WO2014187059A1 - 基于阻变忆阻器的时间关联学习神经元电路及其实现方法 - Google Patents

基于阻变忆阻器的时间关联学习神经元电路及其实现方法 Download PDF

Info

Publication number
WO2014187059A1
WO2014187059A1 PCT/CN2013/084752 CN2013084752W WO2014187059A1 WO 2014187059 A1 WO2014187059 A1 WO 2014187059A1 CN 2013084752 W CN2013084752 W CN 2013084752W WO 2014187059 A1 WO2014187059 A1 WO 2014187059A1
Authority
WO
WIPO (PCT)
Prior art keywords
terminal
excitation signal
transmission gate
neuron
signal
Prior art date
Application number
PCT/CN2013/084752
Other languages
English (en)
French (fr)
Inventor
黄如
张耀凯
蔡一茂
杨帆
潘越
王宗巍
方亦陈
Original Assignee
北京大学
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 北京大学 filed Critical 北京大学
Priority to US14/892,130 priority Critical patent/US20160110644A1/en
Publication of WO2014187059A1 publication Critical patent/WO2014187059A1/zh

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/049Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/06Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
    • G06N3/063Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
    • G06N3/065Analogue means
    • GPHYSICS
    • G11INFORMATION STORAGE
    • G11CSTATIC STORES
    • G11C11/00Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor
    • G11C11/54Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using elements simulating biological cells, e.g. neuron
    • GPHYSICS
    • G11INFORMATION STORAGE
    • G11CSTATIC STORES
    • G11C13/00Digital stores characterised by the use of storage elements not covered by groups G11C11/00, G11C23/00, or G11C25/00
    • G11C13/0002Digital stores characterised by the use of storage elements not covered by groups G11C11/00, G11C23/00, or G11C25/00 using resistive RAM [RRAM] elements
    • G11C13/0007Digital stores characterised by the use of storage elements not covered by groups G11C11/00, G11C23/00, or G11C25/00 using resistive RAM [RRAM] elements comprising metal oxide memory material, e.g. perovskites

