High order synaptic learning in neuro-mimicking resistive memories

TL;DR

This paper introduces an electroforming-free, CMOS-compatible memristor based on oxygen-deficient SrTiO$_{3-x}$ that mimics complex synaptic learning rules, advancing biomimetic neuromorphic architectures.

Contribution

It presents a novel oxygen-deficient SrTiO$_{3-x}$ memristor that emulates high order synaptic learning rules without electroforming, enabling scalable neuromorphic networks.

Findings

Successfully implemented high order time- and rate-dependent synaptic learning rules.

Demonstrated good agreement with biological synaptic measurements.

Achieved electroforming-free operation through oxygen vacancy engineering.

Abstract

Memristors have demonstrated immense potential as building blocks in future adaptive neuromorphic architectures. Recently, there has been focus on emulating specific synaptic functions of the mammalian nervous system by either tailoring the functional oxides or engineering the external programming hardware. However, high device-to-device variability in memristors induced by the electroforming process and complicated programming hardware are among the key challenges that hinder achieving biomimetic neuromorphic networks. Here, an electroforming-free and complementary metal oxide semiconductor (CMOS)-compatible memristor based on oxygen-deficient SrTiO$_{3-x}$ (STO$_x$) is reported to imitate synaptic learning rules. Through spectroscopic and cross-sectional transmission electron microscopic analyses, electroforming-free characteristics are attributed to the bandgap reduction of STO$_x$ by the formation of oxygen vacancies. The potential of such memristors to behave as artificial synapses is demonstrated by successfully implementing high order time- and rate-dependent synaptic learning rules. Also, a simple hybrid CMOS-memristor approach is presented to implement a variety of synaptic learning rules. Results are benchmarked against biological measurements form hippocampal and visual cortices with good agreement. This demonstration is a step towards the realization of large scale adaptive neuromorphic computation and networks.