Metrology and process control for semiconductor manufacturing

A semiconductor metrology system comprising:

• a spectrum acquisition tool configured to
• collect, in accordance with a first measurement protocol, a baseline set of scatterometric spectra on a first set of semiconductor wafer targets, and
• for each of one or more predefined sources of spectral variability, collect, in accordance with the first measurement protocol, a variability set of scatterometric spectra on a second set of semiconductor wafer targets corresponding to the first set of semiconductor wafer targets, wherein the variability set of spectra embodies the spectral variability;
• a reference metrology tool configured to collect, in accordance with a second measurement protocol, values of predefined parameters of the first set of semiconductor wafer targets;
• a training unit configured to
• create a training set of data from the collected sets of spectra and parameter values,
• using one or more generative models to increase the size of the training set of data, and
• use the training set of data to train a prediction model using machine learning and minimize a loss function associated with the prediction model,
• wherein the prediction model is configured to be used to predict values for any of the predefined parameters using production spectra of a third set of semiconductor wafer targets, wherein the production spectra are collected using the spectrum acquisition tool and in accordance with the first measurement protocol, and
• wherein the loss function is minimized by incorporating, for each of the one or more predefined sources of spectral variability, a term representing the spectral variability.