus helps reduce the overall power consumption [2]. These applications need novel photodetectors with a broad wavelength range, high detectivity, low power, low production cost, and easy integration with Si CMOS technology. Graphene and the related two-dimensional (2D) material-based photodetectors fulfill all these requirements and offer a fast-broadband response from ultraviolet to Graphene for Post-Moore Silicon Optoelectronics, First Edition. Yang Xu, Khurram Shehzad, Srikrishna Chanakya Bodepudi, Ali Imran, and Bin Yu. © 2023 WILEY-VCH GmbH. Published 2023 by WILEY-VCH GmbH.

1  Graphene for Silicon Optoelectronics Market opportunity

Transistors

Front-end-of-line (high complexity) Back-end-of-line (low complexity)

2

High

FEOL opto electronics (optical interconnects)

Neuromorphic computing

BEOL opto electronics (modulators and detectors) Interconnects

Low

Heterogeneous integration of memory

Sensors

Color key Functional devices

Power-gate transistors BEOL opto electronics (cameras and spectrometers)

2018

2020

2022

2024

Medium

Opto electronics

2026

2028

2030

Logic

Time

Figure 1.1  Application of two-dimensional materials‑silicon technology depicted in time and the complexity for integration. Source: Reproduced with permission from Akinwande et al. [2]; Springer Nature.

terahertz spectral range (Figure 1.1). The emerging applications in optoelectronics and future neuromorphic and quantum computing devices seek 2D materials integrated with the Si CMOS process to address the most pressing challenges and limitations faced by miniaturization in Si technology. Low power consumption, high operating speeds, and efficient energy transduction are the primary requirements for the next-generation electronic and optoelectronic devices, where graphene made its mark as a future material with the potential to resolve the limitations of silicon electronics. Developing device schemes by integrating graphene and silicon allows us to continue to use the most successful Si technology while exploiting the benefits of graphene to improve the existing device’s performance and efficiency; graphene electronics is strongly connected with silicon and CMOS technology. For instance, the ability to visualize atomic layer thick graphene on SiO2/Si substrate is the most crucial step in investigating the transport properties of graphene. Let us start our discussion with this fundamental step of identifying graphene and absorbing light.

1.2 ­Optical Absorption The contrast in the optical image can be used to identify graphene in the Si/SiO2 substrate, which results from interference with SiO2 as a spacer layer and scales with the number of graphene layers. This contrast can be further tuned by varying the spacer thickness or wavelength of light. Graphene with fixed universal optical conductance G0 = e2/(4ℏ) = 6.08 × 10−5 gives the transmittance, T

1 0.5

2

1

97.7%

(1.1)

where α = e2/(4πε0ℏc) = G0/πε0c = 1/137 is the fine-structure constant. The optical absorbance is proportional to the layer number, with each layer absorbing 2.3% of the visible spectrum. Graphene exhibits an almost flat absorption spectrum from

1.3  ­Emergence of Graphene in Silicon Optoelectronic T = 300 K Γ = 0 meV τ = 40 fs THz

∣Ef –EDirac∣ = 100 meV ∣Ef –EDirac∣ = 200 meV IR

Visible

Re{σ(ω)} (e2/4–h)

100 10 1 0.1

Increasing ∣Ef –EDirac∣

Inter band 0.01 0.1

Intra band (a)

(b)

1 10 100 Frequency (THz)

1000

Figure 1.2  (a) Optical transitions in graphene. (b) Optical conductivity versus frequency of graphene from terahertz to visible region. Source: Reproduced with permission from Sensale-Rodriguez [5]; IEEE.

300 nm to 2.5 μm, with a peak in the ultraviolet region (~270 nm) originating from the exciton-shifted van Hove Singularities in graphene density of states (DOSs) [4]. Another key parameter that defines the role of graphene in optoelectronics is optical conductivity. Since the light absorption in graphene occurs in both inter- and intraband transitions, as shown in Figure 1.2a, optical conductivity is the combination of inter- and intraband conductivities [4]. Ultrafast optical pulses generate a nonequilibrium carrier population in both valence and conduction bands. The cooling or bringing these excited carriers to equilibrium with lattice consists of two cooling steps, ultrafast carrier–carrier intraband collisions followed by phonon emissions and a slower cooling that corresponds to electron interband relaxation and cooling of hot phonons [6]. At low frequencies or low photon energies, the intraband transitions dominate, and interband transitions dominate in graphene at high frequencies. Further increasing the frequency, the optical conductivity reduces to a constant value e2/(4ℏ) that includes the visible region, resulting in the universal absorbance of ~2.3% for graphene as displayed in Figure 1.2b. The dependence of optical conductivity on carrier concentration or the Fermi level allows electrostatically controlling optical absorption. For instance, in the infrared region, the optical absorption can be slightly varied from 0% to 2.3%, leading to tunable devices such as IR modulators, phase shifters, beam steerers, and mode-lock lasers [5].

1.3 ­Emergence of Graphene in Silicon Optoelectronics Optoelectronics based on 2D materials integrated with the CMOS process is the most growing area of 2D electronics in the present decade (Figure  1.1), where

3

4

1  Graphene for Silicon Optoelectronics

graphene-based devices dominate both in terms of investigating fundamental physics and developing real-world applications. 2D material-integrated optoelectronics such as cameras and spectrometers are already on the verge of commercialization [2]. Next in the row are optical modulators and photodetectors that are gaining the attention of many researchers. Here, we discuss these emerging applications and concepts of graphene-based optoelectronics compatible with CMOS integration in back-end-of-line (BEOL) and front-end-of-line (FEOL). The complexity of integration increases from BEOL to FEOL. Therefore, most of the existing device schemes of 2D material-based optoelectronics are compatible with the BEOL process  [2]. We start our discussion with the role of graphene in BEOL optoelectronics. The simple transfer process, transparent yet high carrier mobility, and mechanically robust while flexible are graphene’s critical merits that allow easy integration with Si CMOS technology in the BEOL fabrication stage. In 2D optoelectronics, most image sensors, spectrometers, modulators, and photodetectors are based on BEOL and show higher potential for commercialization, as shown in Figure 1.1. The cost of image sensors and spectrometers in infrared and terahertz regions is very high. The integration of conventional infrared sensors in Si CMOS is complex and not monolithic. Graphene offers immediate benefits in this sector, leading to low-cost ultrafast optical modulation, extreme broadband photodetection (ultraviolet  –  terahertz), high-performance LEDs, and data communications [2, 4, 5]. Some of these applications and their working principles are discussed here.

1.4 ­Photodetection in Graphene Graphene photodetectors are first categorized based on the intrinsic mechanism combined with the role of graphene and silicon. The intrinsic device mechanism plays a crucial role in developing the device scheme for integrating graphene with the CMOS circuit. Here, we discuss some well-studied graphene photodetection mechanisms and related 2D material-based photodetectors such as photovoltaic, bolometric, photothermoelectric (PTE), photogating, and plasmon-wave-assisted mechanisms [5, 7]. The photovoltaic effect emerges from the built-in electric field at graphene–metal interfaces, gate-induced potential differences, or spatially nonuniform intrinsic doping in graphene, whereas, if the incident light induces local variation in temperature resulting in a change in resistance followed by the current passing through the device in the presence of an electric field, it is termed as bolometric effect. On the other hand, the PTE effect refers to a light-induced thermal gradient across the junction of two materials with different Seebeck coefficients. In a phototransistor structure, when graphene is used as a channel, a finite DC voltage can be generated in the device as a response to the oscillating radiation field in a plasma wave-assisted photodetection mechanism. Here, graphene acts as a cavity for plasma waves – collective density oscillations. To be specific, when the plasma waves are weakly damped, in other words when the plasma waves launched in one of the electrodes of the graphene channel propagate to the other electrode within

1.4  ­Photodetection in Graphen

the momentum relaxation time, then these plasma waves form standing waves, leading to resonantly enhanced response [5, 7]. The physical mechanism and crucial parameters of these different types of photodetection in graphene are discussed in detail below, which are the key mechanisms to understand the role of graphene in Si CMOS-integrated photodetection.

1.4.1  Performance Metrics To evaluate the performance of photodetectors and compare them with other detectors, it is essential to understand some of the crucial performance metrics such as responsivity  , quantum efficiency (η), response time, and noise equivalent power (NEP). We briefly discuss these parameters to understand the general concepts associated with the photodetector. First, we consider the power of the incident light on the photodetector as P with the photon energy, EPh  =  hϑ. Then, we can P . Based on the operating mode, the photovoltage estimate the photon flux, h or photocurrent can be defined as ∆V = Vl − Vd or Ipc = Il − Id. In the same way, the responsivity can be defined in terms of either photocurrent or photovoltage, 

I pc

P

or DV P

(1.2)

The external quantum efficiency is the ratio of photoinduced charges per second to the number of incident photons per second, ext

PC

eo



(1.3)

In the same way, the internal quantum efficiency (IQE) is defined by considering the number of photons absorbed instead of the total incident photons,

PC int

eo



(1.4)

abs

Here, Φabs is the total absorbed photon flux. IQE can also be used to calculate the current gain of a photodetector. The characteristic time of a photodetector to switch between the illuminated (ON) and dark state (OFF) is known as photoresponse or photoswitching time. The bandwidth B of the photodetector can be defined as the cutoff frequency of the photodetector fc ≈ 0.55/τ but often limited by the RC time constant of the circuit [8]. NEP corresponds to the incident optical power required to produce the signal-tonoise (SNR) ratio of one in 1 Hz bandwidth.

1.4.2  Photovoltaic Effect The photovoltaic effect generally emerges when photogenerated carriers are separated and transported due to the built-in potential in the photoactive material. This built-in or internal electric field can be created either by contacting a semiconductor

5

6

1  Graphene for Silicon Optoelectronics

with metal as a Schottky junction or between differently doped (n- or p-type) semiconductors as a p–n junction [7]. These devices, in principle, function in both photovoltaic and photoconductive modes. The external bias is zero in PV mode, and the only electric field is internal. Thus, the current measured in this mode is the shortcircuit current (Isc). In the open-circuit mode, the generated photocarriers accumulate at the terminals to compensate the built-in electric field, leading to finite open-circuit voltage (Voc), often used to measure the generated electrical power in solar cells. The lower dark current in this mode allows higher specific detectivity and lower NEP [2, 4, 7]. Photoconductive mode is activated in the device when a finite bias is applied to the terminals, which is commonly observed in Schottky diodes and p–n junctions in reverse bias. In this mode, photocarriers are quickly swept out of the device, leading to higher photocurrent, and hence exhibit lower response time and higher responsivity than in photovoltaic mode. The combination of photoconductive and PV affect has been observed in both planar and vertical junctions of metal–graphene and other graphene–2D planar junctions. In the case of PV effect in electrostatically doped p–n junctions of graphene or related 2D material, the typical width of the inplane depletion region extends in the range of 200 nm, facing challenges with relatively low efficiencies (>30%) as difficult to maintain large depletion widths to capture the photogenerated carriers, where a large fraction of carriers recombine before moving into either the p or the n region [9, 10]. Vertical junctions formed by graphene and related 2D heterostructures circumvent these issues like the depletion region confines within one or two atomic thicknesses of 2D materials and extends for the entire area of the 2D junction. For instance, Schottky junctions made of graphene-TMDs can accommodate large built-in fields, reaching responsivity larger than 10 mA W−1 and IQEs of ~30% [11]. Much larger responsivities and IQEs have been observed in vertical 2D heterostructures, mostly due to their large active area.

1.4.3  Photoemission in Graphene Schottky Junctions The most basic physical concept that needs to be understood in this context is the electron emission mechanism in heterointerfaces of graphene and other related 2D materials, their multilayers, and heterostructures, along with bulk 3D systems. The initial understanding of 2D-based interfaces and junctions relies on classic thermionic and field-emission theories, such as Richardson–Dushman and Fowler– Nordheim models  [11]. However, electron emission in 2D materials and their interfaces is significantly different from bulk materials that have to be incorporated in transport models. Several models describing the charge transfer at 2D interfaces propose fundamentally different charge-transfer mechanisms. Therefore, it is essential to understand the difference between these models and the classic thermionic and field-emission models. A brief discussion is provided in this chapter on various charge transport models of 2D interfaces, with an emphasis on graphene-based Schottky junctions. Another challenge in this area is identifying the dominating emission mechanism that decides the different charge transport and emission mechanisms at the

1.4  ­Photodetection in Graphen

interface. In any Schottky junction, there are three main electron emission pathways, (i) thermionic emission, (ii) quantum mechanical tunneling, and (iii) photoemission. We first discuss these three emission pathways in graphene-Schottky junctions in different conditions and explain the differences between these models and the classical electron emission models. In general, multiple electron emission mechanisms can coexist simultaneously, thus identifying that the dominating emission mechanism is crucial for designing a device with optimum performance. The coupling of multiple emission mechanisms reveals exciting charge transport phenomena such as the photoenhanced thermionic emission – a combination of thermionic emission and photoemission, leading to the detection of photons with energies lower than the potential barrier at Schottky interfaces. In the same way, when the photoemission is coupled with field emission, a photoassisted field emission emerges, or a thermal-assisted field emission dominates at a finite temperature. Electron emission dynamics are also dependent on the parameters of the incident laser and applied bias across the interface, where ultrafast laser with optimum pulse width and intensity can enhance the coupling between different emission mechanisms in materials such as graphene with unique electronic properties. In such conditions, novel quantum mechanical phenomena such as photothermionic (PTI) emission, multi-photon field emission, and dynamical optical field emission can be investigated in solid state systems. In the case of large interface currents, the electron emission is limited by the space charge, where the electric field of the charge carriers in transit restrains further charge transfer at the interface.

1.4.4  Thermionic Emission in Graphene-based Interfaces Heating electrons in a material to overcome the potential barrier is defined as thermionic emission, a fundamental physical process in electronic devices. This leads to various functionalities in electronics, optoelectronics, and energy conversion. In thermionic emission, electrons are thermally excited to overcome the potential barrier in a heterostructure. The probability of occupation in higher energy states for an excited electron can be estimated by the Fermi–Dirac distribution. There are different ways to excite electrons in a system, such as thermal excitation, photoexcitation, and scattering of electrons in the presence of the electric field. In all these processes, when the excited electrons are in thermal equilibrium with the lattice, the emission process can be described with the thermionic emission model. By analyzing the charge density (J) as a function of applied bias, the charge transport in a Schottky junction can be approximated by the simple diode equation, J

J0 e

qV

kT

1 , J0

A* T 2e

B

/ kT



(1.5)

where J0 is the reverse saturation current density, q is the elementary charge, η is the ideality factor, k is the Boltzmann constant, V is the applied bias, T is the temperature, ΦB is the Schottky barrier height, and A* is the effective Richardson’s constant. The ideality factor η is used to determine the quality of the Schottky interface, and

7

8

1  Graphene for Silicon Optoelectronics

Table 1.1  Comparison of properties of 2D and bulk materials. Property

Bulk material

2D material

Crystal periodicity

Well defined in all 3D

Well defined only in 2D

E–k dispersion

Parabolic

Parabolic/linear

Electrical contact

No qualitative difference

Different (edge/surface)

Electron k∥ component

Strictly conserved

Nonconserved

high-quality graphene interfaces often show near-unity ideality factor. In a metal– semiconductor Schottky junction, the barrier height is constant for reverse bias current, indicating Fermi-level pinning due to large interface states, whereas in graphene–Schottky junction, for low-energy electrons, the Dirac cone approximation provides a reliable model for thermionic emission in graphene. It is essential to evaluate the limitations of the thermionic model for 2D materials. The validity of the Richardson–Dushman thermionic emission model in 2D materials must be carefully scrutinized. Table 1.1 presents the critical differences between bulk and 2D materials. The schematic drawings of one-dimensional edge contact and 2D surface contact of graphene are shown in Figure  1.3a. The initial models inspiring these studies are from semiconductor quantum well structures as displayed in Figure 1.3b. The general assumptions of the RD thermionic emission model are listed below. 1) Electrons follow parabolic energy–momentum dispersion. 2) Crystal periodicity is uniform in all directions. Thus, electrically contacting 3D material in any order does not change the thermionic emission mechanism. 3) The tunneling probability of electrons can be expressed entirely as a function of the out-of-plane component of the momentum (kz). For graphene, due to finite DOSs at the Dirac point, much lower A* (in the order of 10−2 A cm−2 K−2) than the theoretical value of 112 A cm−2 K−2 have been observed and thus can be represented by the Landauer transport model as represented by the equation, J

q

T E D E

fg

fSi dE



(1.6)

where τ is the carrier injection time at the contact, T(E) is the transmission probability over the Schottky barrier ΦB, D(E) is the DOSs of graphene, and fg and fSi are the Fermi–Dirac functions of graphene and silicon, respectively. By incorporating the Landauer transport model in J0 as shown below, the standard diode Eq. (1.5) can be applied to the graphene–Schottky junctions, J 0

qDo

kT

2

B

kT

1 e

B

/ kT

(1.7)

The additional temperature dependence in Eq. (1.7) is the main difference with J0 in Eq. (1.5). The experimental studies show that the Richardson–Schottky model

1.4  ­Photodetection in Graphen Ti/Au Schottky metal contacts

W

Doped AlGaAs Spacer layer 2DEG GaAs

(a)

(b)

L

X

Z

Figure 1.3  (a) Schematic drawing of the one-dimensional edge and two-dimensional surface contact to graphene. (b) Schematic of the initial model of side-way thermionic charge injection of quasi-2D-electron gas in a semiconductor quantum well structure. DEG, dimensional electron gas. Source: Ang et al. [11]; John Wiley & Sons/CC BY SA 4.0.

and the Landauer transport model are indistinguishable as the dominant temperature dependence comes from the e B / kT in both the models. However, since this model mainly considers the limited DOS of graphene near the Dirac point, it is only suitable for low-energy range when ΦB >>kT and the low energy of excited carriers. A significant deviation in the fitting of the experimental data with this model has been observed for higher energy excited carriers much above the Dirac point [11].

1.4.5  Hot Electron-based Photodetection In photocarrier-driven detectors, optical transitions between excited and ground states are crucial as well as with a finite energy gap, whereas in the case of thermally driven detectors, both inter- and intraband transitions can contribute to the resulting photocurrent, eliminating the need for photon energy threshold and therefore exhibiting a large spectral responsivity. One of the pathways for relaxation of these exciting carriers is via carrier–carrier scattering resulting in the redistribution of energy within the electronic system leading to electronic bath temperature. Such a rise in average carrier temperature can generate a photocurrent or -voltage, leading to various hot-electron-based photodetection mechanisms such as PTE effect, bolometric effect, and PTI emission. The redistribution of excited carrier energy between the surrounding carriers can be viewed as thermalization within the electronic system but does not necessarily include lattice heating and thus can be represented by Fermi–Dirac distribution with elevated average electronic temperature. The overall ultrafast photocarrier dynamics are depicted in Figure  1.4. Photoexcitation in graphene can happen within  T0). The total carrier relaxation or cooling includes emission of strongly coupled optical phonons at hot-electron energies higher than ~200 meV, followed by acoustic phonons at low- energy range. How different stages of hot electron cooling leads to various photoresponse effects in graphene- based heterostructures are discussed below.

9

1  Graphene for Silicon Optoelectronics Te > T0

(a)

SCOP

(b)

(c)

Carrier multiplication

Photoexcitation

Acoustic phonons

(d) Phonon-dominated cooling

ΔTe

10

Time (ps)

e–e scattering (< 100 fs)

SCOP emission (~1 ps)

Phonon emission (acoustic, substrate), and supercollision, ~20 ps)

(e)

Figure 1.4  Ultrafast photocarrier dynamics in graphene. (a) Photoexcitation in graphene with the incident of ultrashort light pulse ( T0). Carrier relaxation or cooling in ~1.3 ps via emission of (c) strongly coupled optical phonons (SCOP) at hot-electron energies higher than ~200 meV and (d) acoustic phonons at low-energy range. (e) Evolution of differential electron temperature in the time domain.

1.4.5.1  Photothermoelectric Effect (PTE)

If the light absorbed in a part of the photoactive material heats the electronic temperature Te, a thermal gradient ∇Te emerges between the hot and cold ends of the material, resulting in a finite potential gradient ∇V originating from the net charge imbalance from the thermal diffusion of carriers. The relation between ∇V and ∇Te is proportional to the Seebeck coefficient as V

S

Te

(1.8)

For metals, the Mott formula can be used to estimate the Seebeck coefficient: S

2 2 kBTe

3e0

1 EF

E E



(1.9)

Here, σ(E) is the energy-dependent conductivity. The above formula is valid only for kBTe EF . In graphene PTE photodetectors, based on the differences in the Fermi-level positions, a sixfold pattern of photocurrent can be observed that helps to distinguish it from the PV effect. The PTE effect in graphene with sixfold pattern of photoresponse is clearly evident in a single-layer graphene (SLG) connected by source (S) and drain (D) contacts and dual gates where the top gate is separated by the h-BN layer, and the bottom gate by SiO2, as shown in Figure 1.5. It has been verified that the PTE current becomes zero when there is no difference between the positions of Fermi levels in

11

1  Graphene for Silicon Optoelectronics TG BN D

S

0.4

Si SLG-Contact

p–n junction

SLG-Contact

–3

3 nA

2 m

PC (nA) 1

pn 2

0

0

–0.2 –0.4 –10

(a)

1

0.2

SiO2 VTG (V)

12

–5

0 5 VBG (V)

10

15

–1

(b)

Figure 1.5  Photothermionic effect in graphene-based device. (a) Schematic of single-layer graphene (SLG) connected by source (S) and drain (D) contacts and dual gates where the top gate is separated by the h-BN layer, and the bottom gate by SiO2. (b) Photocurrent mapping by varying top and bottom gating at the position of yellow mapping in the panel in (a). Source: Mathieu [8]/CC BY SA 4.0.

the two regions. Unlike the PV effect, electrons and holes contribute to the PTE effect and cancel each other. In the PV case, the polarity of photocurrent changes only once when both Fermi levels are equal. In graphene-based PTE photodetectors, there is always a trade-off between performance metrics such as IQE and the relaxation time based on the cooling time. For instance, long cooling times give higher IQE but with lower response times. The relaxation of hot carriers in graphene depends on various relaxation pathways, and in general, the cooling times are ~1 ps, implying large theoretical bandwidth of ~500 GHz and IQE within 20% [7, 16, 17]. Graphene with almost uniform optical absorption from MIR to UV region is highly promising in designing broadband photodetectors based on the PTE effect. 1.4.5.2  Photobolometric Effect (PBE)

When the transport conductance (Ge) of material changes due to heating induced by the photon absorption, I PC

Ge V Te T

(1.15)

One of the crucial differences between PTE and photobolometric effect (PBE) is that the photoinduced heat modifies the transport properties of the material rather than the transport of charge carriers, as in the case of the PTE effect. In this case, the conductivity change is due to the carrier mobility’s dependence on temperature. So, unlike the PTE effect, external bias is required to drive the photocurrent in PBE and can emerge in homogeneous materials without a junction. The photoresponse time of the PB effect is very similar to the PTE effect as they both rely on the cooling of the hot electrons. Several studies have investigated the PBE in graphene and other 2D materials  [18–20]. However, the responsivities reported in these studies are relatively low (~0.2 mA W−1) compared with PTE-based devices.

1.4  ­Photodetection in Graphen

1.4.5.3  Photothermionic (PTI) Effect

Ever since the discovery of the photoelectric effect, the emission of electrons from a solid by the absorption of photons, gaining importance both in terms of fundamental physics and real-world applications. Schottky junction provides a simple device scheme to photoelectric effect with the emission of photoexcited electrons into another material, known as internal photoemission (IPE). This IPE led to the development of visible to infrared photodetectors [21, 22]. However, the efficiency of these devices drops sharply if the photon energy drops lower than the interfacial barrier height. Thermalized carriers or hot electrons, via electron–electron scattering, increase the transfer efficiency of IPE by raising a fraction of electrons above the interfacial barrier. This phenomenon, generally called the PTI effect [23, 24], is inefficient in metals due to strong electron–phonon interaction. High thermal conductivity instantly brings electrons to thermal equilibrium with the surroundings in ~100 fs. Choosing a photoabsorber with a weak electron–phonon interaction can hold most of the absorbed photon energy within the electronic system, resulting in a large population of hot electrons with energies much higher than that of direct transitions. There is a promising approach to photodetection of photon energies much lower than the Schottky barrier height. By choosing a material where hot carriers are weakly coupled with the surrounding phonon bath, such as graphene [23, 24], thermalization of photoexcited carriers with other carriers results in hot carrier distribution with a well-defined temperature, Te [25, 26]. Realizing sufficient Te that can overcome the Schottky barrier height, a large pool of electrons that can be emitted as photon energies much lower than the barrier height can still contribute to the hot electron distribution and thus lead to the photocurrent over the barrier. 1.4.5.4  Photogating Effect

Low-intensity light detection, even when the detection limit approaches the singlephoton level, requires a gain mechanism for photocurrent generation where a single-photon absorption gives multiple carriers. Photogating is one such approach for gain in photodetectors. Instead of being used as a light-absorbing medium, graphene can also play a role as a transparent, high carrier mobility channel material and sense the photogenerated carriers in the adjacent semiconducting absorber, thus extending the operational bandwidth to midinfrared with high-performance metrics. In graphene-based phototransistors, the principle of photogate is the detection of change in resistance of graphene with the corresponding light-induced change in the substrate potential. The photogate effect can leverage the large transconductance of graphene-based transistors. In this case, graphene is capacitively coupled to the substrate, and the evolution of photogenerated carriers in the substrate reflects in the conductivity of graphene. For this, the photon energy of the detected radiation must be higher than the bandgap of the semiconductor substrate that generates electron–hole pairs and separates into different zones by the electric field. These separated charges appear at the semiconductor’s surface and alter the effective chemical potential in the substrate. Due to capacitive coupling, this change in chemical potential in the substrate reflects in the charge carrier density with

13

14

1  Graphene for Silicon Optoelectronics

opposite polarity in graphene. The resultant voltage on the graphene–oxide–semiconductor (GOS) structure [27] can be represented as e Vg

eFd

gr

Si

1 4 e2 nd

eFd

(1.16)

Here, Vg is the gate voltage, F is the electric field between the substrate and graphene, d is the thickness of the oxide, μgr, Si is the chemical potential of graphene and silicon, n is the excess interface charge density, and ε is the dielectric constant. To further improve the sensitivity of this device scheme, the substrate is kept in a nonequilibrium deep-depletion state, which significantly improves the photogenerated electron–hole pair separation and holds one type of charge that capacitively couples with the opposite charge in the graphene channel [28]. In this concept, to detect the photogenerated carriers, the semiconductor in a GOS device is kept in deep depletion by rapid application of back-gate voltage. The energy band diagram of GOS device with n-type Si biased in the deep-depletion state is depicted in Figure 1.6. Positive gate voltage drives mobile electrons into the bulk Si and space charge into the deep-depletion region of width W. The depletion region separates the electron–hole pairs, and the holes collect at the silicon–oxide interface as the surface potential goes negative. As the graphene capacitively couples with the Si

Sou

Graphene

rce

Dra

in

Graphene

e

xid

O

rb

so

Ab

er Gate

(a)

(b) Ef-G –qΨ

EC EV

W

(c)

Depletion Vbg > 0 V

Space charge

Photons

Ef-G –qΨ

(d)

–q(Vbg+Vfb)

W

–q(Vbg+Vfb) EC EV

Hole photo charge Vbg > 0 V

Figure 1.6  Deep-depletion GOS device and operating principle. (a) Schematic of the GOS device. (b) Optical image of several GOS devices integrated in an Si/SiO2 substrate. (c) Band diagram of GOS device in the deep-depletion state. Ef-G is the Fermi level of graphene, and the combination of both gate and channel bias (Vbg + Vfb) acting on the band bending near the interface, assuming lightly n-doped silicon. (d) Under illumination, a large density of holes accumulates in the interface inducing a negative charge and thus the n-type doping and Fermi-level shift in graphene. Source: Howell et al. [28]; Springer Nature/CC BY 4.0.

1.4  ­Photodetection in Graphen

substrate, holes collected at the interface induce electrons in the graphene channel and modify the Fermi level reflected in the channel conductivity. This change in conductivity can be sensed from the drain–source current when a slight bias is applied. This phenomenon is well described in Figure 1.6. Here, it is worth noting that the deep-depletion state occurs when the gate voltage sweep is fast enough to avoid the formation of the inversion layer or if the device cannot maintain thermal equilibrium  [27, 28]. The rate of gate voltage change can be expressed with the relation, dVg  qni dt 2Cox

SiVt



(1.17)

p

where Vg is the gate voltage, n is the interface charge density, Vt is the thermal voltage (~26 mV), Cox is the oxide capacitance, and τp is the minority carrier lifetime. dVg For lightly doped Si (Nd ~9 × 1013cm−3), the typical values are ~0.8 V s−1. dt There is another approach for the photogate effect in graphene with an absorber material such as quantum dots (QDs), semiconductors, and pyroelectric materials. Since the role of graphene in these detectors is only charge sensing, the choice of the absorber decides the spectral response  [29–31]. Therefore, this device concept is equally applicable and suitable for detecting infrared to ionizing radiation. The sensitivity of this photodetector is determined by both the transconductance of the graphene and the generation recombination of the absorber. Thus, the full functionality of the device can be realized only when graphene and the absorber are incorporated into a single device, as shown in Figure 1.7a. The key response of the device is the shift in the channel resistance versus the gate voltage, as shown in Figure 1.7b. The short carrier transit time reflected in the fast response time makes graphene an

R (kΩ)

40

Light Quantum dots

Graphene

SiO2

–40 VDS

(a)

IDS

0

10

VBG (V)

40

20 35 R (kΩ)

10 fW 23 fW 0.16 pW 0.84 pW 34.9 pW

10

Si

20 2 0.2 0.02

30 20

Au

VBG

Power (pW)

Light

–20

0 20 VBG (V)

40

60

(b)

Figure 1.7  (a) Schematic of a graphene-quantum dot hybrid phototransistor. (b) Channel resistance versus gate voltage of the graphene-quantum dot hybrid phototransistor. Source: Reproduced with permission from Konstantatos et al. [29]; Springer Nature.

15

16

1  Graphene for Silicon Optoelectronics

excellent material for high-gain photodetection at room temperature [29]. Some of these light-absorbing particles showing favorable responses in sensitizing graphene are colloidal QDs made of PbS, ZnO, and CdS [29–31]. By varying the size of these quantum detectors, strong light absorption and bandgap tunability can be achieved along with spectral response range from UV to short-wave infrared (SWIR). Another advantage of these QDs is the easy integration as they can be processed in solution and deposited by spin-coating, ink-jet-printing, and contact-printing techniques. In addition, the surface of these particles can be optimized for efficient charge transfer within QDs or with graphene. Large QE values >25% have been demonstrated in these hybrid photodetectors.

1.4.6  Infrared Modulators Integrating with other materials is inevitable to extend the operational bandwidth of Si modulators into the midinfrared region. With its straightforward CMOS foundry-compatible processing, graphene became a viable choice for IR modulators. Based on the Fermi-level dependence of the optical absorption in graphene, several graphene-Si waveguide-integrated IR modulators have been demonstrated. The working principles of these modulators are simply based on tuning the optical absorbance in graphene by varying the Fermi-level dependence of the optical absorption in graphene. For instance, in graphene waveguide-integrated optical modulators, graphene is placed on Si/SiO2 ridge waveguide where Si is used to electrostatically tune the Fermi-level position in graphene (Figure  1.8a,b). The waveguide is designed explicitly for the operation at 1.53 μm wavelength, and most of the optical mode intensity lies in the graphene/SiO2 region [32]. The blue dotted lines in Figure 1.8c in both the valence and conduction bands of graphene indicate the incident photon energy range where Fermi level can be tuned with applied gate bias or drive voltage from below to above the two lines. If EF is placed below the blue bottom line, there are no available electrons for interband transition, giving the lower absorption. When the Fermi level rises above the blue bottom line, the optical absorption increases gradually and decreases when EF approaches the top dotted blue line. If the EF crosses the top blue line, then there are no available states to transfer the electrons in the conduction band; thus, the absorption is dropped. This is depicted in the transmission versus drive voltage plot in Figure 1.8c. The switching speed of these modulators is ~1 GHz, which is mainly limited by the RC time constant. With this device structure, modulation of ~0.09 dB μm−1 can be achieved [32]. The optical absorption can be increased by systematically increasing the number of graphene layers, where the configurable regions with graphene– insulator–graphene structures demonstrated modulation depth of ~0.16 dB μm−1.

1.4.7  Photovoltaic Devices Converting light to electricity is the primary function of a photovoltaic cell. Pmax and can also be The energy conversion efficiency can be defined as Pin represented as

1.5  ­Outloo

ne

Au

Pt

Graphe

Pt

Si

Graphene Al2O3

Graphene Al2O3 Si

Si SiO2

(b)

Transmission (dB μm–1)

(a) 0.00

EF EF

–0.03 –0.06

hv/2 –hv/2

EF

–0.09 –6

–4

(c)

–2

0 Drive voltage (V)

2

4

6

Figure 1.8  (a) 3D sketch of graphene-integrated Si waveguide optical modulator. (b) Cross section of the device with the optical mode overlay. (c) Optical transmission as a function of the drive voltage. Source: Liu et al. [32]; Springer Nature.



VOC ISC FF Pin

(1.18)

VM ax I MAX , and VOC is the maxiVOC ISC mum open-circuit voltage. VMax and IMAX are the maximum voltage and current values. Similar to photodetectors, the fraction of absorbed photons converted to current defines the internal photocurrent efficiency. Graphene can fulfill many roles in photovoltaic devices as transparent conductive window, charge transport channel, and the broadband photoabsorbing material. Graphene and its derivatives have been used as transparent electrodes in inorganic, organic, and dye-sensitized solar cells  [33, 34]. The present photovoltaic market is strongly dominated by silicon-based solar cells with a maximum quantum efficiency of ~25%  [4]. Even though the quantum efficiencies of these solar cells are much lower than the silicon solar cells, fabrication costs can be significantly minimized as they can be prepared in a roll-to-roll process. Here, FF is the fill factor and defined as FF

1.5 ­Outlook Heat transduction into electrical energy is pivotal in designing next-generation miniaturized electronics. Distributing the absorbed photon energy within the electron bath and effectively transporting it out of the system is an essential but challenging task that must be addressed with emerging materials and device schemes. Graphene with weak electron−phonon coupling provides an ideal platform to harness

17

18

1  Graphene for Silicon Optoelectronics

excessive heat in the electronic system. The vertical and lateral graphene-bulk (3D) semiconductor heterostructures open a reliable path to commercialization while enabling a testbed to explore hot electron-based transport from graphene to 3D systems. Therefore, future research needs to be focused on developing large-scale, CMOS-compatible graphene-Si optoelectronic devices suitable for a wide range of applications discussed above with room temperature operation.

