M. Nafria et al., CHARACTERIZATION OF SIO2 DIELECTRIC-BREAKDOWN FOR RELIABILITY SIMULATION, I.E.E.E. transactions on electron devices, 40(9), 1993, pp. 1662-1668
Breakdown statistical distributions of thin oxide MOS capacitors are a
nalyzed to reveal the limitations of accelerated procedures for reliab
ility simulations at circuit level. The detailed analysis of the break
down distributions corresponding to different stress voltages shows th
e presence of three modes of breakdown, associated with different kind
s of defects. These breakdown modes control the shape of the distribut
ions at different oxide field ranges. The mean values of time-to-break
down, charge-to-breakdown, and energy-to-breakdown and their dependenc
e on oxide field are found to be directly related to the mode dominati
ng the distribution. Since the shape of the breakdown distribution cha
nges with stress voltage, an accelerated testing procedure must provid
e a way to extrapolate to operation conditions not only the mean value
, but also the shape of the distributions. Our results indicate that d
ifferent physical mechanisms cause the breakdown at low and high field
s. Consequently, the acceleration conditions have to be low enough to
avoid different breakdown mechanisms in the accelerated and normal ope
ration field ranges. The importance of localized latent defects is als
o demonstrated. These defects, which require a certain stress time (ro
ughly independent of stress conditions) to cause the oxide failure, do
minate the low-probability tail of the breakdown distribution and dete
rmine the failure rate of complex IC's. In addition to the intrinsic b
reakdown model, the latent-defect-related tails have to be also correc
tly accounted for in reliability simulators.