A former hockey player founded a company to give athletes and families a better way to identify brain injuries.

By Susan Young

Kelly Gee knows all too well the devastating effects of concussion. A former minor-league hockey player, he says that repeated concussions cut his playing career short. Then, while he was coaching for theChicago Steel junior-league team, a puck struck him between the eyes—an injury he thinks caused severe depression. “After that my whole life fell apart,” he says. He went to dozens of doctors in a variety of specialties but ultimately found few options for tracking and treating the effects of his injury.

That, he says, pushed him to found Quantum Institute, a company that now offers a mobile brain scan for concussions. Gee hopes it will help athletes and their families better identify brain injuries and track their recovery. Currently, concussions can be identified on the sidelines by physicians and athletic trainers who run through a series of questions like “What’s your name?” and watch for other signs that a player is dazed or disoriented. But these tests can miss some injuries, and not all teams have the budgets to keep medical experts at the ready.

Quantum Institute’s first product is a brain-mapping system based on electroencephalography, or EEG. This test, commonly used to monitor neurological disorders like epilepsy, detects electrical activity in the brain through sensors placed on the scalp. Gee’s company uses quantitative EEG, which includes computer-aided analysis of the wave patterns in the scan.

This technology has a mixed history and is often done incorrectly, says Marc Nuwer, a neurophysiologist with the Brain Research Institute at the University of California, Los Angeles, who is not involved with the new company.

But this kind of brain test is well suited to identifying the often subtle injury of concussion, which are caused when the brain slams against the skull. Though other methods, like CAT scans, can detect anatomical changes that may be associated with more severe brain injury, concussions don’t usually cause those kinds of changes. A test for concussion needs to be able to detect changes in brain function, and quantitative EEG, when done correctly, can provide an objective measure of brain activity. This could be particularly valuable for team doctors or trainers who are under pressure to get a player back in the game after a blow to the head, says Nuwer, who has served as a neurologist for the Los Angeles Kings hockey team. Such a test could help bolster an argument that a player should stay out.

An important key to making the technology suitable for identifying concussion is to have a baseline measurement, says Nuwer, since the brain activity signals picked up in an EEG test can vary greatly from person to person. Quantum Institute’s test does start with a baseline—a three-dimensional map of brain activity made before an athlete’s season starts. Then, if the player later gets hit, he or she could get scanned again to look for changes in brain function. This test could happen on the sidelines, in a training facility, or at a doctor’s office.

The test takes about half an hour or less, says Gee. Twenty electrodes on the scalp measure brain activity while the athlete sits in a chair. The software can then confirm whether there is a concussion and identify its location and severity, he says.

The startup has run about 300 tests on hockey players in the Chicago area since it began offering its concussion-detecting technology to teams there in October, Gee says. He says the company will offer the test nationally late this year.

Gee wants the test to be available to all athletes, even kids, and he believes that at $100 to $150 per scan, it will be readily affordable. “Our mission is to make sure that there is nobody out there that takes a second guess if they could afford to have the test,” he says.

BY PAUL CARSTEN

(Reuters) – Apple Inc is finally launching its iPhone on ChinaMobile Ltd’s vast network on Friday, opening the door to the world’s largest carrier’s 763 million subscribers and giving its China sales a short-term jolt.

Underlining how much the launch means for Apple, Chief Executive Tim Cook was on hand as the carrier’s flagship store opened in Beijing’s financial district. With China Mobile Chairman Xi Guohua alongside, Cook gifted signed iPhones to a handful of customers and posed for pictures.

The long-awaited deal could trigger a limited turnaround for Apple, whose fortunes have wavered in China in the face of stiff competition from market-leader Samsung Electronics Co Ltd and up-and-coming local rival Xiaomi Tech.

Samsung had a 21 percent share of China’s smartphone market in the third quarter of 2013, with Apple trailing in fifth place with just 6 percent, according to research firm Canalys.

But the arrival of the iPhone could be a double-edged sword for China Mobile, with some analysts predicting a costly subsidy war as rival carriers compete to lure customers.

