Machine Learning in Deeptech: Machine Learning Before Machine Learning
Abstract:
Machine learning depends on data. However, many deeptech and EIC Pathfinder projects involve early-stage process development with limited data and few experiments or samples.
Machine learning and AI are useful for development, but research activities are only one link in a larger chain. Optimizing a single process step may miss strategically important data connected to manufacturing, sustainability, traceability, and future exploitation. What data should be considered before large datasets are available? The webinar aims to open discussion and exchange around “machine learning before machine learning” and how early-stage deeptech projects can prepare for machine learning, AI, and data-driven value creation. As a case study, experiences from CS-PVT pilot activities in rePowerSiC will be presented.
More details:
Machine learning depends on data. However, many deeptech and EIC Pathfinder projects are still in early-stage development, where data is limited, experiments are few, and processes are still being developed.
Machine learning and AI are often proposed for developing a specific activity. However, the activity is one link in a larger (value) chain. Optimizing only a single process step may overlook strategically important data connected to the broader value chain. This is especially important in Pathfinder projects, where research is expected to progress towards impact and exploitation.
What data should be created already now, before large datasets are available?
This challenge exists in rePowerSiC and many other deeptech projects. The CS-PVT process used to fabricate SiC base layers (epiwafers) for high-power laser conversion may eventually reach high technical performance. However, technical performance alone is not sufficient. A real challenge is that once there is a commercial market, PVT manufacturing (like China) will learn about CS-PVT and rapidly scale epiwafer production capabilities.
Competitiveness may therefore depend not only on process performance itself, but also on data-driven added value.
Machine learning should extend beyond optimisation of individual process steps towards value-chain optimisation, predictive maintenance across manufacturing systems, process intelligence, traceability, sustainability positioning, and other aspects that emerge across the chain rather than within a single technology step.
The webinar discusses how deeptech projects can think about:
- machine learning with limited datasets,
- identifying strategically important data early,
- linking technical development to future exploitation,
- integrating sustainability and value-chain perspectives,
- and building foundations for post-project effect and long-term impact.
We will present experiences from our own pilot in CS-PVT.
Organized by the EIC Pathfinder project rePowerSiC.