Intel Corp (INTC)

[Unkwn]

While Borusa and Andy are patting each other's backs, here is some more information about the actual densities of upcoming processes of the foundries and Intel:

http://www.eetimes.com/document.asp?doc_id=1331185&page_number=2

Samsung already announced 10nm chips, its own Exynos and the Snapdragon. According to the article, this is with bad yields. It also states that Intel 10nm already has better yields than TSMC and Samsung, with TSMC somewhere in between.

Comparing the actual nodes, Samsung 10nm is less dense than TSMC 10nm, which again is less dense than Intel 10nm. From the chart it seems that they are all pretty close to each other this generation. At 7nm, the picture looks different (though I think there's even much more speculation involved), with Intel having its node close to what TSMC and Samsung claim as 5nm. According to the chart, Intel 10nm will be actually the same as TSMC 7nm, with Samsung 7nm being a bit denser than the two (this also tells you what TSMCs 7nm chest pumping is worth). Regarding the actual density of products, that's still a different story, I guess (also depends on tools, cell library etc.).

What is important in my opinion: All the 10nm nodes don't seem to be too far away from each other, with Intel slightly leading but Samsung introducing the first products at this nodes (at bad yields) as it seems. Can we call that an even? If not lost, Intel has almost no lead left. Bye bye the chance of "flooding" the mobile market and positioning itself as a high end foundry with premium prices. Screwed nicely ... Intel 14nm would have been a nice foundry killer, if, just if ...

I have had the privilege to visit a modern fab recently, including its clean room. It wasn't a leading edge 28 or 14nm fab but also not too old, able to process sub 90nm 300mm wafers. What I think to have learned from this visit is that it much more boils down to logistics and automatisation than I have thought before. It is a huge logistic challenge to process all those different products with different requirements, slightly different process steps, layers, quality requirements, testing etc. I think to understand a lot better what Intel faces as challenge when competing against a foundry like TSMC, which is able to mass produce a vast variety of products, from analog large structure ICs, mixed signal middle density to high density logic chips. This is much more of a logistic and automatization problem than it is a "be capable to produce small structures" issue. I have ever higher doubts that Intel has just close the skills and attitude to compete in this space, especially since they lost their process lead. Processing their own highly uniform products is a completely different animal.

Ah, and following the article, it seems to me that trying by brute force to be the leader in process tech no matter what the costs are is soon turning into a looser's game. Maybe the winners will be the ones staying at the nodes where the masses of customers find the best compromises instead of paying big for some prestige products nobody is going to pay the money for. Mask costs at 16nm are said to top 5 million now, compared to 2 million at 28nm and 1 million at 40nm. Sure, costs are expected to get lower with time but those are figures only a certain market share of the foundry business will economically benefit from - at some stage, the total costs of owning such prestige process tech will be higher than the return on that market share that is left to pay for it. We'll see who is acting the smartest from here.