In the May 15 issue of Future Fab News!, Aaron Hand, contributing editor, asked for opinions on Mark Bohr’s (Intel) now famous EE Times interview, with Rick Merritt, where he said the fabless model is collapsing and a return to the IDM is inevitable. (I addressed a similar topic a year ago in a post titled Will 3D Integration Keep 2nd Tier Foundries Alive?) So for what it’s worth, here’s my input, from the perspective of the 3D IC industry, where things are progressing quite nicely, thank you, and the fabless/foundry/OSAT model seems to be the most desired for a multitude of reasons.
But before I extol the virtues of the fabless model, I suggest you consider the source of this smear campaign. Of course Intel would love to see the demise of a fabless world, and a return to the IDM. The potential for controlling the semiconductor industry is heady stuff. They would have a lot less competition, for one thing. It’s well known that the transition to 450mm and future scaling is too expensive for all but a few top tier players, of which only one, TSMC, is not an IDM. It was no accident that Bohr’s comments were aimed at TSMC and Qualcomm, when he said TSMC will “only serve one flavor of 22nm” and “won’t be able to make the kind of 3-D transistors needed mitigate leakage current”, and therefore, “Qualcomm won’t be able to use that process.”
Mark Bohr is not entirely wrong when he told Merritt “Being an integrated device manufacturer really helps us solve the problems dealing with devices this small and complex,” But while that may be true when it comes to scaling below 22nm and developing 3D transistors like Intel’s tri-gate transistor, it doesn’t mean the fabless model as a whole is headed for collapse.
In his April 29 blog, Intel says fabless model collapsing… really? Daniel Nenni, of SemiWiki, provided an excellent blow-by-blow analysis of this discussion. He sums it up saying, “Either way, I do not see this as a zero sum game, both TSMC (foundry) and Intel (IDM) will thrive in the new geometries. The fabless model has brought us many new innovations and a very rich ecosystem which will be very hard to break.” I couldn’t agree more.
In fact, in the 3D IC world, which is being touted by many as an alternative to future scaling, a way to scale further, or an orthogonal arm of scaling, the fabless/foundry/OSAT model is emerging as the most viable solution for high volume manufacturing (HVM) of 3D ICs; and a way for second tier, pure play foundries and OSATS to compete with the Intels, Samsungs, IBMS, and even TSMC (the only pure play foundry attempting to become an end-to-end provider of 3D ICS).
Zvi Or-Bach, of MonolithIC 3D, touches nicely on this in his recent blog post, Reversal from the foundry model back to IDM? Zvi Or-Bach when he states, “this might explain why both TSMC and GlobalFoundries recently announced investment in 3D IC processing lines…. As the current scaling trend works against them, they both chose to move the game to a court where an ecosystem would be more powerful than corporate vertical integration.”
So in answer to your question, Aaron, I would say it’s a gross exaggeration for Bohr to predict the demise of a thriving ecosystem based purely on one example. He’s definitely blowing smoke. ~ F.v.T.
It is always preferable when opimions / Business speculations are grounded in technical facts rather than emotion or name calling.
Fact :
1. As yet Intel has a very low share in the large and fast growing market for Smart Phone / Tablet processors
2. They would very much like a share of that pie as the fickle consumer moves away from traditional PCs and even Ultrabooks may not be all that popular ( software developers have definitely jumped ship to Mobile Apps )
3. Given the history of WinTel, players in the Smart Phone / Tablet field used to the wide-open eco-system are wary of walking into the WinTel duopoly once again especially since ARM, in spite of being the sole source for the design of most RISC based low power chips, has been so non-manipulative and are now flaunting their comparatively benign business model as a huge plus.
4. It would take really compelling technical / business advantages for Smart Phone makers to walk back into the Intel camp. This would start with newcomers in the segment e,g. Lenovo of China or rank newbies like Lava of India ( require a lot of handholding and even some finance provided by Intel ).
4. Right now Intel Fab technology is two generations ahead of even the best Foundries and is expected to stay that way for the next 3 to 4 years, Intel leads in technical factors critical to Handheld applications : transistor & die size ( cost ), power dissipation by leakage in ultra small transistors at 22 nm, performance ( e,g. real time graphics for on-line games ).
4. But for performance parameters of interest to SoCs in handhelds, early comparisons of Ivy Bridge ( 22 nm Trigate FinFET transistors ) to Sandy Bridge ( 40 nm, planar ) are not exactly earth-shaking.
5. Technical challenges of drilling and filling holes in live silicon ( TSVs ) and then sticking them together are not quite in the same league as developing & putting into HVM a new breed ( 3D ) of transistors. So this crude mechanical mode of integration would be left to those not quite able to keep up with Moore’s law or afford next gen Fabs ( sub 28 nm, 450 mm wafer ) or design their way out of the Memory Bandwidth Barrier.
6. Once they are perceived to be indispensable and thus technical leaders in their own segment, even the Foundries that have persistently come up short in transition to new nodes, start to emulate the hauteur of IDMs like Intel. You may recall that even the leading Foundry who you are championing here, has since last Nov been making adamant public statements about doing ALL of TSV 3D themselves ( instead of sharing some activities with the OSATs ).
7. In the end 2.5D might delay 3D for quite a few years because 2.5D is ready and makes sense ( easier heat dissipation at not too big a penalty to speed or size ) for Graphics processors etc. Acceptable solutions for 3D stacking of Memory over hot processors are still in the works and the Fabless wonders & their favorite Foundries are waiting for IBM etc. to show them the way.
Hi Dev –
Thank you for adding technical facts and your perspective to this post. Although I thought I was defending the fabless model and not championing TSMC!) Mostly, I champion the Foundry/OSAT model for 3D ICs, as indicated in this post, 3D Panel Discusses the evolving 2.5D/3D Infrastructure.
To paraphrase Mark Twain… Death of fabless is greatly exaggerated. And Nemmi is right. It is not a zero sum-game. Plenty of products do not require 22nm or 3um vias for interconnection.