­References 1 Castro Neto, A.H., Guinea, F., Peres, N.M.R. et al. (2009). The electronic properties of graphene. Reviews of Modern Physics 81 (1): 109–162. 2 Akinwande, D., Huyghebaert, C., Wang, C.-H. et al. (2019). Graphene and twodimensional materials for silicon technology. Nature 573 (7775): 507–518. 3 Ferrari, A.C., Bonaccorso, F., Fal’ko, V. et al. (2015). Science and technology roadmap for graphene, related two-dimensional crystals, and hybrid systems. Nanoscale 7 (11): 4598–4610. 4 Bonaccorso, F., Sun, Z., Hasan, T., and Ferrari, A.C. (2010). Graphene photonics and optoelectronics. Nature Photonics 4 (9): 611–622. 5 Sensale-Rodriguez, B. (2015). Graphene-based optoelectronics. Journal of Lightwave Technology 33 (5): 1100–1108. 6 Kampfrath, T., Perfetti, L., Schapper, F. et al. (2005). Strongly coupled optical phonons in the ultrafast dynamics of the electronic energy and current relaxation in graphite. Physical Review Letters 95 (18): 187403. 7 Koppens, F.H.L., Mueller, T., Avouris, P. et al. (2014). Photodetectors based on graphene, other two-dimensional materials and hybrid systems. Nature Nanotechnology 9 (10): 780–793. 8 Mathieu, M. (2017). Ultrafast Optoelectronics in 2D Materials and Their Heterostructures. Universitat Politècnica de Catalunya. 9 Peters, E.C., Lee, E.J.H., Burghard, M., and Kern, K. (2010). Gate dependent photocurrents at a graphene p–n junction. Applied Physics Letters 97 (19): 193102. 10 Lemme, M.C., Koppens, F.H.L., Falk, A.L. et al. (2011). Gate-activated photoresponse in a graphene p–n junction. Nano Letters 11 (10): 4134–4137. 11 Ang, Y.S., Cao, L., and Ang, L.K. (2021). Physics of electron emission and injection in two-dimensional materials: theory and simulation. InfoMat 3 (5): 502–535. 12 Gabor, N.M., Song, J.C.W., Ma, Q. et al. (2011). Hot carrier assisted intrinsic photoresponse in graphene. Science 334 (6056): 648–652. 13 Mueller, T., Xia, F., and Avouris, P. (2010). Graphene photodetectors for high-speed optical communications. Nature Photonics 4 (5): 297–301. 14 Gan, X., Shiue, R.-J., Gao, Y. et al. (2013). Chip-integrated ultrafast graphene photodetector with high responsivity. Nature Photonics 7 (11): 883–887. 15 Freitag, M., Low, T., and Avouris, P. (2013). Increased responsivity of suspended graphene photodetectors. Nano Letters 13 (4): 1644–1648. 16 Xia, F., Mueller, T., Lin, Y.-m. et al. (2009). Ultrafast graphene photodetector. Nature Nanotechnology 4 (12): 839–843.

  ­Reference

17 Shiue, R.-J., Gao, Y., Wang, Y. et al. (2015). High-responsivity graphene–boron nitride photodetector and autocorrelator in a silicon photonic integrated circuit. Nano Letters 15 (11): 7288–7293. 18 Schuler, S., Schall, D., Neumaier, D. et al. (2016). Controlled generation of a p–n junction in a waveguide integrated graphene photodetector. Nano Letters 16 (11): 7107–7112. 19 Freitag, M., Low, T., Xia, F., and Avouris, P. (2013). Photoconductivity of biased graphene. Nature Photonics 7 (1): 53–59. 20 Yan, J., Kim, M.H., Elle, J.A. et al. (2012). Dual-gated bilayer graphene hot-electron bolometer. Nature Nanotechnology 7 (7): 472–478. 21 Scales, C. and Berini, P. (2010). Thin-film Schottky barrier photodetector models. IEEE Journal of Quantum Electronics 46 (5): 633–643. 22 Emmanuel Rosencher, B.V. (2002). Optoelectronics. Cambridge: Cambridge University Press. 23 Massicotte, M., Schmidt, P., Vialla, F. et al. (2016). Photo-thermionic effect in vertical graphene heterostructures. Nature Communications 7 (1): 12174. 24 Massicotte, M., Schmidt, P., Vialla, F. et al. (2016). Picosecond photoresponse in van der Waals heterostructures. Nature Nanotechnology 11 (1): 42–46. 25 Tielrooij, K.J., Piatkowski, L., Massicotte, M. et al. (2015). Generation of photovoltage in graphene on a femtosecond timescale through efficient carrier heating. Nature Nanotechnology 10 (5): 437–443. 26 Massicotte, M., Soavi, G., Principi, A., and Tielrooij, K.-J. (2021). Hot carriers in graphene – fundamentals and applications. Nanoscale 13 (18): 8376–8411. 27 Voronin, K.V., Ermolaev, G.A., Stebunov, Y.V. et al. (2021). Photogating in graphene field-effect phototransistors: theory and observations. AIP Conference Proceedings 2359 (1): 020034. 28 Howell, S.W., Ruiz, I., Davids, P.S. et al. (2017). Graphene-insulator-semiconductor junction for hybrid photodetection modalities. Scientific Reports 7 (1): 14651. 29 Konstantatos, G., Badioli, M., Gaudreau, L. et al. (2012). Hybrid graphene– quantum dot phototransistors with ultrahigh gain. Nature Nanotechnology 7 (6): 363–368. 30 Sun, Z., Liu, Z., Li, J. et al. (2012). Infrared photodetectors based on CVD-grown graphene and PbS quantum dots with ultrahigh responsivity. Advanced Materials 24 (43): 5878–5883. 31 Guo, W., Xu, S., Wu, Z. et al. (2013). Oxygen-assisted charge transfer between ZnO quantum dots and graphene. Small 9 (18): 3031–3036. 32 Liu, M., Yin, X., Ulin-Avila, E. et al. (2011). A graphene-based broadband optical modulator. Nature 474 (7349): 64–67. 33 Junbo, W., Héctor, A., Zhenan, B. et al. (2011). Organic solar cells with solutionprocessed graphene transparent electrodes. Applied Physics Letters 92 (1): 2633021. 34 Lewis, G.D.A., Yi, Z., Cody, W.S. et al. (2010). Highly flexible, and transparent graphene films by chemical vapor deposition for organic photovoltaics. ACS Nano 4 (5): 2865–2873.

19

21

2 Growth and Transfer of Graphene for Silicon Optoelectronics 2.1 ­Introduction While Gr and other 2D materials do possess superior or complementary properties to Si, such as high carrier mobilities, more favorable optical properties (broadband absorption, strong light–matter interaction, and transparency), mechanical flexibility, and absence of surface dangling bonds, simple, cost-­effective, and large-­scale production and processing of Si-­based optoelectronic devices along with Si readout circuit make it an irreplaceable semiconductor at least in the foreseeable future. If the exotic properties of 2D materials are ever to be utilized for large-­scale commercial products, a scalable large-­scale manufacturing process of 2D materials is the key. The probability of establishing 2D materials-­specific fabrication lines in the near future is low. In this regard, integrating 2D materials with current Si fabrication lines is an alternate low engineering effort solution that can unleash the true market potential of 2D materials-­based photodetectors. The integration of 2D materials with traditional silicon-­based fabrication lines entails a number of aspects, including synthesis, transfer, dielectric deposition, and contact fabrication. Here, in this chapter, we discuss the details of Gr synthesis, transfer, and integration with Si. Among all the processes, producing Gr on the industrial scale is probably the most important step to realizing practical applications. For this purpose, chemical vapor deposition (CVD) has the potential to produce uniform and large-­area Gr with high quality. In this chapter, we emphasize the progress in the growth and transfer techniques of Gr on different substrates.

2.2  ­Growth of Graphene To understand the chemistry of the growth mechanism, it would be meaningful to have a brief overview of the effect of the interatomic bonding on the bulk properties of the carbon (C). The hybridized bonding among the C–C atoms can predicate the allotropic properties. Graphite and diamond are the most abundant allotropes of C in daily life [1]. Diamond is sp3-­hybridized C-­allotrope with insulating properties, Graphene for Post-Moore Silicon Optoelectronics, First Edition. Yang Xu, Khurram Shehzad, Srikrishna Chanakya Bodepudi, Ali Imran, and Bin Yu. © 2023 WILEY-VCH GmbH. Published 2023 by WILEY-VCH GmbH.

22

2  Growth and Transfer of Graphene for Silicon Optoelectronics Diamond

SP3 Graphene 2

2

2

1S , 2S , 2P Carbyne

Graphite SP

SP2

Figure 2.1  Electronic configuration of C-­atom and three-­phase hybridization states.

while graphite is sp2 hybridized and is a very good conductor with metallic behavior. The Gr can be supposed as the monolayer graphite having Fermi level at the Dirac point, which has the unique property to be doped as n-­or p-­type. A neutral C-­atom in its ground state has six electrons surrounding the nucleus by the orbital arrangement of 1s2, 2s2, 2p2 having two half-­filled p-­orbitals. The valence electrons (2s2, 2p2) are the measure of the reactivity of C-­atoms, as these are responsible for bonding with other atoms through hybridization, which includes sp, sp2, or sp3 hybridization. Diamond consists of tetrahedrally directed σ bonds with sp3 bonds. On the other hand, graphite consists of multiple sp2-­hybridized layers that have π bonds along with the layer, while weakly connected interlayers through the van der Waals forces [2]. A three-­phase schematic of the C-­allotropes with their hybridization is presented in Figure 2.1. The Gr is available in the forms of flakes (~μm in size) and sheets (~cm in size) [3]. The flakes can be easily produced through the exfoliation from the graphite using scotch tape at a laboratory scale. They can also be produced at the industrial scale, which is classified into Gr-­nanoplatelet (GNP) and reduced Gr-­oxide (rGO) [4]. The mechanical exfoliation assisted by the sonication of the graphite can be used to produce GNP, while the oxidation intercalation process using hydroxyl, carbonyl, or carboxyl functional groups can be employed to produce rGO. However, the crystal quality of the rGO produced through the oxidation process is not so good. The CVD growth process for the Gr consists of C-­precursor dehydrogenation at elevated temperatures to produce large-­area continuous Gr [5]. A generalized schematic for the production of Gr flakes and Gr films is presented in Figure 2.2a,b.

2.2.1  Growth Dynamics of CVD Gr and Choice of Substrate Although the crystal quality of the mechanically exfoliated Gr is very high, this method is not convenient for mass production as it is difficult to control the flake size and the number of the Gr layers. On the other hand, the CVD of Gr on metal substrates (Cu, Ni, Pd, Au, Ru, etc.) can produce large-­area high-­quality Gr sheets  [5–7]. Various kinds of CVD approaches can be utilized to synthesize Gr

2.2  ­Growth of Graphen Graphene nanoplatelet

Graphene flakes

Graphene oxide

Reduced graphene oxide

ical han Mec liation exfo

Graphite

Oxi inte dation rcal atio n

OH O

O

O O

O OH OH

O

(a)

OH

OH

Carbon precursor

Graphene film

Reduction O O OH

CVD graphene film

CVD

(b)

Figure 2.2  Schematic for the industrial-­level production of (a) Gr flakes and (b) Gr film. Source: Reproduced with permission from Jia et al. [5]; AIP Publishing.

films such as hydrogen-­free CVD, carbon-­enclosed CVD, plasma-­enhanced (PE) CVD, ultrahigh vacuum-­CVD, and oxygen-­free CVD. The choice of the substrates is also a very critical parameter as it may affect the crystal quality as well as the properties of the grown film. The surface diffusivity of the C-­atoms on the metal substrate is one of the most important parameters for the growth of Gr film, which can be found by the Arrhenius equation, DT

D0e

ED

k BT

(2.1)

where DT is the diffusion constant, D0 is the diffusion at equilibrium,  kB is the Boltzmann constant, ED is the activation energy, and T is the temperature. Hence, the diffusion length for the C-­atom can be found by L

2 DTc

1

2

(2.2)

where DTc is the diffusion of the C-­atom at temperature T, and τ is the time of diffusion. The other important parameters of the growth rate are the flux of precursors and the growth temperature of the substrate. The Gr is formed on the metallic substrates through diffusion, surface mediation, and segregation, where dissociated C-­atoms are dissolved into bulk metal and subsequently segregated to form a layer over the metal surface [8]. A large-­area substrate (which can be etched during the subsequent transfer process) is a key requirement for the production of the CVD-­Gr at an industrial scale. So, the choice of the substrate is also very important with the commercial perspective (cost and

23

2  Growth and Transfer of Graphene for Silicon Optoelectronics

availability). For this purpose, Cu and Ni are the most suitable, which are abundant and easily available. Cu is one of the most common substrates used for the growth of Gr due to low C-­solubility, low cost, and precise controllability for the Gr-­crystal quality as well as layer thickness. The growth follows the self-­assembled surface-­ mediated mechanism, which automatically stops when the whole substrate surface is covered with Gr layer.

2.2.2  Growth on Metallic Substrates

Relative energy (eV)

The basic demonstration for the growth steps of Gr by CVD is presented in Figure 2.3a, which includes gas-­phase reactions along with surface reactions. First, the decomposition of hydrocarbons occurs to produce C-­species, which happens in the gas phase as well as on the Cu substrate surface [11, 12]. The Cu acts as a surface catalyst, which can greatly reduce dehydrogenation energy barriers for the C-­species. The barrier for the decomposition of CH4 to CH3 (~4.85 eV) can be reduced (~1 eV)

I. Gas-phase reactions

H2 CH4

VIII. Desorption

II. Adsorption

5 4 3 2 1 0

III. Decomposition

CH4

CH3+H

CH2+H

CH+H

C+H

Cu(100)

VII. Merging into continuous film

(a)

Cu(110)

(b)

Compact Linear

–6.2

3C

–6.4

8C

4C 10C

–6.6 5C –6.8

–7

(c)

Cu(110) Cu(100) Cu(111)

Cu(111)

IV. Surface diffusion VI. Lateral growth V. Nucleation Cu

Formation energy (eV/atom)

24

6C 3

4

5

6

7

8

9

10

11

Carbon cluster size (number of atoms)

12

13C

13

(d)

Figure 2.3  Growth mechanism of Gr film on Cu substrate in the CVD system. (a) Steps for Gr growth on Cu, including both surface reactions and gas-­phase reactions. Source: Reprinted with permission from Jia et al. [5]; Copyright© 2021, AIP Publishing Ltd. (b) Energy barriers for the decomposition of methane on Cu(110) (black), Cu(100) (red), and Cu(111) (blue) surfaces. Source: Reproduced with permission from Wang et al. [9]; Royal Society of Chemistry. (c) The formation energies per C-­atom of 1D linear chains and 2D compact islands consisting of 3–13C-­atoms on Cu(111). (d) Top views of the relaxed 2D compact structures on Cu(111). Small orange and large gray balls represent C and Cu atoms, respectively. Source: Reproduced with permission from Van Wesep [10]; AIP Publishing.

2.2  ­Growth of Graphen

due to the presence of Cu, which acts as a catalyst  [9]. The energy barriers for Cu(111), Cu(100), and Cu(110) are given in Figure 2.3b. The quick collisions, as well as diffusion of the active C-­content (CH, CH2, or CH3), on the Cu substrate surface would form the bigger clusters (CxHy) due to the catalyzation process by Cu. The CH–CH is the most dominant species on the Cu surface compared to other C-­clusters, as the formation process of CH–CH species is exothermic with the release of 1.94 eV energy [10]. The different reactions over the surface can occur resulting in cluster formations, whose crystal structure depends on the available number of C-­atoms. This phenomenon can be well understood by Figure  2.3c,d. The one-­dimensional structure is energetically more stable for the cluster having more than 13C-­atoms, while clusters with a higher number of atoms may form two-­dimensional structures such as pentagons, hexagons, and heptagons. So, the transition from unstable liner to stable two-­dimensional structure requires a specific amount of energy. Exceeding this energy limit, the diffusion and nucleation process required to form the hexagonal cluster formation of C-­atoms occurs, which is presented in Figure 2.4a. The hexagonal cluster subsequently acts as the Gr seed or nucleating center, where the surrounding active C-­atoms would be diffused and attached, leading to multiplication in the size of Gr ring in a symmetrical fashion [12]. Consequently, a large-­area high-­quality single-­layer sheet of Gr can be achieved. The reactions CH4

CH3

1. Chemisorption 2. Desorption

C3H6

1 0.1

Asat =1

Ceq

Mole fraction

iv

Ccu

50–100% 3 3 30–50% 10–30% 5–10% 1–5% C6H5

(b) Asat 100 μm) into graphene-­based nanofilms by vacuum filtration (Figure 2.18a–f). After chemical reduction, camphor-­assisted cold-­grabbing separation, and 2800°C defect-­ healing processes, a translucent and free-­standing MAG was obtained with easy transferability. The thermal treatment ensures MAG as high-­crystallinity building blocks.

II

I

(a)

GO/AAO

rGO

Camphor/rGO/AAO

(102)

G

Si

(004) 11.4 nm

(103) (112)

MAG

(b)

1.2˚

(105) (104)

III

(002)

Intensity (a.u.)

38

D

2DT MAG 2D1 2D2

1200 1400 1600 2600 C

2800 C HOPG

(101) (110)

(c)

2 nm

(100)

(d)

MAG

5 nm

(e)

–50

(f)

–40

40

50

Raman shift (cm–1)

Figure 2.18  (a, b) Fabrication process of chemical synthesized multilayer Gr (MAG) films. (c–e) HRTEM images of MAG films. (f) Raman spectra of MAG films. AAO, anodic aluminum oxide. Source: Peng et al. [47] / John Wiley & Sons / CC BY 4.0.

2.6  ­Graphene Transfer on Flexible Silico

100 μm (a)

200 μm

2 mm (b)

(c)

Pt MAG Si

50 nm

(d)

Figure 2.19  (a–c) Photographs of MAG films integrated with Si. SEM images of MAG/Si heterojunction. (d) MAG/Si interface. Source: Reprinted with permission from Peng et al. [47]; Copyright© 2022, John Wiley and Sons.

Raman spectrum confirms the existence of a decoupled structure with a 2DT/ (2D2 + 2DT) value of nearly 10%  [48–50]. Compared to the frequency of C mode (43.4 cm−1) in highly oriented pyrolytic graphite (HOPG) (25 l), the C mode in MAG with the similar thickness shows a lower frequency of 42.9 cm−1, indicating that multilayer graphene (about 9–10 l) stacks with different relative orientations [51], which is consistent with the fitting result of the 2D peak. The sublimable camphor-­ assisted transfer method avoids the interface contamination of metal salts and polymers and does not involve mechanical forces, which often damage the film. The MAG can be transferred onto a standard CMOS silicon wafer without any etchant or support. The nanoscale thickness of MAG and vacuum filtration ensured a uniform thickness of MAG with a close-­stacked interlayer structure and its high adhesion due to atomic-­scale contact with Si wafer (Figure 2.18d). The control of the MAG interface reduces the nonuniformity of the contact and doping. The MAG films were etched into the desired size of pixel arrays using plasma and standard lithography steps used in microelectronics (Figure  2.19a). The ability to achieve conformal coating on silicon substrate makes MAG compatible with the top-­down CMOS process flow, as demonstrated in a packed image sensor with a 9 × 9 pixels array (Figure 2.19b,c).

2.6  ­Graphene Transfer on Flexible Silicon Monolayer Gr shows a fracture limit of 25% and Young’s modulus of 1 TPa, which are far superior to Si with a fracture limit of ≈1% and Young’s modulus of ≈130 GPa. The “strain engineering” approach provides a new degree of freedom in designing a wide range of potential applications. Details of electromechanical coupling and mechanical properties of Gr-­related 2D materials can be found in a recent review [50]. Generally, two approaches can build structures that can flex and bend:

39

40

2  Growth and Transfer of Graphene for Silicon Optoelectronics

BOX PDMS stamp

Thin Si bar

(b)

(c)

(a) Polyimide substrate

(d)

(e)

(f)

(g)

Figure 2.20  (a) Schematics of PDMS-­assisted transfer printing of thin Si microstructures to flexible polyimide (PI). (b–f) Schematics of the device fabrication process. (g) Photograph of real devices (photodetector arrays) on PI substrate. BOX, buried oxide. Source: Reproduced with permission from Ali et al. [52]; IEEE.

flexible materials or structures [51]. When a material is sufficiently thin, the bending strain decreases with thickness; placing the thin material into a neutral mechanical plane can minimize the bending strain. Our previous study [52] shows a flexible, ultrathin Si and Gr-­based metal–semiconductor–metal (MSM) UV photodetector with a high UV/Vis rejection ratio. In our photodetector device, single-­layer Gr acts as a thin, active electrode, which allows for the formation of an ultrashallow Schottky junction; thin Si imparts flexibility to the device and yields a high UV/Vis ratio, while the MSM structure allows for high-­speed photoresponse. Figure 2.20a–f shows the digital image of a fabricated MSM device array (with Si thickness of 200 nm) on polyimide (PI) substrate. Due to their flexible nature, fatigue tests, with bending up to 1000 cycles, show little change in performance.

2.7  ­Graphene Integration with Silicon in CMOS Process Integrating 2D materials with traditional silicon-­based fabrication lines entails several aspects, including synthesis, transfer, dielectric deposition, and contact fabrication. Direct CVD growth of 2D materials on Si requires high temperatures (1000 °C) and a metal substrate. The front-­end-­of-­line (FEOL) process, where a high temperature (1000 °C) is needed for the activation of dopants, may seem suitable for Gr growth and integration. However, any presence of metal can lead to highly mobile metal contaminations in Si, which can create deep traps and affect the overall performance of the device. For 2D integration during back-­end-­of-­line (BEOL), 2D materials will be relatively far from the active devices, hence metal contamination

2.8  ­Challenges and Future Prospective

may not be an issue (Figure 2.20 shows BEOL integration of Gr with Si). But processing temperatures for BEOL process are relatively low (104

10–9

(c)

–1.0 –0.5 0.0 0.5 Voltage (V)

300

280

260

240

–32

–36 10–11

320

–28 In (lo/T2)

10–7

1.0

T (K)

10–5

Current (A)

58

1.0

1.5

HOPG/Si:P HOPG/GaAs:Si

2.8

(d)

3.2

3.6 1000/T

4.0

Figure 3.6  (a) Device schematics. (b) Linear fit of the forward I–V curve according to the thermionic emission model. (c) Temperature dependence of the saturation current. Source: Reprinted with permission from Parui et al. [30]; Copyright© 2018, IEEE. (d) Richardson activation plots for HOPG/Si:P and HOPG/GaAs:Si junctions. Source: Reprinted with permission from Tongay et al. [23]. Copyright© 2009, AIP.

accounts for nonideality/ideality. To accommodate nonideality in reverse bias, a bias-dependent SBH is introduced (whose origin, as image-force lowering and interface states, has been discussed in Section 3.2). Figure 3.6 illustrates the extraction of the SBH (ΦB0) and the ideal factor η from the forward I–V curve of MSJ.

3.4 ­2D Materials and Schottky Junctions The most common Schottky junction based on 2D materials is between Gr and Si. Contrary to common metals, where Fermi energy of metals and consequently the SBH remains constant against voltage variations, Fermi energy of Gr is tunable with applied voltage, and hence, its SBH also varies. The Fermi energy variation of Gr with applied voltage and its effect on the SBH is demonstrated in Figure 3.7. When forward biased, Gr becomes p-doped, work function increases (Fermi energy shifts lower), and consequently, SBH also increases. While reverse biased, Gr is n-doped, which decreases the work function (Fermi energy shifts higher), and SBH also reduces.

3.4  ­2D Materials and Schottky Junction

EF

ФBo

++ ++

ФB

EF

Фbi

++ ++ +++

EFS

Фbi + V EC

ФB

EF

++

Фbi – V EFS

EFS (a)

EV

(b)

(c)

Figure 3.7  Fermi-level modulation of Gr in Gr/Si at (a) zero, (b) reverse, and (c) forward bias. Source: Reproduced with permission from Di Bartolomeo et al. [31]; IEEE.

Bias-induced variations in the Fermi level of Gr and SBH can be easily included in the ideal diode equation obtained from the TE model (Figure 3.8). Voltage-dependent change in the Fermi level of Gr can be expressed in the following equation: EF (V ) 

2

h

vF

nin (V )

(3.25)

where nin(V) is the voltage-induced carrier injections in the Gr. At a certain bias voltage V, charges induced in Gr are equal to the variation in density of the positive dopant ion concentration in the semiconductor depletion area. Therefore, nin(V) is given as n in (V )

ndepl

2s N

i

V kT / e / e

2s N

i

kT / e / e

(3.26)

Gr is usually doped during the fabrication and transfer process. It is already doped with a doping density of n0, then the total carrier density (ng) can be given as n0 + nin n g (V ) n0 nin (V )

2s N

i

V kT / e / e

(3.27)

50 0

50 Metal like (EF = const)

–50 –5

0

Richardson * = const) –Dushman (AG

–4

–3

–2 V (V)

–1

–50

ΔEF (meV)

j (mA cm–2)

Using the equation and applying condition nin ≪ n0, the change in Fermi energy can be written as

0

Figure 3.8  Dotted red line shows typical metal with no change in the reverse current or Fermi level with voltage. Solid red and blue line show change in the reverse current and Fermi level, respectively, with voltage for Gr. Source: Trushin et al. [32]; AIP Publishing.

59

60

3  Physics of Graphene/Silicon Junctions

EF (V )

h 2

1

vF

ng n0

n0

a

V kT / e

i

i

kT / e

(3.28)

The corresponding Fermi energy change can be written as

B (V )

0 B

0 B

B (V )

0 B

EF (V )

a

i

i

kT / e

V kT / e



(3.29)

where Φ0 and ΔΦ are the zero bias-Schottky barrier height and the correction to the ZB-SBH due to the applied voltage V. The equation for the reverse saturation current and diode equation can be modified as I 0

AA*T 2e

1 kT

0 B

e V Rs l

I0

I0 e

kT

1

B

V

AA*T 2e

AA*T 2e



1 kT

0 B

kT

a

0 B

i

a

i

V kT / e

V kT / e

i

i

kT / e



(3.30)

kT / e



e V Rs l

e

1 kT

(3.31)

1

Another important consideration for Gr/Si Schottky and Schottky junctions of other 2D materials is the Richardson constant. For a standard 3D metal/bulk semiconductor Schottky junction, the Richardson constant is estimated by considering the velocity vector component of electrons normal to the barrier plane and with kinetic energy greater than the barrier height. However, for 2D graphene on silicon, the current across the G/Si barrier can be estimated by the tails of electron wave functions pointing into the n-Si conduction band [32]. This leads to a much lower effective Richardson constant (Figure  3.7). A lower value of the Richardson constant will also result in a lower SBH value. As mentioned above, the Fermi pinning effect is highly related to the surface defects of the semiconductors caused by dangling bonds, device fabrication, and the MIGSs. Using 2D materials (metals and/or semiconductors), we can suppress the Fermi-level pinning effect by benefiting from the following factors. Firstly, 2D material surfaces are dangling bond free, so no surface states exist. Secondly, with the advancement in transfer techniques, as 2D materials are generally transferred without excessive use of solvents and chemicals, little or no damage is imparted to these materials. Thirdly, 2D metals can be transferred, instead of evaporated onto the semiconductors, hence no damage is done to the semiconductor surfaces. Fourthly, unlike 3D metals, 2D metals can suppress the MIGSs [33–36]. Apart from eradicating Fermi-level pinning, two more conditions need to be fulfilled to construct an excellent 2D semiconductor-based Schottky diode. First, a large depletion width is required, and secondly, SBH should be large enough. A smaller depletion width will increase the tunneling probability, resulting in a large reverse leakage current. In addition to a large SBH, a spacious depletion width, the

3.5  ­Challenges and Future Prospectiv

Schottky barrier width W of the Schottky junction, can be obtained by the following formula: W

2

s

q Nd

bi



(3.32)

The Schottky barrier width of all 2D Schottky junctions (such as MoTe2/MoS2) is generally smaller due to the high carrier concentration of >1019 cm−3 of atomically thin bodies of 2D materials. Generally, the sulfur vacancies are considered as the source of n-type doping of the monolayer MoS2  [37, 38]. By reducing the carrier concentrations through healing the sulfur vacancies, the depletion width can be enhanced, and better rectification characteristics can be obtained. By exploiting the benefits of using a 2D metal and a wider depletion width in 2D semiconductors, Zhang et  al. obtained all 2D Schottky junctions with a high rectification ratio (a rectifying ratio of >5 × 105) and near-unity ideality factor of ~1.6.

3.5 ­Challenges and Future Prospective Though Gr/Si junction has been modeled in several studies to explain the experimental results, a comprehensive theoretical picture of Gr/Si junction is still lacking. Compared to the traditional metal/semiconductors, the Schottky junction, the 2D nature of Gr, and the low density of states at the Fermi level make Gr/Si junctions physically unique and difficult to model. Two of the unique observations include lower than bulk semiconductors Richardson constant due to the 2D nature of Gr and the variation in SBH with the reverse voltage due to the low density of states at the Fermi level. Gr in Schottky junction also offers certain advantages. Fermi-level pinning in traditional semiconductors is a serious issue that limits the control over SBH and hence the electrical current passing through the junction. Fermi-level pinning in Gr/Si junction can be minimized due to the fact that Gr has a dangling bondfree surface, and hence, fewer or no chemical bonds are formed between Gr and Si (Figure 3.9). The absence of interface bonds can minimize the Fermi-level pinning. Moreover, deposition of bulk metals induces defects on the Si surface. However, as Gr is generally transferred rather than grown on Si, no or fewer defects are introduced on the Si surface, which also contributes to minimizing the Fermi-level pinning. Unlike 3D metals, 2D metals can suppress the MIGSs formed from the decaying metallic wave function of the 3D metal itself, again helping to minimize the Fermi-level pinning. Benefitting from the possibility of fabricating Fermi-level pinning, free and nearideal Schottky diodes with Gr, along with the better understanding of physics at Gr/ Si Schottky junction, can make such junctions ideal replacement for traditional bulk MSJs for a variety of electronic and optoelectronic applications including photodetectors and solar cells. Moreover, a deeper understanding of the Gr/Si junction physics will also allow a better control of the contact resistance, which plays a significant role in conventional electronic devices such as Schottky barrier-field effect transistor (SB-FET) and the metal–semiconductor field effect transistor (MESFET).

61

360

Z X

SiO2/p-Si substrate

Cr/Au 10 µm

Y

0

0.9

0.6

ΦSB = 501±26 meV

0.3 0.0

–20 –10 0 10 20 30 Gate voltage (V)

Schottky barrier height (eV)

Barrier height (eV)

1Tʹ-MoTe2/MoS2

Fitted line

(d)

S

(b) 1.2

1.2

Schottky–Mott role

0.9

Te

Ag

Cu

Au Pd

Work function (eV)

10–6 EF

MoS2

Cr

Pt MoS2 (f)

1Tʹ-MoTe2

150

200

250

300

Raman shift (cm–1)

ΦSBʹ

S = 0.09

0 100

(c)

1Tʹ-MoTe2

0.3 0.0

Pd

S=1

0.6

Mo

ΦSB

4.2 4.5 4.8 5.1 5.4 5.7 (e)

180 120 60

X (a)

240

Drain current (A µm–1)

Cr/Au

Top view

1H-MoS2 1Tʹ-MoTe2

1

300 Polarization angle (°)

1Tʹ-MoTe2 1Tʹ-MoTe2 Monolayer MoS2

Side view

Metal

Normalized intensity (a.u.)

1Tʹ-MoTe2

1H-MoS2 Metal

Cr

10–7

Pd/MoS2 diode 1Tʹ-MoTe2/MoS2 diode

10–8 10–9 10–10

Rectifying ratio: 11 and 101

10–11 –1.0 –0.5

(g)

0.0

0.5

1.0

Drain current (V)

Figure 3.9  (a) Schematic (top) and optical micrograph (bottom) of a device. (b) Crystal structures of the 1H-MoS2 (left) and 1T′-MoTe2 (right) monolayers. (c) Angle-resolved Raman spectra of the mechanically exfoliated 1T′-MoTe2 flake. (d) Barrier heights plotted against gate voltage for 1T′-MoTe2/MoS2 Schottky junction. (e) Schottky barrier height (SBH) plotted against the metal work function for evaporated metals and 1T′-MoTe2 on MoS2. The SBH value of MoTe2 sits nicely on the line (dashed line) calculated from the ideal Schottky–Mott model, while the SBH values for evaporated metals (blue line) deviate strongly form the Schottky–Mott mode. Band diagrams (f) and I–V (g) of the monolayer MoS2 diodes with the Pd and 1T′-MoTe2 electrodes. Source: Reprinted with permission from Zhang et al. [4]; Copyright© 2021, Nature Publishing Group.

  ­Reference

­References   1 T  ung, R.T. (2014). The physics and chemistry of the Schottky barrier height. Applied Physics Reviews 1 (1): 011304.   2 Sze, S.M., Li, Y., and Ng, K.K. (2006). Physics of Semiconductor Devices. Wiley.   3 Liu, Y., Guo, J., Zhu, E. et al. (2018). Approaching the Schottky–Mott limit in van der Waals metal–semiconductor junctions. Nature 557 (7707): 696–700.   4 Zhang, X., Liu, B., Gao, L. et al. (2021). Near-ideal van der Waals rectifiers based on all-two-dimensional Schottky junctions. Nature Communications 12 (1): 1–10.   5 Xu, Y., Cheng, C., Du, S. et al. (2016). Contacts between two-and three-dimensional materials: ohmic, Schottky, and p–n heterojunctions. ACS Nano 10 (5): 4895–4919.   6 Butler, S.Z., Hollen, S.M., Cao, L. et al. (2013). Progress, challenges, and opportunities in two-dimensional materials beyond graphene. ACS Nano 7 (4): 2898–2926.   7 Xia, F., Wang, H., Xiao, D. et al. (2014). Two-dimensional material nanophotonics. Nature Photonics 8 (12): 899–907.   8 Fiori, G., Bonaccorso, F., Iannaccone, G. et al. (2014). Electronics based on twodimensional materials. Nature Nanotechnology 9 (10): 768–779.   9 Akinwande, D., Huyghebaert, C., Wang, C.-H. et al. (2019). Graphene and twodimensional materials for silicon technology. Nature 573 (7775): 507–518. 10 Wan, X., Xu, Y., Guo, H. et al. (2017). A self-powered high-performance graphene/ silicon ultraviolet photodetector with ultra-shallow junction: breaking the limit of silicon? NPJ 2D Materials and Applications 1 (1): 1–8. 11 Liu, C.-H., Chang, Y.-C., Norris, T.B. et al. (2014). Graphene photodetectors with ultra-broadband and high responsivity at room temperature. Nature Nanotech­ nology 9 (4): 273–278. 12 Guo, J., Li, J., Liu, C. et al. (2020). High-performance silicon − graphene hybrid plasmonic waveguide photodetectors beyond 1.55 μm. Light: Science & Applications 9 (1): 1–11. 13 Shehzad, K., Shi, T., Qadir, A. et al. (2017). Designing an efficient multimode environmental sensor based on graphene–silicon heterojunction. Advanced Materials Technologies 2 (4): 1600262. 14 Bhopal, M.F., Lee, D.W., ur Rehman, A. et al. (2017). Past and future of graphene/ silicon heterojunction solar cells: a review. Journal of Materials Chemistry C 5 (41): 10701–10714. 15 Won, R. (2010). Graphene–silicon solar cells. Nature Photonics 4 (7): 411. 16 Chen, X., Shehzad, K., Gao, L. et al. (2020). Graphene hybrid structures for integrated and flexible optoelectronics. Advanced Materials 32 (27): 1902039. 17 Di Bartolomeo, A. (2016). Graphene Schottky diodes: an experimental review of the rectifying graphene/semiconductor heterojunction. Physics Reports 606: 1–58. 18 Mott, N.F. (1939). The theory of crystal rectifiers. Proceedings of the Royal Society of London Series A Mathematical and Physical Sciences 171 (944): 27–38. 19 Schottky, W. (1939). Zur Halbleitertheorie der Sperrschicht-und Spitzenglei­ chrichter. Zeitschrift für Physik 113 (5, 6): 367–414. 20 Sajjad, M., Yang, X., Altermatt, P. et al. (2019). Metal-induced gap states in passivating metal/silicon contacts. Applied Physics Letters 114 (7): 071601.

