Quality Evaluation of Single-Crystal Wafer X-Ray Detection Using Random Vector Functional-Link Network.

An X-ray orientation instrument is a primary nondestructive means to detect the quality of single-crystal wafers. However, some problems exist, such as low accuracy and low efficiency, which hinder its industrial application. This article proposes an original approach to wafer quality evaluation. The limited multipoint strategy, instead of the single-point strategy, is used first to evaluate quality. Then, using interval analysis theory for feature expansion, the point-valued data is transformed into the interval data to reduce the number of detection points and cope with small sample situations. Finally, combining interval analysis with an improved cross-layer point-valued random vector functional-link (CP-RVFL) network, a cross-layer interval RVFL (CI-RVFL) network was designed to evaluate the quality of single-crystal wafers using interval data. The proposed approach effectively solves various difficulties in evaluating wafer quality and can be seen as a relatively general quality evaluation method. Several typical industrial data experiments have demonstrated the feasibility of CI-RVFL, especially when the number of training samples is small. Its performance is significantly better than that of general point-valued RVFL (P-RVFL) and CP-RVFL.