• State TI Thinks Resolved
  • Replies 9 replies
  • Answers 1 answer
  • Subscribers 102 subscribers
  • Views 383 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

TDA4VH-Q1: The Training value of DQ is not centered enough in the eye diagram

heng wang
Prodigy 10 points
Part Number: TDA4VH-Q1

Tool/software:

Dear TI Engineer,

The Training value of DQ is not centered enough in the eye diagram. Can you help analyze the reasons and provide improvement methods?

thanks.

DDRSS3_READ_(DBI ENABLE).xlsx

  • 0
    TI__Mastermind 31655 points

    Hi,

    Can you provide your configuration file (inputs to DDR tool) , as well as comment on the memory part number being used?

    Are you observing functional failures, or just would like to improve the centering of the trained value?

    As a note, VREF (vertical axis) is shared not only between the "rising edge" and "falling edge" of the signal, but also across the entire byte lane which could make alignment for any single signal less than optimal. 

    Regards,
    Kevin

  • 0 in reply to Kevin S
    Prodigy 10 points

    Hi Kevin,

    Thanks for your reply.

    We use Samsung 16Gb DDR, K4F6E3S4HM-THCL, DDRSS_RegConfigTool form and Samsung recommendations as attachments. This board has a worse eye pattern than other boards, and the high temperature margin is very small, resulting in occasional "kernel panic" logs. We have also noticed that the VREF (vertical axis) of DQ16 to DQ23 is low, whether it is DQS Rising Edge or DQS Falling Edge. Is there any way to improve it overall?

    Regards.

    eye diagram.tar

    211210 LPDDR4 4GB Validation Issue on TDA4 with CBT Enable.pdfSPRACU8B_Jacinto7_DDRSS_RegConfigTool——20240712.tar

  • 0 in reply to heng wang
    Prodigy 10 points

    Hi Kevin,

    Following the above question, the test found that at high temperatures, the eye diagram becomes relatively small, but the center position along the rising or falling direction of DQS is relatively fixed. I checked the description of the corresponding register of TRM in DRS821 and read out the following registers in DDRS0 ~3. It seems that the DQ3 training VREF of DDRS33 is smaller. Is that true? Help provide detailed descriptions of these registers and how to correspond to specific voltage values.
    DDRSS_PHY_100[27-16] PHY_PAD_VREF_CTRL_DQ_0
    DDRSS_PHY_356[27-16] PHY_PAD_VREF_CTRL_DQ_1
    DDRSS_PHY_612[27-16] PHY_PAD_VREF_CTRL_DQ_2
    DDRSS_PHY_868[27-16] PHY_PAD_VREF_CTRL_DQ_3
    We tested several boards and found that the vref of the eye diagram is relatively fixed, all around 140mv, and it is relatively fixed with high and low temperatures. Can Vref be set manually?
    Best Regards.

  • 0 in reply to heng wang
    TI__Expert 4425 points

    Hi Kevin,

    can you give some detail explanation for those three registers to customer?

    DDRSS_PHY_100[27-16] PHY_PAD_VREF_CTRL_DQ_0
    DDRSS_PHY_356[27-16] PHY_PAD_VREF_CTRL_DQ_1
    DDRSS_PHY_612[27-16] PHY_PAD_VREF_CTRL_DQ_2
    DDRSS_PHY_868[27-16] PHY_PAD_VREF_CTRL_DQ_3
    BR,
    Biao
  • 0 in reply to Biao Li
    Prodigy 10 points

    Hi Kevin，

    The question is why the reading DQ training value is not centered in the eye diagram.And why the DQS Rising Edge eye diagram is better than the DQS Falling Edge.
    Looking forward to reply, thank you~
  • 0 in reply to heng wang
    TI__Mastermind 31655 points

    Hi,

    First, really sorry for the delay.

    Biao Li said:

    can you give some detail explanation for those three registers to customer?

    DDRSS_PHY_100[27-16] PHY_PAD_VREF_CTRL_DQ_0
    DDRSS_PHY_356[27-16] PHY_PAD_VREF_CTRL_DQ_1
    DDRSS_PHY_612[27-16] PHY_PAD_VREF_CTRL_DQ_2
    DDRSS_PHY_868[27-16] PHY_PAD_VREF_CTRL_DQ_3

    Instead of hard-coding, you could reduce the VREF search range, especially on the low end.

    For instance, if you are observing low VREF just on byte lane 2, you could change the following parameters:

    * DDRSS_PHY_597[30:24] - Start value (should default to 0x10)
    * DDRSS_PHY_598[6:0] - Stop value (should default to 0x40)

    heng wang said:
    The question is why the reading DQ training value is not centered in the eye diagram

    The data pattern we use to generate the eye image is more stressful than what is executed as part of training (training is limited to values stored in MR32 / MR40). In other words, the training point might appear more centered if using the same data pattern used by training to generate the eye.

    Regards,
    Kevin

  • 0 in reply to Kevin S
    Prodigy 10 points

    Hi Kevin,

    Thanks for your reply.

    There is another question to help confirm, why the DQS Rising Edge eye diagram is better than the DQS Falling Edge.

    Best Regards.

  • 0 in reply to heng wang
    TI__Guru**** 427585 points

    Hi ,

    Any further update on this thread? 

    Regards,

    Brijesh

  • 0 in reply to Brijesh Jadav
    TI__Expert 4425 points

    Hi Brijesh, 

    Close this thread because of no further update.

    BR,

    Biao