Light-Induced Neuron-Like Bursts

Abstract

Aside from recent advances in artificial intelligence (AI) models, specialized AI hardware has become increasingly important for processing large volumes of unstructured and dynamically evolving data at the edge. The growing demand for on-device learning calls for dynamically reconfigurable systems that are capable of responding to continuously changing environments. Here, we demonstrate a dynamic approach to information processing using devices exhibiting negative differential resistance operated near their folding point, where oscillatory input signals induce controllable bursting dynamics. In this regime, time-varying input signals generate tunable spike responses governed by the waveform characteristics. We systematically investigate the influence of input amplitude, frequency, and DC offset on the resulting device dynamics. The observed sensitivity enables compact encoding and discrimination of temporal signal features without external clocking. These results establish negative differential resistance devices as promising platforms for asynchronous signal classification based on intrinsic nonlinear dynamics.