Neural networks containing variable resistors as synapses

• US 5,422,982 A
• Filed: 01/06/1993
• Issued: 06/06/1995
• Est. Priority Date: 05/02/1991
• Status: Expired due to Fees

• Current Assignee: Dow, Inc.
• Original Assignee: Dow, Inc.
• Inventors: Pernisz, Udo C.
• Primary Examiner(s): Downs, Robert W.
• Application Number:
  US08/002,355
• Time in Patent Office: 881 Days
• Field of Search: 307/201, 395/24, 395/25, 364/807
• US Class Current: 706/33
• CPC Class Codes: G06N 3/063 using electronic means
• CPC Class Codes: H10B 63/82 the switching components having a common active material layer
• CPC Class Codes: H10N 70/021 Formation of the switching material, e.g. layer deposition
• CPC Class Codes: H10N 70/20 Multistable switching devices, e.g. memristors
• CPC Class Codes: H10N 70/826 adapted for essentially vertical current flow, e.g. sandwich or pillar type devices
• CPC Class Codes: H10N 70/883 Oxides or nitrides

Abstract

A synthetic neural network having a plurality of neuronal elements arranged in an input layer, an output layer, and a hidden layer between the input layer and the output layer. The network has a first plurality of synaptic weighting elements interconnecting the neuronal elements of the input layer with the neuronal elements of the hidden layer, and a second plurality of synaptic weighting elements interconnecting the neuronal elements of the hidden layer with the neuronal elements of the output layer. The improvement involves the synaptic weighting elements in the synthetic neural network being in the form of a silicon dioxide film derived from a hydrogen silsesquioxane resin. Such a silicon dioxide film is characterized by a jV curve which includes both linear and non-linear regions.

Claims

1. In a synthetic neural network including a plurality of neuronal elements arranged in an input layer, an output layer, and a hidden layer between the input layer and the output layer, the network further including a first plurality of synaptic weighting elements interconnecting the neuronal elements of the input layer with the neuronal elements of the hidden layer, and a second plurality of synaptic weighting elements interconnecting the neuronal elements of the hidden layer with the neuronal elements of the output layer, the improvement comprising the synaptic weighting elements in the synthetic neural network being in the form of a silicon dioxide film derived from the oxidation of a hydrogen silsesquioxane resin, the silicon dioxide film being characterized by a jV (wherein jV designates current-voltage) curve which includes both linear and non-linear regions, the jV curve of the silicon dioxide film including:

(i) a first non-linear region wherein increasing the voltage applied to the film increases the current to a current maximum;

followed by (ii) a second non-linear region of a negative differential resistance wherein increasing voltage applied to the film decreases the current to a current minimum while the voltage is increasing;

the jV curve of the silicon dioxide film including (iii) a third region capable of being activated by a voltage in excess of the voltage at the current minimum, at which time a rapid removal of the excess voltage causes the current to decrease in a linear fashion; and

the jV curve of the silicon dioxide film including (iv) a fourth region wherein an increase of voltage applied to the film up to a threshold voltage causes a rapid transition from a linear current increase to a non-linear current increase which continues to the current maximum.

2. The network of claim 1 wherein the film thickness is between about 50 and about 5000 nm.

3. The network of claim 2 wherein the film thickness is between about 100 and about 600 nm.

4. The network of claim 1 wherein one or more modifying ceramic oxides are incorporated in the film.

5. The network of claim 4 wherein the modifying ceramic oxide is an element selected from the group consisting of iron, titanium, zirconium, aluminum, tantalum, vanadium, niobium, boron and phosphorous, present in an amount such that the film contains 0.1 to 30 percent by weight of the modifying ceramic oxide.

6. An array comprising a first plurality of neuronal elements arranged in a first layer and a second plurality of neuronal elements arranged in a second layer, the first layer and the second layer of neuronal elements each having inputs and outputs, row electrodes and column electrodes connected to the inputs and outputs of the first and second layers of neuronal elements, and at least one synaptic weighting element in the form of a variable resistive element connected between the output of a neuronal element in the first layer and the input of a neuronal in the second layer, the variable resistive element being a non-fully densified silicon dioxide film derived from a hydrogen silsesquioxane resin and including electrical contacts for applying an electrical potential across the film, the silicon dioxide film being characterized by a conductive and a resistive state between which transitions can be induced by the application of voltage pulses, the silicon dioxide film being further characterized by a jV (wherein jV designates current-voltage) curve which includes both linear and non-linear regions for the resistive element, the jV curve of the silicon dioxide film including:

(i) a first non-linear region wherein increasing the voltage applied to the film increases the current to a current maximum;

the first non-linear region being followed by (ii) a second non-linear region of a negative differential resistance wherein increasing the voltage applied to the film decreases the current to a current minimum while the voltage is increasing;

the jV curve of the silicon dioxide film including (iii) a third region capable of being activated by a voltage in excess of the voltage at the current minimum, at which time a rapid removal of the excess voltage causes the current to decrease in a linear fashion; and

the jV curve of the silicon dioxide film including (iv) a fourth region wherein an increase of voltage applied to the film up to a threshold voltage causes a rapid transition from a linear current increase to a non-linear current increase which continues to the current maximum.

7. The array according to claim 6 in which the variable resistive element is placed in a non-oxidizing environment.

8. The array according to claim 6 in which there are a plurality of said variable resistive elements connected between the outputs of the first layer and the inputs of the second layer.

9. The array according to claim 6 in which the output of each neuronal element in the array is connected to its input.

10. The array according to claim 9 including control circuitry connected to the array for changing the adaptive synaptic weight of the synaptic weighting elements in the array by the application of voltage pulses.

11. The array according to claim 6 in the form of a network in which a third plurality of neuronal elements form a hidden layer with the hidden layer being disposed between and interconnected with the first layer and the second layer of neuronal elements.

12. The network according to claim 11 including control circuitry connected to the array for changing the adaptive synaptic weight of the synaptic weighting elements in the array by the application of voltage pulses.

13. The network according to claim 12 including auxiliary control circuitry interconnected with the array for the control of WRITE and ADDRESS SELECT.