Neural network using photoelectric substance

• US 5,248,899 A
• Filed: 02/07/1992
• Issued: 09/28/1993
• Est. Priority Date: 02/07/1992
• Status: Expired due to Fees

• Current Assignee: Aaron Lewis, Haronian, Dan
• Original Assignee: Aaron Lewis, Haronian, Dan
• Inventors: Haronian, Dan, Lewis, Aaron
• Primary Examiner(s): Hudspeth, David R.
• Application Number:
  US07/832,912
• Time in Patent Office: 599 Days
• Field of Search: 307/201, 395/25
• US Class Current: 706/40
• CPC Class Codes: B82Y 10/00 Nanotechnology for information processing, storage or transmission, e.g. quantum computing or single electron logic
• CPC Class Codes: G06N 3/002 Biomolecular computers, i.e. using biomolecules, proteins, cells using DNA G06N3/123; using neurons G06N3/061
• CPC Class Codes: G06N 3/063 using electronic means
• CPC Class Codes: G06N 3/0675 using electro-optical, acousto-optical or opto-electronic means

Abstract

A neural network, and a method of storing information and retrieving it by such network. The network comprises neurons, synapses and switches, and when required also rectifying means. The network is based on a substance which undergoes a reversible change from stable state A to stable state B, and this substance can also be changed from state A to another state C, which change is also reversible, where each change provides a measurable electrical pulse. The change of state is brought about by means of illumination for a predetermined period of time at a certain wavelength, it being possible to convert a desired part of the substance from one state to the other.

Claims

1. A neural network comprising a plurality of neurons, synapses and switches, which synapses comprise a substance X which can be transferred by illuminating it at a given wavelength from stable State A to stable State B, which change is reversible, and which substance can be transferred from State A to State C which differs from State B, which is also reversible, each such change providing an electrically measurable pulse, said components being interconnected for data storage and retrieval in parallel or serially, wherein said data storage is achieved in individual synapses by selectively illuminating said substance X at said given wavelength to thereby change part of said substance X to stable State B.

2. A network according to claim 1, and further comprising reading means for transforming the rest of the substance to State C and back to State A, resulting in an electrical signal proportional to the quantity of substance X passing from A to C and back which is indicative of the quantity of substance X in State B.

3. A neural network according to claim 1, wherein the synapses are composed of a material which acts as a programmable current source.

4. A neural network according to claim 1, where the synapses are composed of a material which can be altered by means of an external energy source so that the synaptic strength can be varied at will, and can be reprogrammed by the use of this external energy source.

5. A network according to claim 1 or 2, where the active substance is bacteriorhodopsin (bR).

6. A network according to claim 1, where the active substance X is bacteriorhodopsin, where the change from A to B is from a State of bacteriorhodopsin to a State of M by 570 nm irradiation;

and where the change from A to C is from the State of bacteriorhodopsin to a state of K also by 570 nm irradiation, in sequenced timing with a 680 nm pulse for reversal to the State of bacteriorhodopsin, resulting in an electric pulse indicative of the portion of substance X in the M state.

7. A network according to claim 3, comprising means for maintaining a desired temperature and a dry atmosphere.

8. A network according to claim 1, wherein said synapses are arranged in columns and rows, and further comprising operational amplifiers which act as said neurons and connect each column of synapses through one of said switches from a ground to a virtual ground, said one of said switches being connected to and selectively controlled by said operational amplifiers such that:

when the switch is closed, each synapse in a column is connected across the ground and the virtual ground so as to produce a photocurrent in the row which the synapse is located in, which photocurrent depends on the state of the synapse; and

when the switch is open, the column of synapses is left in an open circuit condition such that the synapses therein cannot contribute photocurrent to the rows.

9. A network according to claim 8, wherein said rows and columns are defined by a plurality of electrodes, and wherein said substance X is microfabricated to provide pixels of bacteriorhodopsin sandwiched between the electrodes which define the columns and the electrodes which define the rows.

10. A network according to claim 9, and further comprising a photovoltage or photocurrent sensing device connected from a column of pixels to a row of pixels so as to form a two dimensional light detector.

11. A method of storing information in an artificial neural network comprising neurons, synapses, switching an rectifying means, which contains synapses containing a substance X which exists in a number of energy states, which comprises irradiating any of the synapses to convert part or all of the substance X to stable state M, converting the remaining part of X, if such exists to state K and reverting that part from K to the initial state, and measuring the electrical signal of the conversion to K and back, erasure of the information in the synapse being effected by reversing substance from M to the initial state.

12. A method according to claim 11, where the substance X is bacteriorhodopsin.

13. A method according to claim 12, where the conversion from the initial state to M is by means of irradiation in the bacteriorhodopsin absorption range;

where the conversion from initial state to state K and back is by means of a two-pulse sequence with a certain time interval between the pulses where the wavelength of the first pulse is in the bacteriorhodopsin absorption and the second pulse is in the K absorption.

14. A method according to claim 11, where the neurons are updated in parallel (at the same time) and further comprising the step of rectifying currents from each of the synapses in order to eliminate cross-talk which develops through the synapse and between neurons as a result of such parallel updating.

15. A method according to claim 11, where the neurons are updated serially either by predetermined path or at random, such that no cross-talk appears.

16. A network according to claim 15, wherein said substance X is microfabricated to provide pixels of bacteriorhodopsin sandwiched between an electroluminescent or photoluminescent material and a plurality of electrodes.

17. A network according to claim 16, wherein each pixel is provided with a light source which is selectively activated during data retrieval.

18. A network according to claim 17, and further comprising means for applying appropriately timed voltages over said pixels during data storage, such that data storage can be achieved substantially in parallel.

19. A network according to claim 18, and further comprising data retrieval means for illuminating the entire network with a single light pulse to thereby produce a photovoltage at each pixel which erases part of each pixel and allows data retrieval, and means for compensating in future data retrievals for the percentage of the substance X in the pixels that has been transformed in the previous retrieval operation.

20. A network according to claim 19, and further comprising a connection between the neurons of the network and other networks for performing computational tasks.

21. A network according to claim 16, wherein said substance X comprises mutant or chemically modified protein retinal complexes.