Optical processor for an artificial neural network

• US 20080154815A1
• Filed: 10/16/2006
• Published: 06/26/2008
• Est. Priority Date: 10/16/2006
• Status: Active Grant

• Current Assignee: Alcatel-Lucent USA, Inc. (Nokia Corporation)
• Original Assignee: Lucent Technologies, Inc. (Nokia Corporation)
• Inventors: Martinez, John C.
• Granted Patent: US 7,512,573 B2
• Time in Patent Office: Days
• US Class Current: 706/15
• CPC Class Codes: G06N 3/067 using optical means

Abstract

An optical processor adapted to emulate an artificial neural network (ANN) having a plurality of interconnected layers, each layer having one or more artificial neurons, the processor having a spatial light modulator (SLM) optically coupled, via an optical mask, to a photodetector array. In one embodiment, the SLM has a plurality of pixels, each pixel being configurable to emulate an output portion of a corresponding artificial neuron in a signal-sending ANN layer. The optical mask has a hologram that encodes the weights corresponding to interlayer connections in the ANN and spatially modulates the light transmitted by the SLM. The photodetectors of the array spatially resolve the interference pattern produced by the spatially modulated light, with each photodetector being configurable to emulate an input portion of a corresponding artificial neuron in a signal-receiving ANN layer.

Claims

1. Apparatus, comprising:

a spatial light modulator (SLM) having a plurality of pixels, each pixel being configurable to direct or not to direct light toward a photodetector array;

an optical mask adapted to subject to spatial modulation the light directed by the SLM toward the photodetector array; and

the photodetector array adapted to measure an interference pattern produced by the spatially modulated light.

2. The invention of claim 1, further comprising a processing and control unit (PCU) adapted to:

configure the pixels of the SLM based on an externally generated input signal; and

reconfigure the pixels of the SLM based on measurement results produced by the photodetector array.

3. The invention of claim 2, wherein:

the apparatus is adapted to emulate an artificial neural network (ANN) having a plurality of interconnected layers, each layer having one or more artificial neurons;

each pixel is adapted to emulate an output portion of a corresponding artificial neuron of a signal-sending ANN layer by directing or not directing the light based on an output of said artificial neuron; and

each photodetector is adapted to emulate an input portion of a corresponding artificial neuron of a signal-receiving ANN layer by measuring light intensity produced by the interference pattern at said photodetector, wherein the PCU is adapted to compare the measured light intensity with a respective threshold value to determine an output for the artificial neuron whose input portion is being emulated by said photodetector.

4. The invention of claim 3, wherein:

the PCU is adapted to configure the pixels of the SLM and to process the measurement results of the photodetector array to emulate interlayer signal coupling in the ANN, wherein the apparatus is adapted to recursively emulate signal propagation through successive ANN layers to determine a final output of the ANN.

5. The invention of claim 3, wherein each pixel of the SLM comprises an optical window adapted to become either substantially transparent or substantially opaque based on the output of the artificial neuron whose output portion is being emulated by said pixel.

6. The invention of claim 3, wherein each pixel of the SLM comprises a movable mirror adapted to change orientation based on the output of the artificial neuron whose output portion is being emulated by said pixel.

7. The invention of claim 3, wherein the optical mask comprises a hologram that encodes weights corresponding to signal flow between one or more artificial neurons of the signal-sending ANN layer and one or more artificial neurons of the signal-receiving ANN layer.

8. The invention of claim 1, further comprising:

a laser optically coupled to the SLM and adapted to direct light via the SLM toward the photodetector array.

9. The invention of claim 1, wherein:

the optical mask is adapted to operate in reflection; and

the SLM and the photodetector array are integrated on a single chip.

10. A method of signal processing, comprising:

selectively directing light toward a photodetector array using a plurality of pixels in a spatial light modulator (SLM);

spatially modulating the light directed by the SLM toward the photodetector array using an optical mask; and

measuring an interference pattern produced by the spatially modulated light using the photodetector array.

11. The invention of claim 10, further comprising:

processing measurement results produced by the photodetector array to configure the pixels of the SLM.

12. The invention of claim 10, further comprising:

emulating an artificial neural network (ANN) having a plurality of interconnected layers, each layer having one or more artificial neurons, wherein;

each pixel is adapted to emulate an output portion of a corresponding artificial neuron of a signal-sending ANN layer by directing or not directing the light based on an output of said artificial neuron; and

each photodetector is adapted to emulate an input portion of a corresponding artificial neuron of a signal-receiving ANN layer by measuring light intensity produced by the interference pattern at said photodetector, wherein the PCU is adapted to compare the measured light intensity with a respective threshold value to determine an output for the artificial neuron whose input portion is being emulated by said photodetector.

13. The invention of claim 12, wherein the step of emulating comprises:

configuring the pixels of the SLM and processing the measurement results of the photodetector array to emulate interlayer signal coupling in the ANN; and

recursively emulating signal propagation through successive ANN layers to determine a final output of the ANN.

14. The invention of claim 12, wherein each pixel of the SLM comprises an optical window adapted to become either substantially transparent or substantially opaque based on the output of the artificial neuron whose output portion is being emulated by said pixel.

15. The invention of claim 12, wherein the optical mask comprises a hologram that encodes weights corresponding to signal flow between one or more artificial neurons of the signal-sending ANN layer and one or more artificial neurons of the signal-receiving ANN layer.

16. The invention of claim 15, wherein the step of emulating comprises:

repeating the steps of selectively directing, spatially modulating, and measuring to process one or more training input patterns for the ANN; and

changing the hologram to find optimal weights for one or more artificial neurons of the ANN based on the processing of said one or more training input patterns.

17. A method of emulating an artificial neural network (ANN) having a plurality of interconnected layers, each layer having one or more artificial neurons, the method comprising:

based on one or more outputs of one or more artificial neurons in a first selected layer of the ANN, configuring one or more pixels of a first selected section of a spatial light modulator (SLM) to direct or not to direct light toward a photodetector array;

subjecting the light directed toward the photodetector array by the one or more so configured pixels to spatial modulation using an optical mask; and

spatially resolving an interference pattern produced by the spatially modulated light at the photodetector array to determine one or more outputs of one or more artificial neurons in a next layer of the ANN.

18. The invention of claim 17, wherein:

the first selected layer is an input layer; and

the step of configuring comprises, for each artificial neuron of said input layer, comparing a respective input value with a respective threshold value to determine the output for said artificial neuron.

19. The invention of claim 17, wherein:

the next layer is an intermediate layer; and

for each artificial neuron of said intermediate layer, the step of spatially resolving comprises;

measuring light intensity at a photodetector corresponding to the artificial neuron; and

comparing the measured value with a respective threshold value to determine the output for the artificial neuron.

20. The invention of claim 19, further comprising:

repeating the steps of configuring and subjecting for said intermediate layer; and

repeating the step of spatially resolving for a layer following said intermediate layer.