Please create an account to participate in the Slashdot moderation system

 


Forgot your password?
Close
typodupeerror
×
Intel United States Hardware

Intel's Expensive New Plan to Upgrade Its Chip Technology - and US Manufacturing (cnet.com) 131

Posted by EditorDavid from the made-in-USA dept.
America's push to manufacturer more products domestically gets an in-depth look from CNET — including a new Intel chip factory outside of Phoenix.

CNET calls it a fork in the road "after squandering its lead because of a half decade of problems modernizing its manufacturing..." With "a decade of bad decisions, this doesn't get fixed overnight," says Pat Gelsinger, Intel's new chief executive, in an interview. "But the bottom is behind us and the slope is starting to feel increasingly strong...." More fabs are on the way, too. In an enormous empty patch of dirt at its existing Arizona site, Intel has just begun building fabs 52 and 62 at a total cost of $20 billion, set to make Intel's most advanced chips, starting in 2024. Later this year, it hopes to announce the U.S. location for its third major manufacturing complex, a 1,000-acre site costing about $100 billion. The spending commitment makes this year's $3.5 billion upgrade to its New Mexico fab look cheap. The goal is to restore the U.S. share of chip manufacturing, which has slid from 37% in 1990 to 12% today. "Over the decade in front of us, we should be striving to bring the U.S. to 30% of worldwide semiconductor manufacturing," Gelsinger says...

But returning Intel to its glory days — and anchoring a resurgent U.S. electronics business in the process — is much easier said than done. Making chips profitably means running fabs at maximum capacity to pay off the gargantuan investments required to stay at the leading edge. A company that can't keep pace gets squeezed out, like IBM in 2014 or Global Foundries in 2018. To catch up after its delays, Intel now plans to upgrade its manufacturing five times in the next four years, a breakneck pace by industry standards. "This new roadmap that they announced is really aggressive," says Linley Group analyst Linley Gwennap. "I don't have any idea how they are going to accomplish all of that...."

Gelsinger has a tech-first recovery plan. He's pledged to accelerate manufacturing upgrades to match the technology of TSMC and Samsung by 2024 and surpass them in 2025. He's opening Intel's fabs to other companies that need chips built through its new Intel Foundry Services (IFS). And he's relying on other foundries, including TSMC, for about a quarter of Intel's near-term chipmaking needs to keep its chips more competitive during the upgrades. This three-pronged strategy is called IDM (integrated design and manufacturing) 2.0. That's a new take on Intel's philosophy of both designing and making chips. It's more ambitious than the future some had expected, in which Intel would sell its factories and join the ranks of "fabless" chip designers like Nvidia, AMD and Qualcomm that rely on others for manufacturing...

Shareholders may not like Gelsinger's spending-heavy strategy, but one community really does: Intel's engineers... Gelsigner told the board that Intel is done with stock buybacks, a financial move in which a company uses its cash to buy stock and thereby increase its price. "We're investing in factories," he told me. "That's going to be the use of our cash...."

"We cannot recall the last time Intel put so many stakes in the ground," said BMO Capital Markets analyst Ambrish Srivastava in a July research report after Intel announced its schedule.
Intel will even outpace Moore's law, Gelsinger tells CNET — more than doubling the transistor count on processors every two years. "I believe that you're going to see from 2025 to 2035 a very healthy period for Moore's Law-like behavior."

Although that still brings some risk to Intel's investments if they have to pass the costs on to customer, a Linley Group analyst points out to CNET. "Moore's Law is not going to end when we can't build smaller transistors. It's going to end when somebody says I don't want to pay for smaller transistors."
This discussion has been archived. No new comments can be posted.

Intel's Expensive New Plan to Upgrade Its Chip Technology - and US Manufacturing More

Intel's Expensive New Plan to Upgrade Its Chip Technology - and US Manufacturing

Comments Filter:

  • The thing I want to know (Score:4, Interesting)

    by rsilvergun ( 571051 ) on Sunday November 21, 2021 @07:16PM (#62008685)
    is that given they want to build these plants in the Southwest, where are they going to get the water?

    • Re:The thing I want to know (Score:5, Insightful)

      by cirby ( 2599 ) on Sunday November 21, 2021 @07:41PM (#62008731)

      Part of the plan is incredibly heavy recycling and purification. Costs a bit more, but considering the size of the investment, it's not that much.

      Chip fabs use a lot of water, but it's not like the water disappears when they use it.

