What are the methods for laying out differential pair/node in high speed application?

[sxunxs]

Hello!
what are the techniques to consider in laying out differential pair/differential nodes in a high speed application?
any idea?

thanks and regards,
sxunxs

 

[DZC]

differntial pair

It's all about matching.
Keep everything:size,environment,wiring...as same as possible.
Other techniques including using unit component,split large transistors,dummy,etc

 

[sxunxs]

Re: differntial pair

yeah u are right.. but how about the differential nodes? how can i reduce the parasitic capacitance between them? putting extra lines between them may do, what else?

 

[zhangsai04]

differntial pair

the 2 input signal lines should be as short as possible

 

[mdcui]

differntial pair

big spacing between differential lines are also appreciated, watch the adjacent metals on the same layers as well as the on different layers, shield them if necessary.

 

[williamy]

differntial pair

when frequency is very hign, such as RF , which is more important between parasitic and match?

 

[analayout]

Re: differntial pair

hi
i think in high freequencies like RF paracetic is important
regards

analayout

 

[sxunxs]

Re: differntial pair

big spacing between differential lines are also appreciated, watch the adjacent metals on the same layers as well as the on different layers, shield them if necessary.

if big spacing between diff lines are drawn, noise that might be injected to one line is different from the other one. how could this be avoided?

 

[srieda]

Re: differntial pair

big spacing between differential lines are also appreciated, watch the adjacent metals on the same layers as well as the on different layers, shield them if necessary.

Differential lines MUST be laid out as close as possible... (So that noise affects them in the same way...)
The paths must be very symmetrical and as close as possible... right from the pads to the amplifier inputs. (So that there is no phase difference introduced between the differential signals)
Avoid long routing if possible (obviously signal strength.)

I am a designer and I always insist on lines to be close to each other as possible for the layout guys.. Please correct me if my understanding is wrong!!

 

[sxunxs]

Re: differntial pair

big spacing between differential lines are also appreciated, watch the adjacent metals on the same layers as well as the on different layers, shield them if necessary.

Differential lines MUST be laid out as close as possible... (So that noise affects them in the same way...)
The paths must be very symmetrical and as close as possible... right from the pads to the amplifier inputs. (So that there is no phase difference introduced between the differential signals)
Avoid long routing if possible (obviously signal strength.)

I am a designer and I always insist on lines to be close to each other as possible for the layout guys.. Please correct me if my understanding is wrong!!

thanks.. you are right.. but when the 2 diff lines are close to each other, parasitic capacitance would be seen bet the lines.. do u know any technique to avoid this parasitic except to put a dummy path/line between them? if a had to use different metals in routing the diff lines, diff noise would be injected to them. correct me if im wrong.

 

[forkschgrad]

Re: differntial pair

will that have a big effect? i guess it may not cause a big effect on the whole design as long as those paths are parallel and differential impedance is achieved.

 

[srieda]

Re: differntial pair

thanks.. you are right.. but when the 2 diff lines are close to each other, parasitic capacitance would be seen bet the lines.. do u know any technique to avoid this parasitic except to put a dummy path/line between them? if a had to use different metals in routing the diff lines, diff noise would be injected to them. correct me if im wrong.

interesting point... but I dont think that'll matter much in terms of loading... you are right in that you should not use different metals for routing diff lines...

In fact I have another way of interpreting this.. :idea: The coupling capacitances between these might help to couple the noise between the two lines which is what we want...!!! Thus from this angle, these coupling capacitances are useful!!

Comments and arguments are invited...

 

[sigui777]

Re: differntial pair

Diff. inputs of pair should be close to each other so that noise becomes common to both lines and then rejected since common mode is rejected in diff. pairs.

 

[sxunxs]

Re: differntial pair

thanks.. you are right.. but when the 2 diff lines are close to each other, parasitic capacitance would be seen bet the lines.. do u know any technique to avoid this parasitic except to put a dummy path/line between them? if a had to use different metals in routing the diff lines, diff noise would be injected to them. correct me if im wrong.

interesting point... but I dont think that'll matter much in terms of loading... you are right in that you should not use different metals for routing diff lines...

In fact I have another way of interpreting this.. :idea: The coupling capacitances between these might help to couple the noise between the two lines which is what we want...!!! Thus from this angle, these coupling capacitances are useful!!

Comments and arguments are invited...

yeah u might be right... but this diff amp is used in high speed applications... this capacitance ( i think ) is not beneficial since it has an effect to the speed of the system. correct me if im wrong again... ur comments will also be anticipated and appreciated.. thanks

Added after 3 minutes:

will that have a big effect? i guess it may not cause a big effect on the whole design as long as those paths are parallel and differential impedance is achieved.

yeah it has a big effect.... parasitic capacitance bet the diff lines deteriorates the speed of the system considering that it is used in high speed applications.


comments and reactions are welcome to this topic...

 

[srieda]

Re: differntial pair

@sxunxs

my point is the transmitter is usually designed to drive a load cap of at least 12pF. The coupling capacitances might add up to max a few fF. Thus loading as such is not such a big problem.
Yes, to some extent the speed gets affected. But trust me, this is much better than skew between the lines... skew will mean that the effective eye is very much constricted and this is very bad news when you are trying to sample the recovered signals.

 

[road_0506]

Re: differntial pair

try to route the diff pairs with same metal and as closed as possbile. put dummy path/metal bet the pairs.

 

[protonixs]

Re: differntial pair

does anybody have any other idea about this topic?

 

[Brittoo]

Re: differntial pair

does anybody have any other idea about this topic?

Hi protonixs

I would suggest you to shield the inputs to the differential pair. That would avoid coupling of other signals with the inputs thereby changing the functionality.
Use common centroid matching, add dummies with continuous diffusion,enclose this differential pair in a different guardring. Place the outputs far away from the i/p to avoid coupling between them, take care of symmetry( for routings and contacts as well), keep the i/ps close together(with shielding).

Regards
Brittoo

 

[mdcui]

differntial pair

I was suggesting an approporate distance between differential signal routings just to reduce the line2line cap induced loading, if you want to minimize the loading of your routing of differential signal, putting them as minimum space sometimes is not the best choice.

 

[cmos_dude]

differntial pair

The idea of capacitive coupling as suggested by srieda might be correct as the common mode noise gets rejected.
But can anybody explain in detail, how putting dummy metal between the pair helps ?

Thanks

--cmos_dude