CN109472348A - A kind of LSTM nerve network system based on memristor crossed array - Google Patents
A kind of LSTM nerve network system based on memristor crossed array Download PDFInfo
- Publication number
- CN109472348A CN109472348A CN201811236611.0A CN201811236611A CN109472348A CN 109472348 A CN109472348 A CN 109472348A CN 201811236611 A CN201811236611 A CN 201811236611A CN 109472348 A CN109472348 A CN 109472348A
- Authority
- CN
- China
- Prior art keywords
- memristor
- voltage
- crossed array
- threshold value
- feature extraction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Neurology (AREA)
- Semiconductor Memories (AREA)
- Semiconductor Integrated Circuits (AREA)
- Thin Film Transistor (AREA)
Abstract
The invention discloses a kind of LSTM nerve network systems based on memristor crossed array, it include: input layer, feature extraction layer and classification output layer, feature extraction layer includes: data storage, the first memristor crossed array, the first D/A converter and converter, and the classification output layer includes: the second D/A converter, the second memristor crossed array and voltage comparator；The first memristor crossed array, for carrying out feature extraction；The second memristor crossed array, for carrying out tagsort, the voltage comparator obtains the comparison result of multiple analog voltages for being compared to multiple analog voltages；Using the maximum value in comparison result as the classification results of the digital signal of input layer.Present system is smaller, and power consumption is lower.
Description
Technical field
The invention belongs to field of neural networks, more particularly, to a kind of LSTM nerve net based on memristor crossed array
Network system.
Background technique
Convolutional neural networks (CNN), full convolutional network (FCN) and long-term short-term memory (LSTM) even depth neural network
Development promotes the research of deep neural network hardware design.In deep neural network hardware circuit design field, due to a large amount of
Input data (such as image or text), reasons, the common CMOS projected depth such as the complicated network structure, network parameter be numerous
The problems such as size that neural network faces circuit is too big, and power consumption is high.
2008, the researcher of Hewlett-Packard made the memristor of nano-scale, neuromorphic computing system reality for the first time
Basic challenge be artificial synapse development, and memristor is found to be us and designs neuromorphic computing architecture and provide perfection
The element agreed with.Memristor, non-volatile due to its nano-grade size, high density, low-power consumption and compatible with CMOS technology
Property etc. excellent properties, shown its neural network hardware design wide application prospect.
However, since LSTM neural network has very big otherness with CNN in structure and operation logic, it is entire
It is no longer practical in LSTM based on the hardware circuit of convolutional neural networks, while the prior art there are sizes larger, power consumption
Larger technical problem.
Summary of the invention
Aiming at the above defects or improvement requirements of the prior art, the present invention provides a kind of based on memristor crossed array
Thus LSTM nerve network system solves the prior art there are sizes larger, the larger technical problem of power consumption.
To achieve the above object, the present invention provides a kind of LSTM nerve network system based on memristor crossed array, packets
It includes: input layer, feature extraction layer and classification output layer,
The feature extraction layer includes: data storage, the first memristor crossed array, the first D/A converter and AD conversion
Device, wherein data storage, the digital signal after digital signal and feature extraction for storing input layer；First DA conversion
Device, for converting the first analog voltage for the digital signal of input layer；First memristor crossed array, for the first simulation electricity
Pressure carries out feature extraction, obtains the first electric current；Converter is believed for converting the number after feature extraction for the first electric current
Number；
The classification output layer includes: the second D/A converter, the second memristor crossed array and voltage comparator；Wherein,
Two D/A converters, for converting the second analog voltage for the digital signal after feature extraction；Second memristor crossed array, is used for
Tagsort is carried out to the second analog voltage, obtains multiple analog voltages；Voltage comparator, for being carried out to multiple analog voltages
Compare, obtains the comparison result of multiple analog voltages；Using the maximum value in comparison result as point of the digital signal of input layer
Class result.