      • Can you recycle 100% of water? Some will inevitably be lost.

        Also, as for the chips and moores law? I'll believe it when I see it. I fully expect Intel to do the bare minimum and charge more and more.

        • You wouldn't need to get to 100% recycling. Just get to 95%. Then you only need to draw 1/20th as much water from your providers as you actually use. If you use 1000 gallons for a process, you only need to get 50 from your water provider.

      • Re: (Score:3, Informative)

        by AmiMoJo ( 196126 )

        The tech has already been perfected in Taiwan, where they also have a water shortage. They have very strict regulations for factories using water, which includes having to purify it when they are done with it.

        GamersNexus covers it on a few of their factory tour videos.

        • Re: (Score:2)

          by shanen ( 462549 )

          IBM perfected it first. Their fab near Lake Biwa had almost perfect recycling of the water. So good that IBM dumped the plant. Can't remember who they sold it to. And I doubt it's still producing chips, to boot.

          But my conclusion is that the underlying reality is that IBM couldn't compete with the companies that didn't bother with all that recycling. Competitive advantage will still go to the runner with the cheapest sources of water and the highest local tolerance for such externalities as contaminated wast

    • Re:The thing I want to know (Score:5, Interesting)

      by Local ID10T ( 790134 ) <ID10T.L.USER@gmail.com> on Sunday November 21, 2021 @07:44PM (#62008741) Homepage

      What do you think happens to the water used in wafer fabrication?

      If you think water is used to wash away the chemical sludge of the manufacturing process and dumped... your views of modern manufacturing are wildly out of date.

      Water is pumped from the aquifer (or drawn from municipal water supplies) and filtered via RO/DI to remove as many impurities as possible to avoid contaminating the silicon. It is then used in the processing of the silicon wafers. Then the water is cleaned again and pumped back into the aquifer it came from. Very little water is lost in the process.

      • Then the water is cleaned again and pumped back into the aquifer it came from.

        Why do they do that? Why not just put it around the same circuit again?

        • It does sound like a reasonable idea but it would require additional holding tanks for the water. So it might just be to save space.

        • The RO/DI process separates the minerals/etc that are naturally found in water sources from the pure H2O. The pure H2O is used in the silicon wafer fab process, and filtered again to reclaim manufacturing chemicals. The end water is combined with the filtered out minerals from the intake process (which need to be flushed out of the system) and returned to the aquifer as remineralized water.

          They would need to flush the minerals out of the system, and using the post-process water allows the outtake water to

  • Sounds great to me (Score:4, Interesting)

    by SuperKendall ( 25149 ) on Sunday November 21, 2021 @07:18PM (#62008691)

    At least this plan sounds realistic and forward thinking.

    Still a trremedious task to bring fabs this advanced online, but if they can do it should be really good for Intel, and also for the American continent to be able to have so many chips manufactured here instead of overseas...

    • Indeed, but many-a-company have failed to bring forth good engineering after a long period of neglect. However, good luck to them - at least they're doing *something*, which I'd imagine most people could agree is required.

      I wonder if Intel will also spend as big on new designs though - even maybe breaking away from the x86 designs and into something a bit more interesting? Arm based designs are, frankly, eating their lunch, as Apple, Amazon and Tesla are making their own (with TSMC). To compete with that, I

  • No stock buybacks? How long until shareholders start pondering whether the company needs a new CEO?

    • Truer words have not been spoken (Score:5, Interesting)

      by Somervillain ( 4719341 ) on Sunday November 21, 2021 @08:56PM (#62008869)

      No stock buybacks? How long until shareholders start pondering whether the company needs a new CEO?

      Stock buybacks are a cancer and I don't even know if there's anything that can be done...short of introducing regulatory legislation, that no matter how well written and popular, would have a snowball's chance in hell of passing.

      We've made a deal with the devil. We compensate those who make decisions based on the share price, giving them incentive to starve the company by buying back as much stock as humanly possible. This ransacking the coffers to enrich these piece of shit executives needs to stop. People need to be angrier about this.

      We're trading our long-term economic prospects for short-term gain of Wall Street & C*Os. How much of the failing of American manufacturing and industry is due to other countries being more competitive and how much of it is just our own companies imploding due to incentivizing short term gains and catering to every whim of parasitic Wall Street investors?

