IBM Creates Custom-Made Brain-Like Chip

An anonymous reader writes In a paper published Thursday in Science, IBM describes its creation of a brain-like chip called TrueNorth. It has "4,096 processor cores, and it mimics one million human neurons and 256 million synapses, two of the fundamental biological building blocks that make up the human brain." What's the difference between TrueNorth and traditional processing units? Apparently, TrueNorth encodes data "as patterns of pulses". Already, TrueNorth has a proven 80% accuracy in image recognition with a power consumption efficiency rate beating traditional processing units. Don't look for brain-like chips in the open market any time soon, though. TrueNorth is part of a DARPA research effort that may or may not translate into significant changes in commercial chip architecture and function.

Comments

So we've created George W Bush?

[Anonymous Coward]

I think this chip could govern the US more effectively than our current Congress.

Re:So we've created George W Bush?

[Anonymous Coward]

This chip would be overkill, an old 4004 would be sufficient.

Re:So we've created George W Bush?

[RabidReindeer]

This chip would be overkill, an old 4004 would be sufficient.

A block of wood would be sufficient. A 4004 would be overkill.

Re:So we've created George W Bush?

[tomhath]

But can it read from a teleprompter? If so it should qualify for a Nobel Peace Prize.

Re:So we've created George W Bush?

[blue9steel]

Yes, using S.A.M. (Software Automatic Mouth) via the SID audio chip built in.

Re:So we've created George W Bush?

[ganjadude]

so could an 8008

IBM and chips

[jbolden]

It is getting hard to figure out where IBM is on chips. Arguably the 4 main chips experiencing investment are: x86, ARM, Z-Series processors and POWER series 2 of which are IBM. OTOH there is no roadmap for POWER beyond the current generation. I'd love to know is IBM getting more serious about CPUs or pulling back?

Re:IBM and chips

[jbolden]

No, I'm an IBM partner which means I can sell their stuff. I've never worked directly for them.

So, we are going to have artificial Brains Soon?

[rtoz]

I am just curious to know whether this chip can lead to the development of artificial brains to be used by Humans in future? And, I couldn't understand why this chip will not available in the open market.

Re:So, we are going to have artificial Brains Soon

[K. S. Kyosuke]

Thanks to your capitalization, I've misread it as "development of artificial brains to be used by Hamas". (Yes, I need new glasses.)

to save others googling

[Anonymous Coward]

The number of neurons in the brain varies dramatically from species to species. One estimate (published in 1988) puts the human brain at about 100 billion (10^11) neurons and 100 trillion (10^14) synapses.

100 billion divided by 1 million = 100,000 of these chips to reach the human neuron count.
100 trillion divided by 256 million = 390,625 of these chips to reach human synapse count.

Assuming Moores Law for these chips with a doubling every 24 months to be conservative.
2 of these on a chip in 2016
4 of these on a chip in 2018
8 of these on a chip in 2020
16 of these on a chip in 2022
32 of these on a chip in 2024
64 of these on a chip in 2026
128 of these on a chip in 2028
256 of these on a chip in 2030
512 of these on a chip in 2032
1024 of these on a chip in 2034
2048 of these on a chip in 2036
4096 of these on a chip in 2038
8192 of these on a chip in 2040
16384 of these on a chip in 2042
32768 of these on a chip in 2044
65536 of these on a chip in 2046
131072 of these on a chip in 2048
262144 of these on a chip in 2050

So we could be seeing human brain capabilities on a chip by mid century. Quite possible we'd see similar capabilities built as a supercomputer 10-20 years before that. Don't flame for the wild assumptions I'm making here - i know there are a lot, this is just intended as some back of the envelope calculations.

Re:to save others googling

[Calinous]

Specific activities engage only part of a brain - so we probably only have to go go 10% or so. That cuts less than a decade though, so 2040 something

Re:to save others googling

[K. S. Kyosuke]

Uh, you couldn't possibly be wronger.

Re:to save others googling

[LordWabbit2]

Looks like your grammar is not engaging any part of your brain

Re:to save others googling

[Calinous]

http://en.wikipedia.org/wiki/V...
http://en.wikipedia.org/wiki/C...
http://en.wikipedia.org/wiki/C...
http://en.wikipedia.org/wiki/C... ... so, parts of the brain are specialized for specific activities.

Or you use all your brain for any activity you do, and you can't do two things like Napoleon: sit on a horse and look left ?

