Become a fan of Slashdot on Facebook

 


Forgot your password?
Close
typodupeerror
×
Intel Hardware

Intel Says New Transistor Technology Could Boost Chip Performance 20% (reuters.com) 94

Posted by msmash from the how-about-that dept.
Intel on Thursday disclosed a new method for making transistors on semiconductors that its chief architect said could boost the performance Intel's next round of processors by as much as 20%. From a report: The Santa Clara, California-based company is one of the few remaining in the world that both designs and manufactures its own chips. But its manufacturing operations have become a concern among investors after Intel last month said that its next-generation chip-making process, called its 7-nanometer process node, would be delayed. Analysts believe the delays could cement the lead that rivals such as Taiwan Semiconductor Manufacturing Co have gained in making smaller, more power efficient chips. Intel's shares have fallen nearly 20% since the delays were disclosed. On Thursday, Intel sought to buck the notion that the single-number names given to each generation of chip process node tell the entire story by disclosing improvements on its existing 10-nanonmeter process node. It announced a new way of making what it now calls "SuperFin" transistors, which, along with a new material being used to improve the capacitors on chips, is expected to boost the performance of Intel's forthcoming processors, despite their still being made on 10-nanometer manufacturing lines.
This discussion has been archived. No new comments can be posted.

Intel Says New Transistor Technology Could Boost Chip Performance 20% More

Intel Says New Transistor Technology Could Boost Chip Performance 20%

Comments Filter:

  • Lame (Score:2, Interesting)

    by Anonymous Coward

    I remember a time when each new CPU released by Intel was 200%+ faster than the previous. It hasn't been worth upgrading for years now.

    • Re: (Score:3)

      by Amouth ( 879122 )

      honestly since we got SMP in everyday computers and could drop TDM there has been little reason to upgrade CPUs and chase the speed.. the next limiter was I/O which SSD solved vs. spinning platter.

      So yea rock that old computer with plenty of power (posted on a daily use work laptop with a release date in 2014)

  • Translating from typical bullshit to reality I assume this means 2% or less.

    Who would trust their numbers?

    • Amazing how its a 20% drop in stock value followed by boasts of 20% in hidden value.

      • Re: (Score:1, Troll)

        by flyingfsck ( 986395 )

        It sure sounds like a stock price boost is the main reason for this iNnovation.

        I'm wondering what new iNsecurity features will come along for the ride.

  • Dammit, Moore, I thought you had left (Score:5, Funny)

    by goombah99 ( 560566 ) on Thursday August 13, 2020 @06:13PM (#60399195)

    I was so ready for the end of Moore's law. Now I have to wait till next season to see how it will end.

  • Will it fix all your hardware issues? (Score:3)

    by Snotnose ( 212196 ) on Thursday August 13, 2020 @06:16PM (#60399223)
    If so, will fixing your hardware faults make you faster than, I dunno, existing silicon? We won't even mention AMD here, we're talking your new fastest bestest chipmaking vs your current VHDL designs.

    • Re: (Score:3, Interesting)

      by SuricouRaven ( 1897204 )

      At some point, the x86 family are going to have to hit the limits of the design - forty years of legacy rubbish is going to be an issue. One of the most difficult problems in multiprocessor design is cache coherence, as a consequence of needing to maintain compatibility with software that predates DOS.

      But that point can be postponed. Always postponed, because there's just too much of an existing software base for x86 now for alternatives to catch on.

      • Re:Will it fix all your hardware issues? (Score:5, Interesting)

        by Rockoon ( 1252108 ) on Thursday August 13, 2020 @06:47PM (#60399339)
        Cache coherence is important because some shit really needs to be both low latency as well as coherent across all threads.

        You are trying to shove a gpu/simd cache model onto cpu cache design, while claiming that cpu design needs to "get with the times" or whatever, but in fact you are ignoring the basic reality that there are two diametrically opposed cache designs in use today; one focusing on achieving remarkably low latency (3 cycle L0? no problem), and the other focusing on achieving high bandwidth. You are looking for the high bandwidth model and thats over in the GPU section why not go bitch at nvidia.

      • One of the most difficult problems in multiprocessor design is cache coherence, as a consequence of needing to maintain compatibility with software that predates DOS.