      • Truly competitive companies keep enough stock "locked" in trusted hands to prevent this from happening. Having a "free-float" of 49% or less is a great way for a company to raise money without being in danger of becoming Wall Street's plaything. The non free-floating stock doesn't have to belong all to the founder, just be in the hands of a few key people or organisations. On the contrary, if nobody owns a large enough piece of the company to care about the long-term prospects of the company, nobody will.

    • No stock buybacks? How long until shareholders start pondering whether the company needs a new CEO?

      Most larger companies have no problems with shareholders and don't do stock buybacks. They are normally a short term measure ultimately of little interest to shareholders. Far more interesting to large shareholders of bluechip stocks is the fact that Intel has missed precisely 1 quarterly dividend in the past decade.

  • Last I heard, it was Wil.i.am. It's finally paying off!
    Maybe they can get the guys from OCC to build them another bike for the Grand Opening.

  • Just one not-so-small problem... (Score:3)

    by Todd Knarr ( 15451 ) on Sunday November 21, 2021 @08:41PM (#62008851) Homepage

    As various outlets have noted, a lot of the demand for chips isn't for the cutting-edge chips Intel is planning to build fabs for, it's for older-generation chips that nobody's building new fabs for. I think before I evaluated Intel's plans I'd want to look at product demand by process node to see if those plans produced the right kinds of supply for the demand.

    I think if I were looking for a long-term land-office business, I'd invest in fabs targeting older process nodes used in automotive and industrial-equipment applications. You'd have the advantage of going for a market the other chip makers aren't satisfying plus giving that market a reason to diverge from what the other chip makers want them to do leaving you with less competition for customers.

    • Can they not build low end chips there too? As in, what is stopping them from building say a microcontroller chip like the ATmega328 using a 3 nanometer process? They might actually make more money since it would use less than 1% of the die area of a modern CPU.

      • As in, what is stopping them from building say a microcontroller chip like the ATmega328 using a 3 nanometer process?

        1) Have you tested this chips thoroughly as since it is used in automobiles, it has to be extremely reliable. Testing these chips normally takes years which is why they are older chips. 2) Is Intel going to charge less than $1 for each chip as that is what the automotive is paying.

        They might actually make more money since it would use less than 1% of the die area of a modern CPU.

        The automotive industry pays very little for these chips. I do not see how Intel is going to make a profit.

        • They can easily print a few hundred thousand of those chips on a single wafer since they are only around 50 thousand transistors, compared to say 15 billion on a single apple M1 chip. That is literally 300,000 chips times more than an ATmega328 on one M1 chip (of which a wafer must have many).. that means theoretically they can be 300,000 times cheaper than an M1 cpu .. but let's say because of other issues such as cutting margins and packaging it is only 3000 times cheaper. That means they can easily recou

          • Good luck driving 200mA at 5V (all outputs combined) out of a 7nm process chip.
            Also, good luck cooling that 3500 transistors, 1/1000 or less square millimeter silicon chip.

            • Never heard of a resistor? Voltage regulator circuit? Build it into the chip package. Also, cooling a small area is easier than cooling a bigger chip.

          • And how much set up do you think it takes to make those wafers? Lots of set up leading to a short run for very little money each. You think that is a profitable combination? Most lines are set up to run months if not years.

            • You think Intel doesn't have older facilities around that are looking for stuff to do, because they aren't suitable for the absolute latest node processes? They can probably extend the usefulness of some facilities that they would otherwise look to mothball or demolish, and extend the payback on those investments. They've been doing that for years by taking old processor fabs, and retooling them for flash memory manufacture, or chipset manufacture that doesn't require the latest processes, etc.

              • Except that is not what the poster is advocating. He is specifically advocating using smaller lithographies to make older chips and he does not see why that is not profitable: "They can easily print a few hundred thousand of those chips on a single wafer since they are only around 50 thousand transistors, compared to say 15 billion on a single apple M1 chip". At that scale, each customer needs less than 5 wafers? Normal runs are in the hundreds to thousands of wafers.

        • The make profit because they can fab them in older facilities that are no longer suitable for leading-edge technologies, but already have the capital costs paid off. When you don't have to build a $20B fab to crank out high yields of simple microcontrollers, the ROI arrives a lot sooner. Retooling is a whole lot cheaper.

          • The make profit because they can fab them in older facilities that are no longer suitable for leading-edge technologies, but already have the capital costs paid off. When you don't have to build a $20B fab to crank out high yields of simple microcontrollers, the ROI arrives a lot sooner. Retooling is a whole lot cheaper.