Definitions

  • the present invention relates to a neuron cell circuit, and more particularly to a time-dependent learning neuron circuit based on a resistive memristor and an implementation method thereof . Background technique
  • Digital computers are an important product of the progress of human science and technology civilization in the twentieth century, and their influence permeates all aspects of people's lives.
  • people have been unable to meet the functions of existing computers.
  • High computing speed, large storage capacity, and intelligence have become an inevitable trend in the further development of computers.
  • neural computers Because of its large-scale parallel processing, strong recognition ability, ability to process analog information, and machine self-learning, neural computers have become powerful substitutes for digital computers in the future.
  • the key to hardware manufacturing lies in good weights that can be integrated on a large scale. interconnected.
  • a large number of synaptic connections are required in neuronal cell circuits, and these synaptic connections must have variable weights and a small area to facilitate large-scale integration.
  • a resistive memristive device used as a synaptic connection in a neuronal cell circuit its resistance is the weight of the synapse.
  • the memristor has the characteristics of simple structure, small area, large-scale integration, continuous change of resistance, etc., so its appearance provides a good device basis for the hardware implementation of the neural computer. Therefore, the neuron unit based on the resistive memristor
  • the circuit has been extensively studied. In the prior art, the computing function of the digital computer has been completed in the design stage, and after the computer design is completed, the computer merely reproduces the logic that has been set, does not have an independent learning ability, and does not have a meaningful learning function. Summary of the invention
  • the present invention proposes a novel neuron circuit capable of realizing the basic learning and memory function of biological neurons. It is an object of the present invention to provide a time-dependent learning neuron circuit based on a resistive memristor.
  • the time-dependent learning neuron circuit of the resistive memristor comprises: two neuron cell circuits and a resistive memristor as a synaptic connection therebetween; further, the neuron cell circuit comprises: an excitation a signal terminal, a synaptic connection terminal, a buffer, a control signal inverter, a first transmission gate, and a second transmission gate; wherein The output end of the buffer is connected to the excitation signal end, and the input end is connected to one signal end of the second transmission gate; the input end of the control signal inverter is connected to the excitation signal end, the positive control end of the first transmission gate, and the second transmission a negative control end of the door, the output end being connected to the negative control end of the first transmission gate and the positive control end of the second transmission gate; one signal end of the first transmission gate is connected to the voltage source, and the other signal end is connected to the synaptic connection;
  • the positive control terminal is connected to the excitation signal terminal, and the negative control terminal is connected to the output terminal of the control signal
  • the resistive memristor is a sandwich structure including a top electrode, a bottom electrode, and a resistive material filled therebetween.
  • the material of the top electrode and the bottom electrode is made of metal.
  • the resistive memristor is a resistor programmed by voltage, that is, the resistance of the device can be changed by applying a certain voltage. Such devices have been extensively studied in the current academic field. According to the polarity of the programming voltage, the device can be classified into a unipolar resistive memristor and a bipolar resistive memristor. The correlation between the two neuronal cell circuits will be determined by the resistive memristor that is the synaptic connection between the two.
  • the two neuron cell circuits are a pre-neuron cell circuit and a post-neuron cell circuit, and the control terminal of the first transmission gate of the pre-neuron cell circuit is connected to the positive voltage source, and the first transmission gate of the post-neuron cell circuit is controlled. The terminal is connected to a negative voltage source.
  • the synaptic junction of the anterior neuronal cell circuit is connected to the top electrode of the resistive memristor by a metal connection; the synaptic terminal of the posterior neuron cell circuit is connected to the bottom electrode of the resistive memristor by a metal connection.
  • the pre-neuron cell circuit receives the excitation signal, a positive voltage is applied to the resistive memristor through the synaptic connection end; when the post-neuron cell circuit receives the excitation signal, a negative voltage is applied to the resistive memristor through the synaptic connection end,
  • a large voltage difference is generated at both ends of the resistive memristor, so that the resistance of the resistive memristor becomes small.
  • the excitation signal end of the neuron cell circuit can be used as the input end of the excitation signal or as the output end of the excitation signal.
  • the excitation signal When used as the input end of the excitation signal, the excitation signal is input from the excitation signal end and connected to the positive of the first transmission gate.
  • the control terminal can be turned on by opening the first transmission gate, and the second transmission gate is turned off, thereby applying a voltage source signal to the synaptic connection terminal, and the voltage source signal is given by the voltage source; when used as the output end of the excitation signal, as a buffer
  • the input of the buffer is connected to the synaptic connection.
  • the buffer is an even number of inverters connected in series to improve the driving capability of the next stage circuit and make the voltage more stable.
  • the strength of this learning memory is determined by the strength of the synaptic association, and the strength of the association is determined by the length of the learning time.
  • This time-related learning and memory model is related to the resistance of the resistance-replacement memristor The characteristics are very similar and this is the theoretical basis on which the invention is based. Then, the implementation principle of the present invention will be briefly explained.
  • the two neuron cell circuits are simultaneously connected to their respective excitation signals, they respectively generate a stress signal to the respective excitation signals, and the stress signals are applied to the resistive memristor connected thereto through the metal wires.
  • Another object of the present invention is to provide a method for implementing time-dependent learning based on a time-dependent learning neuron circuit of a resistive memristor.
  • the method for implementing time-dependent learning by using a time-dependent learning neuron circuit based on a resistive memristor comprises the following steps:
  • the first step establish an association
  • any of the two neuron cell circuits When any of the two neuron cell circuits receives the previously learned excitation signal again, it will influence the other neuron cell circuit through its own stress signal, through the resistive memristor, and It generates its corresponding excitation signal.
  • the implementation of time-dependent learning is extremely accurate in mimicking the human learning process. Advantages of the invention:
  • the invention utilizes the switching characteristic of the resistive memristor, and when both ends thereof are synchronously selected by the two excitation signals, a voltage drop can be formed at both ends of the device to cause resistance change, thereby realizing the synapse
  • the breaking of the connection realizes the correlation of the two excitation signals, has memory characteristics, and can reproduce the previous excitation signal, that is, achieve the learning purpose. Due to the simple structure and high integration of the resistive memristor, large-scale physical neuron synaptic connections can be realized to achieve more complex learning and even logic functions, which have a good application prospect in neuron computing. . DRAWINGS
  • FIG. 1 is a schematic structural view of a time-dependent learning neuron circuit based on a resistive memristor of the present invention
  • FIG. 2 is an internal circuit diagram of an embodiment of a neuron cell circuit of the present invention
