Artificial Intelligence Hardware Research Articles

Hardware-accelerated integrated optoelectronic platform towards real-time high-resolution hyperspectral video understanding

Recent advancements in artificial intelligence have significantly expanded capabilities in processing language and images. However, the challenge of comprehensively understanding video content still needs to be solved. The main problem is the requirement to process real-time multidimensional video information at data rates exceeding 1 Tb/s, a demand that current hardware technologies cannot meet. This work introduces a hardware-accelerated integrated optoelectronic platform specifically designed for the real-time analysis of multidimensional video. By leveraging optical information processing within artificial intelligence hardware and combining it with advanced machine vision networks, the platform achieves data processing speeds of 1.2 Tb/s. This capability supports the analysis of hundreds of frequency bands with megapixel spatial resolution at video frame rates, significantly outperforming existing technologies in speed by three to four orders of magnitude. The platform demonstrates effectiveness for AI-driven tasks, such as video semantic segmentation and object understanding, across indoor and aerial scenarios. By overcoming the current data processing speed limitations, the platform shows promise in real-time AI video understanding, with potential implications for enhancing human-machine interactions and advancing cognitive processing technologies.

Perspective: Entropy-stabilized oxide memristors

A memristor array has emerged as a potential computing hardware for artificial intelligence (AI). It has an inherent memory effect that allows information storage in the form of easily programmable electrical conductance, making it suitable for efficient data processing without shuttling of data between the processor and memory. To realize its full potential for AI applications, fine-tuning of internal device dynamics is required to implement a network system that employs dynamic functions. Here, we provide a perspective on multicationic entropy-stabilized oxides as a widely tunable materials system for memristor applications. We highlight the potential for efficient data processing in machine learning tasks enabled by the implementation of “task specific” neural networks that derive from this material tunability.

(Invited) Enabling Bio-Realistic Artificial Intelligence Hardware with Neuromorphic Nanoelectronics

The exponentially improving performance of digital computers has recently slowed due to the speed and power consumption issues resulting from the von Neumann bottleneck. In contrast, neuromorphic computing aims to circumvent these limitations by spatially co-locating logic and memory in a manner analogous to biological neuronal networks [1]. Beyond reducing power consumption, neuromorphic devices provide efficient architectures for image recognition, machine learning, and artificial intelligence [2]. This talk will explore how low-dimensional nanoelectronic materials enable gate-tunable neuromorphic devices [3]. For example, by utilizing self-aligned, atomically thin heterojunctions, dual-gated Gaussian transistors have been realized, which show tunable anti-ambipolarity for artificial neurons, competitive learning, spiking circuits, and mixed-kernel support vector machines [4,5]. In addition, field-driven defect motion in polycrystalline monolayer MoS2 enables gate-tunable memristive phenomena that serve as the basis of hybrid memristor/transistor devices (i.e., ‘memtransistors’) that concurrently provide logic and data storage functions [6]. The planar geometry of memtransistors further allows multiple contacts and dual gating that mimic the behavior of biological systems such as heterosynaptic responses [7]. Moreover, control over polycrystalline grain structure enhances the tunability of potentiation and depression, which enables unsupervised continuous learning in spiking neural networks [8]. Finally, the moiré potential in asymmetric twisted bilayer graphene/hexagonal boron nitride heterostructures gives rise to robust electronic ratchet states. The resulting hysteretic, non-volatile injection of charge carriers enables room-temperature operation of moiré synaptic transistors with diverse bio-realistic neuromorphic functionalities and efficient compute-in-memory designs for low-power artificial intelligence and machine learning hardware [9].[1] V. K. Sangwan, et al., Matter, 5, 4133 (2022).[2] V. K. Sangwan, et al., Nature Nanotechnology, 15, 517 (2020).[3] M. E. Beck, et al., ACS Nano, 14, 6498 (2020).[4] M. E. Beck, et al., Nature Communications, 11, 1565 (2020).[5] X. Yan, et al., Nature Electronics, DOI: 10.1038/s41928-023-01042-7 (2023).[6] X. Yan, et al., Advanced Materials, 34, 2108025 (2022).[7] H.-S. Lee, et al., Advanced Functional Materials, 30, 2003683 (2020).[8] J. Yuan, et al., Nano Letters, 21, 6432 (2021).[9] X. Yan, et al., Nature, DOI: 10.1038/s41586-023-06791-1 (2023).

