Process Drift and Automatic Process Control

7.1 Adjusting for process drift There are times when processes drift, and the best course of action is to make an adjustment. This situation occurs when overlay correctable parameters or focus offsets are changed. Every lithography operation should have controls to detect process drift. As discussed in Chapter 1, the ±3σ rule is not particularly sensitive to shifts in the process mean, compared to other Western Electric Rules. Consequently, it is advisable to include at least one additional Western Electric Rule to supplement the ±3σ rule to increase sensitivity to process drift. An alternative to the Shewhart control chart, with greater sensitivity to shifts in the process mean, is the exponentially-weighted moving average (EWMA) chart, discussed in the next section. Suppose that a process has been under control until a Western Electric Rule is violated. If, after investigation, no assignable cause can be identified, it may be decided that the process should be adjusted. Determining the amount of adjustment is the subject of this section. Each of the Western Electric Rules involves a certain amount of data. The ±3σ rule is violated when a single point exceeds control limits, while the eight-in-a-row rule involves 8× more pieces of data. As discussed earlier, the process mean μ is known to within a “±3σ” level of confidence (99.7%) following measurements: With more measurements there will be a smaller uncertainty in the actual process mean (Fig. 7.1). For this reason, it is useful to repeat measurements when Western Electric Rules have been violated. This is particularly true for the ±3σ rule, for which there may be only a single measurement indicating a loss of process control. As seen in Fig. 7.1, the violation of the ±3σ rule could have been caused by only a very small shift in the process mean. Changing the operating point to move the single violation to the process mean would most likely represent over-compensation. The exponentially weighted moving average, discussed in the next section, provides a means of both detecting and estimating the process mean, particularly in situations involving gradual drift. Suppose only the ±3σ rule has been violated, and it is proposed to adjust the process assuming that the single measurement represents a new shifted process. The process is then adjusted on the basis of this assumption to some percentage of the difference between the single value and the process target.