ReDON: Recurrent Diffractive Optical Neural Processor with Reconfigurable Self-Modulated Nonlinearity

Researchers introduce ReDON, a new recurrent diffractive optical neural processor that overcomes limitations of traditional optical neural networks through reconfigurable self-modulated nonlinearity. The architecture demonstrates up to 20% improved accuracy on image recognition tasks while maintaining energy efficiency, establishing a new paradigm for non-von Neumann analog processors.

• →ReDON addresses computational limitations of static diffractive optical neural networks through dynamic, input-dependent optical transmission.
• →The architecture uses reconfigurable self-modulated nonlinearity inspired by gated linear units from large language models.
• →ReDON achieved up to 20% improvement in test accuracy and mIoU on image recognition and segmentation benchmarks.
• →The system maintains energy efficiency while extending nonlinear representational capacity through recurrent optical hardware reuse.
• →This work establishes a new paradigm combining recurrence and self-modulation in non-von Neumann analog processors.