A Sparse and Spike-Timing-Based Adaptive Photoencoder for Augmenting Machine Vision for Spiking Neural Networks

A Crayfish-Inspired Sensor Fusion Platform for Super Additive Integration of Visual, Chemical, and Tactile Information

The concept of cross-sensor modulation, wherein one sensor modality can influence another's response, is often overlooked in traditional sensor fusion architectures, leading to missed opportunities for enhancing data accuracy and robustness. In contrast, biological systems, such as aquatic animals like crayfish, demonstrate superior sensor fusion through multisensory integration. These organisms adeptly integrate visual, tactile, and chemical cues to perform tasks such as evading predators and locating prey. Drawing inspiration from this, we propose a neuromorphic platform that integrates graphene-based chemitransistors, monolayer molybdenum disulfide (MoS2) based photosensitive memtransistors, and triboelectric tactile sensors to achieve "Super-Additive" responses to weak chemical, visual, and tactile cues and demonstrate contextual response modulation, also referred to as the "Inverse Effectiveness Effect." We hold the view that integrating bio-inspired sensor fusion principles across various modalities holds promise for a wide range of applications.

An artificial visual neuron with multiplexed rate and time-to-first-spike coding

Human visual neurons rely on event-driven, energy-efficient spikes for communication, while silicon image sensors do not. The energy-budget mismatch between biological systems and machine vision technology has inspired the development of artificial visual neurons for use in spiking neural network (SNN). However, the lack of multiplexed data coding schemes reduces the ability of artificial visual neurons in SNN to emulate the visual perception ability of biological systems. Here, we present an artificial visual spiking neuron that enables rate and temporal fusion (RTF) coding of external visual information. The artificial neuron can code visual information at different spiking frequencies (rate coding) and enables precise and energy-efficient time-to-first-spike (TTFS) coding. This multiplexed sensory coding scheme could improve the computing capability and efficacy of artificial visual neurons. A hardware-based SNN with the RTF coding scheme exhibits good consistency with real-world ground truth data and achieves highly accurate steering and speed predictions for self-driving vehicles in complex conditions. The multiplexed RTF coding scheme demonstrates the feasibility of developing highly efficient spike-based neuromorphic hardware.

A Spiking Artificial Vision Architecture Based on Fully Emulating the Human Vision

Intelligent vision necessitates the deployment of detectors that are always-on and low-power, mirroring the continuous and uninterrupted responsiveness characteristic of human vision. Nonetheless, contemporary artificial vision systems attain this goal by the continuous processing of massive image frames and executing intricate algorithms, thereby expending substantial computational power and energy. In contrast, biological data processing, based on event-triggered spiking, has higher efficiency and lower energy consumption. Here, this work proposes an artificial vision architecture consisting of spiking photodetectors and artificial synapses, closely mirroring the intricacies of the human visual system. Distinct from previously reported techniques, the photodetector is self-powered and event-triggered, outputting light-modulated spiking signals directly, thereby fulfilling the imperative for always-on with low-power consumption. With the spiking signals processing through the integrated synapse units, recognition of graphics, gestures, and human action has been implemented, illustrating the potent image processing capabilities inherent within this architecture. The results prove the 90% accuracy rate in human action recognition within a mere five epochs utilizing a rudimentary artificial neural network. This novel architecture, grounded in spiking photodetectors, offers a viable alternative to the extant models of always-on low-power artificial vision system.

Hierarchies in Visual Pathway: Functions and Inspired Artificial Vision

The development of artificial intelligence has posed a challenge to machine vision based on conventional complementary metal-oxide semiconductor (CMOS) circuits owing to its high latency and inefficient power consumption originating from the data shuffling between memory and computation units. Gaining more insights into the function of every part of the visual pathway for visual perception can bring the capabilities of machine vision in terms of robustness and generality. Hardware acceleration of more energy-efficient and biorealistic artificial vision highly necessitates neuromorphic devices and circuits that are able to mimic the function of each part of the visual pathway. In this paper, we review the structure and function of the entire class of visual neurons from the retina to the primate visual cortex within reach (Chapter 2) are reviewed. Based on the extraction of biological principles, the recent hardware-implemented visual neurons located in different parts of the visual pathway are discussed in detail in Chapters 3 and 4. Furthermore, valuable applications of inspired artificial vision in different scenarios (Chapter 5) are provided. The functional description of the visual pathway and its inspired neuromorphic devices/circuits are expected to provide valuable insights for the design of next-generation artificial visual perception systems.