Proton Conducting Neuromorphic Materials and Devices.

Neuromorphic computing and artificial intelligence hardware generally aims to emulate features found in biological neural circuit components and to enable the development of energy-efficient machines. In the biological brain, ionic currents and temporal concentration gradients control information flow and storage. It is therefore of interest to examine materials and devices for neuromorphic computing wherein ionic and electronic currents can propagate. Protons being mobile under an external electric field offers a compelling avenue for facilitating biological functionalities in artificial synapses and neurons. In this review, we first highlight the interesting biological analog of protons as neurotransmitters in various animals. We then discuss the experimental approaches and mechanisms of proton doping in various classes of inorganic and organic proton-conducting materials for the advancement of neuromorphic architectures. Since hydrogen is among the lightest of elements, characterization in a solid matrix requires advanced techniques. We review powerful synchrotron-based spectroscopic techniques for characterizing hydrogen doping in various materials as well as complementary scattering techniques to detect hydrogen. First-principles calculations are then discussed as they help provide an understanding of proton migration and electronic structure modification. Outstanding scientific challenges to further our understanding of proton doping and its use in emerging neuromorphic electronics are pointed out.

Integrating Artificial Intelligence Into the Visualization and Modeling of Three-Dimensional Anatomy in Pediatric Surgical Patients

BackgroundPediatric surgeons often treat patients with complex anatomical considerations due to congenital anomalies or distortion of normal structures by solid organ tumors. There are multiple applications for three-dimensional visualization of these structures based on cross-sectional imaging. Recently, advances in artificial intelligence (AI) applications and graphics hardware have made rapid 3D modelling of individual structures within the body accessible to surgeons without sophisticated and expensive hardware. In this report, we provide an overview of these applications and their uses in preoperative planning for pediatric surgeons. MethodsDeidentified DICOM files containing cross-sectional imaging of preoperative pediatric surgery patients were loaded from an institutional PACS database onto a secure PC with dedicated graphics and AI hardware (NVIDIA Geforce RTX 4070 laptop GPU). Visualization was obtained using an open-source imaging platform (3D Slicer). AI extensions to the platform were utilized to delineate the anatomy of interest. ResultsSegmentations of skeletal and visceral structures within a scan were obtained using the TotalSegmentator extension with an average processing time under 5 min. Additional AI modules were utilized for providing detailed mapping of the airways (AirwaySegmentation), lungs (Chest Imaging Platform), liver (SlicerLiver), or vasculature (SlicerVMTK). Other extensions were used for delineation of tumors within the hepatic parenchyma (MONAI Auto3DSeg) and hepatic vessels (RVesselX). ConclusionAI algorithms for image interpretation and processors dedicated to AI functions have significantly decreased the technical and financial requirements for obtaining detailed three-dimensional images of patient anatomy. Models obtained using AI algorithms have potential applications in preoperative planning, surgical simulation, patient education, and training. Level of EvidenceV, Case Series, Description of Technique.

Molybdenum Disulfide Memristors for Next Generation Memory and Neuromorphic Computing: Progress and Prospects

AbstractIn the last 15 years memristors have been investigated as devices for high‐density, low‐power, non‐volatile, resistive random access memory (ReRAM) beyond Moore's law. They also show potential in neuromorphic logic architectures to overcome the Von–Neumann bottleneck of classical circuitry facilitating better hardware for artificial intelligence (AI) and artificial neural network (ANN) systems. Molybdenum disulfide (MoS2) has emerged as a promising material for memristor devices of monolayer thickness due to its direct bandgap, high carrier mobility and environmental stability. In this review, recent progress in the development of MoS2 memristors the current understanding of the mechanisms behind their function are examined. The remaining obstacles to a commercially viable device principle and how these may be surmounted in light of the rapid progress that has already been made are also discussed.

