1 Reply Latest reply on Aug 3, 2011 11:02 AM by weston_beal

    IBIS-AMI edges too fast for s-parameter bandwidth

    david.royle

      Folks,

       

      I have a device whose IBIS-AMI model produces a 1-2ps rising edge (inferring hundreds of GHz of bandwidth) . However, my s-parameters (QSFP connector for example) are only characterized to 20GHz (and with warnings not to simulate risetimes more than 25ps).

      Does Hyperlynx handle this gracefully, like looking at a 250GHz signal with a 20GHz scope, or will I expect aliasing to render the 20GHz bandlimited result unreliable?

       

      The s--parameter notes also caution that, even with risetimes > 25ps,  the time resolution of 25ps is not adequate for most time domain simulations unless "tricks such as windowing are used to minimize leakage and aliasing".

      Does Hyperlynx include these tricks?

       

      Thanks,

      Dave

        • 1. Re: IBIS-AMI edges too fast for s-parameter bandwidth
          weston_beal

          The first problem is that the 2ps rise/fall time in the model is bogus. The intent is to make a very fast edge as the stimulus and then run that through the S-parameter model that represents the buffer (and package?). Then the resulting filtered waveform should be representative of the signal out of the actual device.

           

          HyperLynx handles this condition of the super-fast edge rate very well with the advanced complex pole fitting (CPF) algorithm. This is one of the strengths of HyperLynx. The warnings in your model assume that the simulator uses convolution to model the S-parameters, which are frequency domain data, in a time domain simulation. HyperLynx doesn't have these problems because it uses CPF to make a causal model of the S-parameters that cleanly extrapolates the data to infinite frequency.