Design quality enhances company survival

by Bernard Murphy , TechOnline India - March 27, 2009

Flawless execution in design is now a basic survival skill. CQC is the state of the art in managing to flawless execution and there is no downside to adopting these methods.

Most projections for 2009 show IC sales shrinking more than 10 percent. Six of the top ten companies are expected to see a decline in revenue. Some have predicted 40 percent of IC companies will either disappear or be absorbed into more successful competitors over time. In this climate, flawless execution is no longer a visionary goal " it has become a basic survival skill.

The most common approach to quality in most engineering teams is event-driven. Designers work in silos, focused on register transfer level (RTL) creation and limited functional testing, with little objective measurement of quality or completeness. Milestone integration events force some higher level of testing, still with limited objective metrics or visibility. And so on, up to full system integration. Latent problems accumulate everywhere. The measure of completeness is largely subjective and, when the time comes to "fix quality" at any of these events, it is not surprising that team leads waive previously agreed-to objectives, sight unseen, in deference to schedule.

These designers are as quality-conscious as any others; their problem is not in intent, but in approach. Like it or not, semiconductor design in the early 21st century has more in common with an assembly line than a research lab. An assembly line with occasional checkpoints to "fix quality" would be down most of the time. That they function smoothly is only possible because manufacturing engineers know, in advance, exactly how they define quality, and they continuously monitor and correct quality problems, at the source. This continuous monitoring is often labeled continuous quality control (CQC).

Some may argue (and a depressing number still do) that this level of monitoring is unnecessary for semiconductor design. After all, teams tape out and build chips without this approach. It is equally true that you can get a car through an assembly line, eventually, without CQC. Eventually, you can work your way through most problems. But semiconductor markets and companies in 2009 have little tolerance (if they ever had it) for "eventually".

A handful of quality leaders is already adapting. These design teams are aggressively adopting CQC processes. Regressions are run every night and, importantly, results from each regression are automatically emailed to all stakeholders. This frequent and closed-loop monitoring provides no opportunity for problems to fester, or corners to be cut.

Across the range that regression suites provide coverage, every problem becomes immediately visible and actionable. Quality becomes manifestly every designer's responsibility, not just the responsibility of the verification and implementation teams.

Equally these teams know in advance exactly how they are going to define quality. Those metrics are rigorously followed no matter what twists and turns the design may take. A common mistake they avoid in their automated monitoring systems is to report only those steps for which results are available.

{pagebreak}It is important to report status on all steps that need to be completed if you are to have a clear picture of progress against the total quality objective. Reporting status on a step for which you have no results is trivially easy " if the data doesn't exist, you report "Not Started".

Closed-loop and group-wide feedback must be a non-negotiable component of any practical CQC process. Measuring completeness by asking designers where they think they are in their task deserves and will always be rewarded with disappointment. Engineering program management should be driven scientifically by measuring and publishing status.

Surprisingly, while designers may initially resist this level of transparency, they quickly become its strongest supporters. Rather than being held to ill-quantified and un-measurable goals, they know exactly what is expected of them, they can demonstrate progress day-by-day, and they can answer with precision how long it will take them to complete.

Few engineering managers would disagree with any of this. And yet CQC-driven design is still rare. The reason never changes: "We would really like to do that, but we just don't have the time on this design. Maybe on the next design". That argument never held water, but now there is a new level of urgency and a reason to reject that excuse as positively dangerous. The effort and minor delay required to plan and script a reasonable CQC process will always be outweighed by extra effort that will be invested and delay incurred in recovering from a poorly controlled process. You should avoid over-reaching though.

A modest and implementable plan with a high level of automation has infinitely more value than an over-ambitious plan which fails under its own weight. Start with rule-checking " best practices for simulation, synthesis, domain crossings, test and power management. Add basic "liveness" simulation regressions. If you feel you can handle it, add synthesis trials for area and timing. The broader your range of regressions, the better you will control your design program.

The benefits of CQC to the profitability and success of the organization are obvious. But consider also the personal benefits to a design team lead or manager. Who wouldn't want to hear "He/she really runs a professional operation. We need to give them more responsibility and have them propagate those ideas more widely".

Executives value managers who are clearly in control even, paradoxically, if their projects are not always successful. They know that, over the long haul, disciplined managers deliver better results on average than their peers.

In summary, flawless execution in design is now a basic survival skill. CQC is the state of the art in managing to flawless execution and there is no downside to adopting these methods. So when are you going to start?

Bernard Murphy is Chief Technology Officer of Atrenta Inc. (San Jose, Calif.)

About Author


blog comments powered by Disqus