Random defects, which account for more than 90% of electronic product manufacturing defects, are defects caused by foreign substances. Although the detection ability has been improved through various efforts using optical inspection and image features, the ultimate enhancement to improve quality and yield cannot be achieved just by detecting defects. In particular, there was a limit to classifying and specifying the failure source because the ingredients were not known, and it was often impossible to eliminate defects even after spending a lot of time. We developed a world first in‐fab Raman spectroscopy microscopy to be applied to the analysis of foreign substances in the entire area of the 8.5th generation glass substrate and applied it in real time between manufacturing. As a result, it is possible to detect defective sources in real time and to improve quality and yield through process control and ultimate improvement.
In the display industry, the technology of the FAB process is continuously being advanced. As process quality management technology improves, data can be stored, analyzed, and monitored in real time using various types of sensors. The manufacturing process is so complex that it is difficult to detect anomalies simply by analyzing data from on e or two sensors. For t his reason, multivariate data analysis is essential, and time‐series data analysis is particularly effective for detecting process state changes. In this study, we propose a multimodal AVA that can detect process anomalies using multivariate time‐series sensor data. In order to minimize t he information loss of the original data, multivariate time‐series data are processed into tabular and image forms and applied to the proposed model. We demonstrate that the proposed model performs better in anomaly detection compared to a general anomaly detection model in highly imbalanced datasets. The proposed method is expected to reduce cost and time due to defects by detecting abnormal situations in the process in real time and responding quickly when abnormal situations occur.
As manufacturing processes become increasingly sophisticated, the abundance of real‐time multivariate data is also increasing. However, the vast majority of data in manufacturing is normal, and abnormal data is scarce, making it difficult to establish correlations between various data sets. While a model‐centric approach holds importance when applying AI in manufacturing, a data‐centric approach, based on domain knowledge of each process equipment, can yield better results. We refer to this approach as engineer's sense. We used the anomaly transformer as an anomaly detection model for time series data, and were able to improve performance through improvements with data centric. The study found that a focus on data‐centric improvement led to higher performance compared to model‐centric improvement. This method identified 88% higher anomaly F1 scores in the evaluation and is expected to be utilized on additional manufacturing equipment in the future. This will result in significant cost and time savings.
'%3e%3cg id='b'%3e%3cpath id='a' stroke='%23000' stroke-width='2' d='M-6-25H6m-12 3H6m-12 3H6'/%3e%3cuse xlink:href='%23a' y='44'/%3e%3c/g%3e%3cpath stroke='%23fff' d='M0 17v10'/%3e%3ccircle r='12' fill='%23cd2e3a'/%3e%3cpath fill='%230047a0' d='M0-12A6 6 0 0 0 0 0a6 6 0 0 1 0 12 12 12 0 0 1 0-24z'/%3e%3c/g%3e%3cg transform='rotate(-123.69)'%3e%3cuse xlink:href='%23b'/%3e%3cpath stroke='%23fff' d='M0-23.5v3M0 17v3.5m0 3v3'/%3e%3c/g%3e%3c/svg%3e)
'/%3e%3cuse xlink:href='%23a' transform='rotate(23.036 .093 25.536)'/%3e%3cuse xlink:href='%23a' transform='rotate(45.87 1.273 16.18)'/%3e%3cuse xlink:href='%23a' transform='rotate(69.945 .996 12.078)'/%3e%3cuse xlink:href='%23a' transform='rotate(20.66 -19.689 31.932)'/%3e%3c/svg%3e)