Ag-doped non-imperfection-enabled uniform memristive neuromorphic device based on van der Waals indium phosphorus sulfide.

Memristors are considered promising energy-efficient artificial intelligence hardware, which can eliminate the von Neumann bottleneck by parallel in-memory computing. The common imperfection-enabled memristors are plagued with critical variability issues impeding their commercialization. Reported approaches to reduce the variability usually sacrifice other performances, e.g., small on/off ratios and high operation currents. Here, we demonstrate an unconventional Ag-doped nonimperfection diffusion channel-enabled memristor in van der Waals indium phosphorus sulfide, which can combine ultralow variabilities with desirable metrics. We achieve operation voltage, resistance, and on/off ratio variations down to 3.8, 2.3, and 6.9% at their extreme values of 0.2 V, 1011 ohms, and 108, respectively. Meanwhile, the operation current can be pushed from 1 nA to 1 pA at the scalability limit of 6 nm after Ag doping. Fourteen Boolean logic functions and convolutional image processing are successfully implemented by the memristors, manifesting the potential for logic-in-memory devices and efficient non-von Neumann accelerators.

AI Patent Approvals in Service Firms, Patent Radicalness, and Stock Market Reaction

Artificial intelligence (AI)-driven automation is of growing interest in the service sector. Using practice theory in service innovation and recombinant uncertainty frameworks, we ask whether AI patent approval for service firms is received positively by the stock market and whether patent radicalness strengthens or exacerbates the stock market reaction. We draw on 650 service industry firms from the years 1976 to 2019 with 133,813 non-AI patents and AI patents, including 7,543 (AI machine learning), 33,804 (AI hardware), and 53,419 (AI planning/control). The results show that the stock market reaction is positive for machine learning AI patents, and increasing radicalness strengthens the positive relationship; however, the reaction is negative to AI-related planning and control patents and increasing radicalness exacerbates the negative reaction. In addition, stock market reaction is insignificant to AI-related hardware patents and increasing radicalness does not influence this relationship. The findings are robust to excluding large firms representing a significant portion of the AI patents. With increasing radicalness, the stock market reaction to machine learning patents is more positive for low temporal depth and exacerbates with higher patent pedigree. The findings have implications for AI patenting among firms in the service sector.

Molecularly Reconfigurable Neuroplasticity of Layered Artificial Synapse Electronics

AbstractBrain‐inspired electronics with multimodal signal processing have been investigated as the next‐generation semiconductor platforms owing to the limitations of von Neumann architecture, which limits data processing and energy consumption efficiencies. This study demonstrates the molecular reconfiguration of plasticity of artificial synaptic devices with tunable electric conductance based on molecular dynamics at the channel surfaces for realizing chemical multimodality. Carrier transport dynamics are adjusted using the density of trapped carriers for the molecular adsorption of HS in the MoSe2 channel, and the results are consistent with the molecular simulations. In molecular dynamics‐controlled devices, enhanced hysteresis enables the engineering of artificial neuroplasticity, mimicking the neurotransmitter release of biological synapses. Owing to the molecular reconfigurability of MoSe2 devices, the synaptic weights of excitatory and inhibitory synapse modes are significantly enhanced. Thus, this study can potentially contribute to the creation of the next generation of multimodal interfaces and artificial intelligence hardware realization.

An LSTM-based Neural Network Wearable System for Blood Glucose Prediction in People with Diabetes.

