Interpretable ML-DFT Framework for Performance Prediction and Structure-Activity Relationship Analysis of Acidic...

Area of Science:

• Materials Science
• Computational Chemistry
• Electrochemistry

Background:

• High-performance levelers are essential for microelectronic copper electroplating.
• Current development is limited by expensive experimental screening and poorly understood structure-activity relationships (SAR).

Purpose of the Study:

• To develop an integrated machine learning (ML) and density functional theory (DFT) framework for rapid discovery of novel electroplating levelers.
• To identify key molecular descriptors governing leveling efficacy.

Main Methods:

• Established a dataset using experimental Dissolution Peak Decrease Amount (DPDA) and DFT-calculated adsorption energies (Eads).
• Employed ML models (XGBoost Regression, CART) to predict DPDA and Eads for 521 prescreened molecules.
• Utilized Pearson correlation and SHAP analyses to identify key molecular descriptors.

Main Results:

• Successfully predicted DPDA and Eads for a large set of molecules, identifying five superior novel levelers (Res-1 to Res-5).
• Res-5 demonstrated significantly superior performance compared to the industrial standard Janus Green B (JGB).
• Identified electrophilic molecular backbone and N-N functional group interactions as critical for leveling efficacy.

Conclusions:

• The integrated ML-DFT framework provides an efficient screening tool for high-performance levelers.
• The study offers new insights into molecular SAR at electrochemical interfaces.
• Novel levelers with potential industrial applications were discovered.