Using reinforcement learning to determine optimal alignment for an optical system

A method for aligning components of an optical system, the optical system comprising an electronic model generated based on a corresponding physical optical system, the method comprising:

• determining, with one or more processors, a state of components of the optical system, the state determined based on input information associated with the components of the optical system;
• determining, with the one or more processors, based on the state of the components, an alignment metric associated with the components of the optical system, the alignment metric comprising an indication of whether alignment requirements for the components of the optical system are satisfied for the state;
• and
• iteratively, with the one or more processors, adjusting the state of the components based on the alignment metric, wherein:
• iterative adjustments are configured to enhance the alignment metric;
• and
• each adjustment during the iteratively adjusting is determined based on a set of possible actions in an action space associated with the components of the optical system, and a corresponding set of observations that are observed for the components of the optical system in response to an adjustment of the state,
• wherein determining the state, determining the alignment metric, and iteratively adjusting the state are performed as part of a reinforcement learning framework wherein some adjustments are rewarded or penalized more than others based on an adjustment speed, semiconductor manufacturing process throughput associated with the optical system, a relative alignment metric enhancement, a relative alignment metric diminishment, and/or an amount of additionally required adjustments to the components of the optical system caused by an adjustment.

A computer-implemented method for aligning components of an optical system, the method comprising:

• determining, with one or more processors, a state of components of the optical system, the state determined based on input information associated with the components of the optical system;
• determining, with the one or more processors, based on the state of the components, an alignment metric associated with the components of the optical system, the alignment metric comprising an indication of whether alignment requirements for the components of the optical system are satisfied for the state;
• and
• iteratively, with the one or more processors, adjusting the state of the components based on the alignment metric, wherein iterative adjustments are configured to enhance the alignment metric,
• wherein determining the state, determining the alignment metric, and iteratively adjusting the state are performed as at least part of a reinforcement learning framework.

A method for aligning components of an optical system, the optical system comprising an electronic model generated based on a corresponding physical optical system, the method comprising:

• determining, with one or more processors, a state of components of the optical system, the state determined based on input information associated with the components of the optical system;
• determining, with the one or more processors, based on the state of the components, an alignment metric associated with the components of the optical system, the alignment metric comprising an indication of whether alignment requirements for the components of the optical system are satisfied for the state;
• and
• iteratively, with the one or more processors, adjusting the state of the components based on the alignment metric, wherein:
• iterative adjustments are configured to enhance the alignment metric;
• and
• each adjustment during the iteratively adjusting is determined based on a set of possible actions in an action space associated with the components of the optical system, and a corresponding set of observations that are observed for the components of the optical system in response to an adjustment of the state, wherein determining the state, determining the alignment metric, and iteratively adjusting the state are performed as part of a reinforcement learning framework, the reinforcement learning framework configured such that iteratively adjusting the state of the components comprises penalizing adjustments that diminish the alignment metric, and rewarding adjustments that enhance the alignment metric, for individual adjusted states during a sequence of adjustments, until final alignment requirements for the optical system are satisfied.