AX-DX

[13]

• Multilayer plasma etching thickness prediction via shared-latent physics multi-fidelity GP (ICP-RIE, CF₄/O₂/Ar)

• Process variables (pressure / CF₄ flow / top RF) + post-etch RGB (smartphone) + stack order as input / layer-resolved remaining thickness of SiNₓ/SiOₓ bilayers (NO, ON) as output

• Single-layer (LF) log etch rates orthogonally decomposed into an affine shared trend + LF-specific GP, with only the shared-trend posterior transferred to the HF log-rate GP; sequential exposure model + variational posterior propagation for predictive uncertainty; 10 repeated splits

• SLP-MF beats HF-only & existing MF models (NARGP, LMGP) at all training sizes: RMSE 4.27 nm vs 7.03 nm for physics-based HF at n_H = 4 (7–39% reduction up to 20 specimens/stack); selective transfer > full LF transfer; lower CRPS

       In Preparation 2026

[12]

• SiO₂ etch depth prediction using multimodal in-situ diagnostics (ICP-RIE)

• OES + QMS time-series of 4 species (CF₃/O/CO/CO₂) as input / SiO₂ etch depth as output

• Multimodal fusion (OES+QMS) via ICA low-dim features + GPR; score-space domain transformation for missing-modality generation; 50× Monte Carlo CV

• Multimodal beats single-modality: OES-only R² 0.973 → OES+QMS R² 0.978, RMSE 0.188 nm; CF₃ = most important feature (permutation importance)

Under Review, Journal of Mechanical Science and Technology 2026

[11]

• Adaptive EPD modeling under limited & imbalanced plasma etching data (NO / ON bilayers)

• Process parameters as input / end-point state & oxide thickness as output

• Random Forest classification (NO) + inverse-form physical regression (ON), selected by data availability

• Robust & interpretable prediction where OES-based methods fail under data scarcity

Under Review Vacuum 2026 

[10]

• EPD & thickness prediction for SiO2/Si3N4 multilayer plasma etching

•  RGB image color + process parameters as input / layer-resolved thickness as output

• Random forest ML, physically validated by SE & XRR cross-check 

• Non-contact, smartphone-based; R2 = 0.9863, EPD 80% → 93.3%

Major Revision ACS Applied Materials & Interfaces 2026 

[9]

• Thickness prediction for HfO2 using PEALD (8-inch wafer)

• Plasma process parameters as input / thickness as output

• D-optimal augmentation of OFAT dataset (42 → 57) for better DOE coverage & lower multicollinearity

• Optimization of various ML (Ridge / RF / XGBoost / GPR): GPR R2 0.25 → 0.78, RMSE ↓ up to 2.34 nm

2nd Minor-Resubmitted Advanced Intelligent System 2026

[8]

• First hetGP-RBRDO framework for Plasma etching that separates aleatoric and epistemic uncertainties

• Uniquely used high-fidelity experimental data from industrial ICP-RIE (Lam Research)

• It enables variance-aware optimization that ensures both process uniformity and reliability

[arXiv: 2505.03826] &

Engineering Optimization 2026

[7]

• Review: ML-enabled in-situ diagnostics for intelligent plasma-based semiconductor manufacturing

• Organized by equipment (PECVD / RIE / sputtering) and by ML application (anomaly detection / diagnostics / predictive maintenance)

• From classical algorithms to deep learning — virtual metrology, real-time monitoring, process optimization

• Future directions: physics-informed learning, multimodal sensor fusion, uncertainty-aware & explainable/agentic AI toward autonomous fabs

ACS Applied Materials & Interfaces 2026 18 (27) 

(Cover)

[6]

• Plasma ions and radicals simultaneously drive physical sputtering and chemical etching of SiO₂

• Optical emission and RGB imaging capture real-time plasma–surface interaction signatures

• Ensemble models integrate process parameters and optical data for precise etch-depth prediction

• SHAP-based analysis reveals RF top power as the dominant variable governing etch depth

 Journal of Vacuum Science & Technology A 2026 44 (2) 023408

Editor's pick

[5]

• Prediction for thickness of SiO2 using RIE

• Digital Image Colorimetry (DIC) M.L for etch depth

• RGB from DIC, RIE parameter as input, etch depth as output

• Optimization of various Machine Learning (ANN, BNN) 

[arXiv: 2505.03826] &

Advanced Intelligent System 2025 8(1) 2500517

(2025년 제10회 KSME-SEMES 오픈 이노베이션 -장려상)

[4]

• Analysis of High-Temperature Fluorocarbon plasma in semiconductor process

• Discovery of new reaction pathways in Fluorocarbon plasma

• Multidimensional data visualization and process clustering

• Scalable methodology of other gas-based processes

Sensors 2024 24(22) 7307

[3]

• Three in-situ plasma dianosis tool (OES, QMS, ToF-MS)

• Plasma process parameter based plasma information using RIE 

• Correlation analysis was applied to cycleGAN (generative adversarial network)

Journal of Sensor and Actuator Network 2024 13(6), 75

특허등록 KR

[2]

• Prediction of atomic layer control for MoS2

• Two models: linear-, polynominal-regression

• Machine Learning 

Applied Science and Convergence Technology 2023 32, 106-109

(Cover)

(Best Paper Award)

[1]

• ML approach for time-varying 10th harmonics of low-k oxide (SiOF)

• High density plasma (HDP) CVD (Ultima, Applied Materials) 

• Artificial Neural Network (ANN) 

• Binary cross-entropy loss (BCEL) function

Sensors 2023 23, 8226