CN105787866A - Cell unit circuit and cellular neural network - Google Patents
Cell unit circuit and cellular neural network Download PDFInfo
- Publication number
- CN105787866A CN105787866A CN201610203007.2A CN201610203007A CN105787866A CN 105787866 A CN105787866 A CN 105787866A CN 201610203007 A CN201610203007 A CN 201610203007A CN 105787866 A CN105787866 A CN 105787866A
- Authority
- CN
- China
- Prior art keywords
- voltage
- cell
- neural network
- resistive device
- source
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/20—Processor architectures; Processor configuration, e.g. pipelining
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The invention discloses a cell unit circuit and a cellular neural network. The circuit comprises an independent voltage source, a metal oxide resistive device, a capacitor which is connected in parallel with the metal oxide resistive device, at least one voltage control current source which is connected in parallel with the metal oxide resistive device and has one end grounded and the other end used for receiving feedback voltage of the voltage control voltage sources of cell units around, an independent current source which is connected in parallel with the metal oxide resistive device and used for providing a bias current, and a voltage control voltage source output voltage source which is used for providing feedback voltage for the voltage control current sources of the cell units around. The invention further discloses a cellular neural network comprising the cell unit circuit. According to the cell unit circuit and the cellular neural network disclosed by the invention, the weight of the neural network can be updated without reconstructing the network, and multiple functions are achieved.
Description
Technical field
The application relates to cell neural network field, is specifically related to circuit and the nerve cell network of a kind of cell factory.
Background technology
1988, LeonO.Chua proposed cell neural network (CellularNeuralNetwork, CNN) on the basis of neutral net and cellular automaton.CNN has the main feature of neutral net, has important application in the field such as image procossing and pattern recognition.For image procossing, it has outstanding advantage: continuous time, feature made it have real time signal processing ability in digital field；Local interconnectivity makes it be applicable to VLSI realization, and is particularly suitable for high-speed parallel process；Its processing speed is unrelated with image scale.
In tradition CNN, neutral net connects weights after determining and fixes, thus is difficult to realize the renewal of weights under the premise not rebuilding network, it is impossible to realize multi-purpose function.
Summary of the invention
The application provides circuit and the nerve cell network of a kind of cell factory, it is possible to realize the renewal of weights under the premise not rebuilding network, thus realizing multi-purpose function.
First aspect according to the application, the application provides the circuit of a kind of cell factory, comprising:
Independent voltage source, one end ground connection of described independent voltage source；
Metal-oxide resistive device；
Electric capacity, it is in parallel with described metal-oxide resistive device；
At least one voltage-controlled current source, it is in parallel with described metal-oxide resistive device, one end ground connection of wherein said voltage-controlled current source, and the other end is for receiving the described voltage controlled voltage source feedback voltage of peripheral cell unit；
Independent current, it is in parallel with described metal-oxide resistive device, is used for providing bias current；And output voltage source, one end ground connection of described output voltage source, the other end of described output voltage source is used for
Feedback voltage is provided to the described voltage-controlled current source of described peripheral cell unit.
In the circuit of the cell factory of the application, the sense of current of described voltage-controlled current source is the one end flowing to its ground connection from its ungrounded one end.The sense of current of described independent current is the one end flowing to its ground connection from its ungrounded one end.
Second aspect according to the application, present invention also provides a kind of cell neural network, and it includes the circuit of above-mentioned cell factory.
Individual cells unit and the N number of described cell factory local interlinkage of surrounding, N=(2r+1) in the cell neural network of the application2-1, wherein r is positive integer.
The application provides the benefit that:
Owing to quoting metal-oxide resistive device, with the resistance in metal-oxide resistive device replacement circuit, thus the renewal of weights also can be realized under the premise not rebuilding network；Meanwhile, also improve integrated level, greatly reduce whole circuit area；Improve the processing speed of nerve cell network circuit；Reduce the power consumption of nerve cell network circuit.
Accompanying drawing explanation
Fig. 1 is the circuit diagram of the cell factory of the present invention；
Fig. 2 is artwork before treatment；
Fig. 3 is the noise pattern that artwork as shown in Figure 2 is added 25% spiced salt by MATLAB emulation；
Fig. 4 is the result figure that the noise pattern shown in Fig. 3 carries out denoising of the RCNN by the MATLAB Simulation Application present invention；And
Fig. 5 is the result figure that the artwork shown in Fig. 2 carries out rim detection of the RCNN by the MATLAB Simulation Application present invention.
Detailed description of the invention
The application is described in further detail in conjunction with accompanying drawing below by detailed description of the invention.
Fix for solving connection weights after neutral net is determined, the problem being difficult to realize the renewal of weights under the premise not rebuilding network, the application proposes circuit and the nerve cell network of a kind of cell factory, present invention design is in that, utilize metal-oxide resistive device under the effect of external electric field, its resistance can at height, reversible transition is there is between low resistance state, and after electric field is revoked, it still is able to the feature kept, metal-oxide resistive device is used to build the circuit of cell factory and follow-up nerve cell network, when needs carry out right value update, only need to apply suitable applied voltage amplitude at the two ends of metal-oxide resistive device and the time just can change the resistance of metal-oxide resistive device to realize the renewal of weights.
