BOE,

PSRR and matching of M0/M1 is present in both the cases 1,2.

what is your intention?

thanks,
yvkrishna

yvkrishna wrote on Oct 16^th, 2012, 2:55am:


yvkrishna,
I've done this many time and it works fine. Like you figured out, you have to account for the TC of the resistor, but this is true even if you use a voltage to generate the current (your voltage TC has to be the same as the resistor TC.

The first consideration is that it is difficult to trim the TC with a single measurement. If you were doing a voltage-mode reference you can room temperature trim to a "magic" voltage that will give you zero TC for all errors except opamp offset. After that you would trim out the error due to the sheet resistance in your V to I circuit.

The second consideration is that with a current reference you would be trimming the magnitude and TC of the current simultaneously. While you can measure the current directly and trim it to spec, the resulting temp co will depend on how much of the error is due to sheet resistance variation or opamp offset, mismatch and bipolar IS variation. One way to get around this problem with the circuit you showed is first trimming the voltage across R20 (e.g. by varying the ratio of M1 and M0) to the "magic" voltage, and then trimming the output current by sizing M22.

Hint - when trimming don't actually change M0/M1/M22 sizes, but instead put a current dac in parallel with them that is biased with a smaller current scaled from the main current.

Take care that you have adequate trim range both ways(ctat slope as well as ptat slope,both abs trim as well as tempco trim).Bandgaps look great in simulation,but model inaccuracies can  make it tough.Also take care to have metal options in case you need to fib and do some testing before you decide to fix a problem.

Yes, it is perfectly feasible to generate a temperature independent current without and bandgap voltage appearing anywhere in the circuit, despite some of the claims in this thread. But why would this need to be done? All real systems need a reference voltage, so might as well use it! On a technical side, there are techniques in BG design for getting very low noise, that would probable not be mimicable in current mode. Regarding TC, 1st order TC of the resister is not important as it can be taken up in the wash when setting the current cancelling TCs. There is actually a simulation example for the current summing bandgap technique in SuperSpice.