Today's neuromorphic community in Europe is leading the state of the art in this domain. The community counts an increasing number of labs that work on theory, modelling, and implementation of neuromorphic computing systems. The NCT includes both digital and analogue VLSI technologies, memristive devices, photonics, spintronics, magnonics, nanotubes, and other nano-technological solutions. The NEUROTECH Project will create a Cartography of NCT in Europe, explore Roadmaps for its development, collect educational and information resources on NCT, inform about upcoming and previous events.
To enable the uptake of this technology and to match the needs of real-world applications in future products that solve real-world tasks in industry, health-care, assistive systems, robotics, and consumer devices, we aim to create a sustainable communication channel between research labs, industry, general public, and other stakeholders.
NEUROTECH project will assemble and curate a coherent collection of educational resources on all aspects of NCT, share core educational events on NCT, to disseminate NCT curriculum to other communities, and to promote NCT among decision makers.
NCT is computing and sensing hardware that is built on the principles of neural computation: event-based, massively parallel, with co-localised memory and computation.
NCT comprises hardware electronic systems that exploit the physics of their devices to directly emulate the properties of biological neurons and synapses, and uses them to build computational primitives that use the same organizing principles (in space and time) used by the brain.
For me, neuromorphic implies spikes. Historically, it is restricted to sub-threshold analogue, which of course excludes SpiNNaker. Today I think it is wider, and covers all brain-inspired computing models that use spikes for primary communication. But we do need to be inclusive, encourage convergence with ANNs and machine learning.
From the viewpoint of someone developing Neuromorphic algorithms, NCT implies massively parallel computing with large numbers of lightweight compute units, which communicate via tiny timed messages. Information is primarily encoded in the source, timing, and sequence of events, rather than included explicitly in complex messages. Information is mainly local to the compute units, and learning happens locally too, at the connections between those units. This happens to be similar to how spiking networks in the brain operate.
Designing and building hardware that is optimized to perform computing tasks that aim to mimic cognitive functions of the brain. This hardware is designed by taking inspiration from the brain both at the architecture and at the components functionalities level. It includes features such as: colocation of memory and computing, high connectivity between elements, the use of stochasticity, massive parallelism, the use of devices and materials which physics directly implement the behavior of brain cells (the non-linearity of neurons and their spiking behavior, the non-volatility and plasticity of synapses).
For IBM, all computing paradigms that are in some way brain-inspired fall under NCT. So, this includes artificial neural networks, spiking neural networks, feedforward neural networks, recurrent neural networks but also other architectures like reservoir computing etc. As NEUROTECH is a CSA network, we must make sure to include the larger community.