Search

Neuromorphic processors have gained momentum recently due to their high energy efficiency in artificial intelligence applications compared to DNN accelerators. Most neuromorphic processors are executing SNNs (Spiking Neural Networks). Liquid State Machine (LSM), as the spiking version of reservoir computing, shows advantages and great potential in image classification, speech recognition, language translation, etc.. Comparing with other SNN models, LSM has the characteristics of easy to train and low resource utilization, which is suitable for low-power and resource-constrained edge computing scenarios. In this paper, we propose a novel design of a neuromorphic processor, LSMCore, aiming at LSM acceleration. LSMCore supports both training and inference of LSM. It consists of 256 input neurons, 1024 liquid neurons, and 1.31M synapses. Besides, multiple optimization techniques, including weight quantization for reducing storage, zero-skipping for decreasing dynamic sparsity, and mini-batch training are adopted in this processor. The experimental results show that the frequency of LSMCore achieves 400 MHz, the power is 4.9W and the area is 18.49 mm2 with a 40nm library. Comparing with the baseline, LSMCore achieves up to $80.7\times $ ($49.6\times $), $91.3\times $ ($56.3\times $), and $83.1\times $ ($56.8\times $) speedup on MNIST, N-MNIST, and Free Spoken Digital Dataset (FSDD) respectively for training (inference), while the accuracy of LSMCore on these three datasets are 96.8%, 97.6%, and 90% respectively. [ABSTRACT FROM AUTHOR]

When applied to artificial intelligence edge devices, the conventionally von Neumann computing architecture imposes numerous challenges (e.g., improving the energy efficiency), due to the memory-wall bottleneck involving the frequent movement of data between the memory and the processing elements (PE). Computing-in-memory (CIM) is a promising candidate approach to breaking through this so-called memory wall bottleneck. SRAM cells provide unlimited endurance and compatibility with state-of-the-art logic processes. This paper outlines the background, trends, and challenges involved in the further development of SRAM-CIM macros. This paper also reviews recent silicon-verified SRAM-CIM macros designed for logic and multiplication-accumulation (MAC) operations. [ABSTRACT FROM AUTHOR]

This paper provides a review of design approaches towards artificial intelligence (AI) System-on-Chip. AI algorithms have progressed over the past decades from perceptron-based neural network (NN) and neuro-fuzzy (NF) system to today’s deep neural network (DNN) and neuromorphic computing. Recent DNN hardware accelerators focus on energy-efficient integration of digital circuits to realize real-time DNN operation while neuromorphic processors deploy new memory technologies with analog computation for low power consumption. However, different design approaches can be applied to such processor implementation with their pros and cons. This paper reviews from the early processor designs for NN and NF in both mixed-mode and digital implementations to the recent DNN SoC designs that we have proposed for a decade. The former content deals with NN and NF processors used as a functional building block of a machine vision SoC, while the latter concentrates on integration of the whole DNN function. We also provide a discussion on the approaches, and provide perspective on future research directions. [ABSTRACT FROM AUTHOR]

Deep neural network (DNN) has been widely studied due to its high performance and usability for various applications such as image classification, detection, segmentation, translation, and action recognition. Thanks to the universal applications and high performance of DNN algorithm, DNN is adopted for various AI platforms, including edge/mobile devices as well as cloud servers. However, high-performance DNN requires a large amount of computation and memory access, making it challenging to implement DNN operation on edge/mobile. There have been several ways to solve these problems, including algorithms as well as hardware for DNN. Algorithms that help accelerate DNN in hardware enable much more efficient operation of high-performance AI. This article aims to provide an overview of the recent hardware and algorithm co-design schemes enabling efficient processing of DNNs. Specifically, it will provide algorithm optimization methods for DNN structure, neurons, synapses, and data types. This paper also introduces optimization methods for hardware architectures, PE array, data-path control, and microarchitecture of PE. And we will also show examples of DNN algorithm and hardware co-designed ASICs. [ABSTRACT FROM AUTHOR]

