The oscillatory retinal neuron (ORN) is a promising technology
for achieving in-sensor cognitive image computing without external
power. While its operation is based on photoinduced negative differential
resistance (NDR) at a graphene/silicon interface to directly convert
the incident optical signal into voltage oscillations, the optoelectronic
mechanism of the NDR remains elusive. Here, nonadiabatic quantum molecular
dynamics simulations show that the interplay of band alignment and
charge transfer rates of photoexcited carriers at varying applied
voltages gives rise to NDR at a graphene/silicon interface under illumination.
Such intrinsic NDR at an interface, along with extrinsic circuit-level
factors, could enable the much needed rational design of desired image
computing functionality of ORN devices in the era of ubiquitous AI
on edge devices.