The growing energy consumption of data centers for artificial intelligence (AI) is becoming an increasingly acute problem. A solution, according to an international group of researchers, may lie in creating computing systems that work on the principle of the human brain.
Scientists from the University of Ottawa, Columbia University and Hanyang University published in the journal Advanced Functional Materials a systematic review of advances in electronic materials, device physics and circuit integration. Their goal is to chart paths toward creating neuromorphic systems, that is, hardware that physically mimics information processing in the brain. The authors did not conduct a separate experiment but assessed the entire landscape of emerging hardware neurons and presented specific technological roadmaps.
"The materials and architectures described in the review represent a fundamentally different way of processing information. Working more like a human brain than a traditional processor, these technologies will lead to devices that are significantly more energy-efficient, faster and more adaptive," explained lead author Dr. Kannan Udaya Mohan, a postdoctoral fellow in the research group of Professor Chang-Hyun Kim at the University of Ottawa.
Traditional computing, according to Professor Kim, runs into a "memory wall": the transfer of data between memory and processor causes delays and energy losses. The researchers identified the most promising post-CMOS materials and device architectures capable of performing the role of artificial neurons. CMOS is the technology on which most modern microchips are built; post-CMOS means new materials and designs that go beyond it. The authors argue that such circuits can process information faster and with a fraction of the energy consumed by today's mass-market chips.
In addition to reducing the energy footprint of data centers, the technology also has potential for personal electronics. New materials are often cheaper, easily scalable and make it possible to create ultrathin, lightweight and flexible devices. "In addition to green AI data centers, this hardware will directly enable smarter low-power wearable devices for personalized medicine and seamless human-machine interfaces," Kim added.
The review is a summary in nature and does not report the creation of a finished product. However, it offers the scientific community a guide for developing scalable brain-like computing systems.