如果是老一些的工艺, 在MMMC中用 ff library+RC best 作为 hold analysis view 是正确的。
新一些的工艺要多查几个 PVT corner。 例如, 65nm -40C can actually be slower, not faster. Slow process, low temp, low voltage might be the worst corner for setup; fast process, nom temp, high voltage might be the worst corner for hold.
set_analysis_view -setup { your list of views for setup analysis} -hold { your list of views for hold analysis }
setAnalysisMode -checkType setup
report_timing -view <one of your views for setup analysis>
...
setAnalysisMode -checkType hold
report_timing -view <one of your views for hold analysis>
...
Innovus is capable of producing fairly accurate SPEF, but it depends on how you set it up.
1) Did you provide various qrcTechFile in your MMMC?
2) setExtractRCMode -coupled true -engine postRoute -effort signoff
I think only when effort is set to "high" or "signoff", Innovus will run iQRC in the background (takes long time). Still, Quantus QRC is more rigorous signoff.
从描述看, Innovus timing 和 PT timing 没有 correlate, 这是不应该的。 两者应该很接近。 差 10ps 可以相信, 差很多就是设置有问题了。 选一个 path, 一个个 gate 对比,也许会给出一些线索。 BTW, SDF is for backannotating gate level sim. SDF should be written out from PT or Tempus. You seem to have used SDF as input to PT. That's odd.
|