That's up to developers. For
IBM's brain-like chips to
see the light of day, software
and hardware engineers will have to write programs for the chips
and build them into devices.
None of those things are possible
with today's computer chips. The processors you have in your
laptop or smartphone are really good at making lightning-fast
computations, but they're really bad at making sense of the world
around them. That's because most chips today are capable of making
just one calculation at a time.
Our brains don't work like that.
They have multiple "cores"
making decisions and inferences at the same time. One
part of the brain might see blue, another might hear waves crashing,
another might smell sea salt, and your brain simultaneously merges
those senses to determine that you're at the beach.
That's kind of how
IBM's new chip works.
Each chip has
4,096 cores, 1 million "neurons" and 256 million "synapses."
That means each core functions at about the scale of
a worm brain, and each chip works at about the same level as
a bee brain. Put lots of these chips together, and things get
of IBM's Neurosynaptic computer chip that can feel, hear, sense,
smell, and taste like a brain.
Amazingly, the chips are so low-powered,
they can easily run on a smartphone battery. But they pack a
punch: They're capable of 46 billion
synaptic operations per second per watt. That's
a supercomputer the size of postage stamp, the weight of a feather
and runs on the power of a hearing-aid," said
Dharmendra Modha, chief scientist for
brain-inspired computing at IBM
Research. "It's a genuinely
research was published in
Science magazine Thursday.The chip was the result
of 10 years of research from
IBM, $53 million in funding
from DARPA, and
200 people working on the project.
this chip was no cakewalk," said
Modha. "Many thought it was impossible. The impossible has
now become possible. We hope the possible will someday become
16 chip board. Source: IBM
unveils computer chip that emulates human brain
Aug 8, 2014
Low power cognitive computing.
IBM project to build a cognitive
computer that emulates how the human brain works has
produced its first part in the postage
stamp-sized TrueNorth chip, which boasts a
million programmable neurons
and only uses a tiny 70 milliwatts
The new chip is the result of
a United States Defence Advanced Research
Projects Agency (DARPA) funded project that started
six years ago.
said the chip breaks away from the traditional
von Neuman architecture that has dominated computing
until now, and works in parallel with over
400 million bits of local,
on-chip memory to store synapses
and neuron parameters.
chips can be
tiled seamlessly to build larger systems. As of today,
IBM said it has built a board with
16 million neurons and four
billion synapses, and Big
Blue researchers are setting their sights on an even
larger system, with 4096 chips
in a single rack.
This would provide
four billion neurons and a
trillion synapses, with a power draw of
In comparison, human brains are estimated to have a
hundred billion neurons and a
hundred trillion synapses, with a power consumption of
just 20 watts.
has long worked on developing
artificial intelligence and built the world's first
cortical simulation in 1956,
with 512 neurons.
It has come along since the
SyNAPSE (Systems of Neuromorphic Adaptive Plastic Scalable Electronics)
project delivered two working
neurosynaptic core prototypes in
2011 that featured 256 programmable
core had 262,144 programmable
synapses and the other 65,536
Today's second generation chip
is substantially more advanced with a
million programmable neurons and
256 million synapses, as well as
4096 neurosynaptic cores.
The long-term goal for
IBM is to build a neurosynaptic
system with ten billion neurons
and one hundred trillion synapses,
which fits into a container of less than
two litres, and draws no more than
one kilowatt of power.
said the technology can be used many areas, both reseach
and industry, such as public safety, improving vision for blind,
home health monitoring and transportation.
© iTnews.com.au . All rights reserved.