In August of 2014, Intel (NASDAQ:INTC) rolled out their Broadwell line of processors - the first to use the new 14 nanometer manufacturing process. I believe that Intel had planned for this to be the last node to use silicon technology. Why? Because they told us. In 2005, Intel announced that they had prioritized Indium Antimonide ("InSb" for short) as a leading contender to replace silicon in transistors. From the press release:
The results of this research reinforce our confidence in being able to continue to follow Moore's Law beyond 2015. As was the case with other Intel technical advancements, we expect these new materials will enhance the future of silicon-based semiconductors.
They expected silicon to run out of gas in 2015 and began the expensive process to replace it more than a decade earlier. An article at the MIT Technology Review publication explains the complexities well: InSb is not an easy material with which to work and certainly not a foregone conclusion. But there is a noteworthy advantage that was outlined:
Compound semiconductors also have optical properties that could help speed up communication between transistors on a chip and multiple chips within a device. These materials easily emit and detect light - a characteristic that has been studied and improved for decades, says David Hodges, electrical engineer at the University of California, Berkeley. Therefore, he says, light emitters and detectors made of compound materials could potentially replace copper wires, which are a major "impediment of speed."
The clues start to line up like Scooby Snacks at this point because we also know that Intel expected to release their silicon photonics ("optical") in 2015 as well. While they did technically release standalone optical products for "select customers" some months ago, they've delayed the CPU-integrated Purley product for some years.
The difference is huge: current optical technology must be produced on a separate chip because the optical manufacturing process is incompatible with the CPU manufacturing processes. Since that separate chip is typically connected to the CPU by way of a slow electrical bus, the performance advantages are compromised.
All of that high-performance stuff at the other side of the fiber optic cable is now rendered much lower in performance by a slow, electrical chip-to-chip interface between the standalone CPU and photonics chip. This problem is illustrated by the 3D XPoint/Optane NVMe performance compromises. If Intel had an on-CPU optical link to Optane (or any high-performance memory like DRAM), performance would be orders of magnitude better.
Optical is the future. But so is tin foil.
During the optical announcement, Intel bragged that this had been in development for 16 years - surely, a competitive advantage. But what were they working on in the year 2000? It just so happens that they were working on chalcogenides with Energy Conversion Devices ("ECD"), a company founded by inventor Stanford Ovshinsky:
US Patent 6,687,427
December 29th, 2000
Chalcogenide materials are used in memory devices for their phase change property, i.e., a property that can be switched between a generally amorphous and a generally crystalline state by the application of thermal energy. The phase change can be exploited in controlling current flow. A further benefit is that such phase change is reversible.
In addition to undergoing a phase change (amorphous/crystalline) in the presence of sufficient thermal or heat energy, it is also observed that chalcogenide materials undergo a significant change in their index of refraction in changing, for example, from a generally amorphous to a generally crystalline state, i.e., in response to a temperature change. The change in index of refraction between an amorphous chalcogenide material and a crystalline chalcogenide material is on the order of 3 to 20 times. This is comparable to an index of refraction change to thermal energy exposure on the order of about a fifth decimal (10−5) for silicon dioxide.
It is important to note that chalcogenides (warning: audio link) are also useful in emitting and detecting light. And we know that Micron's (NASDAQ:MU) Guy Blalock disclosed that they're using chalcogenide technology in 3D XPoint. What else do we know?
- Chalcogenide is highly-compatible with traditional CPU logic "back end of line" manufacturing processes.
- Chalcogenide switching is extremely high-speed ("on the order of tens of picoseconds") as revealed by this characterization in an Intel patent application.
- Chalcogenide photonics are useful in the extremely low-latency broadcast of data within a neuromorphic AI system.
- LinkedIn reveals lots of chacogenide photonics work at Intel.
- The two fundamental chalcogenide switching patents were filed by ECD in March of 2003 (US6967344 and US7186998) so they're valid into 2023.
- ECD did not disclose this technology for more than three years after the patents were filed.
- Intel announced their InSb goals just months before the latent ECD disclosure.
- These patents were transferred to Ovonyx in 2009 (an act of which I believe to be theft).
- Micron now owns Ovonyx.
- Micron has implemented a poison pill in order to avoid acquisition.
- After nearly 5 years, the ECD bankruptcy has yet to finalize.
What do I believe?
I believe that, after ECD developed chalcogenide transistors in 2003, they quietly went into licensing or sale negotiations with Intel. This is why the patents and trademarks were filed in 2003 while the technology wasn't disclosed for another three years. Ultimately, Intel bluffed with their InSb transistors and this prompted ECD to call that bluff by finally announcing their technology.
But then they forced Stan out of the company (he never retired) and transferred the patents to Ovonyx which was subsequently traded to Micron for some Olive Garden gift cards. I believe that this Ovonyx transaction was coordinated by Intel (please note that Numonyx should not be confused with Ovonyx but you should read into the similarity) while Micron remained in the dark. Now that Micron realizes what they have, I think that they're playing hardball. Good for them and their shareholders.
Intel's normal "tick-tock" cadence was not just interrupted - it has been destroyed. While TSMC (NYSE:TSM) and others prep their 10 nanometer products for imminent shipment, we know now that Intel's Coffee Lake product is still going to be produced at 14 nanometers when it ships in 2018. Tick-tock is now tick-tock-tock-tock with Broadwell, Skylake, Kaby Lake and now Coffee Lake all being produced at 14 nanometers over four years - an eternity in semiconducting.
There's a fly in the ointment over at Intel. And, until they sort it out, they need to make 3D XPoint appear to be worthless garbage. Nevermind the man behind the curtain. Once it is finally sorted, Intel can finish Fab 42 with the original goal of chalcogenide transistors and photonics. Because of the advantages, they'll probably be able to use cheap 22 or 28 nanometer equipment and still put the competition out of business. Operating speeds will be in the hundreds of gigahertz or higher. If you own AMD (NASDAQ:AMD), TSMC, Samsung (OTC:SSNLF), NVIDIA (NASDAQ:NVDA), Qualcomm (NASDAQ:QCOM), Xilinx (NASDAQ:XLNX), ARM (OTCPK:SFTBF) or the like, consider yourself warned. I'm not making any of this up - just speculating on the possible outcome.
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. Micron is either in bed with Intel on this deal or they are quite the opposite. There is no middle ground here. I honestly don't know their disposition, but I think that it'd be better for shareholders if they remained independent so that they can realize the true value of this technology (no matter how they fell into it).
What I do know is that change is coming in a big way. CPU-integrated photonics will transform computing dramatically (with Broadcom (NASDAQ:AVGO) sitting pretty on the SerDes patents that will be required for all of it). This includes the pesky low-performance interface that currently separates Optane from high-performance. Both PCIe/NVMe 4.0 and 5G cellular are very much delayed. I believe that photonics will become a big part of both standards once Intel sorts out their intellectual property with Micron and the ECD bankruptcy.
Then we'll be able to see Optane's true colors as photonic far memory.
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.