Three-dimensional integration of two-dimensional field-effect transistors

In the field of semiconductors, three-dimensional (3D) integration not only enables packaging of more devices per unit area, referred to as 'More Moore'1 but also introduces multifunctionalities for 'More than Moore'2 technologies. Although silicon-based 3D integrated circuits are commercially available3-5, there is limited effort on 3D integration of emerging nanomaterials6,7 such as two-dimensional (2D) materials despite their unique functionalities7-10. Here we demonstrate (1) wafer-scale and monolithic two-tier 3D integration based on MoS2 with more than 10,000 field-effect transistors (FETs) in each tier; (2) three-tier 3D integration based on both MoS2 and WSe2 with about 500 FETs in each tier; and (3) two-tier 3D integration based on 200 scaled MoS2 FETs (channel length, LCH = 45 nm) in each tier. We also realize a 3D circuit and demonstrate multifunctional capabilities, including sensing and storage. We believe that our demonstrations will serve as the foundation for more sophisticated, highly dense and functionally divergent integrated circuits with a larger number of tiers integrated monolithically in the third dimension.

A Butterfly-Inspired Multisensory Neuromorphic Platform for Integration of Visual and Chemical Cues

Unisensory cues are often insufficient for animals to effectively engage in foraging, mating, and predatory activities. In contrast, integration of cues collected from multiple sensory organs enhances the overall perceptual experience and thereby facilitates better decision-making. Despite the importance of multisensory integration in animals, the field of artificial intelligence (AI) and neuromorphic computing has primarily focused on processing unisensory information. This lack of emphasis on multisensory integration can be attributed to the absence of a miniaturized hardware platform capable of co-locating multiple sensing modalities and enabling in-sensor and near-sensor processing. In this study, this limitation is addressed by utilizing the chemo-sensing properties of graphene and the photo-sensing capability of monolayer molybdenum disulfide (MoS2 ) to create a multisensory platform for visuochemical integration. Additionally, the in-memory-compute capability of MoS2 memtransistors is leveraged to develop neural circuits that facilitate multisensory decision-making. The visuochemical integration platform is inspired by intricate courtship of Heliconius butterflies, where female species rely on the integration of visual cues (such as wing color) and chemical cues (such as pheromones) generated by the male butterflies for mate selection. The butterfly-inspired visuochemical integration platform has significant implications in both robotics and the advancement of neuromorphic computing, going beyond unisensory intelligence and information processing.

Coexistence of Electrochromism and Bipolar Nonvolatile Memory in a Single Viologen

Viologens are fascinating redox-active organic compounds that have been widely explored in electrochromic devices (ECDs). However, the combination of electrochromic and resistive random-access memory in a single viologen remains unexplored. We report the coexistence of bistate electrochromic and single-resistor (1R) memory functions in a novel viologen. A high-performance electrochromic function is achieved by combining viologen (BzV2+2PF6) with polythiophene (P3HT), enabling a "push-pull" electronic effect due to the efficient intermolecular charge transfer in response to an applied bias. The ECDs show high coloration efficiency (ca. 1150 ± 10 cm2 C-1), subsecond switching time, good cycle stability (>103 switching cycles), and low-bias operation (±1.5 V). The ECDs require low power for switching the color states (55 μW cm-2 for magenta and 141 μW cm-2 for blue color). The random-access memory devices (p+2-Si/BzV2+2PF6/Al) exhibit distinct low and high resistive states with an ON/OFF ratio of ∼103, bipolar and nonvolatile characteristics that manifest good performances, and "Write"-"Read"-"Erase" (WRE) functions. The charge conduction mechanism of the RRAM device is elucidated by the Poole-Frenkel model where SET and RESET states arise at a low transition voltage (VT = ±1.7 V). Device statistics and performance parameters for both electrochromic and memory devices are compared with the literature data. Our findings on electrochromism and nonvolatile memory originated in the same viologen could boost the development of multifunctional, smart, wearable, flexible, and low-cost optoelectronic devices.

An all 2D bio-inspired gustatory circuit for mimicking physiology and psychology of feeding behavior

Animal behavior involves complex interactions between physiology and psychology. However, most AI systems neglect psychological factors in decision-making due to a limited understanding of the physiological-psychological connection at the neuronal level. Recent advancements in brain imaging and genetics have uncovered specific neural circuits that regulate behaviors like feeding. By developing neuro-mimetic circuits that incorporate both physiology and psychology, a new emotional-AI paradigm can be established that bridges the gap between humans and machines. This study presents a bio-inspired gustatory circuit that mimics adaptive feeding behavior in humans, considering both physiological states (hunger) and psychological states (appetite). Graphene-based chemitransistors serve as artificial gustatory taste receptors, forming an electronic tongue, while 1L-MoS2 memtransistors construct an electronic-gustatory-cortex comprising a hunger neuron, appetite neuron, and feeding circuit. This work proposes a novel paradigm for emotional neuromorphic systems with broad implications for human health. The concept of gustatory emotional intelligence can extend to other sensory systems, benefiting future humanoid AI.

