Slashdot Mirror
Intel's Roadmap Includes 4nm Fab in 2022
Precision submits "Intel Corp., the largest maker of chips in the world, has outlined plans to make chips using 4nm process technology in about thirteen years. According to Intel, integration capacity of chips will increase much higher compared to fabrication process."
The next step of the plan: negative-sized chips by 2050!
Funny may not give karma, but +5 Informative never made anyone snort coffee out their nose.
These are long-term business forecasts for 10+ years down the line. They are thought experiments only, in my opinion. They are still valuable, and something to consider, but still very much a "projection" and not a "concrete plan with funding".
Just because I can hook a shark from a boat, I do no offer to wrestle it in the water.
Wait, how is 4nm possible? Isn't this below the atomic radius of silicon?
This is obviously pie-in-the-sky speak from the marketing dweebs, who don't understand the physical limitations that come with a die shrink.
the atomic radius of silicon is about 110pm, or about 0.11nm, or a little less than 1/36th of 4nm.
Let's do away with the hundreds of pins or pads and have chips communicate through ultra-wideband low-power wireless links (with waveguides if necessary). That should drive costs for board, chips and tools down significantly and quite possibly also enable amateurs to once again build hardware with modern chips.
I have a similar problem. The tagging interface works perfectly for me in every way ... except that not a single tag I have ever set has ever once shown up on the main page. Apparently my tags are going straight to /dev/null after being accepted by a perfectly functional interface. Care to trade?
It is a miracle that curiosity survives formal education. - Einstein
You're hardly a "good poster". Your baseline post score is 1, for God's sake. Get back to me when you actually have some positive karma to deal with, troll-boy.
How much data can you transmit in a very localised (~10cm) environment wirelessly?
How about via a high-quality optical cable? Even if the length of that cable was restricted for optimal performance, i.e., connecting a CPU to another piece of logic?
"Intel Corp., the largest maker of chips in the world,
Is it really neccesary to explain who intel is on /.? I think even my parents know that intel makes chips, they put out enough commercials... Are even our taco overlords not really reading TFS before hitting that submit button?
http://greenobyl.com/ please.... think of the children!!
... besides wishful thinking?
Oh, and given at those dimensions quantum noise (e^KT/q) will be greater in signal strength than a 1 or 0 level I am interested to see just how this works.
I'd love to see it but for the moment it's just numbers on a slide. About a gazzilion dollars in research are needed to get to those dimensions.
"TV, a medium as it is neither rare nor well done." Ernie Kovacs
It just happens that my personal roadmap for 2022 includes a flying pony that craps gold. I'm cautiously optimistic.
Even accounting for the successful introduction of new materials for transistors, 12 years to get to 4nm seems a tad ambitious. Also, you have to wonder whether or not they're approaching the top of the S curve.
I scream. You scream. I assume that means we're both acquainted with the problem. We proceed.
It obvious that by then scientists will have found some of that string they've been theorising about for years and will be using that for interconnects.
Compared to what, 0?
a good deal of hope in this projection that someone will have broken timespace and licensed the procedure, silicon will stop being so uncooperative at the 4nm scale, or the higgs boson is getting discovered at the end of the month this year. this resounds with earlier city planning predictions around the start of the 20th century that people would fly through tubes both subterranean and through the air to their destinations. turns out that wont work.
Bill gates, a guy who arguably got rich by being ruthless and in the right place at the right time, is famous for his pot-shot predictions of the future as news media routinely mistakes him as a mystic sage, and computers in turn as a form of arcane magic. a better tile for this article would be "intel exec cracks skull on platinum yacht crapper during champagne island-wager, sees magic future in pool of head-blood."
a question id ask, before said exec bleeds out, is whether he thinks AMD is going to roll over and die around 2010, thereby paving the way for delusions of globe-dominating profit to be realized...and maybe while im at it id inquire if he knows how the word "nano" works outside of the ipod realm.
Good people go to bed earlier.
By 2022, the only integrated circuits you'll have will be the ones you carve yourself, with your bare teeth, out of the bones of your children(during those rare times that you aren't fighting off hordes of monstrous rat-men or scavenging for survival in a grim Malthusian dystopia).
Except for the fact that a lot of the 'marketing dweebs' at tech companies are engineers.
