Predictive Maintenance in semiconductor manufacturing

Over the past two years the Predictive Maintenance (PdM) capability in semiconductor manufacturing has migrated from Proof-of-Concept (PoC) and univariate Fault Detection (FD) extrapolation mechanisms to fab-wide solutions that are (1) robust to typical process and equipment disturbances, (2) extensible so as to provide solution approaches that are portable across instances of a tool type and across tool types, and (3) maintainable so as to provide solutions that are useful for long periods of time. A number of advancements have facilitated this advancement including solutions for porting modeling components across process and equipment types, mechanisms for incorporating process and equipment knowledge into models, mechanisms for determining model context (e.g., recipe) dependency, methods for model optimization to fab financials, and methods for rejecting run-time disturbances in PdM modeling. As a result of these and other innovations, the landscape of PdM in semiconductor manufacturing has rapidly advanced to the point that, from a technical perspective, solutions are now available for fab-wide PdM realization.