Method for reducing memristor noise based on random pulse sequence

Abstract

The invention relates to the technical field of nano electronic devices, in particular to a method for reducing memristor noise based on a random pulse sequence. The method comprises the following steps: creating positive and negative pulse sequences with a certain proportion and amplitude, randomly extracting pulse voltages from the pulse sequences, extracting random time intervals from Poisson distribution, and generating corresponding random pulse sequences according to the number of pulses. Compared with a single-pulse conductance setting mode, the probability of eliminating an incomplete channel can be increased due to simultaneous action of positive and negative pulses in a random pulse conductance setting process, so that the noise of a device is reduced in the conductance setting process. According to the method, the noise reduction process is simplified, the stability and the calculation precision of the device are improved, and the method is suitable for various application scenes including neuromorphic calculation.

Description

Technical Field

[0001] The present application relates to the technical field of nanoelectronic devices, and in particular to a method for reducing memristor noise based on a random pulse sequence. Background Art

[0002] As CMOS transistors face physical size limits, the separation of storage and computing in the traditional von Neumann architecture leads to the "storage wall" problem, which limits the improvement of computing energy efficiency. Compared with traditional memory, memristors, as emerging memory devices, have higher storage density, faster read and write speeds, and lower power consumption, making them one of the key technologies for realizing neuromorphic computing systems. In neural network applications, memristors require high-precision conductance control to achieve high stability of the memristor network, which requires each memristor device to have a stable and distinguishable conductance state.

[0003] Valence change memory is a two-terminal memristor composed of metal oxide sandwiched between metal electrodes with different work functions. When a forward voltage is applied to the device, oxygen vacancies accumulate and form nano-conductive filaments (CF). When a reverse voltage is applied, the CF breaks, and the high and low resistance state changes of the memristor are attributed to the formation and breaking of CF in the oxide layer. However, CF-based switching does not form a single CF unit in the oxide layer, but rather multiple local CFs appear during the CF formation process. Small morphological changes in CF may cause dramatic fluctuations in resistance, thereby causing noise in the device.

[0004] Current fluctuations caused by random telegraph noise (RTN) and other factors can be attributed to the process of charge capture and decapture by local defects, which will reduce the read margin and cause serious calculation errors, thereby reducing the reliability of devices and systems. In principle, an infinite number of conductance levels can be achieved within the dynamic range of the memristor. However, the current fluctuations at each conductance level limit the number of conductances that can be distinguished within a specific conductance range.

[0005] Normal switching operations always produce some incomplete conductive channels, which are distributed in the form of islands or fuzzy edges along the main channel and are not as stable as the main channel. These incomplete conductive channels will cause RTN noise. Existing theories and experiments show that applying positive and negative pulse voltages of a certain amplitude can eliminate these incomplete channels, thereby reducing the RTN noise of the device. However, existing noise reduction methods only target the RTN noise in the device, without involving other noise signals, and require a large number of iterations to determine whether the noise has been reduced, which increases the complexity and time cost of setting the conductance. Therefore, how to effectively and simply reduce the noise in the memristor and achieve stable conductance regulation is still a technical problem that needs to be solved urgently. Summary of the invention

[0006] The technical problem to be solved by the present invention is to invent a universal noise reduction method for the conductance fluctuation caused by noise in a memristor. The present invention provides a method for regulating nanofilaments based on a random pulse sequence, which can reduce the noise of the device during the conductance regulation process, and not only the RTN noise in the device. The present invention reduces the complexity of noise reduction and improves the practicality of noise reduction.

[0007] In order to solve the above technical problems, the present invention provides a method for regulating memristor nanofilaments and reducing noise through random pulse sequences, which includes the following steps: creating a voltage sequence with the same positive and negative pulse ratios and different positive and negative pulse amplitudes; using a random function to extract the pulse amplitude from the voltage sequence; generating a random time interval that is not 1 through a Poisson distribution function; repeating the operation according to a given number of pulses to generate a random pulse sequence; and applying the above-mentioned random pulse sequence signal to the memristor.

[0008] In order to further solve the technical problem to be solved by the present invention, the amplitudes of the positive and negative pulses are gradually reduced.

[0009] Preferably, the pulse width is 10 us and 1 us, and the number of pulses is set to 1000.

[0010] In one implementation, the memristor is a double-layer memristor with a cross structure.

