Slashdot Mirror
Intel Designs Faster, 3D Transistor
lee1 writes "Intel has found a way to keep on the Moore's Law track by making smaller, faster and lower-power computer chips by building 3D transistors. They are already manufacturing microprocessors using this new design, called a FINFET (for fin field-effect transistor), which incorporates a small pillar, or fin, of silicon that sticks up above the surface of the chip. Intel said that it expected to be able to make chips that run as much as 37 percent faster in low-voltage applications and use as much as 50 percent less power. Products based on the new technology may appear some time later this year."
Is it possible to use that fin for cooling as well?
-- Cheers!
You'll still have to cough up an extra $200 for the privilege of using all the transistors in your Intel hardware though. Or maybe this will bring an end to them segregating things like HW virtualisation based on how deep the users' pockets are.
lots of blather about the 'next dimmension' of computer chips... 3d Inside or some such. Still, nice to see that moore's law still holds... at least for now...
Free Pie! The Pie is Also Evil!
When you die are you going to be able to look back at your life and say "job well done?" What little I know about you (this post) points to "no".
If it uses 50% less power, it will be cooler. The Atom processors generate very little heat compared with the more power-hungry Xeon and Core Duo chipsets.
Intel has become very aware of power usage as well as heat. The days of the room heating CPU are hopefully behind us.
"Trademarks are the heraldry of the new feudalism."
they are not 3D, they are just thinner and deeper than the standard, we still dont see transistors on top of each other. the latice is still pretty much 2D. i ussually dont complain too much, but slash dot summaries are batting way below the mendoza line.
Not much. You're talking about stuff on the nano-scale. So small, it won't even look different to the naked eye. The same thermal compounds should work with about the same amount of efficiency.
Millions of free accelerometers! (For high-G's anyway.)
When you die you are no longer able to look back at anything - you're dead. Other than that, nice trolling.
Eloi are stupid, throw morlocks at them!
D'oh - they should have thought of that much earlier! Get perpendicular - soon also available for SSDs. http://www.youtube.com/watch?v=-xPvD0Z9kz8
If it uses 50% less power, it will be cooler. The Atom processors generate very little heat compared with the more power-hungry Xeon and Core Duo chipsets.
Intel has become very aware of power usage as well as heat. The days of the room heating CPU are hopefully behind us.
Nope. Because no matter how efficient these processors become, there will always be somebody wanting a faster processor, and hey! The power envelope just got more room, just cram more of 'em in there!
Especially since this will be inside of the package, and not exposed to the thermal compound.
APK likes to ask for responses to the same things over and over. Maybe he just likes the responses?
What is indeed news is that intel is fielding them first.
Well, while it is nice a slashdot article has finally been written about FinFET's - there may already have been one, I just can't remember - these devices have been widely guessed to be a part of the 22 nm technology node for quite some time. (see: http://www.itrs.net/ and http://en.wikipedia.org/wiki/22_nanometer ).
They offer more effectivity for your gates as the field is not coming from one, but from 3 sides to the channel. That means a bit more scalability, but not much more. There is only a bit of improvement possible for the future in putting the gate below the channel as well (as hard as that may be, i, personally, don't think it would be worthwhile), so this won't save moore's law in the end.
It may not surprise you that they actually haven't been invented by intel, and are not new.
The term has been coined more than 10 years ago ( http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=823848 ) (find one of the free pdf's of this classic paper for yourself)
What is more interesting is how far down these transistors will scale in the extreme ultraviolet processes that are emerging right now.
What other trending marketing buzzwords can we apply to unrelated technology to get attention? Social Transistors? Transistors 2.0? Cloud Transistors? Tablet Transistors? iTransistor. What would truly be remarkable would be to develop a 2-dimensional transistor. Infinite layers of them with no height added to the die - and it'd still be slim enough to fit in a tablet!
