Final Report Summary - COGNET (Cognitive Networks for Intelligent Materials and Devices)
This project provided a fundamental understanding of the opportunities and limitations of networks comprised of nanowires. It measured for the first time the distribution of junction resistances achievable under different processing conditions. It developed a model that described the ultimate performance of networks as conductors and showed (in contrast to the current thinking) that junctions were not always the limiting factor, and providing a criterion that allowed users to test whether their networks were fully optimised.
It showed that networks are arrays of connected interacting junctions that display dynamic and emergent properties as the current is routed across the network, leading to fault tolerant and programmable behaviours. It demonstrated for the first time the existence of a unique scaling relationship between the conductance of an individual junction with the current that passes through it and that this very same relationship (with the same scaling exponent) holds for a network comprised of the same junctions. Crucially it elucidated the importance of this scaling exponent, how it relates to the properties of the junctions themselves and how it impacts on whether the network actives uniformly or selects a particular winner-takes-all path. This is the first proof that junction engineering can control the connectivity within the network. It also showed that, depending on the junction properties, networks are capable of supporting multiple current paths, and that these connectivity paths may form the basis for neuromorphic computing platform.
The project also showed that controlled junctions with individual wires can be exploited to create multilevel memory devices, capable of responding to both electrical and light stimuli, that exhibit neuromorphic STDP learning behaviour similar to that found in biological neurons. Coaxial engineered wires were shown to exhibit both unipolar and bipolar switching behaviours, and the transition between the two behaviours is controlled by the set current. Moreover, the lifetime of memory state associated with these conductance levels can be sensory, short-term or long-term, depending on the input current.
The project developed strategies for the complete elimination of junctions and the fabrication of seamless networks utilising planar films or 2D materials. This enabled the fabrication of flexible transparent photodetectors based on engineered silicon-on-insulator device layers with the incorporation of porous silicon inclusions for enhanced light absorption. The silicon device layer when transferred to a flexible substrate has exceptional performance as a photodetector. A further development allowed the fabrication of a new generation of transparent conductors that have yielded exceptional performance, with films based on aluminium showing performance that surpasses the best silver based materials reported to date. Aluminium’s lower cost and superior corrosion resistance makes it the material of choice. A particular advantage of this approach is the ability to decouple the materials’ optical behaviour from the electrical performance.
It showed that networks are arrays of connected interacting junctions that display dynamic and emergent properties as the current is routed across the network, leading to fault tolerant and programmable behaviours. It demonstrated for the first time the existence of a unique scaling relationship between the conductance of an individual junction with the current that passes through it and that this very same relationship (with the same scaling exponent) holds for a network comprised of the same junctions. Crucially it elucidated the importance of this scaling exponent, how it relates to the properties of the junctions themselves and how it impacts on whether the network actives uniformly or selects a particular winner-takes-all path. This is the first proof that junction engineering can control the connectivity within the network. It also showed that, depending on the junction properties, networks are capable of supporting multiple current paths, and that these connectivity paths may form the basis for neuromorphic computing platform.
The project also showed that controlled junctions with individual wires can be exploited to create multilevel memory devices, capable of responding to both electrical and light stimuli, that exhibit neuromorphic STDP learning behaviour similar to that found in biological neurons. Coaxial engineered wires were shown to exhibit both unipolar and bipolar switching behaviours, and the transition between the two behaviours is controlled by the set current. Moreover, the lifetime of memory state associated with these conductance levels can be sensory, short-term or long-term, depending on the input current.
The project developed strategies for the complete elimination of junctions and the fabrication of seamless networks utilising planar films or 2D materials. This enabled the fabrication of flexible transparent photodetectors based on engineered silicon-on-insulator device layers with the incorporation of porous silicon inclusions for enhanced light absorption. The silicon device layer when transferred to a flexible substrate has exceptional performance as a photodetector. A further development allowed the fabrication of a new generation of transparent conductors that have yielded exceptional performance, with films based on aluminium showing performance that surpasses the best silver based materials reported to date. Aluminium’s lower cost and superior corrosion resistance makes it the material of choice. A particular advantage of this approach is the ability to decouple the materials’ optical behaviour from the electrical performance.