Data Mining can be defined as the task of extracting statistically meaningful information from data. In the past, statisticians have been doing this activity by performing "custom analysis" on an as needed basis. In the semiconductor field, we must generate numerous statistical process control (SPC) charts and associated data summaries on a regular basis to monitor process performance. Consequently, we must find more efficient methods of analyzing process data and making the information readily available to the engineering community. The information must be accessible in a timely and user-friendly manner. At the Merrimack, NH facility of Texas Instruments (previously Unitrode), we have done this by integrating the capabilities of SAS, the Unix operating system and the Promis material tracking system that operates in the OpenVMS environment. The resulting information is updated daily and placed at an "SPC Website" so that the users can view it using only the "point and click" techniques. This paper describes some of the steps in the long "chain of events" that are essential to make this happen without human intervention. The statistician can now devote his energies to other duties that require his personal attention, such as: design experiments, perform gauge studies, develop & deliver statistics / DoE courses for the engineering community, publish / upgrade SPC specification and support the Quality group's continuous improvement projects. This is in sharp contrast to the traditional methods for computing control limits that depended strongly on human interaction to: 1. Extract process data from Promis (engage repeatedly in almost identical dialogue) into a VMS file; 2. Access that file from the Windows operating system using FTP; 3. Convert it in the CSV format using Excel; 4. Determine the structure of that file to invoke just the right SAS 'macro' for making a dataset; 5. Invoke the right SAS macro to make data summaries, compute limits and make SPC charts; 6. Repeat all of the above for each new chart. Given that a semiconductor facility must track a large number of processes daily, the task of creating and maintaining SPC charts would have been a full time job for more than one statistician. There would be no time for other projects that more clearly required statistician's personal attention.
展开▼