Refer to Fig. 1, in one embodiment, the circuit of cell factory includes independent voltage source 11, metal-oxide resistive device (MetalOxideResistiveDevice, MORD) 12, electric capacity 14, at least one voltage-controlled current source (VoltageControlCurrentSource, VCCS) 16, independent current 18 and output voltage source 19.It is specifically described below.
One end ground connection of independent voltage source 11.Metal-oxide resistive device 12 is in parallel with electric capacity 14, voltage-controlled current source 16 is in parallel with metal-oxide resistive device 12, and one end ground connection of voltage-controlled current source 16, the other end is for receiving the feedback voltage of peripheral cell unit voltage controlled voltage source 19, in one embodiment, the sense of current of voltage-controlled current source 16 is the one end flowing to its ground connection from its ungrounded one end.Independent current 18 is in parallel with metal-oxide resistive device 12, for providing bias current, therefore, as can see from Figure 1, independent electrical in a steady stream 18 one end be also ground connection, in one embodiment, the sense of current of independent current 18 also for flowing to one end of its ground connection from its ungrounded one end.One end ground connection of output voltage source 19, the other end provides feedback voltage for the voltage-controlled current source 16 of cell factory towards periphery.
The circuit of cell factory disclosed in the present application, builds circuit owing to introducing metal-oxide resistive device 12, therefore, it is possible to really do not rebuild the renewal realizing weights under the premise of network, and then realizes multi-purpose function.
Disclosed herein as well is a kind of cell neural network, this cell neural network includes the circuit of above-mentioned cell factory.In one embodiment, in the cell neural network of the application, individual cells and the N number of cell of surrounding carry out local interlinkage, N=(2r+1)2-1, wherein r is positive integer.
The cell factory of individual cells unit and surrounding carries out local interlinkage, for 8 cell factory interconnection of r=1, individual cells unit and surrounding.Each cell factory has oneself peculiar fixing input voltage (VS), one voltage (VCVS) of output simultaneously, this output voltage is connected on the voltage controling end of the voltage-controlled current source of other cell factory and other cell factory is formed feedback, can stablize through the state of this cell factory after a period of time.
The cell neural network of the present embodiment utilizes the connection weights that metal-oxide resistive device change in resistance characteristic is come between torage cell, different cloning module is obtained by changing metal-oxide resistive device resistance, construct the novel cell neural network (MORD-BasedCellularNeuralNetwork based on metal-oxide resistive device (MetalOxideResistiveDevice), RCNN), not only contribute to improve integrated level and parallel processing speeds, and power consumption can be reduced.
Refer to Fig. 2 to Fig. 4, in the present embodiment, utilize the RCNN of the present invention to carry out image denoising by MATLAB emulation.Wherein, Fig. 2 is artwork before treatment.Fig. 3 is the noise pattern that artwork as shown in Figure 2 is added 25% spiced salt by MATLAB emulation.Fig. 4 is the result figure that the noise pattern shown in Fig. 3 carries out denoising of the RCNN by the MATLAB Simulation Application present invention.Experimental result shows, when carrying out denoising with RCNN, speed is fast, power consumption is little, and denoising effect is preferably simultaneously.
Refer to Fig. 2 and Fig. 5, in the present embodiment, utilize the RCNN of the present invention to carry out rim detection by MATLAB emulation.Wherein, Fig. 5 is the result figure that the artwork shown in Fig. 2 carries out rim detection of the RCNN by the MATLAB Simulation Application present invention.Experimental result shows, when carrying out rim detection with RCNN, not only speed is fast, and precision is high, the place that aberration is only small in artwork, all can detect that edge.
Above content is further description the application made in conjunction with specific embodiment, it is impossible to assert the application be embodied as be confined to these explanations.For the application person of an ordinary skill in the technical field, under the premise conceived without departing from the present application, it is also possible to make some simple deduction or replace.
Claims (5)
1. a circuit for cell factory, is used for building cell neural network, it is characterised in that including:
Independent voltage source, one end ground connection of described independent voltage source；
Metal-oxide resistive device；
Electric capacity, it is in parallel with described metal-oxide resistive device；
At least one voltage-controlled current source, it is in parallel with described metal-oxide resistive device, one end ground connection of wherein said voltage-controlled current source, and the other end is for receiving the feedback voltage of the described voltage controlled voltage source of peripheral cell unit；
Independent current, it is in parallel with described metal-oxide resistive device, is used for providing bias current；And output voltage source, one end ground connection of described output voltage source, the other end of described output voltage source is for providing feedback voltage to the described voltage-controlled current source of described peripheral cell unit.
2. the circuit of cell factory as claimed in claim 1, it is characterised in that the sense of current of described voltage-controlled current source is the one end flowing to its ground connection from its ungrounded one end.
3. the circuit of cell factory as claimed in claim 1, it is characterised in that the sense of current of described independent current is the one end flowing to its ground connection from its ungrounded one end.
