Intel Optane = Optical + Octane

| About: Intel Corporation (INTC)
This article is now exclusive for PRO subscribers.


3D XPoint definitely requires optical communication technology.

This is actually much bigger than I had anticipated.

Many more educational articles are forthcoming.

After my previous article, Intel Has No Plans For 10 Nanometer Chips, Seeking Alpha's own Mark Hibben politely posited precisely the opposite. His position was characterized by the following quote:

The power dissipation of the photonic devices made it impossible to achieve anything like the integration density that was possible with silicon-based integrated circuits. Silicon won, not because it was faster, but because it was the most energy efficient form of microprocessor, as well as the lowest cost way to build a microprocessor.

20 years later, these fundamental observations are still true, and it's the main reason why we're still on the Moore's law path of making ever smaller silicon integrated circuits.

My position was characterized by the following quote:

The advent of manufacturable nanoscale chalcogenide technology has Earth-shaking implications on half a trillion dollars across various markets annually. Intel and Micron are on the hook for confirming that it is manufacturable.

Essentially, I'm saying that the advent of chalcogenide (warning: audio link) into commercial semiconducting changes things (both optically and in many other ways) dramatically but Mark does not. The good part about this spat is that one of us is wrong and we'll hopefully know the answer by Q2. If Mark is right, then there's nothing to see here. If I am right, then there will be many billions of dollars changing hands in short order.

There's No Time

Unfortunately, there's not enough time for my usual witty-yet-unique rhetorical exhibitionism, clever double entendre and entertaining wordplay. You see, unlike Hibben, I don't have any formal education on this stuff other than my computer science background (which has absolutely nothing to do with hardware other than fundamental operation). But, for some decades, I've crammed quite a bit of informal knowledge into my noggin while following chalcogenide technology as an ECD shareholder.

The thought of regurgitating all of that is daunting. In layman's terms, perhaps impossible. But I feel an urgent need to try because, now that it is all coming together, I really feel like there isn't much time left before the technology hits the market and the story tells itself. If I can do this quickly enough, then I might be able to ride on the coattails of The Giant. This will take the form of several coffee-infused articles in just a short period - perhaps a month (more if time allows).

But, first, a proper rebuttal

Hibben inserted a red herring in his argument:

The power dissipation of the photonic devices made it impossible to achieve anything like the integration density that was possible with silicon-based integrated circuits. Silicon won, not because it was faster, but because it was the most energy efficient form of microprocessor, as well as the lowest cost way to build a microprocessor.

Photonics didn't lose - they only mostly lost. Vast swaths of technology (especially communications infrastructure) would not exist without photonic technologies. The Internet could not exist at today's performance and coverage levels without photonics, period. Home Internet service is largely delivered by fiber optics in a configuration known as fiber-to-the-node (which are often placed in individual neighborhoods).

They just use coax cable to connect the 'last mile' in the most economical manner. Fiber splicing and terminating is high-skill, but installing and fixing coax is beauty school dropout easy. Your cell towers are connected the same way with the last mile happening wirelessly.

It is all photonic.

More importantly, Intel (NASDAQ:INTC) Thunderbolt is photonic. It is real and it is shipping in oodles of devices - including most of Apple's (NASDAQ:AAPL) stuff. But it's on a separate chip. History buffs will point out that, due to integration trends driven by the inevitable invisible hand of capitalism, computers of today require fewer chips than those of yesteryear (given the same apples).

My argument is essentially that chip makers will always leverage a gainful opportunity to combine chips. We have two chips now. If possible, we should combine them into one. Chalcogenide makes that possible. Intel and Micron (NASDAQ:MU) are telling us that chalcogenide-silicon integration is now possible. Give me some rope here, but here are the economics of this possible integration:

  1. Much reduced cost
  2. Much increased performance
  3. Significant competitive advantage
  4. Zero downside (perhaps fab complexity?)

None of what Hibben put forth shoots down the possibility that chalcogenide enables processors with integrated photonics except a rear-view mirror view on the situation. The only forward-looking statement provided is that "chalcogenides probably fail both the efficiency and economy tests, except for certain applications such as 3D XPoint" without any evidence.

The contrapositive: "if chalcogenides don't fail both the efficiency and economy tests for most applications", then we're about to start a party which is worth hundreds of billions of dollars annually.

