@article{Yeh2026, 
author = {Chiunghan Yeh and Shaotian Zhang and Chen Lu and Pei Liu and Kangli Xu and Jieru Song and Jiaming Liu and Jialin Meng and Tianyu Wang and Jiajie Yu and Hao Zhu and Qingqing Sun and David Wei Zhang and Lin Chen},
title = {Enhancing learning of spiking neural networks with lifetime-tunable optical neuron devices},
year = {2026},
journal = {Nano Research},
url = {https://www.sciopen.com/article/10.26599/NR.2026.94909146},
doi = {10.26599/NR.2026.94909146},
abstract = {Temporal heterogeneity is a defining feature of biological neural systems, where neurons with distinct membrane time constants support complementary information-processing functions. However, most hardware spiking neural networks (SNNs) still rely on artificial neurons with fixed membrane dynamics, limiting the ability to reproduce the adaptive temporal processing of the brain. Here, we demonstrate a CMOS-compatible lifetime-tunable optical neuron device based on a TiN/Hf0.5Zr0.5O2/TiO2/ITO heterostructure. Under optical excitation, read-voltage-regulated photocarrier dynamics enable physical modulation of the photocurrent decay lifetime, providing a direct device-level analogue of a tunable neuronal membrane time constant. Incorporating the lifetime tunability into SNNs enables adaptive temporal heterogeneity with coexisting fast and slow membrane-potential dynamics. Simulation results on CIFAR-10, CIFAR-100, Tiny-ImageNet, Caltech101, and CIFAR10-DVS show that this architecture achieves faster convergence and higher classification accuracy than fixed-lifetime neuronal counterparts. This work opens a route toward adaptive SNNs with tunable membrane time constants, providing a promising strategy for biologically inspired, temporally adaptive, and hardware-oriented neuromorphic computing systems.}
}