A. Christou and P. Tang
University of Maryland
College Park, MD 20742
A definitive separation of failure distribution functions for MMICs has been obtained as a result of microwave accelerated life testing. The circuit which was a 2-8 GHz monolithic amplifier showed a very strong early failure pattern related only to passive component degradation. Using the concept of multifunctional distribution it was possible to separate the early failure pattern and to relate it to degradation of the capacitors in the MMIC circuit. The main failure distribution was related to gate and ohmic contact electromigration. If the data analysis included the early failures, an MTF of less than 1*105 hour has been obtained.
EARLY REPORTS concerning MMIC reliability have indicated mean time to failure (MTTF) levels of longer than 107 hours at 110oC. To date, most investigations have only concentrated on the active components due to the low transistor density of the MMIC structure which allows for the treatment of each transistor as an independent device. It has been assumed that the active device failure distribution would dominate and may be used to predict total circuit reliability. Numerous investigations, however have given evidence to the bimodality of IC lifetimes. The early pattern of failures shows a distribution of weak components, whereas the main MMIC population will fail much later in life1-4. For the purpose of analysis, it is often more convenient to deal with a cumulative distribution function, rather than a failure density function and to totally ignore the early failure distribution. However, due to the nature of the integrated circuit, such as the MMIC, two or more cumulative distribution functions must be used to correctly describe MMIC failures5. It is therefore imperative that the shape of the failure distribution function be obtained and analyzed.
Statistically, one may treat the MMIC as a distribution of active components
capable of power dissipation, connected to a second distribution of totally
passive components through an appropriate correlation function. Thus previous
assumptions concerning single failure distribution functions for MMICs
may not be valid. In the present investigation, we have derived an appropriate
cumulative distribution functions for a typical MMIC by considering both
the early failure and the main failure distributions.
article is available to CALCE Consortium Members.