Slashdot is powered by your submissions, so send in your scoop

 


Forgot your password?
Close
typodupeerror
×
Intel Technology

Intel Outsources Core i3 To TSMC's 5nm Process (eenewseurope.com) 77

Posted by msmash from the how-about-that dept.
TSMC is to start making Intel's Core i3 on its 5nm process in 2H21 says TrendForce. From a report: Market analyst Trendforce reports that foundry TSMC is to start making Intel's Core i3 process later in the year on a 5nm process. This follows Intel's well documented problems with its leading edge process technology at 10nm and 7nm. The Core i3 move to a 5nm process is set to be followed by mid-range and high-end CPUs being produced for Intel by TSMC on a 3nm process in 2H22. TrendForce did not give a source for the information, simply referencing "investigations." Intel has long outsourced production significant amounts of its non-CPU chips to TSMC and UMC -- about 15 to 20 percent of its output, according to TrendForce. This is partly because it has often acquired fabless startups that had brought products to market using foundry. It was usually not worthwhile to re-engineer such products to Intel processes. It is also because Intel has wanted to focus on leading-edge specialist processes, although with less success in recent years. That 15 to 20 percent outsource was likely worth $10.5 billion to $14 billion in 2020, given Intel's annual revenue of $70bn.
This discussion has been archived. No new comments can be posted.

Intel Outsources Core i3 To TSMC's 5nm Process More

Intel Outsources Core i3 To TSMC's 5nm Process

Comments Filter:

  • Hell froze over (Score:2, Interesting)

    by gweihir ( 88907 )

    This is basically the end of Intel as a CPU maker, even if that will still take a few decades to happen. Their CPUs were never any good, it was always their own manufacturing process that gave them an edge. Well, and their shoddy, insecure designs that prioritized speed over everything else.

    • Re:Hell froze over (Score:4, Interesting)

      by alvinrod ( 889928 ) on Wednesday January 20, 2021 @01:20PM (#60969128)
      I have a feeling Intel will be around for a good long while. A lot of people wrote AMD off and they came back from a far, far worse situation. If you think Intel can't you're being incredibly short sighted.

      Also, if Intel books at lot of the newest TSMC node, what's AMD going to use for their next CPUs? Every i3 is a Zen CPU that never gets made. Seems like Intel might have found a new way of playing dirty.

      • Re:Hell froze over (Score:5, Insightful)

        by stabiesoft ( 733417 ) on Wednesday January 20, 2021 @01:27PM (#60969182) Homepage
        I don't disagree it will take a long time for Intel to die if indeed it never gets it act back together. That is the way it is with large corps. Kodak, GE, IBM, HP the list is long of essentially dead tech companies. However, the winner is TSMC, who will increase fab production to handle both AMD and Intel. And with each i3 they fab, the ability for Intel to return to manufacturing their own decreases.

        • Re: (Score:2)

          by gweihir ( 88907 )

          And with each i3 they fab, the ability for Intel to return to manufacturing their own decreases.

          Indeed. I just doubt Intel can really compete in that situation.

      • Yes, but Intel doen't have money to spend forever or they would not resort to this. They are in panic.
        And don't forget every other chip maker that uses TSMC will want to see Intel die. And not have its output cut.
        So you can bet your ass, AMD and the gaming console makers, like Sony, and the car chip makers, will pay just that little bit more, to keep TSMC on their side. Until Intel is starved. Even if it may take some time.

        The next weeks will be funny. Watch for AMD press releases in the next 24 hours.

        • Re: (Score:2)

          by xonen ( 774419 )

          I wouldn't call it panic, people have been suggesting for many years that Intel should outsource more. Now they finally do, people say it's a sign of the end.

          I think Intel has many ways to let their fabs make money. Even totally outdated technology can make commercially viable IC's. Look at the Chinese, there's so much more semiconductors than microprocessors and a lot of it does not require small scale at all.

