Feasibility of giga-scale single-chip neuromorphic systems using memristors

Determine whether memristor-based nanomaterial devices can enable truly giga-scale compact neuromorphic chips that integrate billions of neuron-equivalent units on a single chip and support on-chip self-learning algorithms.

Background

This roadmap discusses neuromorphic computing systems and highlights the promise of emerging nano-scale devices, including memristors, for achieving event-driven, energy-efficient computation. While various CMOS and neuromorphic platforms have scaled significantly, whether novel nanomaterial devices can deliver single-chip platforms with billions of neurons and on-chip learning remains uncertain.

The authors emphasize that achieving such scale would require advances across device engineering, array integration, and learning methodologies that can operate directly on neuromorphic substrates.