Re:to save others googling

[K. S. Kyosuke]

Of course we know that, that's how we found these areas work: if you damage a part of the brain, something vital stops working. The problem is, most of these areas are vital to people almost all of the time.

Re:to save others googling

[Henriok]

There's no chance that Moores law can progress for 50 more years. Wouldn't each transistor be substantially smaller that atom nucleus by then? IF you don't mean a chip the size of a table, that is..

Re:to save others googling

[Anonymous Coward]

Here is hoping they actually keep working at it. You know what IBM are like!
All those plans for Cell, all wasted. Then Power went down the drain and one of their largest buyers (Apple) ditch them because Power was lacking compared to x86, which is just holding back everything.

This is a genuinely interesting thing, possibly the best thing they have made in the longest time in fact.
I couldn't see any reason DARPA wouldn't also be very interested in it, if it works as well as they say it does. Already it is a billion dollar idea in a chip just from the summary description with regards to image recognition.
I'd be surprised if Google doesn't see some interest in this as well. (and NSA of course, why wouldn't they love this?)

Re:to save others googling

[Anonymous Coward]

It's true, there is a documentary about it in the cinemas right now.

Re:to save others googling

[Graydyn Young]

It's really hard to say how many of artificial neurons we would need to make a human-like intelligence, but it's certainly going to be less than the number of neurons in a human head. Computers already do a heck of a lot of tasks better than a human. Heck, using traditional computing methods with just a couple of these chips for image recognition and the like would already make a beast of a machine.

Re:to save others googling

[peragrin]

Yep like image recognition, and audio recognition.

Oh wait.

Computers can do logical operations better yes. Computers can't do fuzzy math, real time image recognition or real time audio recognition. Let me know when a computer can "see" with a pair of cameras. Identify an object heading toward the cpu(not just the cameras) and adjust its motors to dodge the incoming. Bugs can do that much yet computers can't.

Re:to save others googling

[Anonymous Coward]

The math: The latest intel processors use transistors that are 22nm across. The width of a hydrogen atom, ~1.1 angstrom, is about 0.11nm, or 110 picometers across. Assuming the transistor size halves every two years(which, from the looks of it, is impossible), we get this:

2016: 11 nm transistors
2018: 5.5 nm
2020: 2.75 nm
2022: 1.375 nm
2024: 687.5 pm
2026: 343.75 pm
2028: 171.875 pm
2030: 85.9375 pm

And then we're smaller than the smallest atom. However, this is not smaller than the nucleus of the hydrogen atom(a single proton), which is still orders of magnitude smaller than the total size of the atom. But it's hard to imagine right now being able to have transistors smaller than an atom's width. We'll likely need something entirely new that can act like a transistor but isn't. Much like transistors act like vacuum tubes but aren't.

Just for completeness, 50 years out we would expect a transistor to be 0.00008392333984375 nm, or ~0.084 picometers. This is 84 femtometers, and the currently theorized size of a proton is 0.87 femtometers, so 50 years from now transistors would still be 2 orders of magnitude away from being smaller than the hydrogen's nucleus. Even the heaviest atoms' nuclei only ranges up to about 15 fm, so it would take ~56 years from now to be able to say transistors are smaller than atomic nuclei. It would be 64 years before transistors are smaller than a hydrogen atom's nucleus. Of course all of this is complete bullshit because Moore's law was really just Moore's observation of a coincidence. Even now we're not halving the size of transistors every two years.

Re:to save others googling

[timeOday]

If neuron-like processing turns out to be advantageous, there will be much more efficient ways to implement them than using tens of thousands logic gates to simulate each one.

Re:to save others googling

[Anonymous Coward]

"IBM has already succeeded in building a 16-chip system with sixteen million programmable neurons and four billion programmable synapses. The next step is creating a system with one trillion synapses that requires only 4kW of energy. After that IBM has plans to build a synaptic chip system with ten billion neurons and one hundred trillion synapses that consumes only one kilowatt of power and occupies less than two liters of volume."

I think the IBM roadmap is more aggressive than Moore's law, and of course getting to "human brain level" in a supercomputer is a big deal even if it doesn't happen in a single chip.

If the ultra low power consumption holds up, maybe it will be easier to make a 3d stacked chip version. It could certainly become comparable to a brain. 100 layers would be the equivalent of ~10-12 years of "Moore's law" scaling.

Re:to save others googling

[ShanghaiBill]

The problem is, most of these areas are vital to people almost all of the time.

