[2P4-J-2] Machine learning: industries and finance

[2P4-J-2-01] Efficient estimation for red-zone in silicon wafers for solar cells using Level Set Estimation

〇Shota Hozumi¹, Kota Matsui², Kentaro Kutsukake², Toru Ujihara⁴, Ichiro Takeuchi^1,2,3 (1. Nagoya Institute of Technology, 2. RIKEN, 3. Center for Materials Research by Information Integration, National Institute for Materials Science, 4. Nagoya Universary)

Keywords:Machine Learning, Gaussian Process, Active Learning

For the task of estimating a spacial distribution of a physical quantity, it is common to x the measurement
positions to meshgrid points evenly allocated along the coordinates of the space. However, such xed measurement
positions often contain redundancy in the sense that not all the measurements in the meshgrid points are required
for the target task. Especially when a measurement of the physical quantity is costly, it is thus benecial to allocate
the measurement points adaptively and reduce the number of measurements. In this study, we applied Level Set
Estimation (LSE), which is a method to efficiently estimate the boundary position, to carrier lifetime mapping of
silicon for solar cells, and estimated the low quality region. Our approach can reasonably estimate the boundary
position by measuring less than 1% position compare to conventional approach.