Slashdot Mirror
Intel Moving Forward With 10nm, Will Switch Away From Silicon For 7nm
An anonymous reader writes: Intel has begun talking about its plans for future CPU architectures. The company is already working on a 10nm manufacturing process, and expects the first such chips to be ready by early 2017. Beyond that, things are getting difficult. Intel says it will need to move away from silicon when it develops a 7nm process. "The most likely replacement for silicon is a III-V semiconductor such as indium gallium arsenide (InGaAs), though Intel hasn't provided any specific details yet." Even the current 14nm chips they're making ran into unexpected difficulties. "While Intel didn't provide any specifics, we strongly suspect that we're looking at the arrival of transistors based on III-V semiconductors. III-V semiconductors have higher electron mobility than silicon, which means that they can be fashioned into smaller and faster (as in higher switching speed) transistors."
Amazing that we're getting to 7nm, and rather than saying we can't do it, there's just casual talk about how they will have to switch away from silicone. Really incredible. Will they just keep marching forward to less than 7nm and into other exotic configs?
My God can beat up your God. Just kidding...don't take offense. I know there's no God.
Nope. They've decided to hit 7nm and then call it a day.
Help build the anti-software-patent wiki
Always has been, always will be.
GaAs was the future of super-fast transistors. The Cray 3 was made from GaAs.
GaAs has a much higher electron mobility than silicon, 8,5000 versus about 1,500 for silicon. This allows for much faster switching. InGaAs has an electron mobility of 10,000 allowing even faster switching.
But that's just electrons which are used in P channel MOSFETs. For CMOS, you also need N channel MOSFETS. The kicker is that GaAs and InGaAs have respectively lower and much lower hole mobility so the N channel FETs switch rather slower than silicon.
CMOS is by far the only architecture. Historically it is the most power efficient since it only spends energy switching. On high speed, small scale CMOS, however, lots of power goes into the switching itself, the switching is fast enough that the devices don't really act very ideally and there's a lot of leakage.
Perhaps at very extreme ends, other architectures can compete, power wise.
SJW n. One who posts facts.
Moore's Law had a good run, but she's dead Jim.
It doesn't look that dead just yet. While that graph shows a straight diagonal line of transistor count over time, there should also be a flat line alongside showing the number of people who predict that Moore's Law is dead.
Maybe they can partner with Apple and make a really skinny macbook.
Why would they need to partner with Apple when they can just shrink their own competing Ultrabook spec? They own the trademark to it after all.
Welcome InGaAs Valley
InGaAs Valley has a nice ring to it too.
Moore's Law had a good run, but she's dead Jim. Two, maybe 3 shrinks at most, and you're at the end of getting benefit from feature size.
Moore's law is really all about "cost" per transistor. While process shrinks are certainly an important enabler they don't have to be the only driver that keeps things going.
I'm surprised Moore's Law lasted this long. Other bottlenecks seem to be more of a factor of late such that I thought CPU's would take a bit of a rest due to diminishing practical returns, analogous to a Ferrari stuck in traffic.
Table-ized A.I.
So they can grow to Borg-ship size to keep Moores going without shrinking transistors?......oh oh
Table-ized A.I.
will involve making chips taller, ie various forms of 3D ICs. That would mean that we could continue to get the apparent effects of higher densities at least for a while, though we'd really just be making taller or chips or better interconnected layers, but it would also mean that the cost of transistors wouldn't go down, it would probably go up.
It's definitely slowing down, Westmere EX was 2.6B in early 2011, Haswell EP 5.69B in late 2014 so roughly 42 months to double (Haswell die is ~20% bigger accounting for the 220% count instead of 200%) . A large part of that slowdown though might be economics, Westmere was surely started before the financial crisis and Haswell likely during or after so Intel might have slowed development (especially since on these large parts they don't have any meaningful competition except at the very high end from IBM and Oracle)
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
With Si, it's easy to create a great gate insulator, just oxidize the silicon,
and you're done. With GaAs/InGaAs, there's still research going on how
to create a good gate inulator. A paper from 2012
(http://iopscience.iop.org/0268-1242/27/11/115002) proposes to use Al oxide,
which complicates things by adding another material and and an extra process
step to the manufacturing.
On a side note: in the paper, the GaAs gate length was 1.5 um. A bit more than 5 nm.
