Optoelectronic sensory neural network

• US 5,130,563 A
• Filed: 06/07/1991
• Issued: 07/14/1992
• Est. Priority Date: 11/30/1989
• Status: Expired due to Fees

• Current Assignee: Washington University In St Louis
• Original Assignee: Washington Research Foundation (University of Washington)
• Inventors: Darling, Robert B., Pinter, Robert B., Nabet, Bahram
• Primary Examiner(s): HUDSPETH, DAVID R
• Application Number:
  US07/713,550
• Time in Patent Office: 403 Days
• Field of Search: 307/201, 364/513, 364/807, 364/822, 364/841
• US Class Current: 706/40
• CPC Class Codes: G06N 3/0675 using electro-optical, acousto-optical or opto-electronic means

Abstract

A neural network for processing sensory information. The network comprise one or more layers including interconnecting cells having individual states. Each cell is connected to one or more neighboring cells. Sensory signals and signals from interconnected neighboring cells control a current or a conductance within a cell to influence the cell'"'"'s state. In some embodiments, the current or conductance of a cell can be controlled by a signal arising externally of the layer. Each cell can comprise an electrical circuit which receives an input signal and causes a current corresponding to the signal to pass through a variable conductance. The conductance is a function of the states of the one or more interconnecting neighboring cells. Proper interconnection of the cells on a layer can produce a neural network which is sensitive to predetermined patterns or the passage of such patterns across a sensor array whose signals are input into the network. The layers in the network can be made sensitive to distinct sensory parameters, so that networks which are sensitive to different wavelengths or polarizations of light energy can be produced.

Claims

1. A neural network for processing light energy having a plurality of characteristics, comprising:

a plurality of layers of cells, each cell in a layer having one or more neighboring cells in that layer, each layer of cells being connected in a neighboring relationship for processing light energy having a distinct one of the plurality of characteristics, each cell comprising;

selective means for receiving an input signal representing an input corespondent to light energy having the characteristic to be processing by the layer in which the cell belongs;

means for transforming the input signal to a corresponding current produced by a current source, the current being transmitted through a capacitive element and producing a resulting change in a voltage across the capacitive element;

programmable means for sensing the states of the one or more neighboring cells;

means for producing a conductance corresponding to each of the states in a feedforward subset of the states of the one or more neighboring cells;

means for producing a conductance corresponding to each of the states in a feedback subset of the states of the one or more neighboring cells, the feedback subset being disjoint from the feedforward subset;

means for producing a positive state voltage corresponding to the state of the cell, the positive state voltage being transmitted through the capacitive element and producing a resulting change in a voltage across the capacitive element;

means for producing a positive feedback voltage corresponding to each of the states in the feedback subset of the states of the adjacent cells, the positive feedback voltage being transmitted through the capacitive element and producing a resulting feedback change in the state voltage across the capacitive element; and

means for producing a negative feedforward voltage corresponding to each of the states in the feedforward subset of the states of the adjacent cells, the negative feedforward voltage being transmitted through the capacitive element and producing a resulting feedforward change in the state voltage across the capacitive element;

the voltage across the capacitive element representing the state of the cell.

2. The neural network of claim 1 wherein means for transforming the input signal to the corresponding conductance is formed from the parallel combination of a plurality of transistors having conductances that are individually controlled in accordance with the states of the one or more neighboring cells.

3. The neural network of claim 2 wherein each of the transistors is formed in an integrated circuit.

4. The neural network of claim 2 wherein the transistors are field effect transistors.

5. The neural network of claim 4 wherein each of the field effect transistors is formed in an integrated circuit.

6. A neural network for processing light energy having a plurality of characteristics, comprising:

a plurality of layers of cells, each cell in a layer having one or more neighboring cells in that layer, each layer of cells being connected in a neighboring relationship for processing light energy having a distinct one of the plurality of characteristics, each cell comprising;

selective means for receiving an input signal representing an input corespondent to light energy having the characteristic to be processing by the layer in which the cell belongs;