“I don’t see a price war coming where Apple is engaged in the war, but I do think you’re going to see a subsidy war coming,” said Michael Clendenin, managing director of Shanghai-based RedTech Advisors.

“China Mobile, if they’re not making their targets on sales for these phones, they’re going to increase the subsidies… It’s like airlines: the other guys will fall like dominoes, so China Unicom will do it and China Telecom will do it.”

China Mobile’s iPhone sales are expected to reach 12 million in its 2014 fiscal year, but its subsidies will leap 57 percent to 42.4 billion yuan ($7 billion), up from 27 billion yuan in its fiscal year 2013, wrote Cynthia Meng, a Jefferies analyst, in a December note.

For the basic 16GB iPhone 5S, with no subscriber contract, China Mobile is charging 5,288 yuan ($870), the same as on Apple’s China website. The carrier is charging 4,488 yuan ($740) for a basic iPhone 5C, again the same price as on Apple’s China site.

China Unicom Hong Kong Ltd and China Telecom Corp Ltd slashed their iPhone prices by as much as 1,288 yuan ($210) following the announcement that a deal had been struck between Apple and China Mobile. The pair have also offered a range of cut-price deals on contracts.

These offerings and the launch of the iPhone on China Mobile come in the weeks running up to Chinese New Year, when people traditionally exchange gifts of money in red envelopes and retail sales jump.

SALES CANNIBALISATION

After taking years to hammer out a deal with China Mobile, Apple’s sales in China should get a short, sharp boost as subscribers make the most of the double-whammy of the iPhone’s arrival and the rollout of high-speed 4G mobile networks.

Reservations for iPhones had already hit 1.3 million on Wednesday, according to a China Mobile spokeswoman, although Reuters checks showed that there were multiple registrations using fake ID numbers.

But the rewards are expected to be short-lived for the Cupertino, California-based company, which faces a deeper problem in China of having fallen out of favor with consumers who are increasingly opting for domestic offerings.

Another issue is the thriving grey market for iPhones, where users can buy handsets typically smuggled from Hong Kong and then sign up for a China Mobile contract. China Mobile already has 45 million iPhone users in China, according to a company spokeswoman.

“You need to consider the cannibalization for sales to China Unicom, China Telecom and the grey market, so even though there’s an addition from China Mobile it will also impact sales from other channels as well,” said CK Lu, a Taiwan-based analyst with Gartner.

“If we really want to see the expansion of sales we’ll have to wait for the next version of the iPhone. If China Mobile gets first launch and their subsidies are attractive people will probably rush to the iPhone that China Mobile can provide.”

($1 = 6.0557 Chinese yuan)

(Additional reporting by Adam Rose; Editing by Alex Richardson and Kenneth Maxwell)

(Reuters) – IBM Corp said it will invest more than $1.2 billion to build up to 15 new data centers across five continents to expand its cloud services and reach new clients and markets.

The new cloud centers will be in Washington D.C., Mexico City, Dallas, China, Hong Kong, London, Japan, India and Canada, with plans to expand in the Middle East and Africa in 2015.

IBM said the investment will bring up its data center count to 40 this year and double cloud capacity for SoftLayer, which leases online storage space to companies and was acquired by IBM last year for $2 billion.

“This global expansion is aimed at accelerating into new markets based on growing client demand for high-value cloud,” the company said in a statement.

IBM said the global cloud market is estimated to grow to $200 billion by 2020.

More companies are opting for cloud computing, which lets them rent computing power, storage and other services from data centers shared with other customers, which is typically cheaper and more flexible than maintaining their own.

IBM also said it will use web hosting technology from SoftLayer for the delivery of its cloud services.

IBM said since its acquisition of Dallas-based SoftLayer, the business has added 2,400 new clients.

(Reporting by Sruthi Ramakrishnan in Bangalore and Nicola Leske in New York; Editing by Gopakumar Warrier)

BY BILL RIGBY AND CHRISTIAN PLUMB

(Reuters) – Speculation over Microsoft Corp’s succession plans refocused on internal choices on Wednesday, a day after the leading outside candidate, Ford Motor Co CEO Alan Mulally, took himself off the list of potential CEOs at the world’s largest software maker.