What's next? A Twisted Transistor?
power and heat are the same thing
https://www.accountkiller.com/removal-requested
Haven't transistors always been 3D? You may draw them 2D in layout, but it's still a gate sitting on top of a channel with stacks of metal... it has L x W x H. I think that's called 3D. Silly Intel marketing.
Actually integrated transistors have been "planar" for the most part (although there have been "vfets" and other types of 3d channels in the past)...
In planar transistors, the field that chokes off the source/drain path has been mediated by a gate which is just on top of the channel on one plane. Imagine an iron on of a ironing board heating it up the board when you turn it on. Although the ironing board and the iron are both 3-dimensional objects, the interface in a "plane" and the heat diffuses across this plane. In this analogy with a planar transistor, the channel is dug into in the ironing board and the iron is the gate.
In finFET, the gate surrounds the channel on 3 side. Imagine now a tube on an ironing board and the iron has a notch cut in it so the iron surrounds the tube on three side. When you turn on the iron, heat diffuses across all around the notch instead of interfacing on 1 plane. This is "3D" or finFET instead of planar. In this analogy with a finFET transistor, the channel is the tube on the ironing board and the iron with a notch is the gate.
As you might imagine, the finFET architecture should have a better capability to turn on and off the channel since there is field is wrapping around the channel instead of just being applied to one side (okay that's simplification, but you get the idea).
Intel took, this finFET idea and added another twist with a "3", called tri-gate (or tri-channel depending on your point of view). This congolomeration of two independent ideas that both revolve around the number 3 is the kind of thing that drives marketing people to be silly ;^)
And for those that don't understand w/o a car analogy, imagine the difference in traction you get with bald tires on ice (planar where tire is the gate and ice is the channel), vs snow tires on dirt (finFET where the tire is the gate and dirt is the channel)...
They may be correlated, but they're not the same thing.
Power=rate at which work is performed
Heat=energy transfer from one place to another
3D? Aren't all transistors 3d? Along with every other physical item?
Coder's Stone: The programming language quick ref for iPad
when you refer to heat in a cpu system, you are referring to heat flux, which is joules per second, which is power. i'm not just playing with units. conceptually they are the same: the amount of heat transfer is identical to the amount of work done.
https://www.accountkiller.com/removal-requested
Perhaps a better car analogy would be a planar transistor is a two-cylinder single-port injector engine and the intel 3d-tri-gate is a six-cylinder multiport injector engine...
i was being neither. i think both of you are focusing too specifically on one domain or the other and not realizing physics as a whole is the same.
even in your equation Q + W, you only gave half of the equation dU is part of the heat transfer equation, but you ignored the transferring part, you know, the part that takes time, the dt.
Try this on: are you telling me that you can put 100W into a processor and less than that comes out as heat? because it all comes out as heat. hence they are the same physical property. if you disagree, congratulations on your nobel prize!
i think you are hung up on car engines vs. transistors, when in reality the heat transfer and power are the same thing in both cases.
this is a huge problem in engineering schools. the concepts are taught too rigidly, and the relationships are never fully explored which leads to misconceptions like the one you are arguing.
https://www.accountkiller.com/removal-requested
No, because the fundamentals are changing. In the power equation, voltage is squared. So if you reduce the voltage, the power goes down more. You can do more operations per watt. That equals less watts for the equivalent operations.
This post comes with a double-your-money-back guarantee!
Any offense taken to this post is at your sole discretion.
Yeah, considering you need several thousand of them grouped together to be perceptible to the naked eye.
This post comes with a double-your-money-back guarantee!
Any offense taken to this post is at your sole discretion.
When you die are you going to be able to look back at your life and say "job well done?" What little I know about you (this post) points to "no".
