Spectre X fails to output accurate nutascii formated .raw output file if +postlpreset is used

Quick follow-up:   scaling up the design, I noticed Spectre X still had trouble parsing the variables and values correctly in the output .raw file.   I was able to resolve this (i think) by essentially disabling the parasitic optimization using the command line option +postlayout=upa.  This may be a bug, but I guess a nutascii formatted .raw output is barely supported, given the file type is not popular with users.  

This is nothing to do with using nutmeg/nutascii format (not sure what you mean by "parsing the variables and value correctly in the output .raw file" - nothing would get parsed from an output file). It is happening regardless of the output format (the same happens with PSF XL for example).

What seems to be happening is that if you explicitly pick a +postlpreset (which you normally do not need to - Spectre X will apply reduction automatically if it makes sense, at the same level as +preset), it is forcing reduction of the resistors into a single series resistor and causing the output nodes to be incorrect (I would have expected the save statements to have prevented the collapsing of the resistors, but this isn't happening).

I've filed CCR 2781881 for this, but with this kind of circuit I would not explicitly turn on post-layout reduction (since it doesn't really make sense). You might want to report the issue yourself to customer support so that it can be tracked under your name as a duplicate (the example is a bit artificial, so maybe you have a more realistic example).

Regards,

Andrew

Interesting.  Thank you for looking into it further.  I hadn't looked at any other output format and assumed it was a nutascii parsing thing based on the output.    In the end, I had to turn off post-processing with  +postlpreset=off.   I can provide a much more complicated, perhaps more realistic, example, but the artificial/simple one posted here makes it easy to understand things are not working right.

Some general points here:

• You do not normally need to pass +postlpreset when using +preset. Our advice is to not do that unless there is a specific reason to alter the global level of post-layout optimisation (for example, you have a low-frequency but high-precision circuit, you may want +preset=cx but can be more aggressive with post-layout optimisation and so may pick +postlpreset=mx or lx).
• In this case, the default is to not do the post-layout optimisation with this circuit - Spectre X determines that it doesn't need to. By specifying the command-line option, you've forced it to do the optimisation. With a real-life RC post-layout design, it would make more sense.
• In this case, the problem is caused by you having ideal resistors - in real-life circuits where the resistor ladder is part of the design, the resistors would be unlikely to be pure linear ideal resistors and so would not be subject to reduction.
• If you have a bigger circuit (with active devices in) and this problem is occurring when just specifying +preset=mx (and not +postlpreset), then you could use the approach described in this blog to ensure that the specific nets are not impacted:
  Spectre Tech Tips: Handling Differential Matching Circuits in Spectre X
  Thiscan be done by adding: postlpreset=off postlnets=[n*] to your options statement as that suppresses the reduction for these nets only.
• However, I agree that this should behave better. We'll see what R&D say as I think that there's an explicit save that it ought to not reduce the nets.

Regards,

Andrew