loose-electron wrote on Mar 26th, 2014, 11:20am:He needs some basic books and classes in the subject first.
Hell, this place would implode if everyone did that. ;)
I don't know why books "calculate" W/L. Mostly what you need to know is that gm =~ 2I/(Vgs-Vt) in strong inversion. In weak inversion gm becomes largely independent of Vgs-Vt so maximum gm will be ~I/75mV.
This is an approximation, but now you have gm in terms of I and Vgs-Vt, which are the only physical design variables that you can change. I is limited by power, and Vgs-Vt is limited to ~150mV for the diff pair and by headroom requirements for the loads and cascodes.
UNITY GAIN Bandwidth is =gm/Cload/2pi =~ so maximum bandwidth is (I/75mV)/Cload/2pi.
From this you can determine how much current you need to achieve your desired UNITY GAIN bandwidth for a given capacitive load. If you are close you can see if you can get enough gm in weak inversion (realizing your input devices are going to get very wide and increase the input capacitance), but if you are off by a factor of two you either have to relax your power requirements and increase I, or decrease your load capacitance. You don't have any other options.
I emphasized UNITY GAIN because if you are configured for a higher gain your bandwidth will be scaled by your feedback factor.
I'm sure there is a paper or book out there that covers this way of designing but I'm not aware of it.
rg