Forward and Inverse Virtual Metrology for Phototransistor Gain: A Hierarchical, Uncertainty-Aware Approach for Small Production Datasets

This record contains the dataset and reproducibility code accompanying the manuscript "Forward and Inverse Virtual Metrology for Phototransistor Gain: A Hierarchical, Uncertainty-Aware Approach for Small Production Datasets" (under review). The dataset supports a virtual-metrology study of phototransistor gain in a semiconductor fabrication process. It comprises wafer-level measurements from two experimental arms — Arm A (strict/primary, 260 wafers across 13 process runs) and Arm B (inclusive/sensitivity, 285 wafers across 14 runs) — linking process and recipe parameters to measured device gain (GAIN), with a multilevel run → wafer → die structure. To comply with the FBK data-sharing policy, all recipe parameters are released as normalized ratios (each value divided by its per-arm reference median), preserving relative variation while withholding absolute process settings. The GAIN target is provided in its original units. The per-arm normalization constants and the raw die-level data are not part of this release. The accompanying code reproduces every result in the paper from the published data: multi-level data-quality assessment, exploratory analysis, batch- and wafer-level predictive models, a linear mixed-effects variance decomposition, Gaussian-process regression with uncertainty-based trust tiers, SHAP feature attribution, an inverse recipe search, and cluster-bootstrap confidence intervals. The pipeline can also be launched and run entirely in a browser with no local installation via Binder. Contents: normalized wafer-level datasets (CSV) for both arms, a column-by-column data dictionary and schema, the Python analysis pipeline, generated figures and tables, and documentation of the normalization method. Data are released under CC-BY-4.0; the source code is released under the MIT License. Code repository: https://github.com/mahshid-amirabgir/phototransistor-virtual-metrology