  • Fig. 4 is a graph showing the operation timing of an embodiment of the resistive memristor of the present invention.
  • the time-dependent learning neuron circuit based on the resistive memristor comprises: two neuron cell circuits 1 and 2 and a resistive memristor 3 as a synaptic connection therebetween;
  • the neuron cell circuit includes: an excitation signal terminal P, a synaptic connection terminal M, a buffer, and a control signal inverter. N1, the first transmission gate T1 and the second transmission gate T2; wherein the output end of the buffer is connected to the excitation signal terminal P, and the input terminal in is connected to a signal terminal of the second transmission gate T2;
  • the input terminal in of the control signal inverter N1 is connected to the excitation signal terminal P, the positive control terminal S of the first transmission gate T1 and the negative control terminal ⁇ of the second transmission gate T2, and the output terminal out is connected to the first transmission gate T1.
  • the terminal P, the negative control terminal ⁇ is connected to the output terminal out of the control signal inverter N1; one signal terminal of the second transmission gate T2 is connected to the input terminal in of the buffer, and the other signal terminal is connected to the synaptic connection terminal M,
  • the positive control port S is connected to the excitation signal terminal P, and the negative control terminal ⁇ is connected to the output terminal out of the control signal inverter N1.
  • the buffer is composed of two inverters N1 and N2 connected in series
  • the resistive memristor is a sandwich structure including a top electrode 31, a bottom electrode 32, and a resistive material 33 filled therebetween.
  • a bipolar resistive memristor is used.
  • the resistance value R of the resistive memristor will change, and the larger or smaller the voltage polarity is determined, when the voltage is positive In the case of time, the resistance becomes smaller.
  • the resistance becomes larger, and the change in resistance changes in a nonlinear manner, and the amount of change is positively correlated with time t and voltage V.
  • the resistance value of the resistive memristor When the voltage difference applied across the resistive memristor is lower than the threshold, the resistance value of the resistive memristor will not change, exhibiting a memory characteristic.
  • the working principle of the resistive memristor is shown in Figure 4.
  • the voltage value in the first period is less than the programmed threshold, and the resistance remains unchanged.
  • the second period t 2 the forward voltage is higher than the programmed threshold, and the resistance value changes. From large to small, and a non-linear gradual change, the change is getting faster and faster with the increase of time (this mechanism has been confirmed by experiments and theory, the specific principle is not detailed here), its characteristics and The patterns of human cognitive learning have very similarities and similarities.
  • the third time period t 3 which is a voltage lower than the threshold, will remain unchanged, which is equivalent to the memory in the learning process.
  • the fourth time period t 4 a reverse voltage is applied, and the voltage value is higher than the programming threshold Vreset, and the resistance value changes from small to large, and also exhibits a nonlinearity. This time, as time goes on, the change is getting slower and slower. Similarly, this is consistent with the law of forgetting in human cognitive learning.
  • the two neuronal cell circuits are the anterior neuron cell circuit 1 and the posterior neuron cell circuit 2, the anterior nerve
  • the control terminal of the first transmission gate of the cell circuit is connected to the positive voltage source Vp
  • the control terminal of the first transmission gate of the post-neuron cell circuit 2 is connected to the negative voltage source Vn.
  • the excitation signal terminal P of the neuron cell circuit can be used as the input end of the excitation signal or as the output end of the excitation signal. When used as the input end of the excitation signal, the excitation signal is input from the excitation signal terminal and connected to the first transmission gate T1.
  • the positive control terminal can be opened by opening the first transmission gate T1, and the second transmission threshold 2 is turned off, thereby applying a voltage source signal to the synaptic connection terminal ⁇ , and the voltage source signal is given by an independent voltage source; as an excitation signal
  • the output as the output of the buffer, the input of the buffer is connected to the synaptic terminal.
  • the specific working process is as follows: When establishing the association, two excitation signals that need to be associated are respectively input through the excitation signal end of the two neuron cell circuits, and then the transmission gate in the circuit is opened to generate a stress signal. The transmission gates are given to their respective ports. Two oppositely polarized stress voltage signals are again transmitted outward through the port to the top and bottom electrodes of the resistive memristor.
  • the resistive memristor Since the voltage difference at this time exceeds the threshold voltage at which the resistive memristor is resistively changed, the resistive memristor is resistively changed, and the resistance value is greatly reduced due to the forward voltage polarity.
  • the resistance value becomes smaller, and the corresponding synaptic connection weight is increased, that is, when the neuron cell circuit receives the excitation signal, the probability that the excitation signal is also affected by the other neuron cell circuit, that is, the two excitation signals Or there is a correlation between the two cellular circuits.
  • the increase in weight is determined by the length of time that two excitation signals are simultaneously applied. The longer the time, the greater the weight, the greater the correlation between the two, and the higher the success rate of the reverse retelling; the smaller the opposite.
  • the excitation signal which in turn causes a stress signal to the neuronal cell circuit, again causes the synaptic connection to increase in weight. That is to say, each reverse retelling is a deepening of the learning process.
  • This is similar to the principle in human cognitive learning.
  • the following describes an implementation method for performing time-dependent learning of a time-dependent learning neuron circuit based on a resistive memristor, which includes two parts:
  • the first step is to establish a correlation.
  • the two stress voltage signals alone cannot cause the resistive memristor to resist, that is, a single voltage.
  • the amplitude of the signal does not reach the threshold for blocking the resistive memristor, and when one positive, one negative and two voltage signals are superimposed, the voltage difference on the resistive memristor will be the absolute value of the two voltage signals.
  • the sum of the values exceeds the threshold, and the resistive memristor will resist, that is, start to change the connection weight between the two neuron cell circuits, which establishes the correlation between the two excitation signals, and this correlation
  • the establishment which requires two excitation signals to be applied at the same time, will occur, and it will satisfy the human learning cognition.
  • the second step is to repeat the above functions as the first step in learning cognition. To complete the learning function, it is necessary to successfully reverse the previous learning content.
  • the retelling process in the neuron cell circuit is when the two excitation signals are only When one occurs, another stimulus is automatically generated. Whether it is the anterior neuronal cell circuit or the post-neuron cell circuit, when one of the re-examination signals is received again, it will be transmitted to the synaptic connection with another neuron cell circuit through the stress signal.
  • the design circuit has completed the original process of simulating human cognition, and can realize the recognition and learning of the two signals, and can successfully reverse the reversal. It is to be understood that the present invention is intended to be a further understanding of the present invention, and it is understood by those skilled in the art that various modifications and changes can be made without departing from the spirit and scope of the invention and the appended claims It is possible. Therefore, the invention should not be limited by the scope of the invention, and the scope of the invention is defined by the scope of the claims.

