26 / 58
edfas.org
ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 18 NO. 2
26
the identical
D
x
before and after turning off the doping-
dependent mobility degradation. The diffusion current
peaks of both devices shift to shorter times, and the peak
heights of both increase. Moreover, the current peak
height of the 16 nm device increases greater than that of
the 20 nm device, suggesting the mobility degradation
is larger for the 16 nm device. Figure 16 compares
Q
coll
as
a function of the relative
D
x
before and after turning off
the mobility degradation for each SRAM device. The
Q
coll
with
D
x
~ 0 (including drift and diffusion currents) does
not change after turning off the model, because charge
mobility is dominated by the high electric field in the
drain region. The
Q
coll
with larger
D
x
(diffusion current
Fig. 16
Comparison for
Q
coll
(
D
x
) before and after turning off the doping-dependent charge mobility model
only) for all nodes increases after turning off the mobility
degradation. Clearly, doping profiles significantly impact
the diffusion component, having an increasingly large role
in decreasing charge collection with scaling. Finally, after
normalizing for the changes in the drain area, as shown in
Fig. 17, it is found that the distributions of cluster length
of all nodes are similar.
SUMMARY AND CONCLUSIONS
The introduction of the FinFET transistor architec-
ture has resulted in a significant reduction in the rate
of radiation-induced soft errors in SRAMs for all known
particle-strikemechanisms. Both single-bit andmulticell
Fig. 17
Q
coll
(
D
x
) distributions with the model off normalized for identical drain area