Service Communautaire d'Information sur la Recherche et le Développement - CORDIS


COGNET Résumé de rapport

Project ID: 321160
Financé au titre de: FP7-IDEAS-ERC
Pays: Ireland

Mid-Term Report Summary - COGNET (Cognitive Networks for Intelligent Materials and Devices)

Materials as we understand them have well defined electrical and optical properties; properties that are ultimately exploited in the fabrication of sensors and devices of all kinds. Much effort is expended worldwide in an on-going search for new materials to enable next generation products. COGNET introduces a new class of materials that do not have specific properties per se. Rather, these materials exhibit behaviours that can be evolved to establish desired properties. This is accomplished by exposing a random network of nanoscale wires to electrical, optical and chemical stimuli. The properties of any network are determined by the strength of the junction connections. By controlling the passivation or chemical functionalization along these wires it is possible to engineer the composition of junctions within the network, which can then be turned ON and OFF in response to the applied stimuli. The objective is to create network materials with arbitrarily controlled properties for a wide range of smart material, sensor and memory applications. This approach will also enable the development of intelligent networks capable of learning that will form the basis for nanoscale neuromorphic devices that mimic aspects of the operation of the human brain.

Progress to date includes for the first time a real understanding as to how conduction pathways in a network become activated and how they are distributed within a network. We now understand what is required to turn the whole network ON so as to make a material with the maximal connectivity and hence highest conductivity. We can also change the level of connectivity to control the conductivity. We have yet to achieve the ability to reversibly modify the connectivity.

We have also demonstrated the ability to encode multiple levels of memory into a single junction. Instead of having just the usual binary 0 or 1 levels available, these junctions can encode upwards of 10 levels, thereby potentially decreasing the time needed for computation. We have also shown that these junctions respond to correlated electrical and optical inputs. That is the junction response to random electrical and optical pulses is low, but there is an anomalously large response whenever the input pulses occur at the same time. Moveover repeated exposure to simultaneous pulses produces a growing response that is similar to the learning response found in biological systems. We are currently exploring the neuromorphic applications of these materials.


Deirdre Savage, (Research Accounting Manager)
Tél.: +35318961942
Fax: +35317071633