Computing-in-memory (CIM) is a promising approach to reduce the latency and improve the energy efficiency of deep neural network (DNN) artificial intelligence (AI) edge processors. However, SRAM-based CIM (SRAM-CIM) faces practical challenges in terms of area overhead, performance, energy efficiency, and yield against variations in data patterns and transistor performance. This paper employed a circuit-system co-design methodology to develop a SRAM-CIM unit-macro for a binary-based fully connected neural network (FCNN) layer of the DNN AI edge processors. The proposed SRAM-CIM unit-macro supports two binarized neural network models: an XNOR neural network (XNORNN) and a modified binary neural network (MBNN). To achieve compact area, fast access time, robust operations, and high energy-efficiency, our proposed SRAM-CIM uses a split-wordline compact-rule 6T SRAM and circuit techniques, including a dynamic input-aware reference generation (DIARG) scheme, an algorithm-dependent asymmetric control (ADAC) scheme, a write disturb-free (WDF) scheme, and a common-mode-insensitive small offset voltage-mode sensing amplifier (CMI-VSA). A fabricated 65-nm 4-Kb SRAM-CIM unit-macro achieved 2.4- and 2.3-ns product-sum access times for a FCNN layer using XNORNN and MBNN, respectively. The measured maximum energy efficiency reached 30.49 TOPS/W for XNORNN and 55.8 TOPS/W for the MBNN modes. [ABSTRACT FROM AUTHOR]

Silicon-based static random access memories (SRAM) and digital Boolean logic have been the workhorse of the state-of-the-art computing platforms. Despite tremendous strides in scaling the ubiquitous metal-oxide-semiconductor transistor, the underlying von-Neumann computing architecture has remained unchanged. The limited throughput and energy-efficiency of the state-of-the-art computing systems, to a large extent, result from the well-known von-Neumann bottleneck. The energy and throughput inefficiency of the von-Neumann machines have been accentuated in recent times due to the present emphasis on data-intensive applications such as artificial intelligence, machine learning, and cryptography. A possible approach towards mitigating the overhead associated with the von-Neumann bottleneck is to enable in-memory Boolean computations. In this paper, we present an augmented version of the conventional SRAM bit-cells, called the X-SRAM, with the ability to perform in-memory, vector Boolean computations, in addition to the usual memory storage operations. We propose at least six different schemes for enabling in-memory vector computations, including NAND, NOR, IMP (implication), XOR logic gates, with respect to different bit-cell topologies − the 8T cell and the 8+T Differential cell. In addition, we also present a novel ‘read-compute-store’ scheme, wherein the computed Boolean function can be directly stored in the memory without the need of latching the data and carrying out a subsequent write operation. The feasibility of the proposed schemes has been verified using the predictive transistor models and detailed Monte-Carlo variation analysis. As an illustration, we also present the efficacy of the proposed in-memory computations by implementing advanced encryption standard algorithm on a non-standard von-Neumann machine wherein the conventional SRAM is replaced by X-SRAM. Our simulations indicated that up to 75% of memory accesses can be saved using the proposed techniques. [ABSTRACT FROM AUTHOR]

Autonomous mobile robots are receiving a lot of attention for many applications, such as package delivery and smart surveillance, however, the battery capacity is limited to implement intelligent decision making in robots because of the heavy computational costs. In this paper, an ultra-low-power artificial intelligence processor (AIP) is proposed for real-time decision making of autonomous mobile robots. To achieve low power consumption while maintaining high performance, it adopts four key features: 1) an 8-thread tree search processor for real-time path planning; 2) a reinforcement learning accelerator for the avoidance of unexpected obstacles; 3) a 3-level transposition table cache for the reduction of duplicated computations; and 4) a PVT compensation circuit for the stable operation at near-threshold voltage. The proposed 16 mm $^{{{2}}}$ AIP is fabricated using 65-nm triple-well CMOS technology. It consumes only 1.1 mW at 0.55 V supply voltage and 7 MHz operating frequency, and 151 mW at 1.2 V supply voltage and 245 MHz operating frequency. The AIP achieves fast search speed (470 000 search/s) and low energy consumption (79 nJ/search), and it is successfully applied to a battery-powered robot system for autonomous navigation without any collision in dynamic indoor environments. [ABSTRACT FROM PUBLISHER]

