Big Data // Hardware/Architectures
News
11/21/2013
08:06 AM
Connect Directly
RSS
E-Mail
50%
50%

Do Smartphones Need Neural Processors?

Qualcomm is pushing to put its neural processor Zeroth in smartphones. Here's what smartphone makers can -- and can't -- do with it.

7 Slick Siri Alternatives
7 Slick Siri Alternatives
(click image for larger view and for slideshow)

Qualcomm is known for its intellectual property. It hopes to become known for its brain processors. The question for the rest of us to think about is: Should we care?

As with everything tech, the extent of our caring depends largely on what we can do with it. With Qualcomm's neural processor, Zeroth, what we can do with it would currently be zero -- the chip and the elements needed to program it won't be available until sometime in 2014. What might one do with a neural processor? "There's a whole class of problems that computers are not efficient at, and you have to apply almost brute force to get them to work," says Matt Grob, Qualcomm's CTO.

He volunteered examples such as motor control, pattern recognition, facial recognition and complex learning behaviors. All are things the human brain does well but computers do not.

Zeroth will go into phones, most likely. Zeroth phones will have more understanding of context, Grob told me recently. "You can feed it information from industries in your sector and it might be able to spot trends," he said.

[ Can Windows Phones provide a better experience than Siri? Read Microsoft's Answer To Siri: Cortana. ]

Trends, of course, can be spotted in other ways, but Grob did come up with something that sounds useful: training phones to recognize data leakage. "You could take a device and use it normally and transfer some documents and say, 'bad, bad!' And then if you tried to transfer something, the phone could say 'I don't want to do that, don't make me violate confidentiality.' "

Zeroth could also work in other mobile devices, such as robots. Qualcomm put Zeroth into a simple robot, in homage to its name, borrowed from Asimov's Laws of Robotics; the Zeroth Law states that robots must not harm humanity. The following video shows the robot using Zeroth's neural networks to "learn" where to go:

First, as always, beware the hype. Rod Brooks' Baxter also "learns," without anybody touting its brain-like qualities.

Second, using nature to inspire product design, or biomimetics, is a popular idea, but Velcro is probably the only smash hit for the concept. Computing's version of biomimetics is artificial intelligence. AI isn't exactly a failure, but it isn't Velcro, either. The brain has fascinated computer scientists for decades. One of the first stories I wrote, way back in 1989, was on Roger Schank, who had received a then-hefty $30 million from Andersen Consulting (now Accenture) to develop software around his ideas on artificial intelligence. I'm still writing those stories, only now it's on people like Jeff Hawkins, who was running a company (then Numenta, now Grok) that would create algorithms modeled after the human brain.  (See this InformationWeek story on the original iteration of Hawkins's idea.)

Despite the brilliance of the people trying to create digital brains, and the ability of computers to far surpass the human brain in some ways, the microprocessor remains the the brain of computing. And it isn't designed much like the brain. "It's conventional, not biologically inspired" and still has that same old Von Neumann architecture to it, Grob told me.

Third, fear AnnouncementWare. Intel dominates the brains of the computer. Intel is now serious about mobile computing. Intel also happens to be working on a neural chip, as is IBM. Qualcomm needs to keep pace to keep its place in the mobile chip pantheon.

Fourth, the quest to make a digital brain is cool, but what if something else comes along, or the Von Neumann architecture simply continues to work better?

Grob acknowledged that there are challenges for Qualcomm's neural processor. But he says it's a matter of steam engine vs. horse, or transistor vs. vacuum tube. "The incumbent technology is better. But it's probably flatter as well," he says. Neural chips have to be better than what we have now, or they won't sell. "We have to show it is better than something else if people are going to buy it. We'll do that."

Every mainstream technology has a moment where it suddenly matters. Perhaps that moment is about to arrive for neural computing. Perhaps Zeroth is the reason, and when its 400 customers get their Zeroth design kits next year, they'll work up fantastic new products that make us all use our phones in new ways.

Or perhaps we need tools that complement our brains, not re-create them.

InformationWeek 500 companies take a practical view of even trendy tech such as cloud, big data analytics and mobile. Read all about what they're doing in our big new special issue. Also in the InformationWeek 500 issue: A ranking of our top 250 winners; profiles of the top five companies; and 20 great ideas that you can steal. (Free registration required.)

Comment  | 
Print  | 
More Insights
In A Fever For Big Data
In A Fever For Big Data
Healthcare orgs are relentlessly accumulating data, and a growing array of tools are becoming available to manage it.
Register for InformationWeek Newsletters
White Papers
Current Issue
InformationWeek Tech Digest September 18, 2014
Enterprise social network success starts and ends with integration. Here's how to finally make collaboration click.
Flash Poll
Video
Slideshows
Twitter Feed
InformationWeek Radio
Sponsored Live Streaming Video
Everything You've Been Told About Mobility Is Wrong
Attend this video symposium with Sean Wisdom, Global Director of Mobility Solutions, and learn about how you can harness powerful new products to mobilize your business potential.