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)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Neurology (AREA)
  • Chemical & Material Sciences (AREA)
  • Materials Engineering (AREA)
  • Computer Hardware Design (AREA)
  • Semiconductor Memories (AREA)

Abstract

本发明公开了一种基于阻变忆阻器的时间关联学习神经元电路及其实现方法。本发明利用阻变忆阻器的开关特性,当其两端被两个激励信号同步选定时,将会在器件的两端形成可以使其发生阻变的电压压降,从而实现这个突触连接的开断,实现两个激励信号的关联与否,并具有记忆特性,而且能够复述出之前的激励信号,即达到学习目的。由于阻变忆阻器的结构简单且可集成度高,能够实现大规模的物理神经元突触连接,以达到更为复杂的学习甚至逻辑功能,本发明在神经元计算中有着很好的应用前景。

Description

基于阻变忆阻器的时间关联学习神经元电路及其实现方法 技术领域 本发明涉及一种神经元细胞电路, 尤其涉及一种基于阻变忆阻器的时间关联学 习神经元电路及其实现方法。 背景技术
数字计算机是二十世纪人类科技文明进步的重要产物, 它的影响渗透到人们生 活的方方面面。然而, 随着计算机工业的发展和微电子产业的进步, 人们已经不能满 足于现有计算机的功能, 高的计算速度、 大的存储容量、 智能化已成为计算机进一步 发展的必然趋势。神经计算机因其具有大规模并行处理、 识别能力强、可处理模拟信 息、机器自学习等特点, 成为未来数字计算机强有力的替代者, 其硬件制造的关键在 于良好的、能够大规模集成的权重互联。在神经元细胞电路中需要用到大量的突触连 接, 而且这些突触连接必须具有可变的权重, 以及较小的面积以便于大规模集成。对 于在神经元细胞电路中用作突触连接的阻变忆阻器件而言, 其阻值即为突触的权重 值。 忆阻器具有结构简单、 面积小、 便于大规模集成、 阻值连续变化等特点, 因而它 的出现为神经计算机的硬件实现提供了良好的器件基础,因此基于阻变忆阻器的神经 元单元电路得到了广泛的研究。 现有技术中数字计算机的计算功能在设计阶段已经完成,而计算机设计完成后, 计算机只是重现已经设定的逻辑, 不具备自主的学习能力, 不具有真正意义的学习功 能。 发明内容
针对现有技术中存在的问题, 本发明提出一种新型的神经元电路, 能够实现生 物神经元的基本学习记忆功能。 本发明的一个目的在于提供一种基于阻变忆阻器的时间关联学习神经元电路。 本发明的基于阻变忆阻器的时间关联学习神经元电路包括: 两个神经元细胞电 路及作为二者之间的突触连接的阻变忆阻器; 进一步, 神经元细胞电路包括: 激励信 号端、 突触连接端、 缓冲器、 控制信号反相器、 第一传输门及第二传输门; 其中, 缓冲器的输出端连接至激励信号端, 输入端连接至第二传输门的一个信号端; 控制信号反相器的输入端连接至激励信号端、 第一传输门的正控制端及第二传 输门的负控制端, 输出端连接至第一传输门的负控制端及第二传输门的正控制端; 第一传输门的一个信号端连接至电压源, 另一个信号端连接至突触连接端, 正 控制端连接至激励信号端, 负控制端连接至控制信号反相器的输出端; 第二传输门的一个信号端连接至缓冲器的输入端, 另一个信号端连接至突触连 接端, 正控制端口连接至激励信号端, 负控制端连接至控制信号反相器的输出端。
阻变忆阻器为三明治结构, 包括顶电极、 底电极以及填充在两者之间的阻变材 料。顶电极和底电极的材料采用金属。 阻变忆阻器为通过电压编程的电阻器, 即可以 通过施加一定的电压而改变器件的阻值, 该类器件在目前的学术领域得到广泛的研 究。 根据编程电压的极性可以将该类器件分为单极型阻变忆阻器和双极性阻变忆阻 器。两个神经元细胞电路之间的关联性,将由作为两者的突触连接的阻变忆阻器所决 定,当阻变忆阻器阻值处在最大值时,突触连接权重最小,两者间的关联性几乎为零, 即互相不会影响; 而当阻变忆阻器的阻值变小时, 突触连接权重增大, 两者之间产生 较大的关联性, 使得对两个细胞电路产生激励的两个激励信号间, 建立了关联性。 两个神经元细胞电路分别为前神经元细胞电路和后神经元细胞电路, 前神经元 细胞电路的第一传输门的控制端连接正电压源,后神经元细胞电路的第一传输门的控 制端连接负电压源。前神经元细胞电路的突触连接端通过金属连接线连接至阻变忆阻 器的顶电极;后神经元细胞电路的突触连接端通过金属连接线连接至阻变忆阻器的底 电极。前神经元细胞电路接收激励信号时,通过突触连接端对阻变忆阻器施加正电压; 后神经元细胞电路接收激励信号时,通过突触连接端对阻变忆阻器施加负电压,这样 当两个激励信号同时接收时,在阻变忆阻器的两端产生较大的电压差, 使阻变忆阻器 的阻值变小。神经元细胞电路的激励信号端即可作为激励信号的输入端, 也可以作为 激励信号的输出端, 作为激励信号的输入端时, 激励信号从激励信号端输入, 连接至 第一传输门的正控制端, 能够通过打开第一传输门, 而第二传输门关断, 从而将电压 源信号施加到突触连接端, 电压源信号由电压源给出; 作为激励信号的输出端时, 作 为缓冲器的输出端,缓冲器的输入端连接突触连接端。缓冲器为偶数个串联的反相器, 用来提高对下一级电路的驱动能力, 使电压更加稳定。 下面简单说明本发明的原理。 首先, 阐述一种人类学习的基本模式, 即认识一种事物, 需要事物本身的影像 信号, 通过眼睛输入给大脑, 同时需要解释这种事物的声音信号, 通过耳朵输入给大 脑。两者作为这种事物学习的基本元素, 并且只有当两个信号同时输入时, 才会在大 脑中建立事物图像与事物意义的关联, 并在下一次两者中的一个信号输入时,可以通 过 "思考", 唤醒另一个信号的记忆, 即实现了对一种事物的学习和记忆功能。 这种 学习记忆的强弱由突触关联的强弱决定, 而关联的强弱由学习时间的长短决定,这种 时间关联的学习记忆模式,与现在广泛研究的阻变忆阻器的阻变特性有着非常相似之 处, 这也是本发明所基于的理论基础。 然后, 简要说明一下本发明的实现原理。 当两个神经元细胞电路同时接到各自 的激励信号时, 分别对各自的激励信号产生应激信号, 该应激信号会通过金属连线将 其施加到与之相连的阻变忆阻器上, 当前后两个激励信号同时激励时,将在阻变忆阻 器上形成使之发生阻变的电压差, 开始时阻值较大, 随着激励持续时间而阻值逐渐变 小, 即从一开始的突触连接的权重较小变为权重增大, 直到激励信号中的一个结束, 因为认知的过程需要两个激励信号同时进行。激励信号结束后, 阻变忆阻器的阻值保 持不变, 相当于学习过程中的记忆。被改变后的阻变忆阻器, 准确来说是阻值变小后 的阻变忆阻器, 使得两个神经元细胞电路间产生较强的关联性, 即突触连接的权重增 大,其中一个细胞电路的激励信号被另一个细胞感知到的概率增大。学习认知的目的 是记忆, 并且能够准确的逆向复述出来之前认识到的事物的信号。当两个神经元细胞 电路中的任何一个, 再次接收到之前学习的激励信号时,将会通过自身的应激信号与 突触连接, 对另一个神经元细胞电路产生影响, 并使之产生其对应的激励信号, 即当 时两个产生关联的两个激励信号中的另外一个。至此, 完成了利用基于阻变忆阻器的 阻变特性, 模仿人类学习认知的神经元网络电路的学习过程。 本发明的另一个目的在于提供一种基于阻变忆阻器的时间关联学习神经元电路 进行时间关联学习的实现方法。
本发明的采用基于阻变忆阻器的时间关联学习神经元电路进行时间关联学习的 实现方法, 包括以下步骤:
第一步) 建立关联
1 ) 两个神经元细胞电路分别从激励信号端接收两个不同的激励信号;
2) 两个激励信号在时间上产生重叠, 在重叠的这段时间里, 阻变忆阻器的 阻值逐渐变小; 3 ) 两个激励信号中的一个结束时, 阻变忆阻器阻值将保持不变; 第二步) 复述
当两个神经元细胞电路中的任何一个, 再次接收到之前学习的激励信号时, 将 会通过自身的应激信号, 经过阻变忆阻器, 对另一个神经元细胞电路产生影响, 并使 之产生其对应的激励信号。 本发明中, 两个激励信号产生重叠的时间越长, 相当于同一个事物的两个信号 同时建立关联的时间越长, 阻变忆阻器的值越小, 相当于建立起关联的概率越大, 激 励信号结束, 阻值保持不变, 相当于对事物信号的记忆, 当其中一个激励信号再次出 现时, 另一个相关联的激励信号被重复出现的概率越高, 由此可见, 本发明的时间关 联学习的实现方法, 极其准确的模仿了人类的学习过程。 本发明的优势:
