Yesterday I wrote a post He Who Goes First...Loses about how being first isn't always a good thing. Either in the short term, being first is a negative, or there is a short-term positive but longer term things are not so great. New EDA Point Tools Things have changed a bit in the last few years in EDA, which I'll get to later, but it used to be extremely difficult to introduce a new product into the channel because the salespeople knew that going first was a recipe for problems. The product was immature, and the salesperson and associated AE team would have to spend time addressing them. Everyone sat on their hands, waiting for others to go first so that when the product was mature, they could rush in. Nobody wanted to be the forlorn hope; the rewards were not big enough. This was brought home to me after Cadence acquired Ambit in the late 1990s (I was VP Engineering). We had a synthesis product (amusingly named BuildGates), and Cadence had recently canceled their own internal development on Synergy. It was important that the product be sold aggressively to justify the high purchase price (nearly $300M if I remember right). But the salesforce would not. The Cadence salesforce was accustomed to closing deals on the basis of trying to get the entire budget except for synthesis. Very few customers (Philips was the only one) were going to make a wholesale switch, but they would probably take a few copies, and then maybe lots later. No salesperson wanted to risk messing up a multi-million dollar deal for a few hundred thousand dollars on a product they knew little about. While there were some sales, the sales goals were missed by a lot. Another Cadence example was back in the same era. We had developed a lot of formal verification technology at the Cadence Berkeley Labs. This was then transferred to Cadence proper and productized under then name "Heck". This was not a good way to work anyway—you can only really move technology in people's heads—but that's a topic for another day. I have no idea if the technology was any good; in that era, formal verification was a niche market and very hard to use. But again, we never found out, since the sales team refused to sell it, and after about a year it was quietly dropped. Hunters and Farmers This is one reason that back in that era when you could sell a point tool without needing it to be completely integrated across the entire flow, startups could sell products. Their sales people were hunters, finding the early adopters to get the product across the chasm, as opposed to the sales forces of the big companies who were farmers, working the fields of the big customers. So the strategy of the ecosystem was to develop new technology in startups. They would get the product moving in the channel, then the big companies would buy the successful companies. Then the farmers would move in. If the big companies developed internal technology, the salesforce would not sell it and it would fail. You can't get a point tool started with the farmer approach. There are only a couple of products I can think of back in that era that were developed by big EDA companies (once they were out of their own startup phase) that became successful; these were Calibre and PrimeTime (and even that second one is dubious since the #1 product in the space, Motive, was acquired and shut down, leaving a vacuum to move into). Today I think the situation today is actually a bit different. Each new process generation brings new challenges (FinFET, double patterning etc) and so any leading edge customer has no option but to adopt the new version. Everyone—Cadence and the customer—knows there will be issues. Everyone knows we will all have to work together to solve them. There are still one or two startups in little niches, but they are the exception and far less numerous than twenty years ago when DAC would not just fill part of the south hall at Moscone, but all of it. And the North Hall, And use the corridors in between as overflow space for still more companies. Solving problems together has always been the case to some extent. When I was one of Cadence's co-CTOs, I used to end up in arguments with the VP Engineering of one of our big customers. He would complain that our software was buggy for leading edge nodes. I would tell him to use an old version, but we both knew that was impossible; the capabilities were not available in the old version. Then we would have some test cases in the new node before we released the software, and we'd test it better. But this was also a non-starter since the test cases can't be produced until the software is—if not up and running—at least up and staggering. So actually the situation was the same back then, but less well acknowledged. The startup approach no longer works. This is partially due to the investment climate, but also because the problems to be solved (such as coloring vias) require you to have a full flow already. There is no niche for the "via coloring company," no matter how compelling the technology. It has to work in layout, place and route, physical verification and so on, so you have to have it all already.![]()
![]()
↧