Neuromorphic Engineering: from Materials to Device Application
J. Joshua Yang,Julie Grollier,R. Stanley Williams,Ru Huang
DOI: https://doi.org/10.1002/adma.202305078
IF: 29.4
2023-01-01
Advanced Materials
Abstract:With the advent of the Internet of Things, an exponential growth in the amount of data has imposed critical requirements on the energy efficiency and processing speed for data-centric tasks, which calls for changes to information-processing paradigms to overcome the von Neumann bottleneck of traditional computing systems. The brain, on the other hand, remains the most efficient computing system known, thanks to its in-memory, parallel and analog computing paradigm. Following some neuroscience principles not yet fully revealed, the brain is also much more efficient at learning and expressing creativity. Neuromorphic engineering, inspired by what little we know about how the brain computes and learns, has shown promise in greatly improving the energy efficiency and computing throughput of machines. Building neuromorphic artificial systems demands complementary research on materials, devices, circuits, architectures, algorithms, systems and applications, which are all touched upon in this special issue that includes 3 Perspectives, 9 Reviews and 12 Research Articles. These 24 contributions can be roughly grouped into three categories. The first group of papers explores different types of materials that can enable efficient computing systems. The second group examines physical phenomena and properties for neuromorphic computing beyond the traditional electrical domain. The third group utilizes new materials and phenomena for a variety of unconventional computing paradigms. Inspired by the fact that migration dynamics of cations (e.g., Ca2+, Mg2+, Na+, K+) play a critical role in the functions of biological networks and components, electrochemical material systems carry a heavy weight in the first group of papers. Prof. Bilge Yildiz and co-workers (article number 2205169) conducted a comprehensive review on electrochemical ionic synapses based on three-terminal devices that operate by electrochemical and dynamic insertion/extraction of ions to control the electronic conductivity of a channel in a single solid-solution phase. Dr. Alec Talin and co-workers (article number 2204771) provide an in-depth perspective on the same type of devices, covering the history, the recent progress in devices spanning organic, inorganic, and 2D materials, circuits, architectures, the rich portfolio of challenging, fundamental questions, and how such devices can be harnessed to realize a new paradigm for low-power neuromorphic computing. Prof. Ilia Valov and co-workers (article number 2301924) present a review on common electrochemical fundamentals of biological and artificial cells, emphasizing their similarities with redox-based memristive devices as well as the driving forces behind the functionalities and how to control them. Dr. Alec Talin and co-workers (article number 2207595) in a research article expand the materials playground for neuromorphic devices to include a mixed-valence inorganic 3D coordination framework, a ruthenium Prussian blue analog (RuPBA), for flexible and biocompatible artificial synapses that reversibly switch conductance by more than four orders of magnitude based on electrochemically tunable oxidation state. Metal oxides represent another large class of neuromorphic materials. Prof. Shriram Ramanathan and co-workers (article number 2203352) perform a comprehensive review on opportunities presented by complex oxides, a class of electronic ceramic materials with properties that can be elegantly tuned by doping, electron interactions, and a variety of external stimuli near room temperature. Prof. Sarbajit Banerjee and co-workers (article number 2205294) focus on VO2 as a metal–insulator transition oxide and review the strategies for tuning the transformation characteristics of VO2 based on modification of material properties, interfacial structure, and field couplings. Prof. Thomas Mikolajick and co-workers (article number 2206042) extensively summarize hafnium-oxide-based ferroelectrics, ranging from the material science principles to the current status of memory devices and neural-network applications. Materials beyond oxides have also been actively investigated for neuromorphic engineering. Dr. Jeffrey Blackburn and co-workers (article number 2205459) survey the recent progress toward memory and neuromorphic functionality in metal halide perovskite materials and devices where photons are used as a critical degree of freedom for switching, memory, and neuromorphic functionality. Prof. Sreetosh Goswami and co-workers (article number 2204551) report molecular memristors with complex interactions between molecules and ions, offering a multidimensional parameter space to embed, access, and optimize material functionalities for target-specific applications. Other physical phenomena than traditional electrical properties of materials can be adopted to enable efficient computing and are explored in 7 articles in this special issue. For instance, the static conductance levels of RRAMs only enable machine-learning accelerator applications, while the rich nonlinear dynamics of memristors can be exploited to implement neuroscience principles in hardware with a much-reduced consumption in energy, time and area. Prof. R. Stanley Williams and co-workers (article number 2205451) use the VO2 memristor as an exemplar system to demonstrate how to correctly characterize the electro-thermal properties of memristors and construct a device model that can accurately predict electrical and thermal conductivities and capacitances and locally active dynamics. Prof. Ivan K. Schuller and co-workers in a review (article number 2205098) shed light on the present state of thermal management of neuromorphic computing technology and the challenges and opportunities of the energy-efficient implementation of neuromorphic devices. Prof. Han Wang and co-workers (article number 2205047) further review the neuronal dynamics by surveying the reports of artificial neuronal devices based on emerging volatile switching materials from the perspective of the demonstrated neuron