Research map / Probabilistic and scientific modeling / Molecules and proteins
Benchmark of Machine Learning Force Fields for Semiconductor Simulations: Datasets, Metrics, and Comparative Analysis
Geonu Kim, Byunggook Na, Gunhee Kim, Hyuntae Cho, Seungjin Kang, Hee Sun Lee, Saerom Choi, Heejae Kim, Seungwon Lee, Yongdeok Kim
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train with MAE-based loss instead of MSE-based loss (ablation) | 10 MLFF models · SiN | EF metric | Improved |
| train with MAE-based loss instead of MSE-based loss (ablation) | 10 MLFF models · HfO | EF metric | Mixed |
| set force loss coefficient lambda_f below 1 (ablation) | GemNet-dT and MACE · SiN and HfO | EF metric | Hurt |
| increase lambda_f to 10 (ablation) | MACE · HfO | force error | No clear effect |
| use direct force prediction (GemNet-dT) instead of energy-derivative forces (GemNet-T) | GemNet · HfO test set | inference time: 2.2 times longer inference for GemNet-T | Improved |
| use direct force prediction (GemNet-dT) instead of energy-derivative forces (GemNet-T) | GemNet · property benchmark (B0, V0) | simulation indicator scores | Hurt |
| increase training set from 80% to 100% of HfO snapshots (ablation) | six MLFF models · HfO | EF metric | No clear effect |
| use more training data (20% to 80%) with fewer epochs (1000 to 250) (ablation) | six MLFF models · HfO | test EF metric: 1000, 500, 334, 250 epochs for 5.6k, 11.2k, 16.8k, 22.4k snapshots | Improved |
| use SCN instead of GemNet-dT | direct-force MLFF · HfO | computational cost: over 10 times higher | Hurt |
| add OOD EF metric instead of ID EF metric as guide | benchmark evaluation · simulation indicators | Pearson correlation | Improved |
Closest papers by meaning
- Towards Fast, Specialized Machine Learning Force Fields: Distilling Foundation Models via Energy Hessians (ICLR 2025 · similarity 0.86)
- Learning the Electronic Hamiltonian of Large Atomic Structures (ICML 2025 · similarity 0.84)
- MLIP Arena: Advancing Fairness and Transparency in Machine Learning Interatomic Potentials via an Open, Accessible Benchmark Platform (NeurIPS 2025 · similarity 0.84)
- The dark side of the forces: assessing non-conservative force models for atomistic machine learning (ICML 2025 · similarity 0.84)
- EDBench: Large-Scale Electron Density Data for Molecular Modeling (NeurIPS 2025 · similarity 0.84)
- Learning Smooth and Expressive Interatomic Potentials for Physical Property Prediction (ICML 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
