Monolithically Integrated VO$_2$ Mott Oscillators for Energy-Efficient Spiking Neurons

TL;DR

This paper presents a monolithically integrated VO₂-based spiking neuron hardware on CMOS platforms, achieving high-frequency oscillations with low energy consumption, advancing neuromorphic computing capabilities.

Contribution

It introduces a CMOS-compatible, monolithic integration of VO₂ memristor neurons with tunable oscillations and stochastic dynamics, enabling scalable neuromorphic hardware.

Findings

Oscillation frequencies from 40 to 410 kHz achieved.

Energy consumption as low as 18 pJ per spike.

Demonstrated voltage-controlled oscillator and tunable coupling.

Abstract

Brain-inspired non-Boolean computing offers intrinsic error tolerance and parallelism, but its practical deployment is limited by the lack of compact, energy-efficient spiking hardware compatible with large-scale integration. Mott phase-transition materials provide a promising route, as their abrupt insulator-to-metal transitions enable neuron-like thresholding and oscillatory dynamics in compact devices. Among these, vanadium dioxide (VO$_2$) stands out for its near-room-temperature transition, fast switching, and scalability. However, existing VO$_2$-based neuristors rely on discrete components, limiting integration density and system applicability. Here, we report monolithic back-end-of-the-line (BEOL) integration of one-transistor-one-VO2-memristor (1T-1MR) spiking neurons on CMOS-compatible platforms. VO$_2$ nanosheet devices are fabricated by pulsed-laser deposition below 430 {\deg}C on dielectrically isolated silicon-on-insulator (SOI) p-type junctionless field-effect transistors (JLFETs) in a compact 1T-1MR configuration. The architecture exhibits gate-tunable oscillations from 40 to 410 kHz in 60 nm-thick VO$_2$ devices with an active area of 6 $\mu$m$^2$, achieving energy consumption as low as 18 pJ per spike at room temperature, with memristor power dissipation of 8 $\mu$W and potential scaling toward sub-3 $\mu$W operation. We further uncover a non-monotonic dependence of oscillation frequency on current and temperature, along with bias-dependent stochastic firing dynamics, highlighting the rich behavior of integrated VO$_2$ memristor systems. Finally, we demonstrate voltage-controlled oscillator functionality and actively tunable resistive coupling of two nano-oscillators mediated by a JLFET. These results establish a pathway toward dense, energy-efficient, and monolithically integrated Mott-based neuromorphic hardware compatible with CMOS technology.