本发明利用阻变忆阻器的开关特性, 当其两端被两个激励信号同步选定时, 将 会在器件的两端形成可以使其发生阻变的电压压降, 从而实现这个突触连接的开断, 实现两个激励信号的关联与否, 并具有记忆特性, 而且能够复述出之前的激励信号, 即达到学习目的。 由于阻变忆阻器的结构简单且可集成度高, 能够实现大规模的物理 神经元突触连接, 以达到更为复杂的学习甚至逻辑功能,其在神经元计算中有着很好 的应用前景。 附图说明
图 1是本发明的基于阻变忆阻器的时间关联学习神经元电路的结构示意图; 图 2是本发明的神经元细胞电路的一个实施例的内部的电路图; 图 3是本发明的作为神经元细胞电路的突触连接的阻变忆阻器的结构示意图; 图 4是本发明的阻变忆阻器的一个实施例的工作时序的曲线图。 具体实施方式
下面结合附图, 通过实例对本发明做进一步说明。
如图 1所示, 本发明基于阻变忆阻器的时间关联学习神经元电路包括: 两个神 经元细胞电路 1和 2及作为二者之间的突触连接的阻变忆阻器 3 ; 进一步, 如图 2所 示, 神经元细胞电路包括: 激励信号端 P、 突触连接端 M、 缓冲器、 控制信号反相器 Nl、 第一传输门 Tl及第二传输门 T2; 其中, 缓冲器的输出端 out连接至激励信号端 P,输入端 in连接至第二传输门 T2的一 个信号端;
控制信号反相器 N1的输入端 in连接至激励信号端 P、第一传输门 T1的正控制 端 S及第二传输门 T2的负控制端 §, 输出端 out连接至第一传输门 T1的负控制端 § 及第二传输门 T2的正控制端 S; 第一传输门 T1 的一个信号端连接至电压源, 另一个信号端连接至突触连接端 M, 正控制端 S连接至激励信号端 P, 负控制端§连接至控制信号反相器 N1的输出 端 out; 第二传输门 T2的一个信号端连接至缓冲器的输入端 in,另一个信号端连接至突 触连接端 M, 正控制端口 S连接至激励信号端 P, 负控制端§连接至控制信号反相器 N1的输出端 out。 在本实施例中, 缓冲器由两个串联的反相器 N1和 N2构成。
如图 3所示, 阻变忆阻器为三明治结构, 包括顶电极 31、底电极 32以及填充在 两者之间的阻变材料 33。 本实施例中采用双极性阻变忆阻器。 当阻变忆阻器的两端施加的电压差超过阈 值 Vset时, 阻变忆阻器的电阻值 R将会发生变化, 变大还是变小由此时的电压极性 决定, 当电压为正向时, 阻值变小, 当电压为负向时, 阻值变大, 且阻值的变化呈一 种非线性的缓变,变化量与时间 t和电压 V呈正相关。而当阻变忆阻器的两端施加的 电压差低于该阈值时, 阻变忆阻器的电阻值将不发生变化, 表现出一种记忆特性。 阻 变忆阻器的工作原理如图 4所示, 第一时间段 电压值小于编程阈值, 阻值保持不 变, 第二时间段 t2, 正向电压高于编程阈值, 电阻值发生变化, 由大变小, 并且呈一 种非线性的缓变, 随着时间的增加, 变化越来越快(这种机理已经得到实验和理论的 证实, 具体原理这里不展开详述), 其特点与人类认知学习的模式有着非常巧合的相 似之处。 第三时间段 t3, 为一个低于阈值的电压, 其阻值将保持不变, 相当于学习过 程中的记忆。 与第二时间段 t2图像想反, 第四时间段 t4, 施加一个反向电压, 且电压 值高于编程阈值 Vreset, 电阻值发生变化, 由小变大, 同样呈现一种非线性, 这次是 随着时间的增加, 变化越来越慢, 同样, 这与人类认知学习中的遗忘规律相吻合。
两个神经元细胞电路分别为前神经元细胞电路 1和后神经元细胞电路 2,前神经 元细胞电路的第一传输门的控制端连接正电压源 Vp, 后神经元细胞电路 2的第一传 输门的控制端连接负电压源 Vn。 前神经元细胞电路 1接收激励信号时, 通过突触连 接端 M对阻变忆阻器施加正电压; 后神经元细胞电路 2接收激励信号时, 通过突触 连接端 M对阻变忆阻器施加负电压, 这样当两个激励信号同时接收时, 在阻变忆阻 器的两端产生大于编程阈值的电压差, 阻值变小。 神经元细胞电路的激励信号端 P 即可作为激励信号的输入端, 也可以作为激励信号的输出端, 作为激励信号的输入端 时, 激励信号从激励信号端输入, 连接至第一传输门 T1的正控制端, 能够通过打开 第一传输门 Tl, 而第二传输门 Τ2关断, 从而将电压源信号施加到突触连接端 Μ, 电 压源信号由独立的电压源给出; 作为激励信号的输出端时, 作为缓冲器的输出端, 缓 冲器的输入端连接突触连接端 Μ。 具体的工作过程如下: 建立关联时, 两个需要建 立关联的激励信号,分别通过两个神经元细胞电路的激励信号端 Ρ输入,进入之后会 使得电路中的传输门打开, 产生应激信号, 通过传输门给到各自的 Μ端口。 两个极 性相反的应激电压信号再通过 Μ端口向外各自传输到阻变忆阻器的顶电极和底电 极。 由于此时的压差超过了阻变忆阻器发生阻变的阈值电压, 因此, 阻变忆阻器发生 阻变, 由于是正向电压极性, 阻值由大变小。阻值变小,相应的该突触连接权重增大, 即一个神经元细胞电路接收到激励信号时,影响到另一个神经元细胞电路也发生激励 信号的概率增大, 也即两个激励信号或者是两个细胞电路之间发生了关联性。权重增 大的多少, 由两个激励信号同时施加的时间长短决定, 时间越长, 权重越大, 两者间 的关联性越大, 继而逆向复述的成功率越大; 反之则越小。这种规律与人类学习认知 的过程相类似。激励信号结束后, 阻变忆阻器的阻值将保持不变, 表现为学习中的记 忆。逆向复述的过程, 即当任意一个神经元细胞电路单独接收到激励信号时, 其产生 的应激信号, 将会乘上突触权重后, 发射到另一个神经元细胞电路的突触连接端 Μ, 突触连接端 Μ作为输入时, 为缓冲器的输入端, 进来的前一级应激信号将通过缓冲 器在输出端形成激励信号,也就是在该神经元细胞电路的激励信号端 Ρ形成了激励信 号, 该激励信号反过来又会使本神经元细胞电路发生应激信号, 再次使得突触连接发 生权重增大。也就是说每一次的逆向复述都是一次再加深学习过程。这与人类认知学 习中的原理相类似。 下面描述基于阻变忆阻器的时间关联学习神经元电路的进行时间关联学习的实 现方法, 包括两部分:
第一步) 建立关联 就两个应激电压信号单独而言, 都无法使得阻变忆阻器发生阻变, 即单个电压 信号的幅值没有达到使阻变忆阻器发生阻变的阈值,而当一正一负两个电压信号叠加 时, 阻变忆阻器上的电压差将是两个电压信号幅值的绝对值之和, 超过了阈值, 阻变 忆阻器将发生阻变, 即开始改变两个神经元细胞电路之间的连接权重, 是两个激励信 号之间建立了关联性,而这个关联性的建立,需要两个激励信号同时施加,才会发生, 也就满足了人类学习认知时, 需要两个学习元素同时出现才能建立认知。 第二步) 复述 以上功能是学习认知的第一步, 完成学习的功能, 还需要成功的逆向复述出之 前学习内容, 本神经元细胞电路中的复述过程, 即是当两个激励信号只出现一个时, 会自动产出另一个激励信号。无论是前神经元细胞电路, 还是后神经元细胞电路, 当 其中一个再次接收到之前的激励信号后,将会通过应激信号, 发射到与另一个神经元 细胞电路的作为突触连接的阻变忆阻器上, 此时因为阻值小, 突触连接权重很大, 即 两者间有较强的关联性,该应激信号会乘上该权重,输入到另一个神经元细胞电路中, 并通过缓冲器,产生了之前建立关联的另一个激励信号, 至此成功完成了逆向复述的 过程。值得一提的是, 复述产生的激励信号又会在神经元细胞电路中产生再一次的应 激信号, 使得两个细胞电路之间的突触连接再次得到学习, 也就是阻变忆阻器的阻值 进一步减小, 突触连接权重进一步增大。 由此可见, 每一次的逆向复述都是对之前学 习认识的一次学习加固, 即更加增大了两个信号之间的关联性,这与人类学习认知中 的, 勤加练习、 孰能生巧的规律相一致。
至此该设计电路完成了原来预想的模拟人类学习认知的过程, 能够实现对两个 信号的认知和学习, 并能够成功的逆向复述出来。 最后需要注意的是, 公布实施方式的目的在于帮助进一步理解本发明, 但是本 领域的技术人员可以理解: 在不脱离本发明及所附的权利要求的精神和范围内,各种 替换和修改都是可能的。 因此, 本发明不应局限于实施例所公开的内容, 本发明要求 保护的范围以权利要求书界定的范围为准。