            He says specifically: "As in, what is stopping them from building say a microcontroller chip like the ATmega328 using a 3 nanometer process?" which is making a very old chip on a new process.

      • Global Foundries considered investment into the "7nm" process too expensive, and is staying at 14nm and similar (down to "12nm" which is a variant of 14nm). Just for example, 14nm was the last spot Intel held a manufacturing advantage (and, supposedly, its 14+++ +++nm process is better than anyone's).
        As for the ATmega328, the size of the silicon chip is limited by the necessity to connect the 32 pins.
        While you can do this with interposers and the like, the "cheap" version is to have the contacts on the edge

        • Aren't the pins connected by a wire to the silicon chip mounted inside the chip carrier? The packaging can take care of having 32 big pins.

          • I'll bite.
            ATmega328p has a single "power" pin and a single "ground" pin (Analog Vcc) and has a recommended maximum amperage (total) of 200 mA.
            In this case, the recommended wire size (https://learn.adafruit.com/wires-and-connections/wire-guages) is 0.127 mm diameter.
            As such, the pads that take the power and ground wires for "Analog pins" must be - let's say double that diameter, or some 0.25mm each, with a 0.25 mm gap.
            For ease of construction, let's say every "inner pad" is the same, and uses the same wires

            • Re: (Score:2)

              by hawk ( 1151 )

              that's just the CPU, though. It also has 32k flash, and 2k of SRAM, the latter of which takes nearly 25k transistors.

              I don't know if the A/D and clocks are part of the CPU count, though.

    • A lot of it is on the cutting edge chips, but who is going to bet their corporate future on Intel when Intel is likely to compete with them? This will fizzle just like Intel's previous foundry attempt. There is exactly one way to make it work: spin off the fabs.

    • Those chips are (or were) far less profitable per unit.

    • "I'd want to look at product demand by process node to see if those plans produced the right kinds of supply": That's why Microsoft is bringing out Windows 11.

      And no, I'm not moving to Windows 11: I don't need rounded corners that badly.

  • I applaud Intel for getting it together and becoming the "Master of it's of Destiny" once again.

    The demand for lower power, high density chips 4nm and smaller is only going to increase, especially as we move into self-driving cars.
    Tesla just partnered with Samsung for 5nm chips and everyone will soon follow,
    Without the lower power chips, AI and the image processing system would be a drain on the battery.

    At least if Intel doesn't make it's own chips, it can make chips for Apple and Tesla.

    Go Intel! Glad you

  • If they received even a penny in the form of government subsidies, then they should be forced to provide AMD fab capacity at cost, as punishment for their many illegal cr@p that they did to AMD and got away with.

    • This is not about AMD. I would say at this point AMD can fight it's own battle. Remind me, how many AMD chips are made in USA again?

    • Exactly which of Intel's nodes would AMD want?

      • And would AMD want to relayout their chips to Intel fab rules? It seems most people think you can just move a part from one fab to another. Of course you can, but it means you have to relayout the design and ensure it meets timing, power, electromigration... Unless of course the fab setup a second source with the same process. But I don't think that happens much anymore.

        • Last time a major fab company licensed its process tech to another company was when Samsung licensed 14LPE/LPP to Globalfoundries.

          Without that tech GF would never have released their own 14nm and 12nm variants.

  • Why it is preferred for fabs or high tech area vs northern US, where abundant water is found ? Don't know but a desert is just asking for logistic problems.

    • Re: (Score:3)

      by Z00L00K ( 682162 )

      Because management decision - it seems like Texas is preferred for tax reasons while northern states aren't. The management also want to have the production close to the head office to have an excuse to leave the office.

      • The advantage of higher taxes is, at least in principle, better infrastructure so more reliable power (always useful if you need to run at 100% capacity to cover costs) and easier access by employees.

        Of course, that's not always going to pan out. Silicon Forest (the border between Oregon and Washington State) isn't as impressive as you'd like for the money.

        • The fab will have redundant power supply for short pauses in electricity.
          However, as soon as that electricity supply vanishes, around a month of production is lost:
          https://wccftech.com/tsmc-plan... [wccftech.com]
          (depends on the process complexity, number of "repeated" phases and a lot more).

        • I think the bigger problem with Intel's campuses in the west Portland areas is that they don't have any more room, and they would have to tear down fabs that are still running on production loads to build new. That's undoubtedly scheduled for the future as those processes are phased out, but they already have a ton of land in Chandler, AZ that probably has obsolete or mothballed fabs sitting on it that they can tear down and build brand new on the footprint.