Further, the first memristor crossed array includes voltage input port, threshold value memristor, voltage inverter, operation
Amplifier and multiplier,
In two voltage input ports of arbitrary neighborhood, pass through voltage between a voltage input port and threshold value memristor
Phase inverter connection, a voltage input port are connected directly with threshold value memristor, and operational amplifier is in parallel with threshold value memristor；Electricity
The output end of one end and operational amplifier for pressing phase inverter connects, and the input terminal of the other end and multiplier connects.
Further, operational amplifier is in parallel with threshold value memristor, for realizing the calculation function of Sigmoid activation primitive
And current signal is converted into voltage signal.
Further, operational amplifier is in parallel with threshold value memristor, for realizing the operation function of tanh activation primitive
Can and current signal be converted into voltage signal.
Further, the second memristor crossed array includes voltage input port, threshold value memristor and voltage inverter, arbitrarily
In two adjacent voltage input ports, connected between a voltage input port and threshold value memristor by voltage inverter,
One voltage input port is connected directly with threshold value memristor.
Further, in order to avoid threshold value memristor short circuit, the intersection point of the first memristor crossed array is described without electrical connection
The intersection point of second memristor crossed array is without electrical connection.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, can obtain down and show
Beneficial effect:
(1) due to the high density of the nano-grade size of memristor and crossed array so that hardware circuit of the invention it is small in size,
It is easily integrated and preferably is promoted in deep neural network hardware design on a large scale.In addition, the low power consumption characteristic of memristor
So that the power consumption of whole system is substantially reduced compared to materials power consumptions such as CMOS.
(2) a usual threshold value memristor can only express a positive weight, and in the present invention arbitrary neighborhood voltage input
In port, connected between a voltage input port and threshold value memristor by voltage inverter, voltage input port with
Threshold value memristor is connected directly, and one can be expressed with positive and negative weight by being designed in this way two adjacent thresholds memristors.
Detailed description of the invention
Fig. 1 is shot and long term Memory Neural Networks schematic diagram provided in an embodiment of the present invention；
Fig. 2 is a kind of structure of LSTM nerve network system based on memristor crossed array provided in an embodiment of the present invention
Figure；
Fig. 3 is shot and long term Memory Neural Networks feature extraction layer unit schematic diagram provided in an embodiment of the present invention；
Fig. 4 is the circuit diagram of the first memristor crossed array provided in an embodiment of the present invention；
Fig. 5 is the circuit diagram of the second memristor crossed array provided in an embodiment of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below
Not constituting a conflict with each other can be combined with each other.
As shown in Figure 1, long-term short-term memory (LSTM) neural network includes: input layer, feature extraction layer and classification output
Layer.As shown in Fig. 2, a kind of LSTM nerve network system based on memristor crossed array, comprising: input layer, feature extraction layer and
Classification output layer, the feature extraction layer includes: data storage, first switch, the first reading circuit, write circuit, the first memristor
Crossed array, the first D/A converter and converter, the classification output layer include: second switch, the second reading circuit, the 2nd DA
Converter, the second memristor crossed array and voltage comparator；When first switch closure, second switch disconnect, system carries out special
Sign is extracted；When first switch disconnects, second switch is closed, system carries out classification output；
The data storage, the digital signal after digital signal and feature extraction for storing input layer；
First reading circuit for reading the digital signal of the input layer in data storage, and is transmitted to the first DA
Converter；
First D/A converter, for converting the first analog voltage for the digital signal of input layer；
The first memristor crossed array obtains the first electric current for carrying out feature extraction to the first analog voltage；
The converter, for converting the first electric current to the digital signal after feature extraction；
The write circuit, for data storage to be written in the digital signal after feature extraction；
Second reading circuit for reading the digital signal after the feature extraction in data storage, and is transmitted to
Two D/A converters；
Second D/A converter, for converting the second analog voltage for the digital signal after feature extraction；
The second memristor crossed array obtains multiple analog voltages for carrying out tagsort to the second analog voltage；
The voltage comparator obtains the comparison result of multiple analog voltages for being compared to multiple analog voltages；
Using the maximum value in comparison result as the classification results of the digital signal of input layer.
