Forum Discussion
Altera_Forum
Honored Contributor
14 years agoGlad you know it. Note that 0.2ns is really tight. TimeQuest models on-die variation into skew analysis, which tends to be pretty large(I would argue too large at times, but you have to follow the model). So even if two paths were laid out perfectly identical, they would have a decent amount of skew. Naturally, things won't end up being perfectly identical.
Also, I find to get the best skew, hand-placement is sometimes necessary. The reason is that the placer always tries to put things as close as possible, i.e. it wouldn't put things further away to meet hold timing/skew requirement. It's the router that adds delays for these constraints. So if you're skew constraint was from an input to 8 registers, the placer might put two of the register in the LAB right next to the I/O, then two further away, etc.(I'm only putting two per LAB if they have different clock sources, as there are a limited number of clocks per LAB). That ends up with pretty bad placement as far as skew goes, and it's up to the router to add delay to fix it. It's pretty hard to balance this out. By hand-placing them equidistant, you can get better results. Definitely requires a test design and some elbow grease.