A sub-1-volt analog metal oxide memristive-based synaptic device for energy-efficient spike-based computing systems

TL;DR

This paper presents a low-voltage, energy-efficient bilayer memristor capable of analog programming and spike-timing-dependent plasticity, advancing the development of brain-inspired, high-density neural networks.

Contribution

It introduces a forming-free, low-voltage memristor that supports analog states and STDP, enabling more scalable and energy-efficient neuromorphic computing.

Findings

Operates at ~0.8V with ~2pJ switching energy

Supports intermediate resistance states for analog programming

Demonstrates spike-timing-dependent plasticity (STDP) with >30x synaptic strength change

Abstract

Nanoscale metal oxide memristors have potential in the development of brain-inspired computing systems that are scalable and efficient1-3. In such systems, memristors represent the native electronic analogues of the biological synapses. However, the characteristics of the existing memristors do not fully support the key requirements of synaptic connections: high density, adjustable weight, and low energy operation. Here we show a bilayer memristor that is forming-free, low-voltage (~|0.8V|), energy-efficient (full On/Off switching at ~2pJ), and reliable. Furthermore, pulse measurements reveal the analog nature of the memristive device, that is it can be directly programmed to intermediate resistance states. Leveraging this finding, we demonstrate spike-timing-dependent plasticity (STDP), a spike-based Hebbian learning rule4. In those experiments, the memristor exhibits a marked change in the normalized synaptic strength (>30 times) when the pre- and post-synaptic neural spikes overlap. This demonstration is an important step towards the physical construction of high density and high connectivity neural networks.