Polishing semiconductor wafers using causal models

A method comprising:

• repeatedly performing the following:
• obtaining a configuration of input settings for polishing a semiconductor wafer;
• determining a measure of a quality of the semiconductor wafer polished with the configuration of input settings;
• and
• providing the measure of the quality of the semiconductor wafer polished with the configuration of input settings to a system that updates a causal model mapped to a distribution over possible control settings such that the causal model that measures current causal relationships between the input settings and the quality of semiconductor wafers, wherein probability matching is utilized to map impact measurements and confidence intervals in the causal model to probabilities and the control settings are assigned based on these probabilities;
• and
• re-computing the causal model by computing overall impact measurements based on confidence intervals around the overall impact measurements and means of d-scores.

A system comprising one or more computers and one or more storage devices storing instructions that when executed by the one or more computers cause the one or more computers to repeatedly perform the following operations:

• select a configuration of input settings for polishing a semiconductor wafer, based on a causal model mapped to a distribution over possible control settings such that the causal model measures current causal relationships between input settings and a quality of semiconductor wafers, wherein probability matching is utilized to map impact measurements and confidence intervals in the causal model to probabilities and the control settings are assigned based on these probabilities;
• receive a measure of the quality of the semiconductor wafer polished with the configuration of input settings;
• adjust, based on the measure of the quality of the semiconductor wafer polished with the configuration of input settings, the causal model;
• and
• re-compute the causal model by computing overall impact measurements based on confidence intervals around the overall impact measurements and means of d-scores.