When Monte Carlo analysis shows device mismatch variation has become problematic, Virtuoso Analog Design Environment (ADE) GXL Mismatch Contribution Analysis can provide useful diagnostics as a next step. Mismatch Contribution helps in identifying the most important contributors to the variance of the outputs. You can also compare the relative importance of the contributing instances. The analysis results can aid in making design changes to reduce the variation. Mismatch Contribution is a variance-based global sensitivity analysis .
Mismatch Contribution is launched from Monte Carlo results.
Here is a flat view of the outputs, and mismatch parameters are displayed and sorted by the swing specification.
Each device instance may be modeled with multiple statistical mismatch parameters. The parameter names themselves are not always of interest. In some cases the PDK models are derived from principal components. Mismatch Contribution provides a hierarchical view where the total contribution of all of the device parameters is displayed for each instance. The hierarchical view reports the contributions by instance for quick identification of important instances.
Cross probing to the schematic is provided. The schematic is opened to the same level of hierarchy, and the selected instances are highlighted. Navigate the table as you would a schematic. Descend into a row (instance) of the table until reaching the leaf level. The leaf level again displays the individual mismatch parameters of the instance.
By contrast, a top-level view with four blocks shows the block contributing the most variation of the specification. Descend to find the lower-level contributors.
When global process variation is applied during the Monte Carlo analysis, the contributions from the process parameters are included in the contribution analysis.
Mismatch Contribution is not limited to linear effects. When a linear model of the data is insufficient, a quadratic model is automatically applied. The R^2 value in the header of the table for each specification is the proportion of variance explained by the model. This is the goodness of fit of the model. Sparse regression techniques allow for computation of the contributions even when the number of parameters is very large compared to the number of Monte Carlo samples simulated .
Mismatch Contribution is available now in Virtuoso ADE GXL, first released in IC6.1.6 ISR3.
 http://en.wikipedia.org/wiki/Variance-based_sensitivity_analysis  J. Tropp and A. Gilbert, "Signal recovery from random measurements via orthogonal matching pursuit," IEEE Trans. Information Theory, vol. 53, no. 12, pp. 4655--4666, 2007.