All the latest quantum computer articles

See the latest stories on quantum computing from eeNews Europe

Friday, December 21, 2018

Production-ready AI chip has embedded MRAM magnetic memory

By Nick Flaherty www.flaherty.co.uk

Gyrfalcon Technology Inc. (GTI) has launched the first production chip for machine learning with embedded MRAM non-volatile magnetic memory.

The 22nm Lightspeeur 2802M is aimed at IoT endpoints, cloud solutions, autonomous vehicles and many other use cases in edge AI. It has 40MB of memory which can support large AI models or multiple AI models. Examples of models that could be supported within the same chip are image classification, voice identification, voice commands, facial recognition, pattern recognition and many others.

“Our proprietary and patented MRAM engine allowed us to bring this first of our company’s many planned MRAM-enabled AI Accelerators to market, to open up new use cases and drive the success of our customers’ product designs,” said Terry Torng, GTI Co-Founder and SVP of Engineering.

The 2802M is the first in a portfolio of MRAM chips in the company’s roadmap using the MPE and APiM technologies. MPE (Matrix Processing Engine) uses CNN with the APiM (AI Processing in Memory), pairing logic with memory like a real human brain. This optimises the speed of processing, achieving high TOPS performance, while also saving tremendous amounts of power by avoiding management of data in discrete memory components.

The MRAM line of chips will use the SDK and the recently announced PLAI Builder from other of the company's AI chips, which simplifies AI Model creation for development teams less familiar with developing solutions using artificial intelligence. AI Models created with Caffe, TensorFlow and other tools can be supported, enabling a wide variety of neural networks, including VGG, MobileNet, ResNet, SDD, ImageNet and others.

“We are pleased to work with an innovative company like Gyrfalcon Technology to create further growth in the AI space,” said Brad Paulsen, Senior Vice President of TSMC North America. “TSMC’s embedded MRAM process enables unique product differentiation in both the Automotive and AIoT markets.”

www.gyrfalcontech.ai.

No comments: