Kalray's Coolidge processor adds deep learning acceleration

May 24, 2017 // By Peter Clarke
Fabless many-core chip company Kalray SA (Grenoble, France) has announced a third generation of its massively parallel processor architecture (MPPA3) under the codename "Coolidge."

The chip is being designed with 80 or 160 64-bit processor cores plus 80 or 160 co-processors optimized to accelerate computer vision and deep learning and is being aimed at embedded applications with artificial intelligence such as self-driving vehicles. The MPPA Coolidge processor is being developed for implementation in a 16nm FinFET process and will be available in mid-2018, Kalray said.

The MPPA3 Coolidge follows on from Kalray's present-day processor, the MPPA2-256 Bostan processor, which is in use by automotive and aeronautic OEMs and their first-tier suppliers.

The Coolidge is expected to operate at a clock frequency of 1.2GHz and achieve peak performance capability of 5TFLOPS and 5TOPS while consuming less than 20W of power.

The MPPA3 Coolidge will be compatible with multiple interfaces – PCIe, Ethernet, DDR, CAN, USB and others. It will also contain security and safety blocks to keep the processor running safely and protect systems against cyber-attack.

Kalray is currently working with various customers to put the Coolidge at the heart of newly developed critical embedded applications.

Related links and articles:

www.kalray.eu

White paper

News articles:

Kalray turns to neural networks

ARM's soft launch for machine learning library

French parallel processing research wins €32m funding

Nvidia seeds education with deep learning teaching kit