In-sensor computing breakthrough enables efficient tactile information acquisition
The convergence of artificial intelligence, Internet of Things, and soft electronics has advanced tactile perception in flexible electronic skins, enabling applications in robotics, healthcare, and human-machine interfaces. However, conventional tactile sensing architectures separate sensing and processing, requiring analog-to-digital converters and data transfer to microcontrollers, which is energy-intensive and introduces latency. In-sensor computing integrates sensing and processing, reducing power consumption and enabling in-situ analog operations such as multiplication-accumulation (MAC) for artificial neural networks. Wang et al. developed a capacitive in-sensor tactile computing system combining a flexible pressure sensor array with electrical switching networks and a fixed capacitor to perform MAC operations in the charge domain. The sensor unit uses an ionic dielectric layer of PVA/H3PO4 prepared via sandpaper-templated molding, sandwiched between Au electrodes on waterborne polyurethane substrates, achieving high sensitivity of 0.36 nF/kPa and excellent stability. A 3×3 kernel of sensors with programmable bias voltages implements averaging and Laplacian filters for noise reduction and edge detection, validated experimentally. The system processes binary and analog tactile stimuli, with output voltage scaling proportionally with pressure. This in-sensor computing approach addresses energy and latency bottlenecks, offering a pathway for real-time, power-constrained e-skin applications in autonomous robotics, prosthetics, and intelligent interfaces.