Tracking the performance of photolithographic processes with excursion monitoring

Eric H. Bokelberg and Michael E. Pariseau, In today’s high-volume semiconductor fabs, photolithographic process steps (wafer prime, resist-apply, expose, bake, and develop) are performed in sequence hundreds of times on hundreds of wafers per day to produce well-defined photoresist patterns. In many cases, these litho process steps are combined into one tool, referred to as a photocluster, that links a resist-process track system with an exposure tool. Because lithography plays a critical role in creating device features throughout the semiconductor production operation, accurate, repeatable photocluster performance is vital, and the individual tools are typically monitored by means of numerous tool checks (inspection procedures designed to evaluate specific components of the process). Examples include resist-film-thickness and hot-plate temperature uniformity measurements on the track, and dose-uniformity and focus-control evaluations on the expose tool. Assessment of defect performance is limited to foreign material (FM) particulate inspections through subprocesses such as resist-apply, develop only, or dry-wafer handling. While the individual tool checks ensure that many of the most critical components of the litho process are within established specifications, they are unable to monitor the interactions between components, which can create out-of-spec conditions in the printed pattern even though individual tools are in spec. Consequently, the integrated litho process is also monitored through inspections and measurements of production wafers. In-line product parametric data are sorted by photocluster and displayed in tool-specific statistical process control (SPC) charts. When out-of-control trends are identified, operation of the problem tools is inhibited until the problem can be investigated and resolved. As long as these inspections occur immediately after the litho process, and the products and levels processed on each tool are consistent from day to day (as in a large-volume, single-part-number fab), this approach to photo tool control works reasonably well. However, as fab product loading becomes more varied, with