Stella original measuring system (ST/LST series) by the CAD data match-up method
Development of AOI system with AI Function
Stella Corporation succeeded to develop an AOI system with AI (Artificial
Intelligence) function. It's possible to enhance inspection accuracy by
learning judgement result automatically using AI function in this system.
First of all, the company will adopt this function as AOI system for the
screen mask.
▲Flow of AI learning
In this system, firstly, the substrate is inspected automatically. As a result, inspection result is displayed on the PC monitor. The next, sort of detected defects is categorized, such as, foreign particle, remained emulsion, and misinformation (quasi-defect), by AI function. An operator confirms this judgement, and then, judges whether or not AI judgement is correct. AI function is improved continuously by accumulation of this review result as the database.
Of course, misinformation occurs inevitably in AOI system. However, it can be minimized in this system, as a result, duty of operator can be reduced. Furthermore, sensual inspection with human sense is expected to be realized in the future by repeated learning.
▲AOI sample of the screen mask
▲Screen of defect list
Outline extraction function of emulsion pattern (for screen mask use)
Picture 1 Image sample by binarization treatment by brightness
If you make use of ST series as measurement for screen mask use, additional
value appears on the scene. It's to delete mesh pattern in screen mask
by original image processing treatment technology and to find out outline
of emulsion pattern exactly.
As you know, in general, if screen mask is manipulated an image, it's made use of binarization treatment by use of difference of brightness. However, in this case, as figure, mesh pattern in emulsion pattern remains because of mixed mesh pattern and emulsion pattern, so that, it is difficult to measure emulsion pattern exactly.
To solve this problem, we have developed more than 20 kinds of image filtering treatment programs such as binarization treatment by color phase and succeeded to delete mesh pattern only and capture emulsion pattern only.
Concretely, first of all, you capture original pattern on screen mask by a CCD camera (Picture 2). Subsequently, you set up above various filtering treatment (Picture 3). As a result, mesh on screen mask vanishes and emulsion pattern only is captured on monitor(Picture 4). In this time, you can find out edge of emulsion pattern exactly by alignment of outline of detected emulsion pattern and that of original pattern (data) (Picture 5). Finally, measurement point is measured automatically. As a result, we could succeed to enhance measurement accuracy compared to the conventional treatment method such as binarization treatment by use of difference of brightness.
Picture 2 Captured original pattern by a CCD camera
Picture 3 Setting picture plane of image filtering treatment
Picture 4 Image after mesh pattern vanished
Picture 5 Original data image which is superposed with extraction outline
AOI (Automatic Optical Inspection) function
Additionally, we could succeed to build in AOI function, too. It's possible to superpose original CAD data and real pattern on device and monitor on a PC monitor. Therefore, you can find out particle and defect by monitoring. Operators can judge admission decision sensuously because of superposing by monitor. For this reason, unskillful operator can treat easily, too. Also, this function is effective for not only PCB but also photo mask, and etc.
Furthermore, it supports pinhole defect in solid film in addition to fine pattern such as electrode. Therefore, it is useful for inspection of mask-blanks.
▲Image sample of pattern ※Red part is ×1, blue part is ×2, orange part and yellow part are ×3. In short, it's possible to set up various scale factors with each other based on area at will and record inspection result with notation of scale factor. Therefore, it's effective for high speed inspection use.
▲Image sample in case of superposing (Green part is CAD data and its below part is real pattern on device)
Correspond to AOI
Furthermore, we have succeeded to inspect automatically by original algorithm. As picture, it's possible to set up scale factor with each other based on area. Therefore, it's possible to inspect area in rough pattern with low scale factor and area in fine pattern high scale factor. Of course, it's possible to set up threshold value of judgment at will. Also, it's possible to indicate difference scale factors of images continuously.
▲Image sample of inspection
Our new AOI system equips an area CCD camera with 5M pixel and a line sensor CCD camera with 7,400 pixel. The former is used for measurement of pattern and review of AOI, on the other hand, the latter is used for AOI. This system makes it possible to find out open and short defect of pattern, pinhole defect, protrusion defect and so on automatically, to realize high speed image processing treatment like a real-time process because high speed image treatment is performed immediately after scanning of line sensor CCD camera. Furthermore, we have succeeded to reduce inspection time by use of multiple algorithms such as DRC (Design Rule Check) method, CAD data comparison method and so on.
Concretely, in case of 5 inch substrate, inspection time is 90 sec with 2 μm resolution capability, 5 min with 1 μm resolution capability, 30 min with 0.5 μm resolution capability. This value is less than 1/5 compared to the traditional system. On the other hand, minimum resolution capability is 0.1 μm. For this reason, it's possible to detect ultra-fine defect such as sub-micron order. Furthermore, it equips a laser repair head, too. In this case, it upgrades a multiple function system, in short, it has four functions such as measurement, AOI, review, repair.
We suppose this system is useful for various photo masks and PCB with fine pitch. We have shipped the first model (ST-600) at autumn of 2010. Of course, we can brush up as AOI system the other model, such as ST-100, ST-600, ST-750, ST-800.
Also, we are focusing on enhancement of inspection resolution, and we are going to develop the new system, in order to enhance its resolution from 1 μm (line with) to 0.5 μm.
