The SRAM transient event count is shown in
Table 8-7 Key takeaways from the data are:
- ECC protected SRAM(Mx and Dx RAMs) are
two orders of magnitude less susceptible to transient upsets. This implies that during the
read/write duration the majority of upsets are single bit correctable errors
- Due to the nature of ECC implementation
on SRAMs, a correctable error is not only corrected prior to use by the CPU, but the
corrected value is also written back to the SRAM avoiding aggregated errors
- Parity protected SRAM(LSx and GSx), while
observing more failures than the ECC protected RAMs, will be able to detect most
transients due to their single bit nature.
- While all SRAMs are constructed
similarly, it is important to keep in mind that due to size and number of instances that
GSx SRAM will have a higher probability of transient events based on its on die area.
Table 8-7 SRAM Transient Upset Summary
| SRAM Type |
LETEFF(MeV·cm2/mg |
Ion Type |
Bit Flips |
Bit Total |
Fluence (# of ions) |
Cross Section(Event Count/LETEFF |
| Mx and Dx (ECC) RAMs |
45 |
109Ag |
0 |
98304 |
1.00 E6 |
1.85 E-61 |
| Mx and Dx (ECC) RAMs |
8.5 |
40Ar |
0 |
98304 |
1.01 E6 |
6.07 E-71 |
| LSx and GSx(Parity) RAMs |
45 |
109Ag |
3230 |
1,245,84 |
1.00 E6 |
2.27 E-4 |
| LSx and GSx(Parity) RAMs |
8.5 |
40Ar |
521 |
1,245,84 |
1.01 E6 |
3.47 E-5 |
- Since there were no upsets observed the
cross section for this section is calculated as a Mean Fluence To Fail (MFTF) at 95%
confidence interval based on the average fluence of the runs. See Single-Event Effects
Confidence Interval Calculations for more information on this
calculation.