For clock trees, the traditional way is to go with zero skew or balanced skew. For each of the sinks, the insertion delay is kept to be equal so that each node receives the clock at the same time(or as close as physically possible). Here, your clock design is completely independent of your data path design. You are ensuring the clock signals come to each data sync point within a permissible variation, and the design can theoretically be timing closed if the data paths also can meet the timing within a permissible range of the clock signal.
However, meeting zero skew in some large designs can be unnecessarily costly. If clock skew is used intentionally to meet timing then it is called useful skew. P&R tools can use useful skew as an optimization option in CTS and other stage optimizations to leverage the clock signal in meeting timing, with acceptable margins.
Let us first see an example of clock being skewed to meet timing.
Given below is a design with zero skew, but setup timing violation.
Let’s take the clock period of the circuit as 10ns, with 2ns insertion to delay to clock pin of each register. Let’s also assume that the setup and hold times of the flops are zero for this example. The path1 combinational delay is 11ns and path2 combinational delay is 5ns.
The setup timing requirement of path 1 is:
`2ns(clock path delay till FF1/CP) + 11ns < 2ns(clock path delay till FF2/CP)+10ns(clock period)` 13ns < 12 ns --> Timing Violation
If we decide to keep the clock network intact, we need to fix this timing violation by reducing the combinational path delay of Path1. However, path2 timing is met with considerable margin.
`2ns+5ns < 2ns+10ns` So, ideally we could borrow some cycle time from path2, if no other paths are affected by it. Let’s introduce some clock skew to do just that.
Now the setup timing requirement is
`2ns(clock path delay till FF1/CP) + 11ns < 4ns(clock path delay till FF2/CP)+10ns(clock period)` `13ns < 14ns` --> Pass
However, now there is a clock skew of 2ns between these registers, and the Path2 timing has now changed to:
`4ns(clock path delay till FF2/CP)+5ns < 2ns+10ns` `9ns < 12ns` –-> Still not violated, even though margin is now less.
PnR tools when enabled ‘useful skew’ can change the clock network and introduce intentional skew to meet timing with smaller cost. Some of the useful commands in Innovus tool are:
setOptMode –usefulSkew false setOptMode -usefulSkew true -usefulSkewPreCTS true setOptMode -usefulSkew true -usefulSkewPostCTS true