Harnessing conversion bridge strategy by organic semiconductor in polymer matrix memristors for high-performance multi-modal neuromorphic signal processing

Abstract

Organic memristors, integrating chemically designed resistive switching and mechanical flexibility, present promising hardware opportunities for neuromorphic computing, particularly in the development of next-generation wearable artificial intelligence devices. However, challenges persist in achieving high yield, controllable switching, and multi-modal information processing. In this study, we introduce an efficient distribution of conversion bridges (EDCB) strategy by dispersing organic semiconductor (poly[2,5-bis(3-tetradecylthiophen-2-yl)thieno[3,2-b]thiophene], PBTTT) in elastomer (polystyrene-block-poly(ethylene-ran-butylene)-block-polystyrene, SEBS). This innovative approach results in memristors with exceptional yield, high stretchability, and reliable switching performance. By fine-tuning the semiconductor content, we shift the primary charge carriers from ions to electrons, realizing modulable non-volatile, and volatile duo-mode memristors. This advancement enables multi-modal signal processing at distinct operational mechanisms—non-volatile mode for image recognition in convolutional neural networks (CNNs) and volatile mode for dynamic classification and prediction in reservoir computing (RC). A fully analog RC hardware system is further demonstrated by integrating the distinct volatile and non-volatile modes of the EDCB-based memristor into the dynamic neuron network and the linear regression layer of the RC respectively, achieving high accuracy in online arrhythmia detection tasks. Our work paves the way for high-yield organic memristors with mechanical flexibility, advancing efficient multi-mode neuromorphic computing within a unified memristor system integrating volatile and non-volatile functionalities.