Just sayin'. Your product management/marketing folks at these firms are often very plugged in to the tech side of things (I should know, being one of them).
I get the same thing and I'm even unable to meta-moderate because of the same issue.
Slashdot, getting more broken every day.
I used to have that problem, now I don't even get the option of being disappointed as the tag disappears into tag-limbo. I think tagging is for personal use, but that common tags get elevated to everyone-sees them. But that doesn't explain why personal tags don't show up anymore.
I don't know, I literally cannot think like the Slashdot coders, as everything they do seems contrary to common sense.
Looking at current processors, never mind the future stuff, we have(in addition to power, which will definitely need pins, though only a few really chunky ones) a HyperTransport or QuickPath connection, both good for ~25gigabytes/s, plus however much bandwidth is needed to/from the system RAM.
I'm not an RF expert; but I'm going to hazard the guess that pushing 30+gigabytes a second over the air, even across short distances, is not something you do to make your life easy.
I would suspect that unforeseen developments, such as big advances in 3d circuit design, would alter this schedule a lot. This is simply daydreaming.
Forget about the limitations of die shrink, what about the limitations of quantum mechanics? I was under the impression that 4 nm is getting awefully close to the point where quantum tunneling makes tansistors unworkable. As in, when you detect a signal, you can't tell if it's there because it should be or because an electron just jumped the gap.
Um, what? They have it going as 1/(process size)^2, which makes perfect sense for same-sized chips. When you halve your process size, each transistor is half as big on a side, or a quarter of the area. So it seems perfectly reasonable to have four times as many. Did whoever wrote this actually stop to think?
They likely have more knowledge surrounding technology for a decade plus roadmap than you do.
“Common sense is not so common.” — Voltaire
Yeah, intel is a bag of shit. I mean, its not as though they're making billions upon billions of moneys - they hardly know what they're talking about. If I wanted to know what was possible in 2022, I'd come to /. to listen to some pathetic failure of a life who actually has the arrogance to believe he knows more than intel when it comes to fabrication
then speak what you know, son!
You humans. When're you gonna learn that size doesn't matter? Just 'cause something's important, doesn't mean it's not very, very small.
-- Frank the Pug
They program in perl. They are ABOVE common sense.
Extreme Programming - Redundant Array of Inexpensive Developers
The atomic radius is not the proper distance to consider. If you do so, you assume that atoms can touch each other, which is very far from the truth. The closest distance "allowed" is the first nearest-neighbor (NN), which is related to the crystal lattice constant (for Si: 0.543 nm), and the crystal structure (Si has a diamond structure). For Si that NN distance is 0.235 nm. This is all very much academic tough. Even if you could make a circuit that small, you would then have to wonder, left alone quantum-size effects, leakage, behavior under oxidation, etc.
It seems to me that rather than the identity and timeframe for the different technology nodes (which anyone who knows Moore's law could have given in advance) the interesting thing from that slide is what it says about delay scaling and energy scaling. Whenever you shrink your process you have a certain amount of gain that can go into either making the chip faster or making the chip more power efficient. For a long time back in the day people wanted to stay at 5 volts to preserve compatibility, so everyone just kept putting it into going faster. Nowadays chipmakers try to go for a more balanced strategy.
But here, on this chart, Intel is saying that they're going to a delay scaling of "~1", staying at pretty much the same speed. And they're looking to increase their energy scaling from "~.5" to ">.5". So it looks like we really have topped out in terms of GHz.
This sig wasn't worth reading, was it.
I don't really pay too much attention to the chip business, so I'm wondering how well, historically, Intel has followed their roadmaps? Are they like an actual roadmap of a, uh, road, that you can follow, or more like a "Roadmap to Peace" that's made because it looks good and people expect you to, even though everyone knows it's not going to work out?
Anyone got a roadmap from 1996 or so, so one can see how well it was followed?
As a 'marketing dweeb' at a big chip company I can tell you that most of us (me included) are former engineers who moved to marketing because we could make significantly more money there, have a nicer office, and generally a better working environment.
We set our own deadlines. If a product fails we decide why (guess what...it is never because it wasn't marketed well). We set our own hours. We travel when we want, where we want. Our co-workers are better dressed and better mannered and have much better hygiene. I can sit in meetings and daydream all day if I want, because my boss doesn't measure my output by piece work standards.
