Prof. Francesco Morichetti, from the Department of Electronics, Information and Bioengineering – Politecnico di Milano, is among the lead contributors to a study published in Nature that opens new perspectives for the future of artificial intelligence. The research, conducted alongside international partners including École Polytechnique Fédérale de Lausanne, Stanford University, University of Cambridge, and the Max Planck Institute, addresses one of the most pressing challenges: the growing demand for energy and computational power generated by increasingly complex AI models, which is unsustainable with traditional digital systems.
To overcome these limitations, the researchers have developed physical neural networks that process information by directly leveraging the laws of physics. In particular, Politecnico di Milano has created integrated photonic chips capable of performing mathematical operations through the interference of light on silicon microchips only a few millimetres in size. This technology drastically reduces energy consumption and processing time, paving the way for applications in strategic fields such as autonomous vehicles and wearable sensors.
A crucial aspect concerns the training of these networks, which cannot follow conventional digital models. The team introduced a fully optical, “in-situ” training technique, making the process faster, more robust, and more efficient. By eliminating digitization, the photonic chips designed at Politecnico enable faster and more energy-efficient computations, opening the door to intelligent devices capable of operating in real time.
These results are part of a broader research effort exploring complementary approaches—photonic, analog electronic, and quantum—to achieve more powerful hybrid systems that better meet human needs. This transformation marks a significant step forward for the evolution of artificial intelligence, with Politecnico di Milano at the forefront.