For many years, there has been consistent advancement in miniaturizing and enhancing the performance of the circuits that drive computers and smartphones. However, the trend known as Moore’s Law is approaching its limits due to physical constraints—like the maximum number of transistors that can be integrated onto a chip and the heat generated from increasingly dense arrangements—which are hindering the speed of performance enhancements. As a result, computing power is reaching a plateau, even though fields like artificial intelligence, machine learning, and other data-heavy applications are pushing for more computational capabilities.
For many years, there has been consistent advancement in miniaturizing and enhancing the performance of the circuits that drive computers and smartphones. However, the trend known as Moore’s Law is approaching its limits due to physical constraints—like the maximum number of transistors that can be integrated onto a chip and the heat generated from increasingly dense arrangements—which are hindering the speed of performance enhancements. As a result, computing power is reaching a plateau, even though fields like artificial intelligence, machine learning, and other data-heavy applications are pushing for more computational capabilities.
To tackle this issue, innovative technologies are essential. A potential avenue is photonics, which can offer improved energy efficiency and lower latency compared to traditional electronics.
One of the most exciting methods involves in-memory computing, which relies on photonic memories. By transmitting light signals through these memories, it becomes feasible to execute operations almost instantaneously. However, earlier proposals for developing these memories encountered obstacles like slow switching speeds and restricted programmability.
Now, a collaborative team of researchers has created a pioneering photonic platform to address these challenges, with their results published in the journal Nature Photonics.
In collaboration with UC Santa Barbara’s electrical and computer engineering professor John Bowers and associate professor Galan Moody, project scientist Paolo Pintus, an assistant professor at the University of Cagliari, led the effort together with Nathan Youngblood from the University of Pittsburgh, Yuya Shoji from the Institute of Science Tokyo, and Mario Dumont, who completed his Ph.D. in Bowers’ lab.
The team utilized a magneto-optical material known as cerium-substituted yttrium iron garnet (YIG), which has optical characteristics that change dynamically under external magnetic fields. By integrating tiny magnets to save data and regulate light travel within the material, they introduced a new category of magneto-optical memories. This advanced platform uses light to execute calculations at significantly faster rates and much greater efficiency than what traditional electronics can offer.
This novel memory type boasts switching speeds that are 100 times faster than the current leading photonic integrated technologies. They consume around one-tenth of the energy and can be reprogrammed numerous times for different tasks. While existing state-of-the-art optical memories are limited in lifespan, allowing for only about 1,000 write cycles, the team has shown that magneto-optical memories can be rewritten over 2.3 billion times, suggesting a practically unlimited lifespan.
“These distinctive magneto-optical materials enable the manipulation of light propagation using an external magnetic field,” Pintus explained. “In our project, we employ an electric current to program micro-magnets for data storage. The magnets govern the light’s journey within the Ce:YIG material, facilitating complex tasks like matrix-vector multiplication, which is fundamental to any neural network.”
The authors are optimistic that these discoveries could initiate a transformative phase in optical computing, laying the groundwork for real-world applications in the near future.
