Temperature-Aware Monolithic 3D DNN Accelerators for Biomedical Applications

TL;DR

This paper presents a temperature-aware Mono3D DNN accelerator optimizer that enhances energy efficiency for biomedical applications by tuning design parameters under thermal and performance constraints.

Contribution

It introduces a novel optimizer for Mono3D accelerators that considers thermal constraints, achieving significant energy efficiency improvements.

Findings

Up to 61% energy efficiency improvement.

Effective tuning of aspect ratios and footprint.

Thermal-aware optimization enhances biomedical DNN inference.

Abstract

In this paper, we focus on temperature-aware Monolithic 3D (Mono3D) deep neural network (DNN) inference accelerators for biomedical applications. We develop an optimizer that tunes aspect ratios and footprint of the accelerator under user-defined performance and thermal constraints, and generates near-optimal configurations. Using the proposed Mono3D optimizer, we demonstrate up to 61% improvement in energy efficiency for biomedical applications over a performance-optimized accelerator.