Fig. 3 illustrates the schematic diagram of LSTM neural network internal element (LSTM cell), c in figuretIt is LSTM cell
State, ct-1It is the state of last moment LSTM cell, htIt is the output of LSTM cell, ht-1It is last moment LSTM cell
Output.ct-1Data width be 1, ctData width be 64, Sigmoid indicate Sigmoid activation primitive, Tanh indicate
Tanh activation primitive, f indicate the output of forgetting door, and i indicates that input gate, j indicate to update door, and o indicates out gate, f_b table
Show forgetting biasing, btIndicate the biasing of forgetting door, biIndicate the biasing of input gate, bjIndicate the biasing of update door, boIndicate output
The biasing of door, WhfIndicate the state weight of forgetting door, WxfIndicate the input weight of forgetting door, WhiIndicate the state power of input gate
Weight, WxiIndicate the input weight of input gate, WhjIndicate the state weight of update door, WxjIndicate the input weight of update door, WhoTable
Show the state weight of out gate, WxoIndicate the input weight of out gate, X indicates the input of LSTM cell.
It is as follows to be converted into mathematic(al) representation for characteristic extraction part in the schematic diagram of LSTM cell:
F=sigmoid (x*Wxf+ht-1*Whf+bf+f_b)
I=sigmoid (x*Wxi+ht-1*Whi+bi)
J=tanh (x*Wxj+ht-1*Whj+bj)
O=sigmoid (x*Wxo+ht-1*Who+bo)
Above data calculates the function of corresponding crossed array, is the completion function of auxiliary circuit below crossed array below
Mathematic(al) representation:
ct=ct-1*f+i*j
ht=o*tanh (ct)
Data in above formula are existed in the form of analog voltage in circuit.
As shown in figure 4, the first memristor crossed array includes the first backbone circuits and the first auxiliary circuit, the first backbone circuits
Including voltage input port 1, threshold value memristor 2, voltage inverter 3, operational amplifier 4, the first auxiliary circuit includes that operation is put
Big device 4, resistance 5 and multiplier 6, the intersection point of the first memristor crossed array is without electrical connection.
A kind of LSTM nerve network system based on memristor crossed array includes n LSTM cell, i.e., a kind of to be based on memristor
The LSTM nerve network system of crossed array includes n the first memristor crossed arrays, and the first analog voltage includes VX1-VXmAnd
Vh1-Vhn, the first analog voltage by voltage input port input the first backbone circuits.One threshold value memristor can only express one
A positive weight, and in the present invention in the voltage input port of arbitrary neighborhood, between a voltage input port and threshold value memristor
It is connected by voltage inverter, a voltage input port is connected directly with threshold value memristor, is designed in this way two adjacent thresholds
Memristor, which can express one, has positive and negative weight.
M1 and M2 is threshold value memristor, Vs1 +For 1V DC voltage, Vs1 -Ground connection, Vs2 +For 1V DC voltage, Vs2 -For -1V
DC voltage, resistance R1、R2、R3、R4And R5Middle resistance value needs are identical, can be used 1K ohm to 10K ohm.Operational amplifier and threshold
It is in parallel to be worth memristor M1, realize Sigmoid activation primitive and current signal is converted into voltage signal, operational amplifier and threshold
It is in parallel to be worth memristor M2, realize tanh activation primitive and current signal is converted into voltage signal.Voltage inverter
The input terminal of the connection of the output end of one end and operational amplifier, the other end and multiplier connects, for converting the direction of voltage.
