Machine vision is an integral part of intelligent systems such as autonomous vehicles and robotics. Visual information is usually captured by a frame-based camera, converted into a digital format and then processed using a machine learning algorithm, such as an artificial neural network. Passing data through such a signal chain, however, results in low frame rates and high power consumption. Thomas Mueller and colleagues now demonstrate an image sensor that itself can constitute an artificial neural network, simultaneously sensing and processing optical images in real time without latency. The sensor is a reconfigurable two-dimensional semiconductor-based photodiode array capable of achieving supervised and unsupervised learning, and the team train the system to classify and encode images. This scalable platform shows the potential of using two-dimensional materials for ultrafast machine vision applications.
Recent Hot Topics
Sign up for Nature Research e-alerts to get the lastest research in your inbox every week.