Can Our Computers Continue To Get Smaller and More Powerful?

aarondubrow (1866212) writes In a [note, paywalled] review article in this week's issue of the journal Nature (described in a National Science Foundation press release), Igor Markov of the University of Michigan/Google reviews limiting factors in the development of computing systems to help determine what is achievable, in principle and in practice, using today's and emerging technologies. "Understanding these important limits," says Markov, "will help us to bet on the right new techniques and technologies." Ars Technica does a great job of expanding on the various limitations that Markov describes, and the ways in which engineering can push back against them.

Comments

Obvious

[Russ1642]

Yes. Next question please.

Re:Obvious

[bobbied]

Actually, the answer is no and that is obvious. Eventually we are going to run into limits driven by the size of atoms (and are in fact already there).

Once you get a logic gate under a few atoms wide, there is no more room to make things smaller. No more room to make them work on less power. We will have reached the physical limits, at least in the realm of our current lithographic doping processes. We are just about there.

This is not to say there won't be continued advances. They are going to get more and more stuff onto each die for quite some time and manufacturing costs will continue to decline as yields go up. It's just that we are about at the limits of lowering the power consumption of the CPU and chipsets.

Re:Obvious

[Shortguy881]

Lol, they meant chip size getting smaller not the human interface.

Re:Obvious

[ShanghaiBill]

Did our jets get faster and lighter and cheaper?

The fastest air breathing aircraft was the SR-71, which went into production in 1962, based on technology from the 1950s. So for at least half a century, jets did not get faster. Aircraft improved enormously between 1903 and 1960. Then the rate of improvements fell off a cliff. That is why Sci-Fi from that era often extrapolated the improvements into flying cars, and fast space travel, but far fewer predicted things like the Internet or Wikipedia.

What's after atoms?

Silicon lithography will hit its limits after a few more iterations. But nano-assembly techniques may allow silicon transistors to be even smaller. After that we may be able to move to carbon nanotube transistors, based on spintronics to lower the heat dissipation. There is still plenty of room at the bottom.

Re:Obvious

[bobbied]

If you read my comment.... I'm saying that we are very close to hitting the physical limits. In the past, the limits where set by the manufacturing process, but now we are becoming limited by the material, the size of the of silicon atoms.

There is basically only one way to reduce the current/power consumption of a device, make it smaller. A smaller logic gate takes less energy to switch states. We are rapidly approaching the size limits of the actual logic gates and are now doing gates measured in hundreds of atoms wide. You are not going to get that much smaller than a few hundred atoms wide. Which means the primary means of reducing power consumption is reaching it's physical limits. Producing gates that small also requires some seriously exacting lithography and doping processes, and we are just coming up the yield curve on some of these, so there is improvement still to come, but we are *almost* there now.

There are still possible power reducing technologies which remain to be fully developed, but they are theoretically not going to get us all that much more, or we'd have already been pushing them harder. So basic silicon technology is going to hit the physical limits of the material pretty soon.

Re:Obvious

[Beck_Neard]

We're eventually going to hit limits, but there's no reason to think that that limit is a logic gate a few atoms wide. There's isentropic computing, spintronics, neuromorphic computing, and further down the road, stuff like quantum computing.

Re:Obvious

[jcochran]

I believe that we can get things smaller. I'll agree that we're approaching the limits as regards what is basically a 2 dimensional layout that we're currently using for chips, but that leaves the 3rd dimension. Of course there is a lot of technical issues to overcome, but I believe that they will be overcome.

Re:Obvious

[GrahamCox]

Then the rate of improvements fell off a cliff

That's only true if you're only judging it by outright speed, height, etc. Things have continued to improve in terms of efficiency, thrust-to-weight ratio, noise, cleanliness of fuel burn and above all, reliability.

The original RB211 turbofan (the first big fanjet of the type that all modern airliners use) had a total lifetime of 1,000 hours. Nowadays it's >33,000 hours. That's an incredible achievement. In 1970, as a young kid with a keen interest in aviation, I would watch Boeing 707s fly in and out of my local airport, all trailing plumes of black smoke, all whining loudly (and deafeningly, on take-off), and understanding where all the noise protesters that frequently appeared on the news were coming from. Nowadays you don't have that, because noise is just not the problem it was, there's no black smoke, and jets slip in and out of airports really very quietly, when you consider how much power they are producing (which in turn helps them climb away more quickly).

As far as computing is concerned, you're right - there's still plenty of room at the bottom. But the current fabrication technology is reaching its limits. Perhaps jet engine manufacturers in the late 60s couldn't see how they would overcome fundamental limits in materials technology to produce the jets we have today, but they did.

