Ball Bond Reliability: Simulation of Pull and Shear Test
1.
This is a student paper, and some names have changed from the original abstract. I’m Electrical Engineering student and the topic is the simulation of pull and shear tests for ball bonds by finite element methods.
2.
Here we see a Au ball bond, and models that we have created in AYSYS at the university to simulate it. The bond ball shown is fully bonded to a pad Al film, which is bonded to other films in the bond pad stack of a 4-level metal IC. We use 3D modeling with about 250-thousand nodes. The lower picture shows a simulated shear tool of W material. Pull test is simulated by applying a force to the top, and shear test is by applying a horizontal force to the W.
Purposes for this project are listed here. We are using 3D FEM to simulate the stresses experienced by the ball bond during pull and shear testing. We want to examine the stress locations and magnitudes, and see what a bond might experience as we change the wire material, pull angle, bond pad metallization, pad Al thickness, and circuitry under pad.
3.
The Bond Pull Strength test typically follows a Mil standard, 883G method 2011. 25-micron Au wire is considered reliable if the bond does not pull off below 3 gram-Force, or about 30milli-Newtons. The pull angle for the bond on the IC can actually vary in practice.
4.
The Bond Shear test follows a JEDEC standard in JESD22, method B116A . In this test, the tool pushes on the bond from the side until it breaks off the IC. Some illustrations of shear test failure modes are shown. The Au ball may shear off without the bond breaking, the IC may “crater” in the process, or the pad Al may shear off or peel off from the barrier film below.
If Cu wire is used instead of Au, then failure mode 2, shown here, will not occur because of the increased stiffness of the Cu. And if the pad Al is thin, and well-bonded with Cu, the most common shear failure becomes type 3, “cratering”. But then it is unclear whether the pad structure had been cracked during bonding, or whether it broke simply due to the shear test. JEDEC has an active committee to develop a method for Cu wire shear testing. Perhaps our modeling data can assist in this.
5.
Many assumptions and simplifications are made in our finite element modeling. First, we must ignore some IC films details because we don’t have sufficient nodes available to model them all at this point. Materials properties such as the elastic moduli are included for each film, and the bond ball. Al and SiO2 have the same elastic modulus at room temperature, and Au is similar, while Cu is almost twice as stiff. Thinking of “cratering”, it’s the SiO2 that cracks, at a fairly low tensile stress of 76 Mega-Pascals, its “fracture toughness” limit.
When we analyze stresses, we’ll use “Equivalent Stress” for the metals, and when we look specifically at SiO2, we’ll use Maximum Principle Stress. We were not able to accomplish our plans for dynamic modeling, so this presentation will only discuss static 3D modeling results. Pull and shear forces used in our simulations are shown here.
6.
Here we see a slice through the center of the bonded ball model, with colors representing stress resulting from a pull force of 30mN upwards. This Au ball is bonded onto a “traditional” IC bond pad with Al metallization and SiO2 dielectric in alternating layers. The red color indicates high stress, at 66MPa equivalent tensile stress, while the dark blue is near 0. This level of stress is not expected to crack or plastically deform any features in this model.
The stress levels we are interested in occur here, at the bond interface of Au to pad Al, inside the pad Al, and inside the top SiO2 film. I’ll show a number of pictures in this same format for qualitative comparisons only. We have not yet refined the simulations to numerically compare the stresses in critical regions.
7.
Now we’ve used the same model, but have applied the pull force at 3 different angles, 60 degrees, 45, and 30 degrees from the bond pad plane. Stress scales are all set to 200MPa for red and 0 for dark blue. Note how the tensile stress shifts higher to the left in the ball as the pull angle decreases to the right. We see some compressive stresses building on the right side as well. We see that in the 30 degree pull, the stress in the pad Al and top SiO2 are highest, on the left side, opposite the pull direction. We’re seeing about 130MPa of tensile stress in the pad Al right at the left ball edge, which will only cause some elastic deformation if the bond is strong. But if the bond is weak, we expect delamination to start right at the ball edge where tension is highest.
8.
Here we’re viewing the 60 degree pull. In the figure below, we ghost-out the bond ball and Al films, then show the maximum principal stress in the SiO2 layers only. We see here that the highest tensile stress, in red, is 66MPa, not far below the 76MPa crack initiation point. Stronger pull on a well bonded ball could easily exceed the SiO2 fracture toughness. This suggests that if we were to see a crater initiating right beneath the ball edge opposite the pull direction, it could possibly be due to a new crack from the test itself.
9.
Now we simulate the same pull tests, but with a Cu ball instead of Au. Stress scales are set to 200MPa as before. The overall stress situation looks about the same here, with the most tensile stress in the films from the 30 degree pull. Stress in the pad Al right beneath the left ball edge is actually a bit lower than we saw before. We think this is indicating that the Cu spreads the stress out more evenly across the bond interface, giving a little more uniform pull then for the more malleable Au.
Of course, we have not modeled the even stiffer intermetallic compound, the IMC, for either Au or Cu, so it may be expected that with IMC in place the stress during pull test will be more even than these simulations indicate.
10.
Here we show the two previous 60 degree simulations, Au ball and Cu ball, with their corresponding SiO2 film stress revealed below. For the Au ball, the highest tension in the top SiO2 was 66 MPa, while for the Cu ball we see only 61MPa, about a 65 decrease. Again, this is evidence that the stiffer ball material spreads the stress out a little more.
11.
