Follow Slashdot stories on Twitter


Forgot your password?
Check out the new SourceForge HTML5 internet speed test! No Flash necessary and runs on all devices. ×

Fin-Fet Transistors on the Horizon 138

MORTAR_COMBAT! writes "According to this 9 September article, IBM scientists have "manufactured a working static RAM chip out of so-called Fin-Fet transistors, which feature two gates, rather than a single one, for conducting electricity". What does this mean for us? 50 percent performance increases, due to increased throughput of electricity, and 50 percent less power usage, due to decreased electrical leakage. Longer battery life for laptops, lower power bills for server farms. Moore's law lives on. More pretty pictures here."
This discussion has been archived. No new comments can be posted.

Fin-Fet Transistors on the Horizon

Comments Filter:
  • Well, this sort of let's the air out of HP's bubble from two stories down! :)
    • Mentioned in that story was that HP scientists are scrambling to publish anything they've been working on because people not putting out anything contributing to the bottom line in a year are going to get laid off.

      I know that IBM just fired thousands of people and has a hiring freeze on their RAM research division (I have a friend that works there), besides letting a number of people go there.

      I suspect the IBM scientists are in the same pickle as HP. When the economy goes down, the first thing to get axed is R&D, and they'd rather not be out on the street.
      • Re:It's like HP (Score:2, Interesting)

        I work for an IBM owned company, luckily we've been sheltered from most of the bad things (at least where I'm at)

        What really helps when working for a large company like IBM, is DONT BECOME A NUMBER. Make sure that you are in a position where customers have to frequently refer to you by "name" and not by position. It also helps if your name becomes almost a cliche' for a particular action or service.

        When the time comes, the numbers will get laid off, the engineer clients know by name won't.
  • by flewp ( 458359 )
    Now when I buy my new machine it'll be eventually outdated!

    Damn you Moore and your laws!
  • Man oh man, the this throws a whole new wrench into the gigahertz wars. Amd and Intel lookout, IBM gonna be rolling out some 33 mhz processers that will whip both your collective asses and further confuse computer owners.

    This is good news but I sense wierdness in the space time contiunuim with this announcement.

    • Sorry, but IBM wins []

      That's 70GHz, folks. And no, this isn't vaporware, I've worked with a chip fabricated using this tech.

      • With SiGe, IBM tackles a new generation of wireless communications and data-conversion components, with application to products such as:

        Inexpensive safety systems for automobiles, including radar at up to 24 GHz for collision
        here's their list of applications for that process.... I don't see CPUs anywhere on there

        -warning or advanced cruise control.
        -Wireless voice and data handsets at 1.8 GHz and beyond, with both RF and digital subsystems on a single chip.
        -High-speed A/D and D/A converters for data acquisition, direct-to-baseband radio receivers, signal synthesis, and more.
        -Low-cost, portable Global Positioning Satellite (GPS) receivers.
        -Other innovative high-frequency products as the imagination and market evolve

        If you could make a CPU out of it, I think they'd list it as a application.

        • I'm sure you could make a CPU out of it, even if it was just a simple one. A simple CPU has less demanding fabrication requirements than all the things you listed. It just isn't economical for general-purpose low frequency use compared to normal silicon CMOS. SiGe is aimed at applications which use e.g. GaAs technology at the moment.
          • My bad, on the IBM site it says that they can't yet manufacture CMOS on SiGe, which is pretty much a requirement for CPUs these days. They can only manufacture bipolar transistors and analog devices.

            They do say that "Shortly, we will be releasing our BiCMOS SiGe technology...", which means that they will be able to manufacture bipolar transistors and MOS transistors on the same chip. In other words, you can't make a modern CPU with it now if you don't work for IBM, but you will be able to soon.

    • so if some large guy with an austrian accent shows up and asks if you are puto, say no, he just left. He went to . don't expect he'll be back, ever.
  • Woop! (Score:2, Insightful)

    Here's for more awesome AMD processors! My computer architecture class was discussing Moore's law and we all agreed that it was reaching a plateau in respect to the law itself. Looks like this is the breakthrough that will take processors to the next level.
  • Windows will have general protection faults twice as quickly.

    (I'm sorry, I just had too...)
  • by azav ( 469988 )
    I love it. The date on the press release says - January 11, 2002.