You, on the other hand, will be able to look upon a lifetime of helpful and insightful posts on /. as ANONYMOUS COWARD. Let me be the first to say "job well done"!
its a complex equation, but you're safe assuming the cube of voltage. you can be a little more specific: only switching power is proportional to the square of voltage. leakage is proportional just a little over the cube, and rushthrough is a little more than squared with power due to the dependency of transition time in th linear region (slope) on voltage. plus there's also frequency and its dependency on voltage, and temperature affect on leakage and Idsat. but the OP has a valid point, power density does increase, so the cooling solution has to be beefed up regardless because even less watts, over a smaller area, requires much more cooling capacity. it's the hot spot that you need to defend against.
https://www.accountkiller.com/removal-requested
I think you're getting hung up on the equations.
Let's keep it simple:
You supply 100W of power to black box. If less than 100W of heat transfers out (minus entropy), you've just observed a violation of physics or made a huge discovery.
Voltage regulator pumps in the energy in, heat sink sucks it out (well, so does the board); energy in energy out over time = watts.
Heat transfer = electrical energy transfer = power
There is no measure of "efficiency" because if a CPU demands 100W, it will expel 100W in heat, there's no way around that (minus entropy). (I'm also assuming a closed boundary around the cpu here, for the sake of the visual.)
I think what you're after when you say "Efficiency" is how much computation is done per Watt. But that's whole 'nother ball o' wax. If you are still thinking Carnot Efficiency, remember that is only a measure of the transfer of energy between two opposing mechanisms in a cycle: the boundary around the entire system remains constant (minus entropy).
https://www.accountkiller.com/removal-requested
thought of another example:
if you're delivering energy to a CPU at some # of Watts, you have to remove it just as fast... and it doesn't just vanish. how is it removed? heat transfer. also watts.
Is this getting boring, should I go back to moving more than -2 meters in QWOP now?
https://www.accountkiller.com/removal-requested
This might be true, technically speaking, if storing the information that results from each calculation required work to be done (for instance, you have to do work to change the magnetic moment of the grains in a hard drive platter). But in the CPU this information is just represented with a potential.
Yep. And in the server world there's also consolidation, you always want fewer and more powerful processors if the cost isn't too high. The typical CPU in desktops and laptops is a different matter though. Maybe the question is more obvious the other way around "Would you like a quieter computer with longer battery life?" and with other marginal benefits like lighter heat sink, lower power bill, probably longer life because of lower temperature and so on. For example many office workers would do exactly the same even if their laptop used half power.
To take another example, if my phone uses half the power I wouldn't make twice as many calls - I pretty much stay connected 24x7 and make all the calls I'd like anyway. Same with say my TV, I wouldn't watch twice as much TV if it used half power. Of course I could start using my phone to play Angry Birds on my way to work instead of reading a book - which may or may not be a power savings depending on how "green" making and distributing that book was - potentially I could now use more power because I have a more efficient CPU. And I could get myself a new and bigger TV that is much bigger and still fills the same power envelope. Except I'll probably have to take into consideration other things like cost, size, weight so maybe I won't.
On the whole though, I think energy efficiency works out as a good thing. Look at fridges, freezers, washing machines and so on. Sure, there's a certain effect that now that we have refrigeration, we'll use it a lot more and have bigger fridges. But in the end it's easier to just make them more efficient and hope the use tops out than to say "Please eat food that doesn't need cold storage and use your clothes until they're really dirty." I think it's same with computers, make them efficient until people say "Actually, this 5w computer is enough for me. Great that there are 100w+ CPUs for those that need them, but I'll rather take the other benefits."
Live today, because you never know what tomorrow brings
You're an idiot, Starscream. When we slip by their early warning systems in their own shuttle and destroy Autobot City, the Autobots will be vanquished forever.
You're an idiot if you think the "entropy" (the bits being 0/1) represents anywhere near the magnitude of the work done to flip those bits. Even if you DO think that, or there is some system where it's true, you can't use high bits as a heatsink (worksink) because the entire storage of the system (every single bit you care to count) only counts when being flipped. You can set all bits to 1 in a few clock cycles, and then everything else is 100% heat.
