NTT Research
Quantum Leap: Groundbreaking Research Shows Single-Photon Coherent Ising Machines Perform Better Than Expected
NTT Research and Tohoku University’s Graduate School of Information Science (GSIS) have jointly published ground-breaking findings in the Quantum Science and Technology journal, marking a substantial advancement in quantum computing and optimization. A single-photon coherent Ising machine (CIM) can significantly outperform traditional CIMs, even ones with significantly higher photon counts, in tackling challenging combinatorial optimization issues, according to their publication, “Single photon coherent Ising machines for constrained optimization problems.” Long-held beliefs about signal-to-noise restrictions are challenged by this unexpected finding, which comes from quantum simulation, showing that quantum entanglement can result in better performance even in extremely low light levels.
Coherent Ising Machine (CIM)
Joint Research Agreement Accelerates Cyber CIM Development
A joint research agreement (JRA) that was signed in 2023 formalizes the partnership between NTT Research and Tohoku University. The goal of this collaboration is to use traditional high-performance computing (HPC) to develop a large-scale CIM simulation platform. The goal of this large-scale CIM is to pave the way for freely available “cyber CIMs” that can solve hard NP, NP-complete, and HP-hard issues.
You can also read Quantum Entanglement Battery Issues Pre-Physics Assumptions
Under the terms of this agreement:
- The JRA’s primary investigator is Professor Hiroaki Kobayashi of GSIS at Tohoku University, and his NTT Research counterpart is Yoshihisa Yamamoto, director of the PHI Lab.
- The goal of Tohoku University’s research is to leverage HPC platforms to optimize the third-generation cyber CIM. This includes improving data management inside the cache memory architecture, investigating vectorization and parallelization of kernels as accelerators, and scaling the cyber CIM to an astounding 100 million spins with sparse connections on a suitable platform.
- This partnership will “unlock energy efficient and optimized machine learning accelerators” and move them closer to implementing a large-scale CIM simulator, stressed Yoshihisa Yamamoto, PHI Lab Director at NTT Research.
- The PHI Lab’s primary goal, which is at the heart of the JRA, is to utilize nonlinear quantum optical technology to develop straightforward, effective, and useful computing devices for everyday problems. This entails reimagining analog/digital hybrid computers that are inspired by the fundamental concepts of quantum physics and neuroscience, similar to the biological computers found in brains.
- An Ising model-mapped network of degenerate optical parametric oscillators (DOPOs) is what is known as the CIM. These oscillators are programmed to tackle combinatorial optimization issues. A mathematical representation of magnetic systems with competitively interacting spins is known as the Ising model. In particular, a combinatorial clustering problem, a typical challenge in unsupervised machine learning, is examined by researchers from both schools as part of their joint study.
A Striking and Unexpected Result from Single-Photon CIMs
The average photon number per pulse in the recently suggested CIM is as low as one. This is an extremely low quantity, eight orders of magnitude lower than the previously reported photon number (10⁸ photons per pulse) commonly found in traditional CIMs. The performance of CIMs must be evaluated using quantum theory rather than traditional heuristic models in such a weak light limit.
At first, it was widely believed that a CIM that only used one photon per pulse would have trouble reliably storing analogue amplitude data and would have a poor signal-to-noise ratio for monitoring internal pulse amplitudes. Its performance was generally anticipated to be noticeably inferior than that of traditional CIMs under this assumption.
Numerical simulations using the quantum model, however, produced an unexpected and entirely different result. In comparison to its traditional, photon-rich equivalents, the research team found that the single-photon CIM performed significantly better, attaining higher probability of success in discovering perfect answers for diverse problem scenarios.
You can also read KISTI and IonQ to Found South Korea’s First Quantum Centre
The Quantum Enhancement Mechanism Explained
One important finding of this collaboration is the improved performance shown in the single-photon CIM, which is ascribed to a special quantum mechanical phenomenon.
- By producing a correlated internal pulse and an extracted pulse for measurement, an extraction beam splitter is essential at the CIM’s measurement port.
- Even when background noise is present, this correlation between the extracted and internal pulses reaches a quantum regime. In the single-photon CIM, this strongly suggests that these two pulses are quantumly entangled.
- Importantly, the single-photon CIM demonstrated the ability to transform the intrinsically weak quantum entanglement which is readily broken by background noise and optical loss into strong classical correlations between a measured pulse and every other pulse. Through its quantum measurement and feedback mechanism, this conversion takes place.
- The single-photon CIM’s enhanced performance is attributed to the generation of quantum entanglement and its instantaneous conversion to classical correlation. This mechanism is conspicuously lacking in traditional CIMs, which depend on numerous photons per pulse.
Future Outlook and Real-World Applications
NTT Research and Tohoku University are dedicated to extending their partnership in the future. Building on the single-photon CIM’s strong theoretical confirmation, their immediate goals include moving closer to its actual realization. At the same time, they will step up their efforts to create Cyber CIM, a massive simulation environment. “Pave the way for fast and energy-efficient solutions to real-world industrial problems” is the anticipated outcome of this coordinated effort, with possible applications extending to energy-efficient machine learning.
You can also read Explaining Quantum Tunnelling News Using Schrodinger Model
