Twisting van der Waals Heterostructures Enables Thermographic In-Sensor Computing and Logics.

Abstract

In-sensor computing and logics are the core for advancing machine vision technology, but face challenges in extending such capability into thermal form. Here, a bioinspired thermoreceptor array based on twisted graphene and titanium diselenide (TiSe₂) heterostructures is presented. The electron-phonon interaction between graphene and few-layer TiSe₂ exhibits strong twist-angle (θ) dependence, as evidenced by varied phonon-induced gap of graphene (114 ± 13 meV for θ = 1° and 67 ± 6 meV for θ = 7°), enabling sensitive thermal reception at temperature range from 20 to 350 K. Integrating such heterostructure into a (64 × 64) thermoperception array with a convolutional neural network (CNN) framework promotes significant advancements in microscopic thermal imaging, enhancing thermal image detection accuracy by 46% and achieving 99% classification accuracy for pathological cell identification. In addition, the "AND" "OR", and,"AND/OR" logic operations are demonstrated using these thermoreceptor arrays. This proof-of-concept thermoperception arrays also simplify the monolithically integrable architecture of machine vision and offer a potential solution for efficient in-sensor computing and logics for next-generation intelligent systems.