Photonic Matrix Computing

Photonic matrix computing for modern neural workloads

This book presents a practical, system-level view of how optical interference, coherent light, and programmable waveguide arrays can be used to perform matrix multiplication with far less energy than conventional digital MAC pipelines. It bridges core physics, hardware design, and neural network deployment, making it useful for readers who want to understand both the promise and the engineering limits of photonic acceleration.

Beginning with the computational role of matrix operations in neural networks, the book explains how dataflow, numerical formats, and matrix shapes influence hardware efficiency. It then builds the optical foundation, covering coherence, phase control, intensity detection, and the linear behavior that makes photonic transforms possible.

From there, the focus shifts to real architectures, including reconfigurable photonic meshes, beam splitters, phase shifters, and calibration methods for realizing target matrices. Detailed coverage shows how to encode inputs, handle positive and negative values, scale outputs, and run batch or tiled workloads for larger layers.

• Hardware-aware training and inference for optical weight constraints
• Nonidealities and mitigation including loss, crosstalk, quantization, and detector noise
• Energy accounting across optical power, modulation, control, I/O, and synchronization
• Integration strategies for combining photonic tiles with digital systems
• Calibration, verification, and maintenance workflows for stable operation

The later chapters bring these ideas together through case studies and design methodology, showing how to estimate energy budgets, choose encodings, and co-design accuracy with power consumption. Worked examples throughout the book make the material concrete and help readers move from theory to implementation.

Ideal for engineers, researchers, and advanced students working in photonics, hardware acceleration, and efficient neural computation.