From paper/text, it usually needs a compensation cap and a nulling resistor(normally realized by transistor biased in linear region) in two-stage op amp. The usual placement is as follows:
output of first stage---cap---resistor---output of second stage
could we switch the cap and the resistor like the below?
output of first stage---resistor---cap---output of second stage
If you use a real resistor (NOT a MOS), I think that the results should be EXACTLY the same.
When you use a MOS transistor in triode the Rds depends on the biasing conditions -> putting it on the 1st stage side or in the second stage side may lead to different Rds. But as long as you tune the MOS dimensions to have the Rds right value, it should be indifferent.
I think 2nd way is better
Because Resistors are more sensitive to node voltage than Cap, and the voltage swing at output node of 1st stage is smaller than that of 2nd stage. Besides, many designs place the resistor at the output node of 1st stage
I think ablue said is very correct, because add the null resistor is to move the RHZ to very large point or RLZ, which make the PM increase. It's principle is make the zero from Gm/Cc to 1/((Rz-1/Gm)Cc), so select Rz larger or equal the 1/Gm to increase the PM. If use MOS as resistor, then it need guarntee the MOS risistor larger than 1/Gm. so the 2nd mos is more stable to get a stable resistor.
ablue said:
I think 2nd way is better
Because Resistors are more sensitive to node voltage than Cap, and the voltage swing at output node of 1st stage is smaller than that of 2nd stage. Besides, many designs place the resistor at the output node of 1st stage
My another question:
If I use real resistor instead of MOS resistor, are the two approches exactly same?
I guess maybe it is better to use the first approach in this case.
If you want to move the zero to a large, then need make the resistor equal to 1/Gm, but since the resistor affected by Process, voltage and temperature, so not guarantee the equal, and use the mos as resistor may guarentee the same variable with the second Gm, so if possible, i think the mos is better.
jordan76 said:
rambus_ddr,
Thanks for your detailed analysis!
My another question:
If I use real resistor instead of MOS resistor, are the two approches exactly same?
I guess maybe it is better to use the first approach in this case.
There is also another point of view to the question where to put the MOS resistor. The zero compensating resistor should be either equal to 1/Gm if we want to move the zero to + infinity or larger than 1/Gm if we need to move the zero in the LHP and make it a good zero. Here Gm is the transconductance of the second stage. However, with process, temperature and weather conditions the value of the MOS resistor will move or 1/Gm will move and it may happen that we don't compensate it as we intended. Because the Rz should be with some relation to 1/Gm and also because to first order Ron of MOS (when transistor is in triode) is equal to 1/Gm (if this same transistor is in saturation), we need to make the size of the driving transistor of the second stage correlated to the size of the MOS transistor playing the role of Rz. This also means we need to provide the same Vgs-Vt for both. In turn this means that if we provide a bias for the gate of the Rz- MOS transistor equal to 2Vgs of the driving transistor and connect the Rz MOST at the output of the first stage (which is the gate of the driving transistor of the second stage), this will guarantee that both transistors work with the same Vgs and also track each other.
I think if you use resistor other than MOS, the second approach also better since many resistor depends on biasing conditions (e.g. diffusion resistor, and even poly resistor has some very small dependency). Connect the resistor at the output will produce some nonlinearity if they are depends on biasing condition.