# US 2021/0406660 A1 · Synaptic Resistors for Concurrent Parallel Signal Processing, Memory and Learning with High Speed and Energy Efficiency

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0406660 A1
- Document ID: US20210406660A1
- Status: Published application
- Kind: A1
- Application number: 16/961,602
- Filed: 2019-01-24
- Published: 2021-12-30
- Earliest claimed: 2018-01-24
- Assignees: The Regents of the University of California
- Applicants: The Regents of the University of California
- Inventors: Yong Chen
- Web page: https://patentgrep.com/patent/US20210406660A1

## CPC

- G06N 3/065: Analogue means
- B82Y 10/00: Nanotechnology for information processing, storage or transmission, e.g. quantum computing or single electron logic
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/049: Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G11C 11/54: using elements simulating biological cells, e.g. neuron
- H10D 64/64: Electrodes comprising a Schottky barrier to a semiconductor
- H10K 85/221: {Carbon nanotubes}
- H10N 70/20: Multistable switching devices, e.g. memristors
- H10N 70/231: {based on solid-state phase change, e.g. between amorphous and crystalline phases, Ovshinsky effect}
- H10N 70/253: {having three or more electrodes, e.g. transistor-like devices}
- H10N 70/823: {adapted for essentially horizontal current flow, e.g. bridge type devices}
- H10N 70/8822: {Sulfides, e.g. CuS}
- H10N 70/8825: {Selenides, e.g. GeSe}
- H10N 70/8828: {Tellurides, e.g. GeSbTe}
- H10N 70/8833: {Binary metal oxides, e.g. TaOx}
- H10N 70/884: {based on at least one element of group IIIA, IVA or VA, e.g. elemental or compound semiconductors (compounds of sulfur, selenium or tellurium, e.g. chalcogenides H10N70/882; oxides or nitrides H10N70/883)}
- H10N 70/8845: {Carbon or carbides}
- G06N 3/048: Activation functions
- H10D 62/882: {Graphene}
- H10K 10/26: Diodes comprising organic-organic junctions
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- B82Y: SPECIFIC USES OR APPLICATIONS OF NANOSTRUCTURES; MEASUREMENT OR ANALYSIS OF NANOSTRUCTURES; MANUFACTURE OR TREATMENT OF NANOSTRUCTURES
- G11C: STATIC STORES (semiconductor memory devices H10B)
- H10D: INORGANIC ELECTRIC SEMICONDUCTOR DEVICES
- H10K: ORGANIC ELECTRIC SOLID-STATE DEVICES
- H10N: ELECTRIC SOLID-STATE DEVICES NOT OTHERWISE PROVIDED FOR

## Abstract

Synaptic resistors (synstors), and their method of manufacture and integration into exemplary circuits are provided. Synstors are configured to emulate the analog signal processing, learning, and memory functions of synapses. Circuits incorporating synstors are capable of performing signal processing and learning concurrently in parallel analog mode with speed, energy efficiency, and functions superior to computers.

## Claims

### Claim 1

A synstor, capable of providing analog signal processing, memory and learning functions of synapse, comprising:
an input electrode and an output electrode;
a semiconducting channel connected between the input and output electrodes;
a dielectric layer disposed adjacent to the semiconducting channel, the dielectric layer disposed adjacent to the single side of the channel or the multiple sides of the channel;
a reference electrode disposed adjacent to the dielectric layer;
a charge storage material disposed within a portion of the dielectric layer;
wherein at least a portion of the dielectric layer is disposed between the charge storage material and the semiconducting channel such that a voltage difference between the channel and the charge storage material must have a magnitude above a threshold value to drive charge through the dielectric layer to the charge storage material to alter a net charge within the charge storage material in an analog mode;
wherein the combination of the input and output electrodes and the semiconducting channel between form resistors, and the combination of the semiconducting channel, dielectric layer, charge storage material, and reference electrode form capacitors;
wherein the synstor is configured to apply a zero or a constant voltage to the reference electrode, and when no voltage signals are applied on the input and output electrodes, a standby zero voltage or a standby constant voltage is applied on the input and output electrodes; and
wherein the synstor is further configured such that sets of voltage signals may be applied on the input and output electrodes with respect to the standby voltage such that:
when a set of input voltage signals is applied on the input electrode and wherein no voltage signal is applied on the output electrode, the set of input voltage signals induce an output current on the output electrode through the resistors and charges the capacitors such that, after the set of input voltage signals end, the charge in the capacitors is discharged and induces an output current on the output electrode, and wherein the output current may last after the set of input voltage signals ends,
when a set of voltage signals is applied on the input electrode and wherein a set of voltage signals with the same or similar amplitudes as the set of the input voltage signals is applied to the output electrode simultaneously, the magnitude of the voltage difference between the channel and the charge storage material produced thereby is above the threshold value, charge is driven through the portion of the dielectric layer disposed between the channel and the charge storage material to modify the net charge within the charge storage material and the conductance of the synstor in an analog mode for learning,
when a set of voltage signals is applied to one of either the input or the output electrode and wherein no voltage signal or a set of voltage signals with the opposite polarity is applied to the other of either the input or output electrode simultaneously, or when no voltage signals is applied to either the input or the output electrode, the magnitude of the voltage difference between the channel and the charge storage material produced thereby falls below the threshold value such that charge is not driven through the portion of the dielectric layer disposed between the channel and the charge storage material such that the net charge in the charge storage material remains unchanged and the conductance of the synstor remains unchanged for learning, and
when a set of voltage signals is applied to one of either the input or the output electrode with respect to the standby voltage and wherein no voltage signal or a set of voltage signals with the opposite polarity is applied to the other of either the input or output electrode simultaneously, or when no voltage signals is applied to either the input or the output electrode, or when no external voltages is applied to either the input or the output or the reference electrode, for the same reason, the conductance of the synstor remains unchanged as memory.

