Multipurpose Deep-Learning Accelerator for Arbitrary Quantization With Reduction

Abstract

Various pruning and quantization heuristics have been proposed to compress recent deep-learning models. How- ever, the rapid development of new optimization techniques makes it difficult for domain-specific accelerators to efficiently process various models showing irregularly stored parameters or nonlinear quantization. This article presents a scalable- precision deep-learning accelerator that supports multiply-and- accumulate operations (MACs) with two arbitrarily quantized data sequences. The