Physical Reservoir\nComputing Using van der Waals Ferroelectrics\nfor Acoustic Keyword Spotting

Acoustic\nkeyword spotting (KWS) plays a pivotal role\nin the voice-activated\nsystems of artificial intelligence (AI), allowing for hands-free interactions\nbetween humans and smart devices through information retrieval of\nthe voice commands. The cloud computing technology integrated with\nthe artificial neural networks has been employed to execute the KWS\ntasks, which however suffers from propagation delay and the risk of\nprivacy breach. Here, we report a single-node reservoir computing\n(RC) system based on the CuInP<sub>2</sub>S<sub>6</sub> (CIPS)/graphene\nheterostructure planar device for implementing the KWS task with low\ncomputation cost. Through deliberately tuning the Schottky barrier\nheight at the ferroelectric CIPS interfaces for the thermionic injection\nand transport of the electrons, the typical nonlinear current response\nand fading memory characteristics are achieved in the device. Additionally,\nthe device exhibits diverse synaptic plasticity with an excellent\nseparation capability of the temporal information. We construct a\nRC system through employing the ferroelectric device as the physical\nnode to spot the acoustic keywords, i.e., the natural numbers from\n1 to 9 based on simulation, in which the system demonstrates outstanding\nperformance with high accuracy rate (>94.6%) and recall rate (>92.0%).\nOur work promises physical RC in single-node configuration as a prospective\ncomputing platform to process the acoustic keywords, promoting its\napplications in the artificial auditory system at the edge.