models and their applications. In a research article, Prof. Huaqiang Wu (article number 2203684) and co-workers take advantage of the rich internal ionic dynamics in memristors to construct a dendritic neuron unit for realizing high-efficiency spatial–temporal information processing. In another research article, Dr. Abu Sebastian and co-workers (article number 2201238) identify an unusual non-ideality of phase change memories that can impact computational precision, namely the bias-polarity-dependent current flow, and provide a scheme to compensate for it. Combining photonics with electronics has become a popular approach to integrate sensing and computing together for efficient sensing-processing systems. Prof. Ru Huang and co-workers (article number 2204844) report a highly linear, light-tunable, cross-talk-free, and silicon-compatible one-phototransistor–one-memristor optic memristor for the implementation of an optical artificial neural network. Prof. Peng Zhou and co-workers (article number 2203909) demonstrate an ultrafast non-volatile photonic memory based on a scandium-doped antimony telluride thin film, which shows the fastest write/erase speed (2 ns) ever reported in integrated phase-change photonic devices. A variety of unconventional computing paradigms enabled by the new materials and phenomena are reported or reviewed in the special issue. Dr. James Bradley Aimone and co-workers (article number 2204569) share their perspective on how to adopt the brain's ubiquitous stochasticity as an additional source of inspiration for expanding the reach of neuromorphic computing to probabilistic neural-computing applications. Prof. Yang Chai and co-workers (article number 2203830) offers their perspective on hardware implementation of in-sensor computing paradigm at the device and array levels to reduce data transfer and decrease the high computing complexity by processing data locally. Prof. Supratik Guha and co-workers (article number 2204944), with an over-arching co-design viewpoint, assess the use of crossbar-based computing-in-memory for neural networks, connecting the material properties and the associated design constraints and demands to application, architecture, and performance. Prof. J. Joshua Yang and co-workers (article number 2206648) demonstrate the first field-programmable analog array for analog computing, where memristors play multiple key roles, including routing networks, neural networks, and analog filters. Prof. Daniele Ielmini and co-workers (article number 2205381) report a reservoir computing system where the input pattern is applied as a stimulation of the MoS2-based charge-trap memories, while the output current after stimulation is processed by a feedforward readout network. Prof. Qiangfei Xia and co-workers (article number 2204778) implement the hierarchy of time surfaces algorithm for event-driven patten recognition using diffusive memristors with a much-improved uniformity in their dynamical relaxation time. Prof. Thirumalai Venkatesan and co-workers (article number 2206128) fabricate crossbar arrays with up to 64 molecular memristors to experimentally demonstrate 8-bit serial and 4-bit parallel adders that operate for thousands of measurement cycles with an estimated error probability of 10−16. Prof. DonHee Ham and co-workers (article number 2205096) experimentally construct an aqueous analog multiply–accumulate machine using a 16 × 16 array of ionic transistors, which manipulate ions in aqueous milieu for signal processing, like in biological circuits, potentially leading to a bioinspired information processing. To close, we appreciate the kind support from the editorial team of Advanced Materials, in particular Dr. Floriano Cuccureddu and Dr. Babak Mostaghaci. We are also very grateful to all the authors, who shared their latest research results and magnificent insights in this exciting special issue of Neuromorphic Engineering: From Materials to Device Application. The authors appreciate the support of TetraMem Inc. (www.TetraMem.com) for the cover image of the special issue. The authors declare no conflict of interest. J. Joshua Yang is a professor of the Department of Electrical and Computer Engineering at the University of Southern California. His current research interest is post-CMOS hardware for neuromorphic computing, machine learning, and artificial intelligence. He is the Founding Chair of the IEEE Neuromorphic Computing Technical Committee. He serves as an associate editor of Science Advances, AAAS. He is an IEEE Fellow and a National Academy of Inventors (NAI) Fellow. Julie Grollier is a Research Director at the joint CNRS, Thales, and Université Paris-Saclay laboratory. Her current research interests are in the field of neuromorphic computing, at the intersections between nanodevice physics, system physics, and machine learning. She serves in the editorial board of Physical Review Applied. She is an APS and IEEE Fellow. She has been awarded the mid-career award of the IEEE Magnetism Society, the Silver medal of CNRS, the Jacques Herbrand and Joliot–Curie prizes of the French Academy of Science. R. Stanley Williams is a professor in the Department of Electrical and Computer Engineering and the Holder of the Hewlett Packard Enterprise Chair at the Texas A&M University. His current research interests are in the areas of nonlinear dynamics of materials and devices, and their utilization for emulating biological processes in neuromorphic computing. He has been awarded the Feynman Prize in Nanotechnology, Julius Springer Award in Applied Physics, Glenn T. Seaborg Medal for Chemistry, Herman Bloch Medal for Industrial Research, and Diels–Planck Medal for Nanotechnology. Ru Huang is a Boya Chair Professor at Peking University. She presently serves as President of Southeast University. Her current research interests include emerging functional integrated devices, emerging nonvolatile memories, neuromorphic computing devices, and new-paradigm computing. She is an elected Academician of the Chinese Academy of Sciences, an IEEE Fellow, and a TWAS Fellow. She served as the General Chair/Co-Chair of many influential international conferences. She is an Associate Editor-in-Chief of Science China: Information Sciences.