vFab-1.0-Photolithography-ML-surrogate-Tensor-Dataset

This dataset contains generated physics simulations for semiconductor photolithography, formatted as Native NCHW .npy arrays for direct ingestion into PyTorch models.

Dataset Structure

The data is split into two simulation phases:

1. Aerial Image Dataset (aerial_image_dataset/)

• Inputs (X_input/): Shape [4, 512, 512]. Contains:
  • Channel 0: Binary Layout Mask (Vertical/Horizontal Gratings, Contact Arrays, etc.)
  • Channel 1: Normalized Dose (mJ)
  • Channel 2: Normalized Defocus (nm)
  • Channel 3: Normalized Numerical Aperture (NA)
• Targets (Y_normalized_aerial/): Shape [1, 512, 512]. Hopkins-Abbe optical intensity simulation.

2. Post-Exposure Bake (PEB) Dataset (peb_dataset/)

• Targets (Y_target_M/): Shape [1, 512, 512]. Chemical reaction matrices (M-matrix) simulating the acid-base diffusion and polymer deprotection during the PEB step.

Usage

Data can be loaded directly using NumPy and PyTorch:

import numpy as np
import torch

x_input = np.load('X_input/sample_0.npy')
tensor_x = torch.tensor(x_input, dtype=torch.float32)

Note: The Hugging Face Dataset Viewer will show an error for this repository because it cannot natively render .npy tensor arrays in the browser. The dataset is fully functional and can be downloaded using the huggingface_hub library.