Oscillation convergence programming method and device for multi-valued resistive random access memory

Through the oscillation convergence programming method, the initial parameters and gate voltage are adjusted, which solves the programming efficiency and accuracy of resistive variable memory, improves the programming efficiency and accuracy of multi-value memory, and is suitable for edge computing and neural network accelerated computing.

CN118760422BActive Publication Date: 2024-11-26ZHEJIANG LAB
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411246402.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-06
Publication Date
2024-11-26
Estimated Expiration
2044-09-06

AI Technical Summary

Technical Problem

In the prior art, the programming efficiency of resistive variable memory is low and the multi-resistive state programming accuracy is insufficient, which makes it difficult to balance programming efficiency and programming accuracy.

Method used

The oscillation convergence programming method of multi-value resistive memory is adopted. By setting initial parameters in advance, the control signal is generated for initialization, and the gate voltage is adjusted according to the measured current value, so that the multi-value programming of the resistive memory is realized until the target conductance is reached. Within range.

Benefits of technology

It improves the multi-value storage accuracy and programming efficiency of large-scale resistive memory arrays, enhances the weight modulation accuracy and speed in the integrated memory process, and is suitable for edge computing and neural network accelerated computing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118760422B_ABST
    Figure CN118760422B_ABST
Patent Text Reader

Abstract

The present invention provides an oscillation convergence programming method and device for a multi-valued resistive random access memory, the method obtains the target value range of the resistive random access memory and initializes it. The current value of the resistive random access memory is obtained, and if it is within the target range, the programming is successful; if the difference between the current value of the resistive random access memory and the target range is within the threshold, the step value is reduced; if the current value of the resistive random access memory is greater than the target value, a pulse signal is applied to the resistive random access memory to make it reach the lowest value, and the current value of the gate voltage minus the gate voltage step value is used as the gate voltage; otherwise, the current value of the gate voltage plus the gate voltage step value is used as the gate voltage; the gate voltage is applied to the resistive random access memory, a set operation is performed, and the above process is repeated until the target value range of the resistive random access memory is reached or the maximum programming number or time is exceeded. The present invention can quickly converge the resistive random access memory to the target value, and can be accelerated by parallel programming to reduce power consumption during the programming process.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of memristor storage and storage-computation integrated computing, and in particular to an oscillation convergence programming method and device for a multi-valued resistive random access memory. Background Art

[0002] With the development of integrated circuits, Moore's Law has gradually come to an end, limiting the further improvement of data computing efficiency under the existing CMOS technology. In addition, computer technology based on the von Neumann architecture has power consumption wall and storage wall problems, which further restricts the increase in integration and computing efficiency. Artificial intelligence technology requires a large amount of matrix numerical calculations, and edge computing requires more low-power and energy-efficient computing devices. Resistive Random Access Memory (RRAM) has the advantages of small size, CMOS process compatibility, non-volatility, multi-value storage, storage and computing integration, and low computing energy efficiency. It has broad application prospects in edge computing, high-density storage, neural network acceleration computing, etc. At present, the integration method of resistive random access memory based on crossbar structure has become the mainstream of practical applications. However, the crossbar based on RRAM alone has problems such as high leakage current and difficulty in device modulation. The crossbar integration method with the transistor-resistive random access memory (one-transistor-one-resistor, 1T1R) structure formed by MOS tube and RRAM in series as the basic storage unit has been unanimously sought after.

[0003] Due to the presence of transistors, 1T1R devices have more controllable characteristics, and when the transistor operates in the saturation region, due to the current limiting characteristics of the transistor, in most cases, the device conductance (or output current) of the set operation has an approximately linear positive correlation with the gate voltage of the transistor, making it possible to control the device.

[0004] At present, there are various multi-value storage programming methods for resistive random access memory. The method based on pulse width to adjust the device conductivity has the problem of complex control circuit due to the continuous change of pulse width. Adjusting the voltage of the bit line BL and the source line SL is a feasible programming scheme, and the gate voltage remains fixed at this time. When the device conductivity is lower than the target conductivity, the BL voltage is increased and SL is grounded; when the device conductivity is higher than the target conductivity, the SL voltage is increased and BL is grounded. By adjusting the voltage, the device conductivity is brought into a tolerable conductivity range. However, due to the high voltage between the device conductivity and the bit line voltage V BL And the source line voltage V SLThere is no deterministic relationship between the device conductance and the gate voltage, so it is difficult to efficiently obtain the corresponding programming voltage. Only a scanning method starting from zero can be used, which reduces the programming efficiency and causes multiple erasures, which reduces the service life of the device. The two-pulse programming strategy based on the linear correspondence between device conductance and gate voltage can accurately determine the voltage value corresponding to the device conductance, thereby performing reset and set operations, and has extremely high programming efficiency. However, the cycle-to-cycle variation characteristics of the device and the nonlinear scanning characteristics of the device make the programming accuracy of the two-pulse programming strategy relatively low, reducing the utilization efficiency of the multi-value storage characteristics of RRAM.

[0005] Therefore, a new programming strategy is needed, which can not only improve the programming efficiency of the resistive random access memory, but also make full use of the multi-value storage characteristics of the resistive random access memory. Summary of the invention

[0006] The purpose of the present invention is to provide a multi-value resistive memory oscillation convergence programming method and device to solve the problems of low programming efficiency and insufficient multi-resistance state programming accuracy in related technologies, and to solve the problem of difficulty in balancing programming efficiency and programming accuracy in the prior art.

