Sorry, you need to enable JavaScript to visit this website.

Memory Corruption on Rev-F07

Zedboard forums is currently read-only while it under goes maintenance.

Unsolved
10 posts / 0 new
q li
Junior(1)
Memory Corruption on Rev-F07

We recently received a new batch of Rev-F07 MicroZed boards.  Memory corruption are detected on these boards. 
After compared Rev-F07 and Rev-F06, We found that the memory chip is changed from Micron to Kingston.  Please help to resolve this problem.

JFoster
Moderator(77)
Hi Q Li,

Hi Q Li,

We are currently looking into your issue. Could you please provide me with a little bit more information.

How many Rev-F06 boards do you have and are all them working?

How many Rev-F07 boards do you have and are they all failing?

How are you determining the Memory corruption?

At what temperature are you seeing this failure?

Are you boards commercial or industrial grade, if you have a mixture of commercial and indstrial grade boards can you be percise with how many of which Rev boards you have?

Thanks,

Josh

q li
Junior(1)
Hi Josh,

Hi Josh,
Thank you for fast response.
We have more than 100 Rev-F06 and earlier versions boards, all of them doesn't have this problem.
Recently we built about 10 products with Rev-F07 boards, all of them can't pass testing due to system crash after a few hours of stress testing.
We noticed that the memory chips are different between Rev-F07 and earlier versions,  A memory testing program is developed to verify DDR memory transactions.  We found that the data error rate is about 10E-8 level on the Rev-F07 boards.  No error is detected on Rev-F06 boards. 
The testing are performed at room temperature about 20C.  We are using industrial grade 7020 microzed boards. 
 
Regards,
Qing

JFoster
Moderator(77)
Hi Qing,

Hi Qing,

Would you be willing to run your memory stress teest with the memory interface at a slighly lower speed to see if that makes any difference and report back your results?

What I mean by slightly lowering the memory interface speed is just having you change the DDR3 clock in the PS Configuration Wizard from 533MHz to a lower frequency (The allowable clock range is 200 to 533MHz).

This will give us another test point to determine if its an issue with the DDR3 controller timing parameters.

Thanks,

Josh

q li
Junior(1)
Hi Josh,

Hi Josh,
 
I just ran the memory stress testing with 466.66MHz DDR clock.  So far there is no error detected on the Rev-F07 board. 
However reduce the DDR clock rate will limit the performance of our product.  We need a solution to operate the DDR at 533.33MHz.
 
Regards,
Qing

JFoster
Moderator(77)
Hi Qing,

Hi Qing,

That is good news, next you should set the clock back to 533MHz and loook into the following DDR3 parameters

Parameter                    Current Setting

CAS Latency                             7

CAS Write Latency                    6

RAS to CAS Delay                    7

Precharge Time                         7

 

Our thoughts are that you may need to increase each parameter by 1.  Try doing so and running your DDR3 test again. If it fails please try tinkering with the Parameters a bit more and testing them. Please post back your results.

Thank you,

Josh

q li
Junior(1)
Hi Josh,

Hi Josh,
We have already tested various combination of Latency settings.  Error rate is not dropped with larger parameter values.
In addition we have tried to adjust tRC, tRAS, tFAW.  And found they are not useful to fix the problem.
We found that uncheck Write leveling train could reduce the error rate to 10E-10 from 10E-8 on one of the board under testing.  However 10E-10 error rate will still lead to a system crash which is no acceptable.
Regards,
Qing
 
 

fletch
Moderator(23)
Micron part markings

Qing,

  Will you please tell me what the Micron DDR3L part markings are on your functioning F06 boards? Or, if you can give me a serial number for one or two boards, I can look it up.

  We know the Kingston memories are rated at 1600. I'm curious to know if your functional boards have the -107 (1866) or -093 (2133) Micron memories. If so, that would explain a margin difference.

  We also run memory tests across temperature when validating our boards. We validated the boards with -125 (1600) Micron memories, and then again with the Kingston memories. I'm very interested to learn more about your memory testing program. I'd like to see if we can duplicate and then alleviate the problem on boards we have. If you prefer, please private message me some details regarding what we could do to duplicate your findings on MicroZed 7020 Industrial F-07 boards that we have.

Bryan

 

q li
Junior(1)
Hi Bryan,

Hi Bryan,
 
I checked one of the Microzed board, the Micron chip label is 4AE77 D9PXV.
 
Regards,
Qing
 

fletch
Moderator(23)
1600 memory

Thanks, Qing. That is a MT41K256M16HA-125:E, which is a 1600 Mbps memory, same as the Kingston. I'm not sure what the difference is. I'm in contact with Wayne from your local field office to work out a resolution for you.

Bryan