• Home
  • News
  • Artificial Intelligence
  • Internet of Things
  • Open Source
  • Hardware
  • Software
  • Security
  • Whitepapers
  • Industry Voice
  • Data Strategy Spotlight
  • Newsletters
  • Whitepapers
    • Inqlogo 120x194
      Five things you should look for in choosing a Testing provider

      Choosing a Testing Partner can be complex.  So what do you look for?  This guide offers insight into the qualities you must look for in choosing a Testing provider.  Download now to learn more.

      Download
      Inqlogo 120x194
      Your questions answered: How to protect your data in the cloud

      The number of successful cyberattacks per year per company has increased by 46% over the last four years. But what really needs to be considered when exploring a solution? What questions need to be asked? Download to find out...

      Download
      Find whitepapers
      Search by title or subject area
      View all whitepapers
  • Follow us
    • Twitter
    • Newsletters
    • Facebook
  • Newsletter
  • Industry Voice
  • Data Strategy Spotlight
The Inquirer
The Inquirer
  • Home
  • News
  • Artificial Intelligence
  • Internet of Things
  • Open Source
  • Hardware
  • Software
  • Security
  • Trending
  • General election
  • Huawei sues FCC
  • Xerox vs HP
  • Galaxy S11
  • McAfee 2020
The Inquirer
  • Boffin Watch

MIT's neural network chip can reduce device power consumption by 95 per cent

Could make it practical to run neural networks locally on smartphones

MIT's neural network chip can reduce device power consumption by 95 per cent
MIT's neural network chip can reduce device power consumption by 95 per cent
  • Lee Bell
  • Lee Bell
  • @llebeel
  • 16 February 2018
  • Tweet  
  • Facebook  
  •  
  •  
  • Send to  
0 Comments

HIGH-TECH INSTITUTION MIT has developed a neural network chip that could reduce the power consumption of devices by a whopping 95 per cent. 

Ideal for battery-powered gadgets to take advantage of more complex neural networking systems, MIT said the chip could even make it practical to run neural networks locally on smartphones or even to embed them in household appliances.

"Most recent advances in artificial-intelligence systems such as speech- or face-recognition programs have come courtesy of neural networks, densely interconnected meshes of simple information processors that learn to perform tasks by analyzing huge sets of training data," the university explained. 

"But neural nets are large, and their computations are energy intensive, so they're not very practical for handheld devices. Most smartphone apps that rely on neural nets simply upload data to internet servers, which process it and send the results back to the phone."

As a result, the MIT researchers have developed the special-purpose AI chip to increase the speed of neural-network computations by three to seven times over its predecessors, while reducing power consumption up to 95 percent.

"The general processor model is that there is a memory in some part of the chip, and there is a processor in another part of the chip, and you move the data back and forth between them when you do these computations," expained Avishek Biswas, an MIT graduate student who led the new chip's development.

"Since these machine-learning algorithms need so many computations, this transferring back and forth of data is the dominant portion of the energy consumption."

However, Biswas - who conducted the research alongside his thesis advisor, Anantha Chandrakasan, dean of MIT's School of Engineering - said these algorithms can be simplified to one specific operation, called "the dot product".

"Our approach was [to] implement this dot-product functionality inside the memory so that you don't need to transfer this data back and forth," he said, which he added improves efficiency by replicating the human brain.

That's because, in the chip, a node's input values are converted into electrical voltages and then multiplied by the appropriate weights.

"Summing the products is simply a matter of combining the voltages," Biswas said. "Only the combined voltages are converted back into a digital representation and stored for further processing."

Related: ARM unveils two new processor architectures designed for on-device AI

The prototype chip can thus calculate dot products for multiple nodes, that is, 16 at a time in a single step, instead of shuttling between a processor and memory for every computation.

IBM's vice president of AI Dario Gil, labelled the development as "a huge step forward".

"The results show impressive specifications for the energy-efficient implementation of convolution operations with memory arrays," he said.

"It certainly will open the possibility to employ more complex convolutional neural networks for image and video classifications in IoT in the future." µ

  • Tweet  
  • Facebook  
  •  
  •  
  • Send to  
  • Topics
  • Boffin Watch
  • chips
  • AI
  • smartphones
  • neural network

INQ Latest

Comet Lake-S leaks keep hitting earth
Intel Comet Lake-S leak teases AMD-chasing six-core Core i5-10600

Hype for HyperThreading

  • Chips
  • 13 December 2019
Apple's parental controls in iOS 13.3 can be easily bypassed
Apple's parental controls in iOS 13.3 can be easily bypassed

Hey kids, leave them iPhones alone

  • Software
  • 13 December 2019
Opera GX brings gaming-led browsing to macOS
Opera GX brings gaming-led browsing to macOS

The Mac lady sings

  • Software
  • 13 December 2019
Google Assistant gets 'Interpreter Mode' on iOS and Android
Google Assistant gets 'Interpreter Mode' on iOS and Android

Babel in yo ear

  • Software
  • 13 December 2019
Back to Top

Most read

Apple's iPhone 12 won't see a significant price increase, claims Kuo
Apple's iPhone 12 won't see a significant price increase, claims analyst
FTC might order Facebook to stop integrating Instagram, Messenger and WhatsApp
FTC might order Facebook to stop merger of Instagram, Messenger and WhatsApp
Comet Lake-S leaks keep hitting earth
Intel Comet Lake-S leak teases AMD-chasing six-core Core i5-10600
Windows 7 goes end-of-life in a month
Windows 7 goes end-of-life in a month
Galaxy S11 specs, release date and price: In-the-wild images show chunky camera hump
Galaxy S11 specs, release date and price: In-the-wild images show chunky camera hump
  • Contact
  • Marketing solutions
  • Enterprise IT Events
  • Incisive Media
  • Terms & conditions
  • Policies
  • Careers
  • Twitter
  • Newsletters
  • Facebook

© Incisive Business Media (IP) Limited, Published by Incisive Business Media Limited, New London House, 172 Drury Lane, London WC2B 5QR, registered in England and Wales with company registration numbers 09177174 & 09178013

Digital publisher of the year
Digital publisher of the year 2010, 2013, 2016 & 2017