Last week at SNUG India, LSI presented a good paper on the topic of Functional Coverage reuse (See: http://www.synopsys.com/Community/SNUG/India/Pages/Abstracts.aspx?loc=India&locy=2011#TA1)
Challenges and Approaches for Functional Coverage in SOC Verification Environments
Manikandan Subramanian, Ron Jacob, Sasidhar Dudyala, Srishan Thirumalai [LSI]This paper describes the complexity in using block level functional coverage at top level and pitfalls and approaches to aid reuse. This also describes controllability on coverage infrastructure from block level to SOC level and how UVM-EA helped in building the layered testbench infrastructure that can be reused.
What I really liked about this is the level of maturity that the SystemVerilog adoption that this paper indicates in India – while functional coverage is one of the top few powerful features in System Verilog, its adoption has been traditionally slower than what we wished. Especially with the boatload of features, knobs/options to control/fine tune, it is clearly one of those features that is waiting to be explored in greater detail. In this paper Ron laid out a nice architecture for “coverage reuse” across levels of verification. The architecture he & his team proposed can be captured into 3 classes:
- Config class – to configure “How much do you want”
- Coverage class – to capture “what and all you want”
- Coverage collector class – to sample the coverage points
In a way the last 2 points have been stressed by VMM for years, and we at TeamCVC have been recommending it to our customers for years.
Specifically during our popular System Verilog training sessions such as VSV (http://www.cvcblr.com/trng_profiles/CVC_LG_VSV_profile.pdf) we compare this to an athletic race and describe how the “field meters” placed/planted in the filed actually measures the speed while the runners/athletes simply RUN RUn & RUN!
Now compare this to a classical VMM environment:
The environment with all the components form the “field” while the “transactions” that flow through match the real “athletes”. It makes a lot of sense to plant the “measuring meters” (in this case the “coverage collectors and the coverage models” away from the actual transactions.
This is what Ron’s team experienced too. Though there are some ACC technologies such as ECHO in VCS that traditionally worked better (see: http://www.cvcblr.com/blog/?p=9) with transactions “embedded with covergroups”, VCS’s ECHO has come a long way and has supported VMM style covergroups as well.
The next big challenge that Ron addressed was the “reusability” and need to “control” the amount of coverage at System Level from block level. He had several good guidelines for the users, recommend highly to take down his paper and keep it handy at work! While some of the “sample_cov” overriding can be better done using SystemVerilog 2009 updates to built-in sample() function, a lot needs to be still done. For instance how do we override a full coverage model/covergroup/coverpoint/bins/cross etc. at System Level?
Ron’s approach was to add disable bits – yes, better than not having it, but it doesn’t scale up. Several years back Vera added such AOP/OOP style extensions to covergroups, but due to slow user adoption, this was never ported to SystemVerilog. Talk to Arturo Salz – friendly known as the “Father of Vera HVL” by many if you are interested.Basically the extensions are to allow things like:
- Add extra coverpoint/bin/cross
- Delete/drop a block level coverpoint/bin/cross
- Re-define the entire covergroup etc.
Now – where do we go from here – IEEE-SA invites sincere participation from end-users to set directions, drive language features/enhancements via active participation. See: https://mentor.ieee.org/stds-india/bp/StartPage to know more.
