Machine learning to predict the environmental footprint of semiconductor processes

/ Machine learning to predict the environmental footprint of semiconductor processes Machine learning to predict the environmental footprint of semiconductor processes Master projects/internships - Leuven | More than two weeks ago Use unsupervised machine learning to help understand the environmental impact of semiconductor manufacturing The semiconductor industry is not only growing rapidly, but its manufacturing processes are increasing in complexity, requiring more and more energy, water, and materials. With those natural resources already under strain, to remain compatible with planetary boundaries (e.g., to align with the drastic emissions reduction required to mitigate climate change), it has become critical for technological advancement to integrate environmental sustainability “by design”. To tackle this challenge, the SSTS program at imecis works along with partners from across the value chain to assess and improve the sustainability of the semiconductor industry. The central tool to this endeavour is a “virtual fab” model that simulates high-volume manufacturing semiconductor fabs with all their related flows and their environmental impacts (accessible through the imec.netzero webapp). This model requires accurate data on the tools and processes used in semiconductor manufacturing (e.g., required energy, processing time, chemicals, ...). However, due to confidentiality concerns this data is not always readily available or shareable. Objective This internship will apply machine learning techniques to tackle two related data sharing issues in the context of semiconductor manufacturing sustainability assessment: First, defining a model that can predict missing values in SSTS databases. Typically, this would involve predicting the flow of a given chemical of a tool, given the chemical flows of similar chemicals and tools as well as the specific data available for the target tool, for example using (sparse) nonnegative matrix factorization (NMF). Second, aggregating tool and processes databases into a representative set of “generic tools/processes” that can be shared without confidentiality concerns, by relying on privacy-aware unsupervised machine-learning tools. An extension of the models (e.g. NMF) from the first objective might be highly relevant. Responsibilities You will actively engage in the research and modelling of a few selected mitigation options, and implementing and testing those models into the imec.netzero virtual fab code. This will involve working closely with both our research team to understand the nuances of semiconductor manufacturing processes and life‑cycle assessment methodology, as well as with the coding team to ensure seamless integration of your models within the imec.netzero system. Your contribution will greatly enhance the environmental impact reduction roadmaps of semiconductor production. Skills and Learning Objectives: Applicants are expected to have a background in applied mathematics with some experience in “traditional” machine learning*, and have solid experience in Python programming. They have the desire to gain proficiency and enhance their skills in the following key areas: Unsupervised ML, privacy‑aware ML, nonnegative matrix factorization Semiconductor Manufacturing Processes Professional Software Engineering (architecture, documentation, clean code) *Here “traditional machine learning” refers to ML approaches besides deep neural networks, which would be unsuitable in this context due to a lack of large training datasets. Type of project : Internship, Thesis, Combination of internship and thesis Required degree : Master of Engineering Technology, Master of Engineering Science, Master of Science Supervising scientist(s) : For further information or for application, please contact: Vincent Schellekens ( ) imec's cleanroom #J-18808-Ljbffr