The CEO panel at the 2nd GTC wasn’t especially enlightening. The theme was that going forward will require cooperation for success and everyone was really ready to cooperate.
The most interesting concept was Aart talking about moving from what he called “scale complexity” aka Moore’s law to what he called “systemic complexity” where we are moving from the age where transistors are cheaper at each process generation to where you can build larger systems, but the per transistor cost will not be less.
Aart also had the most memorable image of the afternoon. He was talking about how amazing it is that you go to a fine-dining restaurant with 8 people and course after course things come to the table at the same time all perfectly prepared. My daughter’s boyfriend is the chef of just such a restaurant so I know about it a bit from behind the scenes and it is still amazing. Delivering a foundry capability is like that: the process, the tools, the IP, the manufacturing ramp and everything else needs to all be ready at the same time. The output isn’t the sum of the factors but the product, and if one is zero the whole thing is a big zero. Global Foundries’ kitchen just happens to cost billions of dollars, a bit more than even the most over the top fine dining restaurant.
Mojy, who was chairing the session, had one question to try and break up the love-in. Global Foundries, IBM and Samsung all compete, and yet they cooperate in process development even going as far as fab-same implementation (same equipment etc in all the fabs of each company). Will the big EDA companies cooperate in the same way? Of course this is a bit of an unfair question. The only reason that semiconductor companies cooperate is that technology development has got too expensive for any one company (except Intel, always the exception) to do it alone as they would have done 15 years ago (and, indeed, did). While foundries and semiconductor companies get some differentiation through process, most comes from what they design (for IDMS) or how they service customers operationally (for foundries). The software that EDA companies create is their differentiation. If Cadence, Synopsys, Magma and Mentor cooperated to build a shared next generation place and route system then it is hard to see how they would differentiate themselves. Yes, some companies have better AEs, some have better geographic coverage in some places etc, but basically they would all be selling a product that they could only differentiate by price. Today, with unique systems but with broadly similar capabilities, they are already close to that situation. So the CEOs largely ducked the question since “no” would have been too direct an answer.