### Claim 2 (depends on 1)

The synstor of claim 1, the voltage signals applied to the input and output electrodes can be voltage pulses with fixed amplitudes and temporal durations for signal processing and learning.

### Claim 3 (depends on 1)

The synstor of claim 1, wherein the set of the input voltage signals induces a dynamic output current for signal processing, and the dynamic output current is the convolution of a kernel function and the product of the set of input voltage signals and the synstor conductance.

### Claim 4 (depends on 1)

The synstor of claim 1, wherein the change rate of the conductance of the synstor is equal to the product of the conductance modification coefficient, the input voltage signal, and the output voltage signal.

### Claim 5 (depends on 1)

The synstor of claim 1, wherein the input and output electrodes and the semiconducting channel form a contact with a contact resistance, which is comparable with or larger than the resistance of the channel.

### Claim 6 (depends on 1)

The synstor of claim 1, wherein the semiconducting channel forms Schottky barriers with the input and output electrodes, and wherein the input and output electrodes and the channel form a contact with a contact resistance which is comparable with or larger than the resistance of the channel.

### Claim 7 (depends on 1)

The synstor of claim 1, wherein the channel is dimensioned to have a length between the input and output electrodes longer than the length of the reference electrode.

### Claim 8 (depends on 1)

The synstor of claim 1, wherein the reference electrode extends beyond the edges of the portion of the dielectric layer containing the charge storage material.

### Claim 9 (depends on 1)

The synstor of claim 1, wherein the thickness of the dielectric layer between the charge storage material and the channel is less than the thickness of the dielectric layer between the charge storage material and the reference electrode.

### Claim 10 (depends on 1)

The synstor of claim 1, wherein the semiconducting channel comprises a semiconducting material selected from the group consisting of carbon nanotube, Si, graphene, Ge, SiC, ZnO, InO, InP, TiO2, Cu2O, GaN, GaAs, MoS2, MoSe2, WS2, WSe2, GaSe, GaTe, FeTe, polymers, molecules, and combinations thereof.

### Claim 11 (depends on 1)

The synstor of claim 1, wherein the semiconducting channel has a channel length between about 5-106 nm, a channel width between about 5-106 nm, and a channel thickness between about 0.1-105 nm.

### Claim 12 (depends on 1)

The synstor of claim 1, wherein the input, output and reference electrodes comprise a conducting or semiconducting material selected from the group consisting of Ti, Al, Au, Ni, Pt, Cu, carbon nanotube, Si, graphene, Ge, SiC, ZnO, InO, InP, TiO2, Cu2O, GaN, GaAs, MoS2, MoSe2, WS2, WSe2, GaSe, GaTe, FeTe, polymers, and combinations thereof.

### Claim 13 (depends on 1)

The synstor of claim 1, wherein the dielectric layers comprises an insulative material selected from the group consisting of HfO2, Al2O3, SiO2, Si3N4, Si, C, Ge, SiC, ZnO, InO, InP, TiO2, Cu2O, GaN, GaAs, polymers, molecules, and combinations thereof.

### Claim 14 (depends on 1)

The synstor of claim 1, wherein the dielectric layers have a dielectric length between about 5-106 nm, a dielectric width between about 5-106 nm, and a dielectric thickness between about 0.2-103 nm.

### Claim 15 (depends on 1)

The synstor of claim 1, wherein the charge storage material is selected from the group consisting of molecules, nanoparticles, semiconductor quantum dots, dopants, implanted ions, defects, vacancies, impurities, semiconducting materials, dielectric materials, and metals.

### Claim 16 (depends on 1)

The synstor of claim 1, wherein the charge storage material is within a volume with a length between about 5-106 nm, a width between about 5-106 nm, and a thickness between about 0.1-103 nm.

