Energy consumption has become a critical concern in a world increasingly driven by artificial intelligence (AI) and digital technology. Data centres, embedded systems, and AI-powered devices demand enormous amounts of power, raising sustainability challenges. Dr Amirreza Yousefzadeh, a researcher at the Faculty of Electrical Engineering, Mathematics and Computer Science (EEMCS), is tackling this issue by developing energy-efficient computing inspired by the human brain. His work can potentially reshape industries ranging from wearables to cloud computing.
The brain as a blueprint for energy-efficient computing
Yousefzadeh’s research focuses on neuromorphic computing—designing processors that mimic the brain’s ability to process signals with minimal energy consumption. “A human brain consumes around 10 watts of power while performing complex tasks like seeing, hearing, learning, and decision-making,” he explains. “Meanwhile, AI models and data centres require thousands of watts to perform similar tasks”.
By developing hardware that replicates biological efficiency, Yousefzadeh aims to revolutionise computing, making it more sustainable for future applications. His work directly aligns with the University of Twente’s role in the national Mission10X collaboration and the IPCEI CES project MISD, which focuses on modular, integrated, and sustainable data centres.
From smartwatches to AI data centres
One of the most immediate applications of Yousefzadeh’s work is in battery-powered devices. “Think about smartwatches and smart glasses,” he says. “A major limitation is battery life—many people stop using these devices because they do not last long enough. If we can significantly reduce processor power consumption, we can extend battery life and improve user experience.”
This research also has profound implications for hearing aids. Many people with hearing loss avoid using these devices because they amplify all sounds, including background noise. By integrating neuromorphic computing with smart glasses, Yousefzadeh envisions a system where users can focus on the voice of the person they are looking at while filtering out other noises.
Beyond personal devices, his work could help reduce the staggering energy consumption of AI systems. “Companies running AI models are even considering building nuclear power plants to sustain their energy needs,” he points out. “We need to rethink the way AI hardware is designed to make it more sustainable.”
Collaboration with industry and startups
Yousefzadeh actively collaborates with startups and industry partners to bring his research into real-world applications. He was part of GrAI Matter Labs, a startup acquired by Snapchat that is developing neuromorphic processors for augmented reality (AR) and virtual reality (VR) glasses. He also works with Innatera, a startup focused on ultra-low-power computing for the Internet of Things (IoT).
While large companies like Qualcomm and NXP are also investing in low-power processors, Yousefzadeh prefers working with startups. “They have a strong motivation to innovate, and they are more willing to explore unconventional solutions,” he says.
Barriers in the field of low-power computing
Despite the promising applications, the field of energy-efficient computing faces several significant challenges. A major issue is technological limitations—most current chip designs are built using existing market technologies, which are not always optimised for low-power applications.
Another challenge is the highly secretive nature of chip design. Unlike AI software, which benefits from open-source collaboration, chip design is dominated by a few large companies that restrict access to technology. “If I read a research paper from another group, I can’t replicate their work because the technology is proprietary,” Yousefzadeh explains. “This slows down innovation.”
However, there is hope for change. There are few recent open-source chip design initiatives making older chip manufacturing technology available to researchers. “This is an exciting development. It allows us to share and improve chip designs collaboratively, much like how open-source software has driven progress in AI,” he notes.
The future of energy-efficient computing at UT
Within the EEMCS faculty at UT, collaboration is key. Yousefzadeh emphasises that energy-efficient computing requires a cross-layer approach, involving everything from application algorithms to chip architecture and circuit design. His research is part of a larger effort within UT to improve chip design capabilities and support sustainable computing initiatives.
Looking ahead, Yousefzadeh is focused on:
- Developing neuromorphic hearing aids integrated with smart glasses for better speech recognition.
- Exploring new memory technologies to reduce power consumption in AI chips.
- Using neuromorphic computing to enhance chip reliability, especially in aerospace applications where radiation exposure can cause errors.
- Advancing open-source chip design to improve innovation and accessibility.
A call for collaboration
For researchers and industry professionals in the EEMCS faculty, Yousefzadeh’s work highlights the importance of interdisciplinary collaboration. “We need to optimise computing across all layers, from hardware to software,” he says. “By working together, we can make computing more sustainable and pave the way for the next generation of energy-efficient AI and digital technology.”
As AI and digital technology continue to expand, Yousefzadeh’s work at UT and its partners makes sure that the future of computing is not only powerful but also smart, efficient, and sustainable.
Amirreza Yousefzadeh is an assistant professor in the CAES research group and has been appointed to a sector plan 2 position as part of the sustainable computing theme.
This interview is part of a series of articles about EEMCS topics. Every two weeks, a new article will be published to highlight topics within our faculty. If you have any suggestions for this series, please contact mc-eemcs@utwente.nl.
