Micron to Sell 3D XPoint Fab to Texas Instruments for $900 Million

By Paul Alcorn
last updated

Micron pops the parachute

3D Xpoint Optane
(Image credit: Intel)

Micron announced today that it has entered into a definitive agreement to sell its Lehi, Utah fab to Texas Instruments for $900 million in cash. In March, Micron announced that it planned to sell off the fab, bringing an end to its production of the radical new 3D XPoint (Optane) memory technology that it developed with Intel. Texas Instruments plans to deploy its own technologies at the site, meaning that it will not be used for 3D XPoint production. Intel currently doesn't have any known high-volume production of the strategically important storage/memory media. However, it is known to produce a small amount of the media for research and validation at its New Mexico facility. As a result, Intel will likely have to establish its own production lines to ensure the supply of its Optane based SSDs and persistent memory DIMMs for its data center clients, though demand has seemed tepid.


Micron chose to exit 3D XPoint manufacturing due to lackluster demand that the company said had "insufficient market validation to justify the ongoing high levels of investments required to successfully commercialize 3D XPoint at scale." The company recently divulged that it lost $400 million this year alone due to the lack of demand for 3D XPoint.

Micron has an agreement to produce 3D XPoint (which Intel brands as 'Optane') for Intel until the end of 2021. However, Intel's own efforts to productize Optane, which uses the 3D XPoint media, have met with slow but steady uptake in the data center but fizzled in the consumer market. As such, Intel ceased production of all Optane devices for desktop PCs in January 2021.

The economic value of the sale weighs in at $1.5 billion for Micron. Texas Instruments will pay $900 million in cash for the fab, while Micron will recoup another $600 million from secondary sales of additional tools and other assets. Micron has already lined up buyers for several of those assets, while other assets will either be sold to other buyers or shipped out to its other Micron manufacturing sites. In addition, TI will convert the 300mm facility for 65nm and 45nm production. It will also attempt to retain all of the Lehi, Utah employees after the closing of the sale, which is planned to occur by the end of the calendar year.

Intel and Micron developed the revolutionary 3D XPoint persistent memory, which melds the speed and endurance of DRAM with the persistence of data storage devices, in a secret joint effort that spanned a decade. The first formal announcement came in 2015.

The Lehi sale marks the end of Micron's 3D XPoint efforts, though the company does retain its IP associated with the technology. The company previously bought Intel out of its stake in their IMFT joint venture for $1.5 billion in 2018. Micron announced several of its own storage devices based on 3D XPoint memory, like the QuantX and X100, but they never came to market, leaving Intel as the sole supplier of 3D XPoint-based products. 

Micron says that it will shift its focus to developing memory products that support the Compute Express Link (CXL) standard, an open memory standard that ties together disparate pools of memory and compute. 

 “Micron’s Lehi, Utah, facility has a strong history of technology innovation and leading-edge semiconductor manufacturing,” said Micron President and CEO, Sanjay Mehrotra. “We are pleased to have reached an agreement with Texas Instruments as it is an industry leader and truly values the talented Lehi team and the capabilities this site offers to deploy its technology effectively. We are greatly appreciative of the contributions that the Lehi team has made to Micron, as well as the collaboration and engagement Micron has had with the local community.”

This is breaking news. We've reached out to Intel for comment and will update as necessary. 

Paul Alcorn
Paul Alcorn
Deputy Managing Editor

Paul Alcorn is the Deputy Managing Editor for Tom's Hardware US. He writes news and reviews on CPUs, storage and enterprise hardware.

4 Comments Comment from the forums
  • Giroro
    Intel fumbled Optane so badly, it's a real shame.
    They create this amazingly fast storage technology for random reads/writes, but then pair it with outdated first-gen controllers that have subpar sequential performance. When you're selling the most expensive SSD (by far), then it needs to excel in every category.

    If you want people to combine an extra module with an HDD, then the pairing needs to be either good enough, cheap enough, or simple enough that people have a reason to buy it over a standard SSD.
    Maybe they could have used optane's longevity to take advantage of people wasting money on Chia?

    3D Xpoint also could have replaced the SLC cache for a high performance optane/flash hybrid SSDs using a single integrated controller. But instead they glued their slowest optane module to their slowest SSD and said "eh, performance, who needs it? ".

    People want this tech, Intel. We have just been waiting for a good-enough product sold at a justifiable price.... With ryzen compatibility.
    Reply
  • escksu
    Giroro said:
    Intel fumbled Optane so badly, it's a real shame.
    They create this amazingly fast storage technology for random reads/writes, but then pair it with outdated first-gen controllers that have subpar sequential performance. When you're selling the most expensive SSD (by far), then it needs to excel in every category.

    If you want people to combine an extra module with an HDD, then the pairing needs to be either good enough, cheap enough, or simple enough that people have a reason to buy it over a standard SSD.
    Maybe they could have used optane's longevity to take advantage of people wasting money on Chia?

    3D Xpoint also could have replaced the SLC cache for a high performance optane/flash hybrid SSDs using a single integrated controller. But instead they glued their slowest optane module to their slowest SSD and said "eh, performance, who needs it? ".

    People want this tech, Intel. We have just been waiting for a good-enough product sold at a justifiable price.... With ryzen compatibility.

    No, I do not think the sequential perform is "subpar" and its a problem. 1 real issues killed Optane, esp for consumer sector.

    ITs PRICE..... Its simply way too expensive compared to NAND SSD.

    As for performance, its actually pretty good but not enough to justify its price. For consumers, space matters even more than just speed. There is no point in having a fast but tiny storage because it will run out of space very soon.

    End-user usage patterns will not allow Optane to shine as well. Optane shines in intensive mixed read/write IO where it will bog down regular SSD. We don't see this in consumer environment. Hence, Optane will perform no different from regular SSD. Sequential Nature of I/O means you won't see much improvement in loading times as well.

    As for sequential performance, its actually not an issue for most users. Most only have 1 drive in their computer. So, unless you transfer a massive file to RAM or to another drive, you will never hit that sequential limit.
    Reply
  • DavidC1
    @Giroro @escksu

    Guys, price and performance would improve over time. You can't get everything right the first time. Actually for sequentials there's evidence that power consumption limited sequential performance. You can see they improved a lot on the second generation.

    Back when the 900P drives were introduced, a reliable leaker said that the 900P drives weren't making Intel much money. You get an idea how expensive it is to make.

    I don't think Intel was planning single controller for Optane at the time. You still need some amount of DRAM buffer because Optane isn't fast enough to replace it entirely. Designing a proper controller would take years and everything around it so that's why they went the simpler way of putting two controllers on the H10 and H20.

    Another problem is they don't have a proper NAND controller anymore. So not only they need to make a competitive NAND controller, but also make it work as a hybrid drive.

    The PCI Express also limits the potential of Optane - ultra low latency comparable with DRAM. At least the DIMM versions make some sense.

    The biggest problem is whether Intel is going to abandon it. Things are not looking good for Optane right now. I think for the sake of Optane they need to find a way to make it open and be compatible with AMD CPUs, because their server CPUs suck.
    Reply
  • Kamen Rider Blade
    Man, I really liked Optane as a technology, but this sucks.
    Reply