The deep neural network classifies nearly 2billion images per second, and the new brain like optical classifier chip is on nature

The research team at the University of Pennsylvania reported an integrated end-to-end photon depth neural network （PDNN）, The network directly processes the light waves impinging on the on-chip pixel array , Sub nanosecond image classification is performed as they propagate through the neuron layer .

In every neuron , Optics performs linear calculations , Photoelectric realization of nonlinear activation function , The classification time is allowed to be less than 570 ps, Equivalent to the single clock cycle of the most advanced digital platform . The uniformly distributed supply light can provide the same light output range of each neuron , So it can be extended to large scale PDNN. The second and fourth categories of handwritten letters are proved , The accuracy is higher than 93.8% and 89.8%.

Direct of optical data 、 Clockless processing eliminates the need for analog-to-digital conversion and large memory modules , So as to provide the next generation of deep learning system with faster 、 More energy-efficient neural networks .

The research 「 An on-chip photonic deep neural network for image classification」 entitled , On 2022 year 6 month 1 Published on 《 Nature》.

Thesis link ：

https://www.nature.com/articles/s41586-022-04714-0