Optical Matrix Computing With Single-MZI Unitary Transformation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing optical chips for matrix operations in artificial intelligence face issues such as high light loss due to beam splitters, increased chip area and pin count, and complexity in packaging and testing, limiting computing capability.

Innovation Solution

A computing apparatus using a unitary matrix represented by a single Mach-Zehnder interferometer (MZI) to perform matrix operations, reducing light loss and chip complexity by projecting the real part of complex amplitudes onto a real axis, and employing feedback loops for recurrent neural networks.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If beam splitters are used in optical chips for matrix operations, then matrix computation capability is achieved, but light loss increases

Engineering Contradiction:

Improvematrix computation capabilityVSAvoidlight loss

Core Design Contradiction:

ProductivityVSLoss of energy

Solution Approach 1:

The patent extracts and removes beam splitters from the optical computing system. By representing real matrices directly through unitary matrices without requiring beam splitter components, the design eliminates the source of light loss while preserving matrix computation capability through alternative optical pathways.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent substitutes the mechanical beam splitter configuration with a unitary matrix-based optical system. This replacement uses phase modulation and direct optical transmission through unitary transformations to achieve matrix operations without the energy-dissipating beam splitting mechanism.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If conventional optical chip architecture is used, then matrix operations can be performed, but chip area and pin count increase

Engineering Contradiction:

Improvematrix operation capabilityVSAvoidchip area

Core Design Contradiction:

ProductivityVSArea of stationary object

Solution Approach 1:

The patent merges multiple functional components into a unified unitary matrix structure. By combining what would traditionally require separate beam splitters, phase shifters, and routing elements into a single unitary transformation layer, the design reduces chip area while maintaining full matrix operation capability.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The unitary matrix structure serves multiple functions simultaneously: it performs matrix multiplication, phase modulation, and signal routing in a single integrated layer. This multi-functionality eliminates the need for separate dedicated components, thereby reducing overall chip area and pin count.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Productivity

If conventional optical chip architecture is used, then matrix operations can be performed, but device complexity increases

Engineering Contradiction:

Improvematrix operation capabilityVSAvoidchip structure complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent extracts and removes complex beam splitter networks from the system architecture. By relying solely on unitary matrix transformations implemented through simpler phase modulation elements, the design reduces structural complexity while preserving computational functionality.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

Instead of building complex beam splitter networks to achieve unitary transformations, the patent inverts the approach by directly implementing unitary matrices through phase modulation. This reversal simplifies the architectural design by eliminating the need for complex interference-based beam splitting structures.

Inventive Principle:

Principle #13The other way round (Inversion)

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

Reduces light loss, simplifies chip structure, decreases pin count, and improves computing performance by using a single MZI to represent any real matrix, enhancing the computing capability of photonic artificial intelligence chips.

Implementation Method 1

a matrix calculation unit having a transmission matrix and configured to use the transmission matrix to process the first optical signal to obtain a second optical signal

Methodology Applied

Scientific EffectInterference: Interference

Data Source

PatentUS12530425B2Computing with optical signal

Publication Date: 2026.01.20 UNITED MICROELECTRONICS CENT CO LTD

AI summary

A computing apparatus is provided. The apparatus includes: an input port to receive a first optical signal; a matrix calculation unit having a transmission matrix and configured to use the transmission matrix to process the first optical signal to obtain a second optical signal, where the second optical signal has a complex amplitude, and the transmission matrix is a unitary matrix M′; a real part obtaining unit optically coupled to the matrix calculation unit and configured to obtain, from the second optical signal, a third optical signal representing a real part of a complex amplitude of the second optical signal; and an output port configured to output an optical signal.