Large orthogonal array-based optimization for high-dimensional parametric systems

Although electromagnetic (EM) simulators are able to simulate systems and devices with a larger number of design parameters, the next challenge is to use such simulators in order to provide parametric simulation and optimization in a high-dimensional design space, while minimizing the number of calls to the EM simulator. This paper presents an optimization approach which can significantly reduce the number of calls to the EM simulator through a case study of a high-dimensional parametric system which contains 127 design factors. Unlike traditional simulated annealing (SA) and particle swarm optimization (PSO) which randomly select values to create all of the combinations of the variables, the proposed optimization approach uses Taguchi's method which employs large orthogonal arrays (OAs) to create a test set that has an even distribution of all combinations. Therefore, it is more efficient, concise and the global optimum can be guaranteed. To further shorten the simulation time, sensitivity analysis is first conducted and unveils that the top 20 most volatile design factors attribute to 80% of the performance variation. Therefore, the 20 parameters are optimized using a large OA of 361 rows and 20 columns. The results show that the proposed optimization approach is capable of optimizing high-dimensional parametric systems swiftly and may potentially have a wide range of applications.