Any diagnosis is inseparable from information collection and analysis. Compared with figures and statistics, people are always better at analyzing problems through graphical information, especially when the data presents problems that change over time. Take GPS, which records your daily walking information, as an example. Although the upper and lower parts of the map below show the journey of the same day, the map track below is obviously easier to understand than the coordinates and time records above.
Process equipment diagnosis
In order to realize the diagnosis of process equipment, Fanlin group has built a practical program named “lamda”. The program can read the log data and show the changes with graphics, which is very suitable for the comparison of different wafers or cavities. Customers and our front-line engineers use this program frequently, and its current valid licenses have exceeded 8000. The best way to use lamda is to use the information provided by lamda to confirm or refute the hypotheses put forward in advance. However, if the scope of the problem is too wide to start with (that is, the problem of “looking for a needle in a haystack”) or if there is a problem but it is not found at all, how to deal with it?
In the face of greater challenges, our coping methods also need to change, just as the classic line of the movie “great white shark” says: “we need to change the ship to deal with the big guy.”. The latest generation of big data analysis tool chamber matching diagnostic tool (d2cm) launched by Fanlin group is such a “big ship”. It adopts machine learning algorithm, which can be used to summarize and analyze the data of all devices in a long period of time. In other words, this tool can actively help you to detect the “needle” hidden in the vast ocean, even if you don’t know the existence of such a “needle”. It can be said that d2cm is “super lamda”.
D2cm is usually used to evaluate the performance of a group of equipment running the same process (or application). Although in mass production, customers can determine whether the equipment operates according to the set parameters by controlling the limit value, there are still errors in different equipment. These differences are not only difficult to identify, but also often difficult to determine the underlying causes. The multivariate (or multidimensional) analysis of d2cm can identify these statistically significant errors and find the direction of improvement. One of the advantages of this method is that it can comprehensively consider the interaction between different signal data (remember, pressure, temperature, power, etc. are interrelated), improve the “signal-to-noise ratio” and find the root cause faster.
Due to the high complexity of machine learning, we often need a lot of training to master its use. If you really need to know how to translate the analysis results into maintenance or corrective actions, you need more skills, because you need to know the process equipment very well.
Is there a better way?
We’ve been trying to develop a “better way.”. Our idea is to comprehensively analyze the basic scenarios of large-scale use of big data and develop them into simpler “applications”. For example, the program may be dedicated to evaluating all configuration settings of all devices in the same group, or a “subsystem health model”, or a program to help you complete the daily operation check of devices. In addition, the program also built a “sandbox” environment, process engineers can easily access the equipment data and the core engine of d2cm, try out different analysis methods, or visualize the data. The first batch of applications of the development project will be put into use in the second half of 2020, and more applications will join in the future, some of which will come from customers’ own sandboxes.
All of these efforts are to improve the quality of data and analysis, so as to better improve the performance and productivity of the equipment.