[0011] like Figure 1 As shown, preferably, the memristor uses Ti as an adhesion layer, Pt as a bottom electrode 1, HfO2 and Al2O3 as a functional layer 2; Al2O3 as an insulating layer 3, and peeling to form a through-hole structure. Ti is used as a functional electrode and Pt is used as a protective layer and a top electrode 4 to fill the through-hole.

[0012] In one embodiment, the memristor is connected to a signal generator and a source meter via a switch matrix, and the source meter is used to test the current of the memristor within a certain period of time to calculate the average conductance of the memristor, and a certain threshold range is set. If it is within this range, the target conductance is reached by default.

[0013] In one embodiment, the memristor forms the basis of neuromorphic computing, mapping the elements in the matrix to the conductance of the memristor, and performing deep learning algorithm calculations through vector-matrix multiplication.

[0014] In one embodiment, the random pulse sequence is implemented through FPGA hardware, and sequential logic code is written to control multiple counters and state machines to generate a high-frequency random pulse sequence.

[0015] Alternatively, the random pulse sequence is implemented in software by a microcontroller in conjunction with a high-resolution DAC.

[0016] By using the above-mentioned method of regulating the memristor nanofilaments and reducing noise through a random pulse sequence, and comparing the method of setting the target conductance by a single pulse and a random pulse sequence, the experimental results show that the target conductance set by a random pulse sequence has lower noise than the target conductance set by a single pulse. Adding random positive and negative pulse signals in the process of setting the target conductance can reduce the noise of the device, and is not limited to RTN noise. This method does not require additional circuits and algorithms, and can achieve the effect of reducing noise in the process of setting the target conductance of the device. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other features of the present invention will be further described below in conjunction with the accompanying drawings and detailed description thereof. It should be understood that these drawings only illustrate several exemplary embodiments according to the present invention and therefore should not be regarded as limiting the scope of protection of the present invention. Unless otherwise specified, the drawings are not necessarily to scale and similar reference numerals represent similar components.

[0018] Figure 1 This is the layout of the device in some embodiments of this application; Figure 2 A flowchart of a method for reducing noise provided in some embodiments of the present application; Figure 3 In some embodiments of the present application, the form of random pulse sequence 1; Figure 4 In some embodiments of the present application, the form of random pulse sequence 2; Figure 5 In some embodiments of the present application, a single pulse is used; Figure 6 It is the target conductance RMS difference between a single pulse and a random pulse in some embodiments of the present application. DETAILED DESCRIPTION

[0019] The following detailed description refers to the drawings which form a part of this specification. The exemplary embodiments mentioned in the specification and drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. It will be appreciated by those skilled in the art that many other embodiments may be adopted and that various changes may be made to the described embodiments without departing from the spirit and scope of protection of the present invention. It should be understood that the various aspects of the present invention described and illustrated herein may be arranged, replaced, combined, separated and designed in many different configurations, all of which are included in the present invention.

[0020] First, to study the noise characteristics of memristors, we designed a cross-structured double-layer memristor (Pt / Ti / HfO2 / Al2O3 / Pt). The device layout is shown in the attached figure. Figure 1As shown. The manufacturing process is as follows: photolithography of the bottom electrode 1 pattern, magnetron sputtering of 10 nm Ti as an adhesion layer, and deposition of 50 nm Pt as the bottom electrode 1. Atomic layer deposition of 10 nm HfO2 and 2 nmAl2O3 films as functional layer 2, the Al2O3 layer can improve the stability and consistency of the device. Photolithography of circular through holes, deposition of 100 nm Al2O3 insulating layer 3 by ion beam coating machine, and peeling to form a through-hole structure. Photolithography of the top electrode 4 pattern, magnetron sputtering of 20 nm Ti as a functional electrode, and then deposition of 100 nm Pt as a protective layer and top electrode 4 to fill the through hole. When the device is wire bonded for testing, the bottom electrode 1 is connected to the ground. The signal generator and the source meter are connected through the switch matrix. The signal generator is used to generate pulse signals, and the source meter is used to read the current signal at a constant voltage.

[0021] In the process of CF (conductive filament) formation, some incomplete conductive channels will be randomly generated in the memristor. These incomplete channels will affect the charge transfer of CF, resulting in noise and unstable changes in conductivity in the device. The incomplete channels generated by a single pulse in the process of setting conductivity cannot be eliminated, but in the process of setting conductivity by random pulses, the simultaneous action of positive and negative pulses can increase the probability of eliminating incomplete channels, thereby reducing the noise of the device in the process of setting conductivity.

