UOC researchers develop a low-power, high-performance AI model

Artificial intelligence cannot exist without energy: a data centre dedicated exclusively to AI products and services currently consumes as much electricity as 100,000 homes, according to the International Energy Agency (IEA). In recent years, artificial intelligence has extended beyond the world of research into many areas of our lives, leading to a significant rise in demand for energy. According to the IEA, data centres currently consume 1.5% of all electricity produced globally and, if nothing changes, their energy demands will have doubled by the end of the decade.

Seeking to counteract this trend and reduce the AI energy footprint, two studies from the Universitat Oberta de Catalunya (UOC), involving researchers Fernando Sevilla Martínez and Laia Subirats Maté from the Cognitive Neuroscience and Applied Data Science Lab (NeuroADaS Lab) group, propose two approaches for more sustainable, efficient and affordable AI model. The articles have been published in open access in IEEE Networking Letters and the International Journal of Intelligent Systems.

"Energy efficiency must become a key parameter in AI design. It’s not just about making models that are faster or perform better; it's also a question of making them sustainable, ethical and accessible", said Subirats Maté, who is also an associate professor in the UOC's Faculty of Computer Science, Multimedia and Telecommunications. "Designing energy-efficient AI models not only benefits the planet, but also makes it possible to deploy AI in small devices such as robots and sensors, reduce operating costs for businesses and data centres, and improve resilience in environments with limited connectivity or energy", she added.

More environmentally-friendly neural networks

The first study published (led by UOC doctoral candidate Fernando Sevilla Martínez, and with participation from the Universitat Autònoma de Barcelona, the Computer Vision Center (CVC/UAB), the Centre Tecnològic de Telecomunicacions de Catalunya and the Volkswagen group) demonstrated that low-power, high-performance spiking neural networks (a type of AI that mimics the functioning of the human brain) can be developed using inexpensive and accessible components such as Raspberry Pi 5 and the BrainChip Akida accelerator. This study paves the way for energy-efficient, distributed artificial intelligence networks that can be applied to fields such as transport, environmental monitoring and the industrial internet of things (IoT).

"The methodology we propose enables these spiking neural network models to be trained, converted and run without a graphics processing unit or being connected to a data centre or the cloud, as they consume less than 10 watts of energy," the authors explained. "Furthermore, thanks to other technologies such as Message Queue Telemetry Transport, Secure Shell and Vehicle-to-Everything communication, various devices can collaborate in real time, sharing results in less than a millisecond with energy expenditure of just 10 to 30 microjoules per operation."

According to the researchers, this will have implications for both energy consumption and social and ethical aspects, as it will make AI accessible to anyone and strengthen data privacy. This will allow schools, hospitals, rural areas with limited infrastructure, and groups of citizens with limited resources to use efficient, sustainable, accessible and distributed artificial intelligence.

Towards efficient autonomous driving

The second study, also led by Fernando Sevilla Martínez from the UOC and with the same participants as the previous project, analysed in detail how spiking neural networks can reduce the energy consumption of autonomous driving systems, compared to convolutional networks, which are widely used in artificial vision systems, such as those found in some autonomous vehicles. To do this, they compared the two technologies in tasks such as predicting steering wheel turning angles and detecting obstacles. The researchers' proposal also involves introducing a new way of measuring the real efficiency of systems, thus achieving a better balance between accuracy and energy consumption.

"The tests we carried out with different architectures show that spiking neural networks with specific encoding achieve an optimal balance between performance and low consumption, using 10-20 times less energy than convolutional networks," said the researchers from the UOC's NeuroADaS Lab group, part of the eHealth Centre. "This shows that neural networks can drive more sustainable AI models even without the need for specialized hardware, marking a key milestone in the development of efficient computing in intelligent and autonomous transportation," they added.

According to the authors, both studies provide valuable data for research into AI systems that consume less energy and are therefore more affordable and more accessible. "The first study provides a practical workflow with lower electricity demand, less heat generation and the possibility of deploying AI directly without data centres – what is known as edge computing," they concluded. "Meanwhile, the second introduces a metric that combines performance and energy consumption, thus enabling the design of more sustainable AI."

Related articles
Martínez, F.S., Casas-Roma, J., Subirats, L., & Parada, R. (2025) "Eco-Efficient Deployment of Spiking Neural Networks on Low-Cost Edge Hardware". IEEE Networking Letters https://doi.org/10.1109/LNET.2025.3611426.

Sevilla Martínez, F., Casas-Roma, J., Subirats, L., & Parada, R. (2025). "Energy-aware regression in spiking neural networks for autonomous driving: A comparative study with convolutional networks". International Journal of Intelligent Systems https://doi.org/10.1155/int/4879993.

Both projects, part of the UOC's Digital health and planetary well-being research mission, support UN Sustainable Development Goals (SDGs) numbers 9, Industry, Innovation and Infrastructure; 11, Sustainable Cities and Communities; and 13, Climate Action.

Transformative, impactful research

At the UOC, we see research as a strategic tool to advance towards a future society that is more critical, responsible and nonconformist. With this vision, we conduct applied research that's interdisciplinary and linked to the most important social, technological and educational challenges.

The UOC’s over 500 researchers and more than 50 research groups are working in five research units focusing on five missions: lifelong learning; ethical and human-centred technology; digital transition and sustainability; culture for a critical society, and digital health and planetary well-being.

The university's Hubbik platform fosters knowledge transfer and entrepreneurship in the UOC community.

More information: www.uoc.edu/en/research