A bio-inspired visuotactile neuron for multisensory integration

Multisensory integration is a salient feature of the brain which enables better and faster responses in comparison to unisensory integration, especially when the unisensory cues are weak. Specialized neurons that receive convergent input from two or more sensory modalities are responsible for such multisensory integration. Solid-state devices that can emulate the response of these multisensory neurons can advance neuromorphic computing and bridge the gap between artificial and natural intelligence. Here, we introduce an artificial visuotactile neuron based on the integration of a photosensitive monolayer MoS2 memtransistor and a triboelectric tactile sensor which minutely captures the three essential features of multisensory integration, namely, super-additive response, inverse effectiveness effect, and temporal congruency. We have also realized a circuit which can encode visuotactile information into digital spiking events, with probability of spiking determined by the strength of the visual and tactile cues. We believe that our comprehensive demonstration of bio-inspired and multisensory visuotactile neuron and spike encoding circuitry will advance the field of neuromorphic computing, which has thus far primarily focused on unisensory intelligence and information processing.

A Peripheral-Free True Random Number Generator Based on Integrated Circuits Enabled by Atomically Thin Two-Dimensional Materials

A true random number generator (TRNG) is essential to ensure information security for Internet of Things (IoT) edge devices. While pseudorandom number generators (PRNGs) have been instrumental, their deterministic nature limits their application in security-sensitive scenarios. In contrast, hardware-based TRNGs derived from physically unpredictable processes offer greater reliability. This study demonstrates a peripheral-free TRNG utilizing two cascaded three-stage inverters (TSIs) in conjunction with an XOR gate composed of monolayer molybdenum disulfide (MoS2) field-effect transistors (FETs) by exploiting the stochastic charge trapping and detrapping phenomena at and/or near the MoS2/dielectric interface. The entropy source passes the NIST SP800-90B tests with a minimum normalized entropy of 0.8780, while the generated bits pass the NIST SP800-22 randomness tests without any postprocessing. Moreover, the keys generated using these random bits are uncorrelated with near-ideal entropy, bit uniformity, and Hamming distances, exhibiting resilience against machine learning (ML) attacks, temperature variations, and supply bias fluctuations with a frugal energy expenditure of 30 pJ/bit. This approach offers an advantageous alternative to conventional silicon, memristive, and nanomaterial-based TRNGs as it obviates the need for extensive peripherals while harnessing the potential of atomically thin 2D materials in developing low-power TRNGs.

Dielectrics for Two-Dimensional Transition-Metal Dichalcogenide Applications

Despite over a decade of intense research efforts, the full potential of two-dimensional transition-metal dichalcogenides continues to be limited by major challenges. The lack of compatible and scalable dielectric materials and integration techniques restrict device performances and their commercial applications. Conventional dielectric integration techniques for bulk semiconductors are difficult to adapt for atomically thin two-dimensional materials. This review provides a brief introduction into various common and emerging dielectric synthesis and integration techniques and discusses their applicability for 2D transition metal dichalcogenides. Dielectric integration for various applications is reviewed in subsequent sections including nanoelectronics, optoelectronics, flexible electronics, valleytronics, biosensing, quantum information processing, and quantum sensing. For each application, we introduce basic device working principles, discuss the specific dielectric requirements, review current progress, present key challenges, and offer insights into future prospects and opportunities.

Hardware and Information Security Primitives Based on 2D Materials and Devices

Hardware security is a major concern for the entire semiconductor ecosystem that accounts for billions of dollars in annual losses. Similarly, information security is a critical need for the rapidly proliferating edge devices that continuously collect and communicate a massive volume of data. While silicon-based complementary metal-oxide-semiconductor technology offers security solutions, these are largely inadequate, inefficient, and often inconclusive, as well as resource intensive in time, energy, and cost, leading to tremendous room for innovation in this field. Furthermore, silicon-based security primitives have shown vulnerability to machine learning (ML) attacks. In recent years, 2D materials such as graphene and transition metal dichalcogenides have been intensely explored to mitigate these security challenges. In this review, 2D-materials-based hardware security solutions such as camouflaging, true random number generation, watermarking, anticounterfeiting, physically unclonable functions, and logic locking of integrated circuits (ICs) are summarized with accompanying discussion on their reliability and resilience to ML attacks. In addition, the role of native defects in 2D materials in developing high entropy hardware security primitives is also examined. Finally, the existing challenges for 2D materials, which must be overcome for large-scale deployment of 2D ICs to meet the security needs of the semiconductor industry, are discussed.