Re:Obvious

[dnavid]

Silicon lithography will hit its limits after a few more iterations. But nano-assembly techniques may allow silicon transistors to be even smaller. After that we may be able to move to carbon nanotube transistors, based on spintronics to lower the heat dissipation. There is still plenty of room at the bottom.

The point of the article and the article it references is that its easy to say stuff like that, but also mostly irrelevant to practical computing because in the history of modern computing its never been absolute physical limits that caused major changes to how computing is implemented. Just because there's room at the bottom, doesn't mean its room we can use. We *may* be able to use nano-assemblers for silicon and *may* be able to use carbon nanotube transistors, but unless that gets translated to someone working on actual practical implementations of those technologies, they will apply as much to the average consumer as the SR-71 that's being discussed in this thread means to the average commercial air traveler. In other words, exactly zero.

When I was in college people were already talking about the exotic technologies we would have to migrate to in order to achieve better performance, and that was the late eighties. In the twenty-plus years since then, we're still basically using silicon CMOS. Granted the fabrication technologies and gate technologies have radically improved, but the fundamental manufacturing technology is still the same. Its been the same because there's hundreds of billion dollars of cumulative technological infrastructure and innovation behind silicon lithography. For these other "room at the bottom" technologies to be meaningful, and not just SR-71s, they need to be able to reach the same point silicon lithography with its multi-decade head start and approaching trillion dollar learning curve. Its not enough to just work in theory, or even in practice one-off. If it can't work at the scale and scope of silicon lithography, its just an SR-71. A cool museum piece of advanced technology almost no one will ever see, touch, use, or directly benefit from.

It isn't trivially obvious there exists a technology commercializable in the next few decades that can replace silicon lithography. Anyone who thinks that's obvious doesn't understand the practical realities of scaling these technologies.

Re:Obvious

[Ottibus]

I wouldn't be surprised to see RAM chips with a part of the die dedicated to CPU/FPU/GPU functions.

The same package is already commonplace, but the same die is problematic because RAM processes are significantly different from CPU processes.

Eventually the concept of a "central" processing unit may give way to passive backplanes and various speed buses, perhaps with a relatively lightweight chip directing everything.

This is a very bad idea. Moving bits uses orders of magnitude more energy than computation, so you need to concentrate the computing behind multiple caches and move the data as little as possible. So the model will continue to be based around islands of high-performance computing connected by slow, expensive busses, but the "CPU" will contain many smaller parallel processors.

Another example, is the x86 architecture. Intel has been amazing in keeping it going, but eventually, moving to something like Itanium with 128+ registers for integer, 128+ for floating point, etc. might be how Moore's "law" keeps going.

More registers means more area and more power for little benefit (though that is not the reason that Itanium failed).

As for x86, it was displaced by ARM a long time ago as the most popular 32-bit architecture.

Battery, Screen, Body

[mythosaz]

Even if the electronics fail to get much smaller, there's plenty of room to be had in batteries, screens, and the physical casings of our handheld devices.

They're pretty small now. Efficiency will improve

[Anonymous Coward]

We're running up against physical limitations but "3d" possibilities will take our 2d processes and literally add computing volume in a new dimension.

So of course it's going to continue, the only question is one of rate divided by cost/benefit.

Bettridge vs Moore in the battle of the laws

[raymorris]

Bettridge's law says no.
Moore's law says yes.

In the battle of the eponymous laws, which law rules supreme? Find out in this week's epoch TFA.

Re:Bettridge vs Moore in the battle of the laws

[riverat1]

In the battle of the eponymous laws, which law rules supreme?

Murphy's Law.

yes. Especially per passenger.

[raymorris]

> Did our jets get faster and lighter and cheaper?

Yes. Especially lighter and cheaper PER PASSENGER, which is the goal for passenger jets.

> it still takes the same amount of energy to fly across the Atlantic.

Nope, fuel efficiency and energy efficiency have improved significantly.

Re:yes. Especially per passenger.

[Kjella]

You're being very dishonest when you leave this part out:

Not by the same degree as computing.

By those standards, airplanes have basically stood still for the last 50 years. Sure they get a bit lighter, a bit better engines, a bit better aerodynamics but they're not radically different nor faster. Already the very first commercial transatlantic flight Berlin-New York was done in 25 hours, like orders faster than a boat and still on the same order - 8.5 hours - today. Same with cars, they've come a long way since the T-Ford but it could do 40-45 mph with 13-21 MPG. What would you get today, 35 MPG? You don't drive cross country on a thimble, that's for sure.