means for transforming the input signal to a corresponding current produced by a current source, the current being transmitted through a capacitive element and producing a resulting change in a voltage across the capacitive element;

programmable means for sensing the states of the one or more neighboring cells;

means for producing a conductance corresponding to each of the states in a feedforward subset of the states of the one or more neighboring cells;

means for producing a conductance corresponding to each of the states in a feedback subset of the states of the one or more neighboring cells, the feedback subset being disjoint from the feedforward subset;

means for producing a positive state voltage corresponding to the state of the cell, the positive state voltage being transmitted through the capacitive element and producing a resulting change in a voltage across the capacitive element;

means for producing a positive feedback voltage corresponding to each of the states in the feedback subset of the states of the adjacent cells, the positive feedback voltage being transmitted through the capacitive element and producing a resulting feedback change in the state voltage across the capacitive element; and

means for producing a negative feedforward voltage corresponding to each of the states in the feedforward subset of the states of the adjacent cells, the negative feedforward voltage being transmitted through the capactive element and producing a resulting feedforward change in the state voltage across the capacitive element;

the voltage across the capacitive element representing the state of the cell.

7. The neural network of claim 6 wherein means for transforming the sensory input signal to the corresponding conductance is formed from the parallel combination of a plurality of transistors having conductances that are individually controlled in accordance with the states of the one or more neighboring cells.

8. The neural network of claim 7 wherein each of the transistors is formed in an integrated circuit.

9. The neural network of claim 7 wherein the transistors are field effect transistors.

10. The neural network of claim 9 wherein each of the field effect transistors is formed in an integrated circuit.

11. A neural network for processing light energy consisting of a plurality of wavelengths, comprising:

a plurality of layers of cells, each cell in a layer having one or more neighboring cells in that layer, each layer of cells being connected in a neighboring relationship for processing light energy having a distinct wavelength in the plurality of wavelengths, each cell comprising;

selective means for receiving an input signal representing an input corespondent to light energy having the distinct wavelength corespondent to the wavelength to be processed by the layer in which the cell belongs;

means for transforming the input signal to a corresponding current produced by a current source, the current being transmitted through a capacitive element and producing a resulting change in a voltage across the capacitive element;

programmable means for sensing the states of the one or more neighboring cells;

means for producing a conductance corresponding to each of the states in a feedforward subset of the states of the one or more neighboring cells;

means for producing a conductance corresponding to each of the states in a feedback subset of the states of the one or more neighboring cells, the feedback subset being disjoint from the feedforward subset;

means for producing a positive state voltage corresponding to the state of the cell, the positive state voltage being transmitted through the capacitive element and producing a resulting change in a voltage across the capacitive element;

means for producing a positive feedback voltage corresponding to each of the states in the feedback subset of the states of the adjacent cells, the positive feedback voltage being transmitted through the capacitive element and producing a resulting feedback change in the state voltage across the capacitive element; and

means for producing a negative feedforward voltage corresponding to each of the states in the feedforward subset of the states of the adjacent cells, the negative feedforward voltage being transmitted through the capacitive element and producing a resulting feedforward change in the state voltage across the capacitive element;

the voltage across the capacitive element representing the state of the cell.

12. The neural network of claim 11 wherein means for transforming the input signal to the corresponding conductance is formed from the parallel combination of a plurality of transistors having conductances that are individually controlled in accordance with the states of the one or more neighboring cells.

13. The neural network of claim 12 wherein each of the transistors is formed in an integrated circuit.

14. The neural network of claim 12 wherein the transistors are field effect transistors.

15. The neural network of claim 14 wherein each of the field effect transistors is formed in an integrated circuit.

16. The neural network of claim 11, wherein the plurality of wavelengths are in the red, green, and blue light spectral regions.