Vc(t-1)Indicate the state of the first memristor of last moment crossed array, VctIndicate the state of the first memristor crossed array,
VhtIndicate the first electric current.The mathematic(al) representation of the completion function of first auxiliary circuit:
ct=ct-1*f+i*j
ht=o*tanh (ct)
As shown in figure 5, the second memristor crossed array includes the second backbone circuits and the second auxiliary circuit, the second backbone circuits
Including voltage input port, threshold value memristor and voltage inverter, in the voltage input port of arbitrary neighborhood, a voltage input
It is connected between port and threshold value memristor by voltage inverter, a voltage input port is connected directly with threshold value memristor.
The intersection point of second memristor crossed array is without electrical connection, and the second auxiliary circuit is a voltage comparator, the voltage comparator,
For being compared to multiple analog voltages, the comparison result of multiple analog voltages is obtained；Maximum value in comparison result is made
For the classification results V of the digital signal of input layero。
The present invention solves the problems, such as LSTM neural network in hardware design, while introducing memristor progress circuit design
The hardware implementation method including activation primitive is provided, The present invention gives LSTM neural networks based on memristor crossed array
Complete implementation.Due to the nano-grade size of memristor and the high density of crossed array, so that the hardware circuit volume of the design
It is small, be easily integrated and preferably promoted in deep neural network hardware design on a large scale.In addition, the low-power consumption of memristor is special
Property the power consumption of whole system is substantially reduced compared to materials power consumptions such as CMOS.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all include
Within protection scope of the present invention.
Claims (6)
1. a kind of LSTM nerve network system based on memristor crossed array, comprising: input layer, feature extraction layer and classification output
Layer, which is characterized in that
The feature extraction layer includes: data storage, the first memristor crossed array, the first D/A converter and converter,
In, data storage, the digital signal after digital signal and feature extraction for storing input layer；First D/A converter is used
In converting the first analog voltage for the digital signal of input layer；First memristor crossed array, for the first analog voltage into
Row feature extraction obtains the first electric current；Converter, for converting the first electric current to the digital signal after feature extraction；
The classification output layer includes: the second D/A converter, the second memristor crossed array and voltage comparator；Wherein, the 2nd DA
Converter, for converting the second analog voltage for the digital signal after feature extraction；Second memristor crossed array, for the
Two analog voltages carry out tagsort, obtain multiple analog voltages；Voltage comparator, for comparing multiple analog voltages
Compared with obtaining the comparison result of multiple analog voltages；Using the maximum value in comparison result as the classification of the digital signal of input layer
As a result.
2. a kind of LSTM nerve network system based on memristor crossed array as described in claim 1, which is characterized in that described
First memristor crossed array includes voltage input port, threshold value memristor, voltage inverter, operational amplifier and multiplier,
In two voltage input ports of arbitrary neighborhood, pass through voltage inversion between a voltage input port and threshold value memristor
Device connection, a voltage input port are connected directly with threshold value memristor, and operational amplifier is in parallel with threshold value memristor；Voltage is anti-
One end of phase device and the output end of operational amplifier connect, and the input terminal of the other end and multiplier connects.
3. a kind of LSTM nerve network system based on memristor crossed array as claimed in claim 2, which is characterized in that described
Operational amplifier is in parallel with threshold value memristor, turns for realizing the calculation function of Sigmoid activation primitive and by current signal
It is changed to voltage signal.
4. a kind of LSTM nerve network system based on memristor crossed array as claimed in claim 2, which is characterized in that described
Operational amplifier is in parallel with threshold value memristor, turns for realizing the calculation function of tanh activation primitive and by current signal
It is changed to voltage signal.
5. a kind of LSTM nerve network system based on memristor crossed array as described in claim 1, which is characterized in that described
Second memristor crossed array includes voltage input port, threshold value memristor and voltage inverter, and two voltages of arbitrary neighborhood are defeated
In inbound port, connected between a voltage input port and threshold value memristor by voltage inverter, a voltage input port
It is connected directly with threshold value memristor.
