Research Bits: Nov. 5

Optical in-memory computing

Researchers from the University of Pittsburgh, University of California Santa Barbara, University of Cagliari, and Institute of Science Tokyo propose a resonance-based photonic architecture which leverages the non-reciprocal phase shift in magneto-optical materials to implement photonic in-memory computing.

“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,” said Nathan Youngblood, assistant professor of electrical and computer engineering at Pitt, in a press release. “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 circuitry – which means it can be integrated into today’s computer technology. Additionally, our technology showed three orders of magnitude better endurance than other non-volatile approaches, with 2.4 billion switching cycles and nanosecond speeds.”

The magneto-optic memory cells are comprised of heterogeneously integrated cerium-substituted yttrium iron garnet (Ce:YIG) on silicon micro-ring resonators, which cause light to propagate bidirectionally. In a press release, Paulo Pintus, assistant professor at the University of Cagliari, likened the effect to sprinters running opposite directions on a track. “It’s 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.”

The team is working to scale up from a single memory cell to a large-scale memory array. [1]

Diffraction casting

Researchers from the University of Tokyo propose diffraction casting as a way to perform optical parallel computing for image processing or machine learning.

Ryoichi Horisaki, associate professor at the Information Photonics Lab at the University of Tokyo, said in a release that the method was inspired by shadow casting, an earlier optical computing technique. “Shadow casting is based on light rays interacting with different geometries, whereas diffraction casting is based on properties of the light wave itself, which results in more spatially efficient, functionally flexible optical elements that are extensible in ways you’d expect and require for a universal computer. We ran numerical simulations which yielded very positive results, using small 16-by-16 pixel black-and-white images as inputs, smaller than icons on a smartphone screen.”

The all-optical system takes an image as a source of data and combines it with a series of diffractive optical elements that exploit the spatial parallelism and wave properties of light, such as diffraction and interference, to perform logic operations. Light is passed through the stack, and the output is the combination of all the layers which is then cast onto a sensor that turns it into digital data.

“Diffraction casting is just one building block in a hypothetical computer based around this principle and it might be best to think of it as an additional component rather than a full replacement of existing systems, akin to the way graphical processing units are specialized components for graphics, gaming and machine learning workloads,” added Ryosuke Mashiko of the University of Tokyo, in a statement. “I anticipate it will take around 10 years to become commercially available, as much work has to be done on the physical implementation, which, although grounded in real work, has yet to be constructed. At present, we can demonstrate the usefulness of diffraction casting in performing the 16 basic logic operations at the heart of much information processing, but there’s also scope for extending our system into another upcoming area of computing that goes beyond the traditional, and that’s in quantum computing.” [2]

Reconfigurable photonic processor

Researchers from Shanghai Jiao Tong University developed a reconfigurable 3D integrated photonic processor specifically designed to tackle the subset sum problem (SSP), an NP-complete problem that involves determining whether a specific subset of numbers can sum to a given target.

Using femtosecond laser direct writing, the researchers constructed a photonic chip composed of 1,449 standardized optical components, onto which the SSP could be mapped. The processor allows photons in a light beam to explore all possible paths simultaneously, providing answers in parallel. The design was able to solve different instances of the SSP with 100% reliability.

The team says the reconfigurable processor could be adapted for tasks beyond SSP, such as optical neural networks and photonic quantum computing. [3]

References

[1] Pintus, P., Dumont, M., Shah, V. et al. Integrated non-reciprocal magneto-optics with ultra-high endurance for photonic in-memory computing. Nat. Photon. (2024). https://doi.org/10.1038/s41566-024-01549-1

[2] Ryosuke Mashiko, Makoto Naruse, Ryoichi Horisaki, “Diffraction casting,” Adv. Photon. 6(5) 056005 (3 October 2024) https://doi.org/10.1117/1.AP.6.5.056005

[3] Xiao-Yun Xu, Tian-Yu Zhang, Zi-Wei Wang, Chu-Han Wang, Xian-Min Jin, “Reconfigurable integrated photonic processor for NP-complete problems,” Adv. Photon. 6(5) 056011 (24 September 2024) https://doi.org/10.1117/1.AP.6.5.056011