        This sounds interesting, why is this harder with x86 compatibility? All processors benefit from cache, and PC software which predates DOS seems unlikely to be MP-aware.

        The x86 decoder in any amd64 or compatible processor is a teeny, tiny part of the CPU. Now we actually have marvelous things like instructions whose operands can be any of multiple plentiful registers.

      • Re: (Score:2)

        by Z80a ( 971949 )

        "x86" is just a tiny portion of the chip that wouldn't give you any significant advantage for stripping it away, just the massive disadvantage of killing the IBM PC platform that is the only current popular platform that is not basically a game console in terms of locked down hardware and obscure documentation.
        Yes, there are some open hardware projects here and there, and raspi is quite popular, but it would be absolutely horrible for software and hardware freedom if the PC went away.

      • Always postponed, because there's just too much of an existing software base for x86 now for alternatives to catch on.

        Pretty soon everything is going to be written in HTML, Javascript, or Webassembly. So architecture won't matter.

        • Except there has to be code somewhere that executes the Javascript/Webassembly. That is written and highly optimised for x86 (and probably ARM as well).

          • Yeah, so for the people who write Chrome and Firefox, architecture matters. For everyone else, it doesn't.

      • Re: (Score:2)

        by AmiMoJo ( 196126 )

        Modern x86 has a legacy mode that disables most of the performance enhancements and allows older stuff to run. As soon as the UEFI firmware has legacy support turned off or boots a modern OS it disables that mode and all the old crud that goes with it.

        In AMD64 mode even more cruft is disabled.

      • I'd like to see a total redesign like how the AS/400 did things. Everything is an address in a 128 bit space, even devices.

      • Are we sure that "legacy rubbish" is significantly impacting the design and performance of x86 chips?
        I'm no CPU expert but we might be whinning about something really small. I guess the real proof will arrive when Apple releases their PCs with ARM CPUs. If they're significantly better than the x86s you might be right.

        • Do we have any CPU design experts here to tell us? The number of people who really understand how a modern microprocessor works in depth is probably quite small.

    • What hardware issues? I hope you're not talking about Spectre and Meltdown. I for one an hoping that AMD adopts these "issues". I could do with a speed boost for a completely irrelevant to me security issue.

  • You mean smaller lithography? (Score:5, Interesting)

    by bobbied ( 2522392 ) on Thursday August 13, 2020 @06:26PM (#60399273)

    You mean smaller lithography ?

    Come on Intel, you guys are getting your heads handed to you by AMD right now because you simply cannot get your smaller manufacturing process online in sufficient yields to keep up. Now you "invent" some new way to build transistors that are faster and use less power (cough, cough), maybe going smaller than 14nm might help? Come on, they already have announced 10nm, I'm wondering if it's the same thing.

    Hummm.. Is it earning season or something? Could they be propping up their stock price with a garbage story like this? Inquiring minds want to know.

    • Re:You mean smaller lithography? (Score:5, Informative)

      by Solandri ( 704621 ) on Thursday August 13, 2020 @06:53PM (#60399353)
      The different fabs measure different parts of the transistor to come up with their "x nm" definitions
      • Intel's 14 nm process is 37.5 million transistors per mm^2
      • TSMC's 10 nm process is 52.5 million transistors per mm^2
      • TSMC's 7 nm process is 96.5 million transistors per mm^2
      • Intel's 10 nm process is 100.8 million transistors per mm^2. That's right, Intel's "10 nm" is actually smaller than TSMC's "7 nm" [wikipedia.org].
      • TSMC's 5 nm process will be 173 million transistors per mm^2
      • Intel's 7nm process will be about 200 million transistors per mm^2

      • Re: (Score:3, Insightful)

        by bobbied ( 2522392 )

        Yea, but if your yield isn't high enough, it doesn't matter what you call it. You will lose your shirt in manufacturing piles of chips you cannot sell because they have too many flaws.

        Intel is struggling to get their 10nm process yields up to profitable levels. AMD is eating their lunch. A situation I don't see changing in the near term.

      • Re: (Score:1)

        by Anonymous Coward

        That might matter a shit if Intel could actually get a real world yield. As usual Intal is all smokescreen.