Seems like you took too long to type yipee there. Better luck next time. Try a few e's less maybe?
The dangers of excessive individualism are nothing compared to the oppressiveness of excessive collectivism
Doubtful, Ga isn't that rare, we mine ~254t per year mostly as a byproduct of Al smelting, this is fairly small compared to ~54,000t for Si use in semiconductors, but is quite high given the fairly small market for it today. To give you an idea Lithium is slightly less common in the crust but annual production is ~30,000t.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
The "rare" in Rare Earth Metals/Minerals says nothing about actual rarity. It's only a statement on whether they can be found in concentrated ores or not.
it's in my head
Silver is 3x more rare and is mined at ~18,000t per year, so again you can reasonably expect ~60kt per year if prices support the effort (though that's a bit misleading since elemental silver veins happen in numerous places but In has not been found in similar streaks)
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
Diamond is the best semiconductor by quite a margin. This link: http://www.evincetechnology.com/whydiamond.html tells part of the story. Some diamond devices have run at 700 degC.
In the mean time, SiC is the next best thing.
> III-V semiconductor such as indium gallium arsenide (InGaAs
I think the french will like it and possibly the swedes. They use Gallium and Indium based semiconductors in airborne electronic warfare systems, which allows for very high RF energy output in physically very small and high temperature tolerant packages. (For example used in the Dassault Rafale and SAAB Gripen fighter jets). The french SPECTRE jamming suite is especially famous: the Rafale plane is not stealthy, only has reduced radar reflection, but the french trusted their system enough so their pilots were already flying deep in lybian airspace by the time the US Navy started to launch Tomahawk cruise missiles at Gaddhafi. Supposedly there is something equal or better in the american F-35 JSF, but that airframe is so buggy one must wonder if it will ever enter service?
On the other hand non-silicon semiconductors, like Ga and IN tend to cost twice the price of pure gold per weight or more. At the most extreme end, the soviet-russians even created diamond-based semiconductors, for use in space weapons and a planned Venus robotic rover. They invented a diamond crystal growing machine for the purpose, which after the Cold War was sold to a US company, which nowadays grows and sells multiple carat "cultured" yellow diamonds for ladyfolk decoration purposes. Beware, that femme fatale may wear a supercomputer on her finger! Now you know why multiple-finger gesture support was developed by Synaptics...
That is actually not correct. ... many of them are actually absolutely not rare.
The comes from the fact that they where considered rare when they where discovered, the whole third group and the Lanthanoids are considered 'rare earth metals'
Their oxydes are rare ores, perhaps you meant that. On the other hand 'deposites' of thise minerals are rare, too. But they are usually mined in quantities together with other ores, the primary ore of the deposite in question.
See e.g. http://en.wikipedia.org/wiki/L....
Cost free eBook I read (by iBook/Kobo/Amazon/ObookO/Gutenberg etc.): "The Green Odyssey" by Philip Jose Farmer.
The prices in my condo development in Indium Gallium Arsenide Valley is going to explode!
Despite their name, rare earth elements (with the exception of the radioactive promethium) are relatively plentiful in Earth's crust, with cerium being the 25th most abundant element at 68 parts per million (similar to copper). However, because of their geochemical properties, rare earth elements are typically dispersed and not often found concentrated as rare earth minerals in economically exploitable ore deposits.[3] It was the very scarcity of these minerals (previously called "earths") that led to the term "rare earth".
http://en.wikipedia.org/wiki/R...
it's in my head
The ingredients are definitely nasty, so there's concern for industrial waste and exposure. However, the finished material has proven to be relatively harmless in animal studies. I was surprised to learn this, but that seems to be the conclusion, so there should be no immediate risk for using the end products.
I'm not sure about the stability of the compounds or how they degrade over time.
Chips that run hotter also have more thermal gradient, which can put mechanical stress on the various delicate layers of the chip. Being able to run hotter means you can support more of a thermal gradient to ambient, and thus support more heat flow and thus more computations/sec. However, at some point you're going to cause mechanical failure of the chip, especially if the stresses cycle.
So not only termperature tolerance, but also coefficient of thermal expansion and strength of all the various materials is going to count when it comes to longevity.
--PM
Except Moore's law is about transistors per square inch, nothing about cost. So if process shrinks don't occur Moore's law fails.
