RobG wrote on Sep 12th, 2011, 10:36am:loose-electron wrote on Sep 12th, 2011, 10:22am:The john-martin structure with the op-amps will have problems at 12 bits due to op-amp offsets.
You had me going for a second, but looking closely at the circuit I don't think it is a problem. He uses the same opamp for each tap. You'd get INL, but it will be monotonic as long as the offset is less than the first dac LSB (6 bits in my example).
However, if he simply walked the opamps down the first resistor string so the "top" opamp was always used for the "topmost" tap the opamp offset would have to be less than 12 bits.
rg
Hmmm, maybe we are talking about different circuits - I am thinking of the one where you mux the LSB on an R-string and buffer it with a pair of op-amps
That thing will have Offset, and as you get up into 12 bit land you will start have more subtle stuff play in like offsets that change slightly due to the common mode voltage.
Thing is the OP is asking for a lot - 12 bits with no alignement and calibration is not easy. The oversampling world takes a lot of the problem out of analog issues and makes it a more robust duty cycle set of problems.
But, they don't want a clock and they want a fairly quick response characterisitic.
Have not even asked the question if R-2R can settle out in the desired time, and if you can get good enough geometric matching.
Opinion? Anything beyond 8 bits should be calibrated. Some could argue 10 bits, and others would argue 6.