In semiconductor chip production, the inner wall of the rapid thermal annealing chamber will be affected by the coating due to long-term product processing, thus affecting the internal temperature control. Therefore, it is necessary to regularly detect the health of the chamber by measuring the silicon dioxide thickness or resistance of the wafer after rapid thermal treatment to determine whether to perform preventive maintenance ( PM ). We found that the stability of the silicon dioxide thickness of the existing machine is different from that of the Golden tool, but it cannot be directly quantified, and the attribution analysis is relatively complicated. Therefore, we used JMP to quantify the process capability of the chamber and assist in the attribution analysis, and finally achieved better performance than the Golden Tool. We first visualized the process capability of the chamber using the IMR control chart, process performance chart, and target chart, and then clearly quantified the process capability of the chamber by calculating the process capability index and stability index. Subsequently, the multi-factor model analysis was used to explore the differences in WtW and WiW uniformity between the existing machine and the Golden tool. The XBar-S control chart and process capability chart were used to compare the performance of silicon dioxide thickness and resistance after annealing between the existing machine and Golden tool. It was shown that the resistance uniformity after annealing was much better than that of Golden tool , which was more stable and capable, so the hardware factor of the machine could be ruled out. After further analysis, it was found that this kind of film thickness change also potentially exists in Golden tool. After process debugging, the stability of silicon dioxide thickness was greatly improved, and the resistance uniformity and stability after debugging were consistent with those before process debugging. The comparison of process performance chart, target chart, process capability index and stability index also showed that the effect of this process debugging was feasible.
