CMOS-compatible 2D semiconductor-nitride ferroelectric based optoelectronic memristive transistors for neuromorphic sensory computing and optical decoding

Optoelectronic memristive devices that enable neuromorphic sensory computing with integrated sensing, memory and processing functionalities, have been identified as a promising element for next-generation artificial vision systems. However, the optoelectronic memristive dynamics of these devices are largely based on engineered structures and/or innovative functional nanomaterials, presenting critical challenges for their silicon complementary metal-oxide-semiconductor (CMOS) technology compatibility at a scalable application perspective. Here we demonstrate a CMOS-compatible optoelectronic memristive device based on the aluminium scandium nitride (AlScN) ferroelectric/two-dimensional (2D) semiconductor heterostructure enabled ferroelectric field-effect transistor (FeFET). We show that such FeFETs can be fabricated under the back-end-of-line (BEOL) thermal budget and exhibit non-volatile electronic memory properties. Harnessing the light-induced ferroelectric domain reconfiguration in the AlScN layer, these devices offer light-tunable short- and long-term memristive properties, directly linking light stimulation history to electronic dynamics. Such a working principle enables rich optoelectronic synaptic functions and the implementation of the psychological human visual memory model, offering a critical demonstration for optical neuromorphic sensory computing. Furthermore, a linear and non-volatile dependence of the photoresponse current on the light intensity is exploited for the optical information decoding. Our results mark new opportunities for CMOS-compatible optoelectronic memristive devices as the hardware basis of next-generation neuromorphic vision architectures.