Data Mining Systems for the Semiconductor Industry

PRESS CENTER
Articles

Filters & Miners Improve Productivity By Jon Goldman

Some of the most powerful, and in my opinion undertilised, tools in helping achieve productivity improvements are software-based information systems. Relatively quick and inexpensive to develop, these suit a variety of special purpose applications, are flexible and capable of being modified to meet changing needs (key caveat: provided the software is well written). In particular, data capture and analysis software interfaced to process equipment helps predict system failures, determine schedules, verify successful maintenance operations, reduce test runs, and eliminate test wafer usage. Easily retrofitted to existing machines with minimal impact on existing process flow, these can have enormous productivity gains. The older approach of relying on individual experience and sixth sense to know the idiosyncrasies of equipment must give way to the new: excellent computer skills but little direct experience and capable of improving equipment productivity, if given access to relevant information. Two important classes of software tools used with data collections are filters and miners. Both are engineering tools that can be used to extract the real truth about process and equipment performance from collected data. HIGH BANDWIDTH DATA TRANSFORMATION VIA FILTERS One category of equipment data is fixed, static and immobile: the other is fleeting, repeating and dynamic. Just as the brain extracts only small amounts of the information an eye "sees" to accomplish vision, the objective in operating fab equipment should be to extract the image of the process that best meets production requirements. Fab equipment can emit machine data at incredibly high rates. The first challenge is obtaining high-level access to this low-level data in a usable form. Process tools in a fab emit different sets of process variables, whether they use a standard language like GEM or a bus like the RS488, they still say different things. Software filters provide the means to collect variables, abstract, normalise and push data into common forms, e.g. fields and timestamps in records, common records into tables. As a wafer passes through process equipment it follows a series of steps, each with a function of varying importance and each with the varying potential to damage the wafer. As an example, many processes are sensitive to gas-phase nucleation problems caused by flow overshoots when an MFC ramps. The critical event is the behaviour of an MFC during flow initialisation. Consider pumpdown and backfill steps common to many types of equipment. Improper venting, poor automation maintenance and a host of causes can result in particle generation. Critical event is the successful load and unload of wafers without introducing particles onto their surface. This is where the focus of data filters needs to be trained. The filter's goal should be to distill out what is critical for successful operation so an effective process becomes one in which its critical events are well behaved. CRITICAL EVENT BEHAVIOUR REVEALED BY MINERS The other challenge is obtaining data mining tools that deliver filtered information in a useful form. Effective data mining starts with ideas, mental models and a nose for where problems lie. You must be able to question "If x is happening, then the data should show me y." The functions of the Miner are to determine which events are critical. Good data mining tools should be capable of extracting the few kernels of useful information from reams of raw and filtered data available from metrology, maintenance logs, and input parameter statistical process control data from equipment. ACTIONS BETTER THAN SUCCESSFUL SPC EXPERIMENTS The final challenge is to assist engineering and manufacturing to implement systems that integrate critical event monitoring with routine production process flows so that improvements become institutionalised. It is also essential (but difficult) to build upon the corporate storehouse of useful knowledge by archiving data analysis success stories to make them available to engineers at other sites and future generations of new engineering recruits who are destined to "rediscover" the same issue on their own. Good filters and miners not only help provide major productivity gains, but can also provide invaluable assistance when new equipment emerges in the process line. ABOUT THE AUTHOR Jon Goldman received his PhD degree in materials science from MIT. In 1973 he joined Motorola SRDL where he worked on LPCVD process development, then in its infancy. Subsequently technical director at Thermco Systems, he founded Jon Goldman Associates in 1987. E-mail him at jongoldman@compuserve.com.