MLC ECC with limited resources (FPGA/MCU)

Hi everyone,

Now, I'm quite new in this, so please forgive me if I say silly things.

I'm seeing whether or not it is possible (feasible) to implement MLC 
NAND error correction coding on our current system. We have a cortex-M3 
(1MB Flash, 128K RAM) and a small FPGA with no DSP. On the FPGA I only 
have ~1K LUTs remaining.

After searching the internet I found that BCH most commonly mentioned as 
ECC scheme for MLC NAND. Also Reed-Solomon is mentioned, but it is said 
that RS is not suitable for MLC NAND with large page sizes (our NAND has 
an 8KByte page size).

What ECC scheme is best suited to be implemented on MCUs? If I would 
have to implement such a scheme, what functionality would you put in the 
MCU and what in the FPGA? Are there simpler ECC schemes which can deal 
with MLC NAND? Any insights are greatly appreciated!

Regards,
Vincent

On Wednesday, January 13, 2016 at 10:24:39 AM UTC-5, Vincent vB wrote:
> Hi everyone,
> 
> Now, I'm quite new in this, so please forgive me if I say silly things.
> 
> I'm seeing whether or not it is possible (feasible) to implement MLC 
> NAND error correction coding on our current system. We have a cortex-M3 
> (1MB Flash, 128K RAM) and a small FPGA with no DSP. On the FPGA I only 
> have ~1K LUTs remaining.
> 
> After searching the internet I found that BCH most commonly mentioned as 
> ECC scheme for MLC NAND. Also Reed-Solomon is mentioned, but it is said 
> that RS is not suitable for MLC NAND with large page sizes (our NAND has 
> an 8KByte page size).
> 
> What ECC scheme is best suited to be implemented on MCUs? If I would 
> have to implement such a scheme, what functionality would you put in the 
> MCU and what in the FPGA? Are there simpler ECC schemes which can deal 
> with MLC NAND? Any insights are greatly appreciated!
> 
> Regards,
> Vincent

Start at the beginning, and state clearly:
What problem are you trying to solve?

Op 13-1-2016 om 16:28 schreef Dave Nadler:
> On Wednesday, January 13, 2016 at 10:24:39 AM UTC-5, Vincent vB wrote:
>> Hi everyone,
>>
>> Now, I'm quite new in this, so please forgive me if I say silly things.
>>
>> I'm seeing whether or not it is possible (feasible) to implement MLC
>> NAND error correction coding on our current system. We have a cortex-M3
>> (1MB Flash, 128K RAM) and a small FPGA with no DSP. On the FPGA I only
>> have ~1K LUTs remaining.
>>
>> After searching the internet I found that BCH most commonly mentioned as
>> ECC scheme for MLC NAND. Also Reed-Solomon is mentioned, but it is said
>> that RS is not suitable for MLC NAND with large page sizes (our NAND has
>> an 8KByte page size).
>>
>> What ECC scheme is best suited to be implemented on MCUs? If I would
>> have to implement such a scheme, what functionality would you put in the
>> MCU and what in the FPGA? Are there simpler ECC schemes which can deal
>> with MLC NAND? Any insights are greatly appreciated!
>>
>> Regards,
>> Vincent
>
> Start at the beginning, and state clearly:
> What problem are you trying to solve?
>
Hello Dave,

We put MLC NAND because we needed more storage on our PCB, but only 
discovered later on that our MCU's hardware ECC support is not 
sufficient. So, now I'm seeing whether or not we can salvage the PCBs by 
implementing the ECC in soft/gate-ware, or whether we need to bite the 
bullet and do a hardware redesign possibly using eMMC.

Regards,
Vincent

Op Wed, 13 Jan 2016 16:24:59 +0100 schreef Vincent vB <embedded@spam.com>:
> I'm seeing whether or not it is possible (feasible) to implement MLC  
> NAND error correction coding on our current system. We have a cortex-M3  
> (1MB Flash, 128K RAM) and a small FPGA with no DSP. On the FPGA I only  
> have ~1K LUTs remaining.
>
> After searching the internet I found that BCH most commonly mentioned as  
> ECC scheme for MLC NAND. Also Reed-Solomon is mentioned, but it is said  
> that RS is not suitable for MLC NAND with large page sizes (our NAND has  
> an 8KByte page size).
>
> What ECC scheme is best suited to be implemented on MCUs?

The one that best fits your application requirements.

> If I would have to implement such a scheme, what functionality would you  
> put in the MCU and what in the FPGA?

I would put everything in the MCU, unless a significant enhancement can be  
had (e.g. in response time or power usage).

> Are there simpler ECC schemes which can deal with MLC NAND?

Why do you need a simple scheme?

> Any insights are greatly appreciated!



-- 
(Remove the obvious prefix to reply privately.)
Gemaakt met Opera's e-mailprogramma: http://www.opera.com/mail/

Hi Boudewijn,

On 15-1-2016 at 13:42, Boudewijn Dijkstra wrote:
>>
>> What ECC scheme is best suited to be implemented on MCUs?
>
> The one that best fits your application requirements.

The issue is that I currently do not know what best meets my application 
requirements. I don't even know it is it feasible at all.

I did not realize until too late that the ECC scheme supported in 
hardware by my MCU is not good enough for the MLC NAND now on our PCB. 
So, I'm seeing if there is a way to salvage the already produced PCBs by 
using a (partial) software ECC implementation.

>
>> If I would have to implement such a scheme, what functionality would
>> you put in the MCU and what in the FPGA?
>
> I would put everything in the MCU, unless a significant enhancement can
> be had (e.g. in response time or power usage).

As far as I know most bit-wise algorithms are better suited in an FPGA. 
For example a BCH encoder uses Galois field polynomial division. 
Something which does not map very well on an MCU, as far as I 
understand. But maybe it does, and I did not realize. However, I do not 
wish to limit myself to BCH, though it does seem to be the most suitable 
at the moment.

>
>> Are there simpler ECC schemes which can deal with MLC NAND?
>
> Why do you need a simple scheme?

I know MLC NAND Flash schemes tend to take up a lot of clock cycles 
and/or memory on an MCU. Maybe there is an algorithm or trade-off which 
uses bigger parity codes, but takes up less cycles.

Anyway, I know I am asking a rather broad and vague question. First of 
this is because I am not very schooled in this matter and second because 
I don't want to narrow the discussion in fear or missing possible solutions.

Just an elaboration of more experience people in this matter would also 
be helpful. Maybe others have tried to do similar things, and can share 
some insights.

Regards,
Vincent