And only Intel and Micron are invited

Micron and Samsung (OTC:SSNLF) have both shipped chalcogenide-based devices to the tune of "tens of millions". Although the subsequent phones were shipped into the third world where people don't have the means to lawyer-up when the technology done fell over, the tech held-up just fine. It took some doing, but I managed to procure a Nokia Asha phone with Micron chalcogenide-based phase change memory almost 5 years ago. I don't use it for anything other than a keepsake for all of the money that I squandered in ECD. And I transfer some data around to check the durability of the chalcogenide phase change memory inside. So far, so good.

But perhaps that's too much insight into my lack of independence on this matter? Maybe the voices that tell me that all this was stolen from ECD shareholders is just stress-induced? Poor life decisions? Regardless, the point is that Micron previously told us that chalcogenides were easy and then they went off and proved it. Now they are telling us that they're tough.

Who do you believe? Micron from 5 years ago or Micron from today? It is important to note that Micron from 5 years ago believed that the demise of the other licensee on this technology (including the optical communication) was in dissolution. Was it a coincidence that Micron changed their tune about the same time that this entity was pulled out of dissolution?


I didn't want to cloud my previous article with senseless over-complication of the matter so I left some important things out. So, it proved helpful that Hibben stated the following:

Breezy claims that "chalcogenide switching is extremely high speed", based on a patent application he cites. I believe that Breezy has incorrectly interpreted the patent, as the patent does not refer to switching time per se.

First of all, the patent language is as follows (realize that these are Intel's words - not mine):

Snap back is a property of the composite memory [cell] that results in an abrupt (e.g., on the order of tens of picoseconds) increase in conductivity (and corresponding decrease in resistance) of the memory element. Sense circuitry, coupled to the memory [cell], is configured to detect the presence or absence of snap back in a sensing time interval. The presence of snap back may then be interpreted as a logic one and the absence of snap back as a logic zero.

Chalcogenide 101: Crosspoint memory differs from other memory technology in that it does not require a transistor to access and protect the contents of the memory. In order to prevent stray electrical current from damaging the content of non-targeted-but-adjacent memory cells, each cell is protected with an inline chalcogenide-based "selector". From the 3D XPoint press materials, you can clearly see the selector and the memory cell:

For clarity, I've changed the term "element" to "cell" in the patent language above. These are interchangeable synonyms and I wanted this to match this diagram. The memory cell is also chalcogenide. The only difference is that the selector is comprised of a stronger alloy which will not change its atomic structure under stress. The described "composite memory cell" is just the combination of the selector and the memory cell - nothing more.

Now, in rereading the patent language, they're reading a memory cell by targeting it with the proper voltages on the corresponding wordline and bitline (both indicated in white). Because they know that the cell will predictably "snap back" within "tens of picoseconds" if there is a '1' in the cell (as indicated in the diagram), they will necessarily know that, if there isn't a snap back, then there's a '0' stored there. They've outlined a deterministic cell read in just picoseconds.

Assuming that they can overcome parasitic chip capacitance and other overhead to do a full read in 100 picoseconds, that works out to 10Ghz. If you aren't a technical person, then you should realize that this makes the hair stand up on our necks.

Check Mate

Since it isn't possible to measure such high switching speeds using traditional equipment, Stanford University actually had to build a chip with integrated measurement equipment on-board so that they could ascertain just how fast it would go:

Their instrumentation detected that the amorphous-on state - initiating the flip from zero to one - occurred less than a picosecond after they applied the jolt.

To comprehend the brevity of a picosecond, it's roughly the time it would take for a beam of light, traveling at 186,000 miles per second, to pass through two pieces of paper.

It is nice to know that the Stanford University researchers have drawn the same implications that I have (where "TS" is "threshold switching" and "ovonic" implies "chalcogenide"):

The observation of picosecond TS is particularly important for ovonic switches, sometimes used as pre-selectors in memory devices [31], which then allow sub-picosecond access to the specific memory cell. In this sense, TS allows the design of ultrafast electronic switches.

So, yes Mark. Even though the patent did not refer to switching time per se, this is definitely what they're discussing. How do you transfer data that you can read at the speed of light?

With light.

Disclosure: I am/we are long INTC, MU, ENERQ.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

Editor's Note: This article discusses one or more securities that do not trade on a major U.S. exchange. Please be aware of the risks associated with these stocks.