Fig. 6: A schematic of a photonic neural network designed to show the scalability of the proposed integrated photonic tensor core, employing a hybrid approach that combines time-division multiplexing (TDM) and wavelength-division multiplexing (WDM).
From: 120 GOPS Photonic tensor core in thin-film lithium niobate for inference and in situ training
This photonic neural network is capable of processing multiple tasks in parallel. As an example, the proposed network includes four layers. Matrix multiplication between the input layer and hidden layer 1 is performed using the TDM method. Subsequent matrix multiplications--from hidden layer 1 to hidden layer 2, and from hidden layer 2 to the output layer--are carried out using the WDM method. A comb source generates multiple wavelengths to facilitate these operations. Complementary metal-oxide-semiconductor (CMOS) comparators are utilized to implement activation functions. TFLN: thin-film lithium niobate. PD: photodetector.
