Not sure what's going on with the second circuit.  You can build a single transistor error amplifier to accomplish the same with the attached ckt.

T2 = Cascode
T3 = error amplifier

I am not completely sure, but the circuit of Steve has the problem that T3 has to  be operated in subthreshold region if vds of T1 is not sufficiently high.

T4 might circumvent the problem by providing a levelshift. However, the bias current through T4 must be small compared to the main branch.

Cheers

Berti wrote on Aug 3^rd, 2009, 12:00am:


Sorry for dominating these conversations, but I hope I'm being helpful.

The circuit on the right has been used for a long time. It is a wide swing regulated current source. As you guessed, M4 provides level shifting so that M1 has the minimal voltage across it without going linear. The current does not have to be small, but M1 must be sized to handle it. In fact, if the current is too small it will reduce the phase margin of the gain boosting circuit.

rg

raja.cedt wrote on Aug 3^rd, 2009, 2:01pm:

M4, M3, and M2 create a local feedback loop. M4 just acts as a level shifter, but it's impedance (1/gm) will react with the gate-source cap of M3 and introduce a pole. Therefore you want the current through M4 to be large enough so that the pole doesn't affect the phase margin of this local feedback loop.

AnalogDE wrote on Aug 11^th, 2009, 1:48pm:


Unfortunately the source follower would need to be a PMOS its Vt would not track with the NMOS amp (M3) Vt. This is needed to provide the proper Vds  for M1. The only way I've been able to do it without the diode connected device is to use a PMOS diff pair opamp.