The BICS research group explores the computational capabilities of networks of spiking neurons using biologically inspired electronic systems. We develop real-time full-custom VLSI devices for building general purpose large-scale spiking neural networks as well as specialized systems for low-power compact solutions for practical applications.
Two main streams can be identified in our research activities: (i) neuromorphic systems and theories for brain inspired computation (learning, cortical inspired neural architectures); (ii) neuromorphic sensing and actuating.
Neuromorphic systems and theories for brain inspired computation
The neuromorphic systems and theories for brain inspired computation research stream focuses on investigating algorithms and analog hardware implementations of intelligent processing. We develop architectures and methods for embedding simple cognitive behaviors in real-time neuromorphic systems. Our research is grounded in both the biological and computational aspects of neuroscience, taking inspiration by cortical structures and working on models of cortical computation. We carry out this research both by means of classical neuromorphic approaches (design of novel circuits and architectures) and through the development of technologies (integration of novel nano-electronic “memristive” devices).
- Recurrent Competitive Networks
- Calcium-based plasticity learning model
- Neuromorphic circuits based on hybrid CMOS-Ferroelectric implementations (H2020 ICT BeFerroSynaptic)
- Neuromorphic Circuits for Novel Devices (H2020 MSCA-ITN MANIC)
Neuromorphic sensing and actuating
Within the neuromorphic sensing and actuating research stream we develop neuromorphic electronic circuits to build compact autonomous sensory and sensory-motor systems specialized for interacting with the environment and solving specific real-world tasks. We focus on invertebrates and lower vertebrates with highly specialized sensory and actuator systems (e.g. optic flow in flying insects) to bridge the gap between basic and applied computational neuroscience. The study of these animals offers the unique opportunity to aim at a neural network model comparable in size with the nervous system under investigation. Genetic tools and the limited number of computational stages offers the possibility to reveal more detailed knowledge of the biological system, when compared to higher vertebrates. Furthermore, given the high specialization for a well defined task, nature can offer optimized solutions for currently unsolved technological challenges which can lead to low-power adaptive artificial sensory-motor systems.
|Last modified:||02 February 2021 2.48 p.m.|