Events - Event View

From Neural Interfaces to Computing Paradigms

Prof. Lorenzo Pavesi

University of Trento, Italy

Location: NS II, 1201

Date: May 9, 2025

Time: 3:30pm

Interest in Artificial Neural Networks (ANNs) has surged in recent years due to their remarkable versatility in addressing a wide range of complex problems. Today, ANNs are predominantly implemented on electronic platforms. While very-large-scale ANN models have been developed that surpass human performance in specific tasks, they do so at the cost of long training times and significant power consumption. A promising alternative lies in Photonic Neural Networks (PNNs), which leverage the inherent advantages of photonics—namely, high-speed processing, massively parallel transmission (enabled by Wavelength Division Multiplexing, WDM), and low power dissipation.

This seminar will explore a series of simple PNN prototypes implemented on a silicon photonics platform, illustrating the fundamental operating principles of silicon-based PNNs. Silicon photonics is particularly attractive due to its compatibility with electronic integration, enabling on-chip training and scalable fabrication of photonic circuits. A key example to be discussed is a basic optical neuron, in which multiple delayed versions of an optical input signal interfere before reaching the output port—demonstrating an essential mechanism for computation in photonic networks.

Finally, the hybrid approach that combines photonic neural networks with biological neurons will be discussed. It enables experiments in which optical signals are transduced into neuronal activity, activating specific neural circuits. This integration paves the way toward hybrid intelligence systems, opening new frontiers in neuromorphic computing and advanced neurological therapies.