Multimodal\n2D Ferroelectric Transistor with Integrated\nPerception-and-Computing-in-Memory Functions for Reservoir Computing

Emerging neuromorphic hardware promises energy-efficient\ncomputing\nby colocating multiple essential functions at the individual component\nlevel. The implementation is challenging due to mismatch between the\ncharacteristics of multifunctional devices and neural networks. Here,\nwe demonstrate an artificial synapse based on a 2D α-phase indium\nselenide that exhibits integrated perception-and-computing-in-memory\nfunctions in a single-transistor setup, serving as a basic building\nblock for reservoir computing. Extending to the array architecture\nenables concurrent image-sensing and memory. Further, we implement\nmultimode deep-reservoir computing with adjustable nonlinear transformation\nand multisensory fusion using this core device. In the lane-keeping-assistance\ntask for an unmanned vehicle, the system demonstrates ∼10<sup>4</sup> times lower energy consumption and significantly boosted\ndata throughput compared to the state-of-the-art graphics processors.\nThe demonstrated perception-and-computing-in-memory (PCIM) functions\nat a single-transistor level shows the feasibility of implementing\nultrascalable, resource-efficient hardware for brain-inspired computing.