STM32F437IIH6 ECC and NAND Flash size

[WLiao.1]

Hello.

1. Whether the hardware ECC for STM32F437IIH6 support error detection and error correction? STM32F437 is connect to NAND Flash through FMC with Bank2. Reference manual RM0090 section 37.6.6 describe it can perform 1-bit error correction and 2-bit error detection. This error detection and error corretion is supported by hardware or realized by software?

2.Figure 457 FMC memory banks in RM0090 describe the Bank2 for NAND Flash is 4x64 MB. Is that means the NAND flash connected to Bank2 should be 256MB max.? However, as NAND flash doesn't have address pins, and the address is defined through the commands, whether I can connect a 512Mbytes or 1Gbytes NAND Flash to Bank2?

Thank you.

[WLiao.1]

Hello, is there anybody could help on this? Thank you.

[WLiao.1]

It's a pity to say that ST community seems not friendly or helpful for developer/engineers. When I ask a question in TI or Microchip community, I can get the response in few hours, that's efficient and surprised to me. However, in here the ST community, 1 week have past but nobody answer my question.

Please try to have a feedback to help on this . Thank you.

[Tesla DeLorean · Accepted solution]

There isn't a size limit as the memory is not entirely mapped, but rather acts as a block storage device.

T​he STM32 does not do correction, that must be done in software. The hardware was able to do some generation and checking functionality, but that is typically done by the NAND memories themselves these days based on the geometry and level requirements.

T​he industry has generally migrated to eMMC at this point as this manages most of the ugliness of managing blocks, wear, ECC, etc of NAND Flash using a common footprint and SDIO/SDMMC interface. Should support 100's of GB with no problems.

[WLiao.1]

Hello DeLorean,

Thank you for your comments.

For the FMC ECC, the reference manual RM0090 page1653 says "The FMC PC Card controller includes two error correction code computation hardware

blocks, one per memory bank. They reduce the host CPU workload when processing the ECC by software.", "The ECC algorithm implemented in the FMC can perform 1-bit error correction and 2-bit error detection". Those not mean the support of hardware ECC for error correction and detection? Or actually FMC is doing the calculation/computation of the error correction code by hardware, and the correction action is finally implemented by software? Whether the main part which takes CPU workload is calculation error correction code, not the correction action?

For the eMMC, Yes it is very convenient to be used, however, the eMMC (usually MLC NAND eMMC) is typically about 3000 program/erase cycles, however SLC NAND flash can up to 100000 program/erase cycles. As we are doing medical devices and the device life cycle shall be 10 years. So eMMC seems not a good choice for us.

Thank you.

[Tesla DeLorean]

As I understand the mechanics of the HW it does the laborious bit level stuff, allowing you to generate the ECC codewords that can be appended, or drop out the error syndromes so you can work out and fix the offending bits/bytes in the RAM copies you've pulled.

Last time I got deep into NAND memories, the newer architectures had the logic to do much the same thing, using algorithms the manufacturer determined most appropriate for the characteristics and failure mechanisms of the cells. This deprecates the usefulness of the ST implementation, and basically why it was dropped from subsequent families, and isn't particularly well documented or supported. At one time they had libraries and demo boards, but then they spun out the NAND group.

Unfortunately I'm not that interested in working on this ECC HW, and I expect the audience is pretty limited.

[WLiao.1]

Hello DeLorean,

Thanks for your help.

[Imen.D]

Hello @WLiao.1​,

Let me first welcome you to the STM32 Community :smiling_face_with_smiling_eyes: and sorry for the delay to reply you.

I just looked at my actions and realized that I had not responded to your post.

Thanks @Community member​ for your help and clarification on this post =)

I will add some notes:

1-It is error detection and possible 1-bit correction, done by CPU and only the ECC computation is done by hardware. In the RM, there is an example use case.

Note that 3, 5 and 6 are done by software.

2- In the RM0090 (page 1648) , you will find this note:

If you still have questions, please don't hesitate to come back to the Community.

Imen