A source close to the company said no appointment was likely until the last week of January or in February. The source asked not be identified because they were not authorized to speak publicly about the process to select a successor to long-time CEO Steve Ballmer, who in August announced his plan to retire.

In late January, many Microsoft executives will be focused on the company’s earnings, which are scheduled to be released on January 23. Also, chairman Bill Gates, a key member of the search committee, will be at the World Economic Forum in Davos.

For months, Microsoft watchers had pegged Mulally as the odds-on bet to succeed Ballmer as chief executive at the dominant producer of software for desktops and laptops, which has faltered in making the transition to the fast-growing mobile phone and tablet markets.

But Mulally formally pulled out of the running on Tuesday, telling the Associated Press he would remain at Ford through 2014.

It was unclear whether Mulally’s withdrawal from consideration was his decision or Microsoft’s. While the company’s intentions remained a guessing game, internal candidates now seemed the obvious ones, especially in light of comments by Microsoft about the importance of tech know-how for the top job.

Mulally appeared to be the front runner in the latter part of last year and seemed to be interested in the job, but Microsoft never made him an offer, suggesting that the prevailing mood on the company’s board was that he was not right for the job, two sources familiar with the process said.

The sources have also told Reuters that Microsoft is down to a “handful” of candidates, including insiders Satya Nadella, executive vice president of the Cloud & Enterprise group and Tony Bates, executive vice president of Business Development and Evangelism.

Former Nokia CEO Stephen Elop, a Microsoft veteran before moving to Nokia, whose mobile phone business Microsoft bought last year, has also been cited as a top candidate, as have one or more outsiders from the tech industry.

As a former Microsoft executive who will soon rejoin the company when the takeover is complete, Elop was mentioned as now “front and center” of the CEO race in a research note from FBR analyst Daniel Ives, who cited his “previous Microsoft experience and demonstrable expertise in the mobile space.”

Evercore analyst Kirk Materne saw it differently, naming Nadella and Bates as the leading internal candidates who “could potentially institute organizational change at a more rapid pace given their insider status.”

Microsoft shares, which have gained 36 percent over the past year, were down 1.2 percent, while Ford shares gained about the same amount.

‘DARK HORSE’ IN THE RUNNING?

Nomura’s Rick Sherlund labeled Mulally’s withdrawal as “disappointing” in a research note, echoing the sentiment of some other investors and analysts.

Sherlund said it was unclear who the board would turn to, saying he viewed Bates and Nadella as capable, but “more likely to take direction from Microsoft’s board rather than redirect the board and take the company in a different direction as we prefer.”

If the company does opt to go with an outsider, candidates could include Facebook Inc Chief Operator Officer Sheryl Sandberg, VMware Inc CEO Pat Gelsinger and Pivotal CEO Paul Maritz, Ives said.

Sherlund agreed that Microsoft could still look at a “dark horse” outsider but noted that some tech sector candidates, including Maritz, may have already declined to take the job.

In a blog post on the company’s website in December, Microsoft lead independent director John Thompson emphasized the need for a CEO with good tech bona fides and “an ability to lead a highly technical organization and work with top technical talent.”

Thompson, who leads the four-member CEO search committee, said at the time he expected the panel to reach a decision “in the early part of 2014.”

(Reuters) – France’s data protection watchdog has fined Google 150,000 euros after the U.S. search engine ignored a three-month ultimatum to bring its practices on tracking and storing user information in line with local law.

The privacy watchdog, known as CNIL, has also ordered Google to post the decision on its google.fr homepage for 48 hours within eight days of being officially notified of the ruling.

At issue was the new approach to user data that Google began in March 2012, in which it consolidated its 60 privacy policies into one and started combining data collected on individual users across its services, including YouTube, Gmail and social network Google+.

It gave users no means to opt out.

“The company does not sufficiently inform its users of the conditions in which their personal data are processed, nor of the purposes of this processing,” CNIL said in a statement.

