IEEE Semiconductor Wafer Test Workshop 2012 – Session 2 (Monday)

Here are the highlights from Session Two “Optimizing Probe Depth Performance” of the 22nd annual IEEE Semiconductor Wafer Test Workshop (SWTW) from Monday June 11, 2012.

Tommie Berry (FormFactor, Inc.), “Actual vs. Programmed Over Travel for Advanced Probe Cards”:

As the number of probes on a probe card increase, the total force required to compress these probes – know as probe force – is increasing. With high force the actual over travel (AOT) – also know as overdrive – of the probe is often significantly different than the programmed over travel (POT) programmed in the prober. Even though memory test engineers with very high probe count cards have been concerned about the difference between AOT and POT for several years, it is becoming increasingly important for other types of probe cards. Other cards, especially system on a chip (SOC) applications, have rapidly increasing probe counts driven by greater probes per device combined with larger number of devices to be tested in parallel.

Mr. Berry reviewed calculations and measurements of AOT vs. POT based upon system spring constant (stiffness) of the test cell and known probe spring rates. The force applied on the wafer prober chuck is always a function of the number of springs and AOT. However, when the ratio of AOT to POT is low, there can be large variations in the over travel on each probe. This can often be seen in the variation in the scrub length on pads across the wafer.

Typically this issue isn’t a large concern for cards with small pin counts, less than one thousand probes. But on higher pin counts cards, two thousand or more, it is critical to understand the AOT as a function of the system stiffness and probe card architecture. By characterizing both, the user can calculate AOT based upon POT and set the POT appropriately.

Questions:

• Have they done a study of hysteresis in the system and their measurements? The load cell used for their measurements has some, but it is very low in comparison to other factors. Keith Breinlinger (a paper co-author) said there were no signs of hysteresis in any of their measurements.
• Have they written a procedure to help teach the “clay puck” technique they used to confirm AOT? There is an older SWTW paper addressing this and they have a written specific procedure.
• The model discussed didn’t include the dampening ratio of the probes as classical springs. How would the dampening ratio effect the system before and after cleaning the probes? Mr. Breinlinger doesn’t believe there is significant friction in the system of probe force from contact to the wafer. They didn’t see any signs of dampening which is more likely to be a factor in lateral force measurements. In the scenario they tested, the wafer chuck is “infinitely” powerful compared to the force of an individual probe (spring).
• Does the test temperature have an impact? Yes, temperature makes a significant difference in AOT versus POT. For this study, they did their measurements at room temperature.
• What is the impact of the tester docking method? In terms of docking method, they treated the tester as a rigid body. But if the tester and docking hardware is not rigid, it should be included in the overall system spring rate (k value).

 

Bernd Bischoff (Texas Instruments – Germany), “Aluminum Probe Pad Thickness and Properties for Stable Parametric ‘Probe‐Ability'”:

Texas Instruments (TI) had experienced poor performance using FormFactor Takumi probe cards for in-process parametric measurements. Problems included large amounts of aluminum debris left on the wafers and frequent cleaning required during use. These problems were exacerbated when the thickness of the aluminum probe pads varied.

Working with International Test Solutions (ITS), the ITS bench top characterization system was used to measure performance of the Takumi probe technology on blank metalized aluminum wafers. The wafers had differing thickness of aluminum (0.6, 1.5, and 3 µm) and they were tested at varying overdrive values (30 and 60 µm).

Instead of using the typical scatter plots of contact resistance (Cres) versus touchdown, Mr. Bischoff used cumulative frequency distribution (CFD) charts to allow easier comparison of large data sets. However, CFD charts do not have an indicator of time sequence.

Through experimentation they were able to optimize the cleaning recipes based upon the metal thickness. By improving the cleaning process, which included more cleaning touchdowns, the Cres was decreased at the lower overdrive value. And the same time it was determined that interval between cleanings could be increased 2 to 9x depending upon the metal thickness. Life expectancy of a probe card is now 2.5 to 3.5x with a similar cost reduction.

Questions:

• Is the thicker layer of aluminum softer? And did they account for what is below the bond pad? The test wafers had the same metal stack as those used in production. Only for the .6 µm thickness wafers is this an issue since it is thinner and sometimes the layers below might come through. However, they didn’t look at active structures below the pads. The grain sizes were different for the different thickness aluminum which is one of the reasons they saw different amounts of debris. And yes, a thicker layer of aluminum is softer.
• What is the influence of temperature on the problem? Parametric tests are done between 25 and 30 C at TI. So, there is no issue from running these experiments at room temperature.
• Is it pure aluminum or alloy? And is it applied via sputtering or electroplating? It is aluminum alloy with a low percentage of copper. Mr. Bischoff doesn’t know how the aluminum was applied to the test wafers.

 

Stevan Hunter (ON Semiconductor), “Comparison of Bond Pad Cracking in Harsh Probing with Three Different Probe Cards”:

Mr. Hunter’s presentation is a follow-on from last year’s presentation where they use both low force cantilever probes with small tips and high force cantilevers to study bond pad cracking. At ON Semiconductor it is easy to accumulate six test touchdowns in production on a device between low temperature test, high temperature test, non-volatile memory test, and retest cycles.

The high force card caused the most cracking but the small tip card had similar cracking results since it applies similar stress to the pad. As the metal layer below the pad metal decreases in density, the rate of cracking – regardless of probe technology – increases once a critical threshold is reached. In addition, there is a sensitivity to the pattern of the metal in the layer below the pad as a function of scrub direction (direction of stress) that may also lead to cracking.

They have determined that in general, the best choice is the low force card since the high force caused the most cracking. However, the small tips caused frequent cracking due to high stress of the small tips. So to reduce cracking, one should use larger tip diameter with lower spring rates and lower prober overdrive.

Questions:

• What type of needles were used? Tungsten rhenium probes.
• For probe tip shape have they considered using rounded tips? Mr. Hunter had thought about it but the focus of the study was to examine crack formation. So their objective is to produce the maximum number of cracks to better study them.