This paper proposes a high-performance and energy-efficient optical near-sensor accelerator for vision applications, called Lightator. Harnessing the promising efficiency offered by photonic devices, Lightator features innovative compressive acquisition of input frames and fine-grained convolution operations for low-power and versatile image processing at the edge for the first time. This will substantially diminish the energy consumption and latency of conversion, transmission, and processing within the established cloud-centric architecture as well as recently designed edge accelerators. Our device-to-architecture simulation results show that with favorable accuracy, Lightator achieves 84.4 Kilo FPS/W and reduces power consumption by a factor of ~24x and 73x on average compared with existing photonic accelerators and GPU baseline.
Works in quantum machine learning (QML) over the past few years indicate that QML algorithms can function just as well as their classical counterparts, and even outperform them in some cases. Among the corpus of recent work, many current QML models take advantage of variational quantum algorithm (VQA) circuits, given that their scale is typically small enough to be compatible with NISQ devices and the method of automatic differentiation for optimizing circuit parameters is familiar to machine learning (ML). While the results bear interesting promise for an era when quantum machines are more readily accessible, if one can achieve similar results through non-quantum methods then there may be a more near-term advantage available to practitioners. To this end, the nature of this work is to investigate the utilization of stochastic methods inspired by a variational quantum version of the long short-term memory (LSTM) model in an attempt to approach the reported successes in performance and rapid convergence. By analyzing the performance of classical, stochastic, and quantum methods, this work aims to elucidate if it is possible to achieve some of QML's major reported benefits on classical machines by incorporating aspects of its stochasticity.
Facial Expression Recognition (FER) plays an important role in human-computer interactions and is used in a wide range of applications. Convolutional Neural Networks (CNN) have shown promise in their ability to classify human facial expressions, however, large CNNs are not well-suited to be implemented on resource- and energy-constrained IoT devices. In this work, we present a hierarchical framework for developing and optimizing hardware-aware CNNs tuned for deployment at the edge. We perform a comprehensive analysis across various edge AI accelerators including NVIDIA Jetson Nano, Intel Neural Compute Stick, and Coral TPU. Using the proposed strategy, we achieved a peak accuracy of 99.49% when testing on the CK+ facial expression recognition dataset. Additionally, we achieved a minimum inference latency of 0.39 milliseconds and a minimum power consumption of 0.52 Watts.
Tensor processing units (TPUs), specialized hardware accelerators for machine learning tasks, have shown significant performance improvements when executing convolutional layers in convolutional neural networks (CNNs). However, they struggle to maintain the same efficiency in fully connected (FC) layers, leading to suboptimal hardware utilization. In-memory analog computing (IMAC) architectures, on the other hand, have demonstrated notable speedup in executing FC layers. This paper introduces a novel, heterogeneous, mixed-signal, and mixed-precision architecture that integrates an IMAC unit with an edge TPU to enhance mobile CNN performance. To leverage the strengths of TPUs for convolutional layers and IMAC circuits for dense layers, we propose a unified learning algorithm that incorporates mixed-precision training techniques to mitigate potential accuracy drops when deploying models on the TPU-IMAC architecture. The simulations demonstrate that the TPU-IMAC configuration achieves up to $2.59\times$ performance improvements, and $88\%$ memory reductions compared to conventional TPU architectures for various CNN models while maintaining comparable accuracy. The TPU-IMAC architecture shows potential for various applications where energy efficiency and high performance are essential, such as edge computing and real-time processing in mobile devices. The unified training algorithm and the integration of IMAC and TPU architectures contribute to the potential impact of this research on the broader machine learning landscape.
With the increased attention to memristive-based in-memory analog computing (IMAC) architectures as an alternative for energy-hungry computer systems for machine learning applications, a tool that enables exploring their device- and circuit-level design space can significantly boost the research and development in this area. Thus, in this paper, we develop IMAC-Sim, a circuit-level simulator for the design space exploration of IMAC architectures. IMAC-Sim is a Python-based simulation framework, which creates the SPICE netlist of the IMAC circuit based on various device- and circuit-level hyperparameters selected by the user, and automatically evaluates the accuracy, power consumption, and latency of the developed circuit using a user-specified dataset. Moreover, IMAC-Sim simulates the interconnect parasitic resistance and capacitance in the IMAC architectures and is also equipped with horizontal and vertical partitioning techniques to surmount these reliability challenges. IMAC-Sim is a flexible tool that supports a broad range of device- and circuit-level hyperparameters. In this paper, we perform controlled experiments to exhibit some of the important capabilities of the IMAC-Sim, while the entirety of its features is available for researchers via an open-source tool.
