Scalable Nanophotonic-Electronic Spiking Neural Networks

Spiking neural networks (SNN) provide a new computational paradigm capable of highly parallelized, real-time processing. Photonic devices are ideal for the design high-bandwidth, parallel architectures matching SNN paradigm. Furthermore, co-integration CMOS and photonic elements combineslow-loss with analog electronics greater flexibility nonlinear elements. We designed simulated an optoelectronic spiking neuron circuit on monolithic silicon photonics (SiPh) process that replicates useful behaviors beyond leaky integrate-and-fire (LIF). Additionally, we explored two learning algorithms potential on-chip using Mach-Zehnder Interferometric (MZI) meshes as synaptic interconnects. A variation Random Backpropagation (RPB) was experimentally demonstrated matched performance standard linear regression simple classification task. In addition, applied Contrastive Hebbian Learning (CHL) rule to network composed MZI random input-output mapping The CHL-trained performed better than guessing but did not match (without constraints imposed by meshes). Through these efforts, demonstrate co-integrated SiPh technologies well-suited scalable computing architectures.