What is the appropriate model for crosstalk control?

Crosstalk is inherently an analog phenomenon, and can be modeled on many levels. These models range from simple to complex, with the more complex models giving more accurate predictions, but requiring more data. In the interest of easier tool development, simpler characterization, or reduced file size, however a simpler model may be used. Such a model must be conservative, since it is acceptable to fix a problem that may or may nor exist, but not acceptable to fabricate a chip that does nor work. This paper examines the tradeoffs between different crosstalk models, with experimental data on the effectiveness and computer demands of different types of analysis and fixing. A lumped crosstalk model with timing windows and slopes, and with characterized cell thresholds, reaches the point of diminishing returns in the examples shown. All errors reported by such a model can be repaired without increasing the chip size or hurting the performance, so a more detailed analysis is unnecessary.