IBM Advances Neuromorphic Chip Technology

 Neuromorphic Computing  

Researchers at IBM have created by far the most advanced brain-inspired neuromorphic computer chip to date. Called TrueNorth, the chip consists of 1 million programmable neurons and 256 million programmable synapses across 4096 individual neurosynaptic cores.

Researchers led by IBM's Dharmendra Modha have unveiled their latest neuromorphic chip that is the size of a postage stamp and capable of processing massive amounts of data while handling inputs from many different sources, the company said.

The announcement of the TrueNorth chip and the accompanying article in the journal Science, comes one month after IBM unveiled a $3 billion investment over the next five years in chip research and development to find a game-changing breakthrough that can help revive its slumping hardware unit.

Unlike most chips, which operate on pre-written paths, IBM's version processes data in real-time and is capable of dealing with ambiguity, the company said. It runs on the energy equivalent of a hearing aid.

Abstract from Science: Inspired by the brain’s structure, we have developed an efficient, scalable, and flexible non–von Neumann architecture that leverages contemporary silicon technology. To demonstrate, we built a 5.4-billion-transistor chip with 4096 neurosynaptic cores interconnected via an intrachip network that integrates 1 million programmable spiking neurons and 256 million configurable synapses. Chips can be tiled in two dimensions via an interchip communication interface, seamlessly scaling the architecture to a cortexlike sheet of arbitrary size. The architecture is well suited to many applications that use complex neural networks in real time, for example, multiobject detection and classification. With 400-pixel-by-240-pixel video input at 30 frames per second, the chip consumes 63 milliwatts.

TrueNorth is built on Samsung Electro-Mechanics Co Ltd's 28-nanometre process technology, the chip consumes a miniscule 72 milliwatts of energy at max load, which equates to around 400 billion synaptic operations per second per watt — or about 176,000 times more efficient than a modern CPU running the same brain-like workload.  This is 769 times more efficient than other state-of-the-art neuromorphic approaches.

Related articles Energy Efficient Brain Simulator Outperforms Supercomputers  The Human Brain Project Has Begun Building the Super Brain  IBM Develops Foundation to Program Cognitive Computing Chips

Indeed, IBM may now a momentous step closer to building a brain on a chip, in TrueNorth's case the estimate is around that of the brain of a bee.

"To underscore this divergence between the brain and today’s computers, note that a 'human-scale' simulation with 100 trillion synapses required 96 Blue Gene/Q racks of the Lawrence Livermore National Lab Sequoia supercomputer," writes Modha on the IBM Research site.

The product of nearly a decade of research, mainly funded under the DARPA SyNAPSE program, the chip aims to bridge the divide between existing computers and the brain's high cognitive power and low energy use.

"After years of collaboration with IBM, we are now a step closer to building a computer similar to our brain," said Professor Rajit Manohar at Cornell Tech, where the chip was designed.

Facebook's AI head Yann LeCun has expressed some skepticism over IBM's approach however. "My main criticism is that TrueNorth implements networks of integrate-and-fire spiking neurons. This type of neural net that has never been shown to yield accuracy anywhere close to state of the art on any task of interest (like, say recognizing objects from the ImageNet dataset)," writes on his Google+ account.

According to LeCun, if the brain were the model for TrueNorth, multiplying the number of spiking neurons would not improve performance. "The advantage of spiking neurons is that you don't need multipliers (since the neuron states are binary). But to get good results on a task like ImageNet you need about 8 bit of precision on the neuron states. To get this kind of precision with spiking neurons requires to wait multiple cycles so the spikes 'average out'. This slows down the overall computation."

Terrence J. Sejnowski, director of the Salk Institute’s Computational Neurobiology Laboratory praises IBM's work. “It will take many generations before it can compete, but when it does, it will be a scalable architecture that can be delivered to cellphones, something that Yann’s G.P.U.s [graphics processing units] will never be able to do.”

TrueNorth is essentially ready for commercial applications. Running parallel with IBM's big data and super-computing solutions, like Watson, the neuromorphic system could really add a new dynamic to computation.

IBM's brain-inspired architecture consists of a network of neurosynaptic cores. Cores are distributed and operated in parallel. Core operate -without a clock- in an event-driven fashion. Cores integrate memory, computation, and communication. Individual cores can fail and yet, like the brain, the architecture can still function. Cores on the same chip communicate with one another via an on-chip event-driven network. Chips communicate via an inter-chip interface leading to seamless scalability like the cortex, enabling creation of scalable neuromorphic systems.

The chip contains one million programmable neurons and could allow a thermometer to scan and smell chemical signals and deliver a diagnosis, or help a search and rescue robot to identify people in need during a disaster, the company said.

Big Blue's long-term goal is to build a neurosynaptic chip system with ten billion neurons and one hundred trillion synapses, all while consuming only one kilowatt of power and occupying less than two liters of volume.

Based on the materials provided, TrueNorth is essentially ready for commercial applications. Running parallel with IBM's big data and super-computing solutions, like Watson, the neuromorphic system could really add a new dynamic to computation. "Over time, our hope is that SyNAPSE will become an integral component of IBM Watson group offerings," writes Modha.

With TrueNorth consuming so much less power than conventional Von Neumann chips, it would make a fantastically efficient processing addition for computer vision systems and sensor input, artificial intelligence and countless other emerging technologies.




SOURCES  IBM Research, New York Times

By 33rd Square

Embed