Claims

权 利 要 求
1. 一种时间关联学习神经元电路, 其特征在于, 所述电路包括: 两个神经 元细胞电路 (1和 2) 及作为二者之间的突触连接的阻变忆阻器 (3 ); 进一步, 神经元细胞电路包括: 激励信号端 P、 突触连接端 M、 缓冲器、 控制信号反相器 Nl、 第一传输门 T1及第二传输门 T2; 其中,
所述缓冲器的输出端 out连接至激励信号端 P,输入端 in连接至第二传输门 T2的一个信号端;
所述控制信号反相器 N1 的输入端 in连接至激励信号端 P、 第一传输门 T1 的正控制端 S及第二传输门 T2的负控制端 §, 输出端 out连接至第一传输门 T1 的负控制端 §及第二传输门 T2的正控制端 S;
所述第一传输门 T1的一个信号端连接至电压源, 另一个信号端连接至突触 连接端 M, 正控制端 S连接至激励信号端 P, 负控制端§连接至控制信号反相器 N1的输出端 out;
所述第二传输门 T2的一个信号端连接至缓冲器的输入端 in, 另一个信号端 连接至突触连接端 M, 正控制端口 S连接至激励信号端 P, 负控制端§连接至控 制信号反相器 N1的输出端 out。
2、 如权利要求 1所述的时间关联学习神经元电路, 其特征在于, 所述阻变 忆阻器(3 ) 为三明治结构, 包括顶电极(31 )、 底电极(32) 以及填充在两者之 间的阻变材料 (33 )。
3、 如权利要求 2所述的时间关联学习神经元电路, 其特征在于, 所述阻变 忆阻器为通过电压编程的电阻器, 根据编程电压的极性, 分为单极型阻变忆阻器 和双极性阻变忆阻器。
4、 如权利要求 2所述的时间关联学习神经元电路, 其特征在于, 两个神经 元细胞电路分别为前神经元细胞电路 (1 ) 和后神经元细胞电路 (2), 所述前神 经元细胞电路 (1 ) 的第一传输门 T1的控制端连接正电压源 Vp, 后神经元细胞 电路 (2) 的第一传输门 T1的控制端连接负电压源 Vn。
5、 如权利要求 4所述的时间关联学习神经元电路, 其特征在于, 所述前神 经元细胞电路(1 ) 的突触连接端 P通过金属连接线连接至阻变忆阻器(3 ) 的顶 电极(31 ); 后神经元细胞电路(2) 的突触连接端 P通过金属连接线连接至阻变 忆阻器 (3 ) 的底电极 (32)。
6、 如权利要求 1所述的时间关联学习神经元电路, 其特征在于, 所述缓冲 器为偶数个串联的反相器。
7、 如权利要求 1所述的时间关联学习神经元电路, 其特征在于, 所述神经 元细胞电路的激励信号端 P 即作为激励信号的输入端, 也作为激励信号的输出 端。
8、 一种权利要求 1所述的时间关联学习神经元电路的进行时间关联学习的 实现方法, 其特征在于, 包括以下步骤:
第一步) 建立关联
1 ) 两个神经元细胞电路分别从激励信号端接收两个不同的激励信号;
2) 两个激励信号在时间上产生重叠, 在重叠的这段时间里, 阻变忆阻器的 阻值逐渐变小;
3 ) 两个激励信号中的一个结束时, 阻变忆阻器阻值将保持不变; 第二步) 复述
当两个神经元细胞电路中的任何一个, 再次接收到之前学习的激励信号时, 将会通过自身的应激信号,经过阻变忆阻器,对另一个神经元细胞电路产生影响, 并使之产生其对应的激励信号。
9、 如权利要求 8所述的实现方法, 其特征在于, 在第一步中, 神经元细胞 电路的激励信号端 P作为激励信号的输入端,激励信号从激励信号端输入,连接 至第一传输门 T1的正控制端 S, 通过打开第一传输门 Tl, 而第二传输门 Τ2关 断, 从而将电压源信号施加到突触连接端 Μ, 电压源信号由电压源给出。
10、 如权利要求 8所述的实现方法, 其特征在于, 在第二步中, 神经元细胞 电路的激励信号端 Ρ 作为激励信号的输出端, 缓冲器的输入端连接突触连接端
Μ。
PCT/CN2013/084752 2013-05-24 2013-09-30 基于阻变忆阻器的时间关联学习神经元电路及其实现方法 WO2014187059A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US14/892,130 US20160110644A1 (en) 2013-05-24 2013-09-30 Time Correlation Learning Neuron Circuit Based on a Resistive Memristor and an Implementation Method Thereof

