Polarization-tunable vdW ferroelectric heterojunction for high-gain photodetection and intelligent visual recognition

Ferroelectric phototransistors have emerged as promising platforms for integrating sensing, memory, and computing toward artificial intelligence systems, however, their practical applications are hindered by limited photoresponse and insufficient device performance. Here, we report a high-performance CuInP₂S₆ (CIPS)/CdS_0.42Se_0.58 ferroelectric heterojunction transistor enabled by a dry-transfer strategy, which synergistically combines ferroelectric polarization modulation with band-engineered heterointerfaces. The device exhibits outstanding optoelectronic performance, including a high light-to-dark current ratio of 4.37×10⁵, a responsivity of 5.38×10³ A/W, an external quantum efficiency of 1.14×10⁶%, and a specific detectivity up to 3.67×10¹⁵ Jones, along with a fast response time of 78/320 μs. Importantly, polarization switching effectively tunes carrier transport by modulating the depletion region at the heterointerface, enabling controllable photoresponse characteristics. Leveraging these advantages, the device is further demonstrated for visible-light imaging and convolutional neural network (CNN)-assisted image recognition, achieving an accuracy of 96.09%. Compared with previously reported CIPS-based devices, this work realizes a substantial improvement in both sensitivity and speed. This study highlights the effectiveness of ferroelectric-semiconductor heterostructure engineering for high-performance photodetection and provides a viable pathway toward highly integrated and intelligent optoelectronic systems.