Structured inverse design: a tile-based approach for practical photonic integration

We present a tile-based inverse design framework for photonic device layout that leverages direct binary search (DBS) with fabrication-aware constraints. In contrast to gradient-based methods, our approach utilizes binary tile primitives to maintain geometric compliance with lithographic limits throughout the optimization process. To enhance the performance of baseline DBS, we introduce four heuristic modifications: (1) macro-initialization for improved starting conditions, (2) resolution variation to balance complexity and tractability, (3) randomized tile traversal to reduce spatial bias, and (4) tile shape diversity to increase design expressivity. We apply this framework to a Y-branch splitter, a 3-port power divider, and a photonic XOR gate, using full-wave FDTD simulations. Systematic evaluation reveals that macro-initialization provides the most substantial performance improvement toward functionally coherent, high-FOM layouts. Randomized traversal improves performance in higher-resolution designs, while resolution variation identifies an optimal grid size that balances layout granularity and efficiency. Tile shape diversity supports directional routing and structural refinement, with its impact depending on device-specific geometry. Our findings demonstrate that tile-based DBS enables scalable, fabrication-compatible photonic integration and supports the synthesis of complex optical functionalities without resorting to continuous or post-processed layouts.