Skip to Main Content
Your organization might have access to this article on the publisher's site. To check, click on this link:http://dx.doi.org/+10.1063/1.353806
This work is principally concerned with the microstructure of electromigration failure in narrow Al‐2Cu‐1Si conducting lines on Si. Samples were patterned from 0.5‐μm‐thick vapor‐deposited films with mean grain size of 2.4 μm, and had linewidths of 1.3 μm (W/G≊0.5), 2 μm (W/G≊0.8), and 6 μm (W/G≊2.5). The lines were tested to failure at T=226 °C and j=2.5×106 A/cm2. Other samples were tested over a range of substrate temperatures and current densities to test the effect of these variables, and 1.3 μm lines were tested after preaging at 226 °C for various times to change the Cu‐precipitate distribution prior to testing. Three failure modes were observed: The 6 μm specimens failed by separation along grain boundaries with an apparent activation energy of 0.65 eV; the 1.3 μm specimens that were preaged for 24 h failed after very long times by gradual thinning to rupture; all other narrow lines failed by the transgranular‐slit mechanism with an activation energy near 0.93 eV. Microstructural studies suggest that the transgranular‐slit failure mechanism is due to the accumulation of a supersaturation of vacancies in the bamboo grains that terminate polygranular segments in the line. Failure occurs after Cu has been swept from the grain that fails. Failure happens first at the end of the longest polygranular segment of the line, at a time that decreases exponentially with the polygranular segment length. Preaging the line to create a more stable distribution of Cu lengthens the time required to sweep Cu from the longest polygranular segment, and significantly increases the time to failure. In the optimal case the transgranular‐slit failure mechanism is suppressed, and the bamboo grain fails by diffuse thinning to rupture. Preaging is particularly effective in increasi- ng the lifetimes of lines that contain very long polygranular segments, and has the consequence that the time to first failure in an array of lines is much longer than predicted by a log‐normal fit to the distribution of failure times.