Neural networks

• US 5,428,466 A
• Filed: 06/11/1992
• Issued: 06/27/1995
• Est. Priority Date: 11/20/1989
• Status: Expired due to Fees

• Current Assignee: British Telecommunications (BT Group Plc) (BT Group PLC)
• Original Assignee: British Telecommunications PLC (BT Group PLC)
• Inventors: Wood, David C., Scott, Edward G., Rejman-Greene, Marek A. Z., Healey, Peter, Webb, Roderick P.
• Primary Examiner(s): Ben, Loha
• Assistant Examiner(s): Collins, Darryl J.
• Application Number:
  US07/859,698
• Time in Patent Office: 1,111 Days
• Field of Search: 359/1, 359/11, 359/15, 359/10, 359/559, 359/561, 359/565, 359/566, 359/567, 395/21, 395/25, 364/807, 364/822, 364/841, 364/359, 364/395
• US Class Current: 359/15
• CPC Class Codes: B82Y 20/00 Nanooptics, e.g. quantum optics or photonic crystals
• CPC Class Codes: G02F 1/017 Structures with periodic or quasi periodic potential variation, e.g. superlattices, quantum wells
• CPC Class Codes: G02F 1/218 using semi-conducting materials
• CPC Class Codes: G06N 3/0675 using electro-optical, acousto-optical or opto-electronic means
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: H01L 31/035236 Superlattices; Multiple quantum well structures
• CPC Class Codes: H01L 31/184 the active layers comprising only AIIIBV compounds, e.g. GaAs, InP
• CPC Class Codes: Y02E 10/544 Solar cells from Group III-V materials

Abstract

A neural network has inputs formed by square array of optical modulators M_ij and outputs by optical detectors D_ij coupled to threshold comparators. A holographic plate includes a spatial modulator whose elements are controlled by a controller to form an array of optical beams from a coherent optical source. Each optical beam optically interconnects a modulator M_ij with a respective detector D_ij. The weight values of the neural network are provided by the intensities of the optical beams. This obviates the need for an optical weighting mask between an array of light emitting diodes and a detector array allowing a higher density of lower power consumption components and reprogrammability of the network.

Claims

1. A neural network comprising:

beam source means for providing a plurality of weighted intensity input optical beams;

an array of modulators, each for modulating a respective one of said input optical beams whereby each said modulator produces a respective modulated beam; and

an array of detectors, each for detecting the modulated beam solely from a respective one of the modulators, said detectors being electrically connected in groups to form nodes, whereby to form sums of the modulated and weighted beam intensities at the nodes;

wherein the beam source means includes a source of laser radiation, and holographic means operative to split the laser radiation from the source into the plurality of input beams.

2. A network as in claim 1 in which the holographic means comprises:

a hologram for generating a plurality of substantially equal intensity optical beams, and

an optical mask between the hologram and the array of optical modulators for providing the weighted intensities of input optical beams.

3. A network as in claim 2 in which the optical mask is removably mounted in the network.

4. A network as in claim 2 in which the mask comprises a spatial light modulator and includes means for altering the spatial light modulator to a desired pattern.

5. A network as in claim 1 in which the holographic means comprises a hologram configured to generate the input optical beams of the required intensities.

6. A network as in claim 1 in which the holographic means is positioned between a pair of collimating lenses.

7. A network as in claim 1 in which the holographic means includes a hologram that is a binary phase pattern.

8. A network as in claim 1 in which the holographic means includes a hologram that is a fixed mask, removably mounted in the network.

9. A network as in claim 1 in which the holographic means comprises a spatial light modulator and includes means for altering the spatial light modulator to a desired pattern.

10. A network as in claim 1 including at least two holograms, any one of which being selectable to provide the plurality of input beams.

11. A network as in claim 10 in which each hologram is illuminatable by a respective optical source.

12. A neural network as claimed in claim 1 wherein the modulators and detectors are arranged in overlying corresponding arrays, with the modulators being electrically interconnected to define array columns, and the detectors being interconnected to define array rows which provide said nodes.

13. A neural network having:

at least one node responsive to the weighted sum of a plurality of inputs to provide an output dependent on said weighted sum,

the node comprising an optical detecting means for detecting each of the inputs, and the inputs comprising input optical beams whose intensities determine the weight values of the inputs, and

the network including a holographic means for generating the input optical beams and an array of optical modulators for controlling which of the input optical beams impinge on the optical detecting means,

wherein the holographic means incorporates the phase structure of a pair of collimating lenses located one either side of the hologram.

14. A neural network having:

at least one node responsive to the weighted sum of a plurality of inputs to provide an output dependent on said weighted sum,

the node comprising an optical detecting means for detecting each of the inputs, and the inputs comprising input optical beams whose intensities determine the weight values of the inputs, and

the network including a holographic means for generating the input optical beams and an array of optical modulators for controlling which of the input optical beams impinge on the optical detecting means,

wherein the array of optical modulators and the optical detecting means are formed on respective sides of a common substrate.

15. A network as in claim 14 in which the optical modulators comprise multiple quantum well devices.

16. A network as in claim 15 including a bias means for adjusting the modulation depth of the multiple quantum well devices.

17. A neural network having:

at least one node responsive to the weighted sum of a plurality of inputs to provide an output dependent on said weighted sum,

the node comprising an optical detecting means for detecting each of the inputs, and the inputs comprising input optical beams whose intensities determine the weight values of the inputs, and

the network including a holographic means for generating the input optical beams and an array of optical modulators disposed between the holographic means and the optical detecting means for controlling which of the input optical beams impinge on the optical detecting means,

wherein the optical modulators and optical detecting means are all formed from multiple quantum well devices.

18. A neural network having:

at least one node responsive to a weighted sum of a plurality of inputs to provide an output dependent on said weighted sum,

the node comprising an optical detecting means for detecting each of the inputs, and the inputs comprising input optical beams whose intensities determine the weight values of the inputs, and

the network including a holographic means for generating the input optical beams and an array of optical modulators for controlling which of the input optical beams impinge on the optical detecting means,

wherein the array of optical modulators and the optical detecting means are formed on a common substrate.