LEARNING METHOD OF NEURAL NETWORK CIRCUIT

• US 20130311415A1
• Filed: 07/25/2013
• Published: 11/21/2013
• Est. Priority Date: 01/20/2012
• Status: Active Grant

• Current Assignee: Panasonic Intellectual Property Management Co., Ltd. (Panasonic Holdings Corporation)
• Original Assignee: Panasonic Corporation (Panasonic Holdings Corporation)
• Inventors: NISHITANI, Yu, KANEKO, Yukihiro, UEDA, Michihito
• Granted Patent: US 8,775,346 B2
• Time in Patent Office: Days
• US Class Current: 706/26
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
• CPC Class Codes: G06N 3/063 using electronic means

Abstract

A neuron circuit in a neural network circuit element includes a waveform generating circuit for generating a bipolar sawtooth pulse voltage, and a first input signal has a bipolar sawtooth pulse waveform. For a period during which the first input signal is permitted to be input to a first electrode of a variable resistance element, the bipolar sawtooth pulse voltage generated within the neural network circuit element including the variable resistance element which is applied with the first input signal from another neural network circuit element is input to a control electrode of the variable resistance element. The resistance value of the variable resistance element changes due to an electric potential difference between the first electrode and the control electrode, the electric potential difference being generated depending on an input timing difference between a voltage applied to the first electrode and the voltage applied to the control electrode.

Claims

1. A learning method of a neural network circuit including a plurality of neural network circuit elements which are interconnected,

wherein each of the plurality of neural network circuit elements includes:

at least one synapse circuit which receives as an input a signal (first input signal) output from another neural network circuit element; and

one neuron circuit which receives as an input a signal output from the at least one synapse circuit,

wherein the synapse circuit includes a variable resistance element which includes a first electrode formed on and above a semiconductive layer;

a second electrode formed on and above the semiconductive layer; and

a control electrode formed on a main surface of the semiconductive layer via a ferroelectric layer and changes a resistance value between the first electrode and the second electrode in response to an electric potential difference between the first electrode and the control electrode;

wherein the synapse circuit is configured to perform switching between a state in which the first input signal is permitted to be input to the first electrode of the variable resistance element, and a state in which the first input signal is inhibited from being input to the first electrode;

wherein the neuron circuit includes a waveform generating circuit for generating a bipolar sawtooth pulse voltage, and the first input signal has a bipolar sawtooth pulse waveform;

wherein for a period during which the first input signal is permitted to be input to the first electrode, the bipolar sawtooth pulse voltage generated within the neural network circuit element including the variable resistance element which is applied with the first input signal from another neural network circuit element is input to the control electrode of the variable resistance element; and

wherein the resistance value of the variable resistance element changes due to an electric potential difference between the first electrode and the control electrode, the electric potential difference being generated depending on an input timing difference between a voltage applied to the first electrode and the voltage applied to the control electrode.

2. The learning method of the neural network circuit according to claim 1,

wherein the synapse circuit includes a first switch for performing switching between a state in which the first input signal input to the synapse circuit is permitted to be input to the first electrode of the variable resistance element and a state in which the first input signal input to the synapse circuit is inhibited from being input to the first electrode of the variable resistance element;

and

wherein the neuron circuit generates a second input signal for causing the first switch to perform switching to form a state in which the first input signal input to the synapse circuit is permitted to be input to the first electrode of the variable resistance element for a period during which the bipolar sawtooth pulse voltage is generated.

3. The learning method of the neural network circuit according to claim 1,

wherein the neuron circuit includes:

an integrating circuit for integrating a value of the current flowing through the variable resistance element included in the at least one synapse circuit; and

a delay circuit for delaying the bipolar sawtooth pulse voltage generated in the waveform generating circuit for a predetermined time and outputting the delayed bipolar sawtooth pulse voltage as a signal output from the neural network circuit element;

wherein the waveform generating circuit generates the bipolar sawtooth pulse voltage when the integrated value of the current which is generated in the integrating circuit exceeds a predetermined value.

4. The learning method of the neural network circuit according to claim 1,

wherein the synapse circuit includes a second switch one end of which is connected to a first reference voltage source and the other end of which is connected to the first electrode of the variable resistance element;

and

wherein the second switch connects the first reference voltage source to the first electrode of the variable resistance element for a period during which the first input signal is input to the synapse circuit.

5. A neural network circuit comprising a plurality of neural network circuit elements which are interconnected,

wherein each of the plurality of neural network circuit elements includes:

at least one synapse circuit which receives as an input a signal (first input signal) output from another neural network circuit element; and

one neuron circuit which receives as an input a signal output from the at least one synapse circuit,

wherein the synapse circuit includes a variable resistance element which includes a first electrode formed on and above a semiconductive layer;

a second electrode formed on and above the semiconductive layer; and

a control electrode formed on a main surface of the semiconductive layer via a ferroelectric layer, and changes a resistance value between the first electrode and the second electrode in response to an electric potential difference between the first electrode and the control electrode;

wherein the synapse circuit is configured to perform switching between a state in which the first input signal is permitted to be input to the first electrode of the variable resistance element, and a state in which the first input signal is inhibited from being input to the first electrode;

wherein the neuron circuit includes a waveform generating circuit for generating a bipolar sawtooth pulse voltage, and the first input signal has a bipolar sawtooth pulse waveform;

wherein for a period during which the first input signal is permitted to be input to the first electrode, the bipolar sawtooth pulse voltage generated within the neural network circuit element including the variable resistance element which is applied with the first input signal from another neural network circuit element is input to the control electrode of the variable resistance element; and

wherein the resistance value of the variable resistance element changes due to an electric potential difference between the first electrode and the control electrode, the electric potential difference being generated depending on an input timing difference between a voltage applied to the first electrode and the voltage applied to the control electrode.