.png){kind=icon}
.png){kind=icon}
Return
MRR-Assisted MZI Crossbar Array for Energy-Efficient Optical Parallel Computing
.png){kind=icon}
PRE
AI
DOI：10.1002/lpor.202501035.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Optical neural networks (ONNs) have emerged as a photonic platform for accelerating artificial intelligence workloads, relying on optical matrix operations as their computational basis. Yet existing implementations face critical challenges in precision, cascaded optical loss, and inefficient weight-programming mechanisms, which limit scalability and practical performance. A micro-ring resonator (MRR)-assisted Mach–Zehnder interferometer (MZI) crossbar architecture (MMCA) that addresses these limitations through parallelized optical computing is presented. By replacing conventional waveguide-based coupling with MRR-enabled wavelength-selective routing, the design suppresses the multiplicative optical losses inherent in cascaded combiner networks. Concurrently, MZI-based crossbar nodes enable deterministic one-to-one weight mapping with enhanced speed and robustness. The architecture synergizes structural merits from mainstream optical computing paradigms while overcoming critical bottlenecks. Experimental validation demonstrates 6-bit weight-loading precision and high energy efficiency. Integrated wavelength-division multiplexing further enables parallel computation, achieving 95.1% classification accuracy on convolutional neural network benchmarks. This approach establishes a scalable framework for low-loss, high-precision optical matrix operations, advancing the development of energy-efficient photonic AI accelerators.
Keywords:
optical convolution computing
optical parallel computation
photonic integrated circuit
photonic neural networks
Journal
L
IF:
10
Papers:
3.7K
Citations:
2.1W