63

64

3  Physics of Graphene/Silicon Junctions

21 Mönch, W. (1999). Barrier heights of real Schottky contacts explained by metalinduced gap states and lateral inhomogeneities. Journal of Vacuum Science &; Technology B: Microelectronics and Nanometer Structures 17 (4): 1867. 22 Nishimura, T., Kita, K., and Toriumi, A. (2007). Evidence for strong Fermi-level pinning due to metal-induced gap states at metal/germanium interface. Applied Physics Letters 91 (12): 123123. 23 Tongay, S., Schumann, T., and Hebard, A.F. (2009). Graphite based Schottky diodes formed on Si, GaAs, and 4H-SiC substrates. Applied Physics Letters 95 (22): 222103. 24 Riazimehr, S., Belete, M., Kataria, S. et al. (2020). Capacitance–voltage (C–V ) characterization of graphene–silicon heterojunction photodiodes. Advanced Optical Materials 8 (13): 2000169. 25 Padovani, F.A. and Stratton, R. (1966). Field and thermionic-field emission in Schottky barriers. Solid-State Electronics 9 (7): 695–707. 26 Schroder, D.K. (2005). Semiconductor Material and Device Characterization. Wiley. 27 Taşçıoğlu, İ., Aydemir, U., and Altındal, Ş. (2010). The explanation of barrier height inhomogeneities in Au/n-Si Schottky barrier diodes with organic thin interfacial layer. Journal of Applied Physics 108 (6): 064506. 28 Yim, C., McEvoy, N., and Duesberg, G.S. (2013). Characterization of graphenesilicon Schottky barrier diodes using impedance spectroscopy. Applied Physics Letters 103 (19): 193106. 29 Dökme, İ., Altindal, Ş., and Bülbül, M.M. (2006). The barrier height inhomogeneity in Al/p-Si Schottky barrier diodes with native insulator layer. Applied Surface Science 252 (22): 7749–7754. 30 Parui, S., Ruiter, R., Zomer, P. et al. (2014). Temperature dependent transport characteristics of graphene/n-Si diodes. Journal of Applied Physics 116 (24): 244505. 31 Di Bartolomeo, A., Luongo, G., Iemmo, L. et al. (2018). Graphene–silicon Schottky diodes for photodetection. IEEE Transactions on Nanotechnology 17 (6): 1133–1137. 32 Trushin, M. (2018). Theory of thermionic emission from a two-dimensional conductor and its application to a graphene-semiconductor Schottky junction. Applied Physics Letters 112 (17): 171109. 33 Gong, C., Colombo, L., Wallace, R.M. et al. (2014). The unusual mechanism of partial Fermi level pinning at metal–MoS2 interfaces. Nano Letters 14 (4): 1714–1720. 34 Liu, Y., Stradins, P., and Wei, S.-H. (2016). van der Waals metal-semiconductor junction: weak Fermi level pinning enables effective tuning of Schottky barrier. Science Advances 2 (4): e1600069. 35 Farmanbar, M. and Brocks, G. (2015). Controlling the Schottky barrier at MoS2/ metal contacts by inserting a BN monolayer. Physical Review B 91 (16): 161304. 36 Farmanbar, M. and Brocks, G. (2016). First-principles study of van der Waals interactions and lattice mismatch at MoS2/metal interfaces. Physical Review B 93 (8): 085304. 37 McDonnell, S. et al. (2014). Defect-dominated doping and contact resistance in MoS2. ACS Nano 8 (3): 2880–2888. 38 Zhang, X., Liao, Q., Kang, Z. et al. (2021). Hidden vacancy benefit in monolayer 2D semiconductors. Advanced Materials 33 (7): 2007051.

65

4 Graphene/Silicon Junction for High-performance Photodetectors 4.1 ­Introduction Graphene (Gr)-based photodetection is increasingly becoming important for a variety of technologies because Gr absorbs light in a broad range of the light spectrum ranging from ultraviolet (UV) to infrared (IR) [1]. It also possesses many other unique properties such as high transparency [2], conductivity [3], mechanical and chemical stability [1], and compatibility with the standard CMOS fabrication process [4], making it a near ideal material for the imaging industry. Despite several unique advantages, Gr-based photodetectors exhibit low electrical response to the incident light due to the ultrashort lifetime of photogenerated carriers in Gr and relatively low light absorption (2.3% only) [1, 2]. To this end, significant efforts have been devoted to develop Gr-based photodetectors with high performance by integrating Gr with different material systems (plasmonic structures, quantum dots [QDs], two-dimensional [2D] materials, and traditional bulk semiconductors) and technologies (waveguides and cavities) [4–11]. Gr integrated with Si (Gr/Si Schottky junction) is perhaps the simplest of the Gr heterojunctions that can be formed for the improvement of the photodetection in Gr. Gr/Si Schottky devices show rectifying characteristics and operate in reverse bias conditions, which results in low dark current/noise. It is important to note that due to zero bandgap, Gr shows a large dark current, and hence, a low dark current heterojunction is extremely useful. In these diode structures, optical light is mostly absorbed by the semiconductor layer (Si, Ge, etc.). Gr absorbs broad wavelength range, but absorption is low, and works as a transparent electrode that collects photoexcited carriers. Furthermore, these photodetectors work in a specific wavelength range depending on the substrate bandgap. Here, we summarize some typical photodetectors from the UV to IR wavelength range and introduce hybrid Gr/Si photodetectors.

4.2 ­Ultraviolet Photodetectors UV detectors are widely used in chemical material analysis, medical, health care, space technology, etc. [12–14]. Traditionally, wide bandgap (WBG) semiconductors Graphene for Post-Moore Silicon Optoelectronics, First Edition. Yang Xu, Khurram Shehzad, Srikrishna Chanakya Bodepudi, Ali Imran, and Bin Yu. © 2023 WILEY-VCH GmbH. Published 2023 by WILEY-VCH GmbH.

66

4  Graphene/Silicon Junction for High-performance Photodetectors

are used for high-performance UV photodetection [15–17]; however, several issues, including but not limited to low crystal quality, lack of flexibility, and high cost, hinder their widespread applications. Using mature Si plane technology to manufacture UV detectors can greatly reduce the production cost and has the advantages of stable product quality and easy control. However, in the range of shortwavelength visible to UV light, Si has a high absorption coefficient of light and a strong reflection of incident light. The penetration depth of light in Si falls sharply with decreasing wavelength. For example, UV light with λ 0.14 A W−1), large on/off ratio (1.2 × 106) at incident light power of 1.0 mW cm−2 and λ = 365 nm, fast response with a fall time 100%, and high detectivity (1.6 × 1013 Jones in vacuum) in the 200–400-nm wavelength range. Here, Gr also acts as a transparent electrode. Traditionally, indium tin oxide (ITO) [16] and ultrathin metallic films are used as transparent electrodes. Apart from low cost, Gr offers another unique advantage over these electrodes. Hot carriers generated by UV light in Gr exhibit a relatively longer life time, when compared to traditional transparent electrodes. Such longlived hot carriers significantly enhance the photocurrent in Gr, resulting in higher IQE. As a result, a record responsivity was observed for these Gr/Si photodetectors, which authors claim is higher than the commercially available Si UV photodetectors. The environmental stability and responsivity of the Gr/Si UV photodetector were further improved with antireflection layer. This experimental work paves the way toward the next-generation UV photodetectors with characteristics of costeffectiveness, ultralow-power consumption, and high performance.

4.3 ­Visible to Near-infrared Photodetector The Si-based photodetector is the device with the longest development time and the most mature technology among all photodetectors, mainly because Si has the advantages of easy production, easy doping, high stability, and low cost. At present, Si-based photodetectors are also in a leading position, but the limitation of the Si bandgap also restricts the application of Si-based devices in the field of IR detection. Therefore, people are looking for devices based on new materials and new structures to break through the problems of traditional devices in terms of spectral response range, sensitivity, fill factor, noise, and cost, and broaden the application scenarios and scope of devices. QDs enhancement is the most commonly used method to broaden the detection wavelength to IR range and increase the photoresponse of Gr/Si photodetectors. In 2016, Yu et  al.  [22] fabricated Si-QD/Gr/Si photodetectors with improved device performance (Figure 4.3a). They observed that Si QDs on the top of Gr enhance the built-in potential across the Gr/Si Schottky junction. Figure 4.3b shows the broadband responsivity curves of the Gr/Si and Si-QD/Gr/Si photodetectors under reverse bias. Clearly, the integration of Gr with Si QDs improved the responsivity at almost all the studied wavelengths. However, the main increase in the responsivity happens at wavelengths shorter than 1000 nm. Same is true for incident photon

4.3  ­Visible to Near-infrared Photodetecto

Graphene Au SiO2 Si

V

Graphene/Si Si-QD/graphene/Si

0.3

60

0.2 0.1

20

100

IQE (%)

40

120 80 60 40 20 0

0

400

600

800

50 40 30 20 10

400

600

800

1000

Wavelength (nm)

0.0

400

600

1000

800

Wavelength (nm) Graphene/Si Si-QD/graphene/Si

0.06

Pulse

0.6

Photocurrent (μA)

60

0.8

Laser input (a.u.)

80

IPCE (%)

0.4

(b)

(a)

(c)

Graphene/Si Si-QD/graphene/Si Optically enhanced

0.5

Reflectance (%)

Responsivity (A W–1)

Si QDs

0.4

0.2

–8 V –4 V –2 V 0V

–8 V –4 V –2 V 0V

0.04

0.02

1000

Wavelength (nm)

400

600

800

Wavelength (nm)

0.0

1000

0.00 0

(d)

10

20

30

Time (ns)

40

0

10

20 30 40 0

10

20 30 40

Time (ns)

Figure 4.3  (a) Schematic of the Si-quantum dots (QDs)-integrated Gr/Si photodetector. (b) The responsivity as a function of the wavelength for Gr/Si and Si-QD/Gr/Si photodetectors under −1 V bias. The inset shows the optical reflectance of the Gr/Si and Si-QD/Gr/Si photodetectors. (c) IPCE and the IQE (inset) of both the photodetectors. (d) Photocurrent response of both the Gr/Si and Si-QD/Gr/Si photodetectors under a single laser pulse at various bias voltages. Source: Reproduced with permission from Yu et al. [22]; John Wiley & Sons.

conversion efficiency (IPCE), as shown in Figure  4.3c. Also the wavelength (λp) where max. R was observed shifts from 860 to 877 nm, while the max. R increases by 18% (from 0.421 to 0.495 A W−1). Figure 4.3d depicts the time-dependent photocurrent of both photodetectors with simple Gr/Si structure and Si-QD-integrated Gr/Si, under a single laser pulse. An important observation is that the photocurrent fall time decreases with the increase in reverse bias. It can be explained by considering that at larger reverse bias a stronger built-in field is formed at the Gr/Si Schottky junction, which makes photogenerated carriers drift faster. Also, given the proportional relationship between the width of the depletion region and the reverse bias voltage, the junction capacitance decreases with the increase in the reverse, which also contributes to fast device response. In summary, a novel 0D (Si-QDs)/2D (Gr)/3D (bulk Si) platform was developed that involved the coupling of SiQDs with Gr. Such an arrangement resulted in photodetectors with excellent responsivity, response speed, and specific detectivity. Photoresponse of these Si-QD/Gr/Si photodetectors can be further tuned by changing the coupling between Si QDs and Gr using Si QDs of different sizes [23] and/or by varying the number of Gr layers [24]. These results not only advance the development of high-performance Gr/Si Schottky photodetectors but also have important implications for the fabrication of other novel Gr-based hybrid structures and

69

4  Graphene/Silicon Junction for High-performance Photodetectors

devices  [25, 26]. When the SiQDs are doped during the preparation process, the light absorption of SiQDs will extend from the conventional UV and VIS light regions to the NIR region. The absorption spectrum of p-type hyper-boron (B)-doped SiQDs (B-SiQDs) shows that it still has a certain absorption in the NIR band of 1100–2000 nm, as shown in Figure 4.4a. Therefore, applying B-SiQDs instead of the abovementioned undoped SiQDs to Gr/Si devices can make up for the weakness of low photoresponse for SiQDs and Gr/Si devices in the long wavelength and thereby preparation of a photodetector with a wide spectrum and high-speed response. To achieve efficient separation of photocarriers, Du et al. [27] prepared a cascade structure for high responsivity at fast speed. High responsivity was the result of the gain from charge traps in B-SiQDs, and the fast operation was the result of inherently fast Gr/Si Schottky diodes (Figure  4.4b,c). The transition work function between B-SiQDs/Si and Gr is estimated from the difference in built-in potentials. The depletion region formed in the QDs film extends to the Si surface. The majority

40 nm

Au/Cr

0.12 20

SiO2

10

nm

0.08

0

–10 –20

0

2

4

μm

6

8

0.00

400

800

1200

Graphene Vb

(b)

–30 nm

0.04

B-SiQDs

n+ Si (b) Electrode

1600

SiO2

2000

Wavelength (nm)

Graphene B-SiQDs n+ Si

200 nm

(a)

(c) Wavelength (μm) 0.9

1.2

Wavelength (μm)

1.5

0.9

1.8 0.6

0.4

40 20

0.2

1.8

16

R (A W–1) D (×1011 Jones)

60

1.5

2.0

12

1.5

8

1.0

*

80

1.2

4

0.5

0 1.4

(d)

1.2

1.0

0.8

Photon energy (eV)

1.4

(e)

1.2

1.0

NEP (×10–13 W Hz–0.5)

Absorption (a.u.)

0.16

IPCE (%)

70

0.8

Photon energy (eV)

Figure 4.4  (a) Optical absorption of B-SiQDs at different optical wavelengths. Inset shows the AFM height and line-scan profiles of B-SiQDs. (b) Schematic of the Gr/B-SiQDs/Si device structure and (c) cross-sectional SEM image of the device. (d) The IPCE and responsivity (R). (e) The D* and NEP as a function of the incident laser photon energy and wavelength. Source: Reproduced with permission from Du et al. [27]; IEEE.

4.4  ­Broadband Photodetector

carrier (hole) injection is suppressed from the Si by the presence of large potential barrier in the valence band, resulting in I–V characteristics with strong rectifying characteristics. To minimize the recombination of carriers during transport through the B-SiQDs layer, the thickness of QD films plays a crucial role. Here, the carrier lifetime must exceed the carrier transit time in the QD film. Trapping one type of charge carriers can also increase the lifetime of the remaining carrier type. In this case, B-SiQDs acts like hole-trapping layer and thus helps to enhance the quantum efficiency (QE) of the photodetector. Figure 4.4d exhibits the IPCE and R as functions of the incident photon energy and wavelength; 90% and 8% IPCE were observed at 1.4 and 0.66 eV photon energy, respectively. Similarly, 0.6 A W−1 and 0.12 A W−1 responsivity values were observed at 900 and 1870 nm wavelength, respectively. Figure.4.4e shows the excellent D* and NEP values at various incident photon energies and wavelengths, indicating reliable performance of the device in the NIR region.

4.4 ­Broadband Photodetectors To obtain the broadband response, photodetector devices are coupled/modified with tricchroic prisms, optical filters, and charge coupled devices, which complicates the system and also increases the cost significantly [28]. Since Gr is inherently a broadband absorber (its absorption spectra covers X-rays to terahertz range), it is considered as a promising material for ultrabroadband photodetector ion  [29]. However, as mentioned earlier, the photoresponse of the pristine Gr photodetectors is not good enough for practical applications, as Gr, due to its atomic thickness can only absorb 2.3% of incident light, despite its strong light–matter interaction. Also, the life time of photogenerated carriers is extremely short, which makes their extraction relatively difficult [1, 2]. Furthermore, its zero bandgap results in substantial dark currents, which is unsuitable for highly sensitive photodetectors [30, 31]. To overcome these challenges, significant efforts have been made to couple Gr with plasmonic nanostructures [5], microcavities [8], Si waveguide [32], and transitionmetal dichalcogenide stacks [33] to improve the responsivities. Despite the improved responsivity, broadband response is still a challenge, mainly because it is a material property, and light absorption relies on the materials, structures, or resonant frequencies. Therefore, enhancing the photoresponse without sacrificing the responsivity or response in wide range of wavelengths remains a challenge. Fluorine functionalization of Gr is demonstrated to change the Gr from conductive to insulator with change in bandgap from 0 to 4.1 eV [34]. Fluorination opens the bandgap in the Gr by removal of bands near the Fermi level of pristine Gr. Carbon atoms where the fluorination takes place change to sp3 hybridization, while the remaining carbon atoms retain their sp2 hybridization states. With the degree of fluorination, the ratio of sp3 and sp2 hybridized carbon will change. This atomic and electronic feature of fluoro graphene (FGr) with variable sp3/sp2 fractions offers a novel tool for enhancing the performance of the photodetectors. Du et al. [35] fabricated an FGr/Gr photodetector with an FGr layer on the top of a

71

4  Graphene/Silicon Junction for High-performance Photodetectors

VDS VGS

Responsivity (A W–1)

105

SiO2

FG

FG/Gr 255 nm 980 nm

4

10

101 100

(b)

532 nm 1870 nm 4290 nm

1010 D* (Jones)

3

375 nm 1319 nm 3459 nm

2 1

109 Gr/Gr FG/Gr

108 107 106

0.10 0.05

0 (c)

101 102 103 Irradiance (µW cm–2)

1011

FG/Gr 255 nm 980 nm 2776 nm

532 nm 1870 nm 3459 nm 4290 nm

102

Gr

(a)

375 nm 1319 nm 2776 nm

103

Au/Cr Si

Photocurrent (µA)

72

2

4 Time (s)

6

0

8 (d)

1

2

3

4

Wavelength (µm)

Figure 4.5  (a) Schematic of FGr/Gr photodetector structure. (b) Responsivity of FGr/Gr photodetectors measured against the laser irradiance. (c) The speed of FGr/Gr photodetector measured at different laser wavelengths. (d) The calculated D* of Gr/Gr and FGr/Gr photodetectors. Source: Reproduced with permission from Du et al. [35]; John Wiley and Sons.

Gr layer, as shown in Figure 4.5a. In contrast to other Gr-based photodetectors, the van der Waals heterostructure consisting of FGr/Gr improves the responsivity of the devices without compromising the broadband photoresponse of Gr. The functionalization of carbon with fluorine results in fluorinated carbon atoms which are sp3-hybridized. As, generally, fluorination is not a very uniform process, especially if Gr is not perfectly single crystalline, random domains of small size consist of sp2 and sp3. These mixture of domains act as discrete quantum confined states and trapping centers for photoexcited charge carriers. Tunable photoresponse has been achieved by modulating the dihedral angles of sp3 sites, the size/shape of the sp3 domains, and the ratio of sp3/sp2-bonded carbon atoms. The photocurrent due to the different photoexcited charge carriers trapping time in sp2 and sp3 nanodomains can be easily distinguished. They demonstrate a prototype device, yielding exceptional photo­responsivities over an ultrabroad spectral range from 255 to 4290 nm. As shown in Figure 4.5b, FGr/Gr devices show a strong photoresponse in a wide wavelength range from 255 nm to 4.3 μm, clearly manifesting that these devices are suitable for broadband detection. In comparison with Gr/Gr devices, the responsivity of the FGr/Gr photodetector is more than three orders of magnitude higher.

4.4  ­Broadband Photodetector

Such strong and broadband response in the FGr/Gr devices is attributed to the photogate effects in both sp2 and sp3 domains. The total responsivity R is a combination of two responsivities Rsp2 and Rsp3. High responsivity and low noise current, as observed for FGr/Gr devices, promise potential for weak light detection, which is measured by calculating the NEP values. The specific detectivity value D* also reaches 4 × 1011 Jones (Figure  4.5d), which is almost three orders of magnitude larger than that of the Gr/Gr device. Strong response of the FGr/Gr devices is due to charge trapping in sp2 and the sp3 domains, which leads to photogating. However, for such photodetectors, there is always a trade-off between responsivity and speed. The same trapping mechanism, which leads to high response due to trapping of one type of carrier for a long time, will lead to slow response, as carrier trapping time is inversely proportional to speed but directly proportional to responsivity [36]. The transient photocurrent response to ON/OFF illumination is displayed in Figure 4.5c. The FGr/Gr device shows stable and moderate switching and reproducibility in UV to mid-IR wavelength region. When the laser is turned on, the photocurrent reaches its maximum value within 80 ms and then decays in 200 ms. To improve the operation speed while keeping the responsivity high is a general problem in photodetection, and we always have to compromise one for the other, depending on the final application. Currently, improving the operating speeds of the device by minimizing the trapped lifetime of the electrons while maintaining relatively high responsivity is under investigation. To overcome the limited light absorption and monotomic thick scattering distances of single-layer graphene (SLG), Gao and coworkers [37] assembled graphene oxide (GO) into a wafer-scale, macroscopic assembled graphene (MAG) nanofilm with high crystallinity to improve light absorption (~40%) and enhance the photogenerated carrier relaxation time ~20 ps. The free-standing MAG can be integrated with Si substrates and form Schottky diodes without any polymer or metal contamination, as shown in Figure 4.6a,b. They improved the performance of mid-IR detectors not by designing a new structure of Schottky junctions but by designing new materials. The MAG/Si Schottky photodetectors exhibit photoresponse in a broad wavelength range (1.5–4.0 μm, Figure 4.6c). From the temperature-dependent I–V characterization, the MAG/Si steady Schottky barrier height was extracted as 0.3 eV, which corresponds to a wavelength of 2.1 μm. Hence, the photoresponse in the wavelength from 1 to 2.1 μm (regime I: direct injection) is attributed to direct photoexcited electron injection over the Schottky barrier through the internal photoemission effect (IPE), while the responsivity in the wavelength from 2.2 to 4.2 μm (regime II: hot-electron injection) is attributed to the hot-electron injection over Schottky barrier through photothermionic effect (PTE). The MAG/Si heterojunction shows high detectivity from 1.6 × 1011 (1.5 μm) to 1.9 × 109 (4.0 μm) Jones at room temperature (Figure 4.6d). Such a platform cannot be realized by the existing technologies due to its stringent requirements for defect control and thickness uniformity. This work opens the avenue for large-scale integration of Gr for high-performance optoelectronic devices, offering a new strategy for room temperature ultrafast mid-IR photodetection.

73

4  Graphene/Silicon Junction for High-performance Photodetectors MIR NIR

MAG MAG SiO2

Vb

Si Iph

A

Si (b) 102

10–3 10–4 10–5 10–6

λ (μm) 1.34 1.55 1.87 2

10–1

(c)

Responsivity (mA W–1)

10–2

2.2 2.5 3 3.5 4 4.2

100

101

(d)

Irradiance (mW mm–2)

10

I

100

10–9

II

10–10 10–1 10–11

10–2 10–3

102

10–8

1

NEP (W Hz–0.5)

(a)

Responsivity (A W–1)

74

10–12 1

3 2 Wavelength (μm)

4

Figure 4.6  (a) A schematic of the MAG/Si device showing the charge transport and current flow. (b) HR-TEM image of the MAG/Si showing an atomic-scale contacting interface. (c) The responsivity of MAG/Si as a function of laser irradiance at different wavelengths. (d) The responsivity and NEP of MAG/Si as a function of laser wavelength. MIR, mid infrared. Source: Reprinted with permission from Li et al. [37]/ John Wiley & Sons / CC BY 4.0.

The good responsivity of the MAG/Si devices results from the following factors: strong light absorption (~40%). The more IR photons absorbed by MAG excite more electron–hole pairs, laying the foundation for a higher responsivity. The zero bandgap enables broadband light absorption of MAG, from X-ray to THz, and long carrier relaxation time (~23 ps). The AB-dominated stacking structure of MAG weakens the electron–phonon coupling, prolonging the carrier relaxation time from 1 ps (single-layer Gr) to ~23 ps. The long carrier relaxation time enhances the electron– electron interaction, thereby prolonging the out-of-plane hot-electron diffusion length, increasing the hot-electron density overcoming the SBH charge transfer efficiency. The hot-electron scattering effect is quantified by calculating the hot-carrier multiplication. When the energy of photoexcited electrons is lower than SBH, they cannot overcome the SBH immediately but thermalize within the electronic system into a Fermi–Dirac (FD) distribution in MAG. When a part of the FD distribution tail extends higher than SBH, some of the hot carriers contribute to the device responsivity and thus greatly extend the detectable wavelength. The atomic-scale interfacial contact with Si helps to suppress the carrier number fluctuation from the surface effect (traps and disorders) and the recombination of carriers at the interface. Furthermore, the proper thickness balances the trade-off between the .

4.5  ­Hybrid Gr/Si Photodetector

out-of-plane charge transfer and the recombination in MAG, optimizing the QE. The thickness enables a large out-of-pane hot-electron diffusion length that redirects some of the backscattered hot electrons into the silicon, mainly through phonon or interface wall scattering. In short, their results are important for two reasons: First, they demonstrate a new methodology for low cost and scalable integration of Gr for CMOS-compatible production of broadband photodetectors that can operate at room temperature. Secondly they explore exciting new physics related to hot-carrier dynamics in multilayer 2D systems. The MAG/Si work represents the conceptionally new advancement in this area and will have a profound impact on multifunctional nanomaterials design, synthesis, engineering, and applications. Especially, the merits of MAG in photodetection are not fully explored. Much work is needed to further improve the detector’s performance, such as matching suitable semiconductor materials and device structures, making it work efficiently as a commercial device. Besides, the first reported free-standing MAG and the corresponding scalable yet low-cost preparation method show great potential in various carbon-dominated electronic devices, such as detector of high-energy radiation (X-ray) and particle (neutron), THz metamaterial, laser-driven ion acceleration, thermoacoustic and acoustic detection devices, magnetoresistance, and magneto-optical sensors. Their work could be an excellent opportunity to set up leading guidance in the area of photoelectric response material and other 2D bulk nanomaterials. In conclusion, Gr/Si heterojunctions not only promise exciting new applications but also act as a platform to explore/investigate the new physics of 2D–3D and 2D–2D interfaces.

4.5 ­Hybrid Gr/Si Photodetectors For realizing high-performance photodetector based on Gr/Si heterojunction, various methods have been developed to form hybrid Gr/Si heterostructures, such as integrating highly photosensitive materials, interface modification using wide bandgap materials, and adopting light-trapping effect  [38–41]. The hybrid Gr/Si photodetectors not only offer effective separation and transport of photocarriers but also could take advantage of the synergistic photoelectric effects between various materials. In this section, we review the efforts of fabricating high-performance hybrid Gr/Si PDs. Yu and coworkers fabricated a high-performance photodetector with high sensitivity to UV light by assembling zinc oxide (ZnO) nanorods arrays with Gr/Si heterostructures (Figure 4.7a) [42]. The device effectively restricts the incident UV light to the surface area and achieves high energy efficiency. Further performance characterization demonstrated that the device exhibited good rectification behavior at different temperatures, with a turn-on voltage of about 0.5–1 V and a breakdown voltage of about −4 V (Figure  4.7b). Based on the thermionic emission theory, the ideality factor (η) and the barrier height (ΦB) of the Schottky junction are calculated and summarized, as shown in Figure 4.7c. From other types of Schottky junctions made of element semiconductors, compound semiconductors, and metals, the ΦB of the Gr/ZnO nanorod arrays/Si Schottky junction

75

4  Graphene/Silicon Junction for High-performance Photodetectors 100

Ag

80

tor

Current (µA)

ula

Ins

ZnO NR arrays

n+

Si

60

300 K 260 K 220 K 180 K 140 K 100 K 80 K

PMMA supported MLG ZnO nanorod array 2 µm

40 20 0 –6

(a)

(c)

150

200

250

Temperature (K)

Photocurrent (a.u.)

1.8

Ideality factor, n

2.1

0.4

100

0

–2

2

4

6

Voltage (V)

1.0

n ΦB

0.6

0.2 50

–4

(b)

0.8

ΦB (eV)

76

0.8

τf = 3.6 ms

0.6 0.4

τr = 0.7 ms

0.2 0.0 0

300

(d)

2

4

10

15

20

Time (ms)

Figure 4.7  (a) Schematic diagram of Gr/ZnO nanorod arrays/Si photodetector. (b) The I–V characteristics of the photodetector at different temperatures in between 80 and 300 K. The inset shows the FESEM cross-sectional image of the photodetector. (c) The change in barrier height (ΦB) and ideal coefficient (η) at different temperatures. (d) The device’s response speed at 50 Hz light illumination. PMMA, poly(methyl methacrylate). Source: Reproduced with permission from Nie et al. [42]; John Wiley & Sons.

increases with increasing temperature, while η increases with decreasing temperature, often referred to as the “T0 anomaly.” The device exhibited fast response times; the rise (τr) and fall (τf ) times could reach 0.7 and 3.6 ms, respectively (Figure 4.7d). In addition, it is worth mentioning that the photodetector can monitor pulsed light at frequencies up to 2250 Hz. This work demonstrated that integrated photosensitive materials with Gr/Si could fabricate a high-performance photodetector. In addition to inserting photosensitive materials between the Gr and Si, developing a coupling effect between materials and Gr is also an important pathway to enhance the responsiveness of the device. Zhou and coworkers fabricated an enhanced Gr/Si Schottky junction photodetector by utilizing the coupling effect between carbon QDs and Gr [43]. The broadband spectral response flatly covered the visible spectrum when the carbon QDS was coupled to the Gr/Si Schottky junction. The optical response is increased by 4.3% with a response time within 2 μs, which is better than most detectors based on Gr-Schottky or QDs structures. Yang and coworkers reported an improved emulsion method for synthesizing high-quality 2D Pt NPs [44]. Then, they put it on the surface of Gr–Si Pd to generate strong plasmon resonant effects. In addition, the high power function of Pt makes Gr doped, which enhances the electric field of the Gr/Si heterojunction. The detector displayed excellent photoelectric conversion properties by combining the plasmon resonant

4.5  ­Hybrid Gr/Si Photodetector

Graphene

h-BN

WG

Ti/Au

Ti/Au

h-BN

SiO2

SiO2

(a)

Ev

(b)

10

0.55

0

0.50

4

0.45

3

–10 –20 –30 0.0

0.2 0.4 Voltage (V)

2 0.40

w/o h-BN with h-BN

0.35

0.6 (d)

PCE (%)

5 VOC (V)

Current density (mA cm–2)

n-Si ΔEV 4.05 eV

1 w/o h-BN

with h-BN

w/o h-BN

with h-BN

0

Figure 4.8  (a) Schematic three-dimensional diagram of Gr/h-BN/Si photodetector structure. (b) The band alignment characteristics of the Gr/h-BN/Si heterojunction. (c, d) Comparison of the I–V and PCE characteristics of the device with/without h-BN. Source: Reproduced with permission from Meng et al. [47]; Copyright© 2016, Elsevier.

effect and physical doping effect, in which the responsivity, response speed, and NEP could reach 26 A W−1, 78 ns, and 4.2 pW Hz−0.5, respectively. Comparatively, lower Schottky barrier height between the Gr/Si results in a large leakage current and thus enhances the recombination of interface carriers. Inserting an ultrathin insulating barrier between Gr and Si layers can alleviate this interface recombination issue, where metal/insulator/semiconductor (MIS) heterostructure has been widely used for this purpose  [45, 46]. Adjusting the band alignment through interface band engineering and reducing interface defects through interface passivation of saturated dangling bonds could effectively improve the device’s barrier height and recombination rate. Yan and coworkers designed a Gr/hexagonal boron nitride (h-BN)/Si structure to enhance the photoelectric performance of the device, the schematic diagram of which is shown in Figure 4.8a [47]. Here, h-BN could be an effective electron-blocking/hole-transport layer, suppressing the interface recombination and significantly increasing the open-circuit voltage. The energy band diagram of the Gr/Si Schottky junction with h-BN is shown in Figure 4.8b. There is almost no lattice mismatch between Gr and h-BN, increasing the conductivity of the Gr on the h-BN. In addition, utilizing the direct-growing Gr/h-BN heterostructure could eliminate the interface defects and contamination generated during layer-by-layer transfer. The device displayed outstanding photovoltaic properties, and power-conversion efficiency could reach 10.93% (Figure  4.8c,d).

77

78

4  Graphene/Silicon Junction for High-performance Photodetectors

Their results demonstrated that h-BN could be an effective interlayer to improve the performance of the photodetector. Aside from the h-BN, other materials, including Al2O3, AlN, GeOx, MoO3, ZnO, MgO, and Gd3Fe5O12, have also been used as the passivation insulation layer to improve the device performance [47–52]. Some organic compounds could act as an electron blocking layer inserted at the interface between Gr and Si. For example, Jie’s group reported a high power-conversion efficiency Gr/Si Schottky junction device introducing a P3HT interlayer. Similar to the role played by h-BN as described above, P3HT could effectively prevent electron transfer from Si to Gr [53]. Further photoelectric performance analysis showed that the power-conversion efficiency of the device reaches 10.56%. These works indicate that band engineering at the Gr/Si junction interface is an effective method for improving the photoelectric response of the device. In order to effectively regulate the work function and electrical conductivity of Gr, chemical doping has recently been developed. Veneri et al. modified Gr with HNO3 molecular, and the conductivity enhanced by about 63% compared with pristine undoped Gr. The schematic diagram of the device structure is shown in Figure 4.8a. The Raman spectra of Gr and HNO3-doped Gr are depicted in Figure 4.8b [54]. The G-band peak of the doped Gr shows a blue shift due to the conventional fluctuation caused by hole doping in Gr. The PCE of the HNO3-doped Gr/Si Schottky junction device increased by about 5%, and it could maintain performance within two weeks. As plotted in Figure 4.8c, the effect of aging and doping on J–V curves under light irradiation is shown. The HNO3-doped Gr/Si junction device exhibited higher PCE and air stability. Similarly, Wu and coworkers reported a high-efficiency HNO3doped Gr/Au NPs/Si hybrid device [55]. Combining the plasmon resonate effect and HNO3 doping impact, the built-in electric field was enhanced and further resulted in higher open-circuit voltage and short-circuit current. The PCE of the Gr/Si junction is increased by 2.6 times, with a maximum value of 10.20% (Figure 4.9). Considering the high-influence affection of HNO3 doping on Gr/Si heterojunction performance, Zhang’s group fabricated high-response Gr/Si nanoarray hybrid photodetector enhanced by HNO3 doping. Prepared by the light-trapping effect, the SiNH array-based device has a larger effective junction region with the same light absorption ability, superior to the Gr/planer Si heterojunction device  [56]. With appropriate surface passivation/modification of Si nanoarrays and careful control of the number of Gr layers and doping times, the maximum PCE of 10.30% was achieved. The device also showed good stability in the air environment, remaining stably photoresponsive after a week. Besides, bis(trifluoro methane sulfonyl) amides (TFSAs), SOCl2, and AuCl3 have also been demonstrated to dope the Gr effectively [57–59]. Moreover, a novel hybrid Gr/Si junction photodetector that operates in a photoconductive mode was proposed by Xu and coworkers, and the specific device structure is shown in Figure 4.10a [60]. Due to the existence of the Gr/Si Schottky junction, lightinduced electron/hole pairs could be separated rapidly. Holes were injected into Gr from Si, and electrons were swept into Si. The I–V curves of the device are plotted in Figure 4.10b. At 5 V, the photocurrent increased from 1.3 to 3.5 mA, while the incident power was increased from 0.112 to 28 μW, and the responsivity was calculated to be

4.5  ­Hybrid Gr/Si Photodetector PG DG

Au/Ti

Graphene

Intensity (a.u.)