[0022] To generate a random pulse signal, we first create a sequence of positive and negative pulses with a certain ratio and amplitude, randomly extract the pulse voltage from the pulse sequence, then extract the random time interval from the Poisson distribution, and finally generate the corresponding random pulse sequence according to the number of pulses. Poisson distribution is a discrete probability distribution that describes the probability distribution of the number of events occurring in a given time period or spatial region. Poisson distribution is applicable to independent and sparsely occurring events, where the occurrence of each event is independent and does not affect each other.

[0023] Embodiment 1: Figure 2 A flowchart of a method for reducing noise provided by at least one embodiment of the present disclosure. Figure 2 As shown, the noise reduction method includes the following steps: creating a voltage sequence with the same positive and negative pulse ratios and different positive and negative pulse amplitudes; using a random function in Python to extract the pulse amplitude from the voltage sequence, and then generating a random time interval (interval is not 1) through a Poisson distribution function, and repeating the operation according to a given number of pulses to generate a random pulse sequence 1. In the device used in the present invention, the pulse width is 10 us and 1 us, the number of pulses is set to 1000, and the pulse form is as shown in the attached figure. Figure 3 For different memristors, the parameters such as pulse amplitude, width and number can be optimized according to the actual device performance.

[0024] As an example, the random function (code) is as follows: positive_state=np.array([pulse_.voltage]*(int(pulse_.points*positive_ratio)))#Generate the corresponding number of positive pulses negative_state=np.array([-pulse_.voltage]*(int(pulse_.points*(1-positive_ratio)))#Generate the corresponding number of negative pulses waveform._data=np.concatenate((positive_state,negative_.state))#positive and negative pulse list #Create a new waveform final_waveform[] for _in range(pulse_points): # Randomly pick a pulse pulse_choice np.random.choice(waveform_data) final_waveform.append(pulse_choice) #Randomly extract the time interval of Poisson distribution Lambda_poisson 1 interval = np.random.poisson(Lambda_.poisson) + 1 # Add 1 to ensure that the Poisson distribution does not take d final_waveform.extend([0]*interval)#Add a Poisson interval after the pulse (indicated by O) print(f"p_ratio:{(np.sum((np.array(final_waveform))>0)) / pulse_points}") Embodiment 2: Create a voltage sequence with the same ratio of positive and negative pulses and different amplitudes. Use the random function in Python to extract the pulse amplitude from the voltage sequence, and then generate a random time interval (interval is not 1) through the Poisson distribution function. Repeat the operation according to the given number of pulses to generate the form of random pulse sequence 2, and the amplitude of positive and negative pulses gradually decreases. The pulse width is set to 10 us and 1 us, the number of pulses is set to 1000, and the pulse form is as shown in the attached figure. Figure 4For different memristors, the parameters such as pulse amplitude, width and number can be optimized according to the actual device performance.

[0025] Device noise testing process and how to compare the noise level: The conductance state of the memristor will increase or decrease according to the applied positive and negative pulse voltages. When setting a random pulse sequence, the amplitude of the positive pulse is greater than the amplitude of the negative pulse, which increases the conductance. Conversely, the amplitude of the negative pulse is greater than the amplitude of the positive pulse, which decreases the conductance. Apply different forms of random pulse sequences to the device to increase the device conductance to the target conductance value. If the conductance exceeds the target conductance value, apply a random pulse sequence with a higher negative pulse. After setting the target conductance, read the current signal at a low voltage. Based on the read current data, calculate the normalized RMS of the current to measure the relative fluctuation of the current, which is used as a criterion for judging the size of the noise. We used a single pulse to set the target conductance and a random pulse to set the target conductance, and compared the results to find that the conductance noise of the random pulse setting is lower than that of the single pulse setting.