          Also, as for the background why Intel can't catch up, it might be technical difficulties but likel

          • Re: (Score:2)

            by tlhIngan ( 30335 )

            Even totally outdated technology can make commercially viable IC's.

            A ton of ICs don't need the latest and greatest. And they're still needed even in a modern computer.

            Memory devices (RAM and flash) require latest technology nodes because smaller transistors mean more per area, aka, more storage. Everyone wants more RAM, more storage and smaller nodes give you this. (These structures are so regular, they're often able to be a half-node ahead).

            High speed cores - CPU and such and caches. The data path is proba

            • I've seen some maskless IC shops, where the "clean room" is encapsulated to the trolley and the stations that the silicon travels on. The dies are not big, about the 3-5 centimeters in diameter, but if someone is wanting a unique ASIC, or wanting to create something that is to be kept in-house, it might be good enough.

              Wish that technology would improve, so maskless, one-off ICs could be more commonplace.

        • Re: (Score:2)

          by gweihir ( 88907 )

          Yes, but Intel doen't have money to spend forever or they would not resort to this. They are in panic.

          Obviously. This move is an admission of extreme incompetence. Obviously the manufacturing process improvement is possible as both TSMC and Samsung have made it. But Intel cannot, to the degree that at this time they think they cannot for quite some time to come.

          And don't forget every other chip maker that uses TSMC will want to see Intel die. And not have its output cut.
          So you can bet your ass, AMD and the gaming console makers, like Sony, and the car chip makers, will pay just that little bit more, to keep TSMC on their side. Until Intel is starved. Even if it may take some time.

          The next weeks will be funny. Watch for AMD press releases in the next 24 hours.

          I would hope for merciless mocking, but unlike Intel, AMD has too much style for that.

      • I think you're absolutely right. Soak up more capacity on TSMC and Intel has leveled the play field of suffering for the others. ...while also getting the space to get them selves sorted out a bit.

      • Re:Hell froze over (Score:4, Interesting)

        by lprint222 ( 7648378 ) on Wednesday January 20, 2021 @06:35PM (#60970980)
        You're out of your mind. Production slots are negotiated with plenty of time in advance, think years, and contracts are water tight. Were TSMC to "dump AMD" in order to satisfy Intel production needs then two things would happen immediatly. The first one would be a massive and perhaps irrecoverable hit to TSMC as a reliable supplier. No one would dare their trust their roadmap execution to TSMC. It would be a bonanza for TSMC competitors. Not that there are many presently, mostly Samsung Electronics on the advanced nodes only. The second one would be the well deserved lawsuit for break of contract that AMD would swiftly file. A lawsuit that would be easily won. Both things would not serve TSMC in the least, and whatever is negotiated with Intel would not in the least compensate TSMC for this. I doubt the article is correct. TSMC production is fully booked for 2021 for a start. Production for 2022 is 80% booked. And in 2021 AMD has *trial runs* of Zen 4 on 5nm. At the very best, by some miracle, perhaps Intel would get some trial runs at 5nm at the end of 2021. I doubt it. It's more likely whomever wrote the article mistook the dates. 2022 seems likely, 2021 not so much.

      • Re: (Score:2)

        by otuz ( 85014 )

        Think of future Intel as of past Motorola. They'll be gone, or maybe aquired by someone else to brand their generic ARM chips "Intel" for some marketing points.

    • Totally spot on. Once the tables got turned - when other foundries got ahead of Intel - they stopped things like pushing x86 for mobile devices, IoT, etc., as their process didn't allow them to compete in those power budgets.

      It's also partly what allowed AMD to start eating Intel's lunch again in the short term, and for ARM to put an end to x86 in the foreseeable future .

      I don't see Intel winning through innovation or superior engineering any time soon, if ever - arguably, they never did win that way, unles

      • Re: (Score:3)

        by kriston ( 7886 )

        Smartest thing would be for Intel to jump on a new architecture and use their substantial, but diminishing industry position, to attempt a comeback.

