I think checkmate pretty much said everything I was going to add. Summarizing:
- The NM14/NM17/PM11/PM12 branch is not really needed, you can tap off the biasing voltages for PM3/PM4 right from PM9/PM10 and for PM7/PM8 from PM13/PM14, just make sure you have right ratios of W/L to get the current values you want.
- Vds of NM16 basically depends on the voltage drop across the resistor, the current you run through the transistor and its W/L value. What you need to do is size either the transistor's W/L or the resistor (or both) in such a way that you get the desired value of Vdsat for the current your running through (write down the equations for that branch and solve, it should be pretty straight forward).
- The margin between vds and vdsat depends on how critical the transistor is through out the signal path. Since this transistor (NM16) is just part of the biasing branch is not that critical because the voltages across its nodes are fairly constant (DC values). Now, you should also keep in mind that leaving vds close to vdsat makes the transistor more vulnerable to any other changes in the circuit, like supply variation, vth variation, etc...
- In practice it is going to be very very difficult to adjust the output value to exactly VDD/2 but usually it's a good idea to leave as close to it as possible to maximize swing. At the end, your DC common mode voltage at the output will be defined by the feedback you apply to the amplifier, but setting its open loop DC value close to VDD/2 is a good practice.
diemilio