Temperature-Aware Optimization of Monolithic 3D Deep Neural Network Accelerators

We propose a design automation methodology to help design of energy-efficient Mono3D DNN accelerators with safe on-chip temperature for mobile systems. We introduce an optimizer capable of investigating the impact of different aspect ratios of the chip and chip footprint specifications, and selecting energy-efficient accelerators under user-specified thermal and performance constraints. We also demonstrate that using our optimizer we can reduce energy consumption by 1.6× and area by 2× with a maximum of 9.5% increase in latency compared to a Mono3D DNN accelerator optimized only for performance.