verification metrics on IMC

T are unique modules and instances are the instantiations. Instance coverage will list all the instances in the hierarchy whereas Types coverage sums all the coverage of a mudule everywhere it is used.

Tim

Thank you for the respawns.

Can you expand more about unique modules - what do you mean when you say it?

For me they still look the same.

Thanks very much,

Nir

If you have a NAND2 module instantiated 100 times, the Type will show the overall coverage for NAND2 and Inst will show coverage for each instance of the NAND2. If your coverage is the same, you probably only have one instance of each module.

Tim

If this is true, the overall coverage of Types should always be always greater than the Instances, but looking at the rapid adoption kit of Cadence I see the following result where Types has smaller coverage than instances.

1. What am I missing?

2. Why do the overall coverage is an average of those two and not only Types coverage value?

Thank you for the answers,

Nir

If the module is flat, that would be true. But if the module has hierarchy, then the type coverage will only show for the module. The inst coverage rolls up the coverage for the hierarchy.

For the example you showed, dkm_test module has 61% coverage. That is just coverage for that specific module.

The dkm_test instance has 50% coverage which is coverage for (itself + dkm_i + dkm_cov_inst) / # of instances or something like that -- I don't recall the exact calculation.

Tim

As Tim says, the "average" grades are taking the grade numers from the immediate child nodes and averaging them. So is dkm_i had an average grade of 0, and dkm_cov_inst had an average grade of 50%, the average grade for dkm_test would be 25% = (0% + 50%)/2.
The "overall covered" grade is a simple ratio of the hit bin count / the total bin count for that instance. Both counts are a sum of all the child counts. 

This is why the grades aren't directly comparable between the types and instances views.

Note also that there may be bins excluded by the user in either the types view or in a single instance, this can again skew the numbers and make it impossible to directly compare the results with just a screenshot.

Thank you for replaying,

I can see the difference between Types and Instances, but looking at it from verification perspective, I'm trying to understand why we calculate the average of Types and Instances, instead of just looking at Types / Instances as our coverage score.

Wouldn't it slightly disturb our result? Should we just choose the way we want to look at our coverage rather than calculating the average?

Thanks for answering.

Nir

I don't really understand your concern. Coverage is quite subjective anyway, and different teams choose different approaches to determine what coverage is important for their project. Nobody would just look at the top-level number in IMC and use that on its own as a sign-off metric.