Of course not just any engineer can make the jump. You have to have social skills. Be good looking. Speak well. Not be shy. Have a sense of fashion. Be funny among normal people, not just at Gen-con.
I can see why Slashdotters are envious of marketing people.
...you can't tell if it's there because it should be or because an electron just jumped the gap.
More of this. We should make tunneling the nerd version of this or this
Interesting.
I'm not sure about the common tags thing. For example, First-European-Provider-To-Break-Net-Neutrality is tagged with 'okwellusehttpthen', which I can't believe is that common.
Here's a set of roadmaps generated at three-year intervals. Note that, with the exception of RAM density, each of the charted criteria outran the roadmaps' predictions.
These roadmaps are generated by a consortium of companies. They're routinely betting the future of their entire industry on these roadmaps. They're actually pretty darned conservative.
It's about the size of the channels that gate the flow of electricity across nerve membranes.
Or so you think! ^^
After all there's a reason you're not actually working in enginerring, when you're such a great engineer...
Any sufficiently advanced intelligence is indistinguishable from stupidity.
Assuming that earth is all carbon (and my calculations are correct), earth is built of approximatly 3.6*10^31 atoms.
According to that table (if extended), there would be a processor with more transansistors on a single chip than atoms in earth in 2152.
Blah blah blah. Why is this news? Remember roadmaps from 2002? Do we even care about those now? We won't about this one in the future.
of course intel showed "plans" for this. they have investors who don't understand the limits of miniaturization to snow.
That's great. Planning for the future must truly be what separates man from beast. I do the same thing. Here's my personal roadmap:
2010) - Get in shape, including 6-pack, benchpressing twice my weight and being able to do a Triathlon in Olympic-qualifying time.
2011) - Win Powerball. Quit job
2012) - Use lottery winnings to build self-sufficient compound to survive Mayan apocalypse.
2013) - Now that I'm the only one in the world with means of survival, all the girls will like me. Procreate wildly to start new human race.
"After all there's a reason you're not actually working in enginerring, when you're such a great engineer..."
Yeah - the pay is better.
"As God is my witness, I thought turkeys could fly." A. Carlson
Funny is how everything changes after 2012, they will have a different type of transistors. Maybe the guy really thinks things won't matter after 2012 - nut-case.
Just in case, I ask you to hold them to their other words too:
http://www.design-reuse.com/news/4850/intel-building-blocks-10-ghz-processors.html
Next year we are going to see 10GHz processors, this is going to be an interesting exercise.
Maybe Tom's Hardware or some other brave soul will manage.
Je me souviens.
Firethorn,
I just wanted you to know that I really appreciated the site you link to in your sig, a-human-right.com. It's rare that I see such good accurate information about the gun-control issue and rarer still that I see such a strong love of freedom. Thank you for sharing this. It's an excellent reference, especially for people who are undecided about gun-control or who don't understand the full political implications of it.
It is a miracle that curiosity survives formal education. - Einstein
There have been formal semiconductor roadmaps to the future since 1992. There's an consensus roadmap updated annually by an industry group.
This isn't a blue-sky thing. It tells all the players what they need to do to keep up their part of the technology. The fab-equipment people, the device physics people, the etching people, the mask people, the substrate people, the design tools people, etc. all have to push their parts forward. The roadmap tells them how far each piece has to be pushed.
These roadmaps are available for past years, and you can see how the industry has tracked the roadmap. It's reasonably close for any five year period. The big change in the last decade is that heat dissipation is starting to dominate the problem. The roadmap now focuses on memory devices, which have low activity per cell compared to compute elements and aren't yet power-limited.
The current consensus is that the improvements to known technology can get down to 22nm, and then it gets hard. The roadmap assumes CMOS transistors; other devices are discussed, but aren't factored into the mainline predictions.
Last time I checked, tansitors, resitors, condenators and tutti quanti are all pretty unworkable already, even without the limitations of quantum mechanics.
I'm not a coward by any name.
First, I work in the same office with marketing people with similar perks to what you mentioned... I'd never want to trade places with them, their work includes way too much ass kissing and bending over.
Second: "nice working environment", "setting your own hours", "setting your own deadlines", "output not measured by piece work standards" -- maybe you couldn't achieve these as an engineer, but I can tell you it's possible. Now, I know I'd get a better pay as a marketing guy but it's not even nearly enough to cover the lost freedom and control.
