Fast and Accurate Sparse Coding of Visual Stimuli With a Simple, Ultralow-Energy Spiking Architecture

Abstract

Memristive crossbars have become a popular means for realizing unsupervised and supervised learning techniques. In previous neuromorphic architectures with leaky integrate-andfire neurons, the crossbar itself has been separated from the neuron capacitors to preserve mathematical rigor. In this work, we sought to design a simplified sparse coding circuit without this restriction, resulting in a fast circuit that approximated a sparse coding operation at a minimal loss in accuracy. We showed that connecting the neurons directly to the crossbar resulted in a more energy-efficient sparse coding architecture, and alleviated the need to pre-normalize receptive fields. This work provides derivations for the design of such a network, named the Simple Spiking Locally Competitive Algorithm, or SSLCA, as well as CMOS designs and results on the CIFAR and MNIST datasets. Compared to a non-spiking, non-approximate model which scored 33 % on CIFAR-10 with a single-layer classifier, this hardware scored 32 % accuracy. When used with a state-ofthe-art deep learning classifier, the non-spiking model achieved 82 % and our simplified, spiking model achieved 80 %, while compressing the input data by 92 %. Compared to a previously proposed spiking model, our proposed hardware consumed 99 % less energy to do the same work at 21 × the throughput. Accuracy held out with online learning to a write variance of 3 %, suitable for the often-reported 4-bit resolution required for neuromorphic algorithms; with offline learning to a write variance of 27 %; and with read variance to 40 %. The proposed architecture's excellent accuracy, throughput, and significantly lower energy usage demonstrate the utility of our innovations.

Improving the Performance of Deep Learning Model‐Based Classification by the Analysis of Local Probability

“…We performed our framework and the existing methods on the samples of CIFAR-10 [15][16][17], CIFAR-100 [18][19][20], and Mini-ImageNet [21][22][23]. All of these evaluations proved the effectiveness of our framework.…”

Towards Efficient Processing and Learning with Spikes: New Approaches for Multi-Spike Learning

“…Spikes are believed to play an essential role in low-power consumption which would be of great importance to benefit devices such as mobiles and wearables where energy consumption is one of the major concerns [3], [11]. The efficiency is now one of the major bottlenecks of deep learning methods [6], and thus attracts more attention to neuromorphic computing [10], [14], [30], [56], [57]. In this work, we make our contributions towards this direction.…”

“…We provide systematic insight into various learning properties of our methods with synthesized experiments, while leaving possible extension to larger, more complex and practical problems unexplored in this study. This leaves a room for future developments where a proper encoding scheme is required to convert external stimuli into spikes [18], [26], [28], [35], [57]. Another limitation of our work is that only single-layer learning is examined.…”

Memristor-based signal processing for edge computing

“…The whole memristor-based system exhibited 16 improvement in power efficiency compared to the stateof-the-art digital computing system. In addition, many new memristor-based algorithms [54][55][56][57] and spintronic devices [56] have been attempted to implement the SC algorithms. Future exploration and optimization on devices, architectures, and algorithms are needed to realize a much more practical and widely used memristor-based signal encoding system.…”

“…N/A Signal transform DFT [40][41][42] E Time-frequency transformation [40] and speech recognition [42] N/A 10 in speed, 109.8 in power efficiency [40] DCT [5,45] E Image compression and processing [5] Energy efficiency: 119.7 TOPs 1 W 1 [5] N/A DWT [44] S Image compression [44] Energy: 6.4 nJ/image Time: 15 s/image [44] 11 in energy efficiency, 1.28 in speed [44] Signal encoding CS [46,48,49,51] E Image compression and reconstruction [51] Power dissipation: 16.2 mW/read [51] 50 in power consumption [51] SC [47,[54][55][56] E Sparse representation of natural images [47] Energy: 719 J/image Time: 0.036 s/image [47] 16 in energy consumption [47] Component analysis PCA [58][59][60] E Classification of breast cancer [60] Power dissipation: 0.27 W/feature [60] N/A ICA [62][63][64] E Blind image source separation [64] N/A N/A Classification and regression N/A SVM [66,67] S Wake-up system [66] Energy: 0.7 nJ for potentiation, 0.5 pJ for depression [66] N/A SLP [68,…”