James J. Filliben
Statistical Engineering Division, ITL
Semiconductor Electronics Division, EEEL
Semiconductor manufacturers have made great strides in increasing chip speed while maintaining chip reliability. An increasingly important potential size and speed bottleneck is not the chip, but the aluminum alloy thin film stripes that interconnect one chip with another. For the system to be speed-efficient as a whole, it is important that the interconnects-not just the chips-be made smaller.
Such thin film interconnect miniaturization suffers from electromigration: the phenomenon in which ion-electron collisions intrinsic to current flow induce internal physical movement causing physical deformations and voids being formed and de-formed. Such migration has of course a deleterious effect on the lifetime and reliability of the interconnect. The understanding of such microstructure mechanics is essential for future reliable miniaturization.
Santos Mayo of the Semiconductor Electronics Division of EEEL is addressing this problem. Specifically, he is investigating and characterizing the damage/reliability mechanisms associated with electromagnetic processes in aluminum alloy thin film interconnect stripes on a silicon dioxide substrate.
Santos has carried out an accelerated experiment whereby he has applied high temperature (in the 175 to 200 degree C range) and high current (in the 1.5 to 2 million amperes per square centimeter range) to such thin film interconnect stripes over the course of several hours and recorded the resistance trace at .001 second intervals. Such a resistance trace is relatively uneventful for most of the time period. At "random" points in time, however, the traces have resistance surges which are signatures of certain types of electromigration activities (void formation, void dispersion, etc.) and which cumulatively serve to degrade the current-carrying capacity of the stripe and ultimately shorten the lifetime of the stripe. Santos' data collection system has been designed to ignore periods of no activity (when a Santos-defined threshold has not been exceeded, but to heavily record (1000 observations per second) when a "possible-event" has been detected. Such several-hour data sets typically contain about 200,000 observations-200 "possible-events" with 1000 observations per "possible-event".
SED is analyzing such records for event/peak identification; peak area calculation; and categorization of events into different types (corresponding to different electromigration mechanisms). A future experiment design will check the generality of the conclusions.
Figure 26: Height, width, and area of event peaks.
Date created: 7/20/2001