SemiAI Unveils AI Technology for Sub-1nm Overlay Prediction

SemiAI (CEO Taekwon Jee) unveiled an AI technology that predicts sub-1nm overlay—a key challenge in next-generation semiconductor processes—at SEMICON Korea 2026, the country's largest semiconductor industry exhibition.

"According to the International Roadmap for Devices and Systems (IRDS), the 3-sigma (3σ) overlay budget for future semiconductor processes is projected to drop below 1nm," said SemiAI CEO Taekwon Jee. "As process noise exceeds control limits, securing atomic-scale precision with conventional feedback control becomes difficult."

To address the expanding process noise and data scarcity that arise in advanced process environments requiring atomic-scale precision, SemiAI presented the Virtual Fab framework, which leverages physics-based synthetic data. Virtual Fab augments real process data with large-scale synthetic data and trains models on extreme process variation conditions, improving prediction stability.

The company also introduced a domain-driven multimodal fusion model that combines scanner log data with process domain knowledge, improving prediction performance over conventional linear regression and XGBoost approaches.

Applying the overlay prediction AI, mean error (RMSE) dropped from around 0.9nm to 0.3nm, and the coefficient of determination (R²) improved from 0.4 to 0.9. The results indicate precision approaching the requirements of sub-1nm processes.

"As semiconductor scaling accelerates, the paradigm of process management is shifting toward prediction," Jee said. "AI-based overlay prediction will define next-generation semiconductor manufacturing competitiveness."