      • Re: (Score:1)

        by Anonymous Coward

        Get the INTEL advantage now with more nm than competitors:

        ------> 7nm
        -------------------->14nm+++++(TM)

        Intel's patented 14nm+++++(TM) process delivers more nm per chip.*

        *chip is the computer brain

      • Re: (Score:2)

        by Entrope ( 68843 )

        Yes, Intel has picked a grand total of two simple gates (a two-input NAND and a single flip-flop) to calculate their transistor density. But why have they stopped publishing the transistor counts for their chips? They have said they do not want to compare achieved transistor densities, which suggests they only compare well on a theoretical number rather than what is delivered.

      • Intel's "10 nm" is actually smaller than TSMC's "7 nm"

        You mean, for a sufficiently restricted definition of "smaller"?

      • Re: (Score:2)

        by robi5 ( 1261542 )

        As Intel's moving from 14 to 10nm, if linear, would give an approx. 2x transistor count increase, yet 37.5 millions to 100.8 millions is much closer to 3x.
        Also, neither density nor nm class tell the whole story, as other important metrics eg. voltage, TDP, peak thermal, frequency etc. are related but still different

    • Re: (Score:3)

      by Zuriel ( 1760072 )
      Intel has been shipping 10nm for a while, it's just not very good. So most of their products are still on 14nm.

    • Come on Intel, you guys are getting your heads handed to you by AMD right now because you simply cannot get your smaller manufacturing process online in sufficient yields to keep up.

      The most laughable part of it all is that the chiplette solution to bad yields is only new as a solution to bad yields. Intel previously used chiplettes as a solution to being without a dual core product back in the day, after all.

      The writing was on the wall ~3 years ago now so Intel is still about ~2 years away from a chiplette redesign (thats the turn-around time in the industry without hacks.)

      Its been a long time since Intel didnt have an absurd transistor budget advantage. Expect absurdly deep pipe

      • While it wasn't explicitly meant to improve yields, rather enabling them to be possible at all (yield > 0 is an improvement, I suppose), Pentium Pro used a separate 256 or 512KB cache die (or two, for 1MB cache); to provide large in-package L2 cache. What's old is new again, I guess.

        Though I can't find any good photos of one of the 1MB chips delidded (they are all the aluminum+fiberglass modules), the datasheet [sandyflat.net] has a low quality shot of the layout.

    • Re:You mean smaller lithography? (Score:5, Interesting)

      by williamyf ( 227051 ) on Thursday August 13, 2020 @07:56PM (#60399519)

      You mean smaller lithography ?

      Come on Intel, you guys are getting your heads handed to you by AMD right now because you simply cannot get your smaller manufacturing process online in sufficient yields to keep up. Now you "invent" some new way to build transistors that are faster and use less power (cough, cough), maybe going smaller than 14nm might help? Come on, they already have announced 10nm, I'm wondering if it's the same thing.

      Hummm.. Is it earning season or something? Could they be propping up their stock price with a garbage story like this? Inquiring minds want to know.

      AMD does not Fab their own processors. And the part of AMD that used to fabricate semiconductors only makes the Northbridges of Zen2. The cores and Graphics cards are done by TSMC (with Samsung as second source). So, I'd say is AMD's great architecture coupled with TSMC's great fab technology that has Intel on the ropes. A smaller process node would not be enough to offset the architectural advantage, and a better architecture would not be enough to offset manufacturing advantage. At this point, intel needs both.

      Actually, "inventing" some way to make transistors (or other parts of the chip) go faster without getting smaller is a technique which has been used time and time again. AMD (when they fabbed chips) and Motorola (when they fabbed chips) were first to market with Cooper Interconnects and Damascene tech. Intel was first with High-K dielectris. Also with FinFets. I do not remeber who was fist with Cobalt interconnects. All those techniques are now used industry wide. The "invention" here is twofold, a technique to improve the FinFets, and a technique to make a capacitor with a higher capacitance per unit of area.

      Intel's earnings call was one or two weeks ago. As TFS says, the share already lost 20% due to the bad news. Next "earnings seasson" comes in more or less 3 months. So I guess this anouncement is not an attempt to pump up share value. It reamins to be seen how big or small the effect of the new technology is, until then, is hard to qualify the story as "garbage".