G

SiO2 n-Si

(a)

Ti/Pd/Ag

D

2000

1000

Current density (A cm–2)

0.02

HNO3 solution

3000

Raman shift (cm–1)

(b)

HNO3 vapor

2D

PD: PCE 1when recombination mechanisms other than the recombination of minority carriers in the quasineutral region occur. Reverse saturation current (J0) can be described as Eq. (5.2), J 0

A*T 2 exp

q SBH kBT

(5.2)

The values of the effective Richardson constant (A*) for n- and p-type Si are 252 and 32 A cm−2  K−2, respectively, and ϕSBH is the height of the Schottky barrier. According to the Schottky–Mott rule, Schottky barrier with the height ϕSBH = ϕmetal – χSi. As clear from the expression, the work function of metal (ϕmetal) should be greater than the electron affinity of Si (χSi). Schottky barrier helps block the recombination of electrons and holes due. Under the light irradiation, the electron–hole pairs are generated due to the absorption of photons by both the Gr and the bulk Si underneath, separated by the built-in potential (V0) and extracted for collection at electrodes. The main parameters commonly used for the performance evaluation of a solar cell are PCE, short-circuit current density (Jsc), open-circuit voltage (Voc), fill factor (FF), external quantum efficiency (EQE), and/or internal quantum efficiency (IQE). When a solar cell is illuminated, electrons and holes are separated and accumulate in different parts of the Gr/Si junction, which gives rise to a photovoltage (V), which is maximum at an open circuit (Voc). When the solar cell is at zero voltage (shortcircuited), it generates a maximum current called Isc. The PCE of the Gr/Si solar cells defines the efficiency of converting the light into electricity and can be expressed as Eq. (5.3), P CE

Voc Isc FF Pin

(5.3)

where Pin is the incident light power. FF can be expressed as in Eq. (5.4), F F

Pmax Voc Isc

(5.4)

EQE is the number of charge carriers generated per number of incident photons with a given energy, which can be calculated as Eq. (5.5),

87

88

5  Graphene/Silicon Solar Energy Harvesting Devices

I ph h

E QE

qPin



(5.5)

where Iph is the photocurrent, and other constants have their usual meanings. IQE is the number of charge carriers generated per photon absorbed. IQE can be calculated from EQE by deducting the photons lost to the reflection and transmission processes by employing Eq. (5.6), I QE

EQE 1 R T

(5.6)

where T is the transmission, and R is the reflection of solar cells. These parameters are tools to compare the performance of Gr/Si solar cells.

5.3  ­Theoretical Efficiency Limits of Graphene Silicon Solar Cells The theoretical efficiency limit of Gr/Si solar cells can be determined by the Shockley–Queisser model, in which the bandgap of a given semiconductor (Si in this case) determines the maximum PCE. By further taking into consideration Auger recombination, efficiency limits have been reassessed, and the maximum theoretical predicted values of Voc, Jsc, FF, and PCE are reported as 0.761 V, 43.31 mA cm−2, 89.26%, and 29.43%, respectively [12–14]. In these models, parasitic resistances such as series resistance (RS) and shunt resistance (RSH) are ignored, which dissipate the converted electrical energy. The current–voltage characteristics of a solar cell without parasitic resistances and with both (series resistance and shunt resistance) are given in Eqs. (5.7) and (5.8), respectively [15, 16], J

J ph

J 0 exp

J

J ph

J 0 exp

qV kBT q V

1 IRS

kBT

J ph V

J 0 exp IRS RSH

qV kBT

(5.7)

(5.8)

where J, J0, and Jph are the current density, the reverse saturation current density, and photocurrent density, respectively, q is (1.6 × 10−19 C), V is the applied voltage, kB is the Boltzmann constant (1.38 × 10−23 J K−1), T is the absolute temperature, Rsheet is sheet resistance, and RSH is the shunt resistance, and η is the ideality factor. Figure 5.2a,b shows the J–V curves of Gr/Si solar cells with different RS and RSH, respectively. The FF of the solar cells has a strong correlation with the Rsheet of Gr, as low Rsheet leads to higher FF. Hence, the improved Rsheet of Gr through high-quality growth, transfer, and controlled doping can significantly improve solar cell efficiency. The effect of shunt resistance only becomes important when the resistance drops below 10 kΩ and otherwise has no or negligible effect [17].

Current density (mA cm–2)

5.4  ­Optimization of Graphene/Silicon Solar Cell

40 Ideal solar cell RS = 10 Ω RS = 30 Ω RS = 50 Ω

30 20 10

Rs (Ω)

0

10

30

50

FF (%)

89.26

79.75

66.66

57.03

0 0.0

0.1

0.2

Current density (mA cm–2)

(a)

0.3 0.4 0.5 Voltage (V)

0.6

0.7

0.8

0.7

0.8

40 Ideal solar cell RSH = 1E6 Ω RSH = 1E4 Ω RSH = 1E3 Ω

30 20 10

RSH (Ω)

Inf

1E6

1E4

1E3

FF (%)

89.26

89.16

88.23

74.54

0 0.0 (b)

0.1

0.2

0.3

0.4

0.5

0.6

Voltage (V)

Figure 5.2  Voltage–current characteristics of an ideal and real graphene/Si solar cell with different (a) series resistance and (b) shunt resistance. The FF values of curve with each resistance are also provided in the inset tables. Source: Reproduced with permission from Huang et al. [10]; John Wiley and Sons.

5.4  ­Optimization of Graphene/Silicon Solar Cells 5.4.1  Doping of Graphene Gr acts as both a transparent conductor and the junction layer. As a transparent conductor, Gr should have low resistance and high transparency. As shown in Figure 5.2a, the change in Gr sheet resistance (Rsheet) significantly increases the efficiency of the Gr/Si solar cells. Resistance of Gr can be decreased by increasing the layer number or doping. However, the process is not that straightforward as, generally, there is a trade-off between transparency and Rsheet. With every additional layer of Gr, transparency will decrease by 2.2–2.3%. Hence, transparency and Rsheet need to be optimized for better efficiency Gr/Si solar cells. For example, by stacking the Gr layers, the Rsheet of bi- and tri-layer Gr was reduced to ≈540 and ≈350 Ω sq−1 from

89

90

5  Graphene/Silicon Solar Energy Harvesting Devices

≈980 Ω sq−1 for single-layer Gr transferred onto the glass. However, the transparency of the tri-layer Gr reduced to 92.9% from 97.6% for monolayer. Similarly, 90.8% transparency for four-layered and 83% for eight-layered Gr (180 Ω sq−1 Rsheet) have been reported. These results clearly show a trade-off between transparency and Rsheet. Despite some positive results, stacking the monolayers of Gr into MLG is not a trivial task. Layer-by-layer stacking is a time-consuming process, cannot be scaled up. Moreover, it also introduces lots of polymeric and environmental impurities during the transfer/stacking process. Apart from increasing the number of layers, doping has also been demonstrated as an effective strategy to reduce the Rsheet of Gr. Various doping strategies such as heteroatom doping, chemical modification, and field and light-induced doping have been employed successfully. In heteroatom doping, atoms, such as oxygen (O), sulfur (S), phosphorous (P), boron (B), and nitrogen (N), either substitute or bond with carbon atoms in the lattice. However, chemical doping is preferred over heteroatom doping for the application in Gr/Si solar cells. Among various dopants, HNO3, SOCl2, and AuCl3 are the most effective. In a study Cui et al. employed different dopants (HNO3, HCl, H2O2, and SOCl2) and found the best improvement for SOCl2. The Rsheet for SOCl2 decreased from 58.5 to 46.5 Ω sq−1, leading to 230% enhancement in PCE over the control sample (2.45%) [18]. In another study of SOCl2-modified Gr, 45% reduction in the Rsheet (from 735 to 405 Ω sq−1) was achieved. HNO3 is the most reported dopant in the literature. Feng et al. [19, 20] reported that doping with HNO3 vapors reduced the RS value of the devices from 6.11 to 4.07 Ω cm2. As a result, the PCE of the solar cell increased from 2.9% to 4.35%, with other performance parameters including Jsc, FF, and Voc also increasing to 17.22 mA cm−2, 51%, and 495 mV, respectively. Though the chemical doping-induced Gr resistance and solar cell efficiency have been reported, the stability of the covalently functionalized Gr in ambient environment is a serious issue, and any deterioration of functionalization could reduce the photovoltaic efficiency up to more than 50%. Doping with gold chloride (AuCl3) is more environmentally stable than the other reported dopants. Transfer of charge from Gr reduced Au3+ to Au, thereby p-doping the Gr. Rsheet improvements from 448 to 11 Ω sq−1 (efficiency increased from 1.85% to 3%) and from 400 to 120 Ω sq−1 (efficiency of 12.4%) for AuCl3-doped MLG have been reported [21, 22]. Besides, organic compounds such as vanadyl-phthalocyanine, 2,3,5,6tetrafluoro-7,7,8,8-tetracyanoquinodi-methane, and trifluoromethanesulfonyl amide (TFSA) can also be used to dope Gr [23]. TFSA-functionalized monolayer Gr has been shown to experience a decrease in the resistance from 1200 to 300 Ω cm2 without any significant change in the transmittance. In another study, after doping with TFSA, due to improvement in resistance, the PCE of the Gr/Si devices increased to 8.6% from 1.9%, the Jsc increased from 14.2 to 25.3 mA cm−2, and Voc increased from 0.43 to 0.54 V, for doped and undoped Gr, respectively [24]. Most of these dopants p-dope the Gr, which decreases the Rsheet of Gr due to a downshift in the Gr Fermi level or an increase in the work function. Since Gr also acts as a junction layer, apart from the improvement in Gr Rsheet, the increases in work function also have other repercussions at the interface. An increase in the Gr work function leads to

5.4  ­Optimization of Graphene/Silicon Solar Cell

increased SBH; hence, a stronger built-in electric field leads to more efficient carrier separation. Figure 5.3a shows that Voc and barrier height (ϕb) increased after doping due to the increased built-in potential V0. After chemical doping, the work function of the MLG (ϕG) increases ϕb, ensuing greater charge collection across the interface. So, doping-induced improvement in the Gr/Si solar cells combines Gr Rsheet and increased Schottky barrier height. Since chemical doping is not stable in ambient environment, more stable doping techniques are highly desirable for practical purposes. Researchers have recently started exploring other doping strategies such as physical and light-induced doping. As an alternative to chemical dopants, metals with Fermi levels lower than Gr, such as gold (Au), can effectively p-dope the Gr. The work function of the metals determines the doping capacity of their NPs. Metals with higher work functions will dope the Gr more effectively. Upon deposition of Au NPs, a blueshift in the 2D Raman peak indicates the successful p-doping (Figure  5.3b)  [26]. By carefully choosing the size of Au NPs, the efficiency of the cells was improved by at least three times to 7.34%. The same group reported Au and HNO3 co-doping and improved the Vacuum level

Vacuum level

X ΦG

e(V0–Vbias) Φb eVbias

e(V0D–VbiasD)

ΦGD

Φn-Si EC EF Doping

Φb

Eg

hv

hv

Ev

MLG

n-Si

EC Eg

eVbiasD

EF Ev

MLG –

(a)

Φn-Si

D

n-Si

+

e

h 6

Au NPs/FLG

2600 2700 2800 Raman shift (cm–1)

FLG Au NPs/MLG MLG

1200 1400 1600 1800 2500 2750 3000 3250 Raman shift (cm–1)

(b)

0 –2 –4 –6 –8

(c)

ΦB = 0.72 V

W/O Au NPs With Au NPs dV/d(InJ)

2 In J

Intensity (a.u.)

4

0.16

RS = 177 Ω

0.12

0.08

ΦB = 0.76 V

RS = 47 Ω

0.2 0.4 0.6 I (mA)

–0.6 –0.4 –0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 Voltage (V)

Figure 5.3  (a) The band diagram of MLG/Si Schottky junction solar cell before and after doping. Source: Adapted from Maldonado et al. [25]. (b) Raman spectra of MLG and FLG before and after doping with Au NPs. (c) Dark J–V curves of the FLG/Si Schottky junction before and after doping with Au NPs. The inset shows the plots of dV/d(ln J) versus I. Source: Reproduced with permission from Liu et al. [26]; AIP Publishing.

91

92

5  Graphene/Silicon Solar Energy Harvesting Devices

efficiency to 10.20% courtesy of postdoping higher work function and enhanced electrical conductivity of Gr [27]. From Figure 5.3c, it is clear that Au NP doping reduced the RS of Schottky diode, which lead not only to increased FF but also enhanced Jsc, consistent with the reports on HNO3 doping. Similarly, 10.69% efficiency is achieved by co-doping of Gr with Au NPs and TFSA. Nanoparticles of other high-work-function metals such as Pt principally can also be used for efficiency enhancement. A PCE value of 7% was achieved for Pt-doped Gr, thanks to a combination of effects, including physical doping of high-work-function Pt nanoparticle, enhanced sunlight absorption of solar cells by the plasmonic product, and the photoinduced doping [28]. Defects can be induced during the Gr-transfer process, and these defects act as traps and facilitate photoinduced doping. Under constant illumination, PCE will increase with time and will finally saturate when traps are filled with holes. P3HT polymer has also been found to donate holes to Gr under light illumination, and carrier concentration changed from 2.9 × 1012 to 1.4 × 1013 cm−2 within 60-minutes illumination  [29]. A PCE value of 12.95% was achieved, which degraded less than 10% after one-month storage. In a similar example of light-induced doping, electron and holes were generated in TiOx under light, and the traps in TiOx captured the holes [30]. Electrons were transferred to Gr. An 8.2% and 10.5% efficiency was achieved for PMMA-free and PPMA antireflection (AR)-coated TiOx/Gr/p-Si devices. No significant change in the performance was observed after 10 days [30]. Gr oxide (GO) is a multifunctional material for Gr/Si solar cells. It works as an AR layer, p-type dopant, and passivation/barrier layer. The PCE of the Gr/GO/Si solar cells, where GO layer acted as a p-type dopant as well as an AR layer, reached up to 10% and remained stable for 20  days  [31]. Jiao et  al. inserted the GO as a passivation layer and observed a significant increase in minority carrier lifetime. At the same time, GO also functioned as a barrier layer; however, its barrier properties diminished at higher temperatures (400 °C) when GO decomposed, and its doping levels changed [32].

5.4.2  Light Trapping in Silicon Due to its high reflective index, planar Si has high reflection, and its absorption losses can reach as high as 40%. The light-harvesting capability of Si can be enhanced by decreasing these losses. One way of mitigating these losses is creating nano- or microstructures on the Si surface, which can significantly suppress the reflection of the light at the junction surface and also improve the charge-collection efficiency. By texturing Si into nanowires, the path length of incident solar radiation can be increased by 73 times [33]. Figure  5.4c shows the transmittance spectra of 2 μm (Figure  5.4a) and 5 μm (Figure 5.4b) Si nanowire arrays. Clearly, compared to planar Si, nanowires show reduced transmission pointing to a strong light trapping. Similarly, for Gr/Si solar cells, to reduce the reflection of incident light and to improve the harvesting of the incident light, Si has been textured into pillars, nanowires, pyramids, etc. The presence of nonplanar structural features (pillars, pyramids, nanowires, etc.) reflects the incident light into the Si, instead of transmitting through the Si, resulting in greater

Transmission (%)

5.4  ­Optimization of Graphene/Silicon Solar Cell

(a)

100 90 80 70 60 50 40 30 20 10 0

(b)

5 µm nanowires 2 µm nanowires Planar Optical model

400

500

(c)

600 700 800 900 Wavelength (nm)

1000 1100

Figure 5.4  Cross-sectional SEM images of tilted (a) 2-μm and (b) 5-μm nanowire array samples. (c) Transmission spectra of thin silicon window structures before and after etching into nanowire arrays. Source: Reproduced with permission from Garnett and Yang [33]; American Chemical Society.

absorption due to light trapping and, consequently, improved PCE. Fan et al. [34] fabricated SiNWs through a metal-assisted etching method and demonstrated that Gr/SiNWs Schottky junction had higher absorption compared with the Gr/planar Si junction and demonstrated an energy conversion efficiency of up to 2.86% at AM1.5 condition. Apart from light trapping, the SiNWs also provide a direct and fast pathway to increase the electrical-hole collection and transport, which also contributed to improved efficiency. Feng at al. [20] fabricated the pillar arrays using photolithography and inductive coupled plasma (ICP). By changing the etching time, the height of the pillars was controlled. The results in Figure  5.5 confirm that the Si pillar substrate shows better light absorption compared with planar Si, and absorption improves with an increase in pillar height (best absorption for 900 nm pillar height). 2 μm 5 μm

Reflectance (%)

(a) 80

Silicon pillar

SiO2

40

0 200

(c)

(b)

Planar Si 200 nm 620 nm 915 nm

400 600 Wavelength (nm)

Ti/Au

Silicon Ti/Pd/Au

800

(d)

(e)

Figure 5.5  (a, b) SEM view of Si pillar with 2-μm diameter and silicon transferred on the pillar. (c) Comparative reflectance spectra of Si pillars with various pillar heights and planar silicon substrates. (d) Schematic view of a graphene/Si pillar Schottky solar cell. (e) Photograph of a graphene/Si pillar device. Source: Reproduced with permission from Feng et al. [20]/AIP Publishing.

93

94

5  Graphene/Silicon Solar Energy Harvesting Devices

The conversion efficiency of Gr/Si pillar devices was 1.96% (improved to 3.55% after HNO3 doping and higher than 1.65% efficiency for Gr/planar-Si devices). Xie et al. [35] fabricated the Si hole array (SiHA) by photolithography and reactiveion etching (RIE). The surface of SiHA was relatively smooth, ensuring a low surfacerecombination velocity. As shown in Figure  5.6, the light absorption of SiHA improved with the hole depth. With a combination of light trapping in SiHA and AuCl3 Gr doping, a high-power PCE of 10.40% was achieved. Additionally, the device showed remarkable stability when stored in the air under ambient conditions for about three months.

5.4.3  Antireflection Coating In addition to texturing of the Si substrate, reflection can be further reduced using an antireflection coating (ARC) (Figure  5.7). With ARC on Gr/Si, we will have a four-layer air/ARC/Gr/Si stack (Figure 5.7a). But since the optical contribution is too low, for practical purposes, a four-layer stack can be simplified to a three-layer air/ARC/Si stack, and its reflectance can be calculated as [38] r

nair nSi

nair nSi

nARC 2

nARC 2

2



(5.9)

where nair, nSi, and nARC are the refractive indexes of air, Si, and ARC, respectively. nSi nair , which gives a rough value of nARC = 2 For zero reflectance conditions nARC as Si (nSi ≈ 4) and air (1 nair ≈ 1). For optimal thickness of ARC in visible range, the expression d  =  λ/4n ARC should hold. Figure 5.7b shows the refractive index of some of ARC candidate materials. Based on its reflective index TiO2 is a candidate for ARC. As Gr is very sensitive to defects, high energy deposition of TiO2 (sputtering) on Gr surface is not suitable. Instead, Shi et al. [36, 39] adopted the spin coating of solution-processable TiO2 colloids on the Gr/Si solar cells as the AR layer. The reflection spectrum revealed that in the optical range of 200–1100 nm, the light reflectance could be reduced to 10% after TiO2 coating, as shown in Figure 5.7c, and subsequently, the cell efficiency was increased to 14.5% after HNO3 doping (Figure 5.7d). High refractive index polymers such as PMMA (n ~ 1.5) are highly transparent to visible light and can be potentially used as ARC for Gr/Si solar cells [40]. Though its refractive index is comparatively low compared to inorganic ARC materials, it offers the advantage of being lightweight, bendable, and flexible. Moreover, traditionally, PMMA is used as a sacrificial support layer in Gr transfer; the use of PMMA as an ARC will eliminate the step of PMMA removal after Gr transfer, which will greatly simplify the device fabrication process. Photovoltaic performance of PMMA-coated Gr films on Gr/Si solar cells was significantly improved compared to bare Gr/Si solar cells obtained after PMMA removal. Figure  5.7e compares the reflectance spectra of PMMA-removed Gr/Si and PMMA-coated Gr/Si devices. Clearly, PMMA-coated Gr/Si devices showed better absorption [37].

Graphene

O2

Si

Graphene Ti/Au In–

Planar Si

(b)

SiNWs

(c)

6 µm

5 µm

J(mA cm–2)

0 –10

SiHA i

60

–20

40 20

–30

10 µm

20 µm (e)

(f)

–0.15 0.00 0.15 0.30 Voltage (V) (g)

20 µm

80

3.8 µm 6.4 µm 10.2 µm 12.8 µm

EQE (%)

(a)

Ga

n–S

0 0.45

3.8 µm 6.4 µm 10.2 µm 12.8 µm

400 (h)

600 800 1000 Wavelength (nm)

Figure 5.6  (a) Schematics of Gr/planar Si and (b) Gr/SiNW junctions. Source: Reproduced with permission from Fan et al. [34]. American Chemical Society. Arrows show the light path in Si. (c) Schematics of the Gr/Si hole arrays Schottky junction. (d) Top-view SEM image of the graphene/Si hole array device. (e) Top-view and (f) cross-sectional SEM images of the as-prepared Si hole arrays. (g) J–V curves under light. (h) EQE spectra of the graphene/SiHA Schottky junction solar cells with different hole depths. Source: Reproduced with permission from Xie et al. [35]. Royal Society of Chemistry.

5  Graphene/Silicon Solar Energy Harvesting Devices

i

tl

n de

t gh

Reflection

ci

In

Air (ñair) x

Multiple reflection

d1

ARC (ñARC)

d2

Graphene

Refractive index

3

2

1

Si (ñSi)

80 60 G–Si cell

40 20 0

TiO2–G–Si cell

400 500 600 700 800 900 10001100

Wavelength (nm)

(c) 100

PMMA removed 1000 rpm 2000 rpm 3000 rpm

80 60 40 20 0

400

600

Wavelength (nm)

800

Current density (mA cm–2)

Reflectance (%)

100

(e)

0 400

TiO2 SiO2 MgF2 500

MoO3 Si3N4 sqrt(nsi)

600

700

800

20 G–Si

10 0

HNO3 doping

–10 –20 –30 –40 –0.2

TiO2/HNO3

0.0

0.2

0.4

0.6

Voltage (V)

(d)

(f)

PMMA ZnO SiC

Wavelength (nm)

(b)

Current density (mA cm–2)

(a)

Reflectancet (%)

96

10 0 –10

PMMA-removed (pristine) PMMA-removed (HNO3 doped) PMMA-coated (pristine) PMMA-coated (HNO3 doped)

–20 –30 –40 –0.2

0.0

0.2

0.4

0.6

Voltage (V)

Figure 5.7  (a) Texturing of the Si substrate. (b) Refractive index versus wavelength of some of ARC materials. (c) Light reflection of a graphene/Si solar cell at various wavelengths before (black) and after (red) coating with TiO2 colloid. Curves clearly show antireflection properties of TiO2. (d) Effect of HNO3 doping and TiO2 coating on J–V characteristics of as-fabricated Gr/Si Schottky junction solar cell. Source: Reproduced with permission from Shi et al. [36]; American Chemical Society. (e) Reflectance spectra of Gr with and without PMMA. (f) Photovoltaic curves of Gr/Si Schottky junction with and without PMMA and HNO3 doping. Source: Reproduced with permission from Gan et al. [37] ; Royal Society of Chemistry.

Similarly, Voc, Jsc, FF, and PCE of PMMA-removed Gr/Si devices obtained from J–V curves in Figure 5.7f were determined to be 0.428 V, 22.61 mA cm−2, 32.22%, and 3.12%, respectively. Values for PMMA-coated Gr/Si were 0.427 V, 33.64 mA cm−2, 45.56%, and 6.55%, respectively. After doping with HNO3 vapor, the PCE of PMMAcoated Gr/Si solar cell devices improved to 13.34%. Most importantly, HNO3 doping did not affect the PMMA structure, which made the doping process of PMMAcoated Gr convenient [37]. Ding et al. [41] used double-layered MgF2/ZnS coating on Gr/Si solar cells, which served the dual purpose. First, for Si solar cells working in the range of 380–1100 nm, ZnS and MgF2 own maximum (n = 2.5) and minimum (n  =  1.4) refractive indexes (Figure  5.8). Hence, the MgF2/ZnS double layer can

5.4  ­Optimization of Graphene/Silicon Solar Cell

40

Red Yellow Cyan Purple

80

30

60 40 20

10 0

(a)

400

500

600 700 800 900 1000 1100 Wavelength (nm)

0

(b)

J (mA cm–2)

–5 –10 –15 –20

Uncoated Si Red Orange Yellow Green Cyan Blue Purple Dark

–25

–5

–15 –20 –25

–35

–30 0.0

SL-Yellow Jsc = 28.1 PCE = 10.6% DL-Yellow Jsc = 30.0 PCE = 11.9%

0.2

0.4

V (V)

0.6

100 80

–10

–30 –40 –0.2

400 500 600 700 800 900 1000 1100 Wavelength (nm)

0 J (mA cm–2)

0

Red Yellow Cyan Purple

Blue Dark

5

5

(c)

Uncoated Si Orange Green

20

EQE (%)

Reflectivity (%)

50

100 Uncoated Si Orange Green Blue Dark

EQE (%)

Simulation measurement

60

60 40 20 0

Wavelength (nm)

0.0

(d)

400 500 600 700 800 900 1000 1100

0.1

0.2

0.3

0.4

0.5

0.6

V (V)

Figure 5.8  (a) Simulated (dashed lines) and actual (solid lines) reflection spectra of DL-MgF2/ZnS-coated Gr/Si Schottky junction solar cells. Dotted and solid lines show simulation and experimental results. (b) Effect of different structural colors on EQE spectra of Gr/Si Schottky junction solar cells. (c) J–V spectra of the devices with different structural colors. (d) Effect of SL-ZnS and DL-MgF2/ZnS coating on photovoltaic characteristics of the Gr/Si device. The inset shows EQE spectra. Source: Reproduced with permission from Ding et al. [41]; Elsevier.

serve the purpose of an effective ARC. Second, by varying the relative thickness of ZnS and MgF2, different colors can be imparted to Gr/Si solar cells. These colored solar modules can find important applications where solar module colors need to be matched with the building fabric for esthetically pleasing designs. For Gr/Si solar cells, the color tuning process was relatively convenient because of the transparent nature of Gr. Figure 5.9 depicts the schematic illustration, the photographs, and the coordinates of seven colors in CIE chromaticity of Gr/Si heterojunction solar cells with DL-MgF2/ZnS films. Figure 5.9 shows the reflectance spectra of various colored MgF2/ZnS-coated Gr/Si devices. Absorption was improved compared to pristine Gr/ Si solar cells. Similar trends were observed for the EQE spectra. A high PCE of 10.7–13.2% was achieved for the multicolor Gr/Si solar cells. The PCE of the device using optimized MgF2/ZnS ARC reached as high as 14.6%.

5.4.4  Interface Engineering Despite several strategies being implemented to improve the Gr/Si devices, their efficiency is still below the commercial Si p–n junction solar cells, which is mainly due to the weaker built-in electric field in the Schottky junction compared with the p–n junction, which cannot fully restrict the recombination of carriers along with

97

5  Graphene/Silicon Solar Energy Harvesting Devices (a)

0.9 0.8

ZnS

0.7

Ag electrode Graphene SiO2 n-Si

540

560

100 µm

0.6

Green

500

Yellow

0.5

5 µm

580

Orange

0.4 0.3

600

490

Red

0.0

620 700

Cyan

0.2

Purple 480

0.1

In–Ga alloy electrode

(c)

520

MgF2

y

98

0.0

Blue

470 460 330

0.1

0.2

0.3

0.4 x

0.5

0.6

0.7

0.8

(b)

Figure 5.9  (a) Schematic illustration of color graphene/Si heterojunction solar cells with DL-MgF2/ZnS films. The inset shows the SEM image of the Ag grid mesh. (b) Photographs of graphene/Si heterojunction solar cells (device area: 1 cm2) with different colors. (c) Coordinates of seven colors in CIE chromaticity. Source: Reproduced with permission from Ding et al. [41]; Elsevier.

the interface. Insertion of an interfacial barrier layer can add a tunneling barrier which can suppress the recombination of carriers by blocking the majority of carriers (electron in n-Si). Besides functioning as an electron-blocking layer, band alignment between the barrier layer and n-Si should be such that it should facilitate the hole transfer to Gr. The thickness of the barrier layer is also important as a relatively thicker barrier layer with a large tunnel barrier can lead to ohmic losses, and a toothin barrier will not be able to block the majority carriers. Hence, the thickness of the barrier layer needs to be optimized to achieve a trade-off between low carrier recombination and ohmic losses. Insulators (SiO2, h-BN, etc.) and semiconductors (P3HT, MoS2, MoO3, etc.) have been used as an interfacial barrier layer. It is found that oxide thickness less than 2 nm (which can be grown by exposing the Si to air for some time) can passivate the interface and increase the Voc and FF [22]. By choos­ ing the optimal oxide thickness (1.5 nm), device efficiency of 12.4% has been achieved after chemical doping, which further increased to 15.6% after applying a TiO2 ARC. However, it is not easy to control the thickness of ultrathin SiO2, as it depends on crystal orientation and environmental conditions. 2D insulator h-BN has merits of dangling bond-free surface, small lattice mismatch with Gr (17%), optically transparent in the visible range, and proper band alignment with n-Si [42, 43]. These merits make it an attractive barrier layer for Gr/ Si solar cells. Figure  5.10a,b shows the band diagram of a Gr/h-BN/SiO2 solar cell [44]. It is clear that compared with a simple Gr/Si interface, insertion of h-BN imparts a large barrier for electron transfer at the interface, but at the same time, holes can transport easily from Si to Gr. By Introducing the BN layer, the Gr/Si solar cell PCE reached 10.93%, which showed a 15% efficiency improvement compared with the reference device.

5.4  ­Optimization of Graphene/Silicon Solar Cell

WG

ΔEc > 4.05 eV

Evac Φ2

ΦB

Graphene

WG

χSi h-BN

Ec EF

Ec EF

Graphene

n-Si

n-Si ΔEv < 0.63 eV

Ev (a)

Ev

(b)

Vacuum level

Vacuum level

4.05 eV –

–

EF +

n-Si (c)

Evac

4.8 eV 4.05 eV

–

Vb

4.8 eV

4.0 eV –

EF

Vb +

+

+

n-Si

Graphene

MoS2 Graphene

(d)

Figure 5.10  (a, b) Effect of h-BN electron-blocking layer on energy band diagrams of Gr/Si Schottky junction under illumination. Source: Reproduced with permission from Meng et al. [44]; Elsevier. (c, d) Energy band diagrams of Gr/n-Si and Gr/MoS2/n-Si solar cells. Source: Reproduced with permission from Tsuboi et al. [45]; Royal Society of Chemistry.

Comparatively thicker (~50nm) semiconductor layers can be inserted as the barrier layer, making the barrier layer processing more convenient compared with the insulators requiring lower thickness. CVD grown/transferred, direct CVD grown on Si without need to transfer, and chemically processed MoS2 have been used as an interfacial barrier layer with PCEs of 11.1%, 6.6%, and 15.8%, respectively [45]. As shown in Figure 5.10c,d, after insertion of the MoS2 layer, bands of MoS2 layer at MoS2/Gr interface bent upward, resulting in an energy barrier MoS2/n-Si at the interface and enhanced total energy barrier between Gr and n-Si. The added energy barrier of the MoS2 layer functions as an extra electron-blocking layer for the photogenerated electrons in the Si. Similar band bending at the semiconductor barrier layer/n-Si interface has been reported for WO3/n-Si. Insertion of WO3 thin films leads to significant upward band bending at the WO3/Si interface, which caused enhanced built-in field (Vbi) and PCE of 10.59%, which was improved from 3.99% efficiency of simple Gr/Si solar cells. Organic semiconductor layers such as P3HT can also selectively block the electrons. As can be seen from Figure 5.11 [46], the lowest unoccupied molecular orbital of the P3HT (3.2 eV) and the conduction band minimum of Si (4.05) provide a large

99

5  Graphene/Silicon Solar Energy Harvesting Devices

3.2

4 Pristine HNO3 5

Graphene 5.1

e– 4.05

4.2

n-Si

3

P3HT

Energy (eV)

100

In/Ga

5.17 h+

Figure 5.11  Schematic illustrations of the band diagram of the hybrid solar cells. Source: Reproduced with permission from Wu et al. [46]; American Chemical Society.

offset to block the electrons. However, a relatively small energy difference between the highest occupied molecular orbital (5.1 eV) of P3HT and the valence band maximum (5.17 eV) allows the transport of holes. PCEs of optimized Gr/P3HT/SiNW array and Gr/P3HT/SiNH array devices were 9.94% and 10.34%, respectively. Another interesting interface between Si and MoO3 layer has been reported, where a PCE value of 12.2% was reported. The large work function of MoO3 resulted in hole injection from the MoO3 to Si, which led to a hole inversion layer at the Si surface, and a strong built-in electric field was developed, which suppressed the recombination. With further device optimization by doping the Gr and employing the ARC, the PCE improved to 12.2%.

5.4.5  Surface Passivation Si surface has lots of dangling bonds, creating surface states that act as recombination centers. Recombination can severely affect the performance of photovoltaic devices. Hence, it is important to passivate the Si surfaces. Si surfaces can be passivated via different chemical groups such as hydrogen (H−), methyl (CH3−), and oxide (SiOX−) [47, 48]. Hydrogen-passivated Si surfaces possess excellent passivating capacity (reduction in surface states by many folds), and the surface recombination velocity of these passivated surfaces is also relatively low. However, hydrogenterminated Si surfaces have low chemical stability and are also not compatible with Gr-related processing. Methyl-terminated Si surfaces have excellent chemical stability and have even lower surface recombination rates (45 cm s−1) than H-terminated surfaces (500 cm s−1)  [25, 47, 49]. Another important difference is that hydrogenterminated Si will suffer from a Schottky barrier decrease, while methyl-terminated Si will see an increase in the Schottky barrier. Hence, methyl-terminated Si surfaces are better for solar cell efficiency improvement. The PCEs of solar cells, with hydrogen, methyl, and oxide passivation, are found to be 1.41%, 1.76%, and 4.42% [50]. Passivation can also be achieved using inorganic layers such as SiO2, Al2O3, HfO2, fluorographene, and SiNx [21, 51–53]. Apart from passivation, these layers also serve as the barriers so their thickness needs to be precisely controlled. GO, BN, and P3HT have also been used as the passivation layer for the improvement of solar cell efficiency [44, 50, 54].

