This research deals with the classification problem (CP) of two-dimensional defect patterns (DP) on semiconductor wafers detected in the final inspection process of semiconductor products such as integrated circuits. In particular, we focus on the classification problem of superposition of different DP classes which is an unavoidable problem in the practice. Previous studies focus on application of image recognition classifier using only DP data, and improvement of the accuracy based on additional transformed features of DP itself has been discussed. However, it is known and concerned that the accuracy deteriorates significantly by such an approach because the patterns and dense of DP superposition become random and high. In this research, we propose a solution by multimodal data analysis using causal variables besides DP data, which is considered to be robust to the superposition of multiple classes in principle. In the numerical evaluation, we prepared the superposition and multimodal data by adding new artificial data to WM-811K, which is the most major open data for semiconductor DP classification study. We will report the verification results to prove the robustness to 2-class SP and of all class SP as typical combinations.