MPC5777M ECC 1b error

[hexshield]

Hello,

I am doing some testings regarding ECC handling on MPC5777M.

Let’s say I first write the following 64 bits of data: 0xDEADBEEF_FFFFFFFF

If I overwrite that 64 bits data sector without erasing first with the following data: 0xDEADBEEF_FFFFFFFE (just 1 bit difference), I get a multibits ECC error (unrecoverable).

Is that because ECC checkbits are also overwritten ? If that is the case, is that like the flash (bits can only go from logic 1 to logic 0) ? 
From my calculations, on first data write ECC checkbits are 0x5C then on the overwrite data checkbits are 0x9D.

[raph]

Hi, I'm also trying to understand better ECC and want to inject ECC at specific addresses to check if my code handles it properly. 

If I understood correctly, you get ECC by writing 64bits of data at specific address twice. 

Can you share the way you did to write 64bits at specific address ? 

[WangShine]

Hello guys,
I want to know where is the checkbits in MCU.I just set the DATA's address,but I can't find the place to store those checkbits.So that I can't understand what happened in checkbits when I use EIM to inject some transport errors into checkbits,however those functions about DATA are exactly.

[petervlna]

Hello,

Is that because ECC checkbits are also overwritten ?

Yes

If that is the case, is that like the flash (bits can only go from logic 1 to logic 0) ?

Yes

From my calculations, on first data write ECC checkbits are 0x5C then on the overwrite data checkbits are 0x9D.

Hard to comment here. But what you are doing is exactly what you have to do if you like to create ECC faults in flash.

If you want to avoid this scenario, erase flash first and then program it.

Best regards,
Peter

[hexshield]

Hi Peter,
Thanks for reply.

Creating ECC faults is intended as i want to learn about the ECC logic used in flash.

So i can conclude that in a case of overwritting, new data is ANDed with currently programmed data, same for ECC checkbits.

I based my ECC checkbits calculation on MPC5777M RM, the Hsiao H-matrix is defined there, but it seems that production devices are using a different matrix.
Is that the case ? If so can it be shared or is it proprietary ?

Thanks !

[petervlna]

Hello,

I have asked about it in application team and typically we don’t release the algo details to customers.

Best regards,
Peter

[hexshield]

Hi, thanks for your answer.

What do you exactly mean by "typically " ? Is that you will never release algo details to a customer by any mean, or, there can be a possibility to get some details for example by signing an NDA ?

Thanks

[petervlna]

Hello,

Well, we can release such info on request based on customer and business related to it.

So for example if Bosch or Conti will request it, our FAE can decide if we are going to share it.

If some unknown community account request it then we wont for sure.

So in case you need such info, please contact your NXP FAE or sales with such request.

Best regards,

Peter