6. a kind of LSTM nerve network system based on memristor crossed array a method as claimed in any one of claims 1 to 5, feature exist
In the intersection point of the first memristor crossed array is without electrical connection, and the intersection point of the second memristor crossed array is without electrical connection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811236611.0A CN109472348B (en) | 2018-10-23 | 2018-10-23 | LSTM neural network system based on memristor cross array |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811236611.0A CN109472348B (en) | 2018-10-23 | 2018-10-23 | LSTM neural network system based on memristor cross array |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109472348A true CN109472348A (en) | 2019-03-15 |
| CN109472348B CN109472348B (en) | 2022-02-18 |
Family
ID=65664136
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811236611.0A Active CN109472348B (en) | 2018-10-23 | 2018-10-23 | LSTM neural network system based on memristor cross array |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109472348B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111460365A (en) * | 2020-03-10 | 2020-07-28 | 华中科技大学 | Equation set solver based on memristive linear neural network and operation method thereof |
| CN111680792A (en) * | 2020-06-18 | 2020-09-18 | 中国人民解放军国防科技大学 | Activation function circuit, memristor neural network and control method of memristor neural network |
| CN111755062A (en) * | 2019-03-26 | 2020-10-09 | 慧与发展有限责任合伙企业 | Self-repairing dot product engine |
| CN112749784A (en) * | 2019-10-31 | 2021-05-04 | 华为技术有限公司 | Computing device and neural network acceleration method |
| CN112884141A (en) * | 2021-04-16 | 2021-06-01 | 安徽大学 | Memristive coupling Hindmarsh-Rose neuron circuit |
| CN116523011A (en) * | 2023-07-03 | 2023-08-01 | 中国人民解放军国防科技大学 | Memristor-based binary neural network layer circuit and binary neural network training method |
| CN112749784B (en) * | 2019-10-31 | 2024-05-14 | 华为技术有限公司 | Computing device and acceleration method of neural network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140172937A1 (en) * | 2012-12-19 | 2014-06-19 | United States Of America As Represented By The Secretary Of The Air Force | Apparatus for performing matrix vector multiplication approximation using crossbar arrays of resistive memory devices |
| CN104916313A (en) * | 2015-06-16 | 2015-09-16 | 清华大学 | Neural network synapse structure based on memristive devices and synaptic weight building method |
| CN106815636A (en) * | 2016-12-30 | 2017-06-09 | 华中科技大学 | A kind of neuron circuit based on memristor |
| CN107273972A (en) * | 2017-05-11 | 2017-10-20 | 北京大学 | It is a kind of based on resistive device and to adapt to excite the neuromorphic system and implementation method of neuron |
-
2018
- 2018-10-23 CN CN201811236611.0A patent/CN109472348B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140172937A1 (en) * | 2012-12-19 | 2014-06-19 | United States Of America As Represented By The Secretary Of The Air Force | Apparatus for performing matrix vector multiplication approximation using crossbar arrays of resistive memory devices |
| CN104916313A (en) * | 2015-06-16 | 2015-09-16 | 清华大学 | Neural network synapse structure based on memristive devices and synaptic weight building method |
| CN106815636A (en) * | 2016-12-30 | 2017-06-09 | 华中科技大学 | A kind of neuron circuit based on memristor |
| CN107273972A (en) * | 2017-05-11 | 2017-10-20 | 北京大学 | It is a kind of based on resistive device and to adapt to excite the neuromorphic system and implementation method of neuron |
Non-Patent Citations (1)
| Title |
|---|
| KAMILYA SMAGULOVA等: "Design of CMOS-memristor Circuits for LSTM architecture", 《2018 IEEE INTERNATIONAL CONFERENCE ON ELECTRON DEVICES AND SOLID STATE CIRCUITS (EDSSC)》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111755062A (en) * | 2019-03-26 | 2020-10-09 | 慧与发展有限责任合伙企业 | Self-repairing dot product engine |
