Inverse design of bandgap for randomly encoded topological photonic crystals based on the SSH model via deep learning

Topological photonics introduces concepts from topological physics and opens innovative avenues for photon manipulation. As a key platform for topological states, photonic crystals leverage their periodic structures to regulate photonic bandgaps, thereby providing an ideal foundation for studying higher-order topological states. However, traditional design methods that rely on experience-driven approaches and numerical trial and error struggle to resolve the intricate mapping relationships between photonic structures and topological bandgaps, thereby limiting the efficiency of developing high-performance devices. This study proposes a randomly encoded photonic crystal structure that overcomes the constraints of conventional periodic designs by incorporating high-degree-of-freedom parameter spaces and complementary topological robustness mechanisms. By leveraging deep learning-based inverse design, we demonstrate wide bandgaps and topological corner states across diverse refractive-index platforms, including III–V compounds and silicon nitride. Our work unlocks possibilities for expanding the applications of various materials in topological photonic devices.