Do people not use common centroid layout anymore?

[sharkies]

I have a chip that is nearly finished in layout, but found out that common centroid layout scheme was not used. I have a lot of differential pairs that are part of CMFB or Gm-C Filters and will have to go through major revisions if I want to implement it in common centroid layout. However, my coworker told me that people don't use common centroid anymore as the matching is generally good, and of course it adds more parasitic capacitances.

Is this true that people don't use common centroid? I'm using TSMC 90nm. Should I fix everything to common centroid? Or is this technique becoming obsolete. I can still imagine couple circuits that will benefit from common centroid layout, such as comparators.. what else???


Let me know, I'm very curious to know how effective common centroids are nowadays.

 

[dick_freebird]

If you don't have much chance of significant thermal gradients
then CC isn't such a help. I think it's less popular in ultralow
power circuits (where additional capacitance will also swing
your bandwidth big time). With devices being so small nowadays
you can pretty much forget about process gradients. And the
gradients you do have, may be less than random fluctuation.

 

[palmeiras]

Hi Sharkies,

My point of view is that this issue is only a question of tradeoff: parasitic capacitances (your frequency specification), area and better matching. I mean, only if you have difficulty to achieve your frequency requirements, you should avoid the centroid common configuration.
I have seen many designs done using 65 nm, 90 or 135 nanometer CMOS technologies, and all of them uses common centroid layout configuration; for example: amplifiers, references, comparators and data converters.

While the transistors dimensions are scaling down, you know that the size of transistors in analog designs (at least part of them) is not directing following this tendency. For example, while using 90 nm tech., it is often using transistor length of 1, 2 or 3 um.

What I would do is to verify the variability of these circuits using Monte Carlo Analysis. This simulation ignores the layout techniques and shows the worst case performance (1, 2, 3 or 4 sigma requirements). So, if your circuits are working in those simulations, I would not be worry about changing those circuits now.

Regards,

 

[MSLayout]

Yes people still use common centroid. Matching is good, it depends on the degree of matching you require. Working on high voltage opamps, the degree required can be
stringent.

 

[EmilioSergio]

While the transistors dimensions are scaling down, you know that the size of transistors in analog designs (at least part of them) is not directing following this tendency. For example, while using 90 nm tech., it is often using transistor length of 1, 2 or 3 um.

Exactly, such dimensions are used even in 28nm tech analog circuits. I did lot of work in 28nm layout lately, surprisingly due to design rules in some 28nm tech analog layout have to be bigger than in 90nm for example (with quite similar effective transistor sizes).

And yes, common centroid is still widely used in analog circuits I'm working at and generally have impact on circuit operation.

 

[love_analog]

I believe at smaller process geometries 28nm etc. common-centroid is no longer that used. The layout complexity of having all these metals running around is too much. I have seen people use interdigitization and then put enough dummies to make sure influence distance is taken care of.
I would assume that in older geometries or where people need super-precision, common-centroid would still be the way to go.

---------- Post added at 23:35 ---------- Previous post was at 23:34 ----------

@Emillio. Have you data to share showing how much matching improved with and w/o common-centroid.

 

[EmilioSergio]

@love_analog:
I have not, I'm layout designer, I base on words of schematics designers, and I was correcting poorly matched circuits to get better results. I assume they know what they're saying.

 

[leo_o2]

Yes, advanced technology adopt some schemes to reduce the dependency on direction. One of them is to spin the wafer under processing. Therefore, common centroid can be waived. However, layout with common centroid is still a good practice if possible.

 

[wanghs1986]

it depends on the layouter, most of time if not dealt properly, comm-centroid just make the layout larger, more parasitic and no better matching