Image sensors capable of in-sensor computing can substantially reduce computational burdens in machine vision systems that require extensive data post-processing. Reconfigurable photodetectors, which serve as the fundamental building blocks for such sensors by enabling pixel-level signal modulation, are therefore central to this development. They have been explored using emerging materials or modified field-effect transistor architectures, but their complex structures and processing limit scalability and resolution. Here we report a reconfigurable photodetector with a simple, vertically stacked two-terminal structure based on an organic donor/acceptor/donor tri-layer active, which functions equivalently as dual photodiodes connected with opposite polarity with a resistor inserted between them. The device exhibits 80 dB linear dynamic range and, furthermore, features responsivity that can be tuned in a quasi-linear fashion from -26 mA/W to 77 mA/W under operation bias from -1 V to 1 V. This continuously variable responsivity can act as weight values in a neural network, as demonstrated with a 4 × 4 photodetector array that performed reconfigured photocurrent map generations and kernel operations, emulating simple image preprocessing filters. Its voltage-controlled tunability and straightforward fabrication process underscore the potential for monolithic integration with various backplane technologies, paving the way toward compact and efficient intelligent image sensors.
Hahn et al. (Mon,) studied this question.