Friday, April 25, 2008

Apple Buys PA Semi—What Was That About?

Apple sent shock waves through the industry with its acquisition this week of PA Semi for a reported $278M. Does this mean that Intel—who had great hopes of seeing next-generation iPhones and iPods based on Atom—and Samsung, who make the iPhone’s ARM-based applications processor, are hosed? Basically, yes.

PA Semi’s PA6T-1682M is a 64-bit low power dual-core version of the PowerPC architecture that Apple used in its computers before switching to Intel. Capable of delivering 8,800 Dhrystone Mips while running at 2 GHz, this chip is clearly much better suited to server blades and high-end embedded applications than it is for portable designs. Apple coyly let on that it was acquiring the firm for its IP and design expertise and had no interest in producing chips, at least for the merchant market. That caused a chorus of cries from mil/aero contractors, who have been scooping up PA’s chips as fast as they could acquire them for a wide range of (presumably rack-mounted) applications. Some have reportedly been whinging to the Department of Defense (DOD), asking them to block the acquisition so that the processors will continue to be available.

Apple’s lawyers can probably deal with any DOD objections by licensing generic versions of PA’s design to silicon manufacturers. What Apple apparently has in mind here is getting tighter control of the hardware, not to mention importing the profits that Samsung and ARM have been enjoying to date. Steve Jobs has long stated that Apple’s winning edge is tight integration between hardware and software, and this acquisition will enable new levels of hardware/software co-design—not to mention a scaled-down version of PA’s current chip that will fit perfectly into Apple’s next-generation of products. I’m sure TMSC will welcome them with open arms.

What does this mean for the rest of the supply chain? Mostly angst. This hits Intel at a bad time; they need some big design wins for Atom, and the iPhone/iPod socket would have really launched the architecture. Intel is counting on Mobile Internet Devices (MIDs) as the Atom launch platform, but MIDs are barely beyond the concept stage, and it’s unclear that they’ll get any real traction. OTOH, sub-notebooks seem to be coming back, thanks to the Asus Eee PC 701 (currently based on a Celeron processor), and this is a market where Atom could take off. So don’t sell your Intel stock just yet. Same for ARM, who have the luxury of being ubiquitous in the embedded and portable space. Shed a tear for Samsung, however. PortalPlayer can feel your pain.

Wednesday, April 9, 2008

CEO Interview: Tom Hart, QuickLogic

After a decade or two of competing head-on with Altera and Xilinx in what CEO Tom Hart refers to as a “Coke and Pepsi market,” QuickLogic looked for other ways to leverage its programmable technology. In 2007 QuickLogic reinvented itself as a purveyor of ‘customer-specific standard products’, or CSSPs. Building on their ArcticLink and PolarPro OTP platforms, QuickLogic offers semicustom parts tailored to customer designs, including proven IP blocks, custom logic and software drivers. Customers stock an inventory of products that may be 80-90% programmed for their designs and customize the rest as the need arises—a sort of ‘roll your own ASSP.”

QuickLogic is primarily focused on portable designs, where cost and low power are of paramount importance. Portable Design sat down with QuickLogic’s CEO Tom Hart to explore how and where this approach fits into the portable arena.

You can watch the video below:

If you'd rather just listen to Tom, click on the link below, or click here if you'd like to download the podcast.

The Road to Barcelona

AMD announced today that it is finally shipping its quad-core Phenom (3/G Opteron) chips, formerly known as Barcelona. Having spent the past year shipping triple-core versions with a crippled fourth core that it took them that long to fix, AMD is reportedly “aggressively pricing” the new chips. In other words, they’re sacrificing margin in an effort to claw back market share from Intel, who have meanwhile been executing flawlessly on their 45-nm quad-core designs.

AMD’s new chip only beats Intel’s on a few benchmarks, so they’re short on options other than competing on price—unpalatable as a tactic and disastrous as a strategy. This is a particularly painful choice for AMD, since they’ve been hemorrhaging money for the past few years and their cost structure is seriously out of whack. The latter means that even with big design wins from H-P, which they also just announced, the spike in revenue probably won’t get them back into the black, and certainly not for long.

AMD’s total operating expenses in 2007, excluding the $5B expense of acquiring ATI, amounted to 121% of sales compared to 79% for Intel. Part of that stems from high SG&A expenses—23% of sales revenues vs. 14% for Intel. When you aren’t selling a winner, it’s a lot more expensive to create demand than to satisfy it.

Meanwhile AMD’s 2007 R&D expenses amounted to 31% of total revenue vs. Intel’s 15%, pretty much the industry standard. Much of the excess had to go into frantically trying to fix Barcelona’s cache flaw; the rest probably went into trying to improve yields at 65 nm and solving new problems at 45 nm—both very costly undertakings.

AMD has promised to look into going fabless, something I’ve long advocated. TSMC is at least one generation ahead of AMD and always will be. Set up a team of designers in Hsinchu and transfer your process design costs—not to mention huge CAPEX—on to the fab. And sell your costly German fab, whatever the political costs. You can’t manufacture in Europe and make money. Just ask Infineon, Qimonda and ST (well, they’ll deny it; look at their balance sheets instead).

Going private is an attractive option, especially now that AMD has a competitive product again. But that option may be dead for now thanks to the turmoil in the financial markets. The Blackstone Group, who took Freescale private, is reportedly in dire straits, as are many if not most private equity investment and hedge funds. This would be a good road for AMD to take when it opens up again.

Meanwhile, Hector should have lunch with NVIDIA’s Jen-Hsun Huang to talk about a buyout. The alternative, if he waits too long, will likely be a major buy-in from Carl Icahn, who’s always looking for underperforming assets. And you know where that leads.

Wednesday, April 2, 2008

Is that the Internet in Your Pocket, or Are You Just Happy to See Me?

Intel today finally rolled out its Atom chip family at IDF Shanghai. This is the same chip family previously ballyhooed at IDF in San Francisco last year. Having never had a chip that could get them into cell phones, Intel is looking to Mobile Internet Devices (MIDs) to deliver a better Internet browsing experience than you can currently get on handsets. Can’t crack a market? Create a new one.

Atom is a good generation away from being able to deliver the same power profile, not to mention the performance, of an ARM Coretex-A8—and that’s if ARM stands still. While not doubting Intel’s engineering chops, it’s far from clear that consumers will be willing to pay $500 for yet another portable gadget to stuff into their pockets along with their cell phones and iPods. They don’t even need iPods anymore—the iPhone is an iPod with a great screen that can also make phone calls. And cruise the web nicely, thank you.

It makes sense that Intel would roll out Atom in China, a hot market for mobile devices and the only one that ARM and its licensees doesn’t entirely own. Good luck with the margin, guys.

[OK, I’ll lay off the MID after this post]