I love the experts at the table series that system-level design has been doing. In general they are well put together and have some interesting points that come out. Often there are differences of opinion, but they are handled in an amicable way. Today I read one that almost made me laugh rather than anything else. If refreshments had been served, I am sure they would have been thrown across the table at each other.

Everything started out with an innocent question about the differences between customers doing the most complex designs and those lagging by a few nodes. I expected the discussion to talk mostly about the different levels of complexity, the challenges of closure, perhaps even to talk about FPGA verification (which they did touch on). Instead the discussion went in two directions. The first was an implication that designers who are not working on the most current nodes are dumb and they need their tools made that way for them. Thanks to Mike Stellfox at Cadence for steering that one in the right direction. It reminded me about an analog design class I had when at college. The subject matter was amplifier design. At the time most commercial home amplifiers on the market were coming from Japan and they were adding more and more knobs and switches. Boost this, filter that – everything could be adjusted. By contrast, the top of the range amplifiers generally had only one switch and one knob. The switch was for power and the knob was for volume. He discussed at length how much more difficult it was to design and build the minimalist system because it had to be able to adapt and compensate for the things around it. This is exactly the same as when building EDA tools. That was what Synplicity based everything on – not making the tool simple, but making it simple to use and to get effective results from it. Many EDA tools are way too complex because the tool developer does not know how to make the tool produce the right results all the time, and so they place that burden on the user.

Then the conversation headed off in a completely different direction. It became a question of how many verification engineers do you need and where should they be placed. Himanshu Bhatnagar of Mindspeed Technologies said “Now the design part has shrunk because you’re buying IP. Verification has also shrunk because of economic conditions. Implementation has expanded.” Well that got things going and he further stoked the fire with “If the design part is pretty much gone, why do you need so many verification resources?” If that weren’t enough to set everyone against him he then adds “Verification is easier to outsource than anything else.” Well, I am not going to reproduce the article, but I was just dumb struck by this discussion. Yes – outsourcing is difficult, but many companies have found ways to make it work and in my observations the ones that work all had very good communications at the heart of it.

------------------------------------------------
Brian Bailey – keeping you covered