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201310197061.7 2013-05-24
CN201310197061.7A CN103246904B (zh) 2013-05-24 2013-05-24 基于阻变忆阻器的时间关联学习神经元电路及其实现方法

Publications (1)

Publication Number Publication Date
WO2014187059A1 true WO2014187059A1 (zh) 2014-11-27

Family

ID=48926417

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2013/084752 WO2014187059A1 (zh) 2013-05-24 2013-09-30 基于阻变忆阻器的时间关联学习神经元电路及其实现方法

Country Status (3)

Country Link
US (1) US20160110644A1 (zh)
CN (1) CN103246904B (zh)
WO (1) WO2014187059A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110428050A (zh) * 2019-08-25 2019-11-08 湖北大学 一种基于忆阻器实现多样化stdp学习规则的突触仿生电路
CN113469348A (zh) * 2021-06-21 2021-10-01 安徽大学 一种联想记忆中多次泛化和分化的神经形态电路
CN114169511A (zh) * 2021-11-11 2022-03-11 山东科技大学 一种基于实物忆阻器的联想记忆电路及方法
CN115456157A (zh) * 2022-11-11 2022-12-09 华中科技大学 一种基于忆阻器的多感觉互联想记忆网络电路
CN116894470A (zh) * 2023-07-28 2023-10-17 常州大学 一种模拟动物操作性条件反射的神经功能性电路
CN117692003A (zh) * 2023-12-11 2024-03-12 广东工业大学 一种基于神经元的模数转换器

Families Citing this family (32)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103246904B (zh) * 2013-05-24 2016-04-06 北京大学 基于阻变忆阻器的时间关联学习神经元电路及其实现方法
CN103580668B (zh) * 2013-10-28 2016-04-20 华中科技大学 一种基于忆阻器的联想记忆电路
CN104979472B (zh) * 2014-04-11 2017-07-28 中国科学院宁波材料技术与工程研究所 一种有机高分子忆阻结构单元
CN104539280B (zh) * 2014-11-23 2017-10-27 华中科技大学 一种基于忆阻器实现多层识别的电路及其控制方法
WO2016085470A1 (en) * 2014-11-25 2016-06-02 Hewlett-Packard Development Company, L.P. Bi-polar memristor
CN106470023A (zh) * 2015-08-18 2017-03-01 华为技术有限公司 神经模拟电路
WO2017068490A1 (en) * 2015-10-23 2017-04-27 Semiconductor Energy Laboratory Co., Ltd. Semiconductor device and electronic device
US10452973B2 (en) 2016-06-22 2019-10-22 International Business Machines Corporation Smart logic device
WO2018100790A1 (ja) * 2016-11-30 2018-06-07 国立研究開発法人科学技術振興機構 ニューロン回路、システムおよびスイッチ回路
CN106779059B (zh) * 2016-12-30 2019-03-05 华中科技大学 一种基于忆阻的巴普洛夫联想记忆的人工神经网络电路
KR101918102B1 (ko) 2017-02-14 2019-02-08 한국과학기술연구원 플로팅 게이트 적분기를 이용한 고집적 저전력 소모 인공뉴런 회로 및 이를 포함하는 뉴로모르픽 시스템, 이의 제어 방법
CN107122828B (zh) 2017-05-09 2020-05-05 清华大学 电路结构及其驱动方法、神经网络
FR3074337B1 (fr) * 2017-11-30 2021-04-09 Thales Sa Reseau neuromimetique et procede de fabrication associe
CN109102071B (zh) * 2018-08-07 2020-12-11 中国科学院微电子研究所 一种神经元电路以及神经网络电路
DE102018213147A1 (de) * 2018-08-07 2020-02-13 Robert Bosch Gmbh Auffrischen von mittels Memristoren gespeicherten Daten
CN109346598A (zh) * 2018-08-31 2019-02-15 南京邮电大学 一种具有生物突触模拟功能的卟啉忆阻器及其制备方法和应用
KR102350877B1 (ko) * 2018-11-29 2022-01-13 한국전자통신연구원 뉴런 회로 및 그 동작 방법
US11977973B2 (en) 2018-11-29 2024-05-07 Electronics And Telecommunications Research Institute Neuron circuit and operating method thereof
CN109754070B (zh) * 2018-12-28 2022-10-21 东莞钜威软件科技有限公司 基于神经网络的绝缘阻值计算方法及电子设备
CN109829194B (zh) * 2018-12-28 2022-12-06 山东科技大学 一种绝对值磁控忆阻器等效模拟电路
CN110428049B (zh) * 2019-08-21 2021-10-26 南京邮电大学 一种基于多态忆阻器的电压型神经网络及其操作方法
CN110728102B (zh) * 2019-09-29 2023-09-22 山东科技大学 一种多稳态忆感器模拟电路
CN111061454B (zh) * 2019-12-18 2022-02-11 北京大学 一种基于双极性忆阻器的逻辑实现方法
CN111967589B (zh) * 2020-08-21 2023-12-26 清华大学 神经元模拟电路及其驱动方法、神经网络装置
CN112071346B (zh) * 2020-09-07 2022-06-21 湖南大学 基于忆阻器簇的3d交叉阵列结构
CN112906880B (zh) * 2021-04-08 2022-04-26 华中科技大学 一种基于忆阻器的自适应神经元电路
CN113311702B (zh) * 2021-05-06 2022-06-21 清华大学 一种基于Master-Slave神经元的人工神经网络控制器
CN113658493B (zh) * 2021-08-20 2023-05-02 安徽大学 一种用于模拟联想记忆的强化学习仿生电路架构
CN114648108B (zh) * 2022-04-20 2024-09-03 华中科技大学 自适应仿生神经元电路及仿生神经元自适应模拟方法
CN115062772B (zh) * 2022-06-10 2023-09-26 常州大学 一种简单的rc型神经元簇放电电路
CN115600665B (zh) * 2022-11-16 2024-04-09 湖南师范大学 一种基于vta-da神经元的忆阻自修复神经网络电路
CN116523012B (zh) * 2023-07-03 2023-09-08 湖南师范大学 一种基于生成对抗神经网络的忆阻器自学习电路

