The spatial complexity of optical computing: toward space-efficient design

Area of Science:

• Physics
• Computer Science
• Engineering

Background:

• Hardware, like algorithms, requires resources such as space, dictated by wave physics for optical computing.
• Determining the spatial requirements for optical computing operations, especially general tasks like classification, remains an open challenge.

Purpose of the Study:

• To investigate the spatial complexity of optical computing systems using scaling laws.
• To propose a novel paradigm for designing space-efficient optical computing systems.

Main Methods:

• Inspired by computational complexity theory, the study analyzes how physical dimensions scale with mathematical operation dimensions.
• A new paradigm, space-efficient neuromorphic optics, is proposed, utilizing structural sparsity and neural pruning based on wave physics concepts like "overlapping nonlocality".

Main Results:

• The proposed methods achieve substantial size reductions, occupying only 1%-10% of conventional designs on free-space and on-chip photonic platforms.
• Theoretical and computational results show diminishing returns in accuracy as structure dimensions increase.

Conclusions:

• The study introduces a new perspective on the ultimate limits of optical computing.
• A balanced trade-off between device size and accuracy is achievable, paving the way for more compact and efficient optical computing systems.