Reviewed by Lexie CornerOct 24 2024
An international team of researchers from Pittsburgh Swanson School of Engineering, Tokyo Institute of Technology, the University of California—Santa Barbara, and the University of Cagliari has created a novel approach to photonic in-memory computing. Their research was published in the journal Nature Photonics, marking a significant step toward the future of optical computing.
A concept image of the group's photonic in-memory computing. Image Credit: Brian Long, Senior Artist, UCSB
Nathan Youngblood, an Assistant Professor of Electrical and Computer Engineering at Pitt, Paulo Pintus, formerly of UC Santa Barbara and now an Assistant Professor at the University of Cagliari in Italy; and Yuya Shoji, an Associate Professor at the Institute of Science Tokyo in Japan, have worked together to coordinate this study.
Researchers' efforts to create photonic memory for AI processing have so far been constrained by the need to trade off one crucial feature, like speed, for another, like energy consumption.
The global team presents a novel approach in the study that overcomes the current drawbacks of optical memory, which have previously been unable to integrate non-volatility, multi-bit storage, fast switching speed, low switching energy, and high durability in a single platform.
The materials we use in developing these cells have been available for decades. However, they have primarily been used for static optical applications, such as on-chip isolators, rather than a platform for high-performance photonic memory. This discovery is a key enabling technology toward a faster, more efficient, and more scalable optical computing architecture that can be directly programmed with CMOS (complementary metal-oxide semiconductor) circuitry – which means it can be integrated into today’s computer technology.
Nathan Youngblood, Assistant Professor, Electrical and Computer Engineering, Pittsburgh Swanson School of Engineering
Youngblood said, “Additionally, our technology showed three orders of magnitude better endurance than other non-volatile approaches, with 2.4 billion switching cycles and nanosecond speeds.”
To create photonic in-memory computing, the authors propose a resonance-based photonic architecture that leverages the non-reciprocal phase shift in magneto-optical materials.
In photonic processing, multiplying a rapidly changing optical input vector by a matrix of constant optical weights is a common method. However, encoding these weights on-chip using conventional materials and techniques has proven challenging.
The researchers address this by using heterogeneously integrated cerium-substituted yttrium iron garnet (Ce:YIG) on silicon micro-ring resonators. These magneto-optic memory cells induce light to propagate in both directions, akin to sprinters running in opposite directions on a track.
Computing by Controlling the Speed of Light
It is like the wind is blowing against one sprinter while helping the other run faster. By applying a magnetic field to the memory cells, we can control the speed of light differently depending on whether the light is flowing clockwise or counterclockwise around the ring resonator. This provides an additional level of control not possible in more conventional non-magnetic materials.
Paulo Pintus, Assistant Professor, University of Cagliari
To accommodate more data for computing applications, the team is now working to scale up from a single memory cell to a large-scale memory array. According to the paper, the non-reciprocal magneto-optic memory cell offers an effective, non-volatile storage solution with the potential for infinite read/write endurance and programming speeds of less than a nanosecond.
We also believe that future advances of this technology could use different effects to improve the switching efficiency and that new fabrication techniques with materials other than Ce:YIG and more precise deposition can further advance the potential of non-reciprocal optical computing.
Yuya Shoji, Associate Professor, Institute of Science Tokyo
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.