Nanowires-based Device for the Implementation of a Reservoir for a Neural Network
Data di pubblicazione19-04-2022
Data di priorità27-07-2021
FaseItalian Patent Application
TitolarePolitecnico di Milano, Politecnico di Torino, Istituto Nazionale di Ricerca Metrologica
DipartimentoDepartment of Electronics, Information and Bioengineering
AutoriIelmini Daniele, Pedretti Giacomo, Ricciardi Carlo, Montano Kevin, Milano Gianluca.
In the context of computing devices, and more in details in the sector of neural networks, the ‘reservoir computing’ paradigm is well known: it is based on the separation between a dynamic part (reservoir) and a non-recurrent part (readout) which needs to be trained. The main problem of reservoir computing systems is the difficulty in adapting to different computationl tasks and the necessity to train several parameters.
The invention provides a device for the implementation of a reservoir for a neural network which is constituted by a network of nanowires arranged on a substrate and a plaurality of electrodes. This device allows to minimize the number of parameters to train in the artificial neural network.