The electronic manufacturing industry has smaller parts and shorter production cycles, making the ability to diagnose abnormalities in equipment and products very important. Its identification capabilities have even become part of a company's competitiveness. How to quickly and efficiently identify these anomalies, as well as provide early warning and analysis, is an indispensable part of improving yield.
今天我們以電子製造行業的一個案例爲例，來看看如何透過JMP強大的神經網絡建模與資料視覺化，幫助數據分析人員大幅减少異常芯片的識別與偵測時間，幫助企業提高生產效率與quality.
Why is data analysis so important?
For high-tech companies, quality means manufacturing costs, shipping cycles and brand value. Therefore, almost every company regards product quality as its top priority. So how to improve the quality of enterprise products has become a difficult task.
(Figure 2) is a typical production step of common high-tech/type enterprises, among which:
1.Wafer manufacturing
Wafer manufacturing refers to the silicon wafer used in the production of semiconductor integrated circuits. Because its shape is round, it is called a wafer. Various circuit element structures can be processed on the silicon wafer to become integrated circuits with specific electrical functions. circuit products.
2.Package testing
Packaging is the process of processing the tested wafers into independent chips according to product type and functional requirements; testing is the process of detecting the quality of each independent chip through a series of inspection items.
Especially in the "testing" process, the traditional method is to use manual identification, that is, to record the problematic wafer diagrams one by one through personal judgment, and then perform subsequent processing. However, in actual work, manual recognition often faces the following three problems:
1. Missed detection pattern
When everyone is faced with hundreds of wafer inspection pictures, it is almost impossible to look at each picture one by one. It is likely that some important graphics are missed due to factors such as excessive fatigue, and these graphics often contain important information.
2. Taking too long
The entire manual identification process is very time-consuming, and no time is spent on solving the core problem.
3. Standards are not unified
For graphics that have not appeared before, the judgment criteria will vary according to personal standards, which can easily lead to the loss of potential information.
Based on the above inevitable factors, JMP visual analysis comes in handy.
For example, we map the inspection results of each independent chip one-to-one according to their position on the wafer. The obtained inspection results are imported into JMP. The following wafer inspection chart can be easily generated through the JMP graphic generator, thus Quickly identify problematic products, as shown in Figure 3.
Figure 3
Among them, different colors represent the size of the final test detection value. Red is the largest, decreasing in sequence, and blue is the smallest.
As can be seen from Figure 3 above, the detection value of the blue part on the lower right side of the wafer is low. What does this mean? This means that during wafer manufacturing, there is an abnormality in this area, which needs to be investigated and resolved as soon as possible. In this way, engineers can immediately start to quickly follow up on such problematic chips. Here we give an example to illustrate.
Case study: use JMP Quickly identify wafer defects
Case background
Engineers from an electronics company collected defect data for 1,559 chips during the final testing phase and wanted to analyze whether the ratio of a certain defect showed a pattern as the location changed.
- If the defect rate of a certain position is '0', it means that this defect has not occurred at this position.
- If the defect rate at this location is high, it indicates that defective products are concentrated in this location.
Through the search for this kind of defect pattern, another quality detection method or indicator can finally be established for many previous processes, which can serve as a breakthrough for overall quality improvement.
Figure 4
In the data table in Figure 4, the LocX and LocY columns are row and column coordinate values respectively, and the numerical type is nominal value. The last column is the defect rate, which is the rate of defects at a certain location over the past period of time. In order to better judge the problem, we choose the neural network algorithm here. The reasons are as follows:
- The project requires algorithms that can automatically discover features that need to be detected from raw data.
- Neural network algorithms generally have multiple levels, and each level performs a transformation, so that there are multiple opportunities to capture different features.
- Generally speaking, the more data used for training, the better the accuracy can be improved.
Next, we will explain how to perform analysis step by step in JMP.
Step1. data import JMP
First, import the data into JMP, convert the collected location defect rates into data values and present them through the graph generator.
Open JMP -> Analyze -> Predictive modeling -> Neural, put the defect rate to be observed in "Y, Response", and put Loc X and Loc Y in "X, Factor ”, as shown in Figure 5 below:
Figure 5
Enter the model configuration window and enter the following parameters. This example starts with the default neural network model and progresses to more complex models with two hidden layers and additional activation functions. Finally, the following configuration was selected, as shown in Figure 6:
In the fitting options section, tell JMP how many times to repeat the model fitting algorithm. Eventually JMP will select the iteration with the best validation statistics as the final model. This article ultimately uses 10.
