Here are the highlights from Session Three – Improving Cost of Ownership of the 20th annual IEEE Semiconductor Wafer Test Workshop (SWTW)
Rey Rincon, Freescale & Jeff Greenberg, Rudolph Technologies, “Optimizing Test Cell Performance Using Probing Process Analysis and Predictive Scrub”:
Rey summarized efforts at Freescale to improve test cell performance with multi-tier cantilever probe cards by investigating prober performance, probe card performance and probe card analyzer correlation to the test cell.
The first pass was to take a probe card that was freshly measured/adjusted on a PrecisionPoint VX3 probe card analyzer and then probe wafers which were then in turn measured with the WaferWoRx system. WaferWoRx examines the scrub marks on the wafer and analyzes the results in terms of errors due to the prober, probe card, and probing process. Overall prober error was determined to be about 12 µm with 5 µm being due to probe to pad alignment however adjusting this was deferred due to low return on investment (ROI).
In terms of the probe card, the system found no XY scaling error and the probe positions are repeatable. However, the probes were scrubbing short of pad center which resulted in a very large 15 µm error. And more importantly was that the multiple tiers with different tip lengths each had different scrub characteristics. Even though the tip widths correlate between the actual scrub marks and the analyzer, the scrub length measured is markedly different (~10 µm delta). This difference is due to the analyzer using a sapphire window on which the probe scrubs while being measured. The sapphire has a lower coefficient of friction than an aluminum pad; hence the scrub motion is different. This is typical of the class of problems when the test or metrology system is not 100% identical to the actual use case.
Jeff then spoke about a new feature in WaferWoRx to generate a scrub correction factor (SCF) to use with the probe card analyzers. The new PrecisionPoint VX4 probe card analyzer uses this correction factor to provide “predictive scrub”. This allows the VX4 to predict where the scrub will be so the probe card is adjusted closer to the targeted location. In the Freescale test case by using the SCF they were able to reduce the error in scrub length and position from 10 µm to 4 µm using SCF. Freescale used the data to produce different SCFs for each tier of the probe card to further tighten the performance of the system.
Thomas Logue, Seagate, “Reducing the Cost of Test on Gold & Copper Pads”:
Historically they used beryllium copper (BeCu) probes since they provide low contact resistance (Cres) for both gold (Au) and copper (Cu) pads used on disk drive magnetic heads. However, as the size of the pad pitch shrunk to 45 µm they experienced significantly reduced probe card lifetime and probe card maintenance increased to the point where a clean was required after each wafer tested. Broken tips accounted for about 50% of their failures and the probes typically had a lifetime of 1.5 M touchdowns.
By switching to tungsten rhenium (W-Re) probes and by using a more effective cleaning media such as International Test Solutions’ ProbePolish and ProbeClean, they were able to increase probe life to an average of 3 M touchdowns and reduced cleaning to once for every four wafers.
Rocky JM Lee, IMT (Korea), “Probe Card Cleaning by Laser”:
Since laser wavelengths are a single frequency of light it is possible to select a laser which will selectively remove only debris from a metal contact surfaces. IMT originally developed the technology for cleaning device test sockets. They recently presented the data at the most recent BiTS Workshop. Now they are applying the technology to probe cards where it takes less than 10 minutes to clean a 300 mm full wafer (1 TD NAND FLASH memory) card.
With MEMS cards, they can clean not just tip but the base area below the tip as well. This is typically not possible with “contact” cleaning media. They have built both mobile (manual) and fully automated solutions in addition to a “hybrid” system where the laser system (on a cart) is attached to an automated stage to get the best of both solutions.
Terence Q. Collier, CVInc, “Impact of Bond Pad Corrosion”:
In order to improve both the probe process and the wire bonding process they recommend cleaning the wafers prior to probe to remove the fluorine corrosion on the bond pads. Terence contends that this fluorine corrosion causes aluminum oxide which is normally 50 to 80 Angstroms thick to grow to 500 to 800 Angstroms. And this thicker aluminum oxide is the major contributor to poor contact resistance (Cres). He recommends using their BPS100 stripper to remove the fluorine corrosion. He shared wire bond test results (Au wires on Al pads) demonstrating the improvements in bonding and no reliability test failures due to cleaning devices prior to probe.
Note: I will post the link for the slides once they become available.