Seminar & Colloquium
[세미나: 11월 15일(금), 오전 10시 30분] Dr. Stefano Brivio, IMM (Institute for Microelectronics and Microsystems) of the National Research Council (CNR)
Title
Single-node reservoir computing through a memristive circuit with complex dynamics
Speaker
Dr. Stefano Brivio, staff researcher, IMM (Institute for Microelectronics and Microsystems) of the National Research Council (CNR), Unit of Agrate Brianza, Italy
* Biography
Stefano Brivio received the master’s degree in engineering physics and the Ph.D. in Physics from Politecnico di Milano, Italy, in 2006 and 2010, respectively. He has been staff researcher at CNR-IMM, Unit of Agrate Brianza, Italy since 2018. His current research interests include the development of novel oxide-based switching devices for neuromorphic and brain-inspired computing applications. He is author of more than 50 publications in international journals and proceedings.
| Date | Friday, November 15th, 2024
| Time | 10:30 ~
| Venue | 33동 226호
[Abstract]
The reservoir computing concept prescribes an ensemble of interacting dynamical objects to preprocess information, through their collective dynamics, in favor of the post-processing by a linear and easily trainable readout. The design, realization and optimization of physical reservoirs is a complex task that led to the proposal of single-node systems (mainly in the photonic realm) that, despite their compactness, show useful complex dynamics.
In this work, we realize an electronic single-node reservoir made of an oscillator circuit with nonlinear and tunable properties enabled by a nonvolatile memristor. Pt/HfO2/TiN devices are programmed through several resistance states featuring different nonlinear current-voltage characteristics required to generate complex dynamics in a circuit inspired from the Murali-Lakshmanan-Chua one.
We demonstrate the hardware system to be able to perform nonlinear classification tasks and extend the results through simulations toward spatial and temporal tasks with increased complexity. We also complement the reservoir system with the implementation of the readout layer through a memristor array performing analog vector-matrix multiplication.
The work is partially supported by the PRIN2017-MIUR project COSMO (Prot. 2017LSCR4K)
| Host | 장호원 교수(02-880-1720)