The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Design >> Analog Design >> L effect on mismatch
https://designers-guide.org/forum/YaBB.pl?num=1089243205

Message started by modern_analog on Jul 7^th, 2004, 4:33pm

Title: L effect on mismatch
Post by modern_analog on Jul 7^th, 2004, 4:33pm

This might sound very basic but here goes

The mismatch depends on WL, or more accurately 1/sqrt(WL). Then why do we need large L. It seems we can get away with small L and large W.

Title: Re: L effect on mismatch
Post by Edward on Jul 16^th, 2004, 10:57am

sometimes, in order to get the desired gain, we use larger L.

then, about mismatch, the larger the transistor is, the smaller the mismatch percentage.

Title: Re: L effect on mismatch
Post by Humungus on Aug 24^th, 2004, 5:52am

Also, L longer than minimum reduces leakage current. For both W and L greater than minumim values, the etch effect is smaller than when minimum values are used

Title: Re: L effect on mismatch
Post by analognut on Sep 12^th, 2004, 6:14pm

First , to minimize errors due to the side-diffusion of the source and drain areas (Ld)
Second, to minmize errors due to the threshold voltage of short-channel devices which exhibits some dependence on the channel length.

above lines are quoted from the book written by razavi

Title: Re: L effect on mismatch
Post by raul on Mar 19^th, 2006, 9:22pm

mismatch also depends on Vgst(Vgs-Vt). Large W and small L will give you a low Vgst which will give you a high gm and the error voltage at the gate of the transistor is multiplied by the gm of the transistor to give the error drain current. Therefore you want to have a large Vgst in order to get a low gm so that the error is multiplied by a smaller number.
bottomline, mismatch has two components; one is size and the other is bias point(gm/Vgst).

Title: Re: L effect on mismatch
Post by RobG on Mar 20^th, 2006, 5:05pm

Expanding on what Raul said... if you do the analysis for a current mirror in strong inversion you will find that the mismatch [of the current, (edit)] is actually independent of width. This is because the Gm of the transistor is increasing as sqrt(W) just as the Vt mismatch is decreasing by sqrt(W). Since the current error dI/I is gm*dVt the effects cancel. So, for a mirror, you want to use as long a channel as speed (and headroom) will allow.

On the other hand, if you have a diff pair (like an opamp diff pair), you generally want to make it as wide as is practical (keeping W*L constant) so that the Gm is as large as possible. This doesn't change the diff pair's contribution, but it decreases the contribution of the mirror because the diff pair mismatch is multiplied by Gm(diff)/Gm(mirror).

So... sometime you want it long, some times you don't. The same analysis holds for 1/f noise...

rg

Title: Re: L effect on mismatch
Post by sheldon on Mar 22^nd, 2006, 4:33am

Modern_analog,

You might also want to read some papers in addition to
Pelgrom's,

"Understanding MOSFET Mismatch for Analog Design",
P. Drennan and C. McAndrew, JSSC, March 2003

"Measurement and Modeling of MOS Transistor Current
Mismatch in Analog IC’s", E. Felt, et al, ICCAD, 1994

The issue of mismatch modeling is complex than mismatch
is proportional to sqrt(W*L), as Rob mentions you also need
to consider the operating point.

This is certainly a useful rule of thumb for designers to keep
in mind. However, it includes some assumptions that designers
need to understand and they should not blindly extrapolate
based on it.

Best Regards,

Sheldon