OK, I just saw this. For those who couldn't upload the paper, the topology appears to be similar to the old trick of putting a fine-bit resistor string that can be placed across two outputs of a coarse string to get the finer resolution. There appears to be some circuitry to offset the loading, but I didn't look at it closely.
First, I'd be wary of trying to cancel the loading... it may introduce more errors than the mismatch. I'd also be wary of the scheme in general... if it is between two resistor strings it may not be guaranteed monotonic... that said, it may also be a great idea... it's hard to say without diving into the circuit.
Matlab Monte Carlo is best for simulating INL of these types of DACs - spice usually takes too long because you need to check many codes. It is about impossible to get a closed form solution for INL (or even DNL) because it is the maximum error of all the codes. Bastos gives a brief discussion of the issue in his Dec 1998 JSSC paper. Steyart also wrote about the subject in a different paper. Read about what INL is in the Johns/Martin book.... that is what you'll want to simulate. Assume a particular yield (say 99%) and figure out how much area is required to achive that yield. Be aware that systematic errors from layout will produce an error on top of the mismatch error. Give yourself plenty of room... 99.9999% yield is a reasonable simulation goal because you aren't including layout errors.
FWIW, Baker's analysis is incorrect in his purple book.... Your dac would need to be a few acres in size to meet his spec. A rough rule of thumb is sigma(INL) = sqrt(2^N)*sigma(LSB), where N is the number of bits and sigma(LSB) is the standard devition of one code.
You will need mismatch data... low sheet (~30 ohm) poly resistors are poor... on the order of 20%-um (meaning the ratio of two poly resistors, each with an area of 1um^2, with have a SD of 20%). High sheet poly is better, and diffusion resistors match very well, but have voltage coefficients. Both high-sheet and diffusion have significant temp cos.... I think some of the newer text books may have information on mismatch.... In a nutshell, the mismatch decreases as the square-root of the area.
Regarding optimium size, a normal D/A wouldn't have an optimal size... bigger would always be better. However, your structure probably has an optimal area allocation between the coarse/fine resistor strings (as well as number bits resolved by each). Lin/Geiger's paper (March 2006, TCAS-I) may help. I published a paper on the subject too
Gregoire: 2004 CICC. However, I expect matlab simulations should be quick, and with a little common sense you can figure it out without knowing the calculus. If you figure it out, publish a tcas-ii paper so I don't have to derive the answer next time
.
I expect your INL will be determined by the coarse dac, so it will require much of the area.
Layout can only make things worse than simulation - poor layout probably won't affect the standard deviation much, but it can introduce large systematic errors... you need to make sure the layout is insensitive to gradients in temperature and process (sheet resistance). Make sure contact, via, and metal resistances don't introduce errors. Other than that, it is just a matter of making the resistors large enough (area wise) to meet your spec.
Vabzter wrote on Jan 15th, 2007, 12:17am:Hello everyone,
I am designing a 10bit DAC in 90nm technology. The topology is Inverted Resistive ladder DAC and I am refering the IEEE paper "A Low Power Inverted Ladder D/A Converter" by Yevgeny Perelman and Ran Ginosar" for the design.I have the following general design questions:
1. How to find out the effects of resistor mismatch in DAC. Which simulations do I perform?
2. Is there and way to find an optimum value of resistances so that the mismatch will be less.
3. Can the mismatch be controlled in the layout only? Or is there and steps I need to so in the design of ladde DAC.
I am new to design so any help would be appreciated..
Thanks a lot in advance,
BR,
Vabzter