[MOVED]better resistor matching

[Dinamovist]

Hi Gentlemen,

I have a question that I found no answer yet.
In a normal bicmos process I want to match 2 poly resistors let's say better than 0.05%. According to the process data is almost impossible to get there only by increasing the size (not taking into consideration second order effects). It
So without drawing them impossibly big, and without trimming is there any method to get better matching?
Let's say second order matching between pairs ?
I'm open to experiment anything. It is just for a simple resistor divider but must be very accurate. What is your experience with this sort of circuits ?

Thank you.

 

[keith1200rs]

0.05% is pretty tight. Are you using the matching parameters for the process resistors, not the absolute dimensional tolerances? Some processes don't give clear matching information, just absolute tolerance information which isn't the same.

Keith.

 

[Dinamovist]

Hi Keith,

Yes I'm using the matching parameters not the W/L tolerances.

Thank you

 

[keith1200rs]

What sort of W/L are you coming up with? Assuming you want 3SD = 0.05% then I calculate WL product as 105000um for a typical 0.35um process. That is obviously a large resistor - say 1050um long by 100um wide depending on how many squares you want! However, it depends on whether poly is the best for matching. For the process I looked at N-well is far better and would come out with WL product as 8800 so a more reasonable 880um x 10um for example. Maybe Erikl will advise later.

Keith.

 

[Dinamovist]

I did look at well resistors and indeed they have better matching but worse voltage effect. For poly, I got with huge numbers similar with yours and that is not feasible for my project. I'm looking for another method and if nothing works I'll use trimming.

Thanks

What sort of W/L are you coming up with? Assuming you want 3SD = 0.05% then I calculate WL product as 105000um for a typical 0.35um process. That is obviously a large resistor - say 1050um long by 100um wide depending on how many squares you want! However, it depends on whether poly is the best for matching. For the process I looked at N-well is far better and would come out with WL product as 8800 so a more reasonable 880um x 10um for example. Maybe Erikl will advise later.

Keith.

 

[erikl]

I suppose you've already considered interdigitizing, common centroid layout and side dummy usage (s. PDFs below).

And perhaps this or that thread may be interesting.

 

[Dinamovist]

Thank you sir,
Right now my concern is just to calculate the best matching possible just as numbers. The layout will be a later concern. Right now it looks like using fab's matching data is hard to get where I want.

I suppose you've already considered interdigitizing, common centroid layout and side dummy usage (s. PDFs below).

And perhaps this or that thread may be interesting.

 

[JoannesPaulus]

What is the resistance you are trying to obtain and why do you need such an accuracy? Have you considered automatic calibration using a mos in triode region instead of trimming?

 

[Dinamovist]

The resistance is around 250k. I'm building multiple references.

That automatic calibration using the triode region sounds quite interesting. Could you give me more details ?

Thank you sir,

What is the resistance you are trying to obtain and why do you need such an accuracy? Have you considered automatic calibration using a mos in triode region instead of trimming?