Ultrascaled Contacts to Monolayer MoS2 Field Effect Transistors

Two-dimensional (2D) semiconductors possess promise for the development of field-effect transistors (FETs) at the ultimate scaling limit due to their strong gate electrostatics. However, proper FET scaling requires reduction of both channel length (L_CH) and contact length (L_C), the latter of which has remained a challenge due to increased current crowding at the nanoscale. Here, we investigate Au contacts to monolayer MoS₂ FETs with L_CH down to 100 nm and L_C down to 20 nm to evaluate the impact of contact scaling on FET performance. Au contacts are found to display a ∼2.5× reduction in the ON-current, from 519 to 206 μA/μm, when L_C is scaled from 300 to 20 nm. It is our belief that this study is warranted to ensure an accurate representation of contact effects at and beyond the technology nodes currently occupied by silicon.

A two-dimensional mid-infrared optoelectronic retina enabling simultaneous perception and encoding

Infrared machine vision system for object perception and recognition is becoming increasingly important in the Internet of Things era. However, the current system suffers from bulkiness and inefficiency as compared to the human retina with the intelligent and compact neural architecture. Here, we present a retina-inspired mid-infrared (MIR) optoelectronic device based on a two-dimensional (2D) heterostructure for simultaneous data perception and encoding. A single device can perceive the illumination intensity of a MIR stimulus signal, while encoding the intensity into a spike train based on a rate encoding algorithm for subsequent neuromorphic computing with the assistance of an all-optical excitation mechanism, a stochastic near-infrared (NIR) sampling terminal. The device features wide dynamic working range, high encoding precision, and flexible adaption ability to the MIR intensity. Moreover, an inference accuracy more than 96% to MIR MNIST data set encoded by the device is achieved using a trained spiking neural network (SNN).

A Monolithic Stochastic Computing Architecture for Energy Efficient Arithmetic

As the energy and hardware investments necessary for conventional high-precision digital computing continue to explode in the era of artificial intelligence (AI), a change in paradigm that can trade precision for energy and resource efficiency is being sought for many computing applications. Stochastic computing (SC) is an attractive alternative since, unlike digital computers, which require many logic gates and a high transistor volume to perform basic arithmetic operations such as addition, subtraction, multiplication, sorting, etc., SC can implement the same using simple logic gates. While it is possible to accelerate SC using traditional silicon complementary metal-oxide-semiconductor (CMOS) technology, the need for extensive hardware investment to generate stochastic bits (s-bits), the fundamental computing primitive for SC, makes it less attractive. Memristor and spin-based devices offer natural randomness but depend on hybrid designs involving CMOS peripherals for accelerating SC, which increases area and energy burden. Here, the limitations of existing and emerging technologies are overcome, and a standalone SC architecture embedded in memory and based on 2D memtransistors is experimentally demonstrated. The monolithic and non-von-Neumann SC architecture occupies a small hardware footprint and consumes a miniscule amount of energy (<1 nJ) for both s-bit generation and arithmetic operations, highlighting the benefits of SC.

An All-in-One Bioinspired Neural Network

In spite of recent advancements in artificial neural networks (ANNs), the energy efficiency, multifunctionality, adaptability, and integrated nature of biological neural networks remain largely unimitated by hardware neuromorphic computing systems. Here, we exploit optoelectronic, computing, and programmable memory devices based on emerging two-dimensional (2D) layered materials such as MoS₂ to demonstrate a monolithically integrated, multipixel, and "all-in-one" bioinspired neural network (BNN) capable of sensing, encoding, learning, forgetting, and inferring at minuscule energy expenditure. We also demonstrate learning adaptability and simulate learning challenges under specific synaptic conditions to mimic biological learning. Our findings highlight the potential of in-memory computing and sensing based on emerging 2D materials, devices, and integrated circuits to not only overcome the bottleneck of von Neumann computing in conventional CMOS designs but also to aid in eliminating the peripheral components necessary for competing technologies such as memristors.

