An Analysis of Elusive Relationships in Floating Zone Growth Using Data Mining Techniques

Abstract

High-purity silicon single crystals, essential in the renewable energy value chain, can be grown using the Floating Zone (FZ) method. Increasing the yield of the FZ process while maintaining its stability is a complex but sought-after goal. This study examines intricate relationships in FZ growth, focusing on how representative crystal quality and process stability measures are influenced by various process parameters simultaneously. Data mining techniques are applied to synthetic data from numerical simulations. Regression Trees identified model parameters and their ranges responsible for complex behavior of the quantities of interest, some of which are undetected by bivariate correlation coefficients. Quantities at the center of the crystal are highly affected by the crystal radius and pulling rate, while quantities near the surface of the crystal are more sensitive to the reflector and inductor parameters due to their proximity. The results illustrate how data mining techniques can support informed parameter engineering of the FZ process toward desirable goals.