| US11532356B2 (en) | 2019-03-26 | 2022-12-20 | Hewlett Packard Enterprise Development Lp | Self-healing dot-product engine |
| CN112749784A (en) * | 2019-10-31 | 2021-05-04 | 华为技术有限公司 | Computing device and neural network acceleration method |
| CN112749784B (en) * | 2019-10-31 | 2024-05-14 | 华为技术有限公司 | Computing device and acceleration method of neural network |
| CN111460365A (en) * | 2020-03-10 | 2020-07-28 | 华中科技大学 | Equation set solver based on memristive linear neural network and operation method thereof |
| CN111460365B (en) * | 2020-03-10 | 2021-12-03 | 华中科技大学 | Equation set solver based on memristive linear neural network and operation method thereof |
| CN111680792A (en) * | 2020-06-18 | 2020-09-18 | 中国人民解放军国防科技大学 | Activation function circuit, memristor neural network and control method of memristor neural network |
| CN112884141A (en) * | 2021-04-16 | 2021-06-01 | 安徽大学 | Memristive coupling Hindmarsh-Rose neuron circuit |
| CN112884141B (en) * | 2021-04-16 | 2022-10-21 | 安徽大学 | Memristive coupling Hindmarsh-Rose neuron circuit |
| CN116523011A (en) * | 2023-07-03 | 2023-08-01 | 中国人民解放军国防科技大学 | Memristor-based binary neural network layer circuit and binary neural network training method |
| CN116523011B (en) * | 2023-07-03 | 2023-09-15 | 中国人民解放军国防科技大学 | Memristor-based binary neural network layer circuit and binary neural network training method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109472348B (en) | 2022-02-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109472348A (en) | A kind of LSTM nerve network system based on memristor crossed array | |
| Biswas et al. | Conv-RAM: An energy-efficient SRAM with embedded convolution computation for low-power CNN-based machine learning applications | |
| CN110352436B (en) | Resistance processing unit with hysteresis update for neural network training | |
| Park et al. | RRAM-based synapse for neuromorphic system with pattern recognition function | |
| CN106374912B (en) | A kind of logical operation circuit and operating method | |
| CN106971372B (en) | Coding type flash memory system and method for realizing image convolution | |
| Tang et al. | A hardware friendly unsupervised memristive neural network with weight sharing mechanism | |
| Pisarev et al. | 3D memory matrix based on a composite memristor-diode crossbar for a neuromorphic processor | |
| CN110007895B (en) | Analog multiplication circuit, analog multiplication method and application thereof | |
| KR101912881B1 (en) | Neuromorphic memristor crossbar circuit | |
| KR102196523B1 (en) | Floating gate memristor and neuromorphic device | |
| CN107545305A (en) | A kind of neuron circuit based on CMOS technology, numerical model analysis, charge-domain | |
| CN110651330A (en) | Deep learning in a two-memristive network | |
| Jarollahi et al. | Reduced-complexity binary-weight-coded associative memories | |
| CN114365078A (en) | Reconstructing MAC operations | |
| CN112465134A (en) | Pulse neural network neuron circuit based on LIF model | |
| WO2019096660A1 (en) | Competitive machine learning accuracy on neuromorphic arrays with capacitor memory devices | |
| CN109524039B (en) | Structure and related method for expanding resistance state number of memristor | |
| CN105741870B (en) | A kind of non-volatile d type flip flop circuit based on memristor | |
| Felder et al. | Reminding forgetful organic neuromorphic device networks | |
| CN114298297A (en) | Memory computing device, chip and electronic equipment | |
| DeSalvo et al. | Emerging resistive memories for low power embedded applications and neuromorphic systems | |
| Khorami et al. | One‐dimensional adiabatic circuits with inherent charge recycling | |
| Van Nguyen et al. | Comparative study on quantization-aware training of memristor crossbars for reducing inference power of neural networks at the edge | |
| CN110991628B (en) | Neuron circuit based on charge pump |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |