Search Here

A team of Massachusetts Institute of Technology (MIT) researchers has introduced a new computer chip at the International Solid State Circuits Conference (ISSCC) on Thursday. The chip, dubbed as 'Eyeriss', has been optimized for deep learning and to be embedded in smart phones. Eyeriss will allow mobile devices to perform natural language processing and facial recognition without any internet connection.

The technology involved, will enable smart phones, wearable or other IoT (Internet of Things) devices to go through deep learning processes. The Eyeriss chip promises neural networks in very low power devices. It consumes 10 times lesser power compared to the graphics processors of the present day's smart phones though containing 168 cores, according to a report published in Engadget.

Data collected by the smart phones, is currently uploaded to the internet for processing by power servers. After completion of processing, the data is then sent back to the device. The complex procedure opens up an array of issues, from latency and network congestion to security and power consumption, reports eWeek while narrating the existing technology.

Eyeriss avoids data swapping through quite a tricky manner. Each of the neuron like core has own memory and capable of compressing data. Nearby cores may communicate with each other and don't require to communicate with central source. Furthermore, a special delegation circuit provides cores as much work as they can handle without fetching the data back, reports The Verge.

The possibilities enabled by running neural networks locally-making decisions regarding the raw data on the devices and sending only their conclusions to the Internet-are broad. Local data processing also offers greater security and privacy and reduces transmission latencies. The Eyeriss devices run the data locally through the algorithms for analyzing instead of sending via internet.