A Google France spokesman told Reuters the company will take note of this decision and consider further action.

“Throughout our talks with CNIL, we have explained our privacy policy and how it allows us to create simpler and more efficient services,” he said.

Spain, Britain, Germany, Italy and the Netherlands have also opened similar cases against Google because the U.S.-based web giant’s privacy policy introduced in 2012 does not conform with local rules protecting consumers on how their personal data is processed and stored.

CNIL said the fine is the highest it has issued until now and is justified by the number and the seriousness of the breaches stated in the case.

But the penalties that France and most other EU countries can impose remain small compared with the $10.7 billion net profit that Google earned in 2012.

Spain can impose fines of up to 1 million euros, while the German Data Protection Act caps penalties at 300,000 euros.

There is no legal framework to levy European-wide fines.

In June, CNIL found Google to be in breach of privacy law on six counts, notably that it posted “insufficient” information for French users about how their private browsing data was collected and used.

Google has said that its privacy policy “respects European law”.

At CES, Ford shows a concept for increasing the amount of solar power a car can generate.

Putting solar panels on an electric car so that you don’t have to plug it in to charge it might sound like a good idea. But the amount of sunlight that hits the surface of a car isn’t enough to completely charge its battery in a day, even if the car is left out in the sun all day, and even if solar panels were far more efficient than they are now. Toyota offers solar panels on its Prius, but they only generate enough power to run a ventilation fan.

At the Consumer Electronics Show in Las Vegas, Ford is showing off an idea for how to get more power out of the solar panels, enough to completely charge the battery in its C-Max plug-in hybrid in 6 hours, providing 21 miles of range. That’s enough to cover the commute of about half of the people in the country. Technology used for automated parking is part of what makes it possible.

Instead of using only the sunlight that falls on the car, Ford suggests building tall carports whose roofs are made of a type of flat lens, called a Fresnel lens. These will gather sunlight over an area about 10 times larger than the surface of a car and focus it on solar panels on the roof of the car, greatly increasing the amount of solar energy those panels can generate.

The idea of concentrating sunlight to generate power is an old one. What’s new is Ford’s strategy for keeping down the cost of the system. As the sun moves across the sky, the point where the lens focuses the light moves. In conventional concentrating systems, the lens is mounted on a costly tracking system so that it can follow the sun throughout the day, keeping sunlight focused on the solar panels. Instead of moving the lens, Ford moves the car.

The car’s software keeps track of the path of the sun on any given day of the year to help determine how the car should move. It also monitors the amount of sunlight the solar panels are generating to make sure the car has moved to the right place.

Why not just put solar panels on the carport and use those to charge the car? In theory at least, Ford’s proposed system could be cheaper. The lens can be made of plastic and be cheaper than the dozens of solar panels you’d need to generate the same amount of power.

So far, the system is just a concept—Ford doesn’t have plans to sell them anytime soon. It has to work out, among other things, how to keep people and objects out of the the path of the concentrated sunlight when the car isn’t there (the beam could burn you).

The system isn’t exactly the ideal of the solar-powered car—which would be to allow you to keep your car charged wherever you go, giving you freedom like you have in a conventional car that can be refueled at ubiquitous gas stations. It’s hard to imagine Ford’s carports installed everywhere (they’re 5 meters tall and bigger than an ordinary carport), and they charge too slowly for road trips. They might be useful at workplaces, but providing outlets for plugging in would probably be cheaper, and it would work with electric cars that don’t have solar panels.

Tesla Motors has another approach to solar powered cars that might work better. It plans to use solar panels to charge batteries stored at its supercharging stations. Those batteries could then quickly charge electric cars, and get them back on the road. Such systems are expensive, but could still produce electricity cheaply enough to compete with powering a conventional car with gasoline, especially as the price of solar panels and batteries continues to fall.

3M’s new silver nanowire films could lead to large, interactive, and ultimately flexible displays.

By Katherine Bourzac

3M will begin selling flexible transparent conductive films made of silver nanowires for use in touch screens. These nanomaterials could enable wider adoption of large touch screens for interactive signs, displays, and personal computers. And the flexible films may come to be used in future foldable, curvy personal electronics, too.

