Ultra-low-power-consuming liquid-water-based optoelectronic computing chip

作者：Minhui Yang, Kangchen Xiong, Xin Chen, Huikai Zhong, Shisheng Lin · 发表于：Device · 年份：2024 · DOI：10.1016/j.device.2024.100547 · 被引用次数：6 · 研究领域：Neural Networks and Reservoir Computing、Advanced Memory and Neural Computing、Photoreceptor and optogenetics research

Traditional solid-state neuromorphic chips use complementary metal-oxide-semiconductor (CMOS) circuits to mimic biological neurons, but their complexity limits efficiency. Here, we introduce a water computing chip with graphene/water/silicon photodetector arrays, leveraging the unique exponential decay of a water molecule's polarization transfer function (PTF) for signal transmission, transcending CMOS architecture. A designed Ising model describes water molecule polarization between graphene and silicon, showing a longitudinal decay PTF for low-energy pulse current output. The lateral decay PTF with a centimeter-scale diameter is validated with superposition output currents in a 3 × 3 water photodetector array. Molecular dynamics simulations indicate a 25 fs flip timescale for water molecules, achieving an ideal energy consumption of just 10 −18 J per operation. As a proof of concept, an 8 × 8 water neuromorphic chip identifies the ASCII code of "ZJU." This innovation significantly reduces energy consumption during inference, outperforming traditional CMOS-based neuromorphic chips and offering promising advancements in brain-inspired computing.