WDM Weighted Sum in an 8x8 SOA-Based InP Cross-Connect for Photonic Deep Neural Networks

We demonstrate high fidelity weighted addition of four 10 Gb/s on-off keyed data channels on a SOA-based monolithically integrated cross-connect, with a reading error dispersion below 0.2 with respect to the target. These results open the route to feasible 8x8 one-layer neuron interconnectivity for photonic integrated deep neural networks.