MIT created new chip for neural networks that increases computation speed by seven times and reduces power consumption by 95%, allowing more efficient integration into mobile and IoT devices

Briefing

MIT created new chip for neural networks that increases computation speed by seven times and reduces power consumption by 95%, allowing more efficient integration into mobile and IoT devices

March 1, 2018

Briefing
  • Neural Network Chip – MIT developed new chip that increases neural network computation speed up to seven times and reduces power consumption up to 95%
  • Simplifying Node Movement – Discovered new operation called dot product, which simplifies node movements between memory and processor, saving significant energy consumption from multiple transfers
  • Performance vs. Conventional Systems – Tests showed results accuracy within two to three percent of those from neural networks built on conventional systems
  • Suitable for Battery-Powered Devices – Can be used to power AI computations directly on mobile devices, rather than uploading to cloud servers, as well as incorporate AI capabilities into Internet-of-Things devices

Accelerator
Artificial Intelligence, Computation and Storage, Internet-of-Things, Mobile and Wearable Devices

Sector
Information Technology

Organization
MIT Media Lab

Source

Original Publication Date
February 13, 2018

Leave a comment