A device made from aluminum gallium nitride can detect light, retain information about it and perform basic processing, according to a study described in a November 11 report in Phys.org. The researchers’ approach uses charge at polarized interfaces to make one component switch between sensing and memory functions.
The study, published in Nature Materials, was led by researchers from the Changchun Institute of Optics, Fine Mechanics and Physics of the Chinese Academy of Sciences, Peking University, East China Normal University, the Shanghai Institute of Technical Physics of the Chinese Academy of Sciences, and Zhejiang University. The report says the device detected light at the nanowatt-per-square-centimeter level and operated stably at temperatures up to 300°C.
Combining sensing and memory could help reduce the movement of raw data between separate detectors, memory units and processors. That matters in in-sensor computing, where processing some information near the point of detection may reduce the burden on the rest of a system. The report describes existing approaches that can depend on defects to store charge, potentially consuming light-generated carriers, or use stacked two-dimensional materials that are difficult to manufacture over large areas. III-nitride semiconductors, by contrast, have established material quality, fabrication processes and industrial supply chains.
The device’s aluminum gallium nitride heterostructure consists of alternating layers with different compositions. At the boundaries between layers, polarized charges create built-in electric fields. Those fields guide light-generated electrons and holes toward different interfaces, reducing the chance that the carriers recombine.
Asymmetric nickel and titanium contacts control how those separated charges move. Under forward electrical bias, holes remain at one interface while electrons produce a readable current, allowing the device to retain information about incoming light. With zero or reverse bias, the stored holes are released and both types of carriers move rapidly, putting the component into detector mode. The same physical structure can therefore provide multiple operating modes through electrical control.
The report describes several forms of memory behavior: a persistent optical state after bright illumination, multiple temporary states from weaker light pulses, and a return to the initial condition through an electrical reset. In detector mode, the device responded on the nanosecond timescale. It also showed a large current contrast and repeatable write-and-erase behavior after extended storage in air.
To assess possible computing uses, the researchers entered experimentally measured device states into a neural network simulation that classified clothing images. The device-based model achieved more than 90% accuracy and remained close to the corresponding full-precision software model. This is a simulation result using measured device behavior, rather than evidence that a complete hardware system has achieved that classification performance. Tests on a fabricated small array also showed that individual pixels could detect and retain optical inputs, supporting the possibility of parallel sensing and local preprocessing.
The work demonstrates individual devices and a small array, not a complete large-scale in-sensor computing system. Direct evaluation in larger hardware is still needed to establish system-level energy use, integration density, readout circuitry and manufacturing yield. The results identify interfacial polarization as a possible design tool for combining functions within a mature semiconductor material family; whether it simplifies practical integration will depend on how the approach performs as hardware scales.