The St. Paul, Minnesota, company will make the films using silver nanowires produced by Cambrios, a Sunnyvale, California, startup founded in 2004 by two materials scientists: Evelyn Hu, now at Harvard University, and MIT’s Angela Belcher. The company’s silver nanowires are a few nanometers in diameter and a few micrometers long, and come suspended in inks. The inks can be spread out onto a surface to make sparse films. The silver wires are designed to spread in random networks, like nano pick-up sticks, so that they won’t cause a pattern that’s distracting to the eye—an irritating problem that plagued earlier metal-mesh touch screens.

The films are mostly empty space, so they’re transparent. But the nanowires and the ink are formulated so that these films are still highly conductive. The company claims that electrodes made from its ink are more conductive and transparent than the most commonly used touch screen material. Cambrios also says the material can be rolled and unrolled over 100,000 times without breaking.

That’s in contrast to today’s touch screens, which are made of brittle indium tin oxide films. These conventional touch screens have made new kinds of interfaces possible in small devices like smartphones and tablets. But touch screens using the current materials are limited in size and design. The electrical signal has to travel from the touch all the way to the edge of the display, where it’s picked up. If the signal has to go very far, it becomes too weak, and the screen isn’t responsive enough. To make indium tin oxide electrodes more conductive and boost that signal, manufacturers can lay on thicker films, but that makes the display less transparent.

The new nano films could also be beneficial for conventional rigid displays. Indium tin oxide is deposited onto sheets of glass in vacuum chambers; half the material lands not on the glass but on the inner surfaces of the chamber and must be recovered and reused. The nanowire inks can be coated more efficiently and can be made on thin, flexible sheets of plastic rather than glass. 3M uses PET, the same plastic found in many water bottles. Glass is getting thinner and lighter, and touch screens built on plastic should be thinner and lighter still—a selling point for cell phone makers.

3M is not the first company to start selling a touch screen product that uses Cambrios inks. In October 2012, Cambrios announced that its inks were being used in LG’s all-in-one computer and some displays; the Cambrios materials are also found in cell phones and tablets made by NEC in Japan and Huawei in China. Supplying nanowires to 3M, however, will enable the company to get into more devices. 3M is making the touch screen films for device makers but has not yet named customers.

A book-sized computer capable of driving a car could help the technology reach the mass market.

By Tom Simonite

Carmaker Audi showed off a book-sized circuit board capable of driving a car on Monday at the International Consumer Electronics Show (CES). Audi claims the computer, called zFAS, represents a significant advance in automation technology because it is compact enough to fit into existing vehicles without compromising design.

Several different Audi vehicles equipped with zFAS drove themselves onto the stage during the presentation, and a new concept car designed to showcase it was also introduced.

The car, called the Audi Sport Quattro Laserlight, is capable of what Ulrich calls “piloted driving” but betrays no outward sign of being different from a conventional vehicle.

“At CES one year ago, the trunk of the demo cars was still full of cables and electronics,” said Audi’s chief technical officer, Ulrich Hackenberg, about the company’s automated driving technology. “The prototype period is almost over. Now it’s time to get ready for serious production.”

Long and mid-range radar systems, several video cameras, a laser scanner, and ultrasonic distance sensors on the front and sides of the car are all small enough to be hidden from view. The best known self-driving cars, the modified Lexus SUVs used by Google, have a large laser scanner visible on top (see “Data Shows Google’s Robot Cars Are Smoother, Safer Drivers Than You or I”).

Audi talked at CES last year about its engineers’ progress in shrinking down laser scanners and other sensors used to monitor the car’s environment (see “Audi Shrinks the Autonomous Car”). Hackenberg said yesterday that his company got help from chipmaker Nvidia on shrinking the zFAS. It is powered by two processors from Nvidia more typically used in tablet computers.

Audi didn’t say when its Piloted Driving technology would be available commercially. No details were provided on the capabilities of a zFAS-enabled car, beyond saying it could drive in traffic and park on its own.