    • I would guess that the January date refers to when the information was first released within IBM. Corporations don't release this type of information before they have all the Intellectual Property stuff hammered out. ;)
    • ... also in related recent news ...

      IBM announced a new line of system processing units, dubbed the "personal computer". These amazing compact systems weigh only 53 lbs and clock in at 4.77 MhZ. This magic box puts more computing power at the fingertips of more hobbyists than ever before.

      IBM expects further growth in the Personal Computer field. "We could see speeds around 8MhZ within a few short years", claims Dick Johnson, chief engineer of the computing division.

  • I would love to get more power out of my laptop battery!!! Sign me up IBM!
  • "Cloning" the existing gates was the plan all along (:

    .. well, according to Jango anyway..
  • by Anonymous Coward
    Is it me or does that term remind me of Phen-Phen which was hyped as a weight loss pill?

    Fin Fet: reduces voltage leakage
    Phen Phen: reduces fat (increases chances of death)
    Olestra: increases leakage while reduces fat absorption

  • by HaeMaker ( 221642 ) on Tuesday September 10, 2002 @07:19PM (#4232872) Homepage
    Link []
  • Wouldn't that mean nothing in terms of the speed at which they can be switched on and off? Wouldn't that also generate moer heat thatn normal CMOSFET chips?
    • Umm... No, think 50% less resistance

      I can transfer 50% more power down my big fat cables than you can down your thin ones........
      • The speed of a transistor is pretty much governed by it's gate capacitance (correct me if I'm wrong, I don't feel like getting out The Art of Electronics). The current flowing through a transistor has no meaning other than to switch the transistor on and off. Don't forget that these are digital circuts. Just because a FET has a lower transmission resistance (I forget what the name is called) it still can allow enough current to flow to turn on the next transistor. This is all that it needed in digital applications because most often the signal is restored to it's original state after every transistor. The lower resistance, however, is good for RAM because it can drive lines with more power, permitting more liberal refresh times (I think.)

  • An improvement in technology that makes computers FASTER? Who could have imagined THAT would ever happen?
  • Intel, though, can boast of research breakthroughs of its own, as well as far higher sales volumes.

    Too many research breakthroughs to mention here, apparently. Also, how does sales volume figure into a discussion about a technological breakthrough? Wouldn't that be something like saying Unix is technically superior, but Windows outsells it. Oh wait, they say stuff like that all the time!

    Despite the downturn in the PC industry, Intel remains the largest semiconductor manufacturer in the world.

    The spin on that one is a little harder to spot, but it's there. Sales across the industry could increase or decrease without changing the relative market share of the various manufacturers, so why even mention it?

    Because C|NET is owned partially by Intel, and is heavily biased towards both Intel and Microsoft. They never say anything positive about IBM or Motorola without getting in a quick mention about Intel, and they never say anything nice about Unix, Linux, or Mac OS X without a tip of the hat to Microsoft. It's kind of fun when you know what you are looking for.

    • by Anonymous Coward
      > Because C|NET is owned partially by Intel

      Ummm. Nope. Intel sold their position some time ago. Here's your lovely parting gift.
  • I looked at the picture, and couldn't figure out what was going on. Where are the drain and source contacts? Does the inversion layer form in the subtrate or in the fin?

    I understand trying to make it simpler, but why remove all information that's meaningful? They might as well say "here's an electromicrograph that looks like a tree, and here's a glossy diagram with some pretty boxes and arrows, but no actual information."


  • by tweakt ( 325224 ) on Tuesday September 10, 2002 @07:36PM (#4233020) Homepage
    Please, the world has enough "Gates" already ;-)
  • I would think that the performance increase OR lower power usage would be the result... not both at once. If you take the performance increase, you need to use the same amount of power... or if you take the power savings, performance needs to stay the same.

    Someone correct me if I'm wrong.
  • by RichMan ( 8097 ) on Tuesday September 10, 2002 @07:42PM (#4233061)
    The first one is very good. It explains the problems with conventional scaling methods then presents the solution to the Gordian knot, the FinFet. Found by searching IBM chips [] (It is on my information resources list)

    Maintaining the benefits of CMOS scaling when scaling bogs down []

    Process requirements for continued scaling of CMOS--the need and prospects for atomic-level manipulation
    • by RichMan ( 8097 ) on Tuesday September 10, 2002 @07:51PM (#4233153)
      Ok, now I understand. I was much confused by all the press writing "two gate" device. Every rational NAND/NOR gate made in a MOS process is made with 2 gates. A 4 input device would have 4 gates.