This time we are using the Au bond ball, but have changed the IC bond pad to one with Cu metal sheets and SiO2 dielectric layers between. The pad Al is still in place above the top Cu layer. This pad is very simplified as compared with actual slotted Cu and lowK dielectrics that you would find in actual ICs. Again, the stress in the Au ball changes with pull angle as we have seen. Stresses in the pad Al are about the same, but stresses in the sub-layers are decreased from what we saw in the Al metallization bond pad. We think this is also evidence that the stiffer Cu tends to spread out the stress a little more than SiO2. Assuming the same film adhesion everywhere, this type of pad may have less tendency to crack and crater during pull testing.
12.
Here we compare the Au ball bonds on the Al metallization pad on the left with the Cu metallization pad on the right. Both are pulled at 30 degrees in this case, and stress scales have red at 200MPa. There is slightly less tensile stress in the Cu pad structure. It has an additional layer, due to the 4 levels of metal plus pad Al on the top.
13.
This time, we have bonded the Cu metallization bond pad with a Cu bond ball. We are seeing the same effect as before, that the stress is slightly less beneath the Cu ball, and with the Cu metallization in the pad, the stress is slightly reduced again. We are pleased that the simulation results seem to be consistent.
14.
Now we will look at 45 degree pulls only, but throw in some variations of the Al metallization bond pad structure. Top left has a reduced pad Al thickness. We don’t see much difference here. Top right is thicker pad Al. Here we see a reduction in tensile stress of the films below, indication that Al mitigates some of the tension by deforming elastically. Bottom left is back to normal pad Al thickness, but we have made the metal layers below as if they are dense circuitry, with overall 50% Al pattern density in all 3 metals below the pad. This did not seem to make an appreciable difference in stresses of these sub-layers in the pull test. Lower right has the Metal 3 layer replaced by SiO2, and we don’t see much difference.
15.
Summarizing the pull test results,
. we clearly see the location of highest tensile stress,
. Cu ball and Cu metallization both seem to spread out the stress, reducing the tension “hot spot” seen with Au and with Al metallization.
. and we saw that thicker pad Al reduced stresses reaching the pad sub-layers
16.
Now we move to shear testing, using the same models and picture format. We are pushing on the little block to the right with a left-wards force, typically enough to shear a 64 micron well-bonded Au ball. On the left are Au ball bonds, top on Al pad, bottom on Cu pad structure. On the right are Cu ball bonds, for the same pad structures. The purple color exceeds the 200MPa limit of the scale, so we’re not seeing any detail in the highest stress region. Remember that 200MPa is the stress required to begin plastic deformation of Al at room temperature, so the red and purple regions indicate plastic deformation of pad Al along the bond interface. The first thing we should notice is the very high tensile stress in the pad sub-layers on the right side. There is plenty of stress to initiate cracking in the top SiO2 layer on the right side, so we should be careful in interpreting failure modes that indicate cracking or cratering beneath the ball near the shear contact point.
Comparing left to right, Au bond ball vs Cu bond ball, we see the compressive stress extending farther in the Cu ball and in the pad Al beneath it, a result of Cu being a stiffer material than Au. But we also see less overall stress in the sub-layers on the left side, indicating that the Cu is likely concentrating more shear stress in the pad Al than the Au ball. Comparing up and down, Al vs Cu metallization bond pads, the Cu metallization clearly has reduced stress in the sub-layers nearest the shear tool. This seems to indicate that Cu helps restrain the shear and compressive stress to the upper layers in the pad.
17.
Here we compare shear stresses for Au ball on Al metallization pads. Upper left has thin pad Al, and upper right has thick pad Al. We see that the thin pad Al results in higher stress in pad sub-layers on the left side, while thick pad Al has a dramatic effect in reducing stress in the pad sub-layers. The lower results are for circuitry under pad and replacing the Metal 3 by SiO2. Both of these pads appear a little more like the Cu metallization pad results, possibly indicating that stress is being spread out more evenly due to more SiO2 in the structure.
18.
Summarizing the shear test simulations, we see
. Large tensile stress in the pad sub-layers near the shear tool contact
. The Cu ball exerts more stress to pad Al along the contact area than Au
. Cu metallization in the pad reduces stress in the sub-layers
. Thin pad Al causes more stress in sub-layers, while thick pad Al reduces stress in the sub-layers
. Circuit under pad structures seem to reduce the stress in sub-layers
19.
Overall conclusions are listed here. For pull testing, we see uneven tensile stress applied to the bond pad films opposite the pull force direction. There could be sufficient stress from pull testing a well-bonded ball to initiate cracks in the top SiO2 film, resulting in a cratering failure mode. Cu bond ball and metallization tends to spread the stress a little more evenly. And thicker pad Al and Cu metallization both help to reduce tensile stress in the pad sub-layers during pull test.
For shear testing, we see again that the high tensile stress could crack the top SiO2 beneath pad Al. The Cu ball causes more stress along the contact area in the pad Al film than the Au ball. Thicker pad Al and Cu metallization and circuit under pad structures all reduce the stress in pad sub-layers.
20.
We spent the majority of our time so far in creating and debugging the complex bond pad models. We now look forward to doing more detailed analyses, especially for the shear test, where we need to carefully analyze the various stress types and directions. We’ll need to include IMC layers and eventually other film layers and vias in the model as we become more skilled in the modeling. Then we can also look at lowK and e-lowK dielectrics in sub-layers. We will try to get more specific regarding where and when a delamination may initiate, and of course move on to dynamic simulations.