        That didn't work out too well for them the last couple [wikipedia.org] of times [wikipedia.org] they tried it.

        • Comment removed based on user account deletion

          • Re: (Score:3)

            by gweihir ( 88907 )

            Wow, I'd never heard about the i860. It looks like they repeated the same mistakes with the Itanium, both using VLIW and both suffering from poor performance because of compilers not being able to generate code to effectively take advantage of it. So much for learning from your mistakes.

            They failed and failed and failed again. There is a reason what we are using today is the AMD64 architecture. Intel could not hack it.

      • " their process didn't allow them to compete in those power budgets."

        But Intel makes multiple processes at each node. Not just the dot processes for different SKUs of processors, but also the odd-numbered process that are more restrained in their performance targets.

      • If they dump x86, it's also just a matter of time. Just... less time.
        Intel that can't run Windows (and by Windows I mean the whole back catalog of weird, niche win32 software that will never see an update) has no advantage over the multitudes of ARM chips for every price point and power budget.

      • You seriously underestimate the difficulty and amount of work to make such a jump. It'd be like a 60 year old plane enginee suddenly having to start from scratch and design buildings.

    • Re: (Score:2)

      by malkavian ( 9512 )

      I'm not so sure.
      The article specifically mentions that there is no source for this, but it's based on speculation from observation of several reports.
      If it's accurate (it may or may not be; at the moment it's 'unknown'), then it would make sense for Intel to have a premium line of chips available at a lower fab size, that could be used in use cases that the regular fab sizes would restrict. There's still a huge market for their regular Intel fabs, so hedging bets and outsourcing has its use.
      It may be that

    • Re: (Score:3)

      by ytene ( 4376651 )
      In fairness to Intel, the decisions they made to prioritize clock speed over everything else was done at the request of one Bill Gates, of Microsoft.

      Go back to around the 1980s and companies like Inmos were releasing innovative computers such as their Transputer [wikipedia.org], a microprocessor that was specifically designed, in harmony with compilers - to support massively parallel processing.

      Intel saw what they thought was a really elegant way to massively scale CPU performance without having to wait for manufactu

      • The problem with Transputer-type parallel processing is that it's only suitable for a subset of programming problems. You can't take a regular C/C++ program, and run it through a compiler, and get an efficient distributed compiled program, no matter how much effort you put into it.

        There was also still a lot of MHz/GHz to be gained, so it makes sense to do that first, and build maximum single processor performance, before going to multiple cores. Even if you do go to multiple cores, shared memory makes more

        • Re: (Score:1)

          by DarkOx ( 621550 )

          There was also still a lot of MHz/GHz to be gained,

          Even more true so when it came to smaller faster to stabilize circuits with fewer interconnects. Gates was very correct on this point, the bigger gains in computer performance were to be had first targeting faster clocks than better IPC.

          Its relatively recently we hit the wall of practical clock speeds with current materials and even more recent that we have really hit IPC limits where the complexity increase starts to force clocks back down. There is a reason we are still mostly working with single and on h

    • I agree that Intel is likely to permanently lose their position as first class chip maker. What surprises me is that they start at the i3 models. Those are relatively low end parts with modest requirements. Intel's 14nm fabs should be quite sufficient for serving that market.

      I really expected that the i7 and i9 parts would get priority. Or maybe the Xeon line of CPUs. Anything where speed and TDP are real considerations.

  • I thought that the Core i3 is a pretty old processor used mostly in entry-level laptops ?
    Why outsource that one ?
    The only explanation I can think of is the in-fighting at Intel has resulted a "we did something" solution, where they did a token gesture to please the shareholders.

    • Re:Outsourcing the Core i3 ??? (Score:4, Informative)

      by UnknowingFool ( 672806 ) on Wednesday January 20, 2021 @01:50PM (#60969376)

      The Core i3 is not an "old" processor. It is the least powerful version in every generation. The last ones were released April 2020 for 10th generation.

