Expectable surprises of silicon substrate mass production data using machine learning
Multi-wire sawing technology is an efficient way of slicing silicon substrates for solar and semiconductor applications in mass production. Over 40 wire sawing parameters and properties influence the substrate quality and interact in a complex manner.
Within the joint project “Wafer 4.0” silicon substrates were produced in an Industry 4.0 environment and an elaborate data analysis of the multi-wire sawing process was carried out. Mass production data acquisition and generation comprised the largest part of the work. The heterogeneity of these data made data collection and mapping very complex. This multitude of data formats was successfully linked together that a cloud-based data analysis via predictive systems models (machine learning) was possible. These models made correlations and predictions about the dependence of substrate geometry on the wire sawing process across space and time. The analysis model showed well known correlations – in order of their influence and gave surprising input of parameters, which no one thought might matter.
In future this knowledge will improve stable substrate mass production and give new insights in multi-wire sawing technology applied to other hard and brittle materials like glasses, ceramics and composites, other crystals and alloys.