Research Bits: June 24

In-sensor visual processing

Researchers from the University of Massachusetts Amherst created silicon-based in-sensor visual processing arrays that can both capture and process visual data in the analog domain to reduce the latency between sensing and identification.

The team created two integrated arrays of gate-tunable silicon photodetectors that share bipolar analog output and low-power operation, enabling one array to capture dynamic visual information, such as event-driven light changes, and one to capture the spatial features in static images to identify what the object is.

“This is very powerful retinomorphic hardware. The idea of fusing the sensing unit and the processing unit at the device level, instead of physically separating them apart, is very similar to the way that human eyes are processing the visual world,” said Guangyu Xu, associate professor of electrical and computer engineering and adjunct associate professor of biomedical engineering at UMass Amherst, in a statement. “Our all-silicon technology lends itself to CMOS integration, mass production and large-scale array operation with low variabilities.”

The device showed a 90% accuracy in classifying human motions and a 95% accuracy in classifying handwritten numbers.  [1]

Perovskite image sensor

Researchers from ETH Zurich and Empa developed a perovskite image sensor. They claim new sensor is more light-sensitive, reproduces colors more accurately, and offers significantly higher resolution than conventional silicon sensors.

The lead halide perovskite material can be modified to absorb different wavelengths of light depending on whether iodine, bromine, or chlorine ions are added to it, without the need for color filters. Since each remains transparent to other wavelengths, individual pixels for red, green and blue can be stacked on top of each other in the image sensor, increasing the amount of light that can be captured for a given surface area. This also helps eliminate certain artifacts of digital photography, such as demosaicing and the moiré effect.

One of the two perovskite-based sensor prototypes that the researchers have used to demonstrate that the technology can be successfully miniaturized. (Credit: Empa / ETH Zürich)

The prototype image sensors with pixel sizes between 0.5 and 1mm were manufactured using common thin-film processes. Because the wavelength range each layer can absorb can be precisely controlled, the researchers anticipate the approach could be used for hyperspectral imaging in machine vision by defining a larger number of color channels. They plan to work to reduce the size of the pixels and create optimized readout electronics. [2]

Self-powered artificial synapse

Researchers from the Tokyo University of Science developed a self-powered artificial synapse capable of distinguishing colors with high precision.

The device integrates two different dye-sensitized solar cells, which respond differently to various wavelengths of light, and generates electricity via solar energy conversion. It can distinguish between colors with a resolution of 10nm across the visible spectrum, which is close to the level of color discrimination in the human eye. The device also exhibited bipolar responses, producing positive voltage under blue light and negative voltage under red light, making it possible to perform complex logic operations.

The team used the device in a physical reservoir computing framework to recognize different human movements recorded in red, green, and blue. The system achieved 82% accuracy when classifying 18 different combinations of colors and movements. [3]

References

[1] Xiong, Z., Liang, W., Zhang, M. et al. Parallelizing analog in-sensor visual processing with arrays of gate-tunable silicon photodetectors. Nat Commun 16, 4728 (2025). https://doi.org/10.1038/s41467-025-60006-x

[2] Tsarev, S., Proniakova, D., Liu, X. et al. Vertically stacked monolithic perovskite colour photodetectors. Nature 642, 592–598 (2025). https://doi.org/10.1038/s41586-025-09062-3

[3] Komatsu, H., Hosoda, N. & Ikuno, T. Polarity-tunable dye-sensitized optoelectronic artificial synapses for physical reservoir computing-based machine vision. Sci Rep 15, 16488 (2025). https://doi.org/10.1038/s41598-025-00693-0