Chip makers are gearing up for 686-class micro-processors and 64Mb production. These require producing <0.30 mu m critical features. They are also focusing on 256Mb DRAMs that require below 0.25 mu m design rules. Therefore, the shift from mercury lamp based i-line to 248nm Deep-Ultra-Violet (DUV) steppers has begun in earnest. Current KrF laser models from several suppliers satisfy the optical requirements necessary for pilot production. However, as the move from DW lithography R&D environment to production occurs, the laser's manufacturability, uptime requirements and cost-of-operation (CoO) become more demanding. The chip maker requires Reliability, Availability and Maintainability CRAM) data from laser manufacturers to support uptime and CoO estimates. Often, such data are required long before the lasers actually go into production. The general approach hitherto followed by laser suppliers in determining these numbers was to lifetest one laser for a certain duration and document the failure mechanisms. The cost-of-operation is then estimated from the cost of replacement parts or sub-systems. The sub-systems that make the highest contribution to the cost-of-operation are then addressed (e.g. laser chamber). However, nuisance failures (such as blown fuse, defective valve, software bugs related shutdowns) are rarely addressed although in some cases these failures alone can cause significant downtime during laser manufacture/test and during use by chip makers. Therefore, the CoO data essentially reflect the cost of replacements parts. The RAM data, however, is at best, an educated guess by the laser manufacturer. In the strict sense, RAM data are measures of equipment performance which have been widely used in industry for several years. Such is not the case for any excimer laser for DW lithography, where the transition to it's use in a production environment from R&D has just begun. The actual usage data, today, from these lasers are negligible. In the approach used here for estimating the RAM data, advantage is taken of the fact that a statistically significant number of lasers are manufactured and tested for DUV lithography at Cymer. The failure data from these lasers are gathered, analyzed and corrected by the techniques described below. In this paper, Cymer reports the first use of Failure Reporting Analysis and Corrective Action System (FRACAS) in determining the manufacturability, reliability and uptime performance of a laser optimized for DUV lithography. FRACAS is a state-of-the-art closed loop process to record, group and analyze failures and preventive maintenance actions. Software to measure the process is granted to Cymer by SEMATECH. It provides the required data for corrective action, highlights developing failure modes, and contributes data for statistical analysis. By far, the most important aspect of FRACAS is that it can be used to ensure closure of all problems. The failures and preventive maintenance events of ELS-5000 lasers are captured by engineering, manufacturing and service personnel. The failures are grouped by problem, and problems are assigned owners who determine the Root Cause Solution (RCS). Problems are then retired only after the RCS has been determined effective by subsequent monitoring. As a result, the number of failure events decrease, resulting in improved manufacturability and reliability of the laser. As an example, in Figure 1 we show the failure rate per laser during the manufacture and test of these lasers, measured over 43 lasers, provided to us courtesy of Cymer Manufacturing. Simultaneously, the key modules of the ELS-5000 laser are reliability tested separately, under stress and under typical operating conditions. Again, their failure mechanisms are analyzed for RCS. Finally, a ELS-5000 laser is tested under typical stepper or scanner operating condition and its performance over a certain period recorded. This laser incorporates all the product corrections and improvements identified as solutions to the failures. According to SEMI E10-96, if reliability is improving (typical during early life of a product), then an overall RAM calculation is misleading. Instead a reliability growth model, such as the one developed by US Array Materials Systems Analysis Activity (AMSAA), must be used. However, in this case, by using FRACAS, Cymer rapidly converged to a stable product, and data from the tests performed on this laser may be analyzed via the techniques outlined in SEMI E10-96. An AMSAA model is not required to analyze the data. In what follows, we describe the laser tests and the subsequent analysis to extract RAM data and the corresponding 80% confidence interval. Although a ELS-5000 laser was used for the study, the discussions are also valid for Cymer's EX-5000 laser which are optimized for scanners with catadioptic optics.