Figure 6
Click the small red arrow on the left side of Model and select Save Formulas, mainly to store the transformed values for later use.
Figure 7
After this row-column conversion, the following new data table is obtained, as shown in Figure 8:
Figure 8
If we use graphics to simply illustrate the conversion process, it would be as shown in Figure 9 below:
Figure 9
Step2. use JMP Graph Builder Analysis
Next, analyze using JMP's unique "Graph Builder".
Here we put LocX on the horizontal axis, LocY on the vertical axis, Transform Value in Color in the upper right corner, click the small red arrow under Graph Builder, and select Local Data Filter, as shown in Figure 10:
Figure 10
Select Label under Local Data Filter:
Figure 11
Then, select the small red arrow - "Animation" to get the following dynamic graphics:
This animation mainly shows how JMP scans the entire wafer one by one during the internal operation process, and finally finds the optimal solution. Where H1_1 refers to the first node of the first hidden layer, and so on.
This is our idea and strategy at the model construction level. So, how should it be deployed and analyzed at the application level of actual projects? Let’s continue the discussion.
JMP Implementation: project deployment and actual results
According to the model generated by JMP above, we can then deploy it at the application level.
At the deployment level, we actually only need to build a ranking model, that is, a ranking model. This model does not need to ensure that every calculation result is very accurate. The most important thing is to be able to capture the most significant graphics in multiple wafer maps. Due to the large amount of data and the server being a Linux system (relatively stable), a python program was used to extract data from the database, and the problematic wafer map was sorted according to RSquare based on the above neural network method.
Figure 14
The reason why we choose RSquare sorting is to represent whether the pattern is caught according to the significance of the model.
That is, the more significant the model is, the greater the possibility of catching the pattern.
Regarding how to calculate Generalized R-Square (shown in the screenshot), please refer to the previous article " To increase speed and efficiency, neural networks help enterprises operate more efficiently 》
For example, there are the following 7 types of defective products (from Defect A to Defect G). Each row represents the X, Y coordinates on the wafer map, and the cumulative defect rate (in a certain week) of different defective product types at the corresponding coordinate positions.
Figure 15
According to the algorithm introduced in the previous section and the RSquare sorting as follows, we find that Defect C and Defect A have the highest values among all types of defective products (if the threshold is defined to be >0.7, of course the threshold can be defined according to the case).
Figure 16
Then the defective graphics of Defect C and Defect A generated by Graph Builder are indeed worthy of study, and the graphics of Defect D, which has the lowest value, are indeed not obvious after drawing the wafer map with Graph Builder.
Defect C (high RSquare value)-right edge graphic:
Figure 17
Defect A (RSsquare value is high) - graphic on the upper right of the center:
Figure 18
Defect D (low RSsquare value) - the graph is almost invisible
Figure 19
The following is a schematic diagram of the system architecture:
Figure 20
It should be noted that according to the sample data and calculated through JMP, the number of nodes in the first and second layers needs to be updated for the model every month (for example, for new patterns, whether the current configuration can be recognized, you need to consider adjustments number of nodes, etc.) in order to make the calculation of the model more efficient.
Through this method, after actual operation, it has been proved that each data analyst in the enterprise reduces the pattern recognition time by 60% per week on average, and the average detection time of abnormal graphics is shortened by 70% compared with before, which makes a contribution to improving the overall level of quality control. made a huge contribution.
From the above practical cases, we can see that relying solely on manual labor for product yield detection is not only time-consuming and labor-intensive, but also prone to errors and missed inspections. However, important quality risks are often hidden in certain areas that cannot be identified manually. place.
This is like an iceberg. People tend to focus on the part above the water surface, but the part below the water surface is often unable to be seen due to various reasons. Through advanced data analysis and technology, problems can be effectively solved and manual energy can be focused more on solving problems, fundamentally saving the cycle time of quality detection.
Instead of just talking, take action, try out JMP, and start your first analysis!
Further reading:
5 reasons to tell you why JMP software is more suitable for you?
A brief analysis of SPC abnormality analysis
SPC controlled product yield rate - quantile method to set product control limits and regular review
Use JMP to purify outliers, taking semiconductor analysis of WAT data as an example