[0007] The technical solution adopted by the present invention is as follows:

[0008] The first aspect of the present invention provides an oscillation convergence programming method for a multi-valued resistive random access memory, the method comprising the following steps:

[0009] (1) Presetting the initial parameters required in the programming process of the resistive memory according to the resistive memory device and array characteristics, and using part of the initial parameters to generate a control signal to initialize the resistive memory; and the control signal determines whether to use the existing gate operating voltage to set the device; and generates a read pulse signal;

[0010] (2) When a read pulse signal of the resistive random access memory is applied, the current output by the array or device is measured and then input into the subsequent decision step for judgment;

[0011] (3) Obtaining the current value of the resistive random access memory from the current measured in step (2), and determining whether the current value is within the target range. If it is within the target range, the programming is completed and the process ends;

[0012] (4) Determine whether the current value of the resistive memory is within the range of the step value threshold constraint, thereby adjusting the size of the step value and the step value threshold; determine the size relationship between the current value of the resistive memory and the target range, decide whether to generate a signal to adjust the initialization conductance, and adjust the gate control voltage value;

[0013] (5) Determine whether the current gate voltage is within the gate voltage constraint range. If it is not within the gate voltage constraint range, exit the current programming process or initialize the gate voltage and start a new round of programming;

[0014] (6) adjusting the initial parameters in step (1) according to the judgment results of steps (3) to (5) to generate a gate voltage value signal to perform a setting operation on the resistive memory cell;

[0015] (7) Repeat steps (3) to (6) to continue adjusting various parameters until programming is completed, that is, the oscillation convergence of the multi-valued resistive random access memory is achieved.

[0016] Furthermore, the initialization in step (1) is specifically as follows: resetting the resistive memory to a minimum conductance value; performing a set operation under a voltage pulse constraint with an initial gate voltage value, and setting the resistive memory to a specific value.

[0017] Specifically, the initial parameters of step (1) are target value, error range constraint, reset voltage, gate voltage initial value, step value, adjustment step value threshold and gate voltage range constraint.

[0018] Furthermore, the step (4) is implemented by the following sub-steps:

[0019] (4.1) If the current value of the resistive memory is within the step value threshold constraint range, the step value is reduced, thereby performing more precise adjustment within a narrower conductance range and reducing the step value threshold constraint;

[0020] (4.2) If the current value of the resistive memory is greater than the upper limit of the target range, a complete reset operation is performed on the resistive memory unit, and the gate voltage value is set to the difference between the current value and the gate voltage step value; the upper limit of the target range is the target value plus the error range constraint;

[0021] (4.3) If the current value of the resistive memory is less than the lower limit of the target range, the gate voltage value is set to the sum of the current value and the gate voltage step value; the lower limit of the target range is the target value minus the error range constraint.

[0022] Specifically, the process of the setting operation in step (6) is to apply a gate voltage to the word line corresponding to the resistive memory and apply a saturated drain voltage of the transistor to the bit line, so that the MOS tube works in the saturation region at this time, and the resistive memory is regulated by the current limiting of the MOS tube.

[0023] Specifically, various parameters are adjusted in step (7), during which the maximum gate voltage V Gmax Constrained to be greater than the initial gate voltage V G , initial gate voltage value V G Greater than the minimum gate voltage constraint V Gmin, that is, V Gmax >V G >V Gmin ; Secondly, the threshold TH of the initial adjustment step value is greater than the preset error range constraint t, that is, TH>t, and it is also compatible with the situation where the threshold TH of the adjustment step value is less than the preset error range constraint t.

[0024] Specifically, the resistive memory cells are cross-arranged to form a resistive memory storage and computing integrated array, which includes a number of word lines, bit lines and source lines. Each intersection of a bit line and a source line is a basic storage unit, which includes a transistor and a resistive memory, wherein the gate of the transistor is connected to the word line.

[0025] Specifically, the initial parameters and configuration parameters in the programming process need to be set according to the actual characteristics of the device. Such parameters are configured by engineering personnel according to actual conditions and are not specifically limited.

[0026] Specifically, a matrix with positive and negative values ​​in practical applications is mapped to a non-negative matrix in a specific corresponding manner, and then mapped to the conductance or current of the resistive random access memory, and the corresponding write conductance matrix parameters are obtained.

[0027] The second aspect of the present invention provides an oscillation convergence programming device for a multi-valued resistive random access memory, comprising:

[0028] Signal generation module: according to the resistive memory device and array characteristics, the initial parameters required in the programming process of the resistive memory are pre-set, and part of the initial parameters are used to generate control signals to initialize the resistive memory; the initial parameters are target value, error range constraint, reset voltage, gate voltage initial value, step value, adjustment step value threshold and gate voltage range constraint; the control signal determines whether to use the existing gate operation voltage to set the device; generates a read pulse signal;

[0029] Current detection module: when the signal generation module applies the RRAM read signal, the current output by the array or device is measured and then input into the decision module for judgment;

[0030] Judgment module: obtain the current measured by the current detection module to obtain the current value of the resistive memory, and judge whether the current value is within the target range. If it is within the target range, the programming is completed and the process ends; judge whether the current value of the resistive memory is within the range of the step value threshold constraint, so as to adjust the size of the step value and the step value threshold; judge the size relationship between the current value of the resistive memory and the target range, decide whether to generate a signal to initialize the conductivity, and adjust the gate control voltage value; judge whether the current gate voltage is within the gate voltage constraint range, if not within the gate voltage constraint range, then exit the current programming process or initialize the gate voltage and start a new round of programming;

[0031] Control module: adjusts various initial parameters in the signal generation module according to the judgment result of the judgment module, generates a gate voltage value signal to perform a setting operation on the resistive memory unit;

[0032] Oscillation convergence module: Repeat the judgment module to the setting operation module to continue adjusting various parameters until the programming is completed, that is, the oscillation convergence of the multi-valued resistive memory is achieved.

[0033] The beneficial effects of the