Low-voltage, long-term stable organic heterojunction transistors with a broadband bidirectional photoresponse for in-sensor vision computing

Abstract

In-sensor vision computing enables visual sensors to perform computations during image acquisition, substantially reducing energy consumption and latency. Conventional phototransistors, however, typically exhibit a unipolar photoresponse and limited logic capabilities, while silicon-based sensors are confined to the visible spectrum, leading to information loss and reduced computational accuracy. Here, we introduce an organic heterojunction transistor incorporating a HfO₂/SiO₂ stacked dielectric. This architecture simultaneously enables low-voltage operation (<5 V) and long-term stability while maintaining well-defined, gate-tunable positive and negative photoresponses. The device exhibits broadband sensitivity spanning 530–1050 nm, allowing effective multi-wavelength detection and processing. Notably, it retains over 80% of its initial responsivity after 70 days of storage, demonstrating exceptional durability. Gate-voltage modulation establishes a linear correlation between responsivity and voltage, enabling real-time in-sensor computing. Using this device, multi-wavelength image processing and classification can be performed efficiently, achieving 99.27% recognition accuracy. This work establishes a compact, low-power, and robust platform for advanced in-sensor vision computing, paving the way for miniaturized, energy-efficient, and multi-spectral imaging systems.