In many industries, the control of fine dust or other foreign matter plays a crucial and pivotal role in the performance and quality of the output products. The areas involved include semiconductor production, LCD screen production, biochemical engineering, biotechnology, precision Among the industries such as equipment, foundry industry, pharmaceuticals, hospitals, etc., especially The highest dust-free level is required in the semiconductor wafer process , Because today's wafers have been developed to the nanometer level, any dust falling on the wafer may cause damage to the entire wafer circuit, and even cause the wafer to be scrapped. All kinds of running water may affect the quality and capability of the overall product, and then affect the external view and evaluation of the company. Although it sounds like a small matter, it has a wide-ranging impact and has to be careful and vigilant.
With the progress of the times, the production time and development period of products have also been greatly reduced, which also means that the industry must be more accurate in risk assessment and production line monitoring, that is, early detection of out of control production lines to reduce possible losses. . Under this trend, How to quickly find abnormal products, and try to catch defective products in the early stage of production, so as to avoid making products that do not meet specifications, is also the top priority faced by production lines and quality management departments in various industries. .
JMP provides many tools for fallout analysis to help customers analyze fallout data and find potential problem points. From preliminary data drawing, to consolidating data for simple statistical analysis, or using statistical tools for in-depth analysis of abnormal projects, JMP can provide perfect support. Today, I will share how to use JMP for semiconductor dust analysis. .
Data drawing and preliminary graphic analysis
JMP's Graph builder platform can provide a fast observation platform, draw data into graphs, and can draw the corresponding point size according to the particle size of the dust point, and use the gradient color to distinguish the particle size, so that you can understand the actual situation more quickly and intuitively. condition.
In the following, we use the data of falling dust on the panel as an analysis case illustration.
Figure 1. Use Graph builder to make a scatter plot of falling dust, and use point size and color to indicate particle size
JMP also has a number of interactive graphing features that allow you to observe the interrelationships between factors by adjusting the graph, and even help you find key influencing factors. The Local data filter under each platform can filter the categories you want to discuss according to the variables you set. Using this function, we can confirm whether there are different distributions between different panels.
Figure 2 Graph builder cooperates with local data filter to filter data
Using the column switcher function, we can quickly switch the parameters discussed on the Graph builder platform, and can quickly find out the factors that have a greater impact. In other words, the factors can better distinguish the data.
Figure 3 Graph builder cooperates with Column Switcher to switch variables
If you want to compare the proportion of dust particles in different lots or between panels, you can also use a pie chart to compare, the following figure is an example, if we want to compare the proportion of dust particles in the number 07 panel is closer to 06 or 12, you can By drawing a simple pie chart, it can be found that the proportion of dust particles of panel 07 is closer to that of panel 12.
Figure 4: The pie chart compares the similarities and differences in the proportion of falling dust
For a small amount of data, we can draw a pie chart to observe the situation of falling dust and find panels with similar proportions of falling dust, but when the amount of data is slightly larger, is there any other way to find the problematic panel?
Group the data, and group those with similar proportions into the same group
In order to solve this problem, JMP provides many simple platforms to help users analyze abnormal points on larger data. First, let's try to use JMP to group the performance based on the proportion of dust fall.
In order to simplify the calculation, we first make a simple grouping of the particle size, and use the distribution platform of JMP to find the overall distribution percentile data. The quantile is used as the dividing point to divide the data into tiny particles, small particles, medium particles, large particles, and ultra-large particles.
Figure 5 Using Distribution to do descriptive statistics analysis
Next, use Hierarchical Clustering to According to the proportion of particles to group, we can find JMP Group similar proportions into the same group , with the help of JMP, we classify the overall panel into three categories, we can find that category 1 seems to have a ring-shaped dust fall trend, while category 3 seems to have some special large particle points and clusters on the upper and lower left corners After the subsequent collection includes more samples, these data can be used as a panel group to judge which kind of dust percentage the new data is closer to. Combined with the subsequent production line results, it can be detected in advance in the previous process. Potential opportunities for defective products to avoid subsequent production of substandard products.
Figure 6 Classification by Hierarchical Clustering according to the proportion of falling dust
After understanding how to group samples with the same trend, we further need to clarify which panels have which particular dustfall problems.
use JMP Analysis of abnormal dustfall data
In order to find those panels with special dustfall conditions, we first simply divide the abnormal dustfall into several items.
- Large Particles - Panels with a much larger fallout particle size relative to the overall fallout particle size data
- High Quantity - The panel has a much higher amount of dust fall than the overall dust quantity data
- Clustered dust - relative to the overall dust data, the panel dust has a higher concentration in some areas
JMP Explore Outliers on The platform can quickly search out the outliers we set , in this case, we use the Quantile Range Outliers interface to find that Tail Quantile is 0.005, and Q is 4, and the point of the outlier is set to be greater than [Percentile 0.995+4*(Percentile 0.995-Percentile 0.005 )] or less than [Percentile 0.005-4*(Percentile 0.995-Percentile 0.005)], it can be found that the upper limit is 541, and the particles that exceed it are regarded as large particles and need to be marked.
Figure 7 Using Explore Outliers to filter outliers
Using Graph builder or Tabulate, we can find that Panels 4~12 have large particles.
Figure 8 uses Graph builder and Tabulate to present the number of large particles of dust on each panel
If you want to select panels with a large amount of dust, use Tabulate to sort out the amount of dust on each panel, and then use the Explore Outliers platform to select the panels with more dust particles. Using this example, we will think that the three panels 06, 07, and 02 have a lot of dust problems.
Figure 9 Using Tabulate to sort the amount of falling dust and using Explore Outliers to filter panels with a large amount of falling dust
Next, we have to deal with the problem of clustered particles. The Normal mixtures platform of JMP provides the ability to judge whether the data points are general random scattered points or special concentrated points. JMP divides the data into several categories according to the degree of aggregation of the points, which is somewhat similar to the method of removing the back of the picture. The general random scattered points are used as the background, and the remaining points are classified and grouped. In this example, we divide all the data after excluding noise points into 5 groups, and then use Graph builder to draw them. It can be found that 02, 06, 07, 09, and 12 have ring-shaped cluster particles, especially in 02, 06 ,07 is particularly obvious on these three panels, while 05, 10, 11, 08 have some clustering points in the upper and lower left corners, 04 has two clustering points on the left, and the clustering points 21~30 are not so obvious , mostly small-scale clustering points or only one clustering point. Such results also correspond to the previous results of Hierarchical Clustering.
Figure 10 Use Graph builder to draw the classification of dust after Normal mixtures classification (distinguished by color)
Dust analysis conclusion
According to the previous analysis and using JMP to organize the following results.
Figure 11 Analysis and arrangement of dust problems
Using the statistical analysis and related interactive graphical tools on JMP, we can quickly and accurately understand the data, Graphical presentation also helps us find problem points more quickly and cooperate with JMP Convenient and highly adjustable tool to help users quickly catch key problem points , in this example, the clustering of annular particles seems to be the main problem. Using JMP, we narrow the problem to a few specific panels, and then we can do follow-up analysis to confirm whether the problem is concentrated in a certain equipment or personnel operation. and so on. By using JMP, you can save time and effort faster, and maintain the quality of your products with less money.
Recommended reading:
Learn how to draw a wafer map with JMP in 7 minutes
How to quickly identify wafer defects? Data Analysis in High-Tech Semiconductors
How to conduct reliability test for IC products? JMP reliability acceleration analysis application