A lot of the troubles that look like fundamental roadblocks (like e^KT/q) become less of an issue at low temperature: quantum tunneling, resistivity, and smallest noticeable voltage change to name a few.
Let me speculate: say we lived in an era where you could run a medium-thick client with hardware like what we have today, but have a fast Internet link to a datacenter with 4 nm chips designed to work at 20 K or cooler. These chips could use much lower voltages and currents, and could have fewer tunneling problems than room temperature computers. Even with the cooling needed, performance/Watt would be a lot higher.
You might not be a fan of "utility" computing, but if you could have a 200 GHz (or so) computer you could get to via NX in some liquid Helium-cooled facility rather than having a poky ~5 GHz machine pushing up against room temperature limits, wouldn't that be tempting?
Expected time to finish is 1 hour and 60 minutes.
most of us (me included) are former engineers who moved to marketing because we could make significantly more money there, have a nicer office,
Not at Intel (the subject of this article). Everyone there gets a cubicle, even the CEO. The executives have slightly nicer and larger cubicles, but that's it. Marketing peons like yourself would get the same 9x9 cubicle the engineers get.
In other words: "we can imagine much more than we can actually produce in this physical reality".
-- Sig down
When you think about all the interference problems, a wireless connection (even a very short-range one, like 1 cm) is just dumb.
When we get to the point that electrical signal interconnects between chips are a problem, the most likely next step is optical interconnects.
Amateurs can already build hardware with modern chips, but there's rarely a good reason for an amateur to work with BGA chips. When working with those, it usually makes a lot more sense to just buy ready-made boards with those chips. For electronic work that doesn't require things like Core2 CPUs, most things are available in standard DIP, SOJ, TQFP etc. chips, all of which can be soldered by hand.
Typically, amateurs simply don't work with anything involving BGAs (where a different package isn't available), unless they're using them in an embedded project and most of their work is on the software. In that case, they buy pre-made hardware, and concentrate on software. Amateurs doing more hardware-type work, or working with microcontrollers, don't have any problems because everything there is available in non-BGA packages.
Yikes!
Would a real engineer trade the joy of building useful things for the benefits you mention? Some of the qualities you mention are desirable for any job and everyone should strive for (e.g. good social skills) but most of your post came off sounding very vain to me. I don't care how much I could make in marketing, I'm not going to switch.
-IOVAR Web Dev Platform
This looks more like a scheme to get some business advantages rather than a traditional roadmap. Hopefully alternative computing models like Photonic computers, Quantum computers and DNA computers will be mature enough to come to normal user before Intel needs to reach 1-figure fab process. It'll be surprising if Intel doesn't get heavily involved in R&D of those technologies in the next 4+ years.
The above post most likely written by a 15 year old hiding in their mom's basement. But you never know, it could be Steve Job's evil twin.
Faster! Faster! Faster would be better!
Pfft...Bitboys had them beat 9 years ago.
Actually not, scrub.
This isn't Intel's first outlandish prediction. In 2000, they predicted they would make a 11Ghz processor by 2011. Instead, they ran into problems reaching beyond 4Ghz, and instead went towards multi-core processors. I have a feeling the same people who made that prediction are behind this one. Whether or not they can accomplish this is to be seen, but it seems quite unlikely.
Having at one time worked for Intel, I can tell you firsthand that the marketing people there are not engineers and are not, to my considerable surprise, any more technically knowledgeable than the marketing people for the average department store chain. But then again, maybe that's not so surprising. After all, they're not marketing to people like us; they're marketing to OEMs. In particular, they're marketing to other marketing people at OEMs. The OEM engineers are involved, of course, but they're not calling the shots.
Doubt me? Have you seen the crap that comes from the major OEMs? ;)
Those who can do. Those who can't, teach. those who can't teach, go into management/marketing.
2022 is when the Qeng Ho are due to release the 4nm technology to us...
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
After all there's a reason you're not actually working in enginerring, when you're such a great engineer...
Or because in addition to understanding the technical details of the work, some people are also skilled at communicating the relevant aspects in an accessible manner. Not all engineers stay in "enginerring" - some, like Milton, have people skills!