This article proposes the first hardware implemen-tation of a low-power LSTM neural network targeting a wearable medical device designed to predict blood glucose at a 30-minute horizon. This work aims to reduce energy consumption by propos-ing new activation functions that target hardware implementation. On top of this proposal, we also prove there is room for improve-ment in energy consumption by applying neural network optimiza-tions at the algorithmic, such as quantization, and architecture level, LSTM hyperparameters, that consider the target hardware. To validate our proposal, we devise an optimized version of the neural network aimed to be wearable and, therefore, to reduce its energy consumption while preserving its accuracy as much as possible. The hardware is implemented on a Xilinx Virtex-7 FPGA VC707 Evaluation Kit. It is compared with (i) a faithful design of the original neural network implemented on the same evaluation kit, (ii) three state-of-the-art LSTM-based FPGA implementations, and (iii) software implementations running in cutting-edge smartphones:OnePlus NordTM and an Apple iPhone 13 ProTM with artificial in-telligence hardware accelerators. Our proposal consumes between ×1020 and ×7 less energy than the software implementations, being the most efficient system compared to the smartphones. On the other hand, its energy efficiency, measured in GFLOP/J, is between ×2.84 and ×7.82 greater than other state-of-the-art LSTM implementations, proving to be the most suitable implementation for a wearable system for blood glucose prediction.

Principles of analog neuromorphic computing: from components to systems and algorithms

This report presents the current state of affairs in the implementation of artificial intelligence hardware accelerators based on practically successful neural network algorithms of the first and second generations based on formal artificial neural networks (ANNs). The shortcomings of existing solutions are noted and ways to overcome them using analog neuromorphic architectures are outlined. The latter are created on the principles of the structuring and functioning of a living nervous system, using artificial neurons and models of synaptic contacts - the so–called memristors, electrically rewritable nanoscale elements of non-volatile memory [1-3]. With the use of these elements, it is possible to significantly increase the performance and energy efficiency of algorithm accelerators based on the ANNs [4-6], as well as the formation of promising computing systems based on bioplausible 3rd generation neural network algorithms - Spiking Neural Networks (SNNs) [7-9]. The original method of substantiating the optimal rules for local tuning SNNs with frequency encoding and the possibility of their implementation in the form of the Spike-Timing-Dependent Plasicity (STDP) are discussed [10]. The results of SNN learning stability to a variability of analog memristors, as well as the use of noise as a constructive factor in the fine-tuning and maintenance of SNN memristive weights are demonstrated [7, 11]. Also, approaches to the implementation of local plasticity rules with dopamine-like modulation as a type of SNN reinforcement learning are discussed. The latter is necessary for the formation of imitative "needs" of an agent in the process of its autonomous functioning [12, 13, 14]. The first results on the creation of a prototype of a memristive implantable neuroprosthesis of the motor activity are considered [15, 16]. Finally, possible hardware solutions for both neuronal elements and synaptic connections based on suitable memristive devices are demonstrated. The concept and first results on the creation of an analog neuromorphic computing system based on the above components are presented. Thus, an attempt is made to systematize the existing and original methods of implementing energy-efficient and compact analog neuromorphic computing systems for real-time and life-learning artificial intelligence.

The hardware is the software

Human brains and bodies are not hardware running software: the hardware is the software. We reason that because the physics of artificial intelligence hardware and of human biological "hardware" is distinct, neuromorphic engineers need to be selective in the inspiration we take from neuroscience.

Open-loop analog programmable electrochemical memory array

Emerging memories have been developed as new physical infrastructures for hosting neural networks owing to their low-power analog computing characteristics. However, accurately and efficiently programming devices in an analog-valued array is still largely limited by the intrinsic physical non-idealities of the devices, thus hampering their applications in in-situ training of neural networks. Here, we demonstrate a passive electrochemical memory (ECRAM) array with many important characteristics necessary for accurate analog programming. Different image patterns can be open-loop and serially programmed into our ECRAM array, achieving high programming accuracies without any feedback adjustments. The excellent open-loop analog programmability has led us to in-situ train a bilayer neural network and reached software-like classification accuracy of 99.4% to detect poisonous mushrooms. The training capability is further studied in simulation for large-scale neural networks such as VGG-8. Our results present a new solution for implementing learning functions in an artificial intelligence hardware using emerging memories.

FinFET 6T-SRAM All-Digital Compute-in-Memory for Artificial Intelligence Applications: An Overview and Analysis.

