Researchers from George Washington University & Optelligence company in the US discovered that using photons within neural network (tensor) processing units (TPUs) could overcome these limitations and create more powerful and power-efficient AI.