A Data-Agnostic, Continuous Machine Learning Framework for Application in High Energy Physics and Beyond

Machine Learning is a revolutionary new technology that has the potential to advance a large number of scientific fields. One such area that can benefit from these advancements is data quality monitoring which is used extensively in both High Energy Physics and in semiconductor device manufacturing. In High Energy Physics, Data quality monitoring involves monitoring the data being recorded by a detector to ensure it meets proper quality standards. In semiconductor manufacturing, data quality monitoring is used to ensure a device being fabricated is free from defects. These tasks have traditionally been performed by human operators, however this process can be slow and error-prone. This proposal therefore looks to develop ensemble machine learning techniques which can automate many of these data quality monitoring tasks, allowing for greater efficiency and fewer errors. The ensemble learning technique in particular allows greater adaptability between types of systems and applications. Phase 1 of this proposal will develop the overall architecture of the ensemble machine leaning model being proposed. This model will initially be trained and tested on data from one sub-detector system of the Compact Muon Solenoid detector, the Electromagnetic Calorimeter. Once this work has been completed, the model will be validated by reusing the same architecture, but by being trained and tested using data collected by a different sub-detector system, the Inner Tracker. The final portion of the phase I effort will be to demonstrate that the developed architecture can successfully be transferred to industrial settings. This will be done by training and testing the system on material analysis data from the semiconductor industry, locally collected by Cerium Labs. The successful results of this project will be commercialized by Cerium labs to the benefit of their commercial partners by increasing the efficiency, accuracy, and throughput of their semiconductor manufacturing facilities. Machine learning improvements to semiconductor manufacturing is projected to grow the semiconductor industry by $38 billion over the next four years. Cerium Labs plans to capture 1% of this new market by leveraging their 200 annual clients from around the semiconductor industry and current market footprint supporting domestic fabrication facilities. This work will make the US semiconductor market more competitive globally, by decreasing costs and increasing efficiency.