Abstract
Neuromorphic devices inspired by the tripartite synapse system offer enhanced modulation of synaptic weight via a third terminal. However, using an electrically independent terminal for memorizing and processing optical information remains unexplored. Here, a ferroelectric-gated phototransistor (FGPT) incorporating ferroelectric polymers and organic photoactive channels is revisited for neuromorphic vision systems. It is demonstrated that partial polarization switching in the ferroelectric gate insulator enables linear control of the photoactive channel. Furthermore, the photogating effect induced by charge trapping at the ferroelectric insulator/photoactive channel interface further enhances the photonic non-volatile (PNV) characteristics of the FGPT. This allows memorized visual information, expressed as photoconductance, to be incrementally potentiated or depressed. The modulated photoconductance fully spans the current level within the dynamic range of the device (153 dB). Finally, the feasibility of the device for all-day face recognition is shown by in-sensor processing of visual information obtained from unstructured environments into the pre-trained range. This approach results in up to a ≈40% improvement in recognition accuracy.