Case Study: ASML achieves virtual overlay metrology and improved yield prediction with MathWorks MATLAB

ASML Develops Virtual Metrology Technology for Semiconductor Manufacturing with Machine Learning

ASML, a leading photolithography equipment manufacturer, faced a throughput-driven metrology challenge: its TWINSCAN systems could only measure overlay on about 24% of wafers, risking undetected layer-alignment errors that reduce yield. To generate overlay estimates for every wafer without slowing exposure, ASML turned to MathWorks products—MATLAB with the Statistics and Machine Learning Toolbox and Deep Learning Toolbox—to prototype a machine-learning–based virtual metrology solution.

Using MathWorks tools, ASML built and trained a nonlinear autoregressive (NARX) neural network with Bayesian regularization on alignment data and YieldStar overlay measurements, then validated the model and developed a prototype real-time overlay controller. The MathWorks-based solution produces virtual overlay metrology for every wafer (vs. only 24% measured), uncovered systematic and random overlay errors that would have gone undetected, supported potential yield improvements down to and below the 5 nm node, and reduced maintenance overhead by leveraging MATLAB’s legacy code base.

ASML

Emil Schmitt

Weaver