How to instantiate Pcell with different cell name?

That shouldn't happen (the article was written by me, by the way). The spectre netlister should produce a unique suffix for each variant. It should look like this:

// Library name: mylib
// Cell name: bussetp
// View name: schematic
subckt bussetp_pcell_0 bus\<2\> bus\<1\> bus\<0\> hi lo
parameters busvalue=0
    vlo2 (mid2 0 lo 0) vcvs gain=1-((busvalue&4)>>2)
    vhi2 (bus\<2\> mid2 hi 0) vcvs gain=(busvalue&4)>>2
    vlo1 (mid1 0 lo 0) vcvs gain=1-((busvalue&2)>>1)
    vhi1 (bus\<1\> mid1 hi 0) vcvs gain=(busvalue&2)>>1
    vlo0 (mid0 0 lo 0) vcvs gain=1-((busvalue&1)>>0)
    vhi0 (bus\<0\> mid0 hi 0) vcvs gain=(busvalue&1)>>0
ends bussetp_pcell_0
// End of subcircuit definition.

// Library name: mylib
// Cell name: bussetp
// View name: schematic
subckt bussetp_pcell_1 bus\<7\> bus\<6\> bus\<5\> bus\<4\> bus\<3\> \
        bus\<2\> bus\<1\> bus\<0\> hi lo
parameters busvalue=0
    vlo7 (mid7 0 lo 0) vcvs gain=1-((busvalue&128)>>7)
    vhi7 (bus\<7\> mid7 hi 0) vcvs gain=(busvalue&128)>>7
    vlo6 (mid6 0 lo 0) vcvs gain=1-((busvalue&64)>>6)
    vhi6 (bus\<6\> mid6 hi 0) vcvs gain=(busvalue&64)>>6
    vlo5 (mid5 0 lo 0) vcvs gain=1-((busvalue&32)>>5)
    vhi5 (bus\<5\> mid5 hi 0) vcvs gain=(busvalue&32)>>5
    vlo4 (mid4 0 lo 0) vcvs gain=1-((busvalue&16)>>4)
    vhi4 (bus\<4\> mid4 hi 0) vcvs gain=(busvalue&16)>>4
    vlo3 (mid3 0 lo 0) vcvs gain=1-((busvalue&8)>>3)
    vhi3 (bus\<3\> mid3 hi 0) vcvs gain=(busvalue&8)>>3
    vlo2 (mid2 0 lo 0) vcvs gain=1-((busvalue&4)>>2)
    vhi2 (bus\<2\> mid2 hi 0) vcvs gain=(busvalue&4)>>2
    vlo1 (mid1 0 lo 0) vcvs gain=1-((busvalue&2)>>1)
    vhi1 (bus\<1\> mid1 hi 0) vcvs gain=(busvalue&2)>>1
    vlo0 (mid0 0 lo 0) vcvs gain=1-((busvalue&1)>>0)
    vhi0 (bus\<0\> mid0 hi 0) vcvs gain=(busvalue&1)>>0
ends bussetp_pcell_1
// End of subcircuit definition.

Note the different subckt names.

Have you got a custom nlSetPcellName function in your environment? Try getd('nlSetPcellName) to see if that returns non-nil (you could also try pp(nlSetPcellName) to pretty-print it). I'm wondering whether the default behaviour has been overridden.

Andrew

Hi Andrew,

Really appreciate your replay and suggestion. 

After putting "getd('nlSetPcellName)" into CIW, it returns "funobj:nlSetPcellName". I don't quite understand this return result.

After putting "pp(nlSetPcellName)", it returns  SKILL script for "procedure(nlSetPcellName(cv paramNames paramValues)" and "nil" at the end.

Also, the version I am using is IC23.1-64b.ISR15.37 for your infomation.

Regards,

Yutao

Yutao,

That means that somebody has defined a custom function which prevents the unique naming of different variants (this is what happens if it returns nil). That's incorrect. 

You should find out who defined that in your environment - it's a custom addition.

You could add this at the end of the .cdsinit to disable it (this undefines the function):

putd('nlSetPcellName nil)

However, it's probably wise to find the definition of the function (which is presumably loaded from a .cdsinit or libInit.il file) and remove that incorrect definition of the function. If the function isn't defined, you would get the default unique naming per variant as I showed in my example above.

Regards,

Andrew

Yutao,

That means that somebody has defined a custom function which prevents the unique naming of different variants (this is what happens if it returns nil). That's incorrect. 

You should find out who defined that in your environment - it's a custom addition.

You could add this at the end of the .cdsinit to disable it (this undefines the function):

putd('nlSetPcellName nil)

However, it's probably wise to find the definition of the function (which is presumably loaded from a .cdsinit or libInit.il file) and remove that incorrect definition of the function. If the function isn't defined, you would get the default unique naming per variant as I showed in my example above.

Regards,

Andrew

Hi Andrew,

Thanks for your explanation.

I attach the detail script below, which returns after inputting "pp(nlSetPcellName)" in CIW.

procedure(nlSetPcellName(cv paramNames paramValues)
let((name value
(pcell_suffix "")
)
foreach((name value) paramNames paramValues
when((name == "variant")
(pcell_suffix = strcat(pcell_suffix "_"
lowerCase(value)
))
)
)
strcat((cv->cellName) pcell_suffix)
)
)

After netlisting the test bench, above script grows to cover analog_cell, mcap etc, I am afraid it is generated by the PDK.

I will talk with AE and see how to deal with it.

Regards,

Yutao

Hi Andrew,

Just some follow-up.

I find nlSetPcellName function is used in a simrc file provided from PDK. I am not sure if I can remove that.

I added "putd('nlSetPcellName nil)" in .cdsini file in my home directory. But it does NOT work.

Thanks and regards,

Yutao

Hi Yutao,

If it's defined in the .simrc or generated dynamically, then overriding this in the .cdsinit won't work.

This is a bug in the PDK. It is assuming that the only pcells are those in the PDK and from the code above it relies on having a parameter called variant which is fairly unlikely. There will be other PCells that are incorrectly netlisted too, I think, if this happens.

I think there is an argument that Virtuoso should treat a return value of nil as using the default naming that would be otherwise used. I will ask R&D about that - but the right thing to do here anyway is to get the PDK provider to fix this. It's clearly wrong what is being done right now.

Andrew

Hi Andrew,

Really appreciate your support. I learn quite a lot on SKILL along the way.

It turns out that it is our CAD team use the function to unify subckt format.

They suggest me to include "bussetp" into a local .simrc file.

I did it and now I can see different subckt name.

But I think this is a temporary solution. 

For long term, is it better to have default subckt naming mechanism still working even after user specify customize subckt naming. 

What do you think?

Thanks and regards,

Yutao

Yutao,

It's not clear how you can have the default naming mechanism in place if you've explicitly overridden it to have a different naming convention.

What I've proposed to R&D (via CCR #3216014) is that if the function returns nil, it should use the default naming (right now it use the cell name with no uniquification). Another idea is to provide some means to limit the nlSetPcellName to specific libraries. Anyway, this would need to be considered for a future release.

In the example you gave above, it would return strcat((cv->cellName) pcell_suffix) where pcell_suffix is "", so it's explicitly returning the cellName. That's pretty hard to determine that it shouldn't do this (yes, maybe we could special case against the function returning the cellName unaltered, but there's no guarantee that the returned name is not unique. 

I think the implementation of the nlSetPcellName in the case from your CAD team looks questionable to me, and I think they should fix it.

Andrew