3 Replies Latest reply on Mar 2, 2017 11:44 AM by Carlos_A

    Unreliable DIMM B operation in Haswell based design

    PVWB Brown Belt

      Dear Embedded Community,

       

      We have a Haswell based processor design.  It has two DIMMs,

      one connected to each of the two DDR controllers.  The "A"

      DIMM appears to work.  The "B" DIMM gets occasional errors

      and the processor will often crash during booting Linux or

      some time afterwards.

       

      We are seeing some differences between the "A" and "B" DIMM

      signals.  We have been triggering on ODT.  We believe that when

      ODT is high, this should mean that the processor is driving DQS

      and data towards the DIMM.

       

      In the attached traces, the pink signal is ODT and the yellow and

      blue signals are DQSxP/N.

       

      In the attached trace for DIMM A DQS7, APR2i_A_DQS7_ODT.tif,

      during ODT we get what we would expect: ODT asserted, followed

      by eight clock edges then ODT deasserted.

       

      But in the attached trace for B DQS2, APR2i_B_DQS2_ODT.tif,

      during ODT we get an initial bad clock transition.  Followed

      by some O.K. clocks, then clock seven seems to be just plain

      missing.  I particularly don't understand the missing clock.

       

      The DIMM "B" trace lengths follow the requirements in the

      Haswell Desktop and Denlow-WS Platform Design Guide, 486711

       

      Is there any explanation that you can give us for the very

      odd behaviour of the clocks on the "B" DIMM?

       

      We can send the Allegro .brd layout file if this would be useful. 

      Does Intel offer a layout examination service?  This would be very

      helpful.  We have had the schematics looked at by Intel in the past.

       

      Thanks, Paul.