Rapid simulation-driven design of compact photonic Y-junction by variable-dimensional sequential approximate optimization

In this paper, a technique for expedited design optimization of a compact photonic Y-junction has been proposed. We utilize sequential approximate optimization where local response surface approximation (RSA) models are constructed and optimized to yield a sequence of designs that converge towards the optimum. The junction geometry is parameterized using splines. In order to reduce the design cost, dimensionality of the design space is gradually increased by adding the number of control points for the spline curves in the subsequent algorithm iterations. The final design exhibits transmission of over 49.5 percent and return loss below 0.3 percent over the wavelength range from 1500 nm to 1580 nm.