Automatic calibration of high-density photonic processor with reconfigurable micro-ring resonators for neuromorphic computing

The prevalent and complex AI-based services have caused the rapid growth of demand for high performance computing (HPC). The conventional accelerators such as graphic processing units (GPU) have faced obstacles regarding energy efficiency and computational resources. Optical approach, utilizing silicon photonic integrated chip provide solutions in terms of speed, parallelism, and energy efficiency. In this work, we propose a photonic processor with coupling ratio variable micro-ring resonators (MRR), consisting of asymmetric Mach-Zehnder Interferometers (MZI) as variable couplers. Considering the number of channels, the coupling ratio variable MMR customizes the FWHM, maximizing the bit resolution while minimizing the error rate. Also, the resonant wavelength is fixed during the entire process which facilitates significant increase in channel density. The entire calibration process is successfully verified with 4-channel photonic processor fabricated with the Si photonic foundry service.