A tactile sensor needs to perceive static pressures and dynamic forces in real-time with high accuracy for early diagnosis of diseases and development of intelligent medical prosthetics. However, biomechanical and external mechanical signals are always aliased (including variable physiological and pathological events and motion artifacts), bringing great challenges to precise identification of the signals of interest (SOI). Although the existing signal segmentation methods can extract SOI and remove artifacts by blind source separation and/or additional filters, they may restrict the recognizable patterns of the device, and even cause signal distortion. Herein, an in-memory tactile sensor (IMT) with a dynamically adjustable steep-slope region (SSR) and nanocavity-induced nonvolatility (retention time >1000 s) is proposed on the basis of a machano-gated transistor, which directly transduces the tactile stimuli to various dope states of the channel. The programmable SSR endows the sensor with a critical window of responsiveness, realizing the perception of signals on demand. Owing to the nonvolatility of the sensor, the mapping of mechanical cues with high spatiotemporal accuracy and associative learning between two physical inputs are realized, contributing to the accurate assessment of the tissue health status and ultralow-power (about 25.1 μW) identification of an occasionally occurring tremor.
Keywords: low-artifact perception; nonvolatile transistor; point-of-care diagnosis; steep-slope; tactile sensor.