SAN JOSE, Calif. - ARM Holdings plc and Intel Corp. are on a collision course in the systems space.
The winner? Intel, according to one analyst. Another analyst disagreed.
''So far, Intel is missing the wave in the ultramobile market and is not likely to gain traction in 2011,'' said Auguste Gus Richard, an analyst with Piper Jaffray, in a recent report. ''The issue for Intel is that software developers are increasingly focusing on ARM platforms. As this momentum grows and the number of applications grows it makes it more difficult for Intel to turn the market in its direction.''
There are other issues. ''We do not expect Intel's 32-nm Medfield platform for smartphones, expected in 1H:11, to gain significant share in 2011 and perhaps 2012. We understand Intel is dropping legacy x86 features in order to lower power consumption in future products, but it is not clear if this is Medfield or the next generation,'' he said. ''We think Intel may have to abandon its legacy x86 architecture or significantly improve its SoC design capability to be competitive. We think the train is leaving the station in the ultramobile era and so far Intel is not on board.''
Intel, however, is leading in some respects. ''We expect that the lead customers for ARM's new A15 core to tape-out around the end of Q1:11. We expect the ARM camp to be in production at 32-nm in 2012 as Intel is moving to 22-nm. Intel leads in manufacturing and ARM leads in mobile architecture,'' he said.
''While we believe over time Intel's process advantage could drive a performance advantage, we do not think this is likely to occur in 2011. Moorestown is facing tough market acceptance due to high power consumption. Moorestown is a two chip solution with the Atom processor (45-nm) on one chip and the I/O (65-nm) on the other,'' he said.
''We understand that the I/O consumes a lot of power and the combination is not likely to be successful in the mobile phone market. The Medfield mobile processor will integrate the I/O with the processor and will be implemented in 32-nm. This will likely significantly improve power consumption and performance. Medfield is
not expected until mid-2011, and we would not expect Medfield based products to enter the market until 2012,'' he said.
The other problem for Intel? The PC market is slowing. ''We expect PC unit growth to decline from 12 percent to 5 percent over the next several years as tablets gain momentum. However, we also expect PC prices to stabilize after 20 years of an 8 percent annual decline. The net is we expect PC and Intel's revenue to grow mid single digits unless the company can move beyond the PC market,'' he said.
Fundamentally, Intel is a company that understands the need to re-invest for the future. If the leadership recognizes the need to change their product mix, I think they will manage to pull it off. Their management is not hamstrung by the accountants as so many other large mature companies are.
Also, in response to the observation that "Intel gears their process towards performance," that is more a function of what we predominantly see from their major product lines. It is useful to look at some of the papers Intel has published. With the control and long range planning of their semiconductor process development, Intel has been putting a lot of emphasis into expanding each technology platform for use in low power and RF applications. The product people at Intel have control over all the process knobs they want, thanks to in-house manufacturing. When they see the right market opportunity, they will be able to deliver the technology to support the product designs.
They would have to ditch their x86 architecture for that, which would take out one of their most appealing value propositions: backward compatibility. Intel would then have to go through a learning and adjustment process to be an ARM-like. This would take large amounts of money and resources with no guarantee of success in the end. I am not too optimistic about Intel's prospects. I think they have started to move a wee bit too late.
I think it's still too early to write Intel off the embedded/mobile world. Intel has enough resources to defeat ARM in that field.
IMHO, Intel needs to do the following to succeed:
1. Be more aggressive on power efficiency with their embedded processors, even if that means completely dropping the legacy code support. Most x86 legacy code was built for Windows anyway. The lack of legacy support won't affect much for embedded systems.
2. If 1 is not enough, adopt a new, RISC architecture to design their next generation embedded processors. The Atom architecture is still CISC, so it can't compete with ARM's RISC in power efficiency. A new, RISC architecture, helped by Intel's advanced process technology, would surpass ARM-based processors in power efficiency.
3. Intel could then port Linux, Android, and other embedded operating systems to its new embedded processor. Intel has enough expertise and resources to create a new software development ecosystem around its new embedded processor.