Interface Charge\nEngineering in Ferroelectric Neuristors\nfor a Complete Machine Vision System

The rapid advancement of artificial intelligence has\ndriven the\ndemand for hardware solutions of neuromorphic pathways to effectively\nmimic biological functions of the human visual system. However, current\nmachine vision systems (MVSs) fail to fully replicate retinal functions\nand lack the ability to update weights through all-optical pulses.\nHere, by employing rational interface charge engineering via varying\nthe charge trapping layer thickness of PMMA, we determine that the\nferroelectric polarization of our ferroelectric neuristors can be\nflexibly manipulated through light or electrical pulses. This capability\nenables dynamic modulation of the device’s optoelectronic characteristics,\nfacilitating a complete MVS. As front-end sensors, devices with the\nthickest PMMA (∼32 nm) demonstrate autonomous light adaptation\nwhile those with the thinnest PMMA (∼2 nm) exhibit bidirectional\nphotoresponse characteristics akin to those of bipolar cells. Furthermore,\nas components of a back-end processor, the conductances of these devices\nwith a moderate thickness (∼12 nm) can be updated linearly\nthrough all-optical pulses. Our MVS, constructed with these neuristors,\nachieved an impressive recognition accuracy of 93% in handwritten\ndigit recognition tasks under extreme lighting conditions. This work\noffers an effective strategy for the development of energy-efficient\nand highly integrated intelligent MVSs.