Adaptive Leaky Integrate-and-Fire Neuron Platform Using VO2 Based Memristors

Details

Supervisors

Flandre, Denis

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil électromécanicien, à finalité spécialisée: mécatronique

Abstract

enNeuromorphic circuits aim to reproduce aspects of biological neural computation using electronic devices and circuits. In this context, volatile vanadium dioxide (VO2) memristors are promising building blocks for neuromorphic systems due to their threshold switching behavior and intrinsic sensitivity to temperature. The objective of this work is to model and experimentally validate an adaptive leaky integrate-and-fire (ALIF) neuron and the VO2 memristor it is based on. First, a compact behavioral model of the VO2 device is developed to reproduce its insulating and metallic resistance states, hysteretic switching behav- ior, and transition voltages. The model is designed for simplicity and VO2 behavior reproduction. The complete ALIF neuron is then modeled in LTSpice and Python, with the latter being used to investi- gate the influence of circuit parameters on spike generation, inter-spike intervals, adaptation and power consumption. A dedicated PCB is designed and used to experimentally characterize VO2 devices and validate the proposed neuron. The measurements demonstrate spike generation and adaptive feedback behavior, while experimental parameter sweeps confirm that the temporal response can be tuned through the circuit components, confirming the memristor and ALIF model. Finally, the intrinsic temperature dependence of the VO2 device is exploited to encode temperature information into the output spike train. Distinct changes in spike count are observed at several temperature thresholds between approximately 30 and 45°C, creating quantization levels for the sensor. These results demonstrate the feasibility of combining temperature sensing, event-based encoding, and neural processing within a single VO2-based ALIF circuit.