      The big advantage of the FinFet device is rather than being an embedded surface device with the gate on top of the channel which is embedded in the substrate, the FinFet uses a channel elevated out of the substrate so the gate wraps three sides of the channel. The papers report access to the top and bottom of the channel as "two gates" it is really a three side wrapping of the source-drain channel which is raised out of the substrate.

      The big advantage is that for a given gate voltage the penetration into the channel in blocking carriers is only so far. With the gate on both(3) sides of the channel the penetration effectiveness for a given voltage is greatly increased.
    • You missed the most important one!
      Beyond the conventional transistor by H.-S. P. Wong []
      5. Conclusion

      From an economic point of view, conventional devices and materials will continue to be employed until they become impractical. Efforts devoted to push conventional approaches as far as they can go [13, 15, 47] continue to be extremely important. In this paper, we review the approaches to circumvent or surmount the barriers to device scaling. These approaches generally fall into two categories: new materials and new device structures. We describe materials innovations for the gate stack and the transistor channel. We also review device structural innovations such as the double-gate FET. While we focus this paper on these unconventional approaches, it is expected that innovations in conventional technologies such as etching, CMP, dopant profile control, and contact formation (silicides) will continue to be advanced and needed even for these "unconventional" approaches. Our application focus is high-performance systems. One may well arrive at a quite different conclusion if considerations are driven by other applications (e.g., low power, analog, sensor systems).

      While nanotechnology may be seen by many as a successor to silicon microelectronics technology, it is clear from the results to date that it will be many years before nanotechnology can reach the level of maturity of the present silicon technology. The deployment of nanotechnology will most probably occur first in niche applications that complement conventional silicon technologies. Future social and application changes will further the shift from microelectronics to nanotechnology on a time scale of decades.

      In the near term, there are many avenues for system performance improvements stemming from device, circuit, and system optimization that are yet to be exploited, which may provide orders of magnitude of power-performance tradeoff. However, this may call for system products based on application-specific device, circuit, and system designs, as opposed to today's general-purpose approach. Despite some speculations as to the impending end of progress, there is still plenty of room for continued technological advancement.
      [Sorry Mr. Wong for posting your conclusion. Here is the
      copyright info [] and it seems to say that it is ok to reproduce the article. Printed form is different than electronic form. :) ]
    • That's funny, nowhere in the CNet article did it mention that the whole idea originated at UC Berkeley. There's even a Lego model of a FinFET transistor in the lobby of Cory Hall.

      A quick search on Google for "FinFET" will get you a whole lot of references.
    • C GATE = ( [epsilon] OX [epsilon] 0/TOX)L GATE + 0.26 fF/m




      [tau] = CGATEVDD/IDSAT

      Oh, of course. I understand much better now.
  • Darlington Melded? (Score:1, Informative)

    by Anonymous Coward
    It's not quite a Melded Darlington, but it certainly reminds me of a Darlington pair.
    A Darlington is two transistors tied together where emitter of one goes straight into the base of the other. This basically sharpens the gain, but you pay a price in speed. Nonetheless, Darlingtons are used, as well as Photo-Darlingtons.

    I had just about forgotten everything about transitors from my EE days until I picked up robotics. Software really isoltated you from how things really work.

    New transitor designs are a dime a dozen. For instance the tunnel diode. (A diode is the most basic semi-conductor, a transitor is basically two diodes.

    Pick up electronics as a hobby. I urge you EE's out there that like me are writing business software. It's very rewarding.
  • Obviously right now this is much more expensive than SRAM because it's not being mass-produced. But let's face it, computers would be faster and cheaper (because of reduced circuitry from lack of refresh stuff) if we could use SRAM, but because it's so much more expensive than DRAM, we don't. My question is this: at the same size (capacity and size in micrometers) how much more/less will this cost than SRAM? Does the price become comparable to DRAM? if it is comparable to SRAM, we will definalty see improvements, but the majop improvements will come if it becomes near DRAM (even if a little more expensive) we could see drastic increases in performance. My 2 cents.
    • This IS SRAM. It has all the features of SRAM. Only instead of being built out of ordinary transistors, it is built out of FinFet transistors, which feature a smaller leakage current. The drawback is they are harder/more expensive to make.

