Paper Details

Communicating Process Architectures (CPA)
 Title: SpiNNaker-2: a 10 Million Core Processor System for Machine Learning and Brain Simulation
 Conference: Communicating Process Architectures 2018
 Authors: Christian Mayr
VLSI-Systems and Neuromorphic Circuits, Technische Universitat Dresden
 Abstract: In the EU flagship HBP, TU Dresden and University of Manchester jointly develop the SpiNNaker-2 processor platform for brain simulation. Characteristics of the SpiNNaker-2 system are 144 ARM cores per chip (including various numerical accelerators), for overall 10 Mio cores with 5PFLOPS CPU performance and 0.6ExaOPS deep learning acceleration. The chips are implemented in a 22nm FDSOI technology, running as low as 0.4V with the usage of our Adaptive Body Biasing methodology. The SpiNNaker-2 system combines highly efficient machine learning, bio-inspired signal processing at millisecond latency and ultra-low energy operation. Thus, besides the main usage as a neuroscience platform, SpiNNaker-2 opens up completely new avenues in areas such as tactile internet, autonomous driving and industry 4.0. 

BibTeX Entry

Full paper