Testing LCD Matrixes with E-Beams

Ralf Schmid, Philipp Herbst and Matthias Brunner Applied Materials-AKT Feldkirchen, Germany -- Semiconductor International, 6/1/2001

Electron-beam technologies, which have proved their applicability for testing electrical function on very small nodes inside of integrated circuits,² offer good flexibility for high-density displays. Electron beams can be used in two ways: 1) passive only, for sensing pixel voltages in response to external driving; or 2) active driving through the beam, eliminating the requirement for external driving, plus sensing pixel voltages.³ The second method offers several advantages as described in this paper. The technology was evaluated by a major flat-panel display manufacturer in its Japanese fab to assess its capabilities for current and future LCD manufacturing requirements.

1. An electron beam replaces a mechanical probe.

Ideally, one would like to contact each pixel with three contacts and measure its proper function, particularly its on-resistance and leakage current. However, only two contacts can be made through the signal and gate lines. It is obvious that the third contact cannot be made by a physical probe on the pixel electrode for an in-process test, but the electron beam provides an ideal contactless solution (Fig. 1). The electrons of the beam supply a current in the same way as a physical probe, and the beam also allows voltage measurements. Positioning and contacting by the beam is much easier and faster than by a physical probe. Electromagnetic deflection is used to vector the beam from one pixel to the next, and beam blanking is used to turn the beam on or off. Positioning the beam and testing are done in 10 µsec for each sub-pixel.

One measurement that can be made by this method is the detection of the voltage drop V_pixel at the transistor caused by the injected beam current (e.g. 0.2 µA) as shown in Figure 2. On-resistance of 10 MW is easily detected by the resulting 2 V at the pixel. Another measurement is the storage of charge in the pixel, which decays to 1/e within the RC time constant t = 2 sec if the isolation resistance is below 10¹³ W. A holding cycle for 2 sec can be introduced without significant increase of test time by using the vector addressing scheme.

2. Leakage and on-resistance detection sensitivity reveal a 10-20 mV grayscale error.

Vector mode addressing, retest

	3. Detected signals outside of threshold limits of pixels 1, 2, 3 and 4 are listed in the first test. Vector addressing allows very fast readdressing of only these pixels in retest a. Additional sequences can be added (...f). All values m, a, ..., f are evaluated per defective pixel for defect confirmation and characterization.

Because the e-beam addresses pixels in vector mode, fast retest of possible defects can be made to significantly increase test accuracy. A first test at each pixel using a first test sequence and set of evaluation parameters results in an initial list of possible defective pixels. These pixels are subsequently tested again in one or more additional test sequences with different parameters by only addressing these defective pixels (Fig. 3).

Since addressing and testing is fast (10 µsec per sub-pixel), and the number of pixels retested is restricted to those found in the first test (e.g. <1000), each of these retests is very fast (10 msec in this example). The final evaluation of the defects uses the signals from all tests done at each pixel and makes a defect characterization based on the detected signal pattern. This retest strategy makes test results very reliable and has particular advantages in eliminating particle counts. Like other test technologies, the e-beam generates a different signal when hitting a particle, but there is no danger of damage. The particle, on the other hand, generates different detector signals in the first test and the retests when appropriate retest parameters are chosen.

Defect detection reliability

The e-beam system was installed at a customer facility to evaluate the tester and its performance with respect to manufacturing requirements. The customer's long-term experience with in-process testing and availability of other test, inspection and review equipment allowed a detailed comparison of test results. Tests were made with a charge sensing test system⁴ and with the e-beam tester to compare both test results. Pixels for which different results were obtained were reviewed under a microscope, and some of the plates were assembled to verify defects in a cell test.

Defect detection reliability consists of two parts. The first part is the percentage of real defects in the display that are detected. Ideally, 100% of the defects are detected. If there are defects that are not detected, the respective missed defect count is termed "underkill." The other part is the opposite — defect messages of the tester that are not true defects. This number of wrong defect messages, called "overkill," ideally should be zero.