      As for why TSMC is making it? I would speculate partially based on pride on Intel's part. They cannot admit they outsourced their flagship CPUs. The other reason could be volume as cheaper desktop CPUs are more likely to be purchased by consumers and OEMs than higher end and pricier Core i9 models. Intel is struggling with yields currently so higher volume parts may be easier to move to a 3rd party if they can make them.

    • Re: (Score:2)

      by kriston ( 7886 )

      Not old, just underfeatured and slower. There are 10th generation Comet Lake Core i3 processors [wikipedia.org] so I wouldn't call a 5nm Core i3 "old" by any stretch of the imagination.

      • OK, so Intel keeps the "Core i3" name and updates the processor with every generation. Thanks for the explanation.

        Anyway, I suspect there is a lot of in-fighting between the processor design (in favor of outsourcing) and the manufacturing sides (against oursourcing) plus the shareholders jumping in and demanding that something be done (so the share price does not suffer).

        • "the processor design (in favor of outsourcing) and the manufacturing sides (against oursourcing)"

          When the manufacturing is in-house you can co-design the process for that the designers want. When you send it to TSMC you'll get nothing and like it.

          The tension is likely between high end vs high volume SKUs. They are likely pulling the process in different directions and trying to cover them all cratered the yield.

        • OK, so Intel keeps the "Core i3" name and updates the processor with every generation. Thanks for the explanation.

          Intel keeps all "Core" naming every generation while updating the processor: Core i9, Core i7, and Core i5. Now naming of individual CPUs may not be consistent at times but the series naming are pretty consistent.

    • Comment removed based on user account deletion

    • The i3 is a series, not a specific model. For instance the Core i3-1125G4 is a brand new low end processor which will be released this quarter on Intel's 11th Gen platform.

  • since this will make them extremely overclockable, assuming they release some K variants.
  • Considering all TSMC's current 7nm customers who want to use 5nm, what did Intel do to get production later this year? Lots of money?

    • Re:How much money did this take? (Score:5, Informative)

      by Solandri ( 704621 ) on Wednesday January 20, 2021 @03:56PM (#60970280)
      Apple bought all of TSMC's initial 5nm capacity. Intel is one of the few companies with a profit margin (about 30%) more obscene than Apple's (about 20%). The rest of the industry has about a 5%-10% profit margin. So Intel is in a position to offer TSMC more money than Apple to get priority access to 5nm (and 3nm).

      TSCM 5nm is about 170 million transistors per mm^2., vs about 100 MT/mm^2 for their 7nm (and Intel's 10nm). That's the main reason Apple's M1 and A14 do so well. And why AMD's current gen (at 100 MT/mm^2) does so well against most of Intel's lineup (manufactured on Intel 14nm, which is only 37.5 MT/mm^2).

      By comparison, the architecture differences appear to be rather small. AMD's Zen+ was manufactured on Global Foundries' 12nm process (about 37 MT/mm^2) which was the same as Intel before 11th gen (Intel 14nm @ 37.5 MT/mm^2). Those performed similarly. And when Apple and Qualcomm were both using TSMC's 7nm, they performed pretty closely with Qualcomm beating out Apple on a few benchmarks (Apple's processors are much larger and have more specialized hardware acceleration - the M1 has more transistors than an octacore Ryzen).
  • As much as I was enjoying AMD and Apple punish Intel for resting on their laurels, Intel is the last major U.S. chip designer that fabbed their own chips. The existing chip fabs are already having a lot of trouble meeting demand and this is probably going to exacerbate that problem. I have confidence that Intel will eventually find a way to compete with AMD and Apple chips, but it's sad to think that we could be looking at the beginning of the end of their fab process.

  • The only reason people are still buying Intel chips is because unlike other manufacturers they don't have a supply bottleneck. Unless the point is to just buy up capacity and prevent AMD and Apple from getting silicon.

