Reviewed by Lexie CornerJan 16 2025
A research team from the University of California, Santa Barbara, has developed a photonic platform for in-memory computing that addresses challenges such as low switching speeds and limited programmability. The findings were published in Nature Photonics.
Artist's concept illustration of a photonic memory array. Image Credit: Brian Long
Over decades, the circuits powering computers and smartphones have consistently reduced in size while improving in performance. However, Moore's Law is approaching its physical limits, as the maximum number of transistors on a chip and the heat generated by densely packed transistors constrain further performance gains.
This plateau in computing capacity poses challenges for data-intensive applications like machine learning and artificial intelligence, which demand ever-increasing processing power. Addressing this requires new technologies. Photonics, which offers lower energy consumption and latency compared to electronics, is a promising solution.
Among the most promising approaches is in-memory computing, which relies on photonic memories to execute operations rapidly by transmitting light signals. However, previous attempts to develop such memories have been limited by low switching speeds and restricted programmability.
A collaborative effort led by Paolo Pintus, Assistant Professor at the University of Cagliari, alongside John Bowers, Professor of Electrical and Computer Engineering (ECE) at UC Santa Barbara, and Galan Moody, Associate Professor of ECE at UC Santa Barbara, has advanced this field. The project also involved Nathan Youngblood from the University of Pittsburgh, Yuya Shoji from the Institute of Science Tokyo, and Mario Dumont, a Ph.D. graduate from Bowers' lab.
The researchers used cerium-substituted yttrium iron garnet (YIG), a magneto-optical material whose optical properties are dynamically influenced by external magnetic fields. Using tiny magnets to store information and control light propagation, the team developed a new class of magneto-optical memories.
This platform uses light to perform computations significantly faster and more efficiently than conventional electronics. The new memory achieves switching speeds 100 times faster than the most advanced photonic integrated technologies, is reprogrammable for various tasks, and consumes roughly one-tenth of the power.
Additionally, the team demonstrated that magneto-optical memories can be rewritten over 2.3 billion times, suggesting an effectively infinite lifespan. In contrast, current optical memories can typically endure only up to 1,000 write cycles.
These unique magneto-optical materials make it possible to use an external magnetic field to control the propagation of light through them. In this project, we use an electrical current to program micro-magnets and store data. The magnets control the propagation of light within the Ce: YIG material, allowing us to perform complex operations, such as matrix-vector multiplication, which lies at the core of any neural network.
Paolo Pintus, Project Scientist, University of California, Santa Barbara
The authors suggest that the results represent a significant advancement in optical computing and could enable practical applications in the near future.
Journal Reference:
Pintus, P., et al. (2024) Integrated non-reciprocal magneto-optics with ultra-high endurance for photonic in-memory computing. Nature Photonics. doi.org/10.1038/s41566-024-01549-1.