Deep-Learning-Assisted Self-Powered Microfluidic Bionic Electronic Tongues

Inspired by the natural mechanism of taste perception, artificial bionic electronic tongues have successfully enabled the detection and classification of various tastes. The liquid-solid contact electrification (LSCE) effect has emerged as a highly effective approach for developing self-powered electronic tongues. However, droplet-based sensing structures often face challenges related to internal and environmental interferences, compromising their stability and repeatability. In this work, we developed a monolithically integrated self-powered microfluidic bionic electronic tongue (SMET), combining the LSCE effect with deep learning algorithms to achieve highly reliable and intelligent sample identification and concentration detection. The incorporation of a multiplexed microchannel structure significantly reduced the required liquid sample volume while simultaneously increasing the electrical output amplitude (up to 10 V at multitone wave excitation), thereby enhancing sensitivity. Instead of micropumps, miniaturized exciters were employed as SMET drivers to generate multiple excitation waveforms, producing various signal types to improve specific algorithmic accuracy. The SMET achieved over 93% classification accuracy for five taste element samples (glacial acetic acid, anhydrous dextrose, quinine, edible chili essence, sodium chloride) and five concentrations of sodium chloride solutions using a single waveform signal, reaching 100% accuracy with the fusion of multiple waveform signals. Furthermore, the SMET was used to detect more than ten different taste samples, each exhibiting distinct signal variations. Thus, due to its ultrahigh sensitivity to the electrical properties of liquids, SMET enables accurate and rapid analysis of liquid samples with high reliability, positioning it as a promising tool in the field of rapid liquid detection.