[0026] Positive and negative pulses will increase or decrease the conductivity of the device. After each pulse signal is applied, the source meter tests the current of the device for 2 s to calculate whether the average conductivity reaches the target conductivity. The threshold range is set to 20 uS. If it is within this range, the target conductivity is reached by default. After setting the target conductivity, the current of the device is tested for 100 s, the sampling rate is 10K, and the normalized RMS of the 100 s current data is calculated as the noise to measure the target conductivity state. The pulse width of a single pulse is 10 us, and the pulse width of a random pulse is 10 us or 1 us. The pulse form is as shown in the attached figure. Figure 3 and 4 As shown. The target conductance was set from 200 uS to 2000 uS in several different forms of single pulse and random pulse. The current data and normalized RMS value were recorded after each setting to the target conductance state. The RMS of the same target conductance state of the single pulse and the random pulse were subtracted, and the mean and standard deviation of the difference of ten cycles of testing were plotted in the same figure as shown in the attached figure. Figure 6 . We found that compared with a single pulse, the conductance RMS of the random pulse setting is lower, that is, the target conductance noise of the random pulse setting is lower, especially in the low conductance state. When we use pulses to control the conductance of the memristor, we add random positive and negative pulse signals to the pulse sequence, and the noise of the target conductance is lower than that of the single pulse setting.

[0027] Controlling memristor conductance and reducing noise via random pulse trains in neuromorphic computing: Memristors are emerging memory devices, and the integrated storage and computing architecture formed by memristor arrays has become the basis of neuromorphic computing. The core of deep learning algorithms is vector-matrix multiplication. We map the elements in the matrix to the conductance of the memristor, and calculate and store them at the same position of the cross array. Vector-matrix multiplication can be achieved based on physical principles such as Ohm's law and Kirchhoff's law. If different matrix multiplication calculations are implemented, the memristors in the array need to be adjusted to different target conductances. In order to reduce the error caused by conductance noise disturbance, the noise of the device can be reduced by adding random pulse signals in the process of adjusting the target conductance, thereby improving the accuracy of neural network calculations. Random pulse sequences can be implemented through FPGA hardware, writing sequential logic code, controlling multiple counters and state machines, and generating high-frequency random pulse sequences. Pulse sequences can also be defined in software through microcontrollers (such as Arduino, STM32, etc.) combined with high-resolution DAC (digital-to-analog converter) and output to DAC to generate pulse signals. The microcontroller controls the frequency and width of the pulse through a timer and outputs a voltage waveform through the DAC. Different hardware methods can be selected according to the actual circuit to achieve the generation of random pulse sequences. This method does not require additional algorithms and a separate pulse noise reduction process, and achieves the purpose of noise reduction during the process of setting the target conductance, thereby reducing the complexity and time cost of noise reduction.

[0028] In the description of the present invention, it is necessary to understand that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0029] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0030] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0031] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0032] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0033] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for regulating memristor nanofilaments and reducing noise by random pulse sequences, comprising the following steps: Create a voltage sequence with the same ratio of positive and negative pulses but different amplitudes of positive and negative pulses; Use a random function to extract the pulse amplitude from the voltage sequence; Generate a random time interval that is not 1 through a discrete probability distribution function; The random pulse sequence is generated by repeating the operation according to the given number of pulses; A signal in the form of the above-mentioned random pulse sequence is applied to the memristor to regulate the memristor nanofilaments and reduce noise.

2. The method according to claim 1, characterized in that The amplitudes of the positive and negative pulses gradually decrease.

3. The method according to claim 1, characterized in that The discrete probability distribution function is a Poisson distribution.

4. The method according to claim 1, characterized in that The memristor is a double-layer memristor with a cross structure.

5. The method according to claim 4, characterized in that The memristor uses Ti as an adhesion layer, Pt as a bottom electrode, HfO2 and Al2O3 as functional layers; uses Al2O3 as an insulating layer, and is peeled off to form a through-hole structure.

6. The method according to claim 5, characterized in that The memristor uses Ti as a functional electrode, Pt as a protective layer and a top electrode to fill the through hole.

7. The method according to claim 1, characterized in that The memristor is connected to a signal generator and a source meter through a switch matrix. The source meter is used to test the current of the memristor within a certain period of time to calculate the average conductance of the memristor. A certain threshold range is set. If it is within this range, the target conductance is reached by default.

8. The method according to claim 1, characterized in that The memristor forms the basis of neuromorphic computing, mapping the elements in the matrix to the conductance of the memristor, and performing deep learning algorithm calculations through vector-matrix multiplication.

9. The method according to claim 8, characterized in that The random pulse sequence is implemented through FPGA hardware, and a timing logic code is written to control multiple counters and state machines to generate a high-frequency random pulse sequence.

10. The method according to claim 8, characterized in that The random pulse sequence is implemented in software by a microcontroller in conjunction with a high-resolution DAC.

Images