The following are the general recommendations on when the error injection on ECC
aggregators must be performed:
- The ECC error injection and monitoring of the error response should be run
at ‘startup’ where possible.
- An alternative recommended approach is to run ECC error injection at
‘shutdown’.
When running of ECC error injection during ‘runtime’, the system integrator must be
aware of the below:
For interconnect endpoints:
- The ECC SEC error injection runs during idle cycles, so can be run during
‘runtime’ for checkers of all types.
- Parity and redundant bit error injections are non-destructive and
can be done at runtime.
- The ECC DED error injection runs during idle cycles, so can be run during
‘runtime’.
- Only single bit error injections are to be done on parity/redundant
checker types. This is because these protect odd number of
injections. A 2 bit injection on parity will not cause a
failure.
For RAM memories:
- ECC self-test for SEC on RAM
memories, can be run at runtime:
- The ECC self-test for SEC injects a single-bit error that is
corrected in-line.
- ECC DED error injection during ‘runtime’ for RAM memories, should only be
done for memories with direct read access
- Two examples of direct read access memories are MSMC SRAM and MCU
SRAM
- Double-bit error injection on all other embedded SRAMS could
possibly result in the downstream IP reading bad data if run during
‘runtime’
- CC error injection on
inject-only memories, should be run at ‘startup’, they are not recommended
to be run during ‘runtime’. For more information on these memories, refer to
Section 12.10.6.2.6.