Machine learning systems for monitoring of semiconductor processing

A computer program product for controlling processing of a substrate, the computer program product tangibly embodied in a non-transitory computer readable media and comprising instructions for causing a processor to:

• receive a plurality of sets of training data, each set of training data including a plurality of training spectra, a timestamp for each training spectrum from the plurality of training spectra, and a starting characterizing value and/or an ending characterizing value for the plurality of training spectra;
• store a plurality of machine learning models, each machine learning model providing at least one different hyperparameter;
• store a plurality of physical process models, each physical process model providing a different function to generate characterizing values as a different function of time and/or a different physical process parameter;
• receive a selection of a machine learning model from the plurality of machine learning models and a selection of a physical process model from the plurality of physical process models to provide a combination of a selected machine learning model and a selected physical process model;
• generate an implemented machine learning model according to the selected machine learning model;
• for each training spectrum in each set of training data, calculate a characterizing value based on the timestamp for the training spectrum, the starting characterizing value and/or ending characterizing value for the set of training data, and the selected physical process model, thereby generating a plurality of training characterizing values with each training characterizing value associated with one of the plurality of training spectra;
• train the implemented machine learning model using the plurality of training characterizing values and plurality of training spectra to generate a trained machine learning model;
• and
• pass the trained machine learning model to a processing control system of a substrate processing system.

A semiconductor fabrication system, comprising:

• a plurality of polishing systems, each polishing system including a support to hold a polishing pad, a carrier to hold a substrate against the polishing pad, a motor to cause relative motion between the substrate and the polishing pad, an in-situ monitoring system to generate a sequence of measurements of the substrate during polishing and a timestamp for each measurement in the sequence of measurements, and a controller,
• wherein at least one controller of at least one of the plurality of polishing system is configured to cause one or more of the plurality of polishing systems to polish a series of training substrates,
• wherein one or more controllers of one or more systems from the plurality of polishing systems are configured to receive a trained machine learning model, to cause the polishing system from the one or more systems to polish a series of device substrates, to receive a sequence of measurements of the device substrates from the in-situ monitoring system of the one or more systems, to generate a sequence of characterizing values based on the sequence of measurements and the trained machine learning model, and to control at least one polishing control parameter of based on the sequence of characterizing values;
• an in-line or stand-alone metrology system to generate a starting characterizing value and/or an ending characterizing value for each of the series of training substrates;
• and
• an algorithm generation platform comprising one or more processors and non-transitory computer readable media storing a computer program product comprising instructions for causing the one or more processors to
• receive, for each training substrate, from the in-situ monitoring system of one or more of the plurality of polishing systems used to polish the training substrate, a plurality of training spectra generated during polishing of the training substrate and a timestamp for each training spectrum from the plurality of training spectra,
• receive, for each training substrate, from the in-line or stand-alone metrology system, the starting characterizing value and/or an ending characterizing value for the training substrate,
• store a plurality of sets of training data, each set of training data including the plurality of training spectra from the training substrate, the timestamp for each training spectrum from the plurality of training spectra, and the starting characterizing value and/or an ending characterizing value for the training substrate,
• store a plurality of machine learning models, each machine learning model providing at least one different hyperparameter,
• store a plurality of physical process models, each physical process model providing a different function to generate characterizing values as a different function of time and/or a different physical process parameter,
• receive a selection of a machine learning model from the plurality of machine learning models and a selection of a physical process model from the plurality of physical process models to provide a combination of a selected machine learning model and a selected physical process model,
• receive at least one hyperparameter value for the selected machine learning model and at least one physical parameter value for the selected physical process model,
• generate an implemented machine learning model according to the selected machine learning model and the at least one hyperparameter value,
• for each training spectrum in each set of training data, calculate a characterizing value based on the timestamp for the training spectrum, the starting characterizing value and/or ending characterizing value for the set of training data, the physical parameter value, and the selected physical process model, thereby generating a plurality of training characterizing values with each training characterizing value associated with one of the plurality of training spectra,
• train the implemented machine learning model using the plurality of training characterizing values and plurality of training spectra to generate a trained machine learning model, and
• pass the trained machine learning model to the controller of the one or more polishing systems for control of polishing of the device substrates.