But they are NOT vital to an AI computer. An intelligent computer should only have to emulate the cerebral cortex, not the entire brain.

Re:to save others googling

[Anonymous Coward]

you are halving when it's area not length - you should be multiplying by .66

that comes out to;
2016: 11nm
2018: 7.26nm
2020: 4.7916nm
2022: 3.162456nm
2024: 2.08722096nm
2026: 1.3775658336nm
2028: 909.193450176pm
2030: 600.06767711616pm
2032: 396.044666896666pm
2034: 261.389480151799pm
2036: 172.517056900188pm
2038: 113.861257554124pm
2040: 75.1484299857217pm
2042: 49.5979637905764pm
2044: 32.7346561017804pm
2046: 21.6048730271751pm
2048: 14.2592161979355pm
2050: 9.41108269063746pm

Yes this creates problems, but is made worse if you assume the current chip starts at the 22nm level. the current chip is probably a few orders of magnitude larger as that makes it much more economical to produce in limited numbers. Also there'e nothing to stop Moores law continuing if physical sizes of chips are just allowed to grow...

Re:to save others googling

[saider]

Moore's law applied to transistor count and the atomic limit only applies if you limit yourself to 2 dimensional chips.

Re:to save others googling

[Anonymous Coward]

Your brain has over a dozen different types of neurons with different functions and individual neurons themselves can have varying structures that can do more complex functions (like AND/OR/NEGATING within the same cell with different groupings of inputs)

Some brain Neurons have thousands of inputs from nearly that many nearby nerve cells and brains have overall layer patterns often with broad regional interconnects of different specific functions.

A million standard IBM-neuron each with 256 synapse inputs (with what variance of signal?) may do only a tiny fraction of what a million real brain cells do.

Human brain is up in the 10 billion neuron range (synapses interconnects in the trillions)

Re:to save others googling

[Anonymous Coward]

Each synapse contains dozens or hundreds of individual receptors that interact with the chemicals (neurotransmitters) being released to transmit the message. Certain types of receptors, called metabotropic, set off a cascade of enzymatic reactions inside the cell that represents further, highly complex, information processing. So when calculating the number of processing units in the brain, you have to go well beyond counting synapses. It's also worth noting that some of the interactions that take place can only be explained with quantum mechanics, so one way of describing the brain is as a quantum computer with quadrillions of processors.

Re:to save others googling

[mark-t]

Although we do really use most of our brain over a prolonged period, only a very small amount of it is used at any one time for cognitive activities. 10% may be a bit of a low estimate, but not by much.

Re:to save others googling

[mark-t]

Babies can't do any of those things very well either.

Re:to save others googling

[kylemonger]

The human brain needs all that parallelism because it's switching rate is abysmal, something like 200Hz. We ought to be able to beat that by a factor of million without setting anything on fire, so adjust your numbers accordingly.

Re:to save others googling

[fnj]

Do you really believe that? Even iIf your transistor is 100 atoms high, it still can't be less than 1 atom wide or deep.

Re:to save others googling

[roman_mir]

You can simulate this guy by 2030 then.

So when are we having skynet?

[thieh]

Is it coming soon?

Now they can replace IBM managers

[gelfling]

Assuming of course this chip can hold 2 hr conference calls with 40 other chips and pound out 240 page Powerpoints.

Re:Is it real or a Real Doll?

[distilate]

real dolls with minor amounts of brains could help solve the navys join the navey, feel a man problem! Ie preventing rampant homosexuality during long missions

HAL 9000?

[Anonymous Coward]

Where are all the HAL 9000 jokes? HAL was built by IBM in "2001: A Space Odyssey", perhaps this is an example of life imitating art?

Re:So, we are going to have artificial Brains Soon

[Anonymous Coward]

I couldn't understand why this chip will not available in the open market.

The millitary boys want to keep their toys secret to prevent the enimies getting them.

ob Heinlein

[Alrescha]

"Human brain has around ten-to-the-tenth neurons. By third year Mike had better than one and a half times that number of neuristors. And woke up." -- The Moon is a Harsh Mistress

A.

