Analysis: Intel-TSMC announcement more complex than reported

Tuesday 03rd March 2009, 03:30:00 PM, written by Arun

Everyone seems to have their own interpretation of the Intel-TSMC deal, with some viewing it primarily as an outsourcing announcement, and others focusing on the IP aspect. But from our perspective, it also seems much more complex than that and Intel's statements are far from straightforward.

Why do they focus on how other companies will now be able to design their own SoCs with Atom, yet claim that 'Products manufactured through the agreement may find adoption in embedded CPU market segments such as mobile internet Devices (MIDs), smart-phones, netbooks, nettops, and AC-powered consumer electronics device'? Surely they don't want, say, Qualcomm to compete with them head-on in x86 netbooks, since that would result in significantly less revenue going to Intel?

And what would their IP licensing contract look like? A traditional IP business is orders of magnitude less lucrative than what Intel is used to; and it's clear that's not what they want to do given that end-products will still be forced to bear the Intel logo. And why does say that 'Intel will work with the end customers' - wouldn't that imply the device manufacturers, rather than just the SoC companies?

So let's stop asking questions, and let's start proposing some real answers. There is one key sentence in the Intel press release that nobody seems to have noticed: "The collaboration is intended to expand Intel’s Atom SoCs availability for Intel customers for a wider range of applications through integration with TSMC’s diverse IP infrastructure." - notice that it's not about potential licensees benefiting from the Atom IP in their own designs. It's about Intel's customers benefiting from a higher level of integration thanks to TSMC's IP infrastructure.

So this could imply that Intel is the one that will design SoCs at TSMC, and they will license the IP from other companies to create solutions to better meet their own customers' needs? Not quite, if you are to trust this statement: "We believe this effort will make it easier for customers with significant design expertise to take advantage of benefits of the Intel Architecture in a manner that allows them to customize the implementation precisely to their needs".

So in summary, what does this all mean? Here's our interpretation: there are two major aspects to this plan. The first is that companies that have both a semiconductor design business and a CE, PC, and/or smartphone business could now create their own SoCs using Atom. These companies are not the likes of Broadcom or Texas Instruments, but rather Samsung, Sony, LG, Toshiba, and so on. Possibly even Nokia, if they truly are interested in the Netbook market. More important, however, is another company that is deeply involved in all of these markets and is still expanding its semiconductor operations: Apple.

That doesn't mean Broadcom couldn't license Atom if they wanted to, and I'm sure Intel would be delighted if they chose to. But that's seemingly not their primary focus right now, and it's not clear whether the license agreement would be tempting enough to make them move away from ARM/MIPS in any of their markets. As for Apple, it feels unlikely that they would be interested in Atom for the iPhone given their long-term technology license agreement with ARM; however, they likely would be for the netbook and nettop markets (and maybe MIDs). In the case of other companies, we suspect they may prefer Intel to synthesize Atom for them and supply it as Hard IP; in Apple's case, it's not obvious what they might want though.

However, there also seems to be a second part to this plan, which is Intel outsourcing some of its own chips to TSMC. Why would they want to do that? There are plenty of points you could make regarding cost and power efficiency (and we will in a second), but there's something else: the IP infrastructure available at TSMC. Right now Intel still uses a separate southbridge, but they want to merge everything with the CPU after Moorestown and its 32nm refresh - so probably around 2011. But there's a catch: to do it on their own process, they'd have to implement all of that I/O and possibly analogue functionality on their own - they couldn't use pre-verified IP from the likes of Synopsys, Mentor, MIPS, or even TSMC itself.

Based on the various reports based on Intel's statements during the briefing as well as their press release, we therefore believe that the single-chip SoC aimed at smartphones and MIDs in the 2011 timeframe is likely to be manufactured at TSMC, likely on their 28LPG process which doesn't have high-k but uses triple gate oxide. Other single-chip SoCs aimed at the netbook/nettop market might also be manufactured at TSMC (Intel's own, that is; it's not even worth mentioning in Apple's case), although given the indications as to how early in the process the two companies are, we are skeptical that it would enter mass production before 2011 either (although it might still sample earlier).

When it comes to cost and power efficiency, one key point to remember is that it's one thing to use a more expensive but higher performance process for a 25mm² chip that exclusively contains the processor. But once you add all the northbridge & southbridge functionality, you might double or even triple that die size and that extra functionality is unlikely to really benefit from the extra performance. Therefore, the economics change quite a bit and analysis of Intel's current Atom products does not tell the whole picture. When it comes to power efficiency, Intel's low-power processes (yes, they do have those!) are actually pretty good and they have surprisingly high performance, but power is also very much about system architecture and Intel probably realizes that it would be extraordinarily difficult to be competitive on that front without a single-chip design.

So that's that. By now it should be clear why it is so difficult to get an exact idea of what Intel's plans are and what some of the complex dynamics affecting this deal are. While the above is partially speculative and it's probably that some details aren't quite spot-on, we still hope that it sheds some light on what exactly it is that Intel is trying to do here.
Discuss on the forums


intel ± tsmc, atom, arm, ip

Related intel News

RWT explores Haswell's eDRAM for graphics
RWT: An Updated Look at Intel's Quick Path Interconnect
32nm sixsome over at RealWorldTech
Intel Core i3 and i5 processors launched
Intel and TSMC join forces to further Atom
Fudzilla: Intel 45nm Havendale MCM replaced by 32nm+45nm MCM
Intel announce Core i7 processors, reviews show up
Intel's Aaron Coday talks to Develop about Larrabee
Larrabee to also be presented at Hot Chips
Intel Larrabee @ SIGGRAPH 2008