Why the extracted capacitor value is so large?

[swgchlry]

I use Calibre to do the LPE, huanderds of capacitors are extracted. Much of the capacitor have a vaule 0.1fF~1pF, but several capacitor get a value 100F~1000F, unbelieveable! I use tsmc 0.18um 1p6m process!

 

[rfsystem]

I assume a simple unit error in the specific capacitor rule set of Calibre. Search for units and compare them to others. The should be in the same range. The simplest is that lengths are defined in um (micrometer) instead in m (meter). Then area specific values get 10e12 off.

 

[swgchlry]

Thank you, rfsystem!
Would you explain the questio further more? I should search what? "um" or "m"? I have review the rules file, the units of length is "U". There are no places in the rules file to explicit utilize units, but all are numbers. You mean these numbers may be wrong?

 

[rfsystem]

Because the Unit-Definition for the layout is in the GDS for all layers at the same time the error could be only in the parasitic cap rule. You have to search which

layer/substrate area&fringe
layer/layer fringe

cap give the 10e12 off result and look at the definition. Some will have U, others no U.

...


I did take a look at the SVRF manual from 1998 there CAPACITANCE is specified in user units! Only a number w/o unit annotation.

So you must define the length unit and the capacitance unit. So please check that the rules match the process specs. I assume that there is a simple mismatch in the specs.

A mixed usage of the UNIT specs in the rule file is not excluded. So search for where the UNIT is defined and how often.

 

[swgchlry]

Very very thanks rfsystem, I have found the problem. The rules doesn't match the process specs, I extract with a older rule file with no error. rfsystem suggests me to check and compare the rule file , hence i found the problem! thank you, rfsystem!

 

[rfsystem]

A nice note is that there was a discussion some years ago about a user driven bug fixing system in EDA. Cadence for instance is mostly programmed in Skill. But most modules are still crypted.

The proposed system is as follows:

Every user which is blocked or should search for workaround register a time delay at the software vendor. Users which fix errors could earn the delay time of all others. The earned time is counted as productivity contribution and is considered as software fee discount.

But together with the usage model never happen.


AND

Rulesets are not different to that, some cost a half million mask set!