Maks wrote on Jan 31st, 2010, 12:48pm:Many precision analog circuits rely on perfect capacitance matching or weighting (i.e. binary weighting). However, even "good" layout techniques do not guarantee perfect matching. Remaining parasitic coupling (at the level of 1%, 0.1%, or 0.01% of the total net capacitance) may be difficult to eliminate, but it may lead to deterioration of circuit performance (linearity, offset, etc.).
Are there any common practical approaches to resolve minute parasitic coupling and to eliminate them? Are there any methodologies, or existing CAD solutions?
There are many design methodologies for making your circuit immune to such imperfections, which primarily find application in data converter circuits where such imperfections manifest themselves in form of nonlinear distortion. Broadly speaking, the following types of methods exist (I am using ADCs as an example class of circuits but similar methods should exist for all/most kinds of circuits):
1. Architectural choices: At a higher level of design, you may choose a scheme which is inherently insensitive to mismatch or where mismatch will give you a more benign error such as offset, gain error instead of nonlinear distortion, e.g. using a single-bit delta-sigma ADC instead of an SAR for realizing higher accuracy.
2. Error cancellation schemes: At a slightly lower level of design you use schemes where error balancing or error cancelling is performed. There are several such schemes for SAR type ADCs where a single amplification process is performed multiple times in order to generate all possible mismatch terms and then combine them to cancel out the errors, e.g. see the paper below and the references contained therein
http://web.engr.oregonstate.edu/~moon/research/files/iscas99_sar.pdf3. Error calibration: The mismatch-induced errors are measured in foreground or background and a correction term applied. There are dozens of examples of this. Just look in the literature.
4. Error randomization: Using dithering or rotating and averaging usually allows improvements in the spectral characteristics of the error, decoupling it significantly from the signal.
5. Designing your circuit in a manner that it operates at the sweet spot where it is least sensitive to mismatch. This point while obvious is the one most likely to be missing in the run-up to most designs. Circuits which are sensitive to mismatch usually are not equally sensitive to mismatch under all conditions. You need to find the best conditions.
If all else fails, then you maybe need to ask yourself why the excessively high levels of matching are needed. Perhaps this should be asked at the beginning.
Vivek