Matrix multiplication using optical processing

Patent metadata

Field	Value
Publication number	US-12174656-B2
Application number	US-202318505602-A
Country	US
Kind code	B2
Filing date	Nov 9, 2023
Priority date	Nov 2, 2018
Publication date	Dec 24, 2024
Grant date	Dec 24, 2024

Abstract

Systems and methods for performing matrix operations using a photonic processor are provided. The photonic processor includes encoders configured to encode a numerical value into an optical signal and optical multiplication devices configured to output an electrical signal proportional to a product of one or more encoded values. The optical multiplication devices include a first input waveguide, a second input waveguide, a coupler circuit coupled to the first input waveguide and the second input waveguide, a first detector and a second detector coupled to the coupler circuit, and a circuit coupled to the first detector and second detector and configured to output a current that is proportional to a product of a first input value and a second input value.

First claim

What is claimed is:

1. A method of performing matrix-matrix and/or tensor multiplication operations using a plurality of optical multiplication devices, the method comprising:
for each clock cycle of a plurality of clock cycles:
encoding, using a plurality of first encoders, a column of a first matrix into a first plurality of optical signals;
encoding, using a plurality of second encoders, a row of a second matrix into a second plurality of optical signals; and
accumulating, from the plurality of optical multiplication devices coupled to the plurality of first encoders and the plurality of second encoders, first electrical signals that represent a matrix comprising an outer product of the column and the row; and

after the plurality of clock cycles, summing the accumulated first electrical signals to generate second electrical signals representing a product of the first matrix and the second matrix; and
outputting the second electrical signals.

2. The method of claim 1 , wherein accumulating the first electrical signals comprises accumulating the first electrical signals using one or more electrical storage devices.

3. The method of claim 2 , further comprising outputting, from the one or more electrical storage devices, the second electrical signals representing the product of the first matrix and the second matrix.

4. The method of claim 1 , wherein:
encoding the column of the first matrix into the first plurality of optical signals comprises encoding each column value of the column of the first matrix onto an optical field of an optical signal of the first plurality of optical signals; and
encoding the row of the second matrix into the second plurality of optical signals comprises encoding each row value of the row of the second matrix onto an optical field of an optical signal of the second plurality of optical signals.

5. The method of claim 1 , wherein:
encoding the column of the first matrix into the first plurality of optical signals comprises encoding each column value of the column of the first matrix into both an amplitude and a phase of an optical signal of the first plurality of optical signals; and
encoding the row of the second matrix into the second plurality of optical signals comprises encoding each row value of the row of the second matrix into both an amplitude and a phase of an optical signal of the second plurality of optical signals.

6. The method of claim 1 , wherein accumulating the first electrical signals comprises detecting, using a plurality of homodyne detectors, the encoded first plurality of optical signals output by the plurality of first encoders and the encoded second plurality of optical signals output by the plurality of second encoders.

7. The method of claim 6 , further comprising phase shifting, using one or more phase shifters, the encoded first plurality of optical signals and the encoded second plurality of optical signals.

8. The method of claim 7 , wherein phase shifting, using the one or more phase shifters, further comprises correcting for phase errors caused by one or more of differences in an optical path length to each optical multiplication device of the plurality of optical multiplication devices and/or temperature fluctuations.

9. The method of claim 7 , wherein phase shifting, using the one or more phase shifters, further comprises encoding values as complex numbers.

10. The method of claim 7 , wherein phase shifting, using the one or more phase shifters, further comprises encoding positive numbers with a first phase and negative numbers with a second phase having a it-phase difference with respect to the first phase.

11. At least one non-transitory computer-readable medium comprising instructions, which, when executed by an at least one photonic processor, cause the at least one photonic processor to perform a method of:
for each clock cycle of a plurality of clock cycles:
encoding, using a plurality of first encoders, a column of a first matrix into a first plurality of optical signals;
encoding, using a plurality of second encoders, a row of a second matrix into a second plurality of optical signals; and
accumulating, from a plurality of optical multiplication devices coupled to the plurality of first encoders and the plurality of second encoders, first electrical signals that represent a matrix comprising an outer product of the column and the row; and

after the plurality of clock cycles, summing the accumulated first electrical signals to generate second electrical signals representing a product of the first matrix and the second matrix; and
outputting the second electrical signals.

12. The at least one non-transitory computer-readable medium of claim 11 , wherein accumulating the first electrical signals comprises accumulating the first electrical signals using one or more electrical storage devices.

13. The at least one non-transitory computer-readable medium of claim 12 , further comprising outputting, from the one or more electrical storage devices, the second electrical signals representing the product of the first matrix and the second matrix.

14. The at least one non-transitory computer-readable medium of claim 11 , wherein:
encoding the column of the first matrix into the first plurality of optical signals comprises encoding each column value of the column of the first matrix onto an optical field of an optical signal of the first plurality of optical signals; and
encoding the row of the second matrix into the second plurality of optical signals comprises encoding each row value of the row of the second matrix onto an optical field of an optical signal of the second plurality of optical signals.

15. The at least one non-transitory computer-readable medium of claim 11 , wherein:
encoding the column of the first matrix into the first plurality of optical signals comprises encoding each column value of the column of the first matrix into both an amplitude and a phase of an optical signal of the first plurality of optical signals; and
encoding the row of the second matrix into the second plurality of optical signals comprises encoding each row value of the row of the second matrix into both an amplitude and a phase of an optical signal of the second plurality of optical signals.

16. The at least one non-transitory computer-readable medium of claim 11 , wherein accumulating the first electrical signals comprises detecting, using a plurality of homodyne detectors, the encoded first plurality of optical signals output by the plurality of first encoders and the encoded second plurality of optical signals output by the plurality of second encoders.

17. The at least one non-transitory computer-readable medium of claim 16 , further comprising phase shifting, using one or more phase shifters, the encoded first plurality of optical signals and the encoded second plurality of optical signals.

18. The at least one non-transitory computer-readable medium of claim 17 , wherein phase shifting, using the one or more phase shifters, further comprises correcting for phase errors caused by one or more of differences in an optical path length to each optical multiplication device of the plurality of optical multiplication devices and/or temperature fluctuations.

19. The at least one non-transitory computer-readable medium of claim 17 , wherein phase shifting, using the one or more phase shifters, further comprises encoding values as complex numbers.

20. The at least one non-transitory computer-readable medium of claim 17 , wherein phase shifting, using t

Assignees

• Lightmatter Inc

Inventors

• Bunandar Darius
• Harris Nicholas C
• Kenney Tyler J

Classifications

• G06E1/045

  Matrix or vector computation · CPC title

• G02F3/00 Primary

  Optical logic elements; Optical bistable devices · CPC title

• G06N3/0675

  using electro-optical, acousto-optical or opto-electronic means · CPC title

• G06N3/084

  Backpropagation, e.g. using gradient descent · CPC title

• G06E3/008

  Matrix or vector computation · CPC title