          The tax advantages of Arizona over Oregon are pro

      • It is more about tax abatements. TX funds from property tax and sales tax. Property tax runs 2% of actual value. So for a fab, figure 20Bx2% or around 400m/yr. Of course no company pays that, the local areas fall all over themselves to give deals to snag plants. In Samsung's case the Taylor site offered Sammy 90% off. https://www.reuters.com/techno... [reuters.com] Of course mere peons pay full tax. I think the state of NY also plays the tax incentive game. They seem to win some deals. I am not a fan of tax abatement. Th

    • I'm pretty sure someone thought about water availability before announcing they would spend $20B to build semiconductor fabrication plants there. As for why Phoenix, Intel already has a huge presence in the Phoenix area. The Chandler, AZ campus used to be huge and have at least 3 operating fabs there - haven't been down there in years but I would assume it still is huge, and that would be where they would be building this.

      Phoenix seems to do alright with water resource management - after all, they have a

  • EUV the limiting reagent (Score:5, Interesting)

    by DrMrLordX ( 559371 ) on Monday November 22, 2021 @12:11AM (#62009175)

    Intel doesn't have enough EUV equipment to compete with TSMC or Samsung. They can build out all the fab capacity they want. If they best they can do is expand capacity in DUV nodes like 10ESF then they're just going to be left behind.

    TSMC already has a significant edge with N5P which is in full production. 10ESF/Intel 7 just came to market with Alder Lake-S and it's already behind in density and performance.

    Intel itself is taking (allegedly) 20kwpm (kilowafers per month) of TSMC N3 to bolster its chip supply. That should tell you that something is very wrong with Intel's next-gen node: Intel 4(formerly known as Intel 7nm). At the very least it's supply limited: estimates put Intel's supply of Intel 4 @ 20kwpm by the end of 2023, which neatly mirrors the number of wafers that TSMC will grant them on what is expected to be a roughly-equivalent node. But that says nothing about yields on Intel 4. Intel's Ponte Vecchio [hardwaretimes.com] was originally meant to be a mashup of Intel 4 and Intel 7 chiplets, but has since transitioned entirely to TSMC N5 and N6. That doesn't speak well of any of Intel's process technology. And it is heavily delayed to boot.

    Intel may be able to fabricate small chiplets on Intel 4 for products like Meteor Lake, but anything else? It is questionable.

    • Additionally, one could also say that Intel is "eating" TSMC manufacturing capacity as a "scorched earth" measure against AMD increasing production.

      • That rumour went around, but as it turns out AMD is basically getting what they want. TSMC is granting very little capacity to Intel, and nobody knows how much Intel is having to pay for those wafers.

  • And the climate crisis will worsen conditions there. Given the massive costs and the need to run at near-100% to break even, I'm not convinced it's the wisest of choices.

    My other thought is that that's a lot of money to burn to continually play catch-up. If they say they're choosing the best path, I guess we have to believe them, they have the expertise after all. Even so, I feel that they're relying too much on repeating a few tricks rather than developing new ones.

    (Even split roles isn't strictly a new tr

    • Texas lacks reliable infrastructure and the climate crisis will worsen conditions there.

      So long as they are in a seismically stable area away from the coast then they can provide their own backups in the event of power outages. It does cost money but it's all part of a financial calculation.

      My other thought is that that's a lot of money to burn to continually play catch-up.

      It is a lot of money but the alternatives are worse. They could rely on another fab but this would have long-term issues. Doing so would ensure they would never have the advantage and put them on equal footing with other chip makers with events like the current fabrication crunch. Giving up a potential

  • It's obvious that improving the fab technology is a worthwhile investment for Intel but the real question is is if they are going to be investing in improving their microarchitectual design/review process. As of now, there are annual vulnerabilities coming out due to negligent design considerations. Making the product faster is only half of the problem with Intel's chips.

  • They way that Intel chases whatever tech is getting press, I am surprised that they are not announcing the upcoming Intel Self Driving Car.

  • This should bring Moore's Law back into scope: People have colloquially equated transistor count with processing power, and so far that's been fairly accurate. But with new SoCs having on-die RAM, the transistor count is going way up with a relatively small increment in raw processing power (but huge increment in other system design parameters).

  • Comment removed based on user account deletion

Slashdot Top Deals

One picture is worth 128K words.

Close