13 January 2026 in International Journal of Extreme Manufacturing
Spiking neural networks (SNNs) rely on precise spike timing for computation, however, their performance is often suffers jitter-induced errors and constraints on synaptic weight updates. In this study, we address this challenge by expanding the neuronal integration window, enhancing temporal robustness while maintaining efficient learning dynamics. We introduce a material-driven approach to expand the operational window of artificial neuron device, which is defined as the difference between the threshold voltage (Vth) and holding voltage (Vhold), in ovonic threshold switch (OTS)-based neurons, demonstrating its direct impact on synaptic weight updates and error mitigation. Carbon doping in GeTe-based OTS devices is employed to systematically modulate trap depth variations under an electric field, achieving precise control over Vth and Vhold. Electrical measurements confirm that an optimal 3.4% carbon concentration maximizes the operational window, stabilizing threshold switching and ensuring reliable neuronal operation. To reveal the atomic-scale mechanisms behind this behavior, we perform density functional theory (DFT) simulations, analyzing coordination number, bond angles to elucidate how carbon incorporation modifies trap distributions and influences device characteristics. Finally, we assess the practical impact of operational window expansion by implementing the optimized OTS neurons in a learning framework based on the tempotron learning rule, revealing enhanced spike timing robustness and reduced synaptic weight update constraints. This study provides a scalable pathway toward more reliable spike-based neuromorphic computing to advance the next generation of artificial intelligence hardware.
12 November 2026 in Nature Communications
Establishing autonomous and energy-efficient feedback control at the hardware level remains challenging because conventional systems rely on centralized processing and continuous external computation, increasing energy consumption and system complexity. Here we show a neuromorphic autapse-inspired system that enables localized and bidirectional feedback control, using thermoregulation as a representative application. An autapse is a self-synaptic connection in which a neuron feeds its output back to its own synaptic input, enabling intrinsic feedback modulation. The system integrates a solid electrolyte-based synaptic transistor capable of multiplexing multiple inputs and a bidirectional leaky integrate-and-fire artificial neuron. By feeding the neuronal output back to an additional synaptic gate, excitatory and inhibitory feedback loops are realized within a compact architecture. Using an artificial thermoregulation platform, temperature-dependent inputs bidirectionally modulate neuronal firing frequency, demonstrating biologically inspired feedback control with a compact set of neuromorphic devices for autonomous and energy-efficient systems.