Big Blue’s cognitive computing chip could enable wide-ranging applications that take advantage of massive performance while using no more power than a hearing aid.
IBM today unveiled what it’s calling the world’s first neurosynaptic computer chip, a processor that mimics the human brain’s computing abilities and power efficiency.
Known as TrueNorth, IBM’s chip could cram supercomputer-like powers into a microprocessor the size of a postage stamp. Rather than solving problems through brute-force mathematical calculations, like today’s processors, it was designed to understand its environment, handle ambiguity, and take action in real time and in context. Plus, it could be among the most power-efficient chips in the history of computing, enabling new types of mobile apps and computing services, IBM principal investigator and senior manager Dharmendra Modha said in an interview.
Modeled after the human brain, the TrueNorth chip incorporates 5.4 billion transistors, the most IBM has ever put on a chip. It also features 1 million programmable neurons and 256 million programmable synapses. That’s far lower than the 100 billion neurons and 100 trillion to 150 trillion synapses in the human brain — but still enough, Modha said, to run devices that could, for example, proactively issue tsunami alerts, do oil-spill monitoring, or enforce shipping lane rules. And all that happens while consuming just 70 milliwatts of power, about the same as a hearing aid.
The TrueNorth chip is the core element of IBM’s cognitive computing program, which is known as SyNapse.
Other potential applications include powering small search-and-rescue robots; helping vision-impaired people move around safely; and automatically distinguishing between voices in a meeting and creating accurate transcripts for each speaker.
IBM revealed the technology in a paper in Science magazine.
Moving beyond von Neumann
The chip is still in the research phase, with Thursday’s announcement describing the second generation of the design. IBM rolled out the first generation a year ago. While the chip is still in the prototype stage, it could be just two or three years from its first commercial use. Experts believe an innovation like SyNapse’s TrueNorth could help overcome the performance limits of the von Neumann architecture, the mathematics-based system at the core of almost every computer built since 1948.
“It is a remarkable achievement in terms of scalability and low power consumption,” said Horst Simon, the director of the US Department of Energy’s Berkeley Lab and an expert on computer science. “The IBM SyNapse project is an indicator of that change that will happen in the next 10 years.”
For eight years, Modha has led the development of the SyNapse project, with $53.5 million in funding from DARPA, the US Defense Advanced Research Projects Agency. IBM expects the chip to one day help “transform science, technology, business, government, and society by enabling vision, audition, and multi-sensory applications.”
‘A really big deal’
IBM’s announcement is “a really big deal,” said Richard Doherty, the research director at the Envisioneering Group, a technology assessment and research firm. The company’s decision to publish its findings in Science rather than a hard-core technology outlet like IEEE Spectrum is a sign that IBM considers this is a major scientific and technological breakthrough, he said.
TrueNorth could fundamentally change computing because its main utility is being able to autonomously figure things out in much the same way a person or animal would, said Doherty, whose firm does not do business with IBM.
Doherty lauded the non-von Neumann nature of the new architecture, in particular because it does not require the heavy computational load needed for complex operations in traditional systems. For example, if a robot run with today’s microprocessors was walking toward a pillar, it would depend on image processing and huge computing resources and power to avoid a collision. By comparison, a robot using a synaptic chip would steer clear of the danger by sensing the pillar, much as a person would.
“We wouldn’t be able to navigate through our world without senses,” Doherty said. “These devices [could one day] see the world around us through vision, smell…sight, and sound, [just] the way we get around the world.”
IBM hopes to make available the TrueNorth ecosystem — which includes a custom programming language — first to universities and later to business customers, Modha said. Given DARPA’s investment in the project, Doherty also imagines the government using the technology in new types of systems meant to evaluate possible incoming dangers.
In the future, Doherty expects that the technology could be integrated into robots that help power autonomous vehicles, work in Amazon warehouses, and ensure home security.
This is just one of the major computing projects that IBM, which spends $6 billion a year on research and development, has launched in the past few years. In 2011, its Watson supercomputer beat the world’s best “Jeopardy” players, an impressive demonstration of computing power since “Jeopardy” is about parsing massive amounts of vague or ambiguous information.
Watson is the best example of a powerful von Neumann machine, but it requires “more electricity than any home I know of,” Doherty said.
Yet for all that computing power, even Watson struggled to solve a question about airports in Chicago. “The category was US Cities, and the [question] was: ‘Its largest airport was named for a World War II hero; its second largest, for a World War II battle,'” IBM wrote at the time on its own blog. “The two human contestants wrote ‘What is Chicago?’ for its O’Hare and Midway [airports], but Watson’s response was a lame ‘What is Toronto???'”
A synaptic chip ideally would have little problem recognizing that the question had to do with airports, or what city was queried, Doherty said.
Another essential element of the SyNapse program is IBM’s decision to make the new architecture available to anyone. “The real miracle,” Doherty said, “is that this will be open for the next generation of [Microsoft co-founder Bill] Gateses and [Apple co-founder Steve] Wozniaks.”
There are questions as to whether IBM will ever be able to bring its new chip to market. But Doherty believes TrueNorth will deliver on its promise. “This is the second generation,” he said. “This is not a Hail Mary.”