MOSFET W size can be lower than his L?

[AMSA84]

Hi guys. I have to design a comparator operating at high frequency (500MHz)
The topology is based on the smetrical ota with crosscoupled load. From rom what I have been exploring I concluded that for the cross coupled MOSFET the value of the W has to be lower than the L, that is, for example, 0.12/0.34. The transistors that I AM using are the 3.3V.

I want to know if someone has any opinion on this, there is any problem to have transistors with the W < L. There are any especial considerations that onde should be aware of?

Any additional comment and opinions are welcome


Thanks in advance.

 

[dick_freebird]

No real fundamental problem, it's been done since people
have been doing CMOS analog. You may find some loss of
matching qualities, especially when you get to where the
narrow W means a dogbone active area (hard to control
the patterning). The narrowest rectangular AA that can
hold a contact and overlaps DRC-clean, may be where you
want to be.

 

[AMSA84]

Okay then.

Thats the only way I think I can get the comparator comparing at that frequency, making those W as small as 0.12um
This way the capacitances are lower and the comparator becomes more fast.

Any other recomendation?

Regards.

 

[dick_freebird]

Making the device small is a step toward maximizing current
density (current vs parasitic capacitance) which is generally
a good thing. But it is not the only thing.

Once other layout parasitics are comprehended, you may find
that larger devices run at higher currents is the way to "bury"
the interconnect and driven-device loading (which are "fixed
losses" more or less). You should put this burden on your
design as early as possible, so you don't optimize yourself
into a perfect (in isolation), but uselessly weak (in application)
design result.

Getting more gain up front is key to high speed operation
at low overdrive levels. The variation of delay with overdrive
can be quite significant in low-stage-count topologies. More
differential high bandwidth stages, there.

Your cross-coupled load is good for higher DC gain (local
regenerative feedback) but may give you a stage-gain-node
impedance which, with local parasitic C, has too low a BW
to meet your goals. Your front end might want to look like
a low noise differential RF amplifier lineup, more than a
classical DC precision comparator (which the back end might
still resemble, somewhat).

Now, the only design value you have articulated is speed.
Of course everybody wants nil power and area and offset,
and infinite gain and bandwidth. But you owe it to yourself
to make more of these dimensions explicit, and see whether
they can in fact be met at once in a relevant application
setup. Point being that this is going to drive you into certain
topology and bias choices, and so should get figured out
early on.

 

[AMSA84]

Thanks Foé the answer.

From what I can understand from what you said is that you are suggesting to increase current in order to increase gain to achieve higher speed and don't opt for small size devices because of the parasítica. Is that it?

Thank you again in advance. Happy new year