The threshold to distinguish between good and bad pixels is not easy to define because pixels are operated with analog signals generating grayscale levels. The question, therefore, is which deviation of characteristics such as leakage, on-resistance, etc., can be tolerated before the pixel is classified as bad.

The correct definition of threshold between good and bad pixels is determined by the visibility of a defect in the cell after assembly. However, even the definition of "visible" is not clearly defined and may depend on the display customers and quality requirements. Not all of the displays could be assembled in these investigations, so other criteria needed to be used for those tests. The customer has extensive experience in testing pixels, reviewing defects under a microscope and reviewing again after assembly; therefore, this experience could be used to judge if defects are true or not without cell assembly.

Engineering lots were passed through both testers for several months, and the results were compared. The conclusion of these investigations is that the overkill and underkill of both testers is similar. A typical example on which this conclusion is based is shown in Figure 4.

	4. Similarity of charge sensing and Puma e-beam test results are shown.

In the sample lot of plates used for these tests, the charge sensing tester detected 60 defects and the e-beam tester detected 61 defects. Fifty-five defects were identical pixels in both testers. Five of the defects found with the charge sensing tester were not detected by the e-beam tester, but optical review showed pixel damage. It is therefore concluded that these are underkill for the e-beam tester. Similarly, six defects found by the e-beam tester were not found by the charge sensing tester, but also showed clear damage in the review, so they are counted as charge sensing underkill.

In this example, the charge sensing tester did detect the 60 defects after increasing its sensitivity slightly. With its initial setting it only found eight fewer defects. That shows that the agreement of results is a matter of setting the threshold between good and bad as explained above. The e-beam tester generated eight more defect messages that turned out to be generated by particles coming from the environment in which the tester was operated for these investigations. It is straightforward to further reduce the particle count by reducing the particle contamination in the cassette station area.

Nevertheless, there is a fundamental difference between the e-beam tester and charge sensing tester. The e-beam tester has some sensitivity to particles that can be reduced but not totally eliminated by the retest method mentioned above. The charge sensing tester, however, has no sensitivity to particles. It should be noted that the e-beam tester may only send an additional wrong defect message — in some cases caused by a particle — but no damage happens to the tester or the display. In the case of the sample test discussed above, the charge sensing tester had no overkill; however, charge sensing tester overkill was observed during tests of other lots during these evaluation tests.

Defect characterization

Defect characterization has several purposes. The first and most important one is to judge if a particular defect can be repaired, or if it is a fatal defect (e.g. a line defect) that requires the matrix to be discarded. Of course, a defect can also be ignored. Another reason for defect characterization is for process monitoring and control.

The defect classification scheme known from the charge sensing tester was used for comparison of both testers. No significant number of fatal defects was present in the displays, but in a few instances those defects were classified as such by the charge sensing tester and by Puma. Some of the other defects were classified differently in both testers. Those pixels with deviating results were further investigated by review under a microscope.

The classifications were compared with the typical optical patterns known for each type of defect. The Table shows the results of one lot of plates as an example. Each defect whose characterization of the respective tester is identical with the optical matrix inspection, and therefore was obviously characterized correctly, is printed in bold.

Defect Classification Reviewed by Optical Inspection
Charge sensing	Etec	Optical matrix inspection	Optical cell inspection
CS LEAK	OTHERS	cs	DAR
EXTR CHAR	CS LEAK
EXTR CHAR	OTH LEAK	slf+oth
EXTR CHAR	SLF LEAK	slf+oth
GT LEAK	CS LEAK	cs
GT LEAK	OTHERS	cs
I OFF	CS LEAK	particle (I OFF)
I OFF	SLF LEAK	loff	BRT
I OFF	SLF LEAK	slf	BRT
LOW VG	CS LEAK
OTH LEAK	OTHERS
OTH LEAK	SLF LEAK	oth
OTHERS	CS LEAK		BRT
OTHERS	OTH LEAK		BRT-S-O
OTHERS	OTHERS	others
OTHERS	SLF LEAK	slf	BRT
PROB ERR	OTH LEAK	particle
SLF LEAK	CS LEAK	slf+oth
SLF LEAK	OTH LEAK	slf+oth
SLF LEAK	OTHERS

Some of the displays were assembled so that optical cell inspection could also be done for further proof. Both charge sensing and Puma testers have the same number of correct and false defect classifications.