Deep neural network (DNN) has recently attracted tremendous attention in various fields. But the computing operation requirement and the memory bottleneck limit the energy efficiency of hardware implementations. Binary quantization is proposed to relieve the pressure of hardware design. And the Computing-In-Memory (CIM) is regarded as a promising method to resolve the memory wall challenge. However, the binary computing paradigm is mismatched with the CIM scheme, which incurs complex circuits and peripheral to realize binary operation in previous works. To overcome previous issues, this work presents Binary Representation Computation-In-Memory (BR-CIM) with several key features. (1) A lightweight computation unit is realized within the 6T SRAM array to accelerate binary computing and enlarge signal margin; (2) The reconfigurable computing scheme and mapping method support extendable bit precision to satisfy the accuracy requirement of various applications; (3) Simultaneous computing and weight loading is supported by column circuitry, which shortens the data loading latency; Several experiments are conducted to estimate algorithm accuracy, the computing latency, and power consumption. The energy efficiency reaches up to 1280 TOPs/W for binary representation. And the algorithm accuracy achieves 97.82%/76.4% on MNIST/CIFAR-100 dataset. [ABSTRACT FROM AUTHOR]

This work presents a fully-digital 64 Kb non-volatile ReRAM based compute-in-memory (CIM) macro for the modern artificial intelligence (AI) edge devices, using 65 nm technology. This digital CIM architecture effectively removes the analog-design issues, related to process variations, noise susceptibility, and data-conversion overhead. Hence, it offers no accuracy loss and high energy-efficiency for the computation. To incorporate the digital computation, a novel NAND logic based 3.25T1R bitcell is proposed. The digital behaviour of this cell makes it superior to the conventional 1T1R based analog bitcell. Also, with the inherent non-volatility of ReRAM, the proposed cell can be a good substitute for SRAM-based CIM architectures with $4.62\times $ , $1.96\times $ , $3.96\times $ , and $5.12\times $ lower area than the XNOR-based 12T, Twin-8T, 8T, and 6T SRAM cell respectively. Moreover, the proposed CIM architecture allows full reconfigurabiliy from 1 to 16b precision for both input and weight. It also allows activating any number of parallel inputs, ranging from 1 to 128. According to simulation results, the proposed macro successfully operates up to 166.6 MHz for 1/8/15b input/weight/output precision and achieves 27.28 TOPS/W without any accuracy loss. Removing sense amplifiers for the ReRAM mode of the proposed work claims additional area and power savings. [ABSTRACT FROM AUTHOR]

Emerging advanced System-on-Chip (SoC) designs contain more and more complicated functions to be accelerated. This presents a challenge to conventional design approaches which use different hardware architectures or separate hardware accelerators to implement the various functions. To tackle this challenge, for the first time, we propose a generalized hardware accelerator called “Huicore” to speed up diverse functions on the same substrate. Through the analysis and transformation of mathematical characteristics, we reveal the commonality of many complicated functions using the CORDIC algorithm. Then we explore a reconfigurable architecture to implement them. The proposed reconfigurable accelerator can not only accelerate the implementation of many complicated functions, but also has small area, low power consumption and high precision. It is very suitable for integration in a SoC system to accelerate the implementation of various applications. [ABSTRACT FROM AUTHOR]

This paper proposes a novel algebraic integer (AI) based multi-encoding of Daubechies-4 and -6 2-D wavelet filters having error-free integer-based computation. Digital VLSI architectures employing parallel channels are proposed, physically realized and tested. The multi-encoded AI framework allows a multiplication-free and computationally accurate architecture. It also guarantees a noise-free computation throughput the multi-level multi-rate 2-D filtering operation. A single final reconstruction step (FRS) furnishes filtered and down-sampled image outputs in fixed-point, resulting in low levels of quantization noise. Comparisons are provided between Daubechies-4 and -6 designs in terms of SNR, PSNR, hardware structure, and power consumptions, for different word lengths. SNR and PSNR improvements of approximately 30% were observed in favour of AI-based systems, when compared to 8-bit fixed-point schemes (six fractional bits). Further, FRS designs based on canonical signed digit representation and on expansion factors are proposed. The Daubechies-4 and -6 4-level VLSI architectures are prototyped on a Xilinx Virtex-6 vcx240t-1ff1156 FPGA device at 282 MHz and 146 MHz, respectively, with dynamic power consumption of 164 mW and 339 mW, respectively, and verified on FPGA chip using an ML605 platform. [ABSTRACT FROM PUBLISHER]