As the technology industry is moving towards implementing tasks such as natural language processing, path planning, image classification, and more on smaller edge computing devices, the demand for more efficient implementations of algorithms and hardware accelerators has become a significant area of research. In recent years, several edge deep learning hardware accelerators have been released that specifically focus on reducing the power and area consumed by deep neural networks (DNNs). On the other hand, spiking neural networks (SNNs) which operate on discrete time-series data, have been shown to achieve substantial power reductions over even the aforementioned edge DNN accelerators when deployed on specialized neuromorphic event-based/asynchronous hardware. While neuromorphic hardware has demonstrated great potential for accelerating deep learning tasks at the edge, the current space of algorithms and hardware is limited and still in rather early development. Thus, many hybrid approaches have been proposed which aim to convert pre-trained DNNs into SNNs. In this work, we provide a general guide to converting pre-trained DNNs into SNNs while also presenting techniques to improve the deployment of converted SNNs on neuromorphic hardware with respect to latency, power, and energy. Our experimental results show that when compared against the Intel Neural Compute Stick 2, Intel's neuromorphic processor, Loihi, consumes up to 27x less power and 5x less energy in the tested image classification tasks by using our SNN improvement techniques.
In this paper, we develop four spiking neural network (SNN) models for two static American Sign Language (ASL) hand gesture classification tasks, i.e., the ASL Alphabet and ASL Digits. The SNN models are deployed on Intel's neuromorphic platform, Loihi, and then compared against equivalent deep neural network (DNN) models deployed on an edge computing device, the Intel Neural Compute Stick 2 (NCS2). We perform a comprehensive comparison between the two systems in terms of accuracy, latency, power consumption, and energy. The best DNN model achieves an accuracy of 99.6% on the ASL Alphabet dataset, whereas the best performing SNN model has an accuracy of 99.44%. For the ASL-Digits dataset, the best SNN model outperforms all of its DNN counterparts with 99.52% accuracy. Moreover, our obtained experimental results show that the Loihi neuromorphic hardware implementations achieve up to 14.67x and 4.09x reduction in power consumption and energy, respectively, when compared to NCS2.
We propose an analog implementation of the transcendental activation function leveraging two spin-orbit torque magnetoresistive random-access memory (SOT-MRAM) devices and a CMOS inverter. The proposed analog neuron circuit consumes 1.8-27x less power, and occupies 2.5-4931x smaller area, compared to the state-of-the-art analog and digital implementations. Moreover, the developed neuron can be readily integrated with memristive crossbars without requiring any intermediate signal conversion units. The architecture-level analyses show that a fully-analog in-memory computing (IMC) circuit that use our SOT-MRAM neuron along with an SOT-MRAM based crossbar can achieve more than 1.1x, 12x, and 13.3x reduction in power, latency, and energy, respectively, compared to a mixed-signal implementation with analog memristive crossbars and digital neurons. Finally, through cross-layer analyses, we provide a guide on how varying the device-level parameters in our neuron can affect the accuracy of multilayer perceptron (MLP) for MNIST classification.
Fully-analog in-memory computing (IMC) architectures that implement both matrix-vector multiplication and non-linear vector operations within the same memory array have shown promising performance benefits over conventional IMC systems due to the removal of energy-hungry signal conversion units. However, maintaining the computation in the analog domain for the entire deep neural network (DNN) comes with potential sensitivity to interconnect parasitics. Thus, in this paper, we investigate the effect of wire parasitic resistance and capacitance on the accuracy of DNN models deployed on fully-analog IMC architectures. Moreover, we propose a partitioning mechanism to alleviate the impact of the parasitic while keeping the computation in the analog domain through dividing large arrays into multiple partitions. The SPICE circuit simulation results for a 400 X 120 X 84 X 10 DNN model deployed on a fully-analog IMC circuit show that a 94.84% accuracy could be achieved for MNIST classification application with 16, 8, and 8 horizontal partitions, as well as 8, 8, and 1 vertical partitions for first, second, and third layers of the DNN, respectively, which is comparable to the ~97% accuracy realized by digital implementation on CPU. It is shown that accuracy benefits are achieved at the cost of higher power consumption due to the extra circuitry required for handling partitioning.
Current state-of-the-art methods of image classification using convolutional neural networks are often constrained by both latency and power consumption. This places a limit on the devices, particularly low-power edge devices, that can employ these methods. Spiking neural networks (SNNs) are considered to be the third generation of artificial neural networks which aim to address these latency and power constraints by taking inspiration from biological neuronal communication processes. Before data such as images can be input into an SNN, however, they must be first encoded into spike trains. Herein, we propose a method for encoding static images into temporal spike trains using edge detection and an adaptive signal sampling method for use in SNNs. The edge detection process consists of first performing Canny edge detection on the 2D static images and then converting the edge detected images into two X and Y signals using an image-to-signal conversion method. The adaptive signaling approach consists of sampling the signals such that the signals maintain enough detail and are sensitive to abrupt changes in the signal. Temporal encoding mechanisms such as threshold-based representation (TBR) and step-forward (SF) are then able to be used to convert the sampled signals into spike trains. We use various error and indicator metrics to optimize and evaluate the efficiency and precision of the proposed image encoding approach. Comparison results between the original and reconstructed signals from spike trains generated using edge-detection and adaptive temporal encoding mechanism exhibit 18x and 7x reduction in average root mean square error (RMSE) compared to the conventional SF and TBR encoding, respectively, while used for encoding MNIST dataset.