SCALE-Sim: Systolic CNN Accelerator Simulator

TL;DR

SCALE-Sim is a configurable, cycle-accurate simulator for systolic-array based DNN accelerators, enabling detailed exploration of design trade-offs and mapping strategies to optimize deep learning performance.

Contribution

It introduces the first systolic-array simulator tailored for DNNs, providing detailed configurability for design space exploration and performance analysis.

Findings

Bandwidth, data flow, and aspect ratio significantly impact runtime and energy.

SCALE-Sim enables comprehensive analysis of systolic-array DNN accelerators.

Insights benefit both hardware architects and machine learning practitioners.

Abstract

Systolic Arrays are one of the most popular compute substrates within Deep Learning accelerators today, as they provide extremely high efficiency for running dense matrix multiplications. However, the research community lacks tools to insights on both the design trade-offs and efficient mapping strategies for systolic-array based accelerators. We introduce Systolic CNN Accelerator Simulator (SCALE-Sim), which is a configurable systolic array based cycle accurate DNN accelerator simulator. SCALE-Sim exposes various micro-architectural features as well as system integration parameters to the designer to enable comprehensive design space exploration. This is the first systolic-array simulator tuned for running DNNs to the best of our knowledge. Using SCALE-Sim, we conduct a suite of case studies and demonstrate the effect of bandwidth, data flow and aspect ratio on the overall runtime and energy of Deep Learning kernels across vision, speech, text, and games. We believe that these insights will be highly beneficial to architects and ML practitioners.