calculation - you see no extra violations from crosstalk windows, only a large number of generic violations from the increased uncertainty, which you then fix blindly. That amounts to over-fixing, and the design may still fail in silicon because real cell-delay changes and noise bumps from crosstalk appear when it is run at the targeted frequency.

KEY Adjust the frequency - it updates crosstalk arrival windows so the tool fixes real SI; changing uncertainty just causes blind over-fixing.

Regular OCV vs AOCV

Yes, regular OCV is more pessimistic, especially for paths with deep logic depth.

• Regular OCV applies a flat derate to every cell regardless of logic depth, so it reports a large number of timing violations.
• AOCV scales the derate by path characteristics - the derate decreases as logic depth increases, and it increases with the cell's distance from its diverging point.
• A longer path with more gates tends to have less total variation because gate-to-gate random variations partially cancel; therefore AOCV applies larger derates to short clock paths and smaller derates to long ones.
• AOCV chooses derates based on path logic depth and the physical distance the path spans - a path crossing a large physical distance tends to have larger systematic variation. Because it uses these path-specific metrics rather than constant factors, AOCV is less pessimistic than traditional OCV.

KEY Regular OCV applies one flat derate; AOCV scales the derate by logic depth and physical distance, making it less pessimistic.

Location-Based Derating in AOCV

Location-based derating means the OCV derate for a cell grows with the cell's physical distance from the diverging point in the clock path - the farther the cell is from where the clock path diverged, the larger the derate applied.