BiLSTM Predicts HPGe Detector Yield
Researchers led by Dongming Mei present a data-driven framework (v1 posted Feb 3, 2026) using a BiLSTM with multi-head attention trained on time-resolved growth data from 48 Czochralski crystal runs to predict final detector-grade fraction of high-purity germanium crystals. Using SHAP, the model finds impurity concentration and growth rate chiefly govern yield, offering interpretable guidance to improve production and scale HPGe detectors for rare-event experiments.
Key Points
- 1Uses BiLSTM with multi-head attention trained on 48 crystal-growth runs to predict detector-grade fraction
- 2Identifies impurity concentration and growth rate as dominant yield drivers via SHAP feature-importance analysis
- 3Enables quantitative, interpretable in-process control to improve yield and scale HPGe production
Scoring Rationale
Practical ML application with interpretable results, limited by niche scope and small (48-run) preprint dataset.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
