Title: Weight Bank Addition Photonic Accelerator for Artificial Intelligence

Abstract: Neural networks powered by artificial intelligence play a pivotal role in current estimation and classification applications due to the escalating computational demands of evolving deep learning systems. The hindrances posed by existing computational limitations threaten to impede the further progression of these neural networks. In response to these issues, we propose neuromorphic networks founded on photonics that offer superior processing speed than electronic counterparts, thereby enhancing support for real time, three dimensional, and virtual reality applications. The weight bank, an integral component of these networks has a direct bearing on their overall performance. Our study demonstrates the implementation of a weight bank utilizing parallelly cascaded micro ring resonators. We present our observations on neuromorphic networks based on silicon on insulators, where cascaded MRRs play a crucial role in mitigating interchannel and intrachannel cross talk, a persistent issue in wavelength division multiplexing systems. Additionally, we design a standard silicon photonic accelerator to perform weight addition. Optimized to offer increased speed and reduced energy consumption, this photonic accelerator ensures comparable processing power to electronic devices.