Artificial intelligence (AI) has revolutionized present-day life through automation and independent decision-making capabilities. For AI hardware implementations, the 6T-SRAM cell is a suitable candidate due to its performance edge over its counterparts. However, modern AI hardware such as neural networks (NNs) access off-chip data quite often, degrading the overall system performance. Compute-in-memory (CIM) reduces off-chip data access transactions. One CIM approach is based on the mixed-signal domain, but it suffers from limited bit precision and signal margin issues. An alternate emerging approach uses the all-digital signal domain that provides better signal margins and bit precision; however, it will be at the expense of hardware overhead. We have analyzed digital signal domain CIM silicon-verified 6T-SRAM CIM solutions, after classifying them as SRAM-based accelerators, i.e., near-memory computing (NMC), and custom SRAM-based CIM, i.e., in-memory-computing (IMC). We have focused on multiply and accumulate (MAC) as the most frequent operation in convolution neural networks (CNNs) and compared state-of-the-art implementations. Neural networks with low weight precision, i.e., <12b, show lower accuracy but higher power efficiency. An input precision of 8b achieves implementation requirements. The maximum performance reported is 7.49 TOPS at 330 MHz, while custom SRAM-based performance has shown a maximum of 5.6 GOPS at 100 MHz. The second part of this article analyzes the FinFET 6T-SRAM as one of the critical components in determining overall performance of an AI computing system. We have investigated the FinFET 6T-SRAM cell performance and limitations as dictated by the FinFET technology-specific parameters, such as sizing, threshold voltage (Vth), supply voltage (VDD), and process and environmental variations. The HD FinFET 6T-SRAM cell shows 32% lower read access time and 1.09 times better leakage power as compared with the HC cell configuration. The minimum achievable supply voltage is 600 mV without utilization of any read- or write-assist scheme for all cell configurations, while temperature variations show noise margin deviation of up to 22% of the nominal values.

Multilayer spintronic neural networks with radiofrequency connections.

Spintronic nano-synapses and nano-neurons perform neural network operations with high accuracy thanks to their rich, reproducible and controllable magnetization dynamics. These dynamical nanodevices could transform artificial intelligence hardware, provided they implement state-of-the-art deep neural networks. However, there is today no scalable way to connect them in multilayers. Here we show that the flagship nano-components of spintronics, magnetic tunnel junctions, can be connected into multilayer neural networks where they implement both synapses and neurons thanks to their magnetization dynamics, and communicate by processing, transmitting and receiving radiofrequency signals. We build a hardware spintronic neural network composed of nine magnetic tunnel junctions connected in two layers, and show that it natively classifies nonlinearly separable radiofrequency inputs with an accuracy of 97.7%. Using physical simulations, we demonstrate that a large network of nanoscale junctions can achieve state-of-the-art identification of drones from their radiofrequency transmissions, without digitization and consuming only a few milliwatts, which constitutes a gain of several orders of magnitude in power consumption compared to currently used techniques. This study lays the foundation for deep, dynamical, spintronic neural networks.

Reversing A Decades‐Long Scaling Law of Dielectric Breakdown using Hydrogen‐Plasma‐Treated HfO2 ReRAM Devices

AbstractDielectric breakdown (BD) is known to cause component failure in electronic devices and high‐voltage power lines over many decades. In recent years, this failure mechanism has been exploited to intentionally form nanoscale filaments in resistive random‐access‐memory (ReRAM) devices for artificial intelligence (AI). The statistical nature of this failure mechanism, known as the inverse size scaling law based on the weakest‐link theory, dictates the ever‐higher forming voltages for ReRAM devices, in turn, requiring additional peripheral transistors. This law also causes diminishing lifetimes of electronic components for future generations of very‐large‐integrated circuits (VLSI) technology. Currently, the semiconductor industry faces these scaling barriers limiting the miniaturization of both VLSI and AI hardware technology. Here, experimental evidence of a reverse area dependence is presented by introducing an innovative fabrication process with a hydrogen‐plasma‐treated layer in a bilayer HfO2 structure. Using joint order statistics rather than traditional extreme‐value statistics, a physics‐based statistical model is developed in agreement with the experimental data, thus demonstrating that there is no fundamental reason preventing this law from being reversed or altered. These findings will have a significant impact on both technology scaling and fabrication innovation for electronic and/or bioinspired nanomaterials; moreover, stimulate much research in physics, statistics, and reliability.

