Simulators for Deep Neural Network Accelerator Design and Analysis: A Brief Review

Abstract

The rapid development of machine learning techniques, especially deep learning, has led to a drastic increase of research attention for domain-specific processors such as deep neural networks (DNN) accelerators. The surge in scale and complexity of DNN accelerators poses great design challenge. Simulators with high simulation speed and accurate performance evaluation capability are pivotal for DNN accelerator design. A number of simulators targeting DNN accelerators have emerged, yet they are not summarized and classified. This paper presents a systematic review of state-of-the-art DNN accelerator simulator from the perspective of simulator performance, target platform, evaluation indicators, input/output characteristics, and implementation details.