2 Replies Latest reply on Apr 9, 2018 3:53 PM by weston_beal

    DDR Batch wizard monotonicity

    pjs

      Hi,

       

      I'm running some DDR4 analysis and I have some DDR Writes which are failing the monotonic test. When I examine the waveform I see the following:

       

       

      One thought as to why this is might be failing is that the default VINH in the IBIS is defined as 1.06V which corresponds directly with the blip we are seeing in the waveform. Having spoken briefly to the DDR chip manufacturer they suggested that in DDR4 the thresholds are dynamic and are based off the DQ Rx mask and the VcentDQ. They insinuated that the VINH value is only really there as a placeholder to meet the IBIS standard.

       

      Do you know if the DDR batch wizard is using the VINH from the IBIS file or will it be using a dynamic value based off the VcentDQ? The results table suggests that the 'Reference voltage used' is 762mV and so I believe this leaves plenty of headroom. The eye density plots also show all the DQ lines meeting the Rx mask with plenty of headroom.

       

      Do you have any idea what thresholds are being used during the monotonic test?

        • 1. Re: DDR Batch wizard monotonicity
          pjs

          Hi,

           

          Just wondering if anyone has any thoughts on this one?

           

          Thanks

          • 2. Re: DDR Batch wizard monotonicity
            weston_beal

            PJS,

             

            I didn't have a test case to test this. I looked at the user guide and found this description.

            For single-ended nets, the monotonicity threshold check is performed while the waveform is between VIL(DC) and VIH(DC). Simulation uses the following receiver threshold information from the IBIS model (highest priority first):

            VIL(DC)

            1. [Receiver Thresholds] keyword, Vth - Vinl_dc subparameter
            2. [Model Spec] or [Model] keyword, Vinl subparameter

             

            VIH(DC)

            1. [Receiver Thresholds] keyword, Vth + Vinh_dc subparameter
            2. [Model Spec] or [Model] keyword, Vinh subparameter

             

            It makes sense to use the Vin[l,h]_dc subparameters, and I hoped that it would make the actual thresholds from the calculated Vref value reported in the DDR report.

             

            I made a small DDR4 test and adjusted the [Receiver Thresholds] subparameters to get some monotonicity errors. It appears that the monotonicity check uses all the values (including Vth) from the IBIS receiver. It appears that the check does not use the calculated Vref value. What does the DDR4 specification say about monotonicity requirements?

             

            Regards,

            Weston