Why Intel is sharing its secret sauce
Since Intel first began cooking up semiconductors, it has taken great pride in its in-house manufacturing chops. If a chip carried the Intel brand, you could be sure it was created in an Intel fab.
Today the pride is the same, but the methods are changing. The Silicon Valley chipmaker on Monday announced a deal that will allow Taiwanese contract manufacturer TSMC to make custom versions of the Atom chip, the first time it's giving another company access to some of its chip cores. How big a deal is this? It's a little like George Lucas agreeing to let someone else make Star Wars movies. (Judging by the most recent ones, that might not be such a bad idea.)
It's not a decision Intel (INTC) came to lightly. For years, companies like TSMC (TSM) have begged Intel for the chance to make chips based on cutting-edge Intel architecture – and Intel has told them to get lost. It was obvious why other chip companies longed to build Intel chips; they serve as the brains of roughly 80% of the world's computers, and oh yeah, they pull in a tidy profit in the process. But why would Intel ever share its secret sauce? By doing everything in-house, the company has grown into one of just a few mega-brands in tech.
In today's environment, though, Intel needs TSMC. What for? In case you haven't noticed, the PC business isn't what it used to be. Technology research firm Gartner predicts that the industry will have its worst year ever, with worldwide shipments dropping 12%. (Until now, the worst decline on record was a 3% dip in 2001.) To keep growing, Intel needs to expand into new markets. Largely, that means mobile: Everything that has a screen and uses battery power is a potential candidate to get Intel inside, including the GPS device in your car, the iPhone in your pocket, and the Nintendo DS in your kid's backpack.
To get that new mobile business, however, Intel knows it has to change some things – including the way it makes chips. Here's why: When it comes to mobile devices, the Nintendos (NTDOF) and Apples (AAPL) of the world don't play by the old PC rules. Instead, they use customizable chip designs from a company called ARM (ARMH).
It turns out, customization and flexibility is a big deal in the mobile world. Intel's chip manufacturing may be the most advanced in the world, but many mobile customers would rather use a chip architecture that lets them tweak the capabilities. (That's why Apple last year bought chip design shop PA Semi for a reported $278 million; word is the engineers are working on custom chip designs for future iPhones and iPods.)
Which brings us back to Monday's announcement. It just so happens that many companies get their custom ARM-based designs manufactured by – you guessed it – TSMC. So by allowing TSMC to make some Atom chips, Intel's sending the message that to be a player in the mobile market, it's prepared to play by the mobile rules, and give customers more flexibility than it ever has before.
This doesn't mean Intel is backing away from its role as a manufacturer. Intel will continue to make chips itself – it recently announced plans to spend $7 billion to enhance its own manufacturing capabilities – and it will place strict limits on how TSMC can use its Atom cores. Any customer who wants to customize Atom will still have to deal with Intel. This deal with TSMC adds a new flexibility to Intel's model, and in the process, Intel also gets access to TSMC's processes, intellectual property, libraries and design flows – the ingredients that let customers tweak their chips.
Could Intel have offered that flexibility without TSMC's help? Not anytime soon. In a way, TSMC is the perfect partner for Intel's needs. Because it's a foundry that makes chips for a legion of customers, TSMC is like a short-order chef set up to make custom meals. Intel, on the other hand, is like a gourmet chef set up specifically to whip up his own menu.
It will be a while before we see whether these two chefs can work well together. Both companies said they're eager for TSMC to start serving up Atom chips, but there are still a lot of business and product details to work out.
And then there are the cultural issues that are sure to arise. Just think: All its life, Intel's been top chef in its own restaurant. It may find that working with a partner feels like too many cooks in the kitchen. (INTC) (AMD) (TSM) (AAPL) (QCOM) (NOK) (MOT) (NVDA)
Intel will find it hard going trying to shift ARM out from its number one position in the embedded space (mobiles, handheld devices, etc.). They tried it once and failed I believe. They have to get over the fact that power and performance is vital in the embedded space. Something ARM have long perfected. Intel will have too move fast, otherwise they will find out that they have missed the party. ARM is the big boy here so Intel will have to get use to being called the new boy on the block.
Your missing the cost side of things. Intel factories are not as cost competitive as TSMC. Putting a low margin product there and keep higher margin products at Intel is what this is really about.
Intel isn't necessarily sharing it's vaunted 'secret sauce' which if you are tech-savvy you'd know was the manufacturing process and high-k 45 nm process. It is mainly sharing the manufacturing load. The title of this article is misleading.
I've read several of Mr Fortt's article… and I get the sense he's dumbing it down for the non-technical readers (I hope). However, I'm not concerned with the watering down of content, but if it's inaccurate… the editors should correct.
The Intel brand has been placed on many "chips" that were mfg outside of their fab network. Examples include flash memory by Sharp (many many years ago), almost all of their communications processors at one-time or another (TSMC, UMC, XFab, Atmel), and the rebranding of chips mfg by someone else.
