Intrinsically Design‐Rule‐Compliant Nanophotonic Inverse Design via Learned Generative Manifolds

ABSTRACT Inverse design has enabled the systematic design of ultra‐compact and high‐performance nanophotonic components. Yet enforcing design‐rule compliance throughout the entire optimization trajectory, without platform‐specific regularization scheduling, remains an open problem. Without explicit fabrication constraints, optimized geometries can violate minimum feature size constraints, producing structures incompatible with lithography processes. Established projection‐filter and morphological constraint methods enforce compliance within topology optimization, but rely on scheduled regularization with process‐dependent design rules. This motivates parameterization approaches that encode fabrication constraints as intrinsic properties of the design space. Here, we demonstrate intrinsically design‐rule‐compliant silicon photonic inverse design through a deep generative reparameterization that confines optimization to a learned manifold of fabrication‐compatible geometries, reducing computational cost by fivefold over unconstrained pixel‐based methods. We validate this approach across representative silicon photonic devices including broadband power splitters, spectral duplexers, and mode converters operating across the 1,500–1,600 nm band, for both electron‐beam lithography and photolithography platforms. Across all devices, the manifold‐based formulation achieves competitive performance metrics, while ensuring design‐rule compliance throughout the entire optimization trajectory. By treating fabrication constraints as a fundamental property of the design representation rather than an external penalty, this work establishes a direct pathway toward automated, platform‐agnostic, design‐rule‐compliant nanophotonic design pipelines.