Despite Google’s public cheerleading of automated vehicles, many established automakers have played down the idea that full automation is near, despite several having technology to match or beat Google’s (see “Driverless Cars Are Further Away Than You Think”). However, Audi chairman Rupert Stadler said that people will routinely let their car do the driving for them in the future, and a promotional video was shown in which a chauffeur sat in the passenger seat while the car did the driving. Stadler described Audi’s Piloted Driving technology as moving the company into new territory: “Today we see a period of major changes, in which we are moving from refining the automobile to redefining mobility.”

Audi also announced partnerships with mobile chipmaker Qualcomm, which will be providing 4G LTE wireless chips in some Audi vehicles, and also with Google. Audi is one of several carmakers that have teamed up with the company to develop a car-centric version of the Android mobile operating system, something Stadler said was already bearing fruit in a new interface for an upcoming concept car that customized itself to each driver. “Thanks to our joint efforts with Google, your interface will feel familiar because it is more intuitive than ever,” he said.

Smartphone battle moves from software to hardware with a crucial component to cut power consumption and allow faster data transmission.

By David Talbot

Beyond the glitz of the International Consumer Electronics Show, the wireless industry faces a fundamental problem: more features and faster data transmission are draining phones’ batteries faster than ever.

Fortunately, there’s room for improvement inside the devices, in parts known as power amplifiers that turn electricity into radio energy. In phones, they typically consumer more power than any other component but waste half of it along the way, as lots of people can attest if they’ve watched their battery die (and their phone get warm) after an hour of streaming video. The same problem bedevils wireless networks’ base stations, which send and receive signals to and from individual phones.

Now a major effort is under way to develop smarter power amplifiers that significantly reduce waste. Eta Devices, an MIT spinoff based in Cambridge, Massachusetts, is preparing a base station module and a chip that it says not only decrease battery drain but work well in high-bandwidth applications for 4G LTE and future ultrafast technologies.

The fundamental problem is that the power needed for radio output fluctuates rapidly when a device is transmitting data at high rates. Existing power amplifiers maintain their voltage at a fairly high level at all times to be prepared for peak needs—but this is wasteful. Newer approaches adjust that level on the fly, following the “envelope” of the actual radio signal.

Such “envelope tracking,” or ET, technologies are the hottest hardware development in the mobile-phone industry. Last fall Qualcomm became the first company to ship a chip with such technology, which it says is the industry’s first for 3G and 4G LTE mobile devices.

The company says the chip helps lower electricity consumption by 20 percent and helps reduce a related problem—heat generation—by up to 30 percent, “allowing for longer battery life for end users, as well as enabling manufacturers to shrink the size of their devices,” says Peter Carson, Qualcomm’s senior director of marketing.

The envelope tracker is already in 10 phones, including the Samsung Galaxy Note 3 and Nexus 5. Many other component makers are scrambling to catch up, including Mediatek, RF Micro Devices, Skyworks, Texas Instruments, Analog Devices, Nujira, and Eta Devices.

The difficulty with ET, though, is that its efficiency plunges at higher data rates. Envelope trackers often require a relatively large capacitor to store and release bursts of energy while maintaining smooth and continuous voltage changes.

Eta Devices takes a radically different approach, favoring fast, abrupt changes with a smaller capacitor. Using a smaller capacitor is more efficient; the downside is that the changes in energy cause more noise in a wireless signal. That problem is overcome by cutting-edge digital signal processing, says Joel Dawson, one of two MIT electrical engineering professors who cofounded the company.

Mattias Åström, the company president, reaches for an automotive analogy to compare the two approaches. “Envelope tracking is basically a continuous variable transmission, compared to our manual gearbox,” he says. “Fuel consumption is always better when you have a manual gearbox.”

The company’s work hasn’t been published and the chip is now being fabricated for the first time, but the concept has been built out for base stations and may be commercialized this year. The Eta module, a little smaller than a shoebox, is the first 4G LTE transmitter in the world to achieve average efficiency greater than 70 percent, a big jump from the 45 to 55 percent in currently available technology, says Dawson.

