CENNA: Cost-Effective Neural Network Accelerator

Abstract

Convolutional neural networks (CNNs) are widely adopted in various applications. State-of-the-art CNN models deliver excellent classification performance, but they require a large amount of computation and data exchange because they typically employ many processing layers. Among these processing layers, convolution layers, which carry out many multiplications and additions, account for a major portion of computation and memory access. Therefore, reducing the amount of computation and memory access is the key for high-performance CNNs. In this study, we propose a cost-effective neural network accelerator, named CENNA, whose hardware cost is reduced by employing a cost-centric matrix multiplication that employs both Strassen’s multiplication and a naïve multiplication. Furthermore, the convolution method using the proposed matrix multiplication can minimize data movement by reusing both the feature map and the convolution kernel without any additional control logic. In terms of throughput, power consumption, and silicon area, the efficiency of CENNA is up to 88 times higher than that of conventional designs for the CNN inference.

Dynamic electric-field-induced magnetic effects in cobalt oxide thin films: towards magneto-ionic synapses

“…Neuromorphic computing emulates the activity of biological synapses by utilizing artificial neural networks in which synapses are massively interconnected in a dynamic and reconfigurable way to process information in an energy-efficient manner. [6][7][8][9] Although the bases of this computation paradigm were formulated in the 1990s, 8,10 a renewed interest has emerged in recent years, jump-started by the discovery and development of advanced materials that might address some of the essential requirements of brain-inspired computing. So far, emulation of artificial synapses has been achieved, to some extent, using phase change materials, 11 superconductors, 12 transistors, 13 spintronic devices 14,15 andmemristors.…”

Verifiable and Energy Efficient Medical Image Analysis with Quantised Self-attentive Deep Neural Networks

“…Therefore, it is key to have an energy-efficient model without degradation in performance. A rough estimate of energy cost per operation in 45nm 0.9V IC design can be calculated using Table 3 presented in [7,23,14]. The number of multiplication and addition operations in a standalone selfattention layer [20] can be calculated as…”

An Approximate GEMM Unit for Energy-Efficient Object Detection

“…Accelerating deep neural network processing in edge computing using energy-efficient platforms is an important goal [ 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 ]. Currently, most object detection and classification models are carried out in graphics processing units.…”

“…Therefore, many lightweight approaches with low-power consumption and low-computational performance have emerged recently. A few dedicated neural network accelerators have been implemented on FPGA hardware platforms [ 12 , 14 , 17 , 21 , 23 ], while several authors proposed ASIC-based neural network accelerators [ 13 , 15 , 16 , 18 , 19 , 22 ]. Samimi et al [ 20 ] proposed a technique based on the residue number system to improve the energy efficiency of deep neural network processing.…”