Physical Neural Networks: Path to Sustainable AI

Published Date : 09/09/2025 

Explore how innovative physical neural networks, using photonic microchips, are revolutionizing artificial intelligence by reducing energy consumption and processing time. 

Artificial intelligence (AI) has become an integral part of our daily lives, powering everything from smart assistants to complex data analysis. However, the increasing demand for larger, more complex AI models is outpacing the computational capabilities of traditional computers. This has led researchers to explore innovative technologies, such as physical neural networks, which directly leverage the laws of physics to process information more efficiently.

The potential of physical neural networks is highlighted in a groundbreaking study published in the prestigious journal Nature. This research is the result of a collaboration between several leading international institutions, including the Politecnico di Milano, the École Polytechnique Fédérale in Lausanne, Stanford University, the University of Cambridge, and the Max Planck Institute.

The article, titled “Training of Physical Neural Networks,” delves into the development and training of these networks, with significant contributions from Professor Francesco Morichetti of the Department of Electronics, Information and Bioengineering (DEIB) at Politecnico di Milano. Morichetti leads the university’s Photonic Devices Lab and played a crucial role in developing photonic chips for the creation of neural networks.

These photonic chips utilize integrated photonic technologies to perform mathematical operations, such as sums and multiplications, through light interference mechanisms on silicon microchips that are only a few square millimetres in size. “By eliminating the operations required for the digitisation of information, our photonic chips allow calculations to be carried out with a significant reduction in both energy consumption and processing time,” explains Morichetti. This advancement is a crucial step towards making AI more sustainable, especially considering the energy-intensive nature of data centres.

The study also addresses the training phase of physical neural networks, which is critical for the network to learn and perform specific tasks. “With our research within the Department of Electronics, Information and Bioengineering, we have helped develop an ‘in-situ’ training technique for photonic neural networks, i.e., without going through digital models. The procedure is carried out entirely using light signals. Hence, network training will not only be faster but also more robust and efficient,” adds Morichetti.

The use of photonic chips will enable the development of more sophisticated AI models and devices capable of processing real-time data directly on-site. For example, autonomous cars and intelligent sensors integrated into portable devices will be able to perform complex tasks without the need for remote processing. This not only enhances efficiency but also improves the reliability and responsiveness of AI applications.

In summary, the research on physical neural networks represents a significant leap forward in the field of AI. By harnessing the power of light and integrated photonic technologies, these networks offer a promising solution to the challenges of high energy consumption and slow processing times, paving the way for a more sustainable and efficient future for artificial intelligence. 

Frequently Asked Questions (FAQS): 

Q: What are physical neural networks?

A: Physical neural networks are analogue circuits that directly exploit the laws of physics, such as properties of light beams and quantum phenomena, to process information. They are designed to overcome the limitations of traditional digital computers in terms of energy consumption and processing speed.

Q: How do photonic chips contribute to sustainable AI?

A: Photonic chips use light interference mechanisms to perform mathematical operations, significantly reducing energy consumption and processing time. By eliminating the need for digitisation, these chips make AI more sustainable, especially in energy-intensive applications like data centres.

Q: What is the significance of the 'in-situ' training technique for photonic neural networks?

A: The 'in-situ' training technique allows the training of photonic neural networks using light signals, without the need for digital models. This makes the training process faster, more robust, and efficient, enhancing the performance and reliability of AI models.

Q: Which institutions were involved in the research on physical neural networks?

A: The research was a collaborative effort involving the Politecnico di Milano, the École Polytechnique Fédérale in Lausanne, Stanford University, the University of Cambridge, and the Max Planck Institute.

Q: What are some potential applications of photonic neural networks?

A: Photonic neural networks have the potential to be used in various applications, including autonomous cars, intelligent sensors in portable devices, and real-time data processing. These applications can benefit from the reduced energy consumption and faster processing times provided by photonic chips. 

More Related Topics :