Artificial intelligence (AI) models are quickly evolving and becoming more complex. To run increasingly sophisticated AI models, computers, servers and large data centres require more powerful electronic processors, which in turn need more electrical power to operate, making AI more costly and taking a serious toll on the environment. This problem inspired Mahdi Chegini, a PhD candidate in electrical and computer engineering, to develop new hardware to more efficiently and sustainably support AI.
More powerful and sustainable AI hardware
Chegini has been passionate about microwave photonics for AI and optical computing since he began studying the subject three years ago. “I was drawn to this field because it offers a promising path toward making future AI systems both faster and more energy efficient.”
Microwave photonics use light and microwave signals to optimize the speed, efficiency and energy consumption of AI models. Chegini explored how to use light instead of electrical signals to efficiently perform the complex calculations used by AI systems. “[This research] tackles one of the most urgent challenges in artificial intelligence today: how to keep increasing computing power without dramatically increasing energy consumption...By designing optical systems that can carry out these calculations more efficiently, my research aims to help make AI hardware more powerful and more sustainable,” he explains.
He credits his research supervisor Professor Jianping Yao for supporting him, as well as CMC Microsystems, for their assistance in testing and fabricating the AI hardware.
A small device with a large impact
Chegini’s research focused on developing AI hardware that can optimize performance. Although he initially considered fibre optics and silicon photonics for this hardware, he later decided to focus on silicon photonics. “The problem with fibre optics is bulkiness. Silicon photonics are especially exciting because they allow us to bring the advantages of large optical systems onto a very small chip, only a few millimeters in size.”
The silicon chip that Chegini designed is known as the photonic integrated convolutional accelerator (PICA) chip. The PICA chip was manufactured in Singapore by Advanced Micro Foundry (AMF), a silicon photonics manufacturing company. The chip was designed specifically to perform convolution, one of the most energy-consuming types of AI calculations. These calculations are widely used for image, text and speech recognition.
Putting it to the test
After a year of design and fabrication, the PICA chip was ready for testing. Chegini reported that thanks to a new algorithm and system design, his chip was able to double AI efficiency, which significantly reduced its energy consumption. “The chip can already operate at speeds of hundreds of billions of operations per second, and it has the potential to reach trillions of operations per second, all while consuming less than a watt of power,” he explains. “This technology can significantly enhance the overall speed and energy efficiency of AI hardware.”
As more individuals, organizations and industries become dependent on AI, it’s important to responsibly manage the amount of power these systems consume. That’s why Chegini hopes to continue improving the PICA chip to increase its performance and reduce its energy expenditure. “In the long term, our goal is to create a photonic chip that can outperform today’s most advanced electronic processors.” Chegini’s research opens the door to the development of AI systems that are both practical and sustainable.
Chegini hopes to continue his engineering career in photonics, computing and hardware design. “I’d like to build a career in research and innovation, whether in academia, industry or a combination of both. I hope to contribute to breakthroughs that help shape the future of AI hardware.”
Photonics for AI, a uOttawa Engineering priority
Photonics for devices, networks and energy is one of the five primary research areas at uOttawa’s Faculty of Engineering. The University is dedicated to optimizing the use of artificial intelligence for a cleaner, greener future.
Mahdi Chegini was awarded first place in the photonics for devices, networks and energy category of the 2026 Engineering and Computer Science Poster Competition, held during Engineering Research Celebration Day. His poster, titled A Next-Generation Silicon Photonic Accelerator for Efficient AI Computing, captivated judges.