Vanu, a company that makes low-power wireless base stations (see “A Tiny Cell-Phone Transmitter Takes Root in Rural Africa”), is testing the technology and may become an early customer. “We think this can give us a ‘green’ benefit as well as an operating cost advantage,” says David Bither, direct of platform engineering at Vanu.

The result could be to expand connectivity and make it affordable to more people in the developing world, where expensive diesel fuel powers at least 640,000 remote base stations at a cost of $15 billion.

The Eta technology was first revealed as a lab-bench setup in late 2012 (see “Efficiency Breakthrough Promises Smartphones That Use Half the Power”). The company was funded by $6 million from Ray Stata, cofounder of Analog Devices, and his venture firm, Stata Venture Partners.

Hardware designed specifically to run complex neural networks could let personal devices make sense of the world.

By Tom Simonite

A powerful approach to artificial intelligence could be coming to smartphones.

Researchers from Purdue University are working to commercialize designs for a chip to help mobile processors make use of the AI method known as deep learning. Although the power of deep learning has inspired companies including Google, Facebook, and Baidu to invest in the technology, so far it has been limited to large clusters of high-powered computers. When Google developed software that learned to recognize cats from YouTube videos, the experiment required 16,000 processors (see “Self-Taught Software”).

Being able to implement deep learning in more compact and power-efficient ways could lead to smartphones and other mobile devices that can understand the content of images and video, saysEugenio Culurciello, a professor at Purdue working on the project. In December, at the Neural Information Processing Systems conference in Nevada, the group demonstrated that a co-processor connected to a conventional smartphone processor could help it run deep learning software. The software was able to detect faces or label parts of a street scene. The co-processor’s design was tested on an FPGA, a reconfigurable chip that can be programmed to test a new hardware design without the considerable expense of fabricating a completely new chip.

The prototype is much less powerful than systems like Google’s cat detector, but it shows how new forms of hardware could make it possible to use the power of deep learning more widely. “There’s a need for this,” says Culurciello. “You probably have a collection of several thousand images that you never look at again, and we don’t have a good technology to analyze all this content.”

Devices such as Google Glass could also benefit from the ability to understand the abundant pictures and videos they are capturing, he says. A person’s images and videos might be searchable using text—”red car” or “sunny day with Mom,” for example. Likewise, novel apps could be developed that take action when they recognize particular people, objects, or scenes.

Deep learning software works by filtering data through a hierarchical, multilayered network of simulated neurons that are individually simple but can exhibit complex behavior when linked together (see “Deep Learning”). Computers are inefficient at running those networks because they are very different from conventional software.

Purdue’s co-processor design is specialized to run multilayered neural networks above all else and to put them to work on streaming imagery. In tests, the prototype has proven about 15 times as efficient as using a graphics processor for the same task, and Culurciello believes that improvements to the system could make it 10 times more efficient than it is now.

Narayan Srinivasa, director of the center for neural and emergent systems at HRL Laboratories, a research lab jointly owned by Boeing and General Motors, says it makes sense to use a co-processor to help implement deep learning networks more efficiently. That’s because in conventional computers, a processor and its memory reside in separate chunks of hardware. By contrast, the operations of deep learning-style neural networks and the real neural networks they are inspired by intertwine memory and processing. Narayan’s own research focuses on addressing that problem with a more extreme solution – designing chips with silicon neurons and synapses that mimic those of real brains (see “Thinking in Silicon”).

The Purdue group’s solution doesn’t represent such a fundamental rethinking of how computer chips operate. That may limit how efficiently their designs can run deep learning neural networks but also make it easier to get them into real-world use. Culurciello has already started a company, called TeraDeep, to commercialize his designs.

“The idea is that we sell the IP to implement this so that a large manufacturer like Qualcomm or Samsung or Apple could add this functionality to their processor so they could process images,” says Culurciello. Yann LeCun, a pioneer of deep learning at New York University who recently started leading Facebook’s research in the area, is an advisor to the company.