Matthew R. Graham

Lithography light source fault detection

High productivity is a key requirement for todays advanced lithography exposure tools. Achieving targets for wafers per day output requires consistently high throughput and availability. One of the keys to high availability is minimizing unscheduled downtime of the litho cell, including the scanner, track and light source. From the earliest eximer laser light sources, Cymer has collected extensive performance data during operation of the source, and this data has been used to identify the root causes of downtime and failures on the system. Recently, new techniques have been developed for more extensive analysis of this data to characterize the onset of typical end-of-life behavior of components within the light source and allow greater predictive capability for identifying both the type of upcoming service that will be required and when it will be required. The new techniques described in this paper are based on two core elements of Cymers light source data management architecture. The first is enhanced performance logging features added to newer-generation light source software that captures detailed performance data; and the second is Cymer OnLine (COL) which facilitates collection and transmission of light source data. Extensive analysis of the performance data collected using this architecture has demonstrated that many light source issues exhibit recognizable patterns in their symptoms. These patterns are amenable to automated identification using a Cymer-developed model-based fault detection system, thereby alleviating the need for detailed manual review of all light source performance information. Automated recognition of these patterns also augments our ability to predict the performance trending of light sources. Such automated analysis provides several efficiency improvements for light source troubleshooting by providing more content-rich standardized summaries of light source performance, along with reduced time-to-identification for previously classified faults. Automation provides the ability to generate metrics based on a single light source, or multiple light sources. However, perhaps the most significant advantage is that these recognized patterns are often correlated to known root cause, where known corrective actions can be implemented, and this can therefore minimize the time that the light source needs to be offline for maintenance. In this paper, we will show examples of how this new tool and methodology, through an increased level of automation in analysis, is able to reduce fault identification time, reduce time for root cause determination for previously experienced issues, and enhance our light source performance predictability.