Answers are grounded in this paper and cite their sources.
Topological states in phononic crystals (PnCs) exhibit remarkable localization and robustness, making them highly advantageous for applications including waveguiding, energy harvesting, and acoustic sensing. However, the precise tailoring of specific topological states remains a considerable challenge. Prior studies have largely focused on band gap characteristics and their associated topological properties, from which topological states are inferred indirectly. To address this limitation, we introduce a deep learning framework for the direct engineering of topological interface states (TISs) and topological corner states (TCSs) at predetermined frequencies within two-dimensional topological PnCs. The unit cells, represented as 32 × 32 pixel matrices with free-form geometries, are compressed into a 10-dimensional latent space via a conditional variational autoencoder, thereby improving design efficiency. Forward prediction models employing multilayer perceptrons are developed to estimate band gap boundaries, their topological properties, TIS gaps, and TCS frequencies. For the inverse design, we propose a Latent-Consistent Tandem Network that enables both direct and multi-objective design of target structural properties and topological states. By integrating the filling ratio as an additional design parameter, the framework enables multi-objective inverse design that simultaneously considers physical properties and geometric features. The approach generates PnCs with prescribed topological band gap boundaries and topological state frequencies, achieving correlation coefficients exceeding 0.96 between the target and achieved properties. Furthermore, the framework supports “one-to-many” inverse design and allows modulation of the filling ratio in the generated unit cells. Customized designs of TIS gaps and TCS frequencies are successfully realized and experimentally validated. This work underscores the potential of deep learning techniques for the intelligent design of topological PnCs and establishes a novel paradigm for the development of engineered functional materials.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: