FA leads to proper ESD tests

At our company, we used failure analysis (FA) to successfully determine what caused GaAs RF ICs to fail during retesting. In our case, the source of the damage turned out to be just as important as the damage itself. Our investigations show the importance of using the Charged Device Model (CDM) for ESD testing of small ICs in plastic packages. In fact, CDM testing proved invaluable in establishing the reliability of a GaAs radio frequency (RF) IC.

The problems we investigated originated when a sample of 200 good ICs was retested and nine of the devices failed the retest. The test results placed the reliability of all the ICs in doubt.

Figure 1. The GaAs RF IC we investigated could fail because of opens at R1 or shorts at C1 or C1.

Figure 2. A curve-tracer's display shows the current-vs.-voltage plot from tests on the RF input pin of the circuit shown in Figure 1. The positive and negative resistance changes indicate problems with internal components.

Previous ESD characterization of the IC had shown that resistor R₁ would likely fail at discharge levels below those that would damage either of the capacitors. Thus, it wasn't immediately obvious that an ESD event could damage a capacitor before it would damage a resistor in the circuit.

After we "decapped" the eight ICs that indicated a short across C₁, we used a liquid-crystal material to help locate the defects. (The liquid-crystal material changes its optical characteristics in response to temperature changes.) Because resistor R₁ provided a low-resistance path to ground, the heat it produced made fault isolation difficult. We decided to destructively "remove" R₁ from the circuit by applying enough voltage—about 2.5 V in this case—to the RF input to "open" R₁.

After removing R₁, we used the liquid-crystal material to locate any hot spots. Our analysis revealed a hot spot in C₁ for all eight ICs, thus each failed capacitor contained a short. On one device we could observe a spot on capacitor C₁ at the same location as the short we identified earlier (Figure 3). This spot looked similar to those seen on other capacitors known to have suffered ESD damage.

To determine whether the spot was a particle, a hole, or a bump, we had to "deprocess" the IC to get down to the top plate of the capacitor. We knew from experience that a plasma etch of the dielectric would leave unwanted residue, a reactive-ion etch (RIE) "grass" on the surface. So we developed a wet-etch method that left only a thin, clean layer of silicon nitride on top of the capacitor.

Because successive dielectric layers of silicon nitride and a low-k dielectric covered the capacitor, we could use the chemical characteristics of the low-k dielectric, along with a laser, to remove the  nitride layers.

Figure 3. A visible-light photo of a failed device shows a small spot revealed by liquid-crystal material. This hot spot indicates a short circuit.

Figure 4. An image from a scanning electron microscope (SEM) shows the trench etched around a defective capacitor by a laser. (The arrow points to the defects' location.)

Figure 5. (a) The bump on a defective GaAs RF IC, as seen in this SEM image, indicated the location of a short circuit. Compare the bump in (a) to the known ESD defect—also a short circuit—shown in (b).

Figure 6. The SEM image of a device tested using the CDM shows a bump similar to those in Figure 5 that indicate short circuits in capacitors.

Because of the similarities, it seemed likely the damage to the nine ICs that failed during retest was due to ESD damage. To test this hypothesis, we devised an experiment to try to reproduce the failure mechanism. Previous ESD testing on the RF IC using the Human Body Model (HBM) and Machine Model (MM) always damaged resistor R₁ rather than either capacitor C₁ or C₂ . So, we included the CDM along with the HBM and the MM in our series of ESD tests. (See "What's an ESD model?").

For our experiment, we chose 45 known-good RF ICs—15 for each ESD model. The test result showed that HBM and MM consistently produced an open or a high resistance on the RF input. On the other hand, CDM yielded a positive or negative bend in the curve trace—the same type of response observed for the original nine test failures we investigated.

An internal visual inspection of the 45 tested ICs verified that testing with the HBM and the MM had overstressed the resistor, thus causing the high resistance. We deprocessed the 15 failed ICs produced during CDM testing and found the same shorted capacitors in each device (Figure 6) as we saw in the original nine retest failures.

Recall, though, that an overstressed resistor, the signature of an HBM or MM event, caused none of the failures seen on the test floor. Problems with the dielectric—our early suspicion—didn't cause problems, either. The nine original devices failed due to a shorted capacitor, the same failure mode found with CDM ESD testing. Thus, for us, the CDM ESD model represents the type of ESD event most likely to occur for a GaAs RF IC, or similar device, on the test floor. Our results underline the importance of testing and characterizing the CDM ESD performance of a device, as well as testing its HBM and MM characteristics. And they show the importance of not only finding a failure but also determining its cause.