CoDR: Computation and Data Reuse Aware CNN Accelerator

TL;DR

CoDR is a CNN accelerator that enhances computation and data reuse by exploiting sparsity, similarity, and repetition, significantly reducing memory access and energy consumption.

Contribution

It introduces Universal Computation Reuse and a customized encoding scheme, optimizing memory access and energy efficiency in CNN accelerators.

Findings

Reduces SRAM access by up to 8x

Consumes significantly less energy than recent accelerators

Exploits weight sparsity, repetition, and similarity simultaneously

Abstract

Computation and Data Reuse is critical for the resource-limited Convolutional Neural Network (CNN) accelerators. This paper presents Universal Computation Reuse to exploit weight sparsity, repetition, and similarity simultaneously in a convolutional layer. Moreover, CoDR decreases the cost of weight memory access by proposing a customized Run-Length Encoding scheme and the number of memory accesses to the intermediate results by introducing an input and output stationary dataflow. Compared to two recent compressed CNN accelerators with the same area of 2.85 mm^2, CoDR decreases SRAM access by 5.08x and 7.99x, and consumes 3.76x and 6.84x less energy.