"It depends on what the meaning of the word 'is' is."
President Bill Clinton - Impeachment Deposition - 9/13/98
Those of us who follow the semiconductor memory industry are accustomed, from long and (mostly) bitter experience, to the random alarms and rumors that noisily blow through our investments in the stocks of Micron (NASDAQ:MU) and SanDisk (SNDK) like a sudden gust in a grove of late Fall trees holding on ever so tenuously to their fading golden leaves. It's one of the unavoidable risks of our focus industry and the stock market, and we're used to it. Life goes on, and happily for us, our grove has seen a good season of growth and we are very confident of many good seasons to come. From time to time, though, another kind of wind gust ruffles the grove and this time the feeling of the air on our skin is weightier and portentous - heavy weather may be in the offing. It's at times like this that the technology investor has to really perk up and take a hard look around, lest an early winter storm do some significant damage to hard-won gains.
We have come to that time. The clouds are swirling around Mount Intel's (NASDAQ:INTC) summit and we semi-tech memory investors can't take that lightly. Our own Russ Fischer has been telling us for some time that Intel has plans for NAND, and these plans may extend well beyond the confines of their long-time partnership with Micron in IMFT. Just two days ago Russ tied a lot of threads together with his analysis of Intel's investor meeting. Those few of you who have not yet read Russ' latest ought to take this opportunity to do so. I'm going to abstract his arguments but there's no substitute for the real thing.
In essence, Russ argues that Intel's move into 3D NAND and subsequent generations of high-performance memory is a strategic inevitability for the following reasons:
- From a business perspective, Intel's size means that real growth can only come from very large addressable markets. Storage, an estimated $65B market, fits the bill.
- The best defense is a great offense. Intel dominates (98% share) the data center server market and will defend that market at all costs. The problem is that the ever-faster CPUs that Intel manufactures are making less and less of a difference for customers. Because of the phenomenon of the CPU/Storage performance gap, substantial ongoing gains in server performance can only come by architecting and provisioning a new type of faster storage which will allow precious CPU cycles to be used more efficiently. Your CPU may be a bullet train, but it's really not going to make a lot of difference to your speed of travel if the train is on 19th century rickety rails. It is in Intel's strategic interest to build new 3D tracks as fast as possible and not wait for others to get around to it.
- Unacceptable technical risk. The "convergence" of logic and memory is a technical necessity that will bring 3rd parties uncomfortably close to Intel's CPU domain. In essence, the real solution to the problem posed in point 2 above is to have memory that is both fast and closely coupled with CPU. Indeed, no less than Intel's own IMFT partner, Micron, has argued that the new industry architecture will be "memory-centric," not processor centric. Intel has argued that just the opposite is true, but it would hardly be prudent to face the risk of being wrong by not hedging the bet and being big in the memory business.
- The solution to the problem lies squarely within Intel's core competency. In effect, the design and fabrication of advanced NAND and next generation memories represent the kind of problem that Intel is supremely (if not uniquely) capable of solving. Need some proof? Rob Crooke, Intel's VP GM of the NVM Solutions Group, claims in his presentation that Intel has already achieved a "breakthrough" with the "First 3D NAND with Disruptive Cost in 2015" (slide 49).
Did he just say "disruptive?" Sounds ominous, doesn't it? Particularly if your company name is anything other than Intel (and possibly Micron). "Disruptive" is, after all, one of those pregnant technology words on which worlds turn, entire industries emerge or vanish, companies implode or explode, and investor fortunes are made and lost.
So ok, Mt. Intel has rumbled, and we have felt the ground shake and the wind blow and we are fully engaged with all of our senses working to the max. Now is not the time for panic - necessarily - so let's take a moment to look at the really big picture and see what we can glean from it.
Several points immediately come to mind:
Intel is already in the memory (planar NAND) business with Micron through the vehicle of IMFT. Per the Intel Investor Meeting slides of Rob Crooke, (slide 54), we are reminded that they own the lion's share of the Enterprise SSD market, with a share greater than double SanDisk/Fusion-IO's. This being the case, Russ' argument devolves to one of scale, not of kind. Does it necessarily follow that Intel can achieve outsized scale in the NVM market? Investors in ARM (NASDAQ:ARMH) and Qualcomm (NASDAQ:QCOM) have surely withstood the Intel juggernaut and lived to tell a very profitable tale in different market segments.
You say "disruptive," I say "cost crossover." Here I refer to the respective statements of the two IMFT partners, Intel and Micron. Here is Scott DeBoer, Micron's VP of Technology at Micron's Summer Analyst Conference:
"Our first product will be a 32-tier 256Gb die [which is initially MLC but soon thereafter will be TLC…] One of the unique things about this technology is that not only does it provide outstanding performance […] but it also provides us with what will be the highest Gb per area density in the industry. [... We envision…] "3D product cost crossover with 16nm in late 2015" [and a] more historical type cost reduction after that."
Here's another comment from Mark Durcan, Micron CEO, at yesterday's Credit Suisse Technology Conference: (this is lengthy but the whole exchange is well worth a careful parsing.)
John Pitzer - Credit Suisse
Q. "Then Mark in general when you think some of your peers and their experiences with 3D over the last 12 to 18 months that was sort of lot of hype and then it kind of slowed down. How comfortable are you or what comfort level can give us as you kind of work through kind of the yield issues or the conceptual legal issues around 3D NAND that this is going to be sort of more difficult transition than you anticipate?"