      So it's never going to be cheaper. It might one day be faster and/or more efficient than current chips.

  • Won't they get sued over that?
  • curious.. anyone know what 'FIN' stands for? 'MOS' in MOS-Fet stands for Metal Oxide Semiconductor.. but FIN?

    Aren't field effect transistors cool? I remember my first MOS-FET HT from Yaesu ( []) - promised and delivered on longer battery life.. can't wait to see the nextgen communications products using this technology....

    -j []
  • Actually the performance benefit from double-gate is minimal. The approximate delay associated with switching a capacitor is CV/I, where C is the capacitance of the gate, V is the source voltage, and I is the on current of the device. Double-gate gives you double (or slightly more) the current, at the expense of twice the capacitance. You don't really gain much at the same gate length. The real advantage is scaling. You can make shorter double-gate FETs, and gain the kind of performance you're used to from following Moore's "Law".
    • I'm not shure what the fin does practically but I would guess it strangles the channel laterally; there isn't much capacitance on that side don't you think so? I think it also helps spreading the E field more evenly to avoid breakdown on voltage swings. This supposing the channel width is that of the gate.
  • More Slashdot hype (Score:4, Informative)

    by trenton ( 53581 ) < minus poet> on Tuesday September 10, 2002 @08:02PM (#4233254) Homepage
    And the hype lives on! Compare and contrast:
    • From the post: "... and 50 percent less power usage."
    • From the web site []: "... new type of transistor which reduces power consumption by 20 - 25%."
    Somewhere along the way, that thing got twice as efficient! Amazing design.
  • ...catchphrases.

    Moore's Law involves the doubling of the NUMBER OF TRANSISTORS in an area (transistor density). If you have a single transistir acting as two, you're acting directly in contravention of Moore's Law. As soon as I had the gist of what was going on for this story i knew some idiot would say something about Moore's Law. Might as well have asked how powerful a Beowulf cluster of processors with these chips might be behind China's firewall while using Google to look for Natalie Portman's case mods.

    - Chris
    • ...for hosting quake servers.
    • Actually, the editor that you are chiding has a valid point in my opinion. Leakage is headed towards the point where it won't matter how many transistors you can pattern in lithography ino a given area if they leak so much that you can never be sure when they turned off. FinFET's are one technique to enable transistors to actually work as MOS transistors when the industry heads below 50nm process technologies. Without some solution to the leakage problem, Moore's law is in some danger of becoming invalid.
    • I'd have to disagree. Moore's law wasn't specific to transistors, but using transistors as a physical evidence was easier then.

      Moore's Law /morz law/ prov. The observation that the logic density of silicon integrated circuits has closely followed the curve (bits per square inch) = 2^(t - 1962) where t is time in years; that is, the amount of information storable on a given amount of silicon has roughly doubled every year since the technology was invented. This relation, first uttered in 1964 by semiconductor engineer Gordon Moore (who co-founded Intel four years later) held until the late 1970s, at which point the doubling period slowed to 18 months. The doubling period remained at that value through time of writing (late 1999). Moore's Law is apparently self-fulfilling. The implication is that somebody, somewhere is going to be able to build a better chip than you if you rest on your laurels, so you'd better start pushing hard on the problem. See also Parkinson's Law of Data and Gates's Law.

      From this, one can see it's the logic density that's important, the actual ability for information to be passed through. If we find a way to do it 16 ways through a single transistor, is that not 16 times more efficient (and therefore 16 times more data that can be 'stored' for use) on a die?
  • ... scientists at IBM created a working prototype static RAM chip using Jango-Fett transistors. Consuming 4 times the power, it will store information with a cold air of dread and competence.

    A secondary line of static RAM chips based on an exact genetic replica (called Boba-Fett transistors) will be developed throughout the year for mobile computing purposes. Support for the Dark Side [] is eminent.
  • Static vs Dynamic (Score:3, Interesting)

    by Nethead ( 1563 ) <> on Tuesday September 10, 2002 @08:47PM (#4233560) Homepage Journal
    You won't see much static RAM (SRAM) in a server farm. SRAM is what most call CMOS RAM (for most of the wrong reasons.) A server farm runs on Dynamic RAM (DRAM) but it's not where the power is chewed up. Disk drives and CPU's take the power, esp. the 10,000 and 15,000 RPM SCSI drives in use today.

