Commentary, Res J Opt Photonics Vol: 5 Issue: 5

Photonic integrated Reconfigurable Linear Processors as Neural community Accelerators

Ahmad Saleem Aktar
Department of Physics, University of Texas, USA.

Keywords: GPUs, DNNs,

Within the discipline ultra-contemporary synthetic intelligence (AI), deep has been distinctly successful over contemporary years, engaging in exquisite milestones in diverse fields, such as photograph reputation, natural language processing, genome evaluation, and independent using. the use of photographs processing gadgets (GPUs) as accelerators for computation required in deep neural networks (DNNs) became pivotal for their fulfillment. However, the computational power wished for education AI-related programs has doubled each 3.5 months inside the fashionable years. Based totally in this, the strength consumption for computation and records motion in GPU-enabled DNNs could likely quick pose a bottleneck. Photonic solutions are being investigated as a promising opportunity for knowing strength-inexperienced DNNs, modern day the inherent parallelism and the opportunity trendy enforcing neural community inference using passive optical elements. several photonic neural networks were lately proposed, based totally both on bulk or covered optics answers. Reconfigurable Photonic Integrated Circuits (PICS), are mainly desirable contemporary their flexibility, compactness, and potential strength-overall performance. in this paper, we gift two photonic included reconfigurable linear processors based mostly on 2 × 2 interferometry factors. The ones gadgets are flexible, being exploited over time for optical processing and filtering, programmable true-time optical delay lines, and quantum information processing .