We're not talking about that kind of improvements when it comes to computers. We're talking about that 30 years ago memory was measured in kilobytes, today it's in gigabytes. If computers double performance in 10 years, we think that's awfully slow progress. 30+ years for a 10x improvement? 100 years until a terabyte is last century's gigabyte? Let's be honest, the kind of marginal - or rather, normal - improvements you're talking about would only be scoffed at. When - not if - we hit that limit that the walls are so thin they can't be thinner they might be roughly as good as they'll ever get.

Re:performance never measured in MHz

[vux984]

three decades in the industry and I've never seen performance measured or stated in MHz

Erm... from the 80286 through the Pentium 3 CPU clockspeed was pretty much THE proxy stat for "PC performance".

Our own computers? In the FUTURE?

[uCallHimDrJ0NES]

Next you'll be telling me they'll let us run unsigned code on processors capable of doing so. You need to get onboard, citizens. All fast processing is to occur in monitored silos. Slow processing can be delegated to the personal level, but only with crippled processors that cannot run code that hasn't yet been registered with the authorities and digitally signed. You kids ask the wrong questions. Ungood.

Re:performance never measured in MHz

[Misagon]

I can't tell if you are being sarcastic or not...

What you say is true only if you bought all your processors from Intel.

Once AMD came along, it was not entirely true if you compared to them. It was not true if you compared to Mac that used 680x0 and later PowerPC.

Re:performance never measured in MHz

[vux984]

Marketing and sales to ignorant consumers don't count.

Originally it was useful enough. Marketing and sales perpetrated it long after it wasn't anymore.

The "MHz Myth" has been time and again a subject in many a PC magazines

Only once the truth had become myth. The Mhz "myth" only existed because it was sufficiently useful and accurate to compare intel CPUs by MHz within a generation and even within limits from generation to generation for some 8 generations.

It wasn't really until Pentium 4 that MHz lost its usefulness. The Pentium 4 clocked at 1.4GHz was only about as fast as a P3 1000 or something; and AMD's Athlon XP series came out and for the first time in a decade MHz was next to useless. Prior to that, however, it was a very useful proxy for performance.

More meaningful benchmarks have existed long before that era (e.g. Whetstone from early 70s) and many were (e.g. Dhrystone in mid 80s) used all through the rise of the microprocessor (8080, 6502, etc.)

Sure they did. But for about decade or so, if you wanted a PC, CPU + MHz was nearly all you really needed to know.

Obligatory: "There's Plenty of Room at the Bottom"

[noidentity]

Feynman's talk on this seems required reading: There's plenty of room at the bottom. None of the linked articles even mention Feynman's name.

Remove the Bloat

[Hamsterdan]

As we're nearing the size limit for IC manufacturing technology, what about reducing bloat and coding in a more efficient manner.

Let's look at the specs of earlier machines

Palm Pilot. 33Mhz 68000 with 8MB of storage, yet it was fast and efficient.
C=64 1Mhz 6510 with 64k RAM (38 useable), also fast and efficient, you could run a combat flight simulator on it (Skyfox)
Heck, even a 16MB 66Mhz 486 was considered almost insane in early 1994 (and it only had a 340 *MB* HDD, and everything was fine. (I bought that in high school for AutoCAD)

Go back to the same efficient and small code, and our devices will seem about 10 times faster and will last longer.

Moore's law

[jafffacake]

Three years ago in the uk i bought my daughter a dell laptop, i5 processor, 6Gb RAM, 500Gb hard drive, £350. Recently it died, so i looked around for a replacement. listed in the bargain forums here (hotukdeals.com) only a couple of weeks ago was a laptop i5, 6Gb RAM, 1Tb hard drive, £380. So in three years the price has barely changed for a remarkably simiar spec. Moore's law seems dead? I agree with the original poster!

Computers Yes. But theres no point

[danknight48]

Computers will get faster, they always do.

But lets be honest, the influx of Java/Ruby/Python and "easy" amature programming are making our computers slower than they were 5 years ago.
- Slower language before we even start.
- Single thread
- No optimizations. Dreadful performance
- Relying on language safety measures, instead of "good logic". Buggy as hell.
- Relying on 50+ library's, just to use 1 function in each.

If only they would learn C++. Our processors probably wouldn't need to be upgraded for another 5 years.
We can but dream, just a shame we live in this "fast food alpha" development world.

Re:Computers Yes. But theres no point

[jawtheshark]

Are you the same guy that labels porn "amature". It's "amateur". "Amature" doesn't even exist, except if you interpret the "a-" prefix as "not", which then the word would mean "not mature".

Re:performance never measured in MHz

[Gavagai80]

I remember the late 90s. My parents unknowingly bought a K6-2 and didn't realize for years that it wasn't an Intel. Nobody was aware of the competition, but that didn't stop them from buying AMD-powered computers.