There is POWER9 on the roadmaps

[Henriok]

I agree i your initial statement, but that's pretty much as it has been for at least 15 years or so. POWER9 is on the roadmaps, and the next generation zArch too. And they are sitting there like proxy boxes with nothing much spced, like it has been for almost all previous generations of their predecessors. What I'm concerned with is the lack of public roadmap for what they are planning in the HPC and super computer space. We had the very public Blue Gene project that began in 2001 with four projects; C, L, P and Q, but since the Blue Gene/Q came to life a couple of years ago, I have no idea what they are planning. It'd be nice to have some clue here.. Why not something from the OpenPOWER Foundation; A P8 host processor with integrated GPU from nVidia, on chip networking from Mellanox and programmable accelerators from Altera. But I haven't seen anything in that direction.

Pulse Pattern

[JimSadler]

Is this pulse recognition closer to Morse code or bar code? Obviously pulse code recognition could go beyond binary bit codes but the hardware must be seriously different from what we have now. It could even be analog.

Re:So, we are going to have artificial Brains Soon

[PolygamousRanchKid]

. . . I wacky-parsed the title as: "IBM Creates Custom-Made Brain-Like Chimp.

. . . so just imagine where that thought train derailed me . . .

Re:HAL 9000 jokes....

[tekrat]

I'm sorry Dave. I'm afraid I can't do that.

Re:Is it real or a Real Doll?

[Anonymous Coward]

real dolls with minor amounts of brains could help solve the navys join the navey, feel a man problem! Ie preventing rampant homosexuality during long missions

And here we have a demonstration of an organism with 256 million synapses.

The Genisys of Skynet!

[wiredog]

As good as any other theory, I guess.

Actually...

[voss]

A commodore 64 running NATO COMMANDER on a datasette could run the George W Bush thought process.

Re:Actually...

[Sciath]

Neoliberal as opposed to a neoconservative? What makes the conservative philosophy superior? Nothing. One of the dons of neoconservativism is GWB. And what was it he accomplished? A twelve year, $2 trillion waste of lives and resources that was and continues to demonstrate a utter failure of foreign policy. During his tenure he created the financial atmosphere of rhinos running loose in a china store. He refused to install import levies even when he knew foreign countries were running slave labor camps and some of those same nations did have import levies on American made products. He knew nothing about diplomacy by essentially telling European countries to go screw themselves with his "with us or again us" mindset. In fact, he should be sitting in federal prison for his incompetent and reckless administration. The ONLY thing he did to benefit anyone but himself was to get Medicare D signed into law.

Re:Is it real or a Real Doll?

[tekrat]

Walking, talking sexbots would absolutely change the world, possibly eliminate most crime, might solve the problem of overpopulation, and as theorized in the manga/anime Chobits, might force real women to have to compete for male attention.

Human Brain is...complicated

[Scottingham]

Every time one of these damn 'neural computers' come out people tend to equate the number of neurons and synapses and think 'hey, if we can get to the number of human neurons... Presto!!!!1'

Brains are waay more complicated than just neurons and synapses. Just taking the neurotransmitters into account makes the whole charade crash down. Then there is the glial network that, surprise surprise, does an enormous amount of complex work. There's even recent research suggesting that the branching patterns of the neurons perform complex computations. There are chemical gradients in the brain that act as a sort of addressing system.



tl;dr Brain on a chip? Yeah fucking right.

Re:Marketing BS

[mark-t]

Evolution got Shakespeare after throwing enough monkeys into the mix.... why can't we?

Re:So, we are going to have artificial Brains Soon

[gewalker]

Can't Purina use them in Zombie Chow? If so, I would rather feed that to the neighborhood zombies instead of my own gray matter?

Re:So, we are going to have artificial Brains Soon

[narcc]

I am just curious to know whether this chip can lead to the development of artificial brains to be used by Humans in future?

That's easy: No.

Kurzweil is the modern equivalent of a televangelist.

Re:Is it real or a Real Doll?

[CastrTroy]

Futurama already covered that issue.

Bad example

[DrYak]

Let me know when a computer can "see" with a pair of cameras. Identify an object heading toward the cpu(not just the cameras) and adjust its motors to dodge the incoming.

That actually do already exist.
It's a car's collision avoidance system.
It's already standard option from some manufacturer (e.g.: Volvo) (and should become mandatory in EU somewhere soonish).

Some like Mobileye rely entirely on camera, while other are integrating other sensors in the mix, like radar, infra red lasers, etc.

But yeah I see your point: complex task require complex network, way much more than this chip.

Re:Flaw in the description

[fisted]

Software can emulate any hardware you build

And what about the performance implications of emulating my massively parallel hardware in software?