CoNNA – Compressed CNN Hardware Accelerator

Abstract

In this paper we propose a novel Convolutional Neural Network hardware accelerator, called CoNNA, capable of accelerating pruned, quantized, CNNs. In contrast to most existing solutions, CoNNA offers a complete solution to the full, compressed CNN acceleration, being able to accelerate all layer types commonly found in contemporary CNNs. CoNNA is designed as a coarse-grained reconfigurable architecture, which uses rapid, dynamic reconfiguration during CNN layer processing. Furthermore, by being able to directly process compressed feature and kernel maps, CoNNA is able to achieve higher CNN processing efficiency than some of the previously proposed solutions. Results of the experiments indicate that CoNNA architecture is up to 14.10 times faster than previously proposed MIT's Eyeriss CNN accelerator, up to 6.05 times faster than NullHop CNN accelerator, and up to 4.91 times faster than NVIDIA's Deep Learning Accelerator (NVDLA), while using identical number of computing units and operating at the same clock frequency.