Is your RTL Linter makes you find needle in the haystack? Here is a smarter approach!
Static design verification through thorough RTL analysis started several years ago, with early stage ones simply being “linters” and the later, advanced ones combining some of formal techniques as well. One of the most common complaints by RTL teams while using linters is the SNR – Signal-2-Noise Ratio of the endless set of errors/warnings from the tool. So much so that we have heard of customers giving up on linters, primarily those freebies/bundled with simulator ones. While it is true that some of the reported ones are indeed serious issues, the ROI (Return On Investment) of having RTL designers navigating through the ocean of messages is too little in many cases.
OTOH there are clear set of issues that a good LINTer can spot for you quite easily, for instance see:
Sledgehammer to crack a nut? – Use right tools for right class of design errors/bugs: http://www.cvcblr.com/blog/?p=377
Essentially it comes to the art of “Finding the needles in the haystack” and in a timely manner:
Don’t loose your heart, there is a silver lining with modern day linters/static analysis tools. We TeamCVC covered the rise of new age linters back in 2010; http://www.cvcblr.com/blog/?p=99. Recently Ascent from RealIntent showed the below case study that proves this very point:
From: http://www.realintent.com/real-talk/979/the-race-for-better-verification:
In the following table you will see the results of the analysis of Ascent IIV on 130K gates of RTL logic that was done by NEC in Japan.
Table 1. Ascent IIV Intent Checks and Failures Report for a 130K Gate Block.
The tool generated 31,186 intent checks that were analyzed by its various engines and 2,999 failures were produced in total. The hierarchical reporting of the tool characterized these failures into several categories. It determined that by fixing the Primary Errors (purple column in the table) this would eliminate the duplicate, secondary and many structural errors. In this benchmark, 181 primary errors were identified. This represents a dramatic contraction of almost 95% of the total failures from the tool. You can read further comments by NEC here.
So next time when you hear of the Linter’s SNR issue, do recollect this “smart categorization” and push for it in your solution or move/upgrade to better ones. you should however check on local support team for such new age products though as these tools do require some hand-holding especially in the initial stages of adoption.
Good Luck,
TeamCVC
Comments
Post a Comment