First-Principles Studies on Spin-Dependent Transport of Magnetic Junctions with Half-Metallic Heusler Compounds

Two types of magnetoresistance (MR), i.e., tunnel magnetoresistance (TMR) and current perpendicular to plane giant magnetoresistance (CPP-GMR), are theoretically discussed for systems with half-metallic (HFM) Co-based full Heusler compounds. In TMR, the non-collinear spin structure at the Co2MnSi(CMS)/MgO(001) interface that results from thermal spin fluctuations significantly reduced the TMR ratio at room temperature (RT). Enhancement of the exchange stiffness of CMS/MgO was essential to suppress the reduction in TMR at RT. Furthermore, it is proposed that inserting of a B2-CoFe layer between CMS and MgO is a promising way to enhance the interface exchange stiffness constant. In CPP-GMR, the interface spin asymmetry γ in the Valet-Fert model was essential to enhance GMR at RT, because the temperature dependence of γ in experiments was much larger than that of the bulk spin asymmetry β. I showed that the experimental CPP-GMR ratio increases with the decrease in the theoretical resistance-area product RPA, which is roughly consistent with the RP dependence of γ. This means that matching of the conductive channel between HMF and a non-magnetic metallic spacer is a key parameter to enhance γ. These findings will be very important for room temperature spintronics devices applied in future artificial intelligence hardware.

Research and Practice on the Engineering Skills Training Pattern in College-Enterprise Cooperative Studios

The development of vocational education is of great significance to the problem of difficult employment in enterprises. The cornerstone of sustainable development for skill studios are solving technical challenges and imparting technology. This will be beneficial for promoting and inheriting the IoT and artificial intelligence hardware technologies that enterprises need. The "Three-Mentor-Mechanism" proposed by the College-Enterprise Cooperative Studios help students personalized and rapid growth. General basic projects, professional skills projects, ability expansion projects, and on-the-job internship projects will be set up and run through the "four stages" of professional talent cultivation. College enterprise collaboration, college park co-education, project penetration finally construct a professional talent training model.

INTEGRATING AI HARDWARE IN ACADEMIC TEACHING: EXPERIENCES AND SCOPE FROM BRANDENBURG AND BAVARIA

Abstract. The field of artificial intelligence (AI) has gained increasing importance in recent years due to its potential to sustain growth and prosperity in a disruptive way. However, the role of special hardware for AI is still underdeveloped, and dedicated AI-capable hardware is crucial for effective and efficient processing. Moreover, hardware aspects are often neglected in university teaching, which emphasizes theoretical foundations and algorithmic implementations. As a result, there is a need for courses that focus on AI hardware development and its diverse applications. In response to this need, the BB-KI Chips consortium aims to develop a series of hardware-oriented courses with real-world AI applications. This consortium includes the Technical University of Munich (TUM) and the University of Potsdam (UP), which both offer a wide range of courses that focus on AI basics, AI algorithmic development, general computer architectures, chip design, and as well applications of AI. In the BB-KI-CHIPS project, these different capacities are planned to be tightly integrated into a unified curriculum covering knowledge from chip design over AI algorithms and techniques to applications.

Special Issue on Testability and Dependability of Artificial Intelligence Hardware

Artificial intelligence (AI) hardware, including AI accelerators and neuromorphic computing processors, emerges as one new frontier in the field of computing. There is an expedited paradigm shift in embracing bold and radical innovation of computer architectures, aiming at the continuation of computing performance improvement despite the slowed-down physical device scaling. Testability and dependability of AI hardware need to be addressed before mainstream adoption, especially in latency or throughput-critical, safety-critical, mission-critical, or remotely controlled applications (e.g., computer vision, autonomous driving, smart healthcare, IoTs, and robotics).