Filtering out redundant visual information at sensor level drastically reduces energy consumption in real-time computer vision
https://www.eurekalert.org/news-releases/1145490 "reconfigurable multi-gate, atomically thin molybdenum disulfide transistor paired with ferroelectric hafnium zirconium oxide allows single device to dynamically switch between image data filtering (control signals specify pixels of which image regions to let through)/ AI brain-inspired spiking neural network functioning... reduces active compute blocks roughly one-third with virtually no lost accuracy... wafer-compatible manufacturing, highly power-efficient... autonomous robots, edge AI"