    • Re:Throwing away their only advantage (Score:4, Interesting)

      by radarskiy ( 2874255 ) on Wednesday January 20, 2021 @02:05PM (#60969508)

      "they don't have a supply bottleneck"

      By diverting their low end SKUs to a different process they can dedicate their manufacturing to the high end and tune the process for those SKUs without abandoning the low end to other processor manufacturers. If those purchasers were lost it will be hard to bring them back.

      • Re: (Score:2)

        by AmiMoJo ( 196126 )

        The high end is where they need 5nm, those things run hot and are issued thermally limited, especially in laptops.

        • High volume is where they need the yield of TSMC N5. By narrowing the range they have to cover, Intel can target their process for the high end but low volume products.
          Intel will trade the performance for an accelerated convergence schedule for the physical design.

  • Value proposition of Intel (Score:3)

    by enriquevagu ( 1026480 ) on Wednesday January 20, 2021 @01:49PM (#60969374)

    So what is exactly the value proposition of Intel? In terms of their processors, Intel has:
    a) a manufacturing process
    b) an ISA
    c) their design & microarchitecture
    d) Inertia, brand value, advertising, integrator lock-in conditions, a bright "Intel Inside" sticker, ...

    a) the manufacturing is falling behind, as is obvious from this post. Note that, if I understand correctly, this is the first time that Intel outsources their own, internal CPU designs that initially targeted their own fabs.
    b) the x86 (actually x86-64) is what, by inertia, is driving their PC and server market. The ISA is not particularly good, quite the opposite, but it has been the standard for so long that there are many many tools (and OSs) tied to it. However, recent movements increase the risk of tying to this ISA, particularly the M1 ARM design from Apple away from x86 (and other server designs, in the cloud environment, such as AWS Graviton).
    c) their design is fine, but others are better, such as AMD Zen.
    d) Inertia and brand name is what is probably still driving many of their current sales.

    What can Intel do to improve their situation, given that a), b) and c) are falling behind? They moved on a) by outsourcing manufacturing of part of their processors, so they can compete in the same arena. However, they are wasting their investment on next-gen fabs. They might modify b), moving to an ARM or RISC-V architecture, but this would break all their backwards compatibility, so vendors would have no particular interest to stick to them (and their "premium" price tag). Regarding c) their microarchitecture, we all know it is plagued with security bugs and they only shine on single-core performance; they should probably use a clean-slate design, but this is not something you can do in months. And regarding d), since they are already market leaders they can simply increase their marketing spending or tighten their lock-in agreements (with potential anti-monopoly fines).

    It seems that Intel has done the only thing that was available to them to remain competitive in the short term. But this will be very problematic in the medium and long terms...

    • The value proposition of Intel is 'Nobody ever got fired for buying IBM' and it's probably going to fare just as well.

    • What is the value proposition for any CPU manufacturer anymore? I don't want to say 640k ought to be enough for anybody, but at some point, people just aren't clamoring for high-end CPUs as much anymore. The refresh cycle keeps getting longer and longer and low-end to mid-range CPUs are an improvement over the equipment being replaced.

      • Re: (Score:2)

        by Luckyo ( 1726890 )

        Between explosion of high end computing in servers ranging from simulations to cloud services, and gaming where high end CPUs are in increasingly high demand, high end CPUs are in extreme enough demand that you will have problems buying them right now, just like a lot of other hardware aimed at the same markets.

        So same applies to mid end as well. But the fact that both high end and mid end is sold out suggests significant demand for both.

  • Intel should contract out to TSMC to have them assist with getting the Intel fabs up and running. This would require TSMC IP so it would not be cheap. TSMC might request a percentage of the silicon value derived from TSMC IP - or perhaps a lump sum. Either way, this method increases the global supply of high-end silicon instead of putting further constraints on what TSMC can output.

    • TSMC laugh them out of the room. Literally.