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020030205A1 (en) * 1998-08-07 2002-03-14 Victor I. Varshavsky Artificial neuron on the base of beta-driven threshold element
CN102543172A (zh) * 2012-02-27 2012-07-04 北京大学 一种适用于神经元电路的阻变忆阻器的控制方法
CN103246904A (zh) * 2013-05-24 2013-08-14 北京大学 基于阻变忆阻器的时间关联学习神经元电路及其实现方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US384010A (en) * 1888-06-05 Percy g
US93001A (en) * 1869-07-27 Improved pitcher for cooling liquids
JP2810202B2 (ja) * 1990-04-25 1998-10-15 株式会社日立製作所 ニューラルネットワークによる情報処理装置
US8605484B2 (en) * 2009-01-29 2013-12-10 Hewlett-Packard Development Company, L.P. Self-repairing memristor and method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20020030205A1 (en) * 1998-08-07 2002-03-14 Victor I. Varshavsky Artificial neuron on the base of beta-driven threshold element
CN102543172A (zh) * 2012-02-27 2012-07-04 北京大学 一种适用于神经元电路的阻变忆阻器的控制方法
CN103246904A (zh) * 2013-05-24 2013-08-14 北京大学 基于阻变忆阻器的时间关联学习神经元电路及其实现方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
CANTLEY, K.D. ET AL.: "Neural Learning Circuits Utilizing Nano-Crystalline Silicon Transistors and Memristors", IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, April 2012 (2012-04-01), pages 565 - 573 *

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110428050A (zh) * 2019-08-25 2019-11-08 湖北大学 一种基于忆阻器实现多样化stdp学习规则的突触仿生电路
CN110428050B (zh) * 2019-08-25 2024-04-05 湖北大学 一种基于忆阻器实现多样化stdp学习规则的突触仿生电路
CN113469348A (zh) * 2021-06-21 2021-10-01 安徽大学 一种联想记忆中多次泛化和分化的神经形态电路
CN113469348B (zh) * 2021-06-21 2024-02-20 安徽大学 一种联想记忆中多次泛化和分化的神经形态电路
CN114169511A (zh) * 2021-11-11 2022-03-11 山东科技大学 一种基于实物忆阻器的联想记忆电路及方法
CN114169511B (zh) * 2021-11-11 2024-03-19 山东科技大学 一种基于实物忆阻器的联想记忆电路及方法
CN115456157A (zh) * 2022-11-11 2022-12-09 华中科技大学 一种基于忆阻器的多感觉互联想记忆网络电路
CN115456157B (zh) * 2022-11-11 2023-02-07 华中科技大学 一种基于忆阻器的多感觉互联想记忆网络电路
CN116894470A (zh) * 2023-07-28 2023-10-17 常州大学 一种模拟动物操作性条件反射的神经功能性电路
CN116894470B (zh) * 2023-07-28 2024-01-23 常州大学 一种模拟动物操作性条件反射的神经功能性电路
CN117692003A (zh) * 2023-12-11 2024-03-12 广东工业大学 一种基于神经元的模数转换器
CN117692003B (zh) * 2023-12-11 2024-05-28 广东工业大学 一种基于神经元的模数转换器

Also Published As

Publication number Publication date
CN103246904A (zh) 2013-08-14
CN103246904B (zh) 2016-04-06
US20160110644A1 (en) 2016-04-21

Similar Documents

Publication Publication Date Title
WO2014187059A1 (zh) 基于阻变忆阻器的时间关联学习神经元电路及其实现方法
US9779355B1 (en) Back propagation gates and storage capacitor for neural networks
Azghadi et al. A hybrid CMOS-memristor neuromorphic synapse
US11222259B2 (en) Counter based resistive processing unit for programmable and reconfigurable artificial-neural-networks
US9396431B2 (en) Network of artificial neurons based on complementary memristive devices
CN102959565B (zh) 用于生物启发型网络的通信和突触训练方法及硬件
US8433665B2 (en) Methods and systems for three-memristor synapse with STDP and dopamine signaling
Xiao et al. GST-memristor-based online learning neural networks
US9330355B2 (en) Computed synapses for neuromorphic systems
Chen et al. Associate learning and correcting in a memristive neural network
Wang et al. Memristor-based synapse design and training scheme for neuromorphic computing architecture
US10319439B1 (en) Resistive processing unit weight reading via collection of differential current from first and second memory elements
Mahalanabis et al. Demonstration of spike timing dependent plasticity in CBRAM devices with silicon neurons
Kang et al. Emulation of spike-timing dependent plasticity in nano-scale phase change memory
Guo et al. A multi-layer memristive recurrent neural network for solving static and dynamic image associative memory
Dong et al. Flexible memristor‐based LUC and its network integration for Boolean logic implementation
Li et al. A reconfigurable bidirectional associative memory network with memristor bridge
CN114169511B (zh) 一种基于实物忆阻器的联想记忆电路及方法
Mostafa et al. Beyond spike-timing dependent plasticity in memristor crossbar arrays
Zhou et al. The characteristics of binary spike-time-dependent plasticity in HfO 2-based RRAM and applications for pattern recognition
Zhang et al. The framework and memristive circuit design for multisensory mutual associative memory networks
Dai et al. An improved memristor model based on the electrochemical metallization effect as a synapse for biomimetic applications
Chakraborty et al. Spike-timing-dependent plasticity for a hafnium-oxide memristive synapse
Hajiabadi et al. Memristor-based synaptic plasticity and unsupervised learning of spiking neural networks
Afifi et al. Efficient hybrid CMOS-Nano circuit design for spiking neurons and memristive synapses with STDP

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 13885435

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 14892130

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 24.03.2016)

122 Ep: pct application non-entry in european phase

Ref document number: 13885435

Country of ref document: EP

Kind code of ref document: A1