Intel Vs. Everyone Else

A report from ABI Research is starting to gain some attention. For cynics, the question of why now seems perfectly reasonable, considering the report was released early last month and promptly fell well under the semiconductor industry’s radar. But cynicism aside, it’s still interesting to compare specs for chips using ARM’s Cortex-A15, A9 and Qualcomm’s ARM-based Krait.

The bottom line is that Intel’s Saltwell chip is faster and uses less current with lower current drain, and that’s an issue that may or may not be improved over the next 12 to 24 months as everyone begins shifting to finFETs.

ABI’s comparison, though, is based on Intel’s “Saltwell” Atom processor, which uses a 32nm process. Intel didn’t start using finFETs until 22nm, and the next version of the Atom processor—Silvermont—a 22nm finFET-based processor wasn’t announced until May 6. If comparable improvements on other Intel chips are any indication, this should be a significant leap for Intel, and it will start showing up in devices this fall. It also should be an interesting comparison for companies considering or currently working with finFETs.

At this point, Intel is the only company commercially shipping finFETs, which it terms TriGate technology. That alone gives Intel at least a couple years head start with 3D transistor structures. Moreover, Intel’s regular layout structures can continue to be a threat to more complex architectures, even without the introduction of EUV lithography.

The foundries, which have been counting on EUV, are wrestling with double patterning and multipatterning issues, and they are still working on test chips involving finFETs. It will take time to improve yields and test the chips, which means the earliest we will start seeing mass production of finFETs will be sometime in 2014, and maybe even after that. That’s many generations in the consumer electronics business, and should be raising blood pressure among Intel’s rivals.

So what happens when the industry really pulls out all the stops? STMicroelectronics has bet heavily on fully depleted SOI at 28nm, because it still uses the same processes without the need for double patterning. It also is developing finFETs—possibly in combination with FD-SOI. And other companies are experimenting with stacked die, which greatly improve throughput with far lower power.

Which approach ultimately wins, and which companies benefit, remains to be seen. But for the first time in many years, Intel is back in the thick of where chips are sold with its laser focus on price and performance now firmly coupled with power. This is where the race will get very interesting over the next 12 to 24 months, and maybe for much longer than that. The sleeping giant has awoken, and where there is a proven path the rest of the industry surely will follow—even if it ultimately is combined with other paths and approaches.

—Ed Sperling