Fully Programmable Spatial Photonic Ising Machine by Focal Plane Division

Area of Science:

• Quantum computing and optimization hardware.
• Development of novel optical computing architectures.
• Advancements in non-von Neumann computing paradigms.

Background:

• Ising machines offer ultrafast, energy-efficient solutions for NP-hard optimization problems.
• Spatial photonic Ising machines (SPIMs) leverage optical computing for accelerated computations.
• Current SPIMs face limitations in programmability, restricting their application scope.

Purpose of the Study:

• To achieve full programmability in SPIMs while maintaining scalability.
• To overcome the critical limitation of partial programmability in existing SPIMs.
• To enable SPIMs to tackle a wider range of combinatorial optimization problems.

Main Methods:

• Introduced a novel operation method based on focal plane division for SPIMs.
• Decomposed general Ising problems into Mattis Hamiltonians for optical computation.
• Simultaneously computed energies by measuring light intensity on a camera sensor.

Main Results:

• Experimentally demonstrated a fully programmable SPIM capable of solving up to 32-spin Ising models.
• Achieved high success probability for ground-state solutions on maximum cut graphs.
• Simulations indicated favorable accuracy scaling with an increased number of spins.

Conclusions:

• The developed SPIM achieves full programmability and scalability, addressing a key challenge in photonic Ising machines.
• This advancement broadens the applicability of SPIMs to various quadratic unconstrained binary optimization problems.
• The fully programmable SPIM reinforces its position as a leading non-von Neumann hardware paradigm for optimization.