A Flexible Precision Scaling Deep Neural Network Accelerator with Efficient Weight Combination

TL;DR

This paper presents a flexible deep neural network accelerator supporting mixed-precision inference from 2 to 8 bits, optimizing hardware utilization and energy efficiency for edge devices.

Contribution

It introduces a novel hardware architecture with weight decomposition, bit-serial MAC operations, and an efficient carry save adder tree for continuous precision scaling.

Findings

Achieves 4.09 TOPS peak throughput at 2-bit precision.

Attains 68.94 TOPS/W energy efficiency.

Supports flexible mixed-precision neural network inference.

Abstract

Deploying mixed-precision neural networks on edge devices is friendly to hardware resources and power consumption. To support fully mixed-precision neural network inference, it is necessary to design flexible hardware accelerators for continuous varying precision operations. However, the previous works have issues on hardware utilization and overhead of reconfigurable logic. In this paper, we propose an efficient accelerator for 2~8-bit precision scaling with serial activation input and parallel weight preloaded. First, we set two loading modes for the weight operands and decompose the weight into the corresponding bitwidths, which extends the weight precision support efficiently. Then, to improve hardware utilization of low-precision operations, we design the architecture that performs bit-serial MAC operation with systolic dataflow, and the partial sums are combined spatially. Furthermore, we designed an efficient carry save adder tree supporting both signed and unsigned number summation across rows. The experiment result shows that the proposed accelerator, synthesized with TSMC 28nm CMOS technology, achieves peak throughput of 4.09TOPS and peak energy efficiency of 68.94TOPS/W at 2/2-bit operations.