Hydrogen bonding networks for tunable organic neuromorphic transistor arrays and in-sensor computing

Abstract

Inspired by neural architectures, synaptic transistors incorporating sensing, memory, and computing functionalities within one device have garnered widespread attention. However, achieving high carrier mobility and enduring synaptic plasticity remains challenging due to the unbalanced charge trapping effect at the dielectric/semiconductor interface. Here, introducing the hydrogen bonding networks, constructed via double dielectric materials, significantly improves the interface characteristics. This approach could modulate the carrier mobility of fabricated synaptic transistors from 0.49 to 22 cm² V⁻¹ s⁻¹ and increase synaptic plasticity from 67 to 10,000 s with an ultra-low energy consumption of 24 aJ per synaptic event. Moreover, we have devised an innovative linear self-attention-based spatial and channel joint attention (LSSCA) network architecture for spiking neural networks (SNNs) that exploits synaptic transistors to enhance image classification accuracy from 76% to 99%. This study provides a direct and effective strategy for engineering optically controlled synaptic transistors that demonstrate superior carrier mobility and prolonged plasticity, promising potential in low-energy, high-precision neuromorphic applications.