| 2 min read

Does UVM sometimes make you feel stupid?

Somewhere in the deep trenches of a UVM based verification project, an engineer teeters on the verge of insanity.

As the saying goes, the faint of heart need not attempt UVM based verification. But what makes it so challenging to learn and adopt UVM in real life?

First off, it is assumed that anybody who attempts to use UVM has a firm grasp on verification-specific features of SystemVerilog. It is a big leap for anyone to get their head around the concepts of OOP –classes, extensions, virtual classes and virtual methods, polymorphism. Add to that constraints, coverage, cover-groups and cover-points, randomization, scoreboards, interfaces and virtual interfaces to the mix, before one can even begin to delve into the basics of UVM.

Does UVM sometimes make you feel stupid?

And then there is the UVM base class library itself. To the uninitiated it is a large library to understand.
There are about 150 SV code files, 460 classes, 40 enums, 440 macros and 1650 unique functions! Of course, we can’t forget Transaction Level Modelling (TLM) and Phases, can we? Very quickly, we realize that understanding the interactions of these classes and methods, and solving the complex verification challenges is a formidable task.

But have no fear, you say –after all we are talking about the Universal Verification Methodology. Universal, as in all EDA tool vendors comply strictly to the same laws, by-laws and guidelines. Yet, in practice we find that there are multiple sub-methodologies or idioms that can be used to set things up and do common tasks. These sub-methodologies have a history in the UVM committee and are backed by different EDA companies in the committee. One sub-methodology relies heavily on the use of macros while another insists on using method calls. One sub-methodology relies on creating virtual sequencers, and the other avoids the term all together.

So, where do you start and how do you avoid creating a hodge-podge of idioms and spaghetti code?

An immediately obvious solution is automation. A moderately complex script could be used to generate the oft-used UVM test-bench structures based on the RTL design being tested. Examples of such structures include interfaces, agents, drivers and monitors for standard buses, and standard sequences. Not to forget is the test-plan, which forms the basis of measuring the success and completeness of your verification efforts. It’s arguable that a home-grown script would suffice for one-time use.

What is not so obvious is that, as specifications change and requirements grow, enhancing and maintaining such a script also becomes an onerous task. “Ownership” of such a house-grown tool also might become an issue as people get re-assigned to different projects. Investing in an industry standard tool, like IDesignSpec with the ARV-Sim module, then becomes a smart choice. Automation is greatly aided if the specification itself is in a standard, executable format. Again, IDesignSpec comes to the fore –it allows you to create a single register specification in a format such as SystemRDL, Word, Excel or IP-XACT. And from that specification, use the ARV-Sim capabilities to generate all UVM test-bench level components, test-plans and scripts (Makefiles). Having generated the UVM test-bench automatically may not be an end in itself –you can use it as springboard to add additional components like third-party VIPs, checkers and monitors. And just like that, your initial trepidations, either real or imagined, about using UVM in your verification environment vanish into thin air. If you compare the time and effort saved in creating, debugging and bringing-up a UVM test environment using a well-designed, robust, industry-proven tool, versus doing it all manually, suddenly makes you feel smart –an entirely different feeling from where you might have started with!

ic designer's guide to automating design through implementation of semiconductors