Wrong. If you stack dies, then you increase transistors per square inch.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
The cost of the raw materials is completely dwarfed by the cost of processing. Even a very large chip (2 cm x 2cm by .5mm thick) masses less than a gram. It's also likely that these high-performance III-V chips will be built on a cheaper substrate, meaning the thickness of the expensive stuff will be much, much smaller.
No, the gp post is right. Moore's law can't break physical laws.
10 nm means the pathways are about 40 silicon atoms wide. 7 nm is 30 silicon atoms wide, but they're planning to move to GaAs or another III IV semiconductor, and those atoms are larger than Si, so even fewer atoms across at that width. Another shrink to 5 nm is about 20 wide.
I don't think we'll go much smaller than that. The smaller you go, the more quantum effects interfere with the electrical properties of the materials. Also, heat means movement, and those chips get really hot. Go too small, and with enough heat, atoms will move out of alignment.
No worries, though. Chips are presently mostly 2D which means a lot of space is taken up by connections between components - like power and clock pulses. 3D opens a doorway for alternative smaller structures and better cooling techniques... maybe liquid cooling between chip components on nano pipes.
No, that is exactly the one thing that Moore's law isn't about.
Moore's law is about the number of transistors you can put onto an IC to achieve the minimal cost per transistor.
If you look at section 4 of Moore's paper he says when you combine the overhead cost of making an IC, the marginal cost of a transistor, and yeild you can calculate a minimum per-transistor cost for each generation of IC. He then provides some examples from 1962, 1965, and 1970, and uses them to hypothesize the number of features on a 1975 IC. If you can develop a technology to reduce the cost of putting a larger number of transistors into an IC then you can extend Moore's law without process shrinks.
That is what I said, but thank you for repeating it :)
Cost free eBook I read (by iBook/Kobo/Amazon/ObookO/Gutenberg etc.): "The Green Odyssey" by Philip Jose Farmer.
Phenomenal cosmic power!!! Itty bitty living space. Nothing like borking amazing hardware with a crappy, virus laden OS.
It's a bit more complicated that that. Even if an element is somewhat abundant but evenly distributed in the earth's crust, then it's difficult to mine. It's only practical to mine something if it's concentrated in some areas. E.g. gold is rare but you can find it in macroscopic flecks or clumps that are concentrated in certain areas. If gold were not concentrated like that but was instead uniformly distributed in the crust, there'd be no economical way to mine it.
That said, it looks like indium is concentrated somewhere: in zinc ores. So large scale production may be possible.
According to Wikipedia 0.075 ton/year is produced of monocrystalline silicon for use in integrated circuits.
That can't possibly be accurate. Here's a paper reporting that total consumption of fully-refined silicon for chip manufacture in 1988 was 750 metric tons. I don't think increasing process efficiencies would have reduced that figure by four orders of magnitude since then...
It's about transistor count.
Go fucking look it the fuck up.
I remember reading about this back in the early 1990's... from what I recall of the article, it wasn't wholy practical at the time owing to the expense of fabrication compared to silicon with the technologies available, but the article writer did talk about the far faster switching speeds than what silicon can achieve... more than an order of magnitude, iirc.
File under 'M' for 'Manic ranting'
If your clock speed is a 4GHz square wave, it's physically impossible to send a signal somewhere on the up transition and have come back before the next up transition if it's more than 32mm away. That's based on the speed of light in a vacuum. Electromagnetic waves travel much slower in doped silicon.
The bigger the die, the harder it is so make a fast chip.
The wheel gets reinvented. Again. The Cray-3 used Gallium Arsenide and 3D chip packaging i.e. chip stacking. You can see 3D chip stacking in use today in things like smartphones. The thing is the Cray-3 is from 1991.
At 7nm, we'll have computers easily 10x faster than today's 16nm fab we're shifting to.
16nm to 7nm we're halving each dimension so 2x2x2 increase in number of transistors in the same space, but going smaller transistors can decrease voltage and increase frequency, so 10x speedup easy.
Hopefully we'll have 8 core CPUs with 4GB of on CPU memory. Having CPU/GPU/RAM/pretty much the whole computer on the main chip = lower memory access timing plus other advantages.
Use that advantage wisely you lazy programmers, cause its your last opportunity to be lazy.
Comment removed based on user account deletion
Comment removed based on user account deletion
Comment removed based on user account deletion