MIT's Eyeriss Low Power Chip Brings Artificial Intelligence To Mobile Devices
MIT has designed a new super chip with 168 cores that is capable of executing advanced artificial intelligence tasks and algorithms locally. Unique quality is that the chip is 10 times more power efficient than graphic processors deployed in mobile phones today. Eyeriss, the super chip, performs complex AI algorithms locally, without communicating with a server over an internet connection, thereby eliminating a major battery-draining operation. Scientists are hopeful of these chips being the neural network of smart devices, and could greatly catalyze the Internet of Things.
Neural networks are computer systems that mimic the human brain and nervous system’s anatomy. The complications of the architecture and consumption of power have limited scientists from bringing AI to smartphones and wearable gadgets. But MIT’s new chip, Eyeriss, is an ultra low-power consuming chip that makes neural networks possible on everyday gadgets. It achieves such high efficiency by minimizing data transfer, both within cores and with an external server.
All 168 cores of Eyeriss have their own memory, as opposed to a central memory bank on graphics processing units. The cores communicate with adjacent cores for information transfer, and compress them before sharing. Another circuit in the chip that allocates tasks to the cores, does so by taking into account what data each core has, and what tasks it can optimally perform without having to reach out for more data.
MIT researchers Vivienne Sze, Yu-Hsin Chen, Prof. Joel Emer and postdoc Tushar Krishna demonstrated Eyeriss performing an image recognition feat at the International Solid State Circuits Conference in San Francisco. Prof. Joel Emer is a senior scientist at NVidia, which suggests that this technology could find its way into consumer electronics sooner than we expect. Their research was partly funded by US defense agency, DARPA, which would be interested in the level of privacy these chips provide by not having to publish all the data they collect to the internet.
In matters related to IoT, this chip could empower smart devices and embedded implants to make decisions locally without having to share all the data they collect with a central core on the internet. The artificial intelligence algorithm and the neural network architecture not only lets the chip act faster and conserve precious battery power, but also learn more about the environment as they evolve. Complex algorithms like face detection and voice recognition could be run on future mobile phones even without an active internet connection.
Source: <a href="https://news.mit.edu/2016/neural-chip-artificial-intelligence-mobile-devices-0203" target="_blank" rel="nofollow noopener noreferrer">Energy-friendly chip can perform powerful artificial-intelligence tasks | MIT News | Massachusetts Institute of Technology</a>
Neural networks are computer systems that mimic the human brain and nervous system’s anatomy. The complications of the architecture and consumption of power have limited scientists from bringing AI to smartphones and wearable gadgets. But MIT’s new chip, Eyeriss, is an ultra low-power consuming chip that makes neural networks possible on everyday gadgets. It achieves such high efficiency by minimizing data transfer, both within cores and with an external server.
All 168 cores of Eyeriss have their own memory, as opposed to a central memory bank on graphics processing units. The cores communicate with adjacent cores for information transfer, and compress them before sharing. Another circuit in the chip that allocates tasks to the cores, does so by taking into account what data each core has, and what tasks it can optimally perform without having to reach out for more data.
MIT researchers Vivienne Sze, Yu-Hsin Chen, Prof. Joel Emer and postdoc Tushar Krishna demonstrated Eyeriss performing an image recognition feat at the International Solid State Circuits Conference in San Francisco. Prof. Joel Emer is a senior scientist at NVidia, which suggests that this technology could find its way into consumer electronics sooner than we expect. Their research was partly funded by US defense agency, DARPA, which would be interested in the level of privacy these chips provide by not having to publish all the data they collect to the internet.
In matters related to IoT, this chip could empower smart devices and embedded implants to make decisions locally without having to share all the data they collect with a central core on the internet. The artificial intelligence algorithm and the neural network architecture not only lets the chip act faster and conserve precious battery power, but also learn more about the environment as they evolve. Complex algorithms like face detection and voice recognition could be run on future mobile phones even without an active internet connection.
Source: <a href="https://news.mit.edu/2016/neural-chip-artificial-intelligence-mobile-devices-0203" target="_blank" rel="nofollow noopener noreferrer">Energy-friendly chip can perform powerful artificial-intelligence tasks | MIT News | Massachusetts Institute of Technology</a>
0