Active pixel sensor matrix based on monolayer MoS2 phototransistor array

In-sensor processing, which can reduce the energy and hardware burden for many machine vision applications, is currently lacking in state-of-the-art active pixel sensor (APS) technology. Photosensitive and semiconducting two-dimensional (2D) materials can bridge this technology gap by integrating image capture (sense) and image processing (compute) capabilities in a single device. Here, we introduce a 2D APS technology based on a monolayer MoS₂ phototransistor array, where each pixel uses a single programmable phototransistor, leading to a substantial reduction in footprint (900 pixels in ∼0.09 cm²) and energy consumption (100s of fJ per pixel). By exploiting gate-tunable persistent photoconductivity, we achieve a responsivity of ∼3.6 × 10⁷ A W^-1, specific detectivity of ∼5.6 × 10¹³ Jones, spectral uniformity, a high dynamic range of ∼80 dB and in-sensor de-noising capabilities. Further, we demonstrate near-ideal yield and uniformity in photoresponse across the 2D APS array.

Two-dimensional materials-based probabilistic synapses and reconfigurable neurons for measuring inference uncertainty using Bayesian neural networks

Artificial neural networks have demonstrated superiority over traditional computing architectures in tasks such as pattern classification and learning. However, they do not measure uncertainty in predictions, and hence they can make wrong predictions with high confidence, which can be detrimental for many mission-critical applications. In contrast, Bayesian neural networks (BNNs) naturally include such uncertainty in their model, as the weights are represented by probability distributions (e.g. Gaussian distribution). Here we introduce three-terminal memtransistors based on two-dimensional (2D) materials, which can emulate both probabilistic synapses as well as reconfigurable neurons. The cycle-to-cycle variation in the programming of the 2D memtransistor is exploited to achieve Gaussian random number generator-based synapses, whereas 2D memtransistor based integrated circuits are used to obtain neurons with hyperbolic tangent and sigmoid activation functions. Finally, memtransistor-based synapses and neurons are combined in a crossbar array architecture to realize a BNN accelerator for a data classification task.

Hardware implementation of Bayesian network based on two-dimensional memtransistors

Bayesian networks (BNs) find widespread application in many real-world probabilistic problems including diagnostics, forecasting, computer vision, etc. The basic computing primitive for BNs is a stochastic bit (s-bit) generator that can control the probability of obtaining ‘1’ in a binary bit-stream. While silicon-based complementary metal-oxide-semiconductor (CMOS) technology can be used for hardware implementation of BNs, the lack of inherent stochasticity makes it area and energy inefficient. On the other hand, memristors and spintronic devices offer inherent stochasticity but lack computing ability beyond simple vector matrix multiplication due to their two-terminal nature and rely on extensive CMOS peripherals for BN implementation, which limits area and energy efficiency. Here, we circumvent these challenges by introducing a hardware platform based on 2D memtransistors. First, we experimentally demonstrate a low-power and compact s-bit generator circuit that exploits cycle-to-cycle fluctuation in the post-programmed conductance state of 2D memtransistors. Next, the s-bit generators are monolithically integrated with 2D memtransistor-based logic gates to implement BNs. Our findings highlight the potential for 2D memtransistor-based integrated circuits for non-von Neumann computing applications.

Bayesian networks are applied to resolve several types of probabilistic problems. Here, Das et al. develop a stochastic computing hardware platform using two-dimensional memtransistors for the implementation of Bayesian network with high accuracy.

Heterogeneous Integration of Atomically Thin Semiconductors for Non-von Neumann CMOS

Atomically thin, 2D, and semiconducting transition metal dichalcogenides (TMDs) are seen as potential candidates for complementary metal oxide semiconductor (CMOS) technology in future nodes. While high-performance field effect transistors (FETs), logic gates, and integrated circuits (ICs) made from n-type TMDs such as MoS₂ and WS₂ grown at wafer scale have been demonstrated, realizing CMOS electronics necessitates integration of large area p-type semiconductors. Furthermore, the physical separation of memory and logic is a bottleneck of the existing CMOS technology and must be overcome to reduce the energy burden for computation. In this article, the existing limitations are overcome and for the first time, a heterogeneous integration of large area grown n-type MoS₂ and p-type vanadium doped WSe₂ FETs with non-volatile and analog memory storage capabilities to achieve a non-von Neumann 2D CMOS platform is introduced. This manufacturing process flow allows for precise positioning of n-type and p-type FETs, which is critical for any IC development. Inverters and a simplified 2-input-1-output multiplexers and neuromorphic computing primitives such as Gaussian, sigmoid, and tanh activation functions using this non-von Neumann 2D CMOS platform are also demonstrated. This demonstration shows the feasibility of heterogeneous integration of wafer scale 2D materials.