      • Quite possibly. But if you were bringing enough cash to the table they might at least think about it. They are all about making money after all. The trick would be making it more profitable for TSMC then the alternative.

        The big advantage for TSMC is that their production is mostly all sold out for the next couple of years. By converting an Intel fab, they could get more production up and running faster. This is important for a customer like Intel because the quantity of silicon required would most c

    • Re: (Score:3)

      by Luckyo ( 1726890 )

      TSMC officials off the record have stated several times that if intel was to ever come knocking even for the manufacturing of their CPUs, they'd give them "whatever we have left, with minimal effort", because they expect intel to just go back to its own fabs as soon as they are competitive again.

      What you're suggesting is to expect TSMC to do something even more stupid. To get a small short term benefit for building up a major long term competitor. They're mostly sold out on the 7nm and 5nm anyway, so they d

  • Why else would TSMC divert valuable resource from its normal clients that are already struggling to fill needs.

    I bet Intel gambles that it will do more to save them, profit-wise, than it costs.
    But they will still have massively lowered profits, and shrink as a business.

    And it will be funny, when everyone else will be paying just a bit more, to starve Intel off. Because if I were pretty much any other business... to which Intel has done exactly that, if they did not obey ... I'd think that'd be some hilariou

  • For the most part the core proper is converged once and instantiated in all the SKUs. To move the i3 SKUs to TSMC N5 means they have to reconverge the ported core. More than just rerunning the place and route, this means redesigning the custom datapath, analog, and memory circuits.

    For this to be viable means there are designers sitting around doing nothing, either in-house or close enough that they can be quickly hired.

  • While this is a landmark, it's not over yet for Intel. They need to skip at least two generations and focus on future improvements, while maintaining at least some kind of production using outsourced fabs.

    There are good indications that it's exactly what they're planning.
  • Maybe we should start outsourcing toilet paper too. Don't leave anything out. Can we outsource the politicians that allow this to continue to happen as well? Our country needs financial aide to stop things like this, oops we outsource our aide to other countries too.

    -Anything left? Then turn those lights off on the way out.
  • Whatever happened to AMD's spun-off fab company, GlobalFoundries? Somehow, everybody only seems to talk about TSMC and Samsung's fabs?

    • GlobalFoundries gave up their plans to go to 7nm, apparently they did not see a good chance to recoup the multi-billion dollar investment.

      Right now they are still making some older Ryzen models in 12nm, and the I/O chips on current Ryzen and Epyc CPUs. Probably some older AMD GPUs too, there are still some Radeon 5xx in the market.

      Plus maybe customers outside of AMD, but I'm not gonna google these now.

  • And it's not the engineers' fault
    Another great company killed by idiot managers

  • https://en.wikipedia.org/wiki/... [wikipedia.org] There is a corollary to Moore's law stating that the cost of the fab plant doubles every 4 years. The cost of the latest fab plants are likely over 20 Billion. There just isn't a business case to have 2 leading edge foundries anymore. Someone has to get squeezed out of the market. Intel and AMD are already spending about 40% of their revenue on R and D. It definitely looks like the world is heading slowly towards a monopoly.
  • This is what Intel can do to get back in the marketing game with both TSMC and Samsung as well as AMD, AAPL and NVDA. - Rename 10+ nm process to 7 nm process - Rename 10++ nm process to 6 nm process - Rename 7 nm process to 5 nm process - Rename 7+ nm process to 4 nm process - Rename 7++ nm process to 3 nm process

  • They clearly paid a large amount to deny fabrication resources to AMD

  • also... (Score:2)

    by smash ( 1351 )
    ... any of TSMC's limited wafers that intel can consume are wafers that aren't used to build Apple M1/M2/etc. or AMD Ryzen. I suspect using TSMC is as much about that as it is trying to build better products.

Slashdot Top Deals

You are always doing something marginal when the boss drops by your desk.

Close