Mal-detection rate and lost detection rate experimentation based on SBP and using an IC

19 -86.40 241.33 16 -84.48 241.24 16 1.92 P

20 -79.00 241.89 17 -79.84 252.90 17 11.05 P

21 -57.25 246.96 20 -55.58 246.84 20 1.67 P

22 -56.34 242.56 20 -54.91 242.96 20 1.48 P

23 -50.91 162.03 55 -38.68 172.84 55 16.32 P

24 -50.32 243.79 21 -41.63 251.26 22 11.46 F

25 -50.28 247.95 21 -48.19 247.29 21 2.19 P

26 -50.44 251.91 21 -47.51 243.50 21 8.90 P

……

5.4 Mal-detection rate and lost detection rate experimentation based on SBP and using an IC with wrong wire bonding

The engineers duplicate the SBP of experimentation 5.2 into other bonding machine and calibrate the bias as a WVM firstly. We can randomly select three wires (wire numbered 16, 20, and 23 in Figure 22(c) and Table 4) and adjust the wire bonding positions manually to let them shift away from the original wire bonding positions in the WVM, but still retain on the same lead (i.e., to ensure correct bonding). Similarly, we can randomly select the other wire (wire numbered 24 in Figure 22(c) and Table 4) and adjust its wire bonding position manually to allow it to shift away from the original bonding lead to cause an incorrect bond in the WVM. Initially, the shift tolerance range R of the lead side is set equal to the lead pitch (3.7 mils). The wire bonding position inspection steps were repeated as described in Section 4, by increasing the value of R by 2 mils each time when the PPC method is applied until no mal-detection can be found (see Table 5).

Figure 22. Experimental results from inspecting a sample multi-layered wire IC.

Table 4. Experimental results of apply PPC method to a multi-layered wire IC with one man-made wrong bonding wire. The lead length is 26 mils. It means that the maximum possible shift distance of the wire is 26 mils and the bonded wire will still be on the targeted lead. In the last two columns on the right-hand side, “P” indicates that the system check is passed and “F” denotes that the system check is failed. The lead pitch is 3.7 mils, R is set to be 3.7 mils first and there are three mal-detected wires. When R is set to 11.99 mils, the mal-detection rate will be reduced. However, releasing R will also cause the lost detection problem.

X Y X Y

……

15 283.61 46.89 284.77 46.20 1.35 Y P P

16 292.84 49.91 295.77 53.37 4.53 Y F(Mal-detection) P

17 300.48 49.48 301.76 48.95 1.38 Y P P

18 307.85 48.97 309.15 48.19 1.52 Y P P

19 315.29 48.17 316.54 48.07 1.25 Y P P

20 322.88 47.95 321.24 36.08 11.99 Y F(Mal-detection) P

21 344.22 42.71 345.5 42.14 1.4 Y P P

22 344.95 46.52 346.17 46.02 1.32 Y P P

23 351.97 117.4 362.4 116.14 10.51 Y F(Mal-detection) P

24 351.01 38.1 359.45 37.72 8.45 N F P(Lost detection)

25 351.71 42.19 352.89 41.69 1.29 Y P P

Table 5. Experimental results from a multi-layered wire IC inspection with one man-made incorrect bonding wire. The 2D image inspection method, the PPC method and the proposed WWBP system based on SBP are compared based on the mal-detection rate and lost detection rate by increasing the shift tolerance range 2 mils each time. The initial shift tolerance range is set equal to the lead pitch (3.7 mils).

Tolerance range

The experimental results are recorded in Table 4, Table 5 and Figure 23. A part of the image is shown in Figure 22. Figure 22(a) shows the full image of the chip. Figure 22(b) is the partially enlarged image of the IC with the correct bonding wire in the first setup machine. Figure 22(c) depicts the partially enlarged image of the IC with an incorrectly bonded wire in the WVM. Figures 22(d), 22(e), and 22(f) illustrate the results from applying the PPC method. The slim lines in Figures 22(e) and 22(f) denote that these wires were verified as correctly bonded wires using the PPC method. The fat lines in Figures 22(e) (wire numbered 16, 20, 23, and 24) and 22(f) (wire numbered 20 and 23), respectively, represent the fact that they are incorrectly bonded wires. These wires were verified using the PPC method. Figures 22(g) and 22(h) are the results from applying the WWBP system. In Figure 22(h), the slim lines express the fact that the wires were verified as being correctly bonded using the WWBP system. The fat line (wire numbered 24) represents the fact that it is an incorrectly bonded wire that was verified using the WWBP system.

Figure 23. The shift distance between WVM and SBP of each wire in lead side and pad side.

With regard to Table 4 and Figure 23(a), when the PPC method was applied, mal-detection occurred in the lead side. Comparing Figure 22(c) with Figure 22(e), we can see that even though the two wires numbered 20 and 23 shifted away from the designated wire bonding position, they were still bonded onto the correct lead. The PPC method verified it as being bonded incorrectly when R is 3.7 mils. The mal-detection problem could be reduced if R could be considered equal to the maximum possible D. However, releasing R might cause the lost detection problem. For example, when R was increased to 9.7 mils (Figure 22(f) and Table 5), wire number 16 is not mal-detected any more. However, the lost detection case occurred for wire number 24. Because the PPC method checks the correctness of wire bonding positions only on the basis of the shift distance, that is, since it does not have the actual information on the lead position, not all incorrect bonds can be detected. Some correct bonds may even be incorrectly detected as defective.

On the other hand, in Figures 22(g) and 22(h), as a result of applying the proposed WWBP system, no mal-detection or lost detection cases occurred. That is, despite the shade from the wires in the upper layer hiding the wires in the lower layer, as Figure 19(c) shows, the proposed WWBP system can successfully be applied for multi-layered wire IC inspection. When the WWBP system was applied, because the actual information on the lead positions was utilized, no mal-detection or lost detection occurred. All of the bonding positions for the wires can be correctly verified.

On the lead side inspection, both PPC method and WWBP system used the same algorithm.

Figure 23(b) showed the pad side comparison result and all wires were identified as correct bonding (D < 0.67 mil). No any mal-detection or lost detection in the pad side inspection was encountered.