5.5  ­Challenges and Perspective

5.5  ­Challenges and Perspectives Since the first report on Gr/Si Schottky junction solar cells, the efficiency of these cells is constantly improving, thanks to strategies such as doping of Gr, ARCs, interface engineering, and light trapping in Si [10, 55, 56]. Apart from efficiency, efforts have also been made to improve the environmental stability and impart new functionalities (flexibility, etc.) to Gr/Si solar cells. Gr, despite its several advantages, has certain drawbacks for solar cell applications, which researchers are focusing on rectifying. For example, Gr is extremely sensitive to the environment, and common environmental gases such as oxygen and carbon dioxide can dope the Gr and change its work function. Hence, the reliability of Gr/Si solar cells is a serious issue. Encapsulation of Gr might protect the Gr from the ambient environment, but cost and manufacturing complexity will increase. Secondly, production and transfer of the Gr to Si for industrial-scale production of solar cells is not possible yet. While significant progress has been made toward large-scale production of Gr, an automatic industrial-scale Gr-transfer method is still lacking. It has been demonstrated that to improve the efficiency of Gr/Si Schottky junction solar cells, the conductivity of the Gr needs to be improved. One solution is to use multilayer Gr; however, there is a trade-off between conductivity and the transparency. Hence, the number of Gr layers and their stacking needs to carefully engineered to optimize the efficiency of the Gr/Si Schottky junction solar cells. The second method to improve the conductivity is doping, which can simultaneously improve the conductivity of the Gr and change its work function; both of the parameters are crucial for improving the solar cell efficiency. But commonly used dopants such as HNO3, though effective, are environmentally not stable, and dopants will evaporate with time. Alternate dopants such as AuCl3 are environmentally stable but expensive. These dopants, when used in large concentrations, tend to decrease the transparency of the Gr. Hence, a careful combination of dopant, doping density, and dopant concentration needs to be developed to optimize the efficiency of Gr/Si solar cells. There is also an urgent need to find new stable dopants that can effectively dope the Gr without significantly compromising the transparency of Gr. As 2D materials is an evolving field, most of the innovation to Gr/Si solar cells comes mainly from the Gr. Efforts to improve the efficiency of Gr/Si solar cells with better contribution from Si have to mainly focus on improving the light absorption in Si using traditional strategies, such as texturing, and ARCs have been employed. Surface passivation of Si surface (especially textured) for better extraction efficiency of photogenerated carriers is also the focus of improvement strategies, but these concepts are already well developed for decades, and the scope of improvement in the near future is limited. Significant progress has been made in the improvement of efficiency of Gr/Si solar cell devices, and efficiency has been improved from 5% to 15.8% within a decade. However, this efficiency is still low compared to other Si and non-Si-based solar cells, and it is a major challenge to improve their efficiency. Despite the lower efficiency, Gr/Si solar cells have great potential for commercialization as the real

101

102

5  Graphene/Silicon Solar Energy Harvesting Devices

advantage of Gr/Si solar cells lies in their cost, easy processing, and better material choices (Si and Gr).

­References   1 P  arida, B., Yoon, S., Jeong, S.M. et al. (2020). Recent progress on cesium lead/tin halide-based inorganic perovskites for stable and efficient solar cells: a review. Solar Energy Materials and Solar Cells 204: 110212.   2 Yoshikawa, K., Kawasaki, H., Yoshida, W. et al. (2017). Silicon heterojunction solar cell with interdigitated back contacts for a photoconversion efficiency over 26%. Nature Energy 2 (5): 1–8.   3 Nakamura, M., Yamaguchi, K., Kimoto, Y. et al. (2019). Cd-free Cu(In,Ga)(Se,S)2 thin-film solar cell with record efficiency of 23.35%. IEEE Journal of Photovoltaics 9 (6): 1863–1867.   4 Yan, N., Zhao, C., You, S. et al. (2020). Recent progress of thin-film photovoltaics for indoor application. Chinese Chemical Letters 31 (3): 643–653.   5 Novoselov, K.S., Geim, A.K., Morozov, S.V. et al. (2004). Electric field effect in atomically thin carbon films. Science 306 (5696): 666–669.   6 Geim, A.K. and Novoselov, K.S. (2007). The rise of graphene. Nature Materials 6 (3): 183–191.   7 Balandin, A.A. (2011). Thermal properties of graphene and nanostructured carbon materials. Nature Materials 10 (8): 569–581.   8 Mahmoudi, T., Wang, Y., and Hahn, Y.-B. (2018). Graphene and its derivatives for solar cells application. Nano Energy 47: 51–65.   9 Zheng, Q., Li, Z., Yang, J., and Kim, J.-K. (2014). Graphene oxide-based transparent conductive films. Progress in Materials Science 64: 200–247. 10 Huang, K., Yu, X., Cong, J., and Yang, D. (2018). Progress of graphene-silicon heterojunction photovoltaic devices. Advanced Materials Interfaces 5 (24): 1801520. 11 Li, X., Zhu, H., Wang, K. et al. (2010). Graphene-on-silicon Schottky junction solar cells. Advanced Materials 22 (25): 2743–2748. 12 Shockley, W. and Queisser, H.J. (1961). Detailed balance limit of efficiency of p–n junction solar cells. Journal of Applied Physics 32 (3): 510–519. 13 Kerr, M.J., Cuevas, A., and Campbell, P. (2003). Limiting efficiency of crystalline silicon solar cells due to Coulomb-enhanced Auger recombination. Progress in Photovoltaics: Research and Applications 11 (2): 97–104. 14 Richter, A., Hermle, M., and Glunz, S.W. (2013). Reassessment of the limiting efficiency for crystalline silicon solar cells. IEEE Journal of Photovoltaics 3 (4): 1184–1191. 15 Walker, G. (2001). Evaluating MPPT converter topologies using a MATLAB PV model. Australian Journal of Electrical and Electronics Engineering 21 (1): 49–55. 16 Sze, S.M., Li, Y., and Ng, K.K. (2021). Physics of Semiconductor Devices. Wiley. 17 Li, X., Zhu, Y., Cai, W. et al. (2009). Transfer of large-area graphene films for high-performance transparent conductive electrodes. Nano Letters 9 (12): 4359–4363.

  ­Reference

18 Cui, T., Lv, R., Huang, Z.-H. et al. (2013). Enhanced efficiency of graphene/silicon heterojunction solar cells by molecular doping. Journal of Materials Chemistry A 1 (18): 5736–5740. 19 Feng, T., Xie, D., Lin, Y. et al. (2012). Efficiency enhancement of graphene/ silicon-pillar-array solar cells by HNO3 and PEDOT-PSS. Nanoscale 4 (6): 2130. 20 Feng, T., Xie, D., Lin, Y. et al. (2011). Graphene based Schottky junction solar cells on patterned silicon-pillar-array substrate. Applied Physics Letters 99 (23): 233505. 21 Ahn, J., Chou, H., and Banerjee, S.K. (2017). Graphene-Al2O3-silicon heterojunction solar cells on flexible silicon substrates. Journal of Applied Physics 121 (16): 163105. 22 Song, Y., Li, X., Mackin, C. et al. (2015). Role of interfacial oxide in high-efficiency graphene–silicon Schottky barrier solar cells. Nano Letters 15 (3): 2104–2110. 23 Wang, X., Xu, J.-B., Xie, W., and Du, J. (2011). Quantitative analysis of graphene doping by organic molecular charge transfer. Journal of Physical Chemistry C 115 (15): 7596–7602. 24 Miao, X., Tongay, S., Petterson, M.K. et al. (2012). High efficiency graphene solar cells by chemical doping. Nano Letters 12 (6): 2745–2750. 25 Maldonado, S., Knapp, D., and Lewis, N.S. (2008). Near-ideal photodiodes from sintered gold nanoparticle films on methyl-terminated Si(111) surfaces. Journal of the American Chemical Society 130 (11): 3300–3301. 26 Liu, X., Zhang, X., Yin, Z. et al. (2014). Enhanced efficiency of graphene-silicon Schottky junction solar cells by doping with Au nanoparticles. Applied Physics Letters 105 (18): 183901. 27 Liu, X., Zhang, X., Meng, J. et al. (2015). High efficiency Schottky junction solar cells by co-doping of graphene with gold nanoparticles and nitric acid. Applied Physics Letters 106 (23): 233901. 28 Huang, K., Yan, Y., Yu, X. et al. (2017). Graphene coupled with Pt cubic nanoparticles for high performance, air-stable graphene-silicon solar cells. Nano Energy 32: 225–231. 29 Xu, D., He, J., Yu, X. et al. (2017). Illumination-induced hole doping for performance improvement of graphene/n-silicon solar cells with P3HT interlayer. Advanced Electronic Materials 3 (3): 1600516. 30 Ho, P.-H., Lee, W.-C., Liou, Y.-T. et al. (2015). Sunlight-activated grapheneheterostructure transparent cathodes: enabling high-performance n-graphene/p-Si Schottky junction photovoltaics. Energy & Environmental Science 8 (7): 2085–2092. 31 Yavuz, S., Kuru, C., Choi, D. et al. (2016). Graphene oxide as a p-dopant and an anti-reflection coating layer, in graphene/silicon solar cells. Nanoscale 8 (12): 6473–6478. 32 Jiao, K., Wang, X., Wang, Y., and Chen, Y. (2014). Graphene oxide as an effective interfacial layer for enhanced graphene/silicon solar cell performance. Journal of Materials Chemistry C 2 (37): 7715–7721. 33 Garnett, E. and Yang, P. (2010). Light trapping in silicon nanowire solar cells. Nano Letters 10 (3): 1082–1087. 34 Fan, G., Zhu, H., Wang, K. et al. (2011). Graphene/silicon nanowire Schottky junction for enhanced light harvesting. ACS Applied Materials and Interfaces 3 (3): 721–725.

103

104

5  Graphene/Silicon Solar Energy Harvesting Devices

35 Xie, C., Zhang, X., Ruan, K. et al. (2013). High-efficiency, air stable graphene/Si micro-hole array Schottky junction solar cells. Journal of Materials Chemistry A 1 (48): 15348. 36 Shi, E., Li, H., Yang, L. et al. (2013). Colloidal antireflection coating improves graphene-silicon solar cells. Nano Letters 13 (4): 1776–1781. 37 Gan, X., Lv, R., Zhu, H. et al. (2016). Polymer-coated graphene films as antireflective transparent electrodes for Schottky junction solar cells. Journal of Materials Chemistry A 4 (36): 13795–13802. 38 Raut, H.K., Ganesh, V.A., Nair, A.S., and Ramakrishna, S. (2011). Anti-reflective coatings: a critical, in-depth review. Energy & Environmental Science 4 (10): 3779–3804. 39 Shi, E., Zhang, L., Li, Z. et al. (2012). TiO2-coated carbon nanotube-silicon solar cells with efficiency of 15%. Scientific Reports 2 (1): 1–5. 40 Jiao, T., Wei, D., Song, X. et al. (2016). High-efficiency, stable and non-chemically doped graphene–Si solar cells through interface engineering and PMMA antireflection. RSC Advances 6 (12): 10175–10179. 41 Ding, K., Zhang, X., Ning, L. et al. (2018). Hue tunable, high color saturation and high-efficiency graphene/silicon heterojunction solar cells with MgF2/ZnS double anti-reflection layer. Nano Energy 46: 257–265. 42 Dean, C.R., Young, A.F., Meric, I. et al. (2010). Boron nitride substrates for highquality graphene electronics. Nature Nanotechnology 5 (10): 722–726. 43 Pakdel, A., Bando, Y., and Golberg, D. (2014). Nano boron nitride flatland. Chemical Society Reviews 43 (3): 934–959. 44 Meng, J.-H., Liu, X., Zhang, X.-W. et al. (2016). Interface engineering for highly efficient graphene-on-silicon Schottky junction solar cells by introducing a hexagonal boron nitride interlayer. Nano Energy 28: 44–50. 45 Tsuboi, Y., Wang, F., Kozawa, D. et al. (2015). Enhanced photovoltaic performances of graphene/Si solar cells by insertion of a MoS2 thin film. Nanoscale 7 (34): 14476–14482. 46 Wu, Y., Zhang, X., Jie, J. et al. (2013). Graphene transparent conductive electrodes for highly efficient silicon nanostructures-based hybrid heterojunction solar cells. The Journal of Physical Chemistry C 117 (23): 11968–11976. 47 Zhang, X., Xie, C., Jie, J. et al. (2013). High-efficiency graphene/Si nanoarray Schottky junction solar cells via surface modification and graphene doping. Journal of Materials Chemistry A 1 (22): 6593–6601. 48 Sieval, A.B., Huisman, C.L., Schönecker, A. et al. (2003). Silicon surface passivation by organic monolayers: minority charge carrier lifetime measurements and Kelvin probe investigations. The Journal of Physical Chemistry B 107 (28): 6846–6852. 49 Zhang, F., Sun, B., Song, T. et al. (2011). Air stable, efficient hybrid photovoltaic devices based on poly(3-hexylthiophene) and silicon nanostructures. Chemistry of Materials 23 (8): 2084–2090. 50 Xie, C., Zhang, X., Wu, Y. et al. (2013). Surface passivation and band engineering: a way toward high efficiency graphene–planar Si solar cells. Journal of Materials Chemistry A 1 (30): 8567–8574.

  ­Reference

51 Rehman, M.A., Akhtar, I., Choi, W. et al. (2018). Influence of an Al2O3 interlayer in a directly grown graphene-silicon Schottky junction solar cell. Carbon 132: 157–164. 52 Alnuaimi, A., Almansouri, I., Saadat, I., and Nayfeh, A. (2018). High performance graphene-silicon Schottky junction solar cells with HfO2 interfacial layer grown by atomic layer deposition. Solar Energy 164: 174–179. 53 Schmidt, J., Merkle, A., Brendel, R. et al. (2008). Surface passivation of highefficiency silicon solar cells by atomic-layer-deposited Al2O3. Progress in Photovoltaics: Research and Applications 16 (6): 461–466. 54 Yang, L., Yu, X., Xu, M. et al. (2014). Interface engineering for efficient and stable chemical-doping-free graphene-on-silicon solar cells by introducing a graphene oxide interlayer. Journal of Materials Chemistry A 2 (40): 16877–16883. 55 Das, S., Pandey, D., Thomas, J., and Roy, T. (2019). The role of graphene and other 2D materials in solar photovoltaics. Advanced Materials 31 (1): 1802722. 56 Ju, S., Liang, B., Wang, J.-Z. et al. (2018). Graphene/silicon Schottky solar cells: technical strategies for performance optimization. Optics Communication 428: 258–268.

105

107

6 Graphene Silicon-­integrated Waveguide Devices 6.1 ­Introduction In recent years, with the development of new-­generation information technologies such as big data, cloud computing, and the Internet of things, the amount of data in the process of information generation, processing, and storage is facing an “explosive” growth. The digital information generated in the past two years alone accounts for 90% of the total amount of the existing data [1]. At the same time, to support data storage and high-­performance computing, the capacity continues to grow according to Moore’s law [2]. The bandwidth density of interchip communication needs to be greatly improved (by 2020, the bandwidth density demand of each interconnect has exceeded 40 Gb s−1 [1]). Due to the disadvantages of limited bandwidth, electrical cross talk, and low input/output pin density, the traditional integrated circuit electrical interconnection technology cannot meet the high-­speed and -­density data-­ processing requirements under the “information explosion”  [3]. Optical interconnection has the advantages of large bandwidth, low power consumption, and low cross talk. It is expected to replace the traditional electrical interconnection and realize high-­speed information exchange [4]. With the development of Si-­based photonic technology, Si-­based optoelectronic chip has become one of the most promising and attractive platforms for applying optical interconnection on-­chip and information interaction between chips [5]. The physical characteristics of Si make Si-­based photonic technology have many significant advantages: (i) there is a high refractive index difference between Si and cladding as optical transmission waveguide, which can realize ultralow loss total reflection optical transmission [6]; (ii) Si has an indirect band gap of 1.1 eV, which can provide an ultrawide transparent window from near-­ to midinfrared band [7]; (iii) Si waveguides have controllable dispersion and nonlinearity [8]; (iv) Si materials have high damage threshold, large thermal conductivity, and mature complementary metal–oxide–semiconductor (CMOS)-­processing technology  [9]; and (v) the strong binding effect of Si waveguides on transmitted light is conducive to the flexible design of optical modes and can be used to realize passive devices such as microring cavities, polarization conversion, and wavelength division multiplexers [10, 11]. Graphene for Post-Moore Silicon Optoelectronics, First Edition. Yang Xu, Khurram Shehzad, Srikrishna Chanakya Bodepudi, Ali Imran, and Bin Yu. © 2023 WILEY-VCH GmbH. Published 2023 by WILEY-VCH GmbH.

6  Graphene Silicon-­integrated Waveguide Devices

Integrated emission intensity

Therefore, Si photonics has been widely valued by enterprises (such as IBM and Intel) and research teams around the world (such as Lipson group of Columbia University, Bowers group of UCSB University, and photonics research group of Ghent University), and a series of breakthroughs have come one after another. In 2019, Intel demonstrated a 400-­GB Si optical transceiver module on interconnect day, including four channels with different wavelengths. Each channel can achieve a transmission rate of 100 GB s−1 and can be applied to 5G and data centers. Acacia, another Si optical company, focuses on providing high-­speed coherent optical interconnection products based on Si optical platform, has a complete DSP solution, and has also launched a 400-­GB pluggable coherent transmission system for long-­ distance, metropolitan, and cross-­data center scenarios. Nevertheless, Si is not an ideal active optical material, and other Si-­compatible materials, including Si nitride (SixNy) and Si oxynitride (SiOxNy), can only be used in passive optical devices. There are still many problems in Si-­based active optical devices. Firstly, Si is an indirect bandgap semiconductor, so its luminous efficiency through overload current recombination is very low, and it is not suitable for making light sources. At present, there are four indirect ways to realize Si-­based light sources [12], as shown in Figure  6.1a–d: (i) hybrid Si laser realized by bonding III–V gain medium (such as indium phosphide) with Si; (ii) germanium Si laser epitaxially growing germanium on Si substrate; (iii) light-­emitting device realized by doping rare earth elements in Si or Si dioxide layer; and (iv) all-­Si Raman laser optically pumped by the stimulated Raman scattering effect of Si. The way of using III–V compound semiconductor, germanium, or rare-­earth element-­assisted luminescence hinders the promotion of commercialization because of its high cost and technical difficulty. However, the application of all-­Si Raman lasers is limited because it cannot realize electric pumping. 25 Emission intensity (a.u.)

108

(a)

20

15

h+

h+

60 40 20 0 0

10 20 30 40 50 Pump energy per pulse (µJ)

50 µJ

5

6.0 µJ

0

1.5 µJ 1560

1580

1600

1620

1640

Ge Si

1660

Pump p-InGaAs p-InP Active region Laser output n-InP

(d)

1700

Directional coupler Iinc Ip(L)Ip(O)

p-region

z Vbias

n-region

(c)

500 nm

1680

Wavelenght (nm)

Bus waveguide

p-contact

SOI

80

10

(b)

n-contact

100

Ring cavity

Figure 6.1  (a) Si Raman lasers. (b) Optically pumped Ge-­on-­Si laser. (c) Hybrid Si-­distributed feedback lasers. (d) Hybrid Si microring lasers. multi-­quantum well, MQW; BOX, buried oxide. Source: Reproduced with permission from Liang and Bowers [12]/Springer Nature.

6.1 ­Introductio

Secondly, as a centrosymmetric crystal, Si lacks the Pockels effect, and because it has only a very weak Kerr effect and Franz–Keldysh effect in the communication band, it is difficult to realize the traditional electro-­optic modulator with Si. The current all-­Si optical modulator  [13] is mainly through the plasma dispersion effect, that is the effect of Si complex refractive index change caused by the injection, accumulation, or depletion of free carrier concentration. The commonly used structures include Mach Zehnder interferometer and ring or disk resonator, as shown in Figure 6.2a,b, respectively. In addition, it is also advantageous to use a germanium or III–V semiconductor electric absorption optical modulator  [13]. These optical modulators usually have large device sizes or can only work in a very narrow wavelength range. The optical modulators assisted by germanium or III–V materials also have the problems of high cost and technical difficulty. The large resistance of the p–n junction in the core of the waveguide Si optical modulator also limits the further improvement of the working speed. Traveling wave electrodes Ground Signal n++

n-Si p++

p-Si

Oxide

p++

Ge

Waveguide

500 nm

500 nm Si waveguide

Si substrate

(a)

260 nm Si slab

220 nm

220 nm

Ge layer Si recess

60 nm

(c)

Output SiO

50 µm

Si p+

n+

n-metal

Waveguide

n-metal

p-metal

BCB

p-metal

n+ doped p+ doped

νF

III–V

1 µm

Ring

InP

Input

(b)

(d)

Figure 6.2  (a) Cross section of the carrier-­depletion-­based Si optical modulator. (b) Si optical modulator using a ring resonator structure. (c) High-­performance Ge-­on-­Si photodetector. (d) InP PIN photodetector integrated on SOI waveguide. BCB, benzocyclobutene. Source: Reproduced with permission from Reed et al. [13]/ Springer Nature.

109

110

6  Graphene Silicon-­integrated Waveguide Devices

In addition, the bandgap of Si at room temperature and pressure is 1.12 eV, and the cutoff wavelength of intrinsic absorption is 1.1 μm. The near-­ to midinfrared bands of M are almost transparent and cannot form effective absorption. To realize photodetectors on Si, people usually need to introduce other absorption media, such as epitaxial growth of germanium  [14] on Si substrate and III–V semiconductor bonded with SOI [15], see Figure 6.2a,b. This also inevitably increases the manufacturing cost and implementation difficulty of devices, and the working wavelength of these devices is also limited by materials, which cannot realize the detection of all communication bands. In general, there is still no perfect solution to realize active optical devices on Si. There is still a certain gap in the performance of Si-­based active optical devices compared with devices made of III–V or lithium niobite (LiNbO3) as displayed in Figure 6.2c,d. Therefore, the large-­scale commercial application of Si photonic-­integrated circuit still lacks enough power and persuasion. It is necessary to make a revolutionary breakthrough such as the Si transistor in the integrated circuit to vigorously promote its development. To further improve the performance of Si-­based active optical devices and reduce the production cost and difficulty, it is necessary to find a kind of CMOS-­compatible material with excellent performance, low cost, and easy integration to assist Si to realize the monolithic integration of all­Si photonic-­integrated devices. The emergence of two-­dimensional materials provides a new possibility for the realization of high-­performance Si-­based active devices. As mentioned earlier in this book, combining Gr with Si photonics platform to make Gr Si hybrid nanophotonic-­integrated devices can just make up for the shortcomings of Si photonics in active devices and also bring many benefits, such as ultrawide optical bandwidth, high-­speed operation, low power consumption, and CMOS process compatibility. On the other hand, it is necessary to use Gr in optical integration system. Gr has a strong interaction with light. Many properties of Gr have been studied by free-­space light system. However, due to the two-­dimensional characteristics of Gr, the interaction distance between free space light and Gr is only one atom thin-­ layer thick. In practical application, such a light response effect is far from enough. As shown in Figure 6.3a, the absorptivity of SLG for vertical incident space light (visible light to infrared light) is only about 2.3%, and the remaining 97.7% of the light is almost transmitted from Gr, which directly leads to the low external quantum efficiency of Gr photodetectors made in this way even when the internal quantum efficiency can be very high [16]. Therefore, to give full play to the potential of Gr in optoelectronic devices and to further study and characterize the basic optical properties of Gr, it is necessary to further enhance the interaction between Gr and light. One method is photon integration, as shown in Figure 6.3b,c. After Gr is integrated into the waveguide, complete light absorption can be achieved as long as the waveguide is long enough. Therefore, Gr is an excellent choice to further improve the performance of Si-­based active optical devices, reduce the production cost and difficulty, and assist Si in realizing the monolithic integration of all-­Si photonic-­ integrated devices. Based on the Si photonics-­Gr hybrid platform, people have realized the applications of the photodetector, an electro-­optic modulator. These aspects are described below.

6.2 ­Hybrid Waveguide Photodetecto

y

~0% z (a)

97.7%

(b)

x

(c)

Figure 6.3  (a) The absorption rate of Gr to vertical incident light is only 2.3%. (b) After Gr is integrated into the waveguide, complete light absorption can be achieved as long as the waveguide is long enough. (c) Simulation result of the fundamental TE mode in an Si waveguide with a layer of Gr on top. Source: Reproduced with permission from Li et al. [16]/ AIP Publishing.

6.2  ­Hybrid Waveguide Photodetector IBM developed the world’s first Gr photodetector for space light in 2009 [17]. This device works based on the photovoltaic effect in the contact area between Gr and metal. However, its maximum response is only 0.5 mA W−1 due to the limitation of Gr absorption. To improve the response, people have adopted the scheme of combining Gr with microcavity resonance  [18] or surface plasmon resonance  [19]. Although these schemes can greatly improve the response of the device, they sacrifice the broadband width of Gr. Integrating quantum dots on Gr  [20] can also improve the response, but the working speed is not high. Because Gr is only the thickness of the monatomic layer, laying Gr on the optical waveguide can prolong the action distance between Gr and light and improve the absorption of Gr. In addition, there is almost no mode mismatch between silica Gr hybrid waveguide and pure Si waveguide, and there is no mode loss from the passive region to the active region. Based on this idea, in 2013, Gan et  al.  [21] (as shown in Figure  6.4a), Pospischil et al. [22] (as shown in Figure 6.5a), and Wang et al. [23] (as shown in Figure  6.6a) independently reported their waveguide-­integrated Gr photodetector for the first time. Gan et al. at Columbia University showed Gr waveguide-­integrated photodetector with the asymmetric position of metal contact electrode relative to waveguide [21]. As shown in Figure  6.4a, the response rate is significantly improved. The device accurately transfers the mechanically stripped double-­layer Gr to the Si waveguide structure through the wet transfer technology. On the one hand, by coupling the Gr with the evanescent field on the surface of the Si waveguide in the 53-­μm device, the optical absorptivity of more than 60% is achieved in the coupling length. On the other hand, by placing the electrodes asymmetrically on both sides of the waveguide, the transverse metal-­doped junction is formed using the difference in work function between Gr and metal, so that the device can separate the photogenerated electron–hole pair and generate photocurrent under zero bias voltage, see Figure  6.4b. A high-­performance Gr detector without external bias is realized,

111

6  Graphene Silicon-­integrated Waveguide Devices 0

+ G

–

Au

Au

Au

Graphene

EF

Au Si

Relative response (dB)

A

Au

–2 –4 –6

12 Gbit s

–1

–8 60 ps

SiO2 Si

(a)

–10

Graphene

(b)

λ = 1550 nm

100 Frequency (GHz)

1

10

(c)

Figure 6.4  (a) Si waveguide-­integrated Gr photodetector with an asymmetric metal electrode. (b) Potential profile across the Gr channel, showing band bending around the two metal electrodes. (c) Dynamic optoelectrical response of the Gr photodetector. Source: Reproduced with permission from Gan et al. [21]/Springer Nature.

SiO2 Graphite S

2500

S Si

GND

Signal out

L Waveguide

4 αM

1500

y x

1000

Bilayer αG

500

Monolayer

0

1 6 4

Pin

2

6 50

50 µm

100

150

200

250

300

350 –25

W (nm)

(a)

lPC

2

0.1 GND

Graphene

W

2000 Photocurrent (μA)

GND GND

Absorption (cm–1)

112

(b)

(c)

–20

–15

–10

Optical power (dBm)

Figure 6.5  (a) Scanning electron micrograph of the waveguide-­integrated Gr photodetector. (b) Comparison of the losses in the metallic electrode and the Gr absorption as a function of electrode width. (c) Performance characteristics of a bilayer Gr photodetector. Source: Reproduced with permission from Pospischil et al. [22]/ Springer Nature.

which maintains a flat high response rate (0.1 A W−1) in the wavelength range of 1450–1590 nm; only 1-­dB signal attenuation occurs in 20-­GHz high-­frequency measurement and presents a clear eye diagram on the 12-­GHz data communication link, as shown in Figure 6.4c. To enhance the interaction between Gr and light, it is a better method to enhance the light absorption of Gr by metal plasma structure with a very strong localized field. Pospischil et al. at Vienna University of Technology reported a metal plasma-­ enhanced Gr Si waveguide-­integrated photodetector  [22], which placed a metal electrode antenna in the center of the waveguide, as shown in Figure  6.5a. The device is fabricated through a simple three-­step process: etching and passivation of Si waveguide, deposition and patterning of Gr, and preparation of metal electrode. The device uses the metal electrode antenna structure to enhance the light absorption of Gr, and the photogenerated carriers can be effectively separated by the built­in electric field at the metal–Gr interface, as shown in Figure  6.5b. It shows the comparison of the losses in the metallic electrode and the Gr absorption. The response rate of Gr device is about 30 mA W−1, while the response rate of the double-­ layer Gr can be increased to 50 mA W−1. The device has a 3-­dB bandwidth of 18 GHz and can achieve ultrawideband response in an all-­optical communication band,

6.2 ­Hybrid Waveguide Photodetecto 0.03

0.0

@1.55 µm excitation

–0.1

25.4 mW

Current (µA)

Current (µA)

29.8 mW

0.02

18.4 mW 14.5 mW

0.01

5.4 mW Dark

0.00

–0.2

Dark 0.31 µW

–0.3

0.44 µW 0.61 µW

–0.4

0.73 µW

–0.5

0.94 µW @2.75 µm excitation

–0.6 –5

(a)

(b)

0 Bias voltage (V)

5

–1.5

–1.0

0.0 –0.5 0.5 Bias voltage (V)

1.0

(c)

Figure 6.6  (a) Schematic of Gr/Si-­heterostructure waveguide photodetector. Dark and illuminated current versus bias voltage under different incident light powers at (b) 1550 nm and (c) 2750 nm, respectively. Source: Reproduced with permission from Wang et al. [23]/ Springer Nature.

which is far beyond the reported response wavelength range of the strained germanium photodetector (Figure 6.5c). An infrared detector is a photosensitive device that can convert invisible infrared radiation into a measurable signal. It is widely used in the military, meteorology, industry, environmental science, medical diagnosis, and other fields. Currently, most commercial infrared detectors are based on intrinsic structures such as HgCdTe and InSb. It has significant disadvantages such as large power consumption, large volume, and high price. Wang et  al.  [23] of the Chinese University of Hong Kong developed a high-­response infrared detector with waveguide structure using Gr/Si heterostructure, as shown in Figure  6.6a. The dark current of Gr-­PD prepared by metal–Gr increases with the increase in the external field voltage, so it is impossible to realize high-­sensitivity signal detection. Gr/Si heterostructure is used in this book. Due to the existence of a junction potential barrier, the dark current is very small, so it has high sensitivity. The PD prepared by Wang et al. realized the measurement of IR at room temperature, and 1.5 V bias voltage. 2.75 μm light wave has a response rate of 0.13 A W−1, as shown in Figure  6.6c. Meanwhile, the photodetector was also characterized in the near-­infrared experiment. The responsivity was about 10−4 A W−1, see Figure 6.6b. In recent years, Si-­based photoelectrons are moving toward 2 μm and longer band (midinfrared) to meet the major needs of optical communication and advanced optical sensing in the future. Recently, Guo et al. of Zhejiang University [24] proposed a new Si Gr metal hybrid plasmon waveguide structure (as shown in Figure 6.7a), innovatively introduced a special ultrathin and ultrawide Si ridge core design, significantly enhanced the light absorption efficiency of Gr (see Figure 6.7b), effectively reduced the metal absorption loss, and realized that the length of the active region is only 20 μm high-­response high-­speed Gr/Si-­based waveguide photodetector of 1.55 μm-­band and 2 μm. The 3-­dB bandwidth of 1.55/2 μm-­band are >40 GHz and >20 GHz, respectively (both limited by the bandwidth of the test equipment; Figure 6.7c shows the measured frequency response for the fabricated device operating at 2 μm), and its responsivity are 400 and 70 mA W−1, respectively. The detector can be applied to 1.55/2 μm band optical communication and optical sensing.

113

6  Graphene Silicon-­integrated Waveguide Devices

Z

Y

Si

Graphene

Y X

Signal

Au

X

Ground

AI2O3

V m–1 107 106 105

Gate

Ground AI2O3 on Si Graphene SiO2

(a)

(b) 10

Normalized response (dB)

114

0 –10 –20 –30 –40 1.5

(c)

10

20

Frequency (GHz)

Figure 6.7  (a) Schematic configuration of the Si–Gr hybrid plasmonic waveguide photodetector. (b) The electric field distribution of the quasi-­TE0 mode. (c) Measured frequency response operating at 2 μm. Source: Reproduced with permission from Guo et al. [24]/John Wiley and Sons.

6.3  ­Hybrid Waveguide Modulator Optical modulator plays an important role in optical signal modulation in optical fiber communication. It is a device that regulates the absorption index, refractive index, phase, or amplitude of output light through the change in voltage or electric field. At present, the size of the Si-­based electro-­optical modulators is large because of their weak electro-­optical characteristics. Germanium is difficult to integrate with other compound semiconductor modulators on Si substrate; their spectral modulation range is usually narrow. Gr has advantages in light modulators due to its zero-­bandgap ability to absorb light over a wide frequency range and its high mobility. Depending on the properties of Gr, three methods can be used to achieve the modulation function, including gate voltage modulation, optical excitation, and temperature modulation [25]. Among them, the gate voltage modulation and optical excitation methods are used to change the carrier concentration of Gr, which in turn changes the real and imaginary parts of the refractive index to achieve amplitude modulation and phase modulation. As for temperature

6.3 ­Hybrid Waveguide

Modulato

modulation, the main purpose is to change the temperature of the waveguide or substrate by heating the Gr, which in turn modulates the refractive index of the whole structure. The following is a brief description of electro-­optic and thermo-­ optic modulation.

6.3.1  Electro-­optical Modulator Gr-­based electro-­optical modulators are mainly used to change the carrier concentration of Gr by adding a voltage to the capacitive structure of Gr to adjust the Fermi energy level, which in turn leads to changes in the real and imaginary refractive indices of Gr and thus can be made into electroabsorption-­ [26, 27] and phase-­type modulators [28]. In 2011, Liu et al. of the University of California at Berkeley demonstrated the modulator prototype device for the first time [26]. Its three-­dimensional structure and light-­field distribution are shown in Figure 6.8a. A single layer of Gr is laid on the Si waveguide, separated by alumina. The waveguide is doped and connected to the electrode through a thin layer of Si. It controls the absorption and transmission of light by controlling the height of the Fermi level by applying a bias. As shown in Figure 6.8b, there can be light absorption only when the Fermi level is between the threshold positive and negative hv0/2, and transmission in other cases. In this way, the modulator with Gr as the absorption medium can reach 0.1 dB μm−1, and the modulation spectral range can reach 1.35–1.60 μm. The modulation speed can reach about 1 GHz, as shown in Figure  6.8c, and only 25 μm2 of the active area. Furthermore, they proposed a waveguide light modulator using double-­layer Gr as the active region, which adopts a structure similar to p-­insulating layer-­N and directly uses Gr to replace Si to form contact with the metal electrode, as shown in Figure 6.8d [29]. This cannot only increase the light absorption but also reduce the

Si

Pt

–90 –96 –102

Response (dBm)

Au

Response (dBm)

–84 Graphene

–80 –2.0 V –2.5 V –3.0 V –3.5 V

–90

–100

(a)

(c)

–5

–108

–4 –3 –2 –1 Drive voltage (V)

108 Graphene Graphene

Si

Au

Si

SiO2

(b)

109 Graphene Al2O3

Al2O3

Al2O3 Pt

Frequency (Hz)

Si

Au

SiO2

(d)

Figure 6.8  (a) Schematic configuration of the Gr-­based waveguide-­integrated optical modulator. (b) Cross section of the device with the optical mode. (c) Dynamic electro-­optical response of the device. (d) Schematic illustration of the double layer Gr modulator. Source: Reproduced with permission from Liu et al. [29]/American Chemical Society.

