Standardized Functional Verification- P19:Every manager who brings a design to tape-out or who purchases IP must eventually face these questions. The ability to answer these questions based on quantitative analysis is both vital and yet elusive. In spite of the enormous technical advances made in IC development and verification software, the answers to these questions are still based largely on guesswork and hand waving. | Knowledge and Risk 165 Fig. . Data-driven risk assessment A prudent strategy for managers of multiple ICs is to aim for accumulation of empirical data over time to develop a baseline for risk modeling. If you can obtain the standard results you can apply the standard measures and enjoy the standard views. These measures and views can then enable you to evaluate the benefits if any of changes or intended improvements. Create your organizational baseline for risk vs. results. Acquired intuition from frequent detailed interaction with the target of verification can prove invaluable in the search for bugs. Careful consideration of this intuition leads to good engineering judgment and the verification manager who learns to recognize and to nurture this good judgment within the verification team will be rewarded with better verification results. Turnover of engineering staff erodes the team s collective engineering judgment and this should be kept in mind during planning and execution of any verification project. 166 Chapter 7 - Assessing Risk Coverage and Risk In chapter 5 we learned how to normalize functional coverage with respect to complexity of the target. This is what enables us to compare widely different projects. Coverage informs us as to how thoroughly a given target has been exercised. The more we know the less risk we face at tape-out. Risk simply reflects what we do not know. If we knew perfectly the dynamics that will affect the final resting position of a pair of tumbling cubes we would excel at shooting craps. But because we do not know these dynamics perfectly we rely on probability theory and statistics. Graphically we can visualize the relationship between knowledge and risk and its inverse confidence as shown in Fig. . The more thoroughly one exercises the target the more knowledge one gains about the target and therefore the lower the risk of an unexposed bug. Fig. . Knowledge reduces risk increasing confidence Data-driven Risk
