PhononicNet
Live Explorer

Physics-informed deep learning for topological phononic crystal design. Adjust parameters below to compute band structures, predict bandgaps, and classify topological phases in real time — all running client-side in your browser.

Crystal Parameters

0.100.300.45
2.010.020.0
Fast60 k-ptsPrecise
Ready. Adjust parameters and click Compute.

Bandgap Properties

Bandgap Width
Center Frequency
Lower Edge
Upper Edge

Topological Classification

Compute to classify

Band Structure — Dispersion Relation

Unit Cell & Brillouin Zone

Phase Diagram (Bandgap Map)

Sweeps r/a and contrast — takes a few seconds

PhononicNet Architecture

Input r/a, contrast, lattice, shape
LHS Sampling Latin Hypercube Design
PWE Solver Plane Wave Expansion
PINN Model Physics-Informed NN
Predictions Bandgap + Topology

Physics-Informed Loss

L = Ldata + λ · Lphysics

Helmholtz equation residual enforces wave physics during training. Bragg scattering constraints ensure physical consistency.

Residual Backbone

Input(4) → Embed(128) → 4×ResBlock → Heads

Skip connections improve gradient flow in the non-convex physics-informed loss landscape.

Topology Classifier

Attention → MLP(128,64,32) → σ

Feature attention learns which crystal parameters drive topological phase transitions.

← All ML/AI Projects View Source on GitHub