115

6  Graphene Silicon-­integrated Waveguide Devices

insertion loss, and the modulation depth of 0.16 dB μm−1 is obtained, while the response bandwidth is about 1 GHz. To improve the extinction ratio and modulation speed of the electro-­optical modulator, Lipson’s group implemented a high-­performance electro-­optical modulator using a microring resonator in 2015 [27], which can achieve 15 dB of extinction ratio variation in the 10-­V bias range, and the 3-­dB bandwidth of the device can be transported to 30 GHz due to the increased thickness of the isolation layer and the reduced capacitance of the structure, as shown in Figure 6.9a–d. However, the optical bandwidth of the structure is only 0.1 nm, and to achieve a large electrical bandwidth, the device also requires a very high drive voltage. From the process point of view, to ensure the flatness of the Gr, most of the modulators reported so far use the method of depositing the oxide and smoothing it with CMP and then transferring the Gr to the waveguide, which is a complicated process. On the other hand, the currently reported waveguide-­based Gr modulators mainly work in the 1.55-­μm band, and Gr modulators for the midinfrared band need further research. Waveguide Graphene/graphene capacitor

Ring Metallization

Waveguide

(a)

(b)

–5 –10 –15 –20

15 dB over 10 V

–25 1569

(c)

1570 Wavelength (nm)

0V –10 V –20 V –30 V –40 V –50 V 1571

Normalized electro-optic S21 (dB)

3

0

Transmission (dB)

116

0

–3

–6

0.1

1 Frequency (GHz)

10

30

(d)

Figure 6.9  (a) Schematic configuration of the Gr/Gr capacitor integrated along a ring resonator. (b) Optical micrograph showing bus waveguide, Ti/Pd/Au metallization, and ring resonator. (c) Transmission spectra for various applied voltages. (d) Electro-­optic frequency response. Source: Reproduced with permission from Phare et al. [27]/ Springer Nature.

Y X

Metal ene Graph ater e nanoh

Si

Si

Metal ene Graph

SiO 2

BOX

arm

layer

(a)

Electrical voltage (V)

Z

Pheating (mW)

6.4 ­Challenges and Prospective 1.5 0.0 0.08 0.04 0.00 –0.04 –0.08 0.0 0.2 0.4 0.6 0.8 1.0 1.2 Time (ms)

1.4 1.6 1.8 2.0

(b)

Figure 6.10  (a) Schematic configuration of the Gr thermo-­optic modulator based on microdisk. (b) The measured temporal responses of the modulated heating power and the corresponding output signal of the photodetector. Source: Reproduced with permission from Yu et al. [31]/Optica Publishing Group.

6.3.2  Thermo-­optic Modulator Gr has a super high thermal conductivity and can achieve effective thermo-­optical modulation. On the one hand, Gr can be used as a good transparent thermal conductor to transfer the heat generated by the metal to the Si-­based optical device to achieve the tuning of wavelength [30]. However, a part of the heat is lost during the heat-­transfer process, resulting in a relatively low thermal tuning efficiency of the device in this structure. On the other hand, Gr can be used as a micro-­ and nanoheater to generate heat by passing electricity directly on Gr to improve the efficiency of heat utilization and achieve thermal tuning [31]. Figure 6.10a,b shows the measured output signal from the photodetector. Compared with the conventional metal-­ heating electrode, this structure further improves the heating efficiency as Gr does not require an intermediate SiO2 isolation layer, which has lower losses and does not cause the mismatch of the mode field. However, its optical bandwidth is narrower due to the resonance effect, and also the insertion loss of the device needs to be further reduced.

6.4  ­Challenges and Prospectives Gr has a special energy band structure (zero bandgap and zero effective mass), good electron transport characteristics, strong light absorption ability, and so on. Due to its wide absorption spectrum and high mobility, the application of Gr in photodetectors and modulators is gradually being discovered and deeply studied. It is believed that Gr will play a vital role in the realization of Si-­based all-­optical interconnection in the near future. Although a series of progress has been made in Si Gr hybrid-­integrated optical detection and optical modulation devices, there are still a series of problems.

117

118

6  Graphene Silicon-­integrated Waveguide Devices

For example, the large dark current limits the detection sensitivity and device power consumption of the Gr detector; Gr devices prepared by mechanical stripping have the disadvantages of low yield and low repeatability and cannot be used commercially. Therefore, it is necessary to further improve the photonic integration process of Gr and Si and improve the performance of Gr detector. For waveguide Gr photodetectors, although researchers have realized devices with a high response, high speed, and large absorption bandwidth because Gr has no bandgap, these devices usually have high dark current, poor signal-­to-­noise ratio and detection sensitivity ratio, and high energy consumption. For the waveguide Gr electro-­optic modulator, although there have been some achievements in the current research, on the one hand, people need to continue to optimize the capacitor resistance structure to improve the bandwidth, and on the other hand, they can study the higher order modulation formats of Gr modulator, such as PAM4 and QPSK to achieve the modulation speed of 100 Gbps. The future research of Si photonic-­integrated detectors based on two-­dimensional materials can focus on the following aspects: (i) Fabrication of large-­scale two-­ dimensional material Si photonic-­integrated devices. Currently, the two-­dimensional materials in most two-­dimensional photodetectors integrated with Si photonic structure are prepared by the mechanical stripping method. To enable two-­ dimensional materials to be used in the manufacture of large-­scale Si photonic-­ integrated devices and realize commercial applications as soon as possible, researchers should actively explore the growth of two-­dimensional materials or two-­dimensional material heterojunctions to prepare two-­dimensional materials with large area, high quality, and low cost. (ii) Emerging two-­dimensional materials are integrated with Si photons. Emerging two-­dimensional materials such as indium selenide, Gr, and perovskite nanosheets have many excellent characteristics, including appropriate bandgap, good photoelectric conversion efficiency at high power, and excellent light absorption energy. These new two-­dimensional materials are expected to be applied in the field of Si photonic-­integrated photoelectric detection. (iii) Optimization of device interface and contact resistance. For the disadvantage of the large dark current of Gr detector, a proven solution is to combine Gr with other two-­dimensional materials. For example, TMDs are stacked vertically to form van der Waals heterojunction [32].

­References 1 Urino, Y., Nakamura, T., and Arakawa, Y. (2016). Silicon optical interposers for high-­density optical interconnects. In: Silicon Photonics III (ed. L. Pavesi and D.J. Lockwood), 1–39. Berlin/Heidelberg: Springer. 2 Pavesi, L. and Lockwood, D.J. (2016). Silicon Photonics III, Topics in Applied Physics, vol. 122, V–VII. Berlin/Heidelberg: Springer. 3 Rumley, S., Bahadori, M., Polster, R. et al. (2017). Optical interconnects for extreme scale computing systems. Parallel Computing 64: 65–80.

  ­Reference

4 Mengyuan, H., Pengfei, C., Liangbo, W. et al. (eds.) (2014). Development of Si photonics technology: Ge/Si avalanche photodiode for PON applications. OFC 2014, San Francisco, California United States (9–13 March 2014). 5 Ho, R., Mai, K.W., and Horowitz, M.A. (2001). The future of wires. Proceedings of the IEEE 89 (4): 490–504. 6 Fey, D. (2001). Architectures and technologies for an optoelectronic VLSI. Optik 112 (7): 274–282. 7 Lee, K.K., Lim, D.R., Luan, H.C. et al. (2000). Effect of size and roughness on light transmission in a Si/SiO2 waveguide: experiments and model. Applied Physics Letters 77 (14): 2258. 8 Lee, B.G., Chen, X.G., Biberman, A. et al. (2008). Ultrahigh-­bandwidth silicon photonic nanowire waveguides for on-­chip networks. IEEE Photonics Technology Letters 20 (5–8): 398–400. 9 Osgood, R.M., Panoiu, N.C., Dadap, J.I. et al. (2009). Engineering nonlinearities in nanoscale optical systems: physics and applications in dispersion-­engineered silicon nanophotonic wires. Advances in Optics and Photonics 1 (1): 162–235. 10 Orcutt, J.S., Khilo, A., Holzwarth, C.W. et al. (2011). Nanophotonic integration in state-­of-­the-­art CMOS foundries. Optics Express 19 (3): 2335–2346. 11 You, J., Lavdas, S., and Panoiu, N.C. (2016). Theoretical comparative analysis of BER in multi-­channel systems with strip and photonic crystal silicon waveguides. IEEE Journal on Selected Topics in Quantum Electronics 22 (2): 63–72. 12 Liang, D. and Bowers, J.E. (2010). Recent progress in lasers on silicon. Nature Photonics 4 (8): 511–517. 13 Reed, G.T., Mashanovich, G., Gardes, F.Y., and Thomson, D.J. (2010). Silicon optical modulators. Nature Photonics 4 (8): 518–526. 14 Michel, J., Liu, J.F., and Kimerling, L.C. (2010). High-­performance Ge-­on-­Si photodetectors. Nature Photonics 4 (8): 527–534. 15 Sheng, Z., Liu, L., Brouckaert, J. et al. (2010). InGaAs PIN photodetectors integrated on silicon-­on-­insulator waveguides. Optics Express 18 (2): 1756–1761. 16 Li, H., Anugrah, Y., Koester, S.J., and Li, M. (2012). Optical absorption in graphene integrated on silicon waveguides. Applied Physics Letters 101 (11): 111110. 17 Xia, F.N., Mueller, T., Lin, Y.M. et al. (2009). Ultrafast graphene photodetector. Nature Nanotechnology 4 (12): 839–843. 18 Furchi, M., Urich, A., Pospischil, A. et al. (2012). Microcavity-­integrated graphene photodetector. Nano Letters 12 (6): 2773–2777. 19 Echtermeyer, T.J., Britnell, L., Jasnos, P.K. et al. (2011). Strong plasmonic enhancement of photovoltage in graphene. Nature Communications 2: 1–5. 20 Konstantatos, G., Badioli, M., Gaudreau, L. et al. (2012). Hybrid graphene-­quantum dot phototransistors with ultrahigh gain. Nature Nanotechnology 7 (6): 363–368. 21 Gan, X.T., Shiue, R.J., Gao, Y.D. et al. (2013). Chip-­integrated ultrafast graphene photodetector with high responsivity. Nature Photonics 7 (11): 883–887. 22 Pospischil, A., Humer, M., Furchi, M.M. et al. (2013). CMOS-­compatible graphene photodetector covering all optical communication bands. Nature Photonics 7 (11): 892–896.

119

120

6  Graphene Silicon-­integrated Waveguide Devices

23 Wang, X.M., Cheng, Z.Z., Xu, K. et al. (2013). High-­responsivity graphene/ silicon-­heterostructure waveguide photodetectors. Nature Photonics 7 (11): 888–891. 24 Guo, J.S., Ye, C.C., Liu, C.Y. et al. (2020). Ultra-­compact and ultra-­broadband guided-­mode exchangers on silicon. Laser & Photonics Reviews 14 (7): 2000058. 25 Yu, S.L., Wu, X.Q., Wang, Y.P. et al. (2017). 2D materials for optical modulation: challenges and opportunities. Advanced Materials 29 (14): 1606128. 26 Liu, M., Yin, X.B., Ulin-­Avila, E. et al. (2011). A graphene-­based broadband optical modulator. Nature 474 (7349): 64–67. 27 Phare, C.T., Lee, Y.H.D., Cardenas, J., and Lipson, M. (2015). Graphene electro-­ optic modulator with 30 GHz bandwidth. Nature Photonics 9 (8): 511–518. 28 Sorianello, V., Midrio, M., Contestabile, G. et al. (2018). Graphene-­silicon phase modulators with gigahertz bandwidth. Nature Photonics 12 (1): 40–48. 29 Liu, M., Yin, X.B., and Zhang, X. (2012). Double-­layer graphene optical modulator. Nano Letters 12 (3): 1482–1485. 30 Yu, L.H., Dai, D.X., and He, S.L. (2014). Graphene-­based transparent flexible heat conductor for thermally tuning nanophotonic integrated devices. Applied Physics Letters 105 (25): 251104. 31 Yu, L.H., Yin, Y.L., Shi, Y.C. et al. (2016). Thermally tunable silicon photonic microdisk resonator with transparent graphene nanoheaters. Optica 3 (2): 159–166. 32 Flory, N., Ma, P., Salamin, Y. et al. (2020). Waveguide-­integrated van der Waals heterostructure photodetector at telecom wavelengths with high speed and high responsivity. Nature Nanotechnology 15 (2): 118–124.

121

7 Graphene for Silicon Image Sensor ­7.1  Introduction The modern Si image sensors technology is mainly based on charge-­coupled devices (CCDs) and complementary metal–oxide–semiconductor image sensor (CIS), which have dominated the market for decades. The first CCD was proposed by W.S. Boyle and G.E. Smith in Bell’s Laboratories in 1969 [1]. Since then, researchers have developed various types of high-­performance CCD image sensors based on different device structures, which have been widely used in astronomical observation, spectral analysis, and military applications. Recently, the CISs have regained their vitality due to the innovation in manufacturing processes for integrated circuits, while their performance can be compared with ordinary CCDs. These two technologies are widely used in high-­end applications and commercial products due to their higher performance and greater pixel density [2]. CCD and CIS sensors have many limitations along with their advantages. For example, CCD has the characteristics of high fill factor (FF), high linearity, high sensitivity, and low noise  [3]. However, to ensure a high charge-­transfer efficiency (CTE), CCD requires a highly specialized manufacturing process. This hinders the monolithic integration of CCS with complementary metal–oxide– semiconductor (CMOS) devices. The serial transfer also requires a relatively complex multiphase bias clock, which may limit the speed and consume large power [4]. The CIS sensor has the advantages of a simple and fast readout circuit with low cost but usually has problems such as low FF and high noise [5]. CCD plays an important role in the field of detection with its advantages, such as high sensitivity, low noise, and high linearity. The CIS is widely used in the field of consumer electronics with its benefits of low power consumption, high speed, and high integration. Despite many achievements in CCD and CMOS, the Si-­based image sensors still have difficulty working beyond the visible-­light range. The Si has an

Graphene for Post-Moore Silicon Optoelectronics, First Edition. Yang Xu, Khurram Shehzad, Srikrishna Chanakya Bodepudi, Ali Imran, and Bin Yu. © 2023 WILEY-VCH GmbH. Published 2023 by WILEY-VCH GmbH.

122

7  Graphene for Silicon Image Sensor

indirect bandgap of 1.12 eV and is optically transparent for wavelengths above 1.1 μm. Hence, it is not suitable for infrared photodetection. The response spectrum of Si is limited to the visible-­light range due to the basic absorption limitation. Infrared light detection has always been a challenge for Si-­based image sensors. Researchers have made many efforts to improve the spectral performance, such as nonmonolithic integration with metal silicide Schottky barrier detectors [6]. However, these methods usually complicate the device structure and manufacturing process, which is restricted by low-­pixel FF, high cost, and low-­temperature operation. Integrating the other narrow bandgap III–V or II–VI semiconductors may help extend the spectral response range. However, growth of III–V or II–VI on Si is challenging due to their different crystal structures. Also, in these non-­Si narrowband semiconductors, thermal noise is a serious challenge. Two-­dimensional (2D) materials can be transferred to any substrate and have a large bandgap selection range. They have great potential in CMOS monolithic integration and can obtain a broadband response from ultraviolet to far-­infrared [7]. Gr is a promising material for various optoelectronic applications due to its unique optical and electrical properties. Gr has a unique energy band structure, ultrahigh carrier mobility, and broad-­spectrum light absorption extending from visible light to the terahertz range. Another important feature of Gr is its stability in various environments. It is easy to integrate with multiple photonic device structures and can be integrated into the back end of the CMOS process production line  [8]. Focusing on the challenges of new devices in the post-­Moore era and breaking the traditional limits of Si-­based optoelectronic devices, the achievement of high-­ performance broadband detection is the key to the development of sensing information technology. 2D materials represented by Gr have a unique band structure and excellent optoelectronic properties, which are suitable for broadband image sensors. Combining the characteristics of Gr broad-­spectrum absorption, strong field-­effect amplification, ultrafast electron–electron scattering with the advantages of low noise and low cost of Si-­based technology, a new type of photodetector based on the Gr/Si system is expected to achieve a breakthrough in the field of image sensors [9]. In this chapter, we provide an overview of the Gr/Si-­based image sensors for various applications. Section 7.2 describes a quantum dot-­sensitized Gr image sen­ sor that responds in a broad wavelength range and could also be integrated with Si CMOS technology, which could potentially develop into a new cost-­effective chip platform for hyperspectral imaging and spectroscopy. Section 7.3 describes a Gr-­ based thermal imaging system with improved detectivity and noise equivalent temperature difference. Section  7.4 focuses on the terahertz detectors based on antenna-­coupled Gr field-­effect transistors with realistic settings, large area, and fast imaging of macroscopic samples. Section 7.5 explains about the Gr device applications for soft bioelectronics devices. Section 7.6 demonstrates the reconfigurable

­7.  Quantum Dot-­based Infrared Graphene Image Sensor

2D materials image sensor that can constitute the artificial neural networks (ANNs), which can simultaneously sense and process optical images. The details of the working principle of CCDs are provided in Section 7.7. Position-­dependent photodetectors based on graphene are described in Section  7.8. The conclusion and future prospective of novel Gr CCD and CMOS imaging devices are presented in Section 7.9.

­ .2  Quantum Dot-­based Infrared Graphene 7 Image Sensor The rapid development of Si and CMOS technology during the past 40 years has enabled the electronics market to fabricate many devices such as computers, smartphones, and compact and low-­cost digital cameras. However, due to the difficulty of combining other semiconductors with CMOS, the application of nonvisible light cameras has been hampered. The advantage of 2D materials is that they can be transferred to almost any substrate. The 2D materials are suitable for monolithic integration with CMOS-­based Si-­integrated circuits. These monolithic integrations are much improved with the development of wafer-­level chemical vapor deposition growth and transfer methods. The European Union launched the Gr flagship program in 2013 to accelerate the technological research on Gr and related materials. Recently, a hybrid Gr quantum dot (GQD) photodetector array, which is vertically integrated into a CMOS readout chip, has been demonstrated. The researchers demonstrated the monolithic integration of Gr and CMOS-­integrated circuits  [10]. The integration potential for such devices was demonstrated by fabricating an image sensor with a 388 × 288 array of GQD photodetectors, which are operated as a digital camera with high sensitivity for both visible and shortwave infrared light. Almost 110 000 Gr photoconductive channels are individually integrated vertically to connect with each electronic component of a CMOS readout-­ integrated circuit (ROIC). The circuitry in the chip is similar to those used for commercial image sensors in digital cameras, which are commonly used in smartphones. However, it operates for both visible and short-­wave infrared light (300–2000 nm). Pixels of the image sensor can be characterized by a high gain of 107 and a responsivity above 107 A W−1. The large size of pixels (20 μm) can limit the spatial resolution for operation in the visible and IR range compared to the present commercial CMOS images operated in the visible region with a pixel size close to 1 μm. The monolithic CIS fabrication is a milestone for low-­cost and high-­resolution broadband and hyperspectral imaging systems, which can be applied to security systems, smartphone cameras, night vision, automotive sensor systems, pharmaceutical inspection, and environmental monitoring (Figure 7.1).

123

124

7  Graphene for Silicon Image Sensor

VIS and NIR (400–1000 nm) Column select Row select Rcomp Rblind

CTIA VSS VREF

1

Colloidal quantum dots CVD graphene: 2 –1 –1 mobility > 1,000 cm V s VCO

VOUT Output driver

CDS

Vertical interconnects CMOS read out circuit

(a)

288

1

388

(d)

(b)

VIS, NIR, and SWIR (400–1850 nm) 1 Column DAC Bias circuitry and row select

Active area Column amplifiers and sampling stages

388 pixels

17

m

m

288 pixels

Digital control and interface

Column multiplexer Analog output

288

Correlated double sampling

(c)

(e)

1

388

Figure 7.1  (a) Schematics of Gr/QD photoconductor of a readout circuit. (b) 3D impression of the S-­shaped Gr channel, QD layer on top, and CMOS readout circuits. (c) Photograph of the image sensor indicating the functionality for each area. (d, e) Visible and near-­infrared photographs of a standard image reference “Lena” and box of apple at mentioned wavelengths. CTIA, capacitor transimpedance amplifier; CDS, correlated double sampling; DAC, digital to analog converter; CVD, chemical vapour deposition. Source: Reproduced with permission from Goossens et al. [11]/Springer Nature.

­7.3  Graphene Thermopile Image Sensor The MIT researchers applied the unique tunable Seebeck coefficient of Gr to demonstrate a Gr-­based thermal imaging system [11]. The devices achieved room temperature responsivities of ∼7–9 V W−1 (λ = 10.6 μm), with a time constant of ∼23 ms through the integration of Gr-­based photothermoelectric detectors with micro-­machined SiN membranes. The large responsivities resulting from thermal isolation and broadband IR absorption of the underlying SiN membrane enable the detection and stand-­off imaging of an incoherent blackbody target (300–500 K). The performances can be further improved by improving the carrier mobility of Gr which is related to two main figures of merit of IR detectors, i.e. detectivity and noise equivalent temperature difference. Using a series of motorized XY translation stages, the Gr thermopile performs thermal imaging of a blackbody source as the image of the measured magnitude of the photovoltage response by analyzing the scan position. Furthermore, combining the back-­end-­of-­the-­line (BEOL)-­compatible processes with CMOS technology, the future Gr-­thermopile-­based focal plane arrays (FPAs) can be very promising (Figure 7.2).

­7.  Graphene THz Image Sensor (1) Gate metal

(2) Etch bias

SiO2/SiN/SiO2

Ti/Pt

Si (7) XeF2 release

(3) Gate dielectric

Release bias

Gate dielectric

Si

Si

(6) Passivation

(5) Ohmic metal/mesa

PMMA

XeF2

Si

Si

(4) Graphene transfer

Ti/Au

CVD graphene

Si

Si

(a) Graphene

5

0.2

10

0.15

15

0.1

20

0.05

Infrared absorber (SiN/SiO2)

0

25 0

(c)

VPH (μV)

M1

Thermal isolation legs

(b)

0.25 T0 = 472 K

y (mm)

G1

0

M2 G2

5

10

15

20

25

X (mm)

Figure 7.2  (a) Schematic of Gr thermopile fabrication and process flow. (b) Schematic of Gr thermopile. (c) Magnitude of the measured signal as a function of the actuator position. PMMA, poly(methyl methacrylate). Source: Reproduced with permission from Hsu et al. [12]/ American Chemical Society.

­7.4  Graphene THz Image Sensor The 2D Dirac materials, such as Gr, provide excellent characteristics, i.e. high electron mobility, broadband optical absorption, high tunable Fermi level, bipolar carriers, and their nonlinear transport for high-­efficiency mixing detection. Due to its broadband absorption, Gr is a suitable material for developing detectors operating in the THz region. Researchers have demonstrated THz detectors based on antenna-­coupled Gr plasma-­wave field-­effect transistors, as shown in Figure 7.3a [12]. They exploit the nonlinear response to the oscillating radiation field at the gate electrode with contributions of thermoelectric and photoconductive origin. These detectors demonstrated a NEP ∼10−9 WHz−1/2 in the range of 0.29–0.38 THz. The Gr devices work at room temperature operation at 0.3 THz during the imaging process. A transmission image of a fresh leaf is presented in Figure 7.3b, which shows the capability to detect the leaf veins. This exhibits that the Gr THz devices are practical and can be applied in fast imaging of large area for macroscopic samples. The THz image consists of 200 × 550 scanned points collected by raster scan of the object in the beam focus, with an integration time of 20 ms/point.

125

7  Graphene for Silicon Image Sensor 90

3.00 × 10–5

80

2.61 × 10–4

70 4.92 × 10–4

60 Y-axis (mm)

126

S

Vout D

50

7.22 × 10–4

40

9.53 × 10–4

30

0.00118

20 G

0.00141

10 Vgate

(a)

0.00155

0

(b)

0

10

20

30

40

50

60

X-axis (mm)

Figure 7.3  (a) Schematics of the THz detection configuration in a FET embedding the optical image of the central area of a bilayer Gr-­based FET. (b) The 0.3-­THz images of a leaf. Gr THz detector not only images the sample but also reveals the leaf veins. Source: Reproduced with permission from Vicarelli et al. [13]/Springer Nature.

­7.5  Curved Image Sensor Array Soft bioelectronic devices have the potential for the emerging new generation of implantable devices, which minimize the impact of tissue damage and immune response. However, due to the large volume and rigidity of traditional imaging modules/materials, flexible implantable optoelectronic devices for optical sensing as well as retinal stimulation have not yet been developed. Researchers have introduced a method for preparing human eye-­excited soft optoelectronic devices using a high-­density MoS2/Gr-­curved image sensor array  [13]. Light intensity (W m–2) 0.2

Lens

0.1

0

CurvlS array Focused light

Concave hemisphere CurvlS array

5 mm

(a)

(b)

Figure 7.4  (a) Schematics of MoS2/Gr heterostructure-­based high-­density curved image sensor array. (b) Optical image of the high-­density curved image sensor array. The inset shows the image captured by the curved image sensor array. Source: Reproduced with permission from Choi et al. [14]/Springer Nature.

­7.  Neural Network Image Sensors

The ultrathin hemispherical curved image sensors can be used as effective imaging elements in soft retinal implants. This device exhibits infrared blindness and successfully acquires pixelated optical signals. Moreover, the retina shows few mechanical side effects upon applying programmed electrical stimulation to the optic nerve (Figure 7.4).

­7.6  Neural Network Image Sensors The image-­sensing process was classified into several steps in conventional imagers, where the analog signals of the image were collected with sensors and then transferred to the computer for processing after digital-­to-­analog conversion. However, the process was inefficient and time consuming compared to the process in which eyes transmit images to the brain. The processing speed of the visual image information can be greatly improved if the eyes can directly process images without bothering the brain. Machine vision technology in recent years follows the same process where a large amount of data is passed through the entire signal chain. However, it results in low frame rates and high power consumption. The researchers built an ANN directly on the chip and chose the tungsten diselenide (WSe2) as the photosensitive material (Figure 7.5a, b). The photodiode array consists of 27 detectors with good uniformity, adjustability, and linearity, arranged in a 3 × 3 imaging array [15]. The pixel size is about 17 × 17 μm2, and each pixel is composed of 3-­WSe2 subpixels whose response to light can be adjusted through the gate voltage. In other words, the semiconductor response to incident light can be tuned by changing the applied voltage, thereby altering the sensitivity of each diode. V1n –V1n

–VMn

R1n

P2

RM1 P2 RMn

l1n

nth pixel R12 RM2

P1

R1N

lMn

RM1 P2

A

(a)

R2N

PN

lM

RMN Encoder

(d)

y1

l2

y2

lM

yM

Autoencoder: R11 R21

R1N

A

Σ

l1

RMN

P1

M subpixels R 1N RMN

l1

Σ

R2N

PN

(c)

PN

R11 RM1

R11 R21

Σ

(b) P1

Classifier:

VMn

P

Σ

l1

Σ

l2

Σ

lM

W11

P′1

P′2

WNM

P′N

Decoder P' ≈ P

Figure 7.5  (a) Schematics of the artificial neural network photodiode array. Subpixels with the same color in the array are connected in parallel to generate IM. (b) Circuit diagram of a single pixel in the photodiode array. (c) Schematics of the classifier and (d) the autoencoder. Source: Reproduced with permission from Mennel et al. [16]/Springer Nature.

127

128

7  Graphene for Silicon Image Sensor

Unlike other networks, the weight of this system is not stored in the computer’s memory and hard disk but directly integrated into the image sensor. The weight of WSe2 devices can be modulated, which is equivalent to the neural network training. The diode’s sensitivity to light can be varied through different external bias voltages. Two neuromorphic functions can be performed using different algorithms for the neural network, which are classifiers and autoencoders (Figure 7.5c, d). The photodiode array and the chip perceptron work together with the nonlinear function outside the chip in the classifier. This type of neural network represents a supervised learning algorithm that can divide the input image into different output categories. The second type of neural network is an autoencoder, which can learn an effective representation of the input image during unsupervised training. It is used together with the decoder in which the image can be reproduced in its output after training the decoder. The encoder is composed of the photodiode array itself, and the decoder is composed of external electronic devices. In this process, the image transmission data is compressed.

­7.7  Graphene Charge-­coupled Device Image Sensor CCD [1] and CMOS are two main imaging technologies [3–5, 17, 18]. With enormous efforts toward higher performance and larger pixel density, both technologies keep competitive and are widely used in high-­end applications and commercial products. For CCD, the simple MOS photogate pixel and charge-­transfer structure give rise to high sensitivity, high FF, and low noise. But high CTE requires specialized fabrication process, which hinders the CMOS-­compatible monolithic integration  [5, 19]. The serial transfer also requires relatively complicated multiphase biasing/clocking, which could limit the readout speed and increase the power consumption [4]. For CMOS, independent pixel structure allows random access, simple clocking, high-­speed parallel readout, natural antiblooming, low power consumption, and high robustness [5]. CMOS also has the advantage of monolithic integration with multiple functionalities such as readout and data-­processing circuitry [3–5, 17]. On the other hand, CMOS usually has lower FF and higher noise than CCD due to the relatively complex active pixel sensor (APS) [5] structure and subsequent circuitry. Thus, imaging devices combining CCDs MOS photogate and CMOSs independent pixel structure can have significant advantages in integration, performance, and readout [20, 21]. A novel imaging device based on MOS photogate and field effect of 2D materials, termed the Gr-­charge-­coupled device (GCCD), has been demonstrated by Zhejiang University researchers  [22]. They reported a novel detecting/imaging device concept called two-­dimensional-­material charge-­coupled device (2D-­CCD), which is based on CCDs MOS photogate but requires no charge transfer between pixels (i.e. “couple”). The “couple” is redefined as the capacitive coupling [23, 24] between 2D materials (e.g. graphene) and semiconductors. The nondestructive and direct readout is made possible by the strong field effect of 2D materials. Besides, the

­7.  Graphene Charge-­coupled Device Image Sensor

2D-­CCD offers further advantages such as creation of a deeper potential well for large amount of charge integration, linear integration of charges and its monitoring by monitoring of charge in real time by 2D materials, amplification of charge signals by high mobility of 2D materials, and, most importantly, a broadband response from visible to short-­wavelength infrared (SWIR) range. Furthermore, the dark current is suppressed using 2D van der Waals (vdW) heterojunction as the 2D-­CCD pixel. Finally, the 2D-­CCD linear array can work in both conventional charge-­ transfer mode and unique random-­access mode. The random-­access mode allows lower blooming [3, 4, 21] and higher FF than the typical transfer mode. 2D-­CCD is based on the 2D-­material field-­effect transistor (FET) but requires a lowly doped semiconductor substrate. The demonstrated 2D-­CCD consists of a simple Gr/SiO2/n-­Si (GOS) capacitor. Figure  7.6a shows the schematics of 2D-­CCD pixel. As in the typical CCD, we created the potential well (deep depletion) in the 2D-­CCD by applying a fast-­sweeping or pulsed gate voltage Vg. The deep depletion was confirmed by the continuous decrease in the capacitance for Vg > 5 V in the high-­frequency capacitance–voltage (HF-­CV, 100 kHz) curve, when Vg was swept at relatively fast rate of 10 V s−1 from accumulation to inversion (Figure 7.6b). Holes from the photo-­ and thermally generated electron–hole (e–h) pairs are integrated into the well, while the electrons are transferred to graphene through the outer circuit. The photo–hole integration is also demonstrated by the increase in the capacitance with laser power. In order to examine the effect of deep depletion and charge integration on the transfer characteristics of graphene, we measured the drain current Id as a function of Vg of 2D-­CCD at the same voltage sweep rate (Figure 7.6c). A strong correspondence between HF-­CV and the large photoresponse in the transfer characteristics is observed for Vg > 5 V. The derivative of the transfer characteristics gives the quasi-­static capacitance–voltage (QS-­CV, Figure 7.7a), confirming that the variation in the drain current can exactly measure the charge variation in the GOS capacitor. This charge variation in graphene is due to the transfer of photoelectrons, as monitored simultaneously by the gate-­charging current Ig (Figure  7.7b). Thus, compared to the serial transfer and readout in the traditional CCD, direct and nondestructive readout in 2D-­CCD is made possible by the strong field effect of graphene. Vd A

1.0

Source – +

SiO2

Lowly

doped

0.8 t=0 Vg

C/COX

Drain

0.6 0.2

Si

0.0

(a)

(b)

t = t1

(c)

200 μm

1 nW 3 nW 5 nW 8 nW 19 nW

0.4

–5

5

15

25

Vg (V)

Figure 7.6  (a) Schematic of 2D-­CCD pixel. Scale bar 200 μm. (b) Working mechanism of 2D-­CCD. (c) The normalized HF-­CV (100 kHz) of the 2D-­CCD at different laser power (wavelength 532 nm).

129

7  Graphene for Silicon Image Sensor 0

(µA)

1.8

1200

0.6

1.2

800

0.4

0.6

0.0 0

10 Vg (V)

110

0.2

0 1 2 3 4 5

(a)

25

Dark 14 nW

600

20

5 15 Vg (V)

–5

1000

ld

C/COX

120 lg (nA)

(µA)

0.8

115

50 V 0V

Vg

125

ld

5

–5

15

25

–1.5 –1.0 –0.5 0.0 0.5 1.0

Vg (V)

6 8

1 nW 3 nW 5 nW

19 38 nW

t (s) 8 nW 19 nW

(b)

(c)

Figure 7.7  (a) The gate charging current Ig − Vg monitored simultaneously with Id − Vg. In the current–voltage measurement, Ig varies with power nonmonotonically due to the competition between the charging time and measurement delay. (b) The normalized QS-­CV at different power derived from Id − Vg. (c) Typical drain current waveform (Id − t) of the 2D-­CCD in dark and light (14 nW) conditions under periodic gate voltage pulses. The drain voltage Vd is 1.5 V (same for all Id − t measurements).

To demonstrate the practical operation of the CCD, a gate voltage pulse was applied to 2D-­CCD and the periodic drain current waveform (Id − t) was obtained in dark and light conditions, as shown in Figure 7.7c. The variation in the hole number in the well is monitored in real time by the variation in the Id − t. The 2D-­CCD integrates the holes in the pulse region (Vg = 40 V) and clears the holes in the reset region (Vg = 0 V).

Vd

– Vg(t) + lg(t) A

SiO2

CCD integration

EF(t0 + Δt) EF(t0)

MLG

Si

MLG

Vis

IR

ϕh

IR

–

n -Si

SLG SiO2

Field-effect readout

G

ld(t) A

SL

130

IR charge injection

Injection

(a)

(b)

𝑛(E)

Thermalization E

Si Native SiO2

(d)

MLG

(c)

(e)

(f)

Figure 7.8  (a) Schematic of graphene charge injection photodetector. Schematics show a deep depletion well where charges are integrated. Top Gr is used for readout, while MLG as the bottom is used for charge injection from IR absorption. (b) Band diagram of the GCI. (c) SEM image of the various size GCI pixels. (d, e) Enlarged image of the highlighted pixels. (f) Cross-­sectional TEM image of the MLG/Si Schottky junction. Scale bars in (c–f) are 100 μm, 5 μm, 5 μm, and 5 nm, respectively. SLG, single layer Gr. Source: Reproduced with permission from Liu et al. [22]/Springer Nature.

5 0 1

0

2 Time (s)

3

1

4

Normalized/Sch.pc

ld (µA)

8 7 6 0

1

2 Time (s)

0 20 40 mW mm–2 λ = 3 µm

3

4

10 30

1

VSch = 0.1 V

0

Eph = 0.67 eV

0 0

1

0

Normalized P

60 lg (nA)

Normalized/Sch.pc

Vg (V)

­7.  Graphene Charge-­coupled Device Image Sensor

1.33

0.41 eV, P2.77 1.78 0.55 eV, P 1.22 0.67 eV, P 1.13 0.80 eV, P

30

1 Normalized P 0.06 V, P 1.22 0.10 V, P 1.11 0.16 V, P 1.01 0.26 V, P

0

(a)

0

1

2 Time (s)

3

4

(b) 100

102

Vd

10–2

10–6

10–2

EQE/EQE0

EQE

GCI

SiO2 R (A W–1)

10–1 10–4

101

ld(t)

–

Vg(t) + A lg(t)

Si MLG

GOS

(c)

A

VSch

10–8

Charge injection (red) No injection (blue) 1

SLG

2

3

Wavelength (µm)

100

10–3 4

(d)

λ = 1850 nm 0

0.04 VSch (V)

0.08

0.12

Figure 7.9  (a) Time-­dependent Id, Ig − t curves of GCI at 5 V gate pulse and 3 μm laser illumination. (b) I–P curve of MLG/Si junction at constant bias and constant photon energy. (c) EQE of the devices at various wavelengths. (d) EQE enhancement of the GCI under 1850-­nm laser illumination as a function of injection bias. Source: Reproduced with permission from Liu et al. [22]/Springer Nature.

In the pulse region, the dark Id − t keeps almost constant, since the thermal generation requires relatively long time to fill the well (~50 seconds). The light Id − t first decreases (linear region) and then saturates (saturation region) within relatively short time (saturation time, Δtsat ~ 0.5 seconds), indicating that the well is filled by the photoholes. This large difference in the drain current between light and dark results from the integration effect and the signal amplification (photoconductive gain) from the graphene, both of which can give high sensitivity to the 2D-­CCD. 2D-­CCD only operates up to NIR light, due to a certain bandgap of Si (1.1 eV). To expand the operation wavelength to longer wavelengths, a unique multilayer Gr (MLG) structure was added to the backside of the 2D-­CCD (Figure 7.8) [22]. This new MLG-­based 2D-­CCD device has also been termed as Gr charge injection (GCI) photodetector. MLG absorbs around 40% of incident light in the wavelength range of 2.5–5.0 μm. MLG on the backside of Si forms a Schottky junction, and hot holes generated in the Gr are injected into the Si via photothermionic (PTI) emission process (Figure 7.8b). These holes are finally integrated into the depletion well for storage and readout through top Gr (Figure  7.9a). As shown in Figure  7.9b, the

131

7  Graphene for Silicon Image Sensor Ambient light

t = 0.033 s

t = 0.067 s

t = 0.1 s

Reconstructed images

DAQ

GCI Outputs Inputs pixels

Object Light source

375 nm

(a) t = 0.1 s

t = 0.2 s

1342 nm

V

D

t = 0.3 s X Y

(b) t = 0.033 s

t = 0.067 s

t = 0.1 s

t = 0.5 s

t = 1.0 s

t = 2.0 s

(f)

Two-axis scanning platform

Linear array GCI

Vg(t)

(c) 3 µm

(g)

t = 1.0 s

t = 1.5 s

t = 2.0 s

t = 0.067 s

t = 0.1 s

Ambient light

t = 0.033 s

(d) 3.8 µm

132

0

Photocurrent

200 µA

(h)

(e)

Figure 7.10  (a–e) Broadband imaging based on the GCI single pixel and linear array under different light conditions (ambient light, 375 nm, 1342 nm, 3 μm, and 3.8 μm). (f) Homemade imaging setup. (g) Photographs of linear GCI array. (h) Imaging under ambient light using GCI arrays. Source: Reproduced with permission from Liu et al. [22]/Springer Nature.

photocurrent of the MLG/Si junction is fitted with a power law ISch,pc ∝ Pγ, which shows a superlinear-­to-­linear transition when increasing the photon energy at a constant bias (left) and increasing the forward bias at a constant photon energy (right). Figure  7.9c,d clearly shows the enhancement in external quantum efficiency (EQE) and responsivity of GCI compared to the 2D-­CCD (GOS) device (Figure 7.10).

­7.8  Graphene-­based Position-­sensitive Detector Since the discovery of the lateral photovoltaic effect, the research is going on the lateral photovoltaic effect in different device applications, such as position-­sensitive detectors (PSDs), space technology, environmental sensing, optoelectronics, and telecommunication. The device mechanism of a p–n junction-­based PSD is very straightforward, in which the optically excited electron–hole pairs are generated and extracted by the built-­in field, as presented in Figure 7.11a. On the other hand, the PSD with an insulating layer follows the band-­bending phenomena at the SiO2/ Si interface, followed by carrier diffusion in lateral directions [26], which is shown in Figure 7.11b,c. The optically generated carriers compensate the electric field in

­7.  Graphene-­based Position-­sensitive Detector

e– e–

EC

e– e–

EC

e–

p-type semiconductor

e–

e–

EC p-Si

SiO2

EF n-type semiconductor

EV (a)

e– e–

(a)

h+

h+

h+

h+

EV h+

EF EV

h+

h+

h+ (c)

h+

–3 –2 –1 0 1 2 3 Film

e– h+ Substrate

(b)

e– e– e– e– e– e– h+ h+ h+ h+ h+ h+

e–

Electron

h+

Hole

e– h+

Figure 7.11  Schematic of the energy band structure of (a) the p–n junction-­based PSD, (b) the SiO2/Si substrate-­based PSD, and (c) carrier diffusion in the lateral direction. Source: Reproduced with permission from Hu et al. [25]/Springer Nature / CC BY 4.0.

the lateral direction through their diffusion [27]. The carrier mobility and the carrier energy of an element are critical parameters for diffusion in the lateral direction. Gr is an attractive material due to its high carrier mobility and tunable energy bandgap [25, 28]. These materials are suitable to apply as FETs, photodetectors, sensors, and PSDs [29, 30]. These PSDs fabricated from 2D materials show high responsivity in broadband, fast response, and low power consumption, as shown in Table 7.1. Gr is one of the most promising 2D materials for PSD application. So here, the applications of Gr in PSDs and the physical mechanism for position sensitivity and fast response for the PSD structures are explained. Gr is a cost-­effective 2D material having excellent electrical, thermal, and mechanical properties, good chemical stability, and high mobility. The PSD based on Gr may detect weak signals due to high mobility as well as a larger photogenerated carrier’s lifetime  [35]. Figure  7.12a presents the Gr/SiO2 layer over a low p-­doped Si substrate. The photogenerated electron–hole pairs upon illumination are collected by the electric field at SiO2/Si and diffused at the depletion layer laterally. Hence, the position can be measured by the information from the relative voltage collected at the different Gr terminals. The reason is that the electron at the acceptor level decreases, while the density of the holes at the donor level becomes higher because of the large positively charged state generated at the Si/SiO2 interface, which bends the energy level down. So, the electrons gather at the p-­Si interface to form a negative depletion layer. The built-­in electric field separates the photogenerated carriers, which are collected as the output of PSD.

133

7  Graphene for Silicon Image Sensor

Photocurrent (µA)

Light

lSD

SiO

A

2

Ligh

tly p

-dop

Electrons

P P B

PC

16 On Off

12 8 4 0

ed S

i

0

Holes

(a)

1 2 Time (s)

(b) E

80

3 µm 8 µm 18 µm 30 µm 50 µm

Normalized photocurrent (a.u.)

20

VD

0.6 0.4

C B

A

5

10

Position (µm)

(c)

4

0.65

2 0 –2 –4 –4

0.45

80

40

S2 0

a.u. 0.90

D

40

0

S1

–15 –10 –5

0.75

0

S1 S2

0.8

3

0.55

(d)

1.0

Y (mm)

Y (µm)

134

(e)

X (µm)

–2

0

X (mm)

2

4

(f)

Figure 7.12  (a) Schematic diagram of graphene-­based position-­sensitive detector. (b) Photoswitching characteristics of graphene-­based position-­sensitive detector at different incident light wavelengths. (c) Position-­dependent photoresponse of a one-­dimensional (1D) position-­sensitive detector prepared under incident light power of 50 nW. (d) 2D spatial mapping of the position-­sensitive detector photocurrent under 400 nW incident light. (e) Optical image of a 2D position-­sensitive detector and schematic of the light trajectory in the operating area. (f) Measured trajectory (red dots) and actual position (dotted white line) of the laser. Source: Reproduced with permission from Hu et al. [25]/Springer Nature / CC BY 4.0. Table 7.1  Performance parameters of Si-­based PSDs.

Structure

Laser wavelength (nm)

Laser power (mW)

Cr/SiO2/Si

635

Co/Si

832

Position sensitivity (mV mm−1 mW−1)

Response time (μs)

References

5

8.40

–

[31]

5

16.40

–

[32]

Ti/TiO2/Si

632

3

37.67

–

[33]

a-­Si:H/c-­Si

980

8.3

0.88

–

[34]

Gr/SiO2/Si

514

5 × 10−5

–

1.2

[35]

Gr/Si

532

82 × 10−5

365.85

0.44

[36]

a-­MoS2/Si

780

10

18.3

2.1

[28]

So, the electrons will be transferred to the SiO2/Si interface and move laterally to achieve equilibrium, while the photocurrent magnitude decreases exponentially in correspondence to an increase in the distance from the illumination spot. Electrons moving into the Gr change the amount of hole concentration as well as the channel current through capacitive coupling. This gating affects the interfacial amplification, measured by the ratio of the carrier lifetime at the SiO2/Si to the carrier transit

­7.  Graphene-­based Position-­sensitive Detector

time in the Gr channel. Ultrahigh mobility and the longer lifetime of the carrier in Gr-­based PSD result in high amplification and improved device sensitivity. The photoswitching of Gr-­PSD at different laser positions is presented in Figure 7.12b. The photocurrent is negligible when light spot is far away from the Gr channel, and it increases when light spot moves close to the channel. The photoresponse of the PSD under 50 nW laser is shown in Figure  7.12c. The optical image of device S1 and device S2 on the same substrate is shown in the inset, separated by 10 μm. The photocurrent (I2 − I1)/(I2 + I1) decreases from ∼1 to 0.5 for S1, when the light position increases from −5 to 5 μm from the Gr, while S2 increases from 0.5 to 1. So, the Gr-­ PSD presents an excellent performance. Additionally, it also shows a good response to the positional changes in the lateral direction, which is presented in Figure 7.12d. The optical microscope images of another type of Gr-­PSD for 2D measurement with a different structure are shown in Figure 7.12e,f. The large-­area Gr/Ge heterojunction-­based PSD can also be achieved, where a large-­area Gr monolayer over Ge substrate can be deposited, as shown in Figure 7.13a. The energy band bending at the surface states forms a built-­in electric field, which directly affects the photoresponse as well as the speed of the PSD. Furthermore, the photoresponse from the pure Gr channel can extend the photoresponse of the PSD to a broadband up to near-­infrared wavelength. The angular position changes can also be well measured by a structural modification for a PSD, as shown in Figure 7.13b. When the target moves at a small angle, the reflected beam moves for a small length over the PSD, so the angular difference can be obtained as shown in Figure 7.13c. This response can also be employed to differentiate the sound frequency, recording, and storing as presented in Figure 7.13d,e. The PSD can also detect the trajectory of fast-­moving objects accurately due to its high Δθ

h𝜈

L EC Ef

2Δ

V2

II

III

V

Position (mm)

IV

(e)

0.50

1.0

0.45

0.5

0.40 0.35

0.0

–1.0

Δθ

(b)

(c)

0.02 0.00 –0.02

VI

1.5

0.30

–0.5 0

N-type Ge

I

ΔL

H

Al2O3

(a)

(d)

θ

EV Electrode Graphene SiO2

V1

Position (mm)

PSD

0.00

0.05

0.004 0.008 0.012 0.016 Δ θ (º)

10 kHz

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 0.2 0.0 –0.2 0

0.2 0.0 –0.2 0.2 0.0 –0.2

0.25

–0.05

1

2

3

4

5

6

7

8

1 kHz

9 10 250 Hz

0

4

8 12 16 20 24 28 32 36 40 44 Multi frequencies

0

10

20 Time (ms)

30

40

(f)

Figure 7.13  (a) Schematics of the graphene-­Ge position-­sensitive photodetector. (b) Schematics of the measurement setup. (c) The small-­angle measurements. (d) Setup of the vibration frequency measurement. (e) Different vibrations were recorded by the position-­ sensitive photodetector. (f) High-­speed trajectory tracking using the graphene-­Ge position-­ sensitive photodetector. Source: Reproduced with permission from Hu et al. [25]/Springer Nature / CC BY 4.0.

135

136

7  Graphene for Silicon Image Sensor

sensitivity and fast responsivity. This can be done without any source meter, and the minimum interval can be smaller than 110 μs. This tracking has an excellent spatial resolution and can be used as a reliable device for practical applications. The Gr-­ PSD trajectory tracking system is presented in Figure 7.13f. In short, the Gr-­PSDs are highly efficient devices with ultrafast response time, good linearity, high sensitivity, and very low power consumption, making them a potential candidate for future device applications with precise and accurate measurements.

­7.9  Challenges and Perspectives The new physical mechanism of 2D materials is expected to promote the comprehensive innovation of materials, devices, integrated circuits, and semiconductor industries and to realize new functions such as digital logic, optical perception, information communication, and storage. However, the existing mature silicon-­ based material technology will still dominate for a long time in the future due to low cost and complete ecological chain. The industrialization challenges of 2D materials include device performance optimization and large-­scale application technical feasibility as well as the establishment of the corresponding industrial ecological chain and cost issues. The fusion development by comprehensive usage of the advantages of both 2D materials and silicon materials is expected to jointly promote the development of devices in the post-­Moore era. It is necessary to consider the material, device, circuits, and commercialization at the same time for the development of silicon-­based 2D material photodetectors (Figure 7.14). At the material level, the synthesis and transfer of large-­scale, high-­quality 2D materials are important issues to be addressed. In addition, interface engineering at the material level, including contact, doping, and strain issues, also needs to be comprehensively considered. Photodetectors with high overall performance are required at the device level. For practical applications, it is required to consider photodetectors

Material level

Tasks:

Fundamental material research

Device level

Circuit level

Commercialization level

Design, fabrication, test

Wafer-scale integration

Large-scale production

High responsivity

Goals:

High-quality material synthesis

High sensitivity Advanced transfer process

Issues and challenges:

High speed

High linearity High stability

Interface engineering

Novel mechanism

Doping engineering

Novel design

Strain engineering Heterostructure integration

Uniformity Reliability Yield Cost

Multilayer vdW system integration Inline metrology CMOS Compatibility of processes

Figure 7.14  The tasks, goals, and challenges for developing silicon-­2D material-­based photodetectors. Source: Reproduced with permission from Liu et al. [37]/Springer Nature / CC BY 4.0.

  ­Reference

with high comprehensive performance, including response speed, sensitivity, and linearity. For Gr-­based detectors that meet the requirements of high-­performance broadband detection at room temperature in the future, it is necessary to achieve the advantages of high sensitivity, high-­speed detection, and high linearity at the same time. More specifically, there is often a trade-­off between response speed and detection sensitivity. In addition, high linearity is a requirement that should be paid more attention to in the future development of silicon-­based 2D photodetectors. The circuit level requires the development of wafer-­level photonic integrated circuits based on 2D materials. In this case, it is essential to realize high-­performance photodetectors with high uniformity and high reliability. The cost aspect requires the development of system-­level multilayer vdW integration methods and quasi-­ CMOS-­compatible processes. The current demand for image data in the context of the internet of things and big data makes the research on advanced wide-­spectrum imaging technology necessary. Silicon/Gr image sensors offer a viable strategy to monolithically integrate 2D materials into conventional solid-­state imaging technology, which is conducive to exploring the next-­generation broadband photodetectors for high-­quality imaging and other functionalities.

­References 1 Boyle, W.S. and Smith, G.E. (1970). Charge coupled semiconductor devices. The Bell System Technical Journal 49 (4): 587–593. 2 Amelio, G.F., Tompsett, M.F., and Smith, G.E. (1970). Experimental verification of the charge coupled device concept. The Bell System Technical Journal 49 (4): 593–600. 3 Magnan, P. (2003). Detection of visible photons in CCD and CMOS: a comparative view. Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 504 (1–3): 199–212. 4 Luštica, A. (ed.) (2011). CCD and CMOS image sensors in new HD cameras. Proceedings ELMAR-­2011. IEEE, Zadar, Croatia. 5 El Gamal, A. and Eltoukhy, H. (2005). CMOS image sensors. IEEE Circuits and Devices Magazine 21 (3): 6–20. 6 Shepherd, F.D. and Yang, A.C. (1973). Silicon Schottky retinas for infrared imaging. 1973 International Electron Devices Meeting. IRE. 7 Denda, M., Kimata, M., Iwade, S. et al. (1991). 4-­band*4096-­element Schottky-­ barrier infrared linear image sensor. IEEE Transactions on Electron Devices 38 (5): 1131–1135. 8 Bao, Q. and Loh, K.P. (2012). Graphene photonics, plasmonics, and broadband optoelectronic devices. ACS Nano 6 (5): 3677–3694. 9 Akinwande, D., Huyghebaert, C., Wang, C.-­H. et al. (2019). Graphene and two-­ dimensional materials for silicon technology. Nature 573 (7775): 507–518. 10 Chen, X., Shehzad, K., Gao, L. et al. (2019). Graphene hybrid structures for integrated and flexible optoelectronics. Advanced Materials 32 (27): 1902039.

137

138

7  Graphene for Silicon Image Sensor

11 Goossens, S., Navickaite, G., Monasterio, C. et al. (2017). Broadband image sensor array based on graphene–CMOS integration. Nature Photonics 11 (6): 366–371. 12 Hsu, A.L., Herring, P.K., Gabor, N.M. et al. (2015). Graphene-­based thermopile for thermal imaging applications. Nano Letters 15 (11): 7211–7216. 13 Vicarelli, L., Vitiello, M.S., Coquillat, D. et al. (2012). Graphene field-­effect transistors as room-­temperature terahertz detectors. Nature Materials 11 (10): 865–871. 14 Choi, C., Choi, M.K., Liu, S. et al. (2017). Human eye‐inspired soft optoelectronic device using high‐density MoS2‐graphene curved image sensor array. Nature Communication 8: 1664. 15 Lee, C.-­P., Cai, M.-­Y., Wang, J.-­Y. et al. (2021). Bilateral photoresponse of a graphene-­oxide-­semiconductor heterostructure diode. Physical Review Applied 15 (5): 054067. 16 Mennel, L., Symonowicz, J., Wachter, S. et al. (2020). Ultrafast machine vision with 2D material neural network image sensors. Nature 579: 62–66. 17 Bigas, M., Cabruja, E., Forest, J., and Salvi, J. (2006). Review of CMOS image sensors. Microelectronics Journal 37 (5): 433–451. 18 Fossum, E.R. (1997). CMOS image sensors: electronic camera-­on-­a-­chip. IEEE Transactions on Electron Devices 44 (10): 1689–1698. 19 Burke, B., Jorden, P., and Vu, P. (2006). Overview paper – CCD technology. In: Scientific Detectors for Astronomy 2005: Explorers of the Photon Odyssey (ed. J.E. Beletic, J.W. Beletic and P. Amico), 225–264. Dordrecht: Springer Netherlands. 20 Mendis, S., Kemeny, S.E., and Fossum, E.R. (1994). CMOS active pixel image sensor. IEEE Transactions on Electron Devices 41 (3): 452–453. 21 Burke, H.K. and Michon, G.J. (1976). Charge injection imaging: operating techniques and performances characteristics. IEEE Journal of Solid-­State Circuits 11 (1): 121–128. 22 Liu, W., Lv, J., Peng, L. et al. (2022). Graphene charge-­injection photodetectors. Nature Electronics 5 (5): 281–288. 23 Brant, J.C., Leon, J., Barbosa, T.C. et al. (2010). Hysteresis in the resistance of a graphene device induced by charge modulation in the substrate. Applied Physics Letters 97 (4): 042113. 24 Wang, H., Wu, Y., Cong, C. et al. (2010). Hysteresis of electronic transport in graphene transistors. ACS Nano 4 (12): 7221–7228. 25 Hu, C., Wang, X., and Song, B. (2020). High-­performance position-­sensitive detector based on the lateral photoelectrical effect of two-­dimensional materials. Light: Science and Applications 9: 88. 26 Henry, J. and Livingstone, J. (2001). Thin-­film amorphous silicon position-­sensitive detectors. Advanced Materials 13 (12, 13): 1022–1026. 27 Jin, K.-­J., Zhao, K., Lu, H.-­B. et al. (2007). Dember effect induced photovoltage in perovskite p–n heterojunctions. Applied Physics Letters 91 (8): 081906. 28 Shan, H., Yu, Y., Wang, X. et al. (2019). Direct observation of ultrafast plasmonic hot electron transfer in the strong coupling regime. Light: Science and Applications 8: 9.

  ­Reference

29 Hu, C., Wang, X., Miao, P. et al. (2017). Origin of the ultrafast response of the lateral photovoltaic effect in amorphous MoS2/Si junctions. ACS Applied Materials & Interfaces 9 (21): 18362–18368. 30 Wang, X., Zhao, X., Hu, C. et al. (2016). Large lateral photovoltaic effect with ultrafast relaxation time in SnSe/Si junction. Applied Physics Letters 109 (2): 023502. 31 Liu, S., Xie, X., and Wang, H. (2014). Lateral photovoltaic effect and electron transport observed in Cr nano-­film. Optics Express 22 (10): 11627. 32 Qiao, S., Liu, Y., Liu, J. et al. (2015). Large lateral photovoltaic effect in a-­Si: H/c-­Si p–i–n structure with the aid of bias voltage. Applied Physics Express 8 (12): 122201. 33 Yu, C.Q., Wang, H., and Xia, Y.X. (2009). Giant lateral photovoltaic effect observed in TiO2 dusted metal-­semiconductor structure of Ti/TiO2/Si. Applied Physics Letters 95 (14): 141112. 34 Yu, C.Q. and Wang, H. (2010). Large near-­infrared lateral photovoltaic effect observed in Co/Si metal-­semiconductor structures. Applied Physics Letters 96 (17): 171102. 35 Wang, W., Yan, Z., Zhang, J. et al. (2018). High-­performance position-­sensitive detector based on graphene–silicon heterojunction. Optica 5 (1): 27. 36 Wang, W.-­H., Du, R.-­X., Guo, X.-­T. et al. (2017). Interfacial amplification for graphene-­based position-­sensitive-­detectors. Light: Science and Applications 6 (10): e17113. 37 Liu, C., Guo, J., Yu, L. et al. (2021). Silicon/2D-­material photodetectors: from near-­infrared to mid-­infrared. Light: Science and Applications 10 (1): 123.

139

141

8 System Integration with Graphene for Silicon Optoelectronics 8.1 ­Introduction A photodetector is a device which converts the incident photon energy into an electrical output signal, and it can be divided into surface incidence‐type and ­waveguidetype detectors, according to the incidence mode. Although the structure of the Gr detector is relatively simple and has a fast response rate, the Gr can only absorb 2.3% of the vertically incident light due to its single atomic layer thickness, resulting in a low response. To improve the response of Gr detectors, some methods have been used to enhance the strength of the Gr‐light field interaction; for example, Marco et al. designed an optical resonant cavity [1], which increases the absorption of Gr by a factor of 26, with an absorption rate of >60%, and the response can be as high as 21 mA W−1. However, due to the resonance effect, the optical bandwidth of this structure is not as high as that of the Gr detector, and the broadband wavelength range absorption cannot be achieved. Using the carrier capture effect (carrier trapping effect), one photogenerated carrier can be trapped at the defect state, while another carrier can make multiple rounds in the circuit resulting in large photoconductive gain. For example, Fang Luo et al. transferred Gr on an SiO2/n‐Si substrate, where n‐Si was heavily doped [2]. When Si absorbs light, an additional photovoltage is generated to modulate the conductivity of Gr, achieving responsiveness of 500 A W−1 at 450‐nm light. However, the long carrier lifetime in the carrier capture effect makes the device response very slow. There is another type of Gr detector for the Gr–Si Schottky junction detector, where photogenerated carriers can be separated due to built‐in electric field. Zefeng Chen et al. laid the Gr connection on top of the p‐type Si [3]; the responsivity can be as high as 104 A W−1, but the response rate is still relatively slow, only 3 μs. As for the optical waveguide‐type Gr detector, it is convenient to use the optical waveguide that can extend the action distance between Gr and the light field, and the response speed of the detector can be taken into account while improving the responsivity, thus gaining wide attention. Xuetao Gan et  al. fabricated an on‐ chip‐integrated Gr photodetector in 2013 [4], using the structure of asymmetric metal electrodes to achieve a response of 0.1 A W−1 and a response speed of more Graphene for Post-Moore Silicon Optoelectronics, First Edition. Yang Xu, Khurram Shehzad, Srikrishna Chanakya Bodepudi, Ali Imran, and Bin Yu. © 2023 WILEY-VCH GmbH. Published 2023 by WILEY-VCH GmbH.

142

8  System Integration with Graphene for Silicon Optoelectronics

than 18 GHz. As the physical size of microelectronic devices becomes smaller, the size that can be achieved by the manufacturing process will reach the limit. If forced to catch up with Moore’s law, the increasing manufacturing difficulty and sky‐high cost of building manufacturing units will seriously hinder the ­development of the industry. Therefore, the traditional scale‐down development model is about to end [5], but demand for improved device performance will not end. In the foreseeable future, packaging will occupy an increasing proportion of microelectronics design. Three‐dimensional packaging provides a stacked vertical interconnection of multiple chips, which greatly reduces the area occupied by connection paths and chips, increases device density, and makes it easier to achieve organic integration of different modules, with better electrical performance and lower power consumption [6].

8.2 ­Graphene Silicon Flip Chips Flip‐chip, one of three‐dimensional packaging, is a means of chip assembly, whereby the chip is attached face down (active integrated circuit [IC] side down) to the substrate [7]. In the 1960s, the flip‐chip technology was proposed by IBM, and then it began to be promoted and developed in the 1980s. In recent years, it has gradually become a research hot spot in the field of high‐density packaging. The flip‐chip soldering technology originally designed by IBM is with a hard copper ball core solder ball structure; that is, a copper ball is placed on the Sn–Pb solder, and the solder ball is formed after reflow at 350 °C. The hard copper ball provides the distance between the two chips, to prevent the short circuit caused by the solder collapse. But this technology had a greater challenge to the production of copper balls, so it had not been promoted. The technology that followed is named “controlled collapse chip attachment”  [4], also known as (C4). This method abandons the hard copper balls and instead uses a glass passivation layer around the solder joints as a dam to prevent the liquid volatile material from overflowing, thereby preventing short‐circuiting of adjacent points. The surface tension of the solder also supports the weight of the chip as the role of a hard copper ball. This method has been widely used. IBM protected this secret technology as intellectual property (IP) for many years. Si‐based image sensors are widely used. However, the sensitive range is fundamentally limited to below 1100 nm. For detection beyond 1100 nm, other materials such as III–V have been used, which need to be flip‐chip‐bonded to Si readout circuits  [8]. Flip‐chip bonding is an ideal interconnect technology for hybrid pixel detectors as the connections are within the footprint of the chip, and there is no spare space for the chip, and the interconnects, maximizing the sensor’s active area. In flip‐chip bonding, conductive adhesives or solder bumps are required to realize the electrical–mechanical interconnections between the chips. For instance, a solder bump is grown on each pixel on the readout chip, and sensor chips have solderable under‐bump metallization (UBM). Since there is a solder bump interconnection for every pixel, the bumping quality and flip‐chip yields must be very high to avoid visible defects in the radiation image. For the chips with the number of pixels reported in this study, the flip‐chip yield of hybrid pixel detectors is typically determined by the success rate of the bumping process.

8.2  ­Graphene Silicon Flip Chip

Flip‐chip packaging can also be applied to Gr image sensors. The schematic diagram is shown in Figure 8.1. The active surface of the Gr photosensitive chip (with surface bonding points) and the base of the readout circuit chip are pasted and packaged. Flip‐chip interconnection technology has reliable electrical properties, low parasitic impedance, good antielectromagnetic interference capability, and high mechanical reliability. At the same time, the technology has excellent heat conduction characteristics, which is conducive to effective heat dissipation of the device. To increase the density of pixel integration, it is necessary to reduce the pixel pitch as much as possible, and the size of the flip‐chip solder joints needs to be reduced accordingly, which poses certain challenges to the generation of reliable and uniform indium pillars. This process intends to obtain indium pillars with suitable size and height and good uniformity by optimizing the substrate temperature, evaporation rate, and changing the inclination angle of the thick photoresist in the process of evaporating metal to meet the requirements of the device flip‐chip technology. The designed process steps of this project are as follows: use the alignment bonding tool to suck the chip, use the self‐aligned display system to align the chip vertically with the base, and the solder bumps of the chip are positioned at the corresponding base contact points, and then heat and pressurize. The method causes the solder to reflow and forms the electrical and physical connection between the base and the chip. Due to the small size and high density of the solder joints, it is easy to cause problems such as indium column deformation and misalignment. This process is designed according to the characteristics of the Si‐based Gr detection array device. It optimizes the pressure, time, temperature, and other conditions of the pressure‐welding interconnection to control. There is an angle problem between the detector and the readout circuit to achieve a good flip‐chip interconnection effect. Illumination

Detector array

RO

IC Cross section Illumination

Graphene Si substrate Oxide CMOS ROIC

Detector array

UBM Passivation layer Indium pillar

ROIC

Figure 8.1  Schematic diagram of flip‐chip bonding between Gr array detector and ROIC.

143

144

8  System Integration with Graphene for Silicon Optoelectronics

The detailed process steps of generating indium pillars and flip‐chip bonding are described as follows. First of all, the Gr sensor and readout‐integrated circuit (ROIC) are created on the respective wafers. Then, pads are metalized on the surface of the chips. The passivation layer plays an essential key role in the packaging process. For example, it acts as an insulator for solder joints to prevent short circuits; on the other hand, the passivation layer helps to make the solder concentrated to the designed solder joint position and ensures that the solders form an approximately spherical shape in reflow. The designed insulator is Si3N4 with a plasma‐enhanced chemical vapor deposition process. The UBM layer is deposited with three metal layers, including Ti, Ni, and Au. A negative photoresist is used for the lift‐off process. After exposure and development, the patterned shape of the window is obtained, which ensures that the photoresist does not cover the areas for the metal pillars to be retained. After the generation process of indium pillars, including lithography, evaporation, and lift‐off, they can be used as solder bumps. Because the layer of metal solder is relatively thick, the thin‐cake‐shaped metal pillars are used to reflow into balls instead of directly making high metal pillars, which reduces the burden in the generation process. The reflow of the indium column on one side of the sensor into a ball to provide the gap between the two pieces is required for the package. The vacuum reflow oven and the flip‐chip soldering machine can be used separately for reflow, and formic acid is used during the reflow process to provide a reduction effect. Finally, the alignment welding of the two chips on the flip‐chip soldering machine is completed (Figure 8.2). Graphene detector

Metal

Gr

Oxide

CMOSROIC

Si

Passivation layer

Passivation Gr

Oxide

CMOSROIC

Si

Metal layer

UBM Gr

Oxide

CMOSROIC

Si

Solder bumps and fill process

Solder bumps

Filler

Gr

Oxide

CMOSROIC

Si

Align, bonding, reflow CMOSROIC

Gr

Oxide

Si

Figure 8.2  Schematic diagram for the fabrication and bonding process of flip‐chip devices.

8.3  ­Graphene Silicon Heterogeneous Integratio

8.3 ­Graphene Silicon Heterogeneous Integration Apart from Gr, 2D materials family also includes other materials such as hexagonal boron nitride (h‐BN) and transition metal dichalcogenides (TMDs, such as MoS2, WSe2, and MoTe2). The lack of dangling bonds and unique electronic and optoelectronic properties make 2D materials suitable for various high‐end applications from transistors to sensitive photodetectors and sensors. Importantly, these novel devices and systems are complementary metal–oxide–semiconductor (CMOS) compatible, allowing their integration with the conventional Si‐CMOS industry. This integration process brings new functionality to the Si devices. Here, we talk about the heterogeneous integration of various 2D material‐based devices with a standard Si platform for various devices application such as transistors, integrated photodevices, and sensors. Various research indicate that 2D‐based transistors are high‐performance devices with diverse electrical characteristics (Figure 8.3) [10]. 2D material‐based photodetectors and modulators have a wide range of advantages, as they offer high speed, wide spectral bandwidth, low power consumption, and monolithic integration with Si‐CMOS technology [9]. By employing Gr‐quantum dot photodetectors [11], a fully integrated image sensor array of 388 × 288 pixels has

Classical FET

Ferroelectric oxide

G S

G

G

S

D

Reconfigurable FET

NEG-FET

D

S

D

Polarity gate Tunnel FET

Vertical barristor

G

Lateral barristor

G

Semimetal phase D

S

S

D

S

G D

(a) n-type

Reconfigurable

Ambipolar

Anti-ambipolar

Single-electron transistor

lD

p-type

VG (b)

Figure 8.3  (a) Illustration of various 2D planar device structures that can be readily integrated on virtually any level of a high‐rise 3D Si. (b) The diverse electrical characteristics afforded by 2D FETs are based on material choice and engineering of the device structure, contacts, and doping. NEG‐FET, negative‐capacitance FET. Source: Reproduced with permission from Akinwande et al. [9]/Springer Nature.

