Recently a group of researchers developed a breakthrough method for photonic in-memory computing. This invention has marked a milestone in making optical computing a reality in the coming future. The group consists of electrical engineers and professors from the University of Pittsburgh, Swanson School of Engineering, University of California, University of Cagliari, and the Tokyo Institute of Technology.
The researchers published a study in Nature Photonics on October 23rd, titled “Integrated non-reciprocal magneto-optics with ultra-high endurance for photonic in-memory computing.” The findings have marked a significant movement in the books of optical computing.
To date, only limited features or they were only able to attribute one feature for further developing the photonic memory for AI processing. Mostly they had to trade off one essential feature for another one and this has contributed to the potential dealy for this long.
This will be the first time the researchers have come up with a complete solution that will be able to deal with all the existing problems and challenges faced by photonic in-memory computing. They have created one unique solution to solve problems like multibit storage, high switching speed, low-speed energy, and many more.
This has been a collective effort by the researchers to improve further and enhance photonic in-memory computing for the future. The researchers revealed that the materials they used to develop the solution have already been available for centuries.
But they have not been used proportionately and for the right purpose. These materials have already been in use for static optical applications. According to the research these materials were instead used for providing performance for the photonic in-memory computing.
Nathan Youngblood, Assistant Professor and one of the researchers from the team stated that the technological method developed by the team displayed three orders of magnitude with better performance and endurance than the non-volatile approaches that come with higher speed and 2.4 billion switching cycles.
In the article, the authors proposed a resonance-based photonic architecture that can be utilized to enhance and leverage the non-reciprocal phase shift in magneto-optical materials. Such an approach is proposed by the authors to further multiply a rapidly changing optical input vector with a matrix of fixed optical weights. According to the researchers, this conventional method has proven quite challenging.
As of now, the team is working collectively to upgrade from a single-scale memory to a large-scale memory to create more space for the data required for computer applications. Through the article, the researchers show that the new-reciprocal magneto-optic memory offers an effective and essential storage solution that provides programming speeds of less than a nanosecond and infinite reading or writing endurance.
Yuya Shoji, Associate Professor of the Tokyo Institute of Science has put forward the fact to the media that the new technology developed by the team will be upgraded for the future could improve the switching efficiency, and will further enhance the quality of non-reciprocal optical computing.
The findings by the group of potential researchers have already become a topic of discussion in the technological field considering the rapid uprise of optical computing. The researchers have noted down their findings in the article and this is a significant step toward creating a future for the advanced form of optical computing.
For many years researchers have faced limitations and restrictions in creating photonic memory for optical computing due to the challenge of trading one important feature like speed or maybe energy consumption. Yet now the team has succeeded in creating an all-in-one solution that has all the capacities to enhance and better the future of optical computing.
