Highly parallel and energy-efficient exhaustive minimum distance search engine using hybrid digital/analog circuit techniques

Abstract

A minimum distance search engine (MDSE) is presented as a hardware. accelerator for various exhaustive pattern-matching systems. This chip executes highly parallel computations of L-1-norms between an input query and stored multiple reference records, and searches for the minimum distance among them in a highly parallel fashion. Our architectural-level estimation shows that this MDSE can reduce energy dissipation by orders of magnitude as the number of records increases, compared with the conventional systems. We have designed a prototype 4-bit 8-word MDSE composed of merged memory logic (MML) and digital/analog-mixed winner-take-all circuit (DAM-WTAC) by using hybrid digital/analog circuit techniques. It was fabricated with a 0.6-mum single-poly triple-metal CMOS technology. Experimental results show that our chip works properly at 3 V/10 MM and has approximately four times larger throughput as well as four times higher energy efficiency, compared with the existing 8-bit microcontrollers.