ReckOn: A 28nm Sub-mm2 Task-Agnostic Spiking Recurrent Neural Network Processor Enabling On-Chip Learning over Second-Long Timescales

TL;DR

ReckOn introduces a compact, low-power spiking recurrent neural network processor capable of on-chip, task-agnostic learning over seconds, suitable for autonomous edge devices with limited memory.

Contribution

This work presents a novel 28nm spiking RNN processor that enables real-time, on-chip, task-agnostic learning over seconds, addressing memory and power constraints.

Findings

Achieved online learning for navigation, gesture recognition, and keyword spotting.

Maintained only 0.8% memory overhead during operation.

Operates with less than 150 μW training power.

Abstract

A robust real-world deployment of autonomous edge devices requires on-chip adaptation to user-, environment- and task-induced variability. Due to on-chip memory constraints, prior learning devices were limited to static stimuli with no temporal contents. We propose a 0.45-mm$^2$ spiking RNN processor enabling task-agnostic online learning over seconds, which we demonstrate for navigation, gesture recognition, and keyword spotting within a 0.8-% memory overhead and a <150-$\mu$W training power budget.