TECH

Building brain-inspired semiconductor chips for smarter, greener data centres

With artificial intelligence (AI) data centres consuming electricity at a staggering and growing rate, identifying a sustainable computing solution to offset these massive community and environmental disruptors is proving critical.

Researchers from the University of Ottawa, Columbia University and Hanyang University believe building brain-inspired computing systems could provide the answer to this global conundrum, with small, fast, energy-efficient and scalable artificial neurons able to support the large-scale computing systems that are now regularly dotting our landscapes.

“The materials and architectures described in this review represent a fundamentally different way to process information. By operating more like a human brain than a traditional processor, these emerging technologies will lead to physical devices that are dramatically more energy-efficient, faster, and adaptive,” explained lead author Dr. Kannan Udaya Mohanan, a Postdoctoral Fellow in Professor Chang-Hyun Kim’s research group at uOttawa.

Building scalable brain-like systems...Professor Kim and his team performed a sweeping, systematic review of years of global advancements in electronic materials, device physics, and circuit integration to chart a new path for "neuromorphic" systems – systems that physically mimic the human brain's processing. Rather than focusing on a single experiment, they evaluated the entire landscape of emerging hardware neurons, distilling complex material science into concrete, real-world technological pathways to help researchers build scalable, brain-like computing systems.

These brain-inspired hardware findings offer a comprehensive, forward-looking roadmap for the scientific community just as the federal government provides new frameworks for data centre developments in Canada.

“Traditional computing is hitting a bottleneck known as the "memory wall," where transferring data between memory and processors causes massive delays and energy loss. We identified the most promising post-CMOS materials and device architectures that can act as artificial neurons, proving that next-generation circuits can process information faster and with a fraction of the energy used by current mainstream chips,” said Chang-Hyun Kim, an Associate Professor in the School of Electrical Engineering and Computer Science and head of the Semiconductor and Microelectronics Laboratory at uOttawa.

Enabling greener technology...Post-CMOS refers to emerging materials and device architectures that move beyond traditional chips. While these brain-inspired technologies can drastically reduce the energy footprint of modern data centres, they offer an equally powerful advantage for personal technology. This is because these novel materials are often lower cost, highly scalable, and support ultra-thin, lightweight and flexible form factors.

“Beyond green AI data centres, this hardware will directly enable smarter, low-power wearable technology for personalized healthcare and seamless human-machine interfaces,” added Kim.

The development of brain-inspired semiconductor chips, known as neuromorphic chips, has emerged as the primary solution for creating smarter and more sustainable data centers. As the adoption of artificial intelligence grows exponentially, traditional computing architecture faces what is known as the "memory wall" bottleneck—a situation where the constant transfer of data between the processor and memory leads to sluggish performance and massive energy waste. In contrast, the human brain is capable of performing incredibly complex tasks while operating with extreme efficiency, consuming only about 20 watts of power.

These new technologies overcome this barrier by unifying processing and storage within a single component, mimicking the behavior of biological neurons and synapses.

Key advancements in neuromorphic technology:

• Post-CMOS materials: Scientists at the University of Ottawa (uOttawa) have identified new materials and circuit architectures capable of functioning as artificial neurons, requiring only a fraction of the electrical power currently consumed.

• Hafnium memristors: Research from the University of Cambridge has led to the development of a stable form of hafnium oxide that operates with currents one million times lower than previous alternatives, with the potential to reduce energy consumption in AI applications by up to 70%.

• Extreme efficiency: Chips based on materials physics and spiking neural networks can process temporal data directly within the hardware, achieving up to 2,000 times greater energy efficiency for specific tasks.

• Biological computing ("wetware"): In an even more radical approach, partnerships involving the startup Cortical Labs have created systems that integrate living, lab-grown human neurons with traditional silicon chips, giving rise to the concept of "biological data centers." Impact beyond data centers... While the immediate focus is on mitigating the sustainability crisis facing large-scale global servers, brain-inspired chips are highly scalable, lightweight, and low-cost. This will enable local, decentralized processing—or "edge computing"—bringing advanced intelligence directly to wearable health devices, smart sensors, and human-machine interfaces, without any reliance on cloud computing.

Would you like to explore any specific aspect of this technology further? I can provide details on how memristors function, explain the differences between spiking neural networks (SNNs) and traditional AI, or share data regarding the implementation of systems incorporating living neurons ("wetware").

University of Ottawa, Columbia University and Hanyang University