      The new transistor technology (+ capacitor technology) was anounced on the 10nm node, not on the 14nm++++ node.

      TSMC said that they will have a similar technology for their (upcoming) 5nm, but have nothing similar now.

      The commonly held wisdom is that intel's 10nm process is roughly equivalent to TSMCs 7nm. Except, that TSMC has much better yields because Intel went with Multi-Multi-Patterning (double multi intended) while TSMC went with EUV and that (except for AMD and nVidia graphics) most of the chip(lets) TSMC manufactures are smaller than Intel's.

    • Now you "invent" some new way to build transistors that are faster and use less power (cough, cough)

      Are you implying that Intel hasn't invented a new way of building a transistor multiple times over the last 40 years, or are you implying that you don't know history?

      • When you say it like that, what Intel has done is really remarkable. These weren't just incremental improvements either, some of them were quite dramatic.

  • (this is their playbook move to slow the loss of customers to AMD)

  • Distraction (Score:3)

    by Gravis Zero ( 934156 ) on Thursday August 13, 2020 @06:53PM (#60399357)

    Here's a more apt title.

    Intel Trying To Distract Investors From Failing Business With Shiny Object

    • That would make far more sense if they did it before or during their earnings report, rather than a long time before the next one.

      • With Intel, PR is a balancing act. Any bad news that comes out needs a good news chaser announcement. They don't want people focusing on anything bad for too long lest they create plans to shift their investments elsewhere.

        • Well I suppose you think a company can only produce bad news and that bad news always follows bad news, and any example contrary to this must be some kind of *epic deep voice* conspiracy. */epic deep voice*

  • real headline (Score:4, Insightful)

    by slashmydots ( 2189826 ) on Thursday August 13, 2020 @07:19PM (#60399427)
    "Intel releases vaporware futurist nonsense to attempt to salvage stock price"

    • What makes you say it's vapourware? Intel has a long history of semiconductor developments most of which they have actively used in actual products, or even licensed to others.

  • This is an example of a hostile takeover of MBA's. Everything is going to crap and they call the marketing department to spin up some new FUD and a hip name like "SuperFin".

    Here is what I say, "If Intel doesn't get it's sh*t together it is going to be SuperFIN as in Super Final for real."

    Fire the MBAs and let the tech innovators do their jobs.

    • Based on what? I mean Intel has produced new technological developments hundreds of times in the past 40 years. This is literally part of their core business and why they invest more in R&D than the GDP of a small country.

      Your MBA nutjob theory would make more sense if they didn't just have their earnings report. If what you said were remotely true the Intel's MBAs must be the dumbest in the industry based on this timing.

      Or more likely the satellites got into your brain because your tinfoil hat was tune

  • Intel is getting spanked so hard right now that they're throwing everything at the wall.

    • Intel is getting spanked so hard right now that they're throwing everything at the wall.

      This is more about boosting their stock price by 20% than anything else.

  • In other words (Score:3, Funny)

    by sjames ( 1099 ) on Friday August 14, 2020 @02:00AM (#60400195) Homepage Journal

    It leaks all your data 20% faster.

  • Is that it? Just 20% increase? That means the next upgrade of Adobe reader will only be 30% slower
  • ... because if this isn't well-grounded in solid research with promising initial results, we'd pretty much have to conclude this was a blatant attempt at manipulating their stock price.

    A possibility that would look all the more "interesting" given the recent gains of their more nimble competitors.

    Maybe someone should call the SEC, just to be on the safe side.
  • But what are the security risks that are introduced? It's Intel, I gotta ask...

  • Remember that 20% that Intel took back on the intstruction Branch Prediction flaws. Now you can have your 2018 performance back.

  • IBM used to talk about how awesome their next mainframe would be, so please don't buy that Honeywell this year! Well, until the DOJ got in their shit about their monopoly. Under the consent decree, they were VERY careful not to pre-announce features for decades after that.

    So yeah, this very much has a "So please don't buy an AMD processor this year!" feel to it.

  • Believe it ONLY when you see it!
    Intel will do anything to scratch back any ground; even if it is exaggerated truth, or even bullshit!

Slashdot Top Deals

Two can Live as Cheaply as One for Half as Long. -- Howard Kandel

Close