A high voltage (a bit that's on) is potential. There is no work done in keeping a bit on, just as there is no work done holding something above your head. It's the lifting that does the work. In this lifting, the "entropy" as you like to refer to it (the word is "potential", you fuckwit, please get a dictionary and learn basic physics) comes from the work, meaning the rest of the work is heat. You have reduced the amount of work that contributes to heat. But if these voltages drop down, guess what, they work against you.
We store energy in molten salt. This works because it takes forever to heat up the salt and it has a large heat capacity. By the time the heat capacity is full, the sun has gone down.
Flipping bits doesn't work this way because flipping bits is easy (bits have nearly 0 heat capacity) and because the entire system is only useful when it's powered (and heat is being pumped into it).
From the announcement, it looks like a product split. Long battery life for portables such as tablets and smart phones without using a really slow chip, and extreme computing for rendering 3 D movies, server farms, and supercomputing. This too will take advantage of the instructions per watt. Your data center can be half the size, use half the cooling, half the power and still run faster. Or as you pointed out, you could have double the data center in the same old room.
In short, for the same processing the power used is much less. How much power you need will scale. Now it will scale in a smaller space with less power.
The truth shall set you free!
In performance per watt, the efficiency is measured in how much computing you can do per watt of power consumed. A computer sitting idle all day is wasting power generating heat. A computer rendering the latest Pixar film may draw more power and generate more heat, but it is also providing the results of the computation.
http://en.wikipedia.org/wiki/Performance_per_watt
The truth shall set you free!
By having the gate on both big sides of the active region (conventional MOS has the gate on only one big side) the gate control is about twice as effective. The active region can be turned off better, and turned on harder. This means lower power dissipation, other things being equal.
Other things aren't completely equal. The larger gate area means more capacitance, which lowers speed and raises dissipation. That is why they're talking about 37% faster and 50% lower dissipation, instead of (magically) 100% faster and 75% lower dissipation.
Contribute to civilization: ari.aynrand.org/donate
The planar process was developed at Fairchild by Jean Hoerni. (wikipedia).
Contribute to civilization: ari.aynrand.org/donate
Although this provides a superior transistor, there is a drawback from the IC designer's standpoint, if the illustrations are accurate. The designer of an IC with planar FETs can control both the width and the length of the active region. With these FinFETs, only the length is under the designer's control. Width is fixed (it's the vertical dimension in the illustrations). The only way to get stronger drive is to parallel transistors, whereas in planar design the transistor is simply made wider.
Contribute to civilization: ari.aynrand.org/donate
It now makes much more sense that Apple is having Intel manufacture the next generation IOS chips. Based on information from Ars, the finFET shines at lower voltages. So Apple is going to be using Intel not just because they can manufacture smaller than everyone else, but also because the resulting chips will be faster or take less power than anyone else.
Faster *and* lower-powered, from *Intel* ?
Truly, the end is nigh.
What a depressingly stupid machine.
isn't this just a mulitgate finFet. That structure has been around for a long time.
don't cut it off www.mgmbill.org
Newsflash: AMD designs even faster, 4-D Transistor!
Are there articles from nine, four, three, two, and one years ago predicting this breakthrough? To someone not in the business, this seems to have come out of the blue--already in production by the time the news breaks. How long have they been working on this?
Their they're doing there hair.
Answering myself: The idea started in 2002.
Their they're doing there hair.
Intel's 3D Transistors One Step Closer to Reality Posted by ScuttleMonkey on 15:15 Monday 12 June 2006 from the closing-your-leaks dept. http://slashdot.org/story/06/06/12/224207/Intels-3D-Transistors-One-Step-Closer-to-Reality
Their they're doing there hair.
exactly!
"the processor is meant to change the entropy in the system"?
I'm not sure what the parent was trying to claim with this indecipherable statement.
"you only have to remove heat if you want the processor to remain at a constant temperature"
which is ... ALL THE TIME!
https://www.accountkiller.com/removal-requested