Table 1.

Time cost for 1000 shape predictions with ML (ResNet) and FE for different types of design: ‘s’ for seconds and ‘h’ for hours

Time cost for 1000 predictions
Design type, Nx × Ny × Ny	ML (ResNet)	FE
Fully random, 15 × 15 × 2	3.6 s	11 h
Island, 15 × 15 × 2	3.6 s	28 h

For the purpose of dataset generation, the FE simulations can be run in parallel to significantly reduce the time cost.

The estimation for time cost is based on a single CPU core (Intel Core i9-10900) and one GPU (NVIDIA Quadro P620).