    The devices that will gain some power savings are those that we'll enjoy it most in; handheld toys!
    • You forgot about caches, where SRAM speed and power dissipation matters greatly.

      Increasing the performance of these types of memory is pretty damn important if you ask me. Reducing the leakage and therefore heat dissipation in and around the CPU is pretty helpful, too (think L1 cache.)

      - dvd_tude
    • I believe that cpu caches are typically built from static ram. Cray has some beautiful "old" machines which used exclusively static ram, IIRC.

      While IBM mentioned the application of this new transistor to static ram, I would guess it would be useful for any fast silicon with transistors (i.e. most =-). I'd appreciate any replies which knew more about this either way.

      -Paul Komarek
      • Yes, it will be used for general applications, not just SRAM. Static RAM is often used as a test for new manufacturing processes because it is very repetitive. This makes it easy to design wafers completely covered by SRAM, and easy to test so the defect rate can be worked out.

        Also, the upcoming Intel/AMD cores will be using huge amounts of cache (of the order of 1MB, which until now has only been seen on pretty high-end server CPUs), so being able to reliably manufacture large amounts of it is important anyway.

  • by geekoid ( 135745 )
    oh fin-fet transistors, I thought for sure it said bobo fet transvestite.

    of course, I'm the guy who spent a night trying to get laid in a warehouse....

  • AMD just today announced that they have created 10 nm double gate transistors. Here is the AMD announcement []

  • Is this too much different from the dual-gate FETs [] that have been available for decades?
  • Good. It's about time, though 50% of current high-power CPUs may be too little too late. I know of at least one major embedded systems corporation in a panic, because modern CPUs consume way too much power for use in many embedded environments.
  • Slashdot. Vapor for nerds. Stuff that won't matter for years.

    Not that I'm complaining about new breakthroughs, but it sure seems like the vapor:substance ratio is sucking eggs lately, at least on slashdot. When someone offers an actual working product for some reasonable cost, maybe then I'll get excited. Until then I'll just stuff this into the mental round file.
  • One probable reason that the industry is looking closely at finFETs is that the original invention of them at UC Berkeley was not patented originally. Note that there are several patents on fabrication methods for manufacturing them now, but the original invention was not patented.

    From an article [] about the early work on this at Berkeley:

    Hu said the FinFET prototype was successfully fabricated last July and appeared to perform well. He said no patent had been taken out on the device. "We made the decision not to patent," Hu said. "We want the widest possible usage. We hope this becomes a mainstream transistor structure in the future."

    As a VLSI design engineer working in the industry, I can see that finFET's are becoming a serious technology contender in the 50nm process timeframe.
  • and 50 percent less power usage, due to decreased electrical leakage. Longer battery life for laptops,

    Battery life for laptops has always been 2 hours. It will always be 2 hours. This is the minimum we'll put up with, and thus we'll invariably find ways to suck up the power until it *is* two hours. Much like how we're so very willing to load bloatware on our computers until windows takes 5 minutes to load. Any more, and we'll think it's too long. Any less and we'll think there's room to spare.
  • Moore's law lives on.

    Why is Moore's Law always referred to with a shrug, as if it's some amazing, consistent, unstoppable force? The results that are interpreted as "Moore's Law" exist purely through human effort. Surely Moore's Law can't actually be the law of maximum human ability to improve, can it? Surely people in high places at Intel are throttling the engineers back when they get ahead of themselves, and pouring on the cash when they get behind... Setting and meeting expectations is what matters most to the stock market, after all. Moore's Law is just a means to that end.

  • like magnetic ram, and all the other wiz-bang new techs in the news recently.

    Question is, when is it going to be common? Nowadays, it seems even with standards, solid backing, things sometimes still don't take off.

    And until they do, it's got nothing to do with us little people.
  • until I can walk into my corner convenience store and pick up the cheap knock-off:

    Herbal Fin-Fet.

As of next Tuesday, C will be flushed in favor of COBOL. Please update your programs.