in the circuit shown below, the output cascode stage of the folded cascode amplifier has two miror, one up and one down, and there is no current source, how the current is then being defined to 2 Ib? we usually have in the output cascode one mirror and the other side is current source
By W/L ratio (=1) between (P3,P4)P1,P2) and (N3,N4)N1,N2) resp. Don't forget: V2 & V3 must be supplied externally. Also, the common drain voltage of P5,N5 (≈VDD/2) supplies all the gates of N3,N4 & P3,P4, hence this design isn't suited for high VDD voltages, cf. Fig. 3 of the attached document.
By W/L ratio (=1) between (P3,P4)P1,P2) and (N3,N4)N1,N2) resp. Don't forget: V2 & V3 must be supplied externally. Also, the common drain voltage of P5,N5 (≈VDD/2) supplies all the gates of N3,N4 & P3,P4, hence this design isn't suited for high VDD voltages, cf. Fig. 3 of the attached document.
Do you practically prefer to deal with such circuit ? or to use the traditional folded one
my second question, does it worth to have a constant gm amplifier? because it makes the design more complicated and consumes more power, or you think it is not that mandatory
Depends. I used it for supply voltages of 1.0±0.2V and it worked well, as the Vth voltages of the process suited fine. With higher VDD this simple (but admittedly very elegant, because of its perfect symmetry) topology probably isn't well suited, because the circuit unnecessarily will end up with too much current consumption. Actually you should have a process with Vthp+Vthn ≈ VDD .
does it worth to have a constant gm amplifier? because it makes the design more complicated and consumes more power, or you think it is not that mandatory
Again depends. If you are bound to get rather flat open loop gain over your ICMR, you need constant gm architecture. If the application allows for scaling down the gain considerably by neg. feedback (closed loop gain « open loop gain), you can achieve the necessary gain-over-ICMR flatness by this feedback. In this case you don't need constant gm topology.