Conventional machine vision shuttles data between a camera, memory and a processor, a separation that costs time and energy. The brain does not work that way: the retina and visual cortex sense and compute together. Researchers in Nazek El-Atab's group at KAUST have built a thin-film synaptic device that edges electronics towards the same integration, sensing and processing visual information in a single flexible element controlled entirely by light.
The active layer is an ultrathin manganese oxide film on a flexible silicon substrate, chosen for its abundance, low cost and multiple oxidation states. Different wavelengths strengthen or weaken the device's response, mirroring the potentiation and depression of biological synapses, and the mixed oxidation states create the defect physics behind its memory behaviour. In a playful demonstration of associative learning, the team recreated Pavlov's experiment with two light colours standing in for bell and food.
Reported in Materials Science and Engineering R, the device keeps working while bent, performs real-time image detection and carries out logic operations compatible with conventional semiconductor computing. Earlier optical synapses typically required both light and electrical inputs, or relied on unstable materials. By achieving fully photonic control in a robust, cheap oxide, the work points towards low-power neuromorphic vision for robotics, wearables and edge AI hardware.