Machine Learning Could Make Gas Purity Data Do More Than Pass or Fail

At the point where ultra-high-purity gas passes from supplier to fab, a process analyzer’s main job is to confirm the gas meets specification, which makes the data stream “intentionally and ideally kind of boring,” says Jeff Wuenschell, gas analysis applications scientist at AMETEK Process Instruments. In a SEMICON West talk, he looked at how machine learning could extract more value from that data: diagnosing problems upstream, informing downstream processes and predicting maintenance on the analyzer itself.

The obstacle is the data. Events worth detecting, such as an upstream valve failure or purifier breakthrough, are rare by design, leaving little to train a model on, and fabs are reluctant to share proprietary process data. Wuenschell also raises whether better detection could push purity roadmaps lower: if an analyzer can see contamination begin to rise well before the alarm level, fabs gain time to react.

The work draws on AMETEK’s ta7000 trace gas purity monitors, with sub-ppb detection of H2, CO, CO2, CH4 and NMHCs, and its 5910 and 5920 UHP moisture analyzers, which reach 150 pptv and 1.0 ppbv detection limits.

Click here to read the full article in the SEMICON West Show Daily.

The post Machine Learning Could Make Gas Purity Data Do More Than Pass or Fail appeared first on Semiconductor Digest.