4. a cell neural network, it is characterised in that include the circuit of cell factory as described in claim 1-3 any one.
5. cell neural network as claimed in claim 4, it is characterised in that the individual cells unit of described cell neural network and the N number of described cell factory local interlinkage of surrounding, N=(2r+1)2-1, wherein r is positive integer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610203007.2A CN105787866A (en) | 2016-04-01 | 2016-04-01 | Cell unit circuit and cellular neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610203007.2A CN105787866A (en) | 2016-04-01 | 2016-04-01 | Cell unit circuit and cellular neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105787866A true CN105787866A (en) | 2016-07-20 |
Family
ID=56394646
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610203007.2A Pending CN105787866A (en) | 2016-04-01 | 2016-04-01 | Cell unit circuit and cellular neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105787866A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020154677A1 (en) * | 2001-01-12 | 2002-10-24 | Stmicroelectronics S.R.L. | Programmbale chaos generator and process for use thereof |
| CN103294872A (en) * | 2013-06-24 | 2013-09-11 | 杭州电子科技大学 | Memristor equivalent circuit and construction method thereof |
| CN103810497A (en) * | 2014-01-26 | 2014-05-21 | 华中科技大学 | Memristor based image identification system and method |
| CN104573238A (en) * | 2015-01-09 | 2015-04-29 | 江西理工大学 | Circuit design method for memory resisting cell neural network |
-
2016
- 2016-04-01 CN CN201610203007.2A patent/CN105787866A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020154677A1 (en) * | 2001-01-12 | 2002-10-24 | Stmicroelectronics S.R.L. | Programmbale chaos generator and process for use thereof |
| CN103294872A (en) * | 2013-06-24 | 2013-09-11 | 杭州电子科技大学 | Memristor equivalent circuit and construction method thereof |
| CN103810497A (en) * | 2014-01-26 | 2014-05-21 | 华中科技大学 | Memristor based image identification system and method |
| CN104573238A (en) * | 2015-01-09 | 2015-04-29 | 江西理工大学 | Circuit design method for memory resisting cell neural network |
Non-Patent Citations (4)
| Title |
|---|
| 张小红等: "基于二次型的CNN全局渐进稳定性研究", 《计算机科学》 * |
| 李德音: "基于矩阵不等式的细胞神经网络稳定性研究与电路实现", 《中国优秀硕士论文全文数据库（信息科技辑）》 * |
| 李志军等: "改进型细胞神经网络实现的忆阻器混沌电路", 《物理学报》 * |
| 高士勇等: "忆阻细胞神经网络及图像去噪和边缘提取中的应用", 《西南大学学报（自然科学版）》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Duan et al. | Small-world Hopfield neural networks with weight salience priority and memristor synapses for digit recognition | |
| US10055022B2 (en) | Simulating obstruction in a virtual environment | |
| Wang et al. | Synthesis of biquadratic impedances with at most four passive elements | |
| CN105160401A (en) | WTA neural network based on memristor array and application thereof | |
| CN111143702A (en) | Information recommendation method, system, device and medium based on graph convolution neural network | |
| Dlugosz et al. | Parallel programmable asynchronous neighborhood mechanism for Kohonen SOM implemented in CMOS technology | |
| Pershin et al. | Memcomputing with membrane memcapacitive systems | |
| CN108352985A (en) | Signal processing circuit and method | |
| Liu et al. | Reconfigurable neuromorphic computing system with memristor-based synapse design | |
| Alimisis et al. | Analog gaussian function circuit: Architectures, operating principles and applications | |
| Brandstetter et al. | Second order low-pass and high-pass filter designs using method of synthetic immitance elements | |
| Abd et al. | A compact four transistor CMOS-design of a floating memristor for adaptive spiking neural networks and corresponding self-X sensor electronics to industry 4.0 | |
| CN105787866A (en) | Cell unit circuit and cellular neural network | |
| Aatre | Network theory and filter design | |
| Pyle et al. | Self-Scaling Evolution of analog computation circuits with digital accuracy refinement | |
| US11625593B2 (en) | Fixed-weighting-code learning device | |
| CN106464239A (en) | Improving linearity of phase interpolators using capacitive elements | |
| Soleimani Abhari et al. | A novel median based image impulse noise suppression system using spiking neurons on FPGA | |
| DE102022134244A1 (en) | Digital-to-analog converter and method for reducing aging effects on digital-to-analog converter components | |
| Lee et al. | Robust touch screen readout system to display noise using multireference differential sensing scheme for flexible AMOLED display | |
| Soliman | History and progress of the Tow Thomas bi-quadratic filter part III: generation using NAM expansion | |
| Hu et al. | An adjustable memristor model and its application in small-world neural networks | |
| Chen et al. | An improved version of the multiple trajectory search for real value multi-objective optimization problems | |
| Kumar et al. | Duality of Graph and Its Application in Electric Circuit | |
| Gupta et al. | Active filter design using two OTA based floating inductance simulator |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160720 |