ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 18 NO. 2
22
for both types of error. However, the probability of MCU
events shows a dependence on the technology node,
implying a dependence on the details of the processing
sequence because the SRAM design is similar in all cases.
While single-bit errors are readily corrected at the system/
circuit level using a variety of techniques that are com-
monly known as error-correction codes, these techniques
cannot be used to correct MCU events, and so the latter
tend to dominate the soft-error reliability of SRAMs.
The MCU events are caused by two mechanisms,
leaving aside the possibility of multicell errors caused by
a single direct ion strike. First is bipolar amplification of
charge collection caused by the turn-on of the parasitic
bipolar transistor formed by the source/bulk/drain of the
MOSFET. Positive charge accumulation in the silicon tub
region of the transistor is responsible for the turn-on of
the parasitic bipolar transistor. A secondmechanism that
Fig. 9
Schematic 2-D top view for the inverter structure in a
SRAM. The ion is normally incident to the silicon surface
with a relative distance
D
x
(or
D
y
) away from the edge
of the drain.
Fig. 10
Measured neutron-induced MCU cluster length distributions along the (a)
BL
direction and (b)
WL
direction for 40, 28, and
20 nm planar SRAMs and a 16 nm FinFET SRAM. The unit of the MCU cluster length has been scaled by a constant.
(a)
(b)