Neuron-Inspired Computational E-Skin with Device Intrinsic In-Sensor Computing and Self-Powered Comprehensive Tactile Perception

This work reports a neuron-inspired, self-powered electronic skin (E-Skin) that combines device-intrinsic insensor computing with tactile perception for large-area, low-power robotic touch. A soft PVA/PAAm/LiCl ionic hydrogel array converts mechanical stimuli into voltage signals, implements geometry-programmed, pixel-wise ionic conductance weights, and physically compresses 40 sensing pixels into two readout channels with zero external power and no active ICs. Despite a 95 % reduction in channel count, the E-skin preserves salient per-pixel information: single-pixel fingertips distinguish hardness and roughness with 95.0 % and 92.8 % accuracy, resolve 20 visually similar materials with 98.83 % accuracy beyond human performance, and reconstruct contact locations from compressed outputs with 95.33 % accuracy. When deployed on a robotic hand, the two-channel, self-powered readout discriminates 29 everyday objects with 99.25 % accuracy and demonstrates low-latency tactile inference. This architecture shifts computation into the sensor body, alleviates compute bottlenecks that limit large-area Eskins, and offers a generalizable route to ultra-low-power tactile interfaces for future human-robot interaction.