You should plot Vth vs. L to get an idea of how much Vth is varying with L. If you care about mismatch, pick an L that will give a fairly constant Vth. For short-channel devices, Vth is fairly sensitive with min L.
A larger Vgs is going to help reduce the Id variation as Vth varies. Also, a larger overdrive is going to give you higher speed, and will help compensate for the reduction in speed caused by a non-min L. One thing to watch for is increasing W/L, thinking you are going to get a speed increase. If you keep Vgs constant, current goes up, but so does capacitance, so you waste power. And, if you keep Id constant, Vgs has to drop, lowering speed and making mismatch worse.
For short-channel devices, I'd say that the variations in Vds are pretty important for matching. Even if you have matched Vgs and Vth and W/L, that's not going to help much for short-channel devices with very different Vds. Of course, longer L will decrease channel-length modulation and Vds mismatches won't matter as much.
In summary:
large L
larger Vgs
match Vds
Marc
bharat wrote on Nov 12th, 2006, 9:54pm:Also I have deep submicron process narrow channel length so don't know the second and third order of effects resulting this mismatch like Vth variation with device dimension.
This is the background of my question. Now the question is, if I have 2 same designs with different dimensions and I want the same performance, what should be matched in first order assumption. Is it current, gm, Vgs Vds ??
Thanks
-Bharat