HBP and Intel to reveal new neuromorphic chip architectures

    02 February 2018

    Three new neuromorphic chip architectures to be revealed at NICE 2018, in Portland, US.

    Researchers from the Human Brain Project, along with colleagues from Intel in the US, will jointly reveal three new neuromorphic chips during the NICE 2018 conference hosted by Intel Corporation at the Jones Farm campus in Hillsboro, Oregon, February 27 to March 1.

    Neuromorphic chips mimic important aspects of biological brains like their energy efficiency, resilience and in particular their ability to learn. They promise to have a major impact on the future of artificial intelligence.

    The new BrainScaleS-2 chip has been developed by researchers in the Human Brain Project. A prototype will be presented by Johannes Schemmel from Heidelberg University (Germany). BrainScaleS is a mixed analog-digital design operating 1,000 – 10,000 times faster than real-time. The new chip features programmable on-chip learning capabilities and a new concept called dendritic computation developed in close collaboration with neuroscientists.

    Sebastian Höppner from TU Dresden (Germany) will present a prototype of the new SpiNNaker-2 chip. This work is also supported by the Human Brain Project. SpiNNaker-2 builds on the many-core architecture developed by Steve Furber from the University of Manchester (UK). It will feature 144 ARM Cortex M4F cores on a single chip with a very advanced power management delivering a computational power of 36 Billion instructions per second per Watt for a real-time simulation of complex neural networks.

    Intel Corporation will showcase their recently announced Loihi research chip with technical talks, posters, and several live demonstrations of the chip. Manufactured in Intel’s 14nm process technology, Loihi provides an unprecedented level of integration density and performance while supporting an advanced spiking neural network feature set that includes microcode-programmable learning rules. Loihi supports a range of on-chip learning models spanning supervised to unsupervised to reinforcement-based. Intel will launch a research partnership program at this event targeted at groups interested in using Loihi for neuromorphic algorithm and application development.

    All three new chip architectures will be demonstrated at the conference and experts will be available to discuss features and applications with prospective users.

    Several other presentations at the NICE 2018 conference will focus on theories, algorithms and applications of Neuromorphic chips.

    More on NICE 2018