145

146

8  System Integration with Graphene for Silicon Optoelectronics

been realized by integrating Gr with a CMOS ROIC and operation as a digital camera was shown for both visible and infrared light. This wavelength range is currently not available for any commercial photodetector or image sensor. Gr–Si phase and absorption modulators show high‐rate (10 Gbits−1) binary transmission [5]. A monolithic 3D image sensor is demonstrated through hybrid integration of TMD‐based photo­ transistors with high‐density Si electronics [6]. In general, these advances in back‐ end‐of‐the‐line (BEOL) integration of 2D materials with Si‐CMOS pave the way for the development of 2DM‐based photodevices. Benefiting from the last layer of the BEOL integration process, 2DM‐based sensors are with widespread research [7]. Gr resistance‐change devices were employed for toxic‐gas sensing on top of a foundry Si‐CMOS chip containing the Si readout electronics [12]. The rapid development of commercial 2D‐Si sensor platforms is probably imminent as the transfer integration or low‐temperature selective growth of suitable 2DMs on Si progresses. 2D materials, thanks to their atomic thickness and exotic physical/electronic properties, are promising materials for a variety of applications. For example, semiconducting MoS2 is an ideal material to replace Si in sub‐10 nm technology. At the same time, Gr, a gapless semimetal, is an ideal material for ultrafast photodetection and optical modulation on a CMOS platform. The future of Gr and related 2D materials lies in the feasibility of their integration with the current CMOS technology. 2D materials can be grown separately on metal catalyst substrates and then conveniently transferred on the Si platform. Transfer of 2D materials required low thermal budget, which is critical for CMOS integration. Monolithic integration of 2D materials with Si is critical to develop complex 2D material‐based 3D systems, which can also benefit from Si circuitry. Such 3D integration will significantly increase the performance of future Si microsystems, enabling exciting new opportunities for hybrid systems (Figure 8.4). CVD graphene

8

38

288

l

e pix

Light in

pixel

A´

Graphene on Si phase modulators

CMOS readout circuitry CMOS wafer with image sensor dies

(b)

(a) 3D stackable FET

Interconnect

TMD transfer

Back electrode

A

100 µm

(c) Top monolayer TMD PD array

Bar-port Cross-port Bottom 3D stackable circuit

SiO2

Substrate

(d)

Figure 8.4  Integrated Gr and 2D photodevices. (a) A broadband camera setup based on CMOS‐integrated Gr‐quantum dot photodetectors. (b) Schematic representation of the integration process. (c) Gr–Si phase modulators with gigahertz bandwidth. (d) A monolithic 3D image sensor with an integrated phototransistor array fabricated using a large‐area (direct‐bandgap) monolayer TMD realized on a logic/memory hybrid. CVD, chemical vapor deposition; PD, photodetector. Source: Reproduced with permission from Akinwande et al. [9]/Springer Nature.

8.4  ­Graphene Silicon

Monolithic Integration for Optoelectronics Application

8.4 ­Graphene Silicon Monolithic Integration for Optoelectronics Applications The outstanding intrinsic properties of Gr have made it a material of great interest. On a single device level, the intrinsic properties are high carrier mobility, broadband optical absorption, and an extremely high surface‐to‐volume ratio. These properties are already harnessed to design high‐performing devices that outperform their established semiconductor counterparts, including IR photodetectors, Hall‐effect magnetic field sensors, pressure sensors, and gas sensors. With the development of technology, people hope to process, calculate, transmit, and store information faster, and meanwhile, the energy consumption is getting lower. Single‐pixel photodetectors can no longer meet the demands of system‐level applications. Hence, development of photodetector arrays is important. Higher requirements are put forward for the performance of integrated circuits, which play an important role in all aspects of photodetectors. For the next‐generation photodetectors, monolithic integration, miniaturization, improved performance and stability, and reduced costs are the fundamental requirements. Si‐based CMOS process is the current mainstream microelectronics manufacturing technology, which has the advantages of low cost, high integration density, and good compatibility. Solving the integration problem between Gr and Si will help realize the practical application of Gr‐based new photodetectors. Developing a reliable large‐scale production process would not only unleash the potential of Gr in sensor applications but may also help to trigger other vital uses, where its unique properties produce significant enhancements. In this respect, integrating Gr into conventional Si‐based fabrication lines could be a promising focus. 3D integration into the Si‐CMOS platform may enable the combination of high‐performance Gr devices with established CMOS readout circuitry, with production costs as low as for conventional Si technology. The first proof‐of‐concept demonstration of such 3D‐integrated Gr sensor systems was magnetic field sensors [13] in 2014. They developed a low‐temperature process for fabricating Gr devices that is compatible with Si‐based CMOS technology and demonstrate the integration of Gr‐Hall element (GHE) with Si‐CMOS‐ICs on the same chip. Signal amplifying/processing ICs were manufactured via commercial 0.18‐μm Si‐CMOS technology, and GHEs were fabricated on top of the passivation layer of the Si‐CMOS chip via a low‐temperature microfabrication process, as shown in Figure 8.5. The chip consists of an overall size of 5 × 5 mm and an active area of 820 × 1030 μm. In 2017, Goossens et al. [5] reported that a hybrid Gr quantum dot (GQD) photodetector array, which is vertically integrated into a CMOS readout chip. The researchers demonstrated the monolithic integration of Gr and CMOS‐integrated circuits. The integration potential for such devices was demonstrated by fabricating an image sensor with a 388 × 288 array of GQD photodetectors, which are operated as a digital camera with high sensitivity for both visible and shortwave infrared light. Almost 110 000 Gr photoconductive channels are individually integrated vertically to connect with each electronic component of a

147

148

8  System Integration with Graphene for Silicon Optoelectronics

Graphene transfer

CMOS IC chip

CMOS IC chip EBL and oxygen plasma etching

EBL and metal deposition

SU8 passivation

CMOS IC chip

CMOS IC chip

AI interconnect

Graphene

Pad

Pad Pin

Si3N4 passivation

Pin Via hole S

SU8 passivation

G

N+ p Si

0.18 µm CMOS D

S

N+

P+

G

D

P+ n well

Figure 8.5  Fabrication process flow and structure of the Gr–Si hybrid Hall chip. EBL, E beam lithography. Source: Reproduced with permission from Huang et al. [14]/ Springer Nature.

CMOS ROIC, as shown in Figure 8.6. The Gr channels are sensitized to ultraviolet, visible, near‐infrared, and short‐wave infrared light with PbS‐CQDs. On light absorption, electron–hole pairs are generated and separated due to the built‐in electric field. The electrons are trapped in the QDs, while the holes are transferred to the Gr. The schematic represents the CTIA‐based balanced readout scheme per column and global correlated double sampling (CDS) stage and output driver. The chip containing the circuitry is similar to those used for commercial image sensors in digital cameras  [5], commonly used in smartphones. This image sensors operates in the wavelength range of 300–2000 nm, which is not possible for current monolithic CMOS image sensors. Such image sensor can find applications in safety and security, smartphone cameras, night vision, automotive sensor systems, food and pharmaceutical inspection, and environmental monitoring [15, 16]. For the integration of graphene devices and CMOS chips, the BEOL process plays an important role. The typical BEOL process of the CMOS‐Gr chip is

8.4  ­Graphene Silicon

Monolithic Integration for Optoelectronics Application

Colloidal quantum dots CVD graphene: mobility > 1000 cm2 V–1 s–1 Column select Rcomp Rblind

Row select CTIA

VSS VREF

CDS

VDD VOUT

Vertical interconnects CMOS readout circuit

Output driver

Figure 8.6  Side view explaining the Gr photoconductor and the underlying readout circuit. Source: Reproduced with permission from Goossens et al. [5]/Springer Nature.

shown in Figure 8.7. Gr can be grown on a separate catalytic substrate and then transferred on the Si. In this case, the quality of grown Gr is good, but the transfer process adds impurities, which diminishes the quality of Gr. Also, a fully automated transfer process is not available yet. As an alternative, Gr can be grown directly on the targeted substrate with low‐temperature methods, However, Gr quality using low temperature and on Si substrate is not good enough. Post‐Gr transfer on Si includes patterning/etching, dielectric deposition, and making metal contacts to the Gr. All these steps are mastered for conventional semiconductor industry; however, careful modifications need to be made for the 2D materials. For example, in the case of 2D materials, instead of top contact, side contact of metal with 2D materials produces better results. However, depositing side contacts to 2D materials using the current industrial technology is challenging.

149

8  System Integration with Graphene for Silicon Optoelectronics

BEOL

Si CMOS chip

Cu

FEOL

150

W S n+

G

Transfer of graphene

D n+

Barrier

Cu

Dielectric

W S n+

S p+

G

D p+

n-well

p Si

Graphene

Barrier G

D n+

Dielectric

G

D p+

n-well

p Si

(a)

(b) Graphene encapsulation

Graphene patterning Graphene

Cu W S n+

G

D n+

Graphene

Barrier

Cu

Dielectric

W S n+

S p+

G

D p+

n-well

p Si

Barrier G

D n+

Dielectric

S p+

G

D p+

n-well

p Si

(c)

(d) Via etching

Via contact metallization Graphene

Graphene

Cu

Barrier

Cu

Barrier

W S n+

Dielectric

W S n+

Dielectric

G p Si

(e)

S p+

D n+

S p+

G

D p+

n-well

G p Si

D n+

S p+

G

D p+

n-well

(f)

Figure 8.7  Schematic diagram for the fabrication of the back end of the line integration. (a) Si‐CMOS chip. (b) Transfer of graphene. (c) Graphene patterning. (d) Graphene encapsulation. (e) Via etching. (f) Via contact metallization. FEOL, front end of line. Source: Reproduced with permission from Neumaier et al. [17]/Springer Nature.

8.5 ­Challenges and Prospective Traditional photodetectors, charge‐coupled devices, and CMOS sensors are widely used in the image‐sensing field. Some breakthrough progress has been made in photodetecting technology, but there are still some limitations. For example, the wavelength of the detected light is limited because of the materials used in photodetectors, and the intrinsic bandgap of silicon limits the wavelength range of the detected light in silicon‐based devices. In addition, the fill factor of the related devices is relatively low. Thus, silicon‐based CCDs and CMOS image sensors are generally not suitable for the application of invisible light, especially infrared light. Although the relative performance can be improved by the heterointegration detectors of metal silicide‐ based Schottky barrier, it still suffers from the disadvantages of complex structure, low fill factor, and high cost. Heterointegration of III–V or II–VI semiconductors with narrow bandgap helps to extend the response range of detectors in the spectrum but is complex technically and with high production cost. In addition to

8.5  ­Challenges and Prospectiv

silicon, other materials have difficulty in compatibility with the standard CMOS process of semiconductors, which limits the large‐scale application of new devices in the post‐Moore era. In this situation, graphene not only exhibits great potential for compatibility with silicon‐based CMOS technology but also has the advantages of broad spectral absorption, high carrier mobility, and strong field effects. Combining graphene with silicon‐based photodetectors is expected to improve the performances of devices, including responsivity, detected spectrum range, image‐sensing quality, and expanding application scenarios. Meanwhile, the silicon‐based CMOS process is the current mainstream microelectronics manufacturing technology which has the advantages of high integration density and good compatibility. Realizing the high‐ performance integration of graphene and CMOS platform will contribute to advancing the practical application of innovative graphene/silicon‐based photodetectors. The integration of CMOS‐graphene image‐sensing chips is a promising research direction, but the field is also full of challenges. As one of the key parameters of image‐sensing chips, the resolution of Gr‐based devices needs to be improved. Relevant further development depends on not only deeply exploring the working mechanisms of Gr materials and devices but also on related craft processes and structural design to reduce the size and structure of pixels without the side effect on performance, thereby raising the resolution and increasing the number of pixels. Besides, the integration‐level CMOS‐Gr image‐sensing chips need to raise. Compared with the analog front‐end module and the digital‐to‐analog conversion module, more modules are equally necessary to be designed in the chip, including the control module, the register group module, and the bias voltage‐generation module. In addition to the research on graphene pixels and dedicated ROICs, the mode and technology of integration between Gr devices and readout chips are also important. To achieve high‐quality interconnection between Gr devices and readout chips, three‐dimensional stacked integration processes have received a lot of attention from researchers. The three‐dimensional stacked integration process focuses on integrating chips vertically and has the advantages of high density, low delay, low parasitic impedance, and high passed yield. Currently, the mainstream three‐ dimensional stacking integration methods include flip‐chip welding technology and copper–copper interconnection. Among them, the copper–copper interconnection process has high precision, but the corresponding cost is also high, while the flip‐chip welding process has a relatively low cost, and the mode is flexible so that it is convenient for researchers to operate in the laboratory. The breakthrough in the above key points is conducive to not only promoting the in‐depth cooperation between graphene material devices and CMOS platforms, providing new solutions for the integrated system of image‐sensing chips, but also serving as a reliable reference for the integrated application of similar two‐dimensional material devices, to improve the performance of devices based on CMOS platforms and deeply explore the practicability of cutting‐edge 2D material devices in the post‐ Moore era.

151

152

8  System Integration with Graphene for Silicon Optoelectronics

­References 1 Furchi, M., Urich, A., Pospischil, A. et al. (2012). Microcavity‐integrated graphene photodetector. Nano Letters 12 (6): 2773–2777. 2 Luo, F., Zhu, M., Tan, Y. et al. (2018). High responsivity graphene photodetectors from visible to near‐infrared by photogating effect. AIP Advances 8 (11): 115106. 3 Chen, Z., Cheng, Z., Wang, J. et al. (2015). High responsivity, broadband, and fast graphene/silicon photodetector in photoconductor mode. Advanced Optical Materials 3 (9): 1207–1214. 4 Gan, X., Shiue, R.J., Gao, Y. et al. (2013). Chip‐integrated ultrafast graphene photodetector with high responsivity. Nature photonics 7 (11): 883–887. 5 Goossens, S., Navickaite, G., Monasterio, C. et al. (2017). Broadband image sensor array based on graphene–CMOS integration. Nature Photonics 11 (6): 366–371. 6 Sorianello, V., Midrio, M., Contestabile, G. et al. (2018). Graphene–silicon phase modulators with gigahertz bandwidth. Nature Photonics 12 (1): 40–44. 7 Yang, C.‐C., Chiu, K.‐C., Chou, C.‐T. et al. (eds.) (2016). Enabling monolithic 3D image sensor using large‐area monolayer transition metal dichalcogenide and logic/memory hybrid 3D+IC. 2016 IEEE Symposium on VLSI Technology. IEEE, Bengaluru, India. 8 Wale, M.J. (ed.) (1990). Self aligned, flip chip assembly of photonic devices with electrical and optical connections. 40th Conference Proceedings on Electronic Components and Technology. IEEE, Las Vegas, NV, USA. 9 Akinwande, D., Huyghebaert, C., Wang, C.‐H. et al. (2019). Graphene and two‐ dimensional materials for silicon technology. Nature 573 (7775): 507–518. 10 Das, N.C., Taysing‐Lara, M., Olver, K.A. et al. (2009). Flip chip bonding of 68 × 68 MWIR LED arrays. IEEE Transactions on Electronics Packaging Manufacturing 32 (1): 9–13. 11 Koppens, F., Mueller, T., Avouris, P. et al. (2014). Photodetectors based on graphene, other two‐dimensional materials and hybrid systems. Nature Nanotechnology 9 (10): 780–793. 12 Joshi, N., Hayasaka, T., Liu, Y. et al. (2018). A review on chemiresistive room temperature gas sensors based on metal oxide nanostructures, graphene and 2D transition metal dichalcogenides. Microchimica Acta 185 (4): 1–16. 13 Mortazavi Zanjani, S.M., Holt, M., Sadeghi, M.M. et al. (2017). 3D integrated monolayer graphene–Si CMOS RF gas sensor platform. NPJ 2D Materials and Applications 1 (1): 1–9. 14 Huang, L., Xu, H., Zhang, Z. et al. (2014). Graphene/Si CMOS hybrid hall integrated circuits. Scientific Reports 4 (1): 1–6. 15 Theuwissen, A.J. (2008). CMOS image sensors: state‐of‐the‐art. Solid‐State Electronics 52 (9): 1401–1406. 16 Golic, M., Walsh, K., and Lawson, P. (2003). Short‐wavelength near‐infrared spectra of sucrose, glucose, and fructose with respect to sugar concentration and temperature. Applied Spectroscopy 57 (2): 139–145. 17 Neumaier, D., Pindl, S., and Lemme, M.C. (2019). Integrating graphene into semiconductor fabrication lines. Nature Materials 18 (6): 525–529.

153

9 Graphene for Silicon Optoelectronic Synaptic Devices ­9.1  Introduction Optoelectronic synaptic devices mimicking the brain’s functionality are one of the emerging choices for next‐generation neuromorphic computing. Silicon‐based complementary metal–oxide–semiconductor (CMOS) is initially used to realize synaptic functionalities in neuromorphic computing [1–6]. These circuits require complex analog circuitry for synaptic operations and do not consist of inherent memory functionality. Later, memristors are used as synaptic devices, fulfilling both memory and processing features [7, 8]. Although memristors perfectly present various synaptic functionalities, they are mainly limited in device density, bandwidth, and connectivity. Light‐based synaptic devices offer multiple advantages while realizing low power consumption, ultrafast computing, high bandwidth, and robustness. Since Si is a primary choice for very‐large‐scale integration (VLSI) circuits, it inevitably plays a central role in developing large‐scale synaptic circuits for neuromorphic computing  [9]. Interest in neuro‐inspired computing chips was raised long back in the 1980s along with the progress of silicon‐based integrated technologies for VLSI [1]. These neuro‐inspired computing systems that emulate the functionality of the biological brain are based on analog spiking integrated systems and analogous to nonspiking mixed‐signal neural networks [6]. The spiking approach is introduced based on the similarity between the spiking behavior in the biological brain and the physical mechanism of integrated chips (ICs). In recent times, the emergence of artificial intelligence (AI) and the Internet of things (IOTs) demands exceptional computing speed and power efficiency that is applicable to synaptic devices as well. The performance of the existing IC architecture is limited by the “von Neumann bottleneck,” where separate units for memory and processing compromise the power consumption and decrease the operating speed as most of the energy and time are invested in communication between different units [10, 11]. A schematic representation of the von Neumann bottleneck is displayed in Figure 9.1a,b, where the program‐storage paradigm includes storing and fetching information between processing and memory units. The separate units for memory and processing significantly limit the efficiency and speed of the computing process and thus demand an entirely new Graphene for Post-Moore Silicon Optoelectronics, First Edition. Yang Xu, Khurram Shehzad, Srikrishna Chanakya Bodepudi, Ali Imran, and Bin Yu. © 2023 WILEY-VCH GmbH. Published 2023 by WILEY-VCH GmbH.

154

9  Graphene for Silicon Optoelectronic Synaptic Devices Memory

(a)

Processing unit

“Memory bottleneck”

(c)

(e)

Memory Fetch Processing unit

(b)

Synapses

Store

+

–

×

– (d)

Neurons

(f)

Figure 9.1  Schematics of computing architecture and prototypes. (a) Schematic representation of von Neumann architecture and the “memory bottleneck” between memory and processing units. (b) Schematic representation of the program storage. The data is collected from memory, processed in the processing unit, and stored back in the memory. (c, d) A neuro‐inspired architecture and computing, respectively. An array of synapse–neuron networks with random connections between electronic neurons via electronic synapses. (e) Schematic of the biological brain. (f) Neurons and synapses in the biological brain are depicted in the schematic. Spike signals from multiple neurons integrate in a neuron through synapses until it reaches a certain threshold and “fires.” Source: Reproduced with permission from Zhang et al. [10]/Springer Nature.

approach to the computing architecture. The neural‐inspired architecture overcomes these limitations by a well‐integrated synapse neuron core with connections between input and output electronic neurons through electronic or optoelectronic synapses, as shown in Figure 9.1c,d [10]. It is worth noting the architecture of the biological brain with neurons and synapses, where neurons integrate with multiple spiking signals from other neurons connected with synapses and fire after reaching a certain threshold as displayed in Figure 9.1e,f. Metal–oxide–semiconductor field‐effect transistor (MOSFET)‐based circuits that emulate neural behavior have been demonstrated with a nonspiking approach for more efficient matrix multiplication in neural network operation. Bell Laboratories fabricated a 12 × 12  unprogrammable synaptic resistive array to perform vector– matrix multiplication in a neural network. However, these pioneering models are constrained to small‐scale demonstrations due to limitations in the available contact device schemes that emulate biological behaviors. One of the crucial developments in this direction is demonstrating silicon neurons that inspire domain‐specific hardware and devices related to neuromorphic computing.

­9.2  Silicon Neurons Silicon analogue of biological neurons has been realized by the VLSI technology, providing different ion currents. These so‐called silicon neurons are the realistic

­9.  Silicon Neurons

substitution for mathematical and engineering abstractions of neurons [12]. Unlike previously demonstrated neural networks, the basic circuits of silicon neurons mimic the behavior of biological neurons instead of representing the biological responses based on digital and linear design principles of conventional circuits. Silicon neurons can operate >106 times as fast as their biological counterparts and thus have enormous potential to develop machines that interact with real‐world events similar to biological neurons [6, 13]. The communication between neurons happens with nerve impulses and self‐regenerated spikes of membrane voltage. While the nerve impulse is initiated, multiple ionic conductances are activated in the membrane, driven by a potential drop between the membrane voltage and the equilibrium potential of the ions. These silicon‐based circuits emulate these ionic currents in neurons. The nerve impulse itself is generated in the nervous system by the current carried by sodium and potassium ions [10]. The voltage‐dependent conductance of biological membranes consists of a sigmoidal conductance–voltage relationship that can be mimicked with the current–voltage relationship generated by CMOS transistors when arranged in a differential pair. We believe that the optoelectronic and nonvolatile memory functions and the mechanical robustness and flexibility of two‐dimensional (2D) materials open up new opportunities in artificial vision and neural networks  [14, 15]. A fair trade‐off is expected from the crucial performance parameters of photodetectors used as optical synaptic devices. For instance, the key features of artificial vision are higher contrast between various wavelengths of light and storing and processing of the information collected from the vision. To realize the contrast that matches the biological eye requires extremely low noise and dark current and higher sensitivity and responsivity. Storing and processing information need optical memory function with high mobility and switching speed, leading to sharp pattern recognition ability [15]. This chapter discusses these crucial parameters of the graphene and silicon photodetectors that provide a comprehensive design guideline for choosing suitable  2D materials and heterostructures from optical synaptic devices to optoelectronic neural networks. Incorporating graphene and related 2D materials can overcome these limitations and provide an alternative path to developing neuromorphic computing systems. The modern neuromorphic computing architecture must fulfill four major benchmarks  –  computing density, energy efficiency, computing accuracy, and on‐chip learning capability – to succeed in edge computing and IOT. In view of realizing faster, low power‐consuming memory integration systems, many researchers are trying to develop devices that can include optical sensing and synaptic and memory functions to process information analogous to the brain. The widely used neuromorphic computing circuity is based on silicon‐CMOS architecture which requires additional clocking to reset the synaptic weights, leading to higher power consumption [3, 10, 11]. Recently, J. Feldmann et al. demonstrated an all‐optical version of a neural network system with better energy efficiency than most conventional computing approaches. The all‐optical implementation of deep neural networks (DNNs) that incorporates many layers of neurons and optical synapses mimicking the connections between neurons in this work inspires such implementation with all‐optical 2D material‐based neural networks. 2D materials

155

156

9  Graphene for Silicon Optoelectronic Synaptic Devices

and their heterostructures exhibit broadband photodetection in terms of optical sensing and higher ON/OFF ratio with fast switching speeds and multibit data storage when configured into a memory device. The optical transparency and mechanical flexibility of 2D materials are ideal for developing optical synaptic memory devices leading to artificial eyes. Large‐scale neuromorphic computing systems require the right combination of mature technology of Si integrated with exciting optoelectronic features of 2D materials leading to highly efficient and faster synaptic functionalities. The standard approach for circuits to solve complex and structured mathematical problems is the von Neumann (VN) architecture [10]. However, with increasing demand for intelligent computing combined with low power consumption, VN architecture’s benefits have begun to fade out. Specifically, the separate functional units used for memory and central processing unit (CPU) in VN architecture are incompatible with large and complex neuromorphic computing systems that enormously increase power consumption. In addition, these traditional architectures cannot fulfill the need to incorporate AI in complex circuits that can independently handle the unplanned, dynamic response to real‐time problems such as the human brain. Therefore, the modern computing system must be equally efficient as the human brain and able to handle intelligent activities such as self‐learning, language processing, and image recognition and overcome the limitations of VN architecture. Mimicking various biological functionalities is pivotal in neuromorphic computing and thus the emerging interest in developing synaptic devices. As mentioned above, initial synaptic devices are based on CMOS technology, and many synaptic functionalities have been rapidly developed for neuromorphic computing. Large‐scale circuits and chips have been designed for this purpose but approaching the theoretical limit of miniaturization and high power consumption.

­9.3  Synaptic Devices Synaptic devices function to fulfill a certain synaptic weight, which is essentially the connection strength between pre‐ and postsynaptic neurons. In principle, a biological synapse is mimicked by one of the device parameters in synaptic devices, such as a change in resistance or conductivity. Synaptic weight change is called synaptic plasticity, critical to neural activities. Some of the necessary forms of synaptic plasticity are short‐term plasticity (STP), long‐term plasticity (LTP), spike timing‐ dependent plasticity (STDP), and spike rating‐dependent plasticity (SRDP). The recognition and processing of neural response decides STP or LTP synaptic responses or even lead to STP‐to‐LTP transitions [14, 16, 17]. The STP of synaptic devices can be evaluated by the change in paired‐pulse facilitation (PPF) or paired‐pulse depression (PPD) index versus the time interval between two successive stimuli (∆t). The PPF index decreases with ∆t as represented by the below equation: PPF decay

c1 exp

t

t1

c2 exp

t

t2



(9.1)

­9.  Silicon Optoelectronic Synaptic Devices ΔW

ΔW

Δt

Δt (a)

(b)

Figure 9.2  Different types of spike‐time‐dependent plasticity (STDP) in neural systems are crucial for learning neural systems’ rules. (a) Asymmetric STDP is often observed in cortical neurons. (b) Symmetric STDP where synaptic weight change (∆wt) depends on the proximity of the spike. Source: Reproduced with permission from Whittington and Bogacz [18]/ Elsevier Ltd.

Here, t1 and t2 are the characteristic relaxation times of PPF decay, and c1 and c2 are the initial facilitation magnitudes of rapid and slow phases, respectively. It is important to note that in the biological synapse, t2 is often an order of magnitude higher than t1, which can be used as a metric to develop ideal optical synaptic devices [16]. These concepts of synaptic plasticity are briefly discussed before going into optoelectronic synaptic devices based on silicon and graphene. Learning and memory that persists for several hours in the brain are enabled by LTP. In a biological synapse, synaptic weight change also has a polarity, showing long‐term potentiation and depression, representing excitatory and inhibitory changes in the synaptic weight. The essential characteristics of biological synapses include STP and LTP, where STP can be turned into LTP by a few rounds of external information training. However, to interpret the interaction between neurons, it is crucial to explore the connection between multiple synapses via STDP or SRDP. Here, STDP relies on the temporal order and the interval between spikes, where different forms can be derived, as shown in Figure 9.2, based on the various changes in the synaptic weight (∆wt % ) caused by either varying spike order or spike interval (∆t). Potentiation (depression) of the synaptic weight can be represented when the presynaptic spike is observed before the postsynaptic weight, i.e. ∆t > 0(∆t  0 Vpre

Δt < 0

1

2

3

4

5

6

0.1 mW 1 mW 5 mW

50 40

Vpost

30

tpre tpost Pre spike Post spike

20 10

Synapse

0 –10

Vpre–Vpost (c)

0

Applied bias (V)

60

p-Si ΔSW (%)

158

–50 –40 –30 –20 –10 0

(d)

10 20 30 40 50

Spike timing Δt (ms)

Figure 9.3  Basic configuration of synaptic devices based on planar photoconductive silicon synapse and Si–HfO2 heterostructure. (a) Schematic of a planar photoconductive synapse unit and the corresponding (b) I–V characteristics. (c) Schematic of the device. (d) STDP under different illumination conditions are displayed. The relative change of synaptic weight (ΔSW%) is defined as (I2−I1)/I1. Source: Reproduced with permission from Yin et al. [6]/IOP Publishing Ltd.

­9.  Graphene for Silicon Synaptic Devices

displayed in Figure 9.3d. Here, the presynaptic electrical spikes applied at Si electrodes are coupled with optical spikes, followed by the postsynaptic spikes applied at Pt electrodes. In these devices, the synaptic weight (∆wt % ) can be modulated by the power density of optical spikes. With transient optoelectronic stimulus, nonvolatile memory and photoresponse are realized in this device, whereas STP and LTP functionalities are not realized.

­9.5  ORAM Synaptic Devices Optical random‐access memory (ORAM)‐based synaptic devices consist of neuromorphic computing and the optical sensory. These systems can respond to optical stimuli and perform a series of light‐tunable synaptic functions [18–20]. These circuits can be designed with optically controllable switching in either volatile or nonvolatile operation. The ORAM synaptic systems with light‐tunable synaptic behavior are initially realized by the heterostructures of graphene and carbon nanotubes (CNTs)  [15]. The channel conductance in this heterostructure device can be tuned by either the optical pulse or the gate voltage. The different conductive states in this device are analogous to the various forms of biological synaptic weight. WSe2/boron‐doped Si nanocrystals act as the ORAM synaptic system with broad absorption spectra ranging from ultraviolet (UV) to near‐infrared (NIR) while consuming less power (∼75 fJ), laying steps toward artificial visual intelligence. Image sensors have been transformed into an ANN with ultrafast sensing [21]. This can be scaled into a chip with a throughput approaching 20 mbps and, therefore, compatible with various applications. A monolithic CMOS readout IC developed by integrating CVD graphene with Si is of high resolution, broadband, and can be used as a digital camera for a broad wavelength range (0.3–2 μm). Thus, it finds applications in various areas such as integrated photonics and high‐frequency electronics. There is a growing interest for multifunctional synaptic devices that can accommodate both information processing and memory functions. Optical resistive random‐access memory (ORRAM) circuits enable image sensing and memory functions with neuromorphic visual sensing, including preprocessing images and a higher image recognition rate in the consecutive processing steps. The strong light–matter interaction and ultrafast carrier dynamics of 2D materials enable 2D‐based ORAM synaptic systems with ultrafast, broadband photoresponse, multibit data storage, and enhanced energy efficiency.

­9.6  Graphene for Silicon Synaptic Devices The advantages of using graphene in silicon synaptic devices can be drawn from the fundamental advantage of integrating graphene with Si CMOS electronics. Graphene photodetectors, modulators, and switches offer wide spectral bandwidth, high speed, low power consumption, and monolithic integration with CMOS

159

160

9  Graphene for Silicon Optoelectronic Synaptic Devices

electronics that include a similar fabrication process [9]. Graphene and 2D materials offer a wide range of physical characteristics that can potentially improve the performance of Si electronics and memory devices. Therefore, similar benefits can be expected for synaptic devices as well. For instance, chemical doping in gating conventional semiconductors can be avoided as a graphene layer can be gated by another. Such device designs can be used for multifunctional optoelectronic synaptic devices with photodetection and memory functions. Graphene’s broadband characteristics can be used to extend the functionality of Si‐ and SiN‐based devices beyond telecommunication wavelengths. In addition, integrating graphene‐based photodetectors and modulators with conventional large‐scale CMOS circuity is not difficult as these devices can operate within 2 V of drive voltage [9]. For instance, phase and absorption modulators of graphene have been integrated with Si CMOS circuits with a high rate of binary transmission of 10 Gbit s−1. Ultrafast photocarrier dynamics of photothermoelectric (PTE) effect in graphene‐enabled photodetection with the zero‐bias operation have been demonstrated, where operating speeds exceed above 180 Gbit s−1. A fully integrated image sensor array of 388 × 288 pixels where graphene integrated with Si CMOS readout circuit has been realized by employing graphene‐quantum dot photodetectors, operating from visible to infrared range (300–2000 nm). Therefore, high‐density synaptic device schemes based on the PTE effect can significantly improve the operation bandwidth, while minimizing the power consumption [9, 15].

­9.7  Synaptic Phototransistor In a synaptic phototransistor, the channel conductance is interpreted as the synaptic weight, where a light pulse acts as external stimuli or presynaptic spikes. The gate‐ induced tunability of charge transport properties of the channel is the crucial function for dynamically tunable synaptic plasticity. A phototransistor with a combination of atomically thin single‐walled carbon nanotubes (SWNTs) and graphene exhibits STP achieved via charge transfer between graphene and SWNT [18]. This is confirmed as this behavior is absent in the case of optically gated pure CNT‐ programable devices. The Gr–CNT heterostructure is chosen due to its inherent robustness, enhanced light absorption, and different ways to tune the charge transport properties. As a result of the rich charge‐trapped interface between the Gr/CNT hybrid film and the SiO2/Si substrate (Figure 9.4a), the synapse incorporates optoelectronic nonvolatile memory, exhibiting retention time relied on the applied gate bias displayed in Figure 9.4b. This device perfectly emulates the LTP by optical coupling mediated by the trapped charge at the interface. Optical logic operation can be realized with this device by exciting an array of these devices with multiples light spikes, such as an advanced optical spike processing. The inhibitory postsynaptic current (IPSC) is enabled by the presynaptic spike at zero gate voltage in this device. Then, immediately after applying the excitation spike, it reaches its peak value and slowly returns to its initial value. This dynamic response well emulates the IPSC process in the biological synapse. In the same way, by setting the gate bias to 20 V,

­9.  Synaptic Phototransistor 1

Light spike

eu

ra

VDS

l

ig

na l

Source

SiO

ina

ls

2

0

Excitation VG = 0 V

–2 1.0

–2 –4

1.5 2.0 Time (s)

Inhibition

Iph (µA)

IDS

ΔPSC (µA)

N

Drain

te ter m

Light spike

0

–1

Axon hillock

Ga

Iph (µA)

2

VG

VG = 20 V

1 0 1.0

Dendrites

(a)

2

–50

(b)

–25

0

1.5 2.0 Time (s)

25

50

Gate bias (V)

3

 –5  –6

(c)

0

20

40

60

80

Spike duration (ms)

100

Illumination Evac Ec VG = Vcross Ev

2

Evac Ec

Ev

Excitation

4

 –4

Dark

Inhibition

5  –3

EPSC amplitude (µA)

IPSC amplitude (µA)

 –2

(d)

Figure 9.4  Light‐stimulated synaptic Gr/SWNT hybrid phototransistor. (a) Schematic of the synapse of the device and the corresponding representation of the neural signal transmission in a neuron. (b) The change in PSC is induced by a presynaptic light spike (50 μW, 5 ms) while varying the gate bias from −50 to +50 V. Typical changes in IPSC and EPSC are activated by the light spike when Vg = 0 V and Vg = 20 V, as shown in the top and bottom insets, respectively. (c) IPSC and EPSC for an inhibitory synapse (blue line with blue balls, Vg = 0 V) and an excitatory synapse (red line with red balls, Vg = 20 V) when the light spike power is 50 μW are displayed as a function of spike duration. The inset is the magnified image of the device (scale bar, 20 μm). (d) The interface band alignment of the Gr/SWNTs. Electrons transferred from the SWNTs layers under illumination induce changes in the built‐in electric field. Purple solid and open circles represent electrons and holes, respectively. Source: Reproduced with permission from Qin et al. [19]/IOP Publishing Ltd.

the input light spike triggers an excitatory postsynaptic current (EPSC) in biological systems (Figure 9.4b, bottom inset). The recovery of channel conductance is fitted with the biexponential decay with the relaxation times (τ1  =  38  ms and τ1  =  528  ms)  [18]. In neuroscience, it is crucial to realize synaptic modification based on the spike time or spike rate between stimuli. The maximum values of ∆IPSC and ∆EPSC are varying between 5 and 100 ms. It is worth noting that the expected spike duration observed in this device is very similar to the stimulation in biological neurons, which is in the order of tens of milliseconds (Figure 9.4c). The adjustable gate plasticity is based on the band alignment between graphene and CNTs, as displayed in Figure 9.4d. PPF refers to a synaptic process where the two closely spaced stimuli activate the enhancement of transmitter release. Along with the gate‐tunable control, this synaptic device is also programmable by the excitation pulses. Variation in synaptic weight in these devices can be triggered by the duration of the presynaptic light spike under a negative gate bias. The resistance change triggered by the excitation does not relax to its original value at Vg = −20 V, indicating LTP formation, where it can be suppressed by applying a lower negative

161

9  Graphene for Silicon Optoelectronic Synaptic Devices

bias (