0 Replies Latest reply on Jul 7, 2016 2:30 AM by dontpanic

    PEX: How to choose parameters for "REDUCE ANALOG" option?

    dontpanic

      Hello! I would be grateful if someone could please give me further insight on how to appropriately choose the parameters for the "REDUCE ANALOG" option for R-C-Cc extractions.

       

      I'm trying to reduce the size of extracted R-C-Cc netlists for critical blocks in my design. I have noticed that using the "REDUCE ANALOG" option provides better reduction than "Reduce TICER", but I don't understand how this reduction method works and how to use it properly. The documentation says that it combines the "Reduce TICER" and "Reduce CC" methods, but it does not give any further details on how it does it. In particular, I would like to know the relationship between the parameters DELAY_ERROR and NOISE_ERROR for "REDUCE ANALOG" and the parameters for TICER and CC reductions ("frequency"; "threshold", "percent", etc). I understand the meaning of the latter when using the "Reduce TICER" and "Reduce CC" methods separately, but it is unclear to me how to select the former (i.e. DELAY_ERROR and NOISE_ERROR) when using "REDUCE ANALOG".

       

      The documentation says:

       

      DELAY_ERROR delay: Optional keyword set specifying the timing delay threshold used for parasitic reduction. The delay units are seconds. The default is 0.5e-12 seconds (0.5 picoseconds).

      ...Does this mean that parasitics branches with time constants smaller than DELAY_ERROR are combined/trimmed out?

       

      NOISE_ERROR noise: Optional keyword set specifying the error threshold used for parasitic reduction. The noise error is defined as the amount of change in noise amplitude relative to a full strength signal. The value for noise is a ratio, a unitless floating number. The default is 0.01, or 1%.

      ...what is noise in the context of an extracted netlist? I cannot really understand what this noise error means...

       

      Could somebody please elaborate on these parameters?

       

      Thanks,

      Jorge.