A. "So I have said for a while now that we were very happy and comfortable with our 3D NAND technology position. This isn't like there was some breakthrough and all of a sudden we are pedal to the metal and there are more dominoes to fall. We have been relatively consistent and methodical and investing in this technology for a considerable period of time. And we are very, very comfortable with where we sit and our ability to execute. The timing is about what we said it was going to be since the beginning that it would be a second half of 2015 phenomenon. That's really where you will start to see the first manufacturing output from Micron out of the existing Singapore fab and then over time as we fill up the white space there we will bleed into the additional space that we are building." -
SA Transcripts, - 12/3/2014 (emphasis added)
This language of caution regarding 3D roll out velocity and difficulty is very consistent and entirely congruent with the public comments of Micron and every other memory purveyor and industry consultant over the course of the last few years. Note particularly the comment from DeBoer regarding 3D NAND cost trajectory over time. SanDisk has consistently mirrored this language as well. Bottom line, I hear Durcan telling investors and analysts to tap the brakes. Clearly the IMFT partners are at loggerheads over 3D.
Notwithstanding the notably less enthusiastic Micron position on 3D, the achievement of "cost crossover" of the IMFT flavor of 3D NAND with the 16nm Micron planar node is a tremendous achievement. Even more tantalizing is the consistency of the statements of the two partners regarding the performance characteristics of the node. Here is Durcan from the Credit Suisse conference:
"We think we are going to have technology that addresses a broad swath of market and segments with some interesting performance parameters and other competitors won't necessarily be able to match. And we also think we will have very, very tough competitive offerings with the very high density in terms of bits per square centimeter of silicon, not only for our initial deployment of vertical NAND but as we move through subsequent generations after that."
I don't know about you but as a Micron investor I'm excited by that quote. As those of you who follow me know I have been quite intrigued by the potential of Micron's 3D offering to bring them out of the woods and into the light of profit recovery in this business. But when? Let's assume everything the IMFT partners are telling us is true - when can the bottom lines of the companies start to see real benefit from this new product? One thing for sure, the Enterprise/hyperscale market customers are a tough lot to introduce new product to. Here's Kip Bedard of Micron speaking at this year's Raymond James Conference:
"[The Enterprise and Hyperscale buyers…] want a thousand drives running a thousand hours [… and] if you fail somewhere […] you have to start over. It's going to be a long qualification time […]."
Bottom line, it may well be late 2016/early 2017 before this new technology gets real traction in the target market.
Now we turn to the question of quantity and cost. How many bits - at what cost? I'm going to address cost in detail in the next article, but there is one top-level point that is pertinent to this discussion. What, exactly, does "cost crossover" mean? Does it mean that as of pilot production in late 2015 the 3D bits coming off the line will cost less to make than the now mature 16nm planar node bits? The answer is surely not. Costing in the memory business is a function of the learning curve, and it depends on timing and yields at full production. With planar learning curves running at a year and 3D learning curves estimated at 2-3x planar, we are potentially talking about 2018 before the 3D bits being shipped are actually cash flow positive for Intel and Micron. Regarding quantity, the answer is not many, not soon. Durcan is clear regarding Micron's ability to supply 3D, which will come out of the present Singapore fab in 2016 and a new fab at that location starting in 2017 but only reaching substantial quantities in 2018. Can Intel start up production in one of its underutilized Fabs? Certainly, but then the central question will price maintenance in a suddenly unhinged oligopoly. This is well-traveled ground but I will return to it in the next article as an aspect of the costing discussion.
Finally, competition. Most analysts have concluded that Samsung's (OTC:SSNLF) "VNAND" offering is not cost effective but no one should underestimate their capability to come up with a fix. SK Hynix's 3D product is on the same schedule as IMFT's with customer samples available in 2H of 2015. SanDisk/Toshiba's BiCs offering is the real dark horse. SanDisk is extremely confident that their technology will be cost superior to any offering when they production ship their second generation of 3D in 2016. It is important to note that, except for Samsung, this is all speculation and vendor hype. No one has seen the 3D offerings of the other 3 MIPO suppliers, so investors are going to have to exercise some patience. The central point is this: Intel and Micron are not going to be the only game in town when it comes to high performance 3D.
So what are we to conclude from all this? Personally, I think three points stand out:
- Intel has good reasons to be a bigger player in the storage market in order to grow its business and protect and expand its enterprise/data center market domination.
- Right now the IMFT partnership of Micron and Intel seems strong and Micron will be a big beneficiary of it. Long term, who knows?
- SanDisk, being NAND only, has the most to fear from Intel's newly aggressive posture in 3D. It is also the strongest overall player in NAND and has the greatest ability to respond to IMFT's threat.
- Nothing is going to happen soon. Investors will want to watch closely for 3rd party reporting on the IMFT 3D product when it becomes publicly available in 2015.
So I guess it all comes down to what the meaning of disruptive is. Right now we can only guess and wait. Join me now as we trudge to the base of Mt. Intel and wait for the fog to clear.
Disclosure: The author is long MU, SNDK, INTL.
The author wrote this article themselves, and it expresses their own opinions. The author is not receiving compensation for it (other than from Seeking Alpha). The author has no business relationship with any company whose stock is mentioned in this article.