17. A neural network for processing light energy having a plurality of distinct polarizations, comprising:

a plurality of layers of cells, each cell in a layer having one or more neighboring cells in that layer, each layer of cells being connected in a neighboring relationship for processing light energy having a distinct one of the distinct polarizations, each cell comprising;

selective means for receiving an input signal representing an input corespondent to light energy having the distinct polarization to be processing by the layer in which the cell belongs;

means for transforming the input signal to a corresponding current produced by a current source, the current being transmitted through a capacitive element and producing a resulting change in a voltage across the capacitive element;

programmable means for sensing the states of the one or more neighboring cells;

means for producing a conductance corresponding to each of the states in a feedforward subset of the states of the one or more neighboring cells;

means for producing a conductance corresponding to each of the states in a feedback subset of the states of the one or more neighboring cells, the feedback subset being disjoint from the feedforward subset;

means for producing a positive state voltage corresponding to the state of the cell, the positive state voltage being transmitted through the capacitive element and producing a resulting change in a voltage across the capacitive element;

means for producing a positive feedback voltage corresponding to each of the states in the feedback subset of the states of the adjacent cells, the positive feedback voltage being transmitted through the capacitive element and producing a resulting feedback change in the state voltage across the capacitive element; and

means for producing a negative feedforward voltage corresponding to each of the states in the feedforward subset of the states of the adjacent cells, the negative feedforward voltage being transmitted through the capacitive element and producing a resulting feedforward change in the state voltage across the capacitive element;

the voltage across the capacitive element representing the state of the cell.

18. The neural network of claim 17 wherein means for transforming the sensory input signal to the corresponding conductance is formed from the parallel combination of a plurality of transistors having conductances that are individually controlled in accordance with the states of the one or more neighboring cells.

19. The neural network of claim 18 wherein each of the transistors is formed in an integrated circuit.

20. The neural network of claim 18 wherein the transistors are field effect transistors.

21. The neural network of claim 20 wherein each of the field effect transistors is formed in an integrated circuit.

22. The neural network of claim 17 wherein the plurality of polarizations are orthogonal polarizations.

23. The neural network of claim 22 wherein the plurality of polarizations are orthogonal linear polarizations.

24. The neural network of claim 22 wherein the plurality of polarizations are orthogonal circular polarizations.

25. A neural network comprising a plurality of cells and means for connecting each cell to one or more neighbor cells, each cell comprising:

a photodetector for receiving optical illumination and for producing a corresponding input current; and

circuit means having a variable conductivity, the circuit means comprising means for receiving the input current for the cell and for conducting the input current such that a voltage is produced across the circuit means, the voltage comprising a state value for the cell, the conductivity of the circuit means being in part a function of the product of the state value of the cell and the state value of one of the neighbor cells.

26. The neural network of claim 25, wherein each cell is connected to at least first and second neighbor cells and comprises first and second conductance elements connected in parallel, the first conductance element having a conductance that is a function of the product of the state value of the cell and the state value of the first neighbor cell, and the second conductance element having a conductance that is a function of the product of the state value of the cell and the state value of the second neighbor cell.

27. The neural network of claim 26, wherein each conductance element comprises a field effect transistor having a gate coupled to the respective neighbor cell.

28. The neural network of claim 27, wherein each field effect transistor is formed in an integrated circuit.

29. The neural network of claim 26, wherein each circuit means comprises a capacitive element connected in parallel with the conductance elements.

30. The neural network of claim 25, wherein each cell further comprises a voltage shifter circuit for producing an output voltage that differs from the state value by an amount independent of the state value, and means for coupling the output voltage to at least one neighbor cell, and wherein the conductivity of the circuit means is in part a function of the product of the state value of the cell and the output voltage of the neighbor cell.

31. The neural network of claim 26, wherein the cells are arranged in a sequence in which for each cell, the first and second neighbor cells are positioned immediately adjacent to the cell in the sequence, and wherein each cell further comprises means for attenuating the state voltage coupled to the first neighbor cell with respect to the state voltage coupled to the second neighbor cell, whereby the response of the network is asymmetric along said sequence.