Question about resistor matching and current flow

[norayra]

Hi experts,

I have a question about resistor mismatch.

If we have a unit element R and we want to get 2R (or du better match), is it required that current flow of matching devices should be the same. Please, refer to the attached pic.

Thanks in advance...

 

[Colbhaidh]

silicide block resistor

For best match, current flow should be in the same direction. This may seem like overkill however there are a few reasons why this is the case.
The largest contribution to resistor mis match is not L or W variations especially in close proximity. Even the lack of dummy resistors either side of an array can have secondary effects to interface resistance. In technologies that use poly resistors with silicide end caps (always the case) and silicide block over the body of the resistor, there is an interface resistance between the silicided end cap and the unsilicided poly of resistor body and this becomes significant. Mis-align between the silicide block and one or other side of the resistor will always be in one direction. This changes the effective interface resistance left vs right or whatever. In addition to this, there is also an implant mask between the silicide block and the end cap in the direction of the contacts. This also causes a mismatch one side vs the other. Both of these can act against one another to enhance the mis-match. In addition to this, even if the heavy implant on the end caps is done at zero degrees, no implanter is exactly zero degrees. So there is an additional mis-match.
(This is also true of current mirrors or matched xtors that look to be optimally matched in layout except that current flow S to D is opposite for matched pairs. This is bad practice no matter what text books say.)

However this is a trade off. The layout of your resistor stack will be much greater if you use current flow in same direction.
You must never overlap metal over one resistor and not its neighbour as this may be worse still. So the metal routing must lie between resistors increasing the space between them, unless all are covered by the same metal layer.

So you must ask yourself just how good the matching has to be given these compromises.
If all resistors are 2R and all layout is the same then it should not matter as the mismatch should be the same across all matched resistors.

So you should try to think about - not how you reduce the mis-match to zero or minimise it, but how you make all the mis-matches the same such that they have the same effect on each matched component. This is the true skill of the layout Engineer.

 

[erikl]

matching current should run in same direction

... Mis-align between the silicide block and one or other side of the resistor will always be in one direction. This changes the effective interface resistance left vs right or whatever. In addition to this, there is also an implant mask between the silicide block and the end cap in the direction of the contacts. This also causes a mismatch one side vs the other. Both of these can act against one another to enhance the mis-match. In addition to this, even if the heavy implant on the end caps is done at zero degrees, no implanter is exactly zero degrees. So there is an additional mis-match. ...

Very true - but what has this to do with the current direction? There will be an overall mismatch between these resistors, but it's still a resistor, and the direction of current flow IMHO shouldn't matter. A resistor is a resistor is a resistor ... Or did I miss another physical effect, may be of 2nd order?

... This is also true of current mirrors or matched xtors that look to be optimally matched in layout except that current flow S to D is opposite for matched pairs...

I don't understand this, sorry. Why should the current flow from S to D be opposite for matched pairs?

 

[Colbhaidh]

resistor match mask

You must consider the poly resistor not as a uniform sheet resistance but as discrete grains and boundaries, that when considered as a whole, share the same behaviour as a continuous sheet resistance. This is OK and in general does not cause any issues even where matching is concerned.
When we use a silicide process a couple of spanners are thrown in.
(1) The end caps either side of the resistor are low resistive silicided poly in series with the bulk unsilicided poly.
(2) Between the silicide and the high resistance poly is a interface resistance which is dependent on the silicide grain structure, the active dopant loss at the silicide/poly silicon interface due to the silicide processing (dopant boiled off or dopant deactivated by piling up at the interface), and the active dopant.
(3) The silicide block process creates more alignment issues.

Problem is all 3 of these effects may differ from one side to the other because of the reasons in my 1st post. But that should not matter, it is a resistor after all?
Well not quite. There is an effect of the silicide grains and the polysilicon grains (and indeed poly-Si grain to poly-Si grain). A voltage drop across a grain and its boundary can disturb the symmetry of near boundary depletion regions causing barrier resistance changes. Problem is the silicide interface on one side will always be different on the other side so there is a directional effect. Locally all the "same" sides will be very similar or more similar that one side to other.
This effect may be too small to matter but it is real. It is worse for Titanium silicide than it is for Cobalt silicide as Titanium disturbs the dopant at the silicide-silicon interface more than cobalt does. It also affects the temperature coefficients.

This may be second order but can cause problems in Flash ADCs where the THD contribution from INL, DNL is sensitive to the resistor network.


The second point is where, for example in a current mirror LTP, the two matched transistors are drawn with common source and mirror each other. They are drawn matched in every respect except orientation and fail because of this. Of course using 2 transistors per side crossed diagonally would circumvent this but I still come qcross designs with this violation more often than the latter. Even after 20 years of screaming !!

 

[erikl]

resistor matching in layout

Thank you for your extensive explanation, Colbhaidh!

...
Problem is all 3 of these effects may differ from one side to the other because of the reasons in my 1st post. But that should not matter, it is a resistor after all?
Well not quite. There is an effect of the silicide grains and the polysilicon grains (and indeed poly-Si grain to poly-Si grain). A voltage drop across a grain and its boundary can disturb the symmetry of near boundary depletion regions causing barrier resistance changes. Problem is the silicide interface on one side will always be different on the other side so there is a directional effect. Locally all the "same" sides will be very similar or more similar that one side to other.

If I understand you correctly, the various participating resistances could be different depending on the direction of the current flow.

...
This may be second order but can cause problems in Flash ADCs where the THD contribution from INL, DNL is sensitive to the resistor network.

AFAIK the poly resistor contact resistances are in the order of 1 to several Ohms, so the differences between them should be at least an order of magnitude lower. For practical reason this would mean, that resistors ≥ 1kOhms, that are layouted to mismatch < 0.1% (for a 10bit converter) shouldn't be adversely affected by the current direction.

The second point is where, for example in a current mirror LTP, the two matched transistors are drawn with common source and mirror each other. They are drawn matched in every respect except orientation and fail because of this.

Of course they shouldn't be mirrored, but repeated. This is good and well known layout practice, and so they also have the same orientation. Of course mixing (interdigitation) - keeping the repeating pattern - and the use of side dummies is still a better solution against mismatch.