### Claim 17 (depends on 1)

The synaptic resistor of claim 1, wherein:
the semiconducting channel extends laterally beyond the edges of the portion of the first dielectric layer containing the charge storage material;
the semiconducting channel extends laterally beyond the edges of the reference electrode;
the reference electrode extends laterally beyond the edges of the portion of the first dielectric layer containing the charge storage material; and the thickness of the dielectric layer between the charge storage material and the channel is less than the thickness of the dielectric layer between the charge storage material and the reference electrode.

### Claim 18

A synstor capable of providing analog signal processing, memory and learning functions of biological synapse, comprising:
a body connected with an input electrode, an output electrode, and a reference electrode;
wherein the synstor is configured to apply a zero or a constant voltage to the reference electrode, and when no sets of voltage signals are applied on the input and output electrodes, a standby zero voltage or a standby constant voltage is applied on the input and output electrodes; and
wherein the synstor is configured such that sets of voltage signals may be applied on the input and output electrodes with respect to the standby voltage such that:
when a set of input voltage signals is applied on the input electrode with respect to the standby voltage and wherein no voltage signal is applied on the output electrode, the set of the input voltage signals induces a dynamic output current on the output electrode for signal processing,
when a set of voltage signals is applied on the input electrode and wherein a set of voltage signals with the same or similar amplitudes as the set of the input voltage signals is applied to the output electrode simultaneously, the conductance of the synstor is changed in analog mode; when a set of voltage signals is applied to one of either the input or the output electrode and wherein no voltage signal or a set of voltage signals with the opposite polarity is applied to the other of either the input or output electrode simultaneously, or when no voltage signals is applied to either the input or the output electrode, the conductance of the synstor remains unchanged for learning, and
when a set of voltage signals is applied to one of either the input or the output electrode with respect to the standby voltage and wherein no voltage signal or a set of voltage signals with the opposite polarity is applied to the other of either the input or output electrode simultaneously, or when no voltage signals is applied to either the input or the output electrode, or when no external voltages is applied to either the input or the output or the reference electrode, the conductance of the synstor remains unchanged as memory.

### Claim 19 (depends on 18)

The synstor of claim 18, wherein the dynamic output current is a convolution of a kernel function and the product of the input voltage signals and the synstor conductance.

### Claim 20 (depends on 18)

The synstor of claim 18, wherein the change rate of the conductance of the synstor is the product of the conductance modification coefficient, the input voltage signal, and the output voltage signal.

### Claim 21

A synaptic circuit capable of providing analog signal processing, memory and learning functions comprising:
a plurality of synstors, each in turn comprising a body connected with an input electrode, an output electrode, and a reference electrode;
wherein each synstor is configured to apply a zero or a constant voltage to the reference electrode, and when no sets of voltage signals are applied on the input and output electrodes, a standby zero voltage or a standby constant voltage is applied on the input and output electrodes; and
wherein each synstor is configured such that sets of voltage signals may be applied on the input and output electrodes with respect to the standby voltage, thereof such that:
when a set of input voltage signals is applied on the input electrode with respect to the standby voltage and wherein no voltage signal is applied on the output electrode, the set of the input voltage signals induces a dynamic output current on the output electrode for signal processing,
when a set of voltage signals is applied on the input electrode and wherein a set of voltage signals with the same or similar amplitudes as the set of the input voltage signals is applied to the output electrode simultaneously, the conductance of the synstor is changed in analog mode; when a set of voltage signals is applied to one of either the input or the output electrode and wherein no voltage signal or a set of voltage signals with the opposite polarity is applied to the other of either the input or output electrode simultaneously, or when no voltage signals is applied to either the input or the output electrode, the conductance of the synstor remains unchanged for learning, and
when a set of voltage signals is applied to one of either the input or the output electrode with respect to the standby voltage and wherein no voltage signal or a set of voltage signals with the opposite polarity is applied to the other of either the input or output electrode simultaneously, or when no voltage signals is applied to either the input or the output electrode, or when no external voltages is applied to either the input or the output or the reference electrode, the conductance of the synstor remains unchanged as memory; and
wherein each of the plurality of synstors is a component of a circuit comprising an M×N network of said synstors connected with M input electrodes and N output electrodes such that each input electrode is connected with multiple output electrodes by one or more synstors such that a set of voltage signals applied on the M input electrodes induces dynamic output currents via synstors flowing to the N output electrodes in analog parallel mode for signal processing, wherein the dynamic output currents on the N output electrodes define a current vector which is equal to the convolution of kernel functions and the product of the synstor conductance matrix and the vector of the input voltage signals on the M input electrodes;
wherein the circuit is configured such that when a set of voltage signals is applied on the M input electrodes and a set of voltage signals is applied on the N output electrodes, concurrently the synstor conductance matrix is modified in an analog parallel mode for learning, wherein the modification rate of the synstor conductance matrix is proportional to the outer product between the vectors of input and output voltage signals; and
wherein concurrently, the circuit is configured to process the voltage signals on the input electrodes and is modified by the voltage signals on the input and output electrodes.
