Dw. Malone et Re. Hummel, ELECTROMIGRATION FAILURE OF INTEGRATED-CIRCUIT METALLIZATIONS SUBJECTED TO HIGH-FREQUENCY PULSED CURRENTS, Journal of applied physics, 83(11), 1998, pp. 5750-5760
Electromigration life tests were performed on copper-alloyed aluminum
test structures that were representative of modern integrated circuit
metallization schemes. A total of 18 electrical stress treatments were
investigated. One was a steady de current, and the others were pulsed
de currents varied according to duty cycle and frequency. The duty cy
cle was varied from 33.3% to 80%, and the frequency was varied such th
at three distinct orders of magnitude were roughly represented-100 kHz
, 1 MHz, and 100 MHz. The median time to failure, t(50), was used as t
he primary basis of comparison between test groups of six-nine samples
. There was no discernible dependence of t(50) on the pulse frequency.
In contrast, the duty cycle, d, played a strong role. Duty cycles gre
ater than 50% produced median lifetimes that varied as 1/d(2), while d
uty cycles equal to and less than 50% produced a shift toward (but not
reaching) a lid dependence. This shift was most pronounced at the sma
llest duty cycle of 33.3%, for which lifetimes were midway between the
predictions of the 1/d(2) and 1/d relationships. Post-test optical mi
crographs were obtained for each test stripe, and these suggested that
the location of electromigration damage was influenced by the pulse d
uty cycle. Most damage occurred near the cathode contact in all instan
ces, but there was an increased incidence of damage farther downwind w
ith decreasing duty cycle. The results are explained in terms of a Ble
ch length effect that may operate as a result of the bamboo test strip
e structure. (C) 1998 American Institute of Physics. [S0021-8979(98)02
111-2].