Fig. 1: Neuroshaper overview. | npj Computational Materials

Fig. 1: Neuroshaper overview.

From: Shaping freeform nanophotonic devices with geometric neural parameterization

Fig. 1

a Visualization of gradient-based optimization in the high dimensional Neuroshaper parameterization space. The optimizer avoids forbidden zones imposed by constraints to reach a suitable local minima. b Overview of the Neuroshaper framework for nanophotonic device design. The pipeline consists of: (1) Neural design representation combining sampling points with lookup tables and multilayer perceptrons to generate continuous level set functions; (2) Level set operations to obtain binary/grayscale patterns; (3) Multi-resolution conversion between high (M1) and low (M2) resolutions; (4) Design evaluation based on geometric constraints and electromagnetic performance; (5) Parameter gradient computation via backpropagation.

Back to article page