4:30 PM - 4:45 PM
▲ [18p-212A-11] Applying Simulated Annealing Method to Optimization Problems in Silicon Photonics
Keywords:simulated annealing, silicon photonics, modulator
The simulated annealing (SA) algorithm offers efficient global optimization in a large multivariate space for large-scale complex nonlinear systems where we have no knowledge of the problem structure and solution cannot be explicitly described or achieved by exhaustive traversal. Due to its robustness of global optimum, the SA algorithm has been applied to training artificial neural network for machine learning towards automatic driving [1] and many tough engineering fields such as VLSI design [2]. Photonic devices and systems face similar optimization problems because many determinative variables in both design and control consist of an N-dimension solution space that is impossible be explored analytically. We propose to apply the SA algorithm to deal with the optimization problems in photonics including automatic design, calibration, and intelligent control towards functionality reconfiguration, all of which are becoming important with integration being scaled up. According to the desired applications, achieving adaptive functionality reconfiguration without changing photonic devices can offer great system flexibility and decrease the cost. In this study, we utilize the SA algorithm to globally optimize the five control parameters (5-dimension solution space) of a single silicon Mach-Zehnder modulator (MZM) to realize an arbitrary 4-level optical waveform generation from an arbitrary initial state. We report this 5-multivariates 4-objectives optimization as an application example of SA algorithm for function reconfiguration of single MZM. We developed the SA algorithm with adaptively adjusting the random step selection according to the algorithm progress and achieved fast global convergence.