In addition, it was found that both testers provided identical classification for ~50% of the classified defects (53% in this tested lot). Since both testers have the same classification accuracy, it is concluded that each tester provides 75% correct classification.

Investigating the impact of classification for the repair decision, it was found that ~70% of the decisions were made identically by both testers. Because both testers have about the same classification error, each tester is 85% accurate in its repair decision.

Conclusion

Test comparisons prove that the e-beam tester provides defect detection capability similar to the charge sensing tester, as well as similar defect characterization capability. However, it does not require complex probers but only simple probe frames. High prober cost and lead time are avoided. The e-beam test principle using the beam as a current source and probe, and using vector addressing to position the beam, has great potential for further improvement of test performance and for extension to larger displays with higher resolution.

Matthias Brunneris managing director of AKT Electron Beam Technology (EBT), a division of AKT Inc., where he is responsible for development and production of e-beam LCD test systems. After receiving his Ph.D. in physics from the Technical University of Berlin in 1981, Brunner worked for IBM Research, then Siemens AG, where he was manager of the e-beam technology group. He joined ICT GmbH in 1993 as manager of product development for LCD testing. ICT was eventually purchased by Etec Systems, which in turn was acquired by AKT Inc. in 2000.

Ralf Schmidis manager of the development and applications laboratory at AKT EBT. He received his diploma in physics, focusing on electron-beam metrology and secondary electron detection. After working in Siemens' corporate research department on e-beam CD measurement and e-beam short/open testing, he received his Ph.D. from the University of Tübingen (Tübingen, Germany).

Philipp Herbstis a system engineer for e-beam active-matrix LCD test systems at AKT EBT. He received his degree in physics from the Technical University of Munich, and a diploma from Walter Schottky Institute for his work in experimental semiconductor physics.

---

REFERENCES

1. M. Ikeda, "LCD Panel Process Inspection Technology: Build of Inspection Information System," FPD Intelligence, Jan. 2000, p. 86.
   
2. J. Frosien, et al, "Electron Beam Testing of Submicron Structures," J. Electrochem. Soc.: Solid-State Science and Technology, Vol. 135, No. 8, 1988, p. 2038; D. Winkler, R. Schmitt, M. Brunner, B. Lischke, "Flexible Picosecond Probing of Integrated Circuits with Chopped Electron Beams," IBM J. of Research and Development, Vol. 34, No. 2/3, 1990, p. 189.
   
3. M. Brunner, R. Schmitt, D. Winkler, R. Schmid, "Electron Beam AM LCD Testing," Proc. Of the 13th Int. Disp. Research Conf. Eurodisplay, 1993, p. 387; M. Brunner, "Applying Electron Beam Technology to AM LCD Array Testing," Display Manufacturing Technology Conference — Display Works - Digest of Technical Papers, 1999, p. 144.
   
4. S. Kimura, Y. Ichioka, K. Suzuki, R.J. Polastre, "High-Speed Testing of TFT Array," SID 92 Digest, 1992, p. 628.
   

Acknowledgement

The authors would like to acknowledge the cooperation and support of Masato Ikeda and Shuichi Nishira from DTI (Display Technologies Inc.), who offered the opportunity for this evaluation.

Talkback

We would love your feedback!

Post a comment

» VIEW ALL TALKBACK THREADS

Related Content

• Topics
• Author

Related Content

 

By This Author

There are no other articles written by this author.

SPONSORED LINKS