Seriously though, as a research engineer, I'm quite glad that my thesis/dissertation advisor emphasized development of writing and presenting skills in addition to expecting good technical analysis. It's served me well, even as someone who's stayed as far towards the technical side of things as possible. On the other hand, plowing through the technical papers that some engineers and scientists publish, the language skills are simply terrible. When someone has the background and intelligence to understand the material, and is capable of writing it so as to have some effect other than inducing a coma in the readers, that's a valuable thing indeed! If some of them choose to work in sales and marketing, that's a plus for all of us. We can get more accurate information than a business or English major would likely be able to give us, but written so that lay people can understand the gist of it.
When a small percentage of engineers are good writers, and there are a great many engineers available, it makes sense that those few who can write well would do that while those who can't fill other jobs. Just because someone can do one job well doesn't mean he shouldn't do another job that he may enjoy more, and which many of his peers couldn't do as well as he can.
And you may even get a real office rather than being herded into a cubicle/stall as if you were livestock.
Fatal error:
MySQL: Can't create a new thread (errno 67); if you are not out of available memory, you can consult the manual for a possible OS-dependent bug
New Economic Perspectives
Sure, but if the electron randomly jumps the gates 40% of the time, you just have to wait for 40 electrons to pass by and there's no problem at all.
When quantum mechanics dominate, we just have to take this into account, it doesn't mean that devices are impossible to design, and people have been studying the effects of quantum mechanics rather thoughroly for decades now.
Which somehow is very wrong. I mean who does the more complex job? Certainly not the manager.
But hey, I'm still happy for you. :)
Any sufficiently advanced intelligence is indistinguishable from stupidity.
I agree. I don't see what RF wireless would solve. It would probably consume more power, have less bandwidth, be more vulnerable to interference, etc. Optical interconnects, on the other hand, offer potentially immense amounts of bandwidth, produce no heat (outside of the light emitters), allow a longer distance between chips, are invulnerable to rf interference, etc. Last I heard, Intel is actually pouring money into researching this kind of technology, trying to find ways to build laser diodes into the fabric of chips.
But wall-mounted water bottle and the food pellets are great!
Interesting.
Complexity:
1) Applying the laws of nature and mathematics (which don't change) to materials (which don't change) to get an outcome (which is generally predictable.
2) Applying various techniques of persuasion and coercion (which may or may not work) to people (who are all different, from day to day) to get an outcome (which, if you are lucky, might resemble what you wanted to begin with).
Hmmmmmm...
(Oh, and I'm a project manager - I have no direct reports, but that doesn't mean I envy my boss)
"As God is my witness, I thought turkeys could fly." A. Carlson
They (Intel) did say that the process variability on the nodes starting at 16 nm will probably be extreme. So getting the necessary chip bins for marketing purposes could be problematic, at least.
Assuming that you mean a 14mm x 14mm chip, divide 14mm by 4nm to get 3.5x10^6. Then square that to get 1.2x10^{13} 4nm x 4 nm cores in a 14mm x 14mm chip.
I'm so sure Intel pays them to sit around making bullshit claims that defy the laws of physics. I love how a bunch of Slashdot members think they know more about chip manufacturing than one of the world's largest and most successful chip manufacturers. I'm sure Intel execs are busy combing this forum for ideas on how to do things.
Intel marketing is amazingly ignorant, in my experience.
All Intel workers (including CEO) work in cubicles.
So, apparently that big chip company don't do integrated marketing, don't do analysis and don't manage their marketing costs. A real recipe for success!
Hardware about beautifull and high-quality
cellphone, TV Mobile Phones,Mobile Phone Watch, Dual SIM Dual Standby phone, PDA Smart Phones, Double SIMs Phones,Dual Band Cellphones
Actually, quantum tunnelling starts at 10 nm, so I don't see how it could be possible with FinFET tech...
In 2004 or so Samsung announced their FinFET design and claimed that they were going to incorporate their FinFET design into their DRAM chips
It's 2009 now, and Samsung has yet to incorporate FinFET into any of their design
Hmmm........
Muchas Gracias, Señor Edward Snowden !
Tunneling isn't completely random. Hell, without it, modern flash memory wouldn't be possible. The trick is to stop thinking in terms of absolutes. Modern EDA tools at 45nm and below already treat gate-delay as a probability function rather than absolute min/max. A chain of gates would produce a probability distribution and you'd simply design it to be ~98% inside your timing envelope.