FAQ 8: How do element studies relate to integrated circuits?"


Previous Previous
Next Next

TriQuint results have concentrated on the elements because those elementary structures help simplify discussions of failure mechanisms, failure distributions, and acceleration factors. In fact, IC life tests typically only provide data for the predominant failure mechanism, and the three key questions can't be answered very well by looking only at ICs. For example, accelerated current density experiments are not possible on digital ICs since voltages are required which typically exceed breakdown thresholds before current densities can be significantly accelerated on internal nodes. The life test results on a particular material are very similar in spite of the differences between elements and ICs. In fact, element data has proven to be such an accurate predictor of ICs, that we can estimate the life test results of a circuit once the thermal analysis has been completed. This is possible because the thermal analysis provides the location of the hotspot. Once the hottest element is located, all the other circuitry can be basically ignored, and the knowledge of element failure mechanisms, distributions, and acceleration factors can be used to predict the median life of the IC.

Examples using both a microwave and digital IC will be used to illustrate this example.

For the MMIC example, a standard TQ9111 will be used. The TQ9111 is a distributed 2-8 GHz amplifier. The thermal analysis of this device indicates that the hottest element is a large drain line terminating resistor made of NiCr. This element runs about 50°C hotter than the distributed MesFETs and it carries the equivalent of 0.9 milli-Amp per micron of width when biased at the maximum rated conditions. It is desirable to life test the amplifier with the FETs running at 240°C, so the NiCr would be at 290°C. The TriQuint data base includes NiCr element tests at 290°C and 1 milli-Amp per micron bias. Although the current densities are slightly different, the same result for these examples would be expected. In fact, the lower current density in the amplifier could be accounted for, since the J exponent is known to be 3 for NiCr. After running the life test on the amplifier, the results of the two life tests can be compared by examining the failure accumulation in time at the accelerated test temperature. The results are very similar. If either result were corrected for the differences in current density, the results would be identical. This result shows two things: the hottest spot on the IC caused the degradation, and that the IC lifetimes can be predicted based upon the elemental results.

For the digital example, we have selected an ASIC 12:1 multiplexer circuit. The IC thermal analysis indicates a much more uniform temperature distribution than for the MMIC amplifier, but there is still a "warm" area of the device which is composed of clock driver MesFETs. As with the MMIC, the result for this IC could be predicted based upon our knowledge of MesFET elements. To evaluate this capability, the ASIC was tested at 275°C, so the failure accumulation could be directly matched with a MesFET result.

Although the results don't match as well as the NiCr example, there is still reasonably good agreement between the two life tests. All the data matches our original introduction into accelerated testing. In fact, this data base is becoming so well-defined, that reliability problems can be spotted quite easily with just a single life test.