What is it about EDA that makes it different from other software businesses? When the CFO of Texas Instruments buys Oracle or SAP he or she doesn’t study what algorithms they use in their relational database. EDA purchasers are the only people who “take the cylinder head off and look at the valves” before buying.
I think it is the speed of change. EDA, as we are all tired of hearing, is driven by Moore’s law. But the effect is that every few years a complete technology re-investment needs to be made and the incumbent does not have much of an advantage in discontinuous change. It’s not like that in other software industries. For Oracle, the last major change in databases was the relational database superseding hierarchical databases, and that was starting in the 1970s (first at IBM and then when Larry Ellison founded Oracle under its original name, Software Development Laboratories). It looks like there may be another change starting, driven by the internet, to schema-free databases which scale better to thousands of servers. So one turn of the handle of Moore’s law in 30 years, 15 to 20 times slower.
You could give me half a billion in VC money and I’m not going to be able to put together a startup to displace Oracle, no matter how many of the best database programmers I hire. Because it is not mainly about technology. And even if, by some miracle, I succeeded it would take 20 years. By contrast, when EDA companies miss a transition they tend to vanish quickly. Calma was not a major force in gate-level design. Daisy missed synthesis despite Synopsys being staffed with many of the same people, and, to add insult to injury, in the same buildings. Cadence lost its Dracula physical verification franchise to Mentor’s Calibre as quickly as it took 0.35um to come online, a few years. Yes, mistakes were made. Daisy made an ill executed acquisition of Cadentix. Cadence decided to make its own hierarchical DRC, Vampire, incompatible with Dracula so it didn’t harm its cash-cow. But mistakes will always be made (“new” Coke, Ford Edsel..) but in other industries one mistake is rarely fatal.
The speed of change also means that startups can get traction in a way that they don’t in other industries. The barriers to entry are really low, just a few people who really know the technology and it is possible to build a world-class product that is better than anything else out there. Most importantly, customers will buy it since the risk of using a product from a startup was lower than the risk of not doing so. Other businesses don’t move so fast. Waiting for the big guys to have it is usually the safest approach.
This may be changing as the importance of integration increases and so the important of point technology diminishes. I spent years at Compass Design Automation with a fully-integrated toolset that was very productive. But customers would only buy “best-in-class point tools” and do all the integration themselves. We were selling an engine when people wanted to buy their own ignition system, their own fuel-injectors and make the wiring harness themselves. Despite the limited commercial success (Compass only reached $55M before it was acquired) I still believe that the integrated approach really was superior for everything except unusual chips like memories and microprocessors. The evidence was that every chip VLSI Technology produced until the late 1990s was produced entirely on Compass tools on very short time-frames. In that era, for example, a huge percentage of mobile phone chips, then and now a market with short product cycles, were done that way.
It is no longer clear that most semiconductor companies have the inclination or manpower to look at tools from startups. Their focus is more on reducing cost and reducing the number of vendors they use. It is completely unclear how EDA will evolve in the current downturn. In Mike Santorini’s memorable image, semiconductor is the speedboat that pulls the EDA water-skier. When it is at full speed we ski well. Right now I think the water is around our waist and we are still sinking.