IBM Builds Brain-Like Chip With 1 Million Neurons
August 11, 2014 Alex Woodie
IBM scientists last week rolled out SyNAPSE, a “neurosynaptic” chip that boasts the equivalent of 1 million programmable neurons and 256 million programmable synapses. The miniscule chip could hasten the transformation of society by enabling new brain-like “neuromorphic” applications, such as helping the blind see again, and perhaps other less philanthropic adventures.
The introduction of the postage stamp-sized SyNAPSE chip, which is built on Samsung Electronic‘s 28 nanometer technology, is the culmination of a decade of research by IBM into non-von Neumann architectures, which have been the standard in computer design since 1946.
IBM says it has made a lot of progress in this area in a short amount of time. Just three years ago, IBM had a single-core SyNAPSE chip that boasted 256 programmable neurons and about 262,000 programmable synapses. With more than 4,000 cores, the second-gen SyNAPSE chip that it unveiled last week offers more than 1,000X the performance of that first-gen chip.
The key to the breakthrough is the combination of low power consumption, parallelism, and event-driven processes. Chips built on traditional von Neumann architectures run all the time, and have inherent I/O limitations that put a cap on performance, IBM says. By contrast, the SyNAPSE cores–which integrate memory, computation, and communications–run in an event driven, parallel, and fault-tolerant manner. The SyNAPSE chip consumes just 70 milliwatts, which is orders of magnitude less than traditional microprocessors.
There’s still a lot of room to grow, and IBM has plans to get there. The SyNAPSE chip was designed in such a way that it can easily be scaled by tiling chips next to each other, IBM says. When tiled, each chip shares resources with surrounding chips. IBM demonstrated a 16-chip system that boasted 16 million programmable neurons and 4 billion programmable synapses, and revealed plans to build supercomputers using the chips.
Eventually, IBM hopes to build a neurosynaptic system with 10 billion neurons and 100 trillion synapses. That would put it nearly in the same ballpark as the human brain, which is estimated to have about 100 billion neurons and 100 trillion synapses, and runs on about 20 watts of electricity.
IBM worked with Samsung, Cornell Tech, and iniLabs on the development of the chip, which is technically known as System of Neuromorphic Adaptive Plastic Scalable Electronics (or SyNAPSE). DARPA has put $53 million into the project since 2008 (and will need to spend millions more before it gets its robot army).
“This is a huge architectural breakthrough that is essential as the industry moves toward the next-generation cloud and big-data processing,” says Shawn Han, vice president of foundry marketing for Samsung. “After years of collaboration with IBM, we are now a step closer to building a computer similar to our brain,” said professor Rajit Manohar of Cornell Tech, which is a graduate program at Cornell University.
The Big Blue plan calls for fusing traditional von Neumann architectures and new SyNAPSE architectures, with the eventual goal of creating a unified computing architecture that can solve cognitive computing challenges. IBM has already started building an ecosystem of languages, applications, and algorithms to serve this new style of computing.
“We foresee new generations of information technology systems–that complement today’s von Neumann machines–powered by an evolving ecosystem of systems, software, and services,” Dr. Dharmendra S. Modha, an IBM Fellow and IBM Research’s chief scientist for Brain-Inspired Computing, stated in a press release. “These brain-inspired chips could transform mobility, via sensory and intelligent applications that can fit in the palm of your hand but without the need for Wi-Fi.”
It’s tough to conceptualize what sorts of applications may come out of all this. IBM says it’s all about moving computations closer to the data, which will open up a whole host of new capabilities in mobile devices, sensor development, and real-time multimedia cloud services. (IBM inevitably brings everything it does back to the four horsemen of cloud, mobile, social, and big data).
IBM talks about “vision assistance for the blind” as one possible use case. Other applications could be found in the areas of public safety, home health monitoring, and transportation, according to IBM’s infographic on Flickr. But the full breadth of possibilities may only limited only by our imaginations.