Fab 5

Your foundry planet graph is a bit off, but captures the essence. The consolidation on the fab space actually makes life for EDA easier... fewer technologies to support. TSMC, with it's reference flow and proprietary formats for DFM is driving differentiation, but other than Blaze, has been largely neutral.

Seems to me that foundry specific tools are going the way of the dinosaur. I used LSI tools a long time ago and also used IBM tools for doing 3 ASICs with them. All had weird idiosyncrasies and, of course, locked you into a single supplier. As a engineer one is better off learning horizontal tools as Grant suggests, since there is no guarantee that your next project or employer will be with the same foundry.

Indeed, as SteveM commented, the reason for such a re-integration of EDA (back end) and Fab will no doubt be when, and if, there is strategic advantage (for example, yield improvement = cost reduction for the fab) that would differentiate one fab over another. And if yield improvement is not done just at the last stages of the back end, but has deep interactions higher up the flow, then this might happen if there is some interesting breakthrough there. So this possibility will depend perhaps on some real innovative breakthroughs in what EDA algorithms can do - breakthroughs that make it worth it to privatise what is now a more open oligopoly.

Martin - Semiconductor co''s view EDA like auto co''s view auto parts and steel companies. The horizontal stack disaggregation (horizontalization) is working perfectly for them. They have an open market with interoperable standards and hold regular open benchmarking to drive competitive performance and pricing. This lets them focus on their specialties and command premiums for their product. What''s not to like ? Externally EDA companies can try serve all the fab clubs and can hope to gain a dominate market share. The trigger which will make your idea take root is when a foundry develops the strategic view that with EDA can offer a unique differentiated product to their customers which will yield them more customers at a higher premium. It''s much easier for a semi CAD manager to manage external firms rather than internal firms. If the technology doesn''t deliver he can just throw his vendor under the bus rather than be personally responsible. Meanwhile he gets free tickets to the ball game and golf outings to boot. IBM is the last of the internal (back-end) CAD species ... so time will tell whether this fab club excels with this unique trait.

Dr. Martin Reading your blog with interest here in India. Would you care to comment on the following question ? Should TSMC come up with its own STA tool for signoff? It is a question many of my acquaintances have asked. It it might relieve the customer from the deadly claws of SNPS and its Primetime. TSMC can easily do a better job than SNPS and have a far better understanding of interconnect physics like inductanctance extraction with accuracy and stable numerical methods and ditto for capacitance. Likewise for DRC/LVS. EDA engineers do not necessarily write the best code. Almost all EDA code is buggy and they are sold with the expectation that they will fail in a real tapeout.

One theory for idle speculation - which I wondered about several years ago - was whether the Fab clusters (clubs in Paul's analysis) would eventually be equivalent to the number of EDA companies with back ends (RTL to GDS II in the old days, now including no doubt mask manipulation etc.). Arguably there are 3.5 of these companies if you think Mentor has .5 of a back end flow with Sierra - or at least 3 - Cadence, Mentor, Synopsys. And Intel probably doesn't need EDA for its back end, as If the fab clubs end up equalling the number of EDA back ends, wouldn't they save a lot of money by each privatising the back end - by buying out one of the EDA companies, keeping the back end R&D, getting rid of the enormous overheads of marketing, sales etc., and spinning back out as small EDA companies the front end bits (like simulation, etc.). Certainly, this was worth an idle speculation or two over lunch several years ago. What might happen now?