Is AMD about to take up ARM?

Several recent stories have jump-started rumors that AMD will design and sell a processor based on ARM's technology. This would arguably be the fastest and least-costly way for AMD to develop a competitive chip for smartphones and tablets. But there are also lots of reasons why AMD may decide it's safer to stick with x86 for now.

The rumors began with reports that AMD was looking for engineers to work on Android. Then AMD announced this week that an ARM official would deliver a keynote at its first Fusion Developer Summit this June. Finally during a quarterly earnings call this week ARM execs confirmed they have long been courting AMD. "AMD is a company that is very capable of deploying ARM technology," CFO Tim Score said.

ARM is now downplaying the comments, but there is no doubt that AMD now sees mobile devices as a big opportunity. At one time AMD had a division that developed technology for mobile devices but it sold the unit in 2009 to Qualcomm, which now uses the graphics cores in its Snapdragon processors. This was part of broader strategy, which also included spinning off its manufacturing into a separate company known as GlobalFoundries, to focus on AMD's core market of x86 processors for PCs and servers and high-end graphics.

Then earlier this year AMD abruptly changed course. The board dismissed CEO Dirk Meyer reportedly because it felt management wasn't moving quickly enough on a strategy for mobile devices.

Both Intel and more recently AMD have been developing x86 processors that use less power--at the expense of performance--in an attempt to get into smaller devices. Intel's Atom chip created the netbook category, but so far it has made little headway with tablets and smartphones. Earlier this month Intel announced a new Atom platform, code-named Oak Trail, for tablets running Android, Windows and MeeGo. Intel manufactures these Atom processors using an older 45nm process, but it believes future 32nm designs will be suitable for high-end smartphones.

Meanwhile AMD designed a low-power Fusion platform, code-named Brazos, based on an APU (Accelerated Processing Unit) that combines a CPU and Radeon graphics on a single chip. There are two Brazos APUs, the C-Series (Ontario) and E-Series (Zacate), both manufactured by TSMC, a semiconductor foundry in Taiwan, using 40nm process technology. AMD has sold more than 4 million of these low-power APUs, but like Atom nearly all of them are going into netbooks and low-end notebooks, not tablets.

Like Intel, AMD could continue to work on its x86 platform, moving it to a more advanced 28nm process at either TSMC or GlobalFoundries to reduce power and porting it to Android, in order to make it a more compelling tablet solution. AMD already has some 28nm prototypes, some of which may be the Krishna and Witchita APUs that will replace the C- and E-Series, respectively, in 2012. Or it could switch to ARM, an entirely different architecture.

ARM isn't a fabless chip designer like AMD (or Nvidia) nor is it a manufacture like Intel. Rather it develops the recipes for processor cores and supporting technologies and licenses them to companies that design chips. The vast majority of tablets and smartphones already use ARM-based processors. Furthermore ARM already works closely with TSMC and GlobalFoundries--AMD's two partners--to make sure its technology works well with their manufacturing process down to 20nm. So it should be relatively easy for AMD to come up with a competitive 28nm ARM-based SoC (System-on-Chip) for tablets and smartphones, leapfrogging Intel.

It looks good on paper and it must be tempting, but there are several reasons why AMD may not go down this road.

First, the time and resources required to develop an ARM SoC depend on whether you simply want something off the shelf or need to add your own special sauce. A small fabless company in China can license ARM's cores and Mali graphics (many of them are doing just that) and have a decent prototype SoC from TSMC in a little more than six months. But Qualcomm spent years developing its ARM-based Snapdragon because it is highly customized. AMD wouldn't need to build an SoC from the ground up like Qualcomm, but it would almost certainly want to use its own Radeon graphics, and that integration would take real work.

Second, ARM is a RISC design that uses a different instruction set. Microsoft's recent announcement that it would develop versions of Windows 8 and the Office productivity apps for ARM is a big win for the platform, but AMD would still need to do a lot of software development whether it chose Android, Windows or both. Many semiconductor companies such as Marvell are even taking things a step further and adding their own platform and apps supposedly to help customers differentiate their Android devices.

In short, it takes a lot of hardware and software engineering to support two different architectures. Even Intel couldn't juggle both x86 and ARM, and ended up selling its XScale PXA mobile processors to Marvell in 2006. Since AMD already has a promising low-power x86 platform, it could make more sense to focus its limited resources there.

But the biggest issue may be that the market for mobile application processors--while growing at a rapid rate--is already very crowded. AMD would need a new salesforce that knows how to sell application processors to companies such as HTC, LG Electronics, Motorola, Nokia, RIM and Samsung. And they would be competing not just with Intel, but with Marvell, Nvidia, Qualcomm, Samsung and Texas Instruments--all of whom offer high-performance ARM SoCs. It takes a long time to design and qualify a tablet or smartphone-unlike a PC--so once handset makers choose a processor line they tend to stick with it. The experiences of both Intel and Nvidia show just how hard it is to break in to this world.

Ultimately AMD doesn't have to choose.

In the short term, it will almost certainly continue to develop its low-power x86 APUs, which already have some traction. The C-Series seems like a good fit not only for netbooks but other mobile devices and if AMD can get Android running on it--like Intel has done with Oak Trail--you could see it or its successor, Krishna, in a few tablets over the next 12 to 18 months.

In the long-term, AMD could also offer an SoC that combines ARM Cortex-A9 or Cortex-A15 cores with Radeon graphics to get into more tablets, as well as high-end smartphones and other consumer electronics. This would give AMD a product line that covers the entire continuum from handsets all the way up to servers. But even if they ultimately do it, it will take time and I don't expect to see an ARM-based processor from AMD in a tablet anytime soon.