Neuromorphic physics
In a nutshell
Neuromorphic physics combines brain-inspired methods with physics for efficient information processing. Using nanomaterials and technologies such as spintronics, superconductivity and photonics, it enhances the performance of classical and quantum neural networks. Physics comes into play both to perfect the devices that mimic neurons and synapses, and to improve the algorithms. The aim is to design energy-efficient devices that are able to analyse data and learning in real time.
As AI processors require more and more energy, our team is exploiting advances in nanotechnology to develop sustainable circuits using spintronic, superconducting and oxide materials. We are building neural networks that take advantage of the unique capabilities of these materials, with the aim of creating hybrid accelerators that surpass CMOS efficiency on a large scale.
Quantum neuromorphic computing
Quantum neuromorphic computing implements neural networks on quantum systems. These networks project data into a high-dimensional Hilbert space for efficient classification and can automatically process and recognize quantum states, reducing the need for classical measurements. In our team, we are exploring the properties of superconducting circuits to develop new learning methods for these networks.
Neuromorphic algorithms
We are developing new learning algorithms that combine the precision of mathematical methods with the hardware compatibility of brain-inspired techniques. By exploiting spintronics, oxides and quantum mechanics, we aim to create efficient and energy-saving hardware neural networks. Using the brain as an example, we are taking up the challenge of achieving high performance while managing the uncertainties of nanodevices.
Coupled nano-oscillators capable of recognizing vowels according to a learning rule
Physicists have succeeded in fabricating a network of four coupled nano-oscillators capable of recognizing spoken vowels by tuning their frequencies according to an automatic rule...
Neural-like computing with populations of superparamagnetic basis functions
In neuroscience, population coding theory demonstrates that neural assemblies can achieve fault-tolerant information processing. Mapped to nanoelectronics, this strategy could allow for reliable computing with...
