An Event-Driven E-Skin System with Dynamic Binary Scanning and real time SNN Classification

A research team has unveiled a breakthrough in tactile sensing with a novel event-driven electronic skin (e-skin) system. The hardware, built around a 16x16 piezoresistive tactile array, introduces a dynamic binary scanning strategy that only reads data when a touch event occurs. This approach slashes the data acquisition overhead, achieving a 12.8x reduction in scan counts, a 38.2x data compression rate, and an impressive 99% data sparsity. The system effectively creates a high-speed, sparse data stream optimized for neuromorphic processing.

This sparse stream is fed directly into a multi-layer convolutional Spiking Neural Network (Conv-SNN) implemented on an FPGA. The neuromorphic architecture is dramatically more efficient than traditional Convolutional Neural Networks (CNNs), requiring only 65% of the computation and 15.6% of the weight storage. Despite these massive efficiency gains, the system maintains a high classification accuracy of 92.11% for real-time handwritten digit recognition. The researchers also constructed a real neuromorphic tactile dataset using Address Event Representation (AER) to train and validate their model.

The work demonstrates a fully integrated, end-to-end pipeline from analog sensing to AI classification. By combining event-driven sensing with neuromorphic computing on-chip, the system offers a powerful and efficient solution for applications requiring real-time, low-power tactile perception. This has significant implications for advancing robotics, prosthetics, and next-generation human-computer interfaces where responsive, energy-efficient touch is critical.