This paper proposes a novel constructive training algorithm for cascade neural networks. By reformulating the cascade neural network as a linear-in-the-parameters model, we use the orthogonal least squares (OLS) method to derive a novel objective function for training new hidden units. With this objective function, the sum of squared errors (SSE) of the network can be maximally reduced after each new hidden unit is added, thus leading to a network with less hidden units and better generalization performance. Furthermore, the proposed algorithm considers both the input weights training and output weights training in an integrated framework, which greatly simplifies the training of output weights. The effectiveness of the proposed algorithm is demonstrated by simulation results. [ABSTRACT FROM PUBLISHER]

In this paper, an on-chip implementation of the active contour technique called pixel-level snakes is proposed. This is based on an optimized cellular neural network (CNN) algorithm with capabilities to support changes in the contour topology. The entire algorithm has been implemented on a 64 × 64 CNN universal machine chip-set architecture for which the results of the time performance measurements are given. To illustrate the validity and capabilities of the proposed implementation some on-chip experiments are also included. [ABSTRACT FROM AUTHOR]

Efficient deployment of deep neural networks (DNNs) emerges with the exploding demand for artificial intelligence on edge devices. Mixed-precision inference with both compressed model and reduced computation cost enlightens a way for accurate and efficient DNN deployments. Despite obtaining mixed-precision DNN models at the algorithmic level, there still lacks sufficient hardware support. In this work, we propose BitSystolic, a neural processing unit based on a systolic array structure. In BitSystolic, the numerical precision of both weights and activations can be configured in the range of 2 - 8 bits, fulfilling different requirements across mixed-precision models and tasks. Moreover, BitSystolic can support various data flows presented in different types of neural layers (e.g., convolution, fully-connected, and recurrent neural layers) and adaptive optimization of data reuse by switching between the matrix-matrix mode and vector-matrix mode. We designed and fabricated the proposed BitSystolic composed of a $16\times 16$ systolic array. Our measurement results show that BitSystolic features the unified power efficiency of up to 26.7 TOPS/W with 17.8 mW peak power consumption across various layer types. [ABSTRACT FROM AUTHOR]

Convolutional neural network (CNN) offers significant accuracy in image detection. To implement image detection using CNN in the Internet of Things (IoT) devices, a streaming hardware accelerator is proposed. The proposed accelerator optimizes the energy efficiency by avoiding unnecessary data movement. With unique filter decomposition technique, the accelerator can support arbitrary convolution window size. In addition, max-pooling function can be computed in parallel with convolution by using separate pooling unit, thus achieving throughput improvement. A prototype accelerator was implemented in TSMC 65-nm technology with a core size of 5 mm2. The accelerator can support major CNNs and achieve 152GOPS peak throughput and 434GOPS/W energy efficiency at 350 mW, making it a promising hardware accelerator for intelligent IoT devices. [ABSTRACT FROM PUBLISHER]

A multiplier-less architecture based on algebraic integer representation for computing the Daubechies 6-tap wavelet transform for 1-D/2-D signal processing is proposed. This architecture improves on previous designs in a sense that it minimizes the number of parallel 2-input adder circuits. The algorithm was achieved using brute-force numerical optimization of the algebraic integer representation. The proposed architecture furnishes exact computation up to the final reconstruction step, which is the operation that maps the exactly computed filtered results from algebraic integer representation to fixed-point. Compared to our recent work, this architecture shows a reduction of 27·n-16 adder circuits, where n is the number of wavelet decomposition levels. The design is physically implemented for a 4-level 1-D/2-D decomposition using a Xilinx Virtex-6 vcx240t-1ff1156 field programmable gate array (FPGA) device operating at up to a maximum clock frequency of 344/ 168 MHz. The FPGA implementation of 1-D/2-D are tested using hardware co-simulation using an ML605 board with clock of 100 MHz. A 45 nm CMOS synthesis of 2-D designs show improved clock frequency of better than 306 MHz for a supply voltage of 1.1 V. [ABSTRACT FROM AUTHOR]