At the end of the day, we don't have enough information of the deal to truly understand the implications. Is there a royalty fee or rev share? what are the use limitations? is there a vol cap? what are the termination clauses? contract expiration / renewal date(s)? etc. Without answers to these questions (and more)… everything is just speculation.
HUGE LOLs to tom in seattle…
"people who like starwars know there are only 3, true starwars movies…"
what a crock of BS…for somebody that 'THINKS' they know starwars, you speak like you've never seen a single one.
i suppose the only reason you think the first three were good is because you were an impressionable, dim-witted, booger picking little kid for the first three…..
well, i'm most certain you don't know that the starwars saga actually consists, as it did originaly from the very first starwars in the 70s, of 9 episodes….not 3….not 6, but 9!!!!
(yes, there are 3 more in the mind of lucas but i wonder if it'll take another 20 years for those to be made)
perhaps you were to enamoured as a simple-minded, easily influenced person to realize the saga is more than the three movies you like.
the last three, episode 1 through 3 were absolutely great movies!!!
(as good if not better than the original three)
of course, i'm not caught up in my own selfish, ignorant, simple-minded point of view so that i express to the world my narrow-minded view of such things as movies nonetheless.
no, i figure you dislike the new starwars movies because you didn't get the same nostalgic euphoria coming over you that you eagerly anticipated the minute you heard the fourth starwars movie was being produced.
(it just goes to show….a lemming is a lemming is a lemming and they just dont change….but only keep on blindly following the others!)
Yes, the 80286 was outsourced to AMD back when customers needed to feel that if the chip manufacturer went down, they wouldn't be stranded. More relevant is the deal with IBM wherein the 80386 core was traded for ASIC design capability for exactly the same reasons as this partnership has been announced. The fact that the author did not know this is a testament to how fruitful it was for both companies.
Brilliant observation, Tony, but I doubt you have to worry about the author watching a lot of star wars (or at least the newer, crappy ones).
Anyhow… the cooking analogy was working… and then you killed it. Seems like the last three paragraphs were just a lead-in to a corny joke.
From Jon Fortt: Last three grafs? Maybe the last two.
People who like starwars know there are only 3, true Starwars movies…and they don't feature jar jar binks.
This move makes no sense. TSMC is a mfg house, not a custom-design firm. If customers want to customize they will still have to deal with Intel, so what are they gaining? Another level of red tape it sounds like.
Intel Is outsourcing the biggest growth opportunity in years, and will have little to fill it's fabs with (due to declining PC market), for what? To save customers some shipping costs, and give them a warm fuzzy sense that Intel wants to work closer with them?
Atom still uses a ton more power than ARM, and power consumption is the critical success factor for mobile devices. So this move smells of trying to fit a square peg in a round hole. ARM based solutions are going to eat Intel's lunch in the smart phone/MID space.
Intel just committed $7B to factory upgrades. This expenditure and factory underutilization will contribute significantly to declining margins over the next few years.
Interesting announcement, given Intel said the following at the Goldman Sach’s Technology and Internet conference on February 25, 2009:
"Question: Your thoughts on their (AMD’s) ability to compete with Intel on an outsourced model?"
“Well, it depends what they choose to do. There’s no question that the Mubadala guys will want a return on their investment, and if you read the details of the contract, which are public, it’s a cost plus contract, and I don’t think they can run their factory any cheaper than AMD ran their factory. So, at a minimum, their costs go up, and then they can decide whether that hits their margins or whether the pricing is less aggressive then it has been – they have a choice to make. The choice will be bounded by their ability to have competitive product, which gets tougher and tougher to do the more you decouple your design from your fabs.”
Minute 31 here:
http://cc.talkpoint.com/gold006/022509a_mg/?entity=10_UEYXDEV
This article is misguided. Intel has licensed nothing to no one as of yet. They have merely agreed to port a few variants of the atom design from Intel fabs to the TSMC process. If you want to license the design, you go through intel. Why did they do this? Simply because Intel is not a merchant semiconductor manufacturer.
Let me give you an example. If Cisco ever wanted to use the atom core in a product, today they would be limited to buying a fully packaged discrete part from Intel. This part may not be 100% ideal for the application but Cisco will have to live with it because Intel will not customize the part unless you are a marque customer (think Apple) or are buying in huge volumes (think Apple).
By licensing the core design to Cisco, Intel is effectively handing over the design and packaging to Cisco. Cisco will have a lot of flexibility in say the package dimension, the speed of the part (low power/low speed vs high power/high speed), etc, etc, etc. And Intel will not have to go through the hassle of tailoring the part for each and every customer. The upside for Intel is that the licensee is a new customer that Intel would not otherwise have had AND Intel doesn't have do a thing about guaranteeing yields. Win-win for TSMC/Intel/customer.
This agreement is nothing but a new revenue stream for Intel. It's unclear though how much this will actually bring in. ARM/Marvell, TI, and to a lesser extent MIPS and Renesas/ST (SHX core) have the market locked up for embedded CPU cores. These vendors span the range from the very high end (2 GHz ARM/Qualcomm Snapdragon) to the mundane (Renesas/ST SHX found in almost every laser printer in the world).