Analysis yields secrets of probe-assembly failures

Figure 1 (a) This photograph of a ceramic ring shows cracking of the ceramic on the left wing under the epoxy. (b) The ring is dyed with a violet color to help distinguish the regular machining groove marks from machining and handling scratch marks.

Figure 2 (a) This view shows the back of a probe card with the corner epoxy removed and the ceramic dyed to expose crack paths. Optical images of the ceramic ring indicate cracks in the (b) inner corner and (c) crack branching at the top corner sections. It appears that the ceramic bottom surface was chipped, and its origin probably lies at the opposite surface under the epoxy. (d) The arrows point to a cracking pattern.

Figure 3 Finite-element analysis yields this stress distribution on a ceramic ring at 10 mils overdrive. The Von Mises value is the square root of the sum of the squares of the x, y, and z stress components.

Epoxy probe cards are key tools for IC wafer test. The probe cards themselves can fail, and when they do, it’s important to determine why so you can minimize subsequent failures. Part 1 of this two-part series considered failures in the probes themselves (Ref. 1). In this article, we will cover the machinable glass-ceramics used in the probe-card ring-build process.

Ceramics are brittle in nature and can fail because of stress concentrations amplified by the design, thermal stresses generated by the epoxy curing process, or process-induced surface flaws. Some critical flaws at the corner of a ceramic ring may go undetected during manufacturing and may then grow to become large and visible cracks in the ring on the epoxy probe card at the test site. In addition, the surface condition of ceramic blades is critical for good adhesion of the metallization layers of attached probes (some nonuniform post-laser microstructures can lead to unpredictably poor bonding of sputtered metals). Probe solder-joint failures in an epoxy probe card often result from the contamination on the PCB surface or a lack of probe strain relief.

Failure analysis and methodology

Ceramic-ring failures generally happen during ceramic-ring production. The machined ceramic usually fails from surface flaws induced by machining and handling. You can minimize thermal stresses by modifying the process to include multiple cure cycles and slower cooling; you also can design the ring to eliminate stress concentrators such as sharp corners. Sharp corners can induce higher normal and shear stresses in epoxy-ceramic bonds due to mismatch in thermal expansion coefficients, and ceramic rings are likely to fail in such locations (Ref. 2).

The deep scratches on the ceramic surface lead to ceramic fracture as cracks propagate under thermal and handling stresses during manufacturing. Such cracking can be observed optically. Microcracks and surface scratches can be revealed by a dye penetrant (Figure 1).

Although rare, ceramic cracking in the corners of a square ring can happen during wafer test. Figure 2a illustrates a ring on a PCB with epoxy removed on the corners to expose the ceramic surface and crack path through use of a dye. Failure analysis, in this case, involves both fractography of fracture surfaces (to locate the origin of the crack) and estimation of deflections during wafer test (to see if such loading may lead to this cracking). Figures 2b and 2c show crack paths that start from underneath the epoxy and propagate outward and downward through the ceramic. The wafer-test loading and the flexural and handling stresses during manufacturing are possible mechanisms for this sort of cracking.

We performed a finite-element analysis (FEA) to determine if the ceramic ring fracture was caused by probe loading during the test. We assumed an evenly distributed array of 750 probes at the ring periphery. By assuming a per probe load of 1 gram-force per mil overdrive and assuming 10 mils overdrive (well beyond the specification), the simulations predicted a maximum stress of 1.4 N/mm² (on the corners) where the flexural strength of the ceramic is 100 N/mm² (a safety factor of 100). This suggests that the ceramic fracture probably is not caused by stresses from probe loading. And the stresses at the corners of the ring seemed to be higher than other sections (Figure 3). Thus, the failure is due to microcracks generated in ring-production operations and flexural stressing of the parts that lead to crack growth during manufacturing.

Analysis of a failed probe card may raise the likelihood of a design error in the layout. For instance, we once noticed the premature wear of a certain number of probes in a probe card during wafer test after thousands of touchdowns. The engineer thought the power probes might have had material defects that were making them wear out faster than the signal probes. Failure analysis in this case involved collecting the initial measurements of probe parameters (tip diameter, tip length, beam length, tip angle, and balanced contact force) and comparing them with measurements made after probe wear had rendered the probe card unusable. We determined from the analysis of the initial and final tip diameters that the probe wear was highest on the corner probes in the card.

Figure 4 indicates that trend for the device clearly in 3-D with dips in the beam length (in.) and higher probe wear (mils) for corner probes. Note that as the beam length increased, the change in tip diameter (Dd = d_FINAL – d_INITIAL), in mils, decreased—the shorter the beam length, the higher the amount of wear in the probes. Figure 5 presents a scatter plot that shows a relation between beam length and change in tip diameter. Individual data points are represented in a 2-D space where axes represent the variables (beam length on the horizontal axis, x, and change in tip diameter on the vertical axis, y). The two coordinates (x and y) that determine the location of each point correspond to specific values for the two variables. A significant negative correlation (–0.6530) was observed, indicating a strong negative relationship between beam length and change in tip diameter. The solid red line on the figure represents the line of best fit (Dd = 9.52 – 42.38 x beam length), which can be used to predict the probe wear given a beam length. This relationship is valid only for the specific geometry of tungsten probes used in the probe card detailed in Figure 4.

In this configuration, because the corner probes had shorter beam lengths, they initally experienced higher forces. These forces remained high until the tips wore down, causing their diameters to increase and making them unable to accurately probe the wafer targets. The probe card and the probe fan-out needed to be redesigned to fix this problem.                                             

The role of failure analysis

An organized FA approach (Figure 6) should be an integral part of the probe-card manufacturing and wafer-test processes. The alternative is a time-consuming and costly trial-and-error methodology. A definitive failure analysis should include these important elements:

• the availability and extent of initial probe card QA data and service records,

• accurate lab tests and measurements of relevant final probe card parameters, and

• the right combination of analytical techniques.

Figure 6 This cause-and-effect diagram lists the various probe card failure types and their possible causes. It can be used as an effective trouble-shooting aid for failure analysis.

The failure analyst should be familiar with the probe-card manufacturing process and with the wafer-test conditions under which the failures appeared. Information gathered from FA helps the probe-card manufacturer improve the card design and manufacturing processes, and it helps test engineers improve test procedures. T&MW

References

1. Tunaboylu, Bahadir, Chander Sekar, and Hank Scutoski, “Investigation Conquers Probe-Card Problems,” Test & Measurement World, November 2000. p. 19. 

2. Mulville, D.R., P.M. Mast, and R.N. Vaishnav, “Strain Energy Release Rate for Interfacial Cracks between Dissimilar Media,” Engineering Fracture Mechanics, Vol. 8, 1976. pp. 555–565.

Acknowledgement

The authors gratefully acknowledge Dr. S. Clegg for the finite-element analysis work. 

Bahadir Tunaboylu is a materials scientist at Cerprobe. He has a PhD in materials science from the University of California, San Diego, and a MS in ceramic engineering from Alfred University, NY.

Chander M. Sekar has been the corporate statistician at Cerprobe since 1996. He has a PhD in statistics from the University of Madras (India).

Henry P. Scutoski is the vice president of Quality & Process Management at Cerprobe. He has an MBA from the University of Dallas and a BS in Industrial Engineering from Bradley University.