KeyRAM: A 0.34 uJ/decision 18 k decisions/s Recurrent Attention In-memory Processor for Keyword Spotting

Abstract

This paper presents a 0.34 uJ/decision deep learning-based classifier for keyword spotting (KWS) in 65 nm CMOS with all weights stored on-chip. This work adapts a Recurrent Attention Model (RAM) algorithm for the KWS task, and employs an in-memory computing (IMC) architecture to achieve up to 9× savings in energy/decision and more than 23× savings in EDP of decisions over a state-of-the art IMC IC for KWS using the Google Speech dataset while achieving the highest reported decision throughput of 18.32 k decisions/s.