Intel (INTC) is hiding something, and as a technology geek and as an investor, I'm not particularly happy about it. Intel hasn't said a peep about its 14 nanometer process technology other than that it's in full development and that it would be ready for production in Q4 2013. This leads me to believe either one of three things:
- The process isn't really all that great and may provide minimal scaling/performance improvements over its 22 nanometer process
- The process is business as usual, with the standard power reduction and performance improvements. No need to brag, particularly as its competitors will just be catching up with its 22nm process at their "20 nanometer FinFET" nodes that are (somewhat misleadingly) called "16 nanometer"
- There's something big coming
Keep in mind that this article is somewhat speculative in nature, but investing is about connecting the dots with established facts -- even if Intel hasn't been great about spilling the beans.
Why This Is Suspicious
Intel has a tendency to "brag" about its next generation process technologies. At the 45 nanometer generation (products launched in 2007), it was shouting from the mountaintops that it had a fully functional high-K metal gate process (TSMC (TSM) and GloFo didn't ship their own HKMG products in volume until 2012, and even then the bulk of their outputs have been non-HKMG). At the 32 nanometer generation, Intel talked up its superior drive currents and tight contacted gate pitch (and, again, was shipping products ~2 years ahead of its nearest competition). And, of course, Intel held an entire press conference back in May 2011 to show off that it not only had a fully functional 22 nanometer process, but that it had full, multi-billion transistor products actually built on it (in 2011, Intel showed off both "Ivy Bridge" as well as "Haswell").
Regarding 14 nanometer, Intel has been suspiciously quiet, so that leads me to cycle through the three "possibilities" that I listed above.
Is It Lackluster And/Or Not Going To Provide A Density Improvement?
The first possibility is that the 14 nanometer process may just be terrible and provide no cost/transistor or performance benefit. Well, luckily Intel has given us enough information to render this claim false.
Performance Improves, Active Power Goes Down
We see from the slide above, the active power per function goes down, which is useful for mobile devices/power sensitive applications. But what about transistor performance? Does that go down?
Well, nope. The claim is that transistor performance increases at every given leakage value from the 22nm generation.
Next, we know that while the foundries expect no density improvement at the "16nm FinFET" (again, it's really "20nm: Revenge of the FinFETs"), Intel actually expects a rather marked improvement in density -- better than what it saw from 32nm -> 22nm.
For investors, the most important question is, "does cost per transistor go down?"
The answer is also a "yes."
So we know that the 14nm generation is expected to be quite good, and that while the other foundries (TSMC, GlobalFoundries) are likely to simply match Intel's 22nm performance characteristics at 20 nanometer FinFET ("16nm"), Intel will jump ahead another generation.
Wait, Wait, Wait - That's A Bold Claim
You've probably noticed that I'm making a rather bold claim that at "16 FinFET" the foundries are likely to catch up with Intel's 22nm in terms of transistor performance. This claim is not baseless; it is based on TSMC's very own slides given at a recent investor presentation:
Notice that the improvements cited over the 28nm HPM (this is the mobile high-K metal gate process) are 38% more speed at the same power OR 54% improvement at the same speed. Does this sound familiar? Indeed, these are the numbers that Intel gave in 2011 when describing its 22nm FinFET process improvements over its 32nm HKMG generation:
Now, it is well known that Intel's 32nm process generally exhibited higher performance than TSMC's 28nm HKMG process, but TSMC's 28nm process was denser. For a quantification and justification of this claim, see the following chart comparing various 28nm/32nm processes (the chart is courtesy of David Kanter's RealWorldTech):
We can see here that Intel's 32nm HKMG process, at constant leakage and voltage, exhibited superior saturated drive current. So assuming that Intel's 32nm process exhibits superior performance metrics against TSMC's 28nm HKMG process, and given that the percent improvements cited as a result of moving to FinFET on TSMC's part and Intel's part for their "16nm" and "22nm" processes, respectively, it stands to reason that TSMC's "16nm" process will be comparable in performance to Intel's 22nm process, albeit it will have a density advantage.
The broader implication is that at the "14nm/16nm" generation, Intel will have a significant performance and power advantage, as TSMC's "16nm" is really "20nm FinFET" while Intel's is a next generation process following its own "22nm FinFET."
With that out of the way, let's get back on track.
Is Intel's 14nm Process Business As Usual?
The first possibility is that Intel's 14nm is simply the logical continuation/shrink of the 22nm process node. Performance improves as usual, power improves as usual, and no real special "tricks" come into play here. We know that the 14nm node is also FinFET (so no TunnelFETs or Gate All Around devices just yet), but the words of Intel Senior Fellow Mark Bohr continue to echo through my mind:
In this era, where you continually have to invent new materials and new structures, it's a lot tougher being a separate foundry and maskless design house. Being an IDM, we have design and process development under one roof. That's really a significant advantage.
To keep performance, power, and size scaling, Mark Bohr is very clear that new materials and structures need to keep being invented, so I'm not yet ready to assume that 14 nanometer is "business as usual."
Intel May Have Something Big That It Doesn't Want To Share
The other possibility is that Intel has another high-K metal gate/FinFET level innovation baked into its 14nm process that it doesn't really want to talk about. See, Intel sort of gave away what its competitors should work on -- FinFETs -- when it boasted that it had gotten them ready for high volume production in 2011. While all of the major foundries research all of the same things (TSMC has published frequently on FinFETs, for instance), there is a huge leap between having a theoretical device in a lab and being able to take something to high volume production with good yields within the right timeframe. In this business, there are a lot of smart people, but having them explore and then develop the right things at the right time is incredibly hard.
If Intel were to give its next generation innovation away too early, then others would know what the "right" thing is to work on and then focus all of their R&D efforts there. Who knows what TSMC and GloFo would have been working on if Intel had shut up about actually getting FinFETs in high volume manufacturing?
I think that Intel will finally disclose what it's doing at 14nm at IDF 2013 in San Francisco, particularly as "Broadwell" -- the first 14nm product -- is scheduled to ship in 1H 2014 for Ultrabooks and fanless tablets:
Further, I expect that Intel's next generation processor for low power applications known as "Airmont" will debut in 2H 2014, which should arrive at a time while competitors are just shipping their first 20nm planar chips (this means Intel will be at least 1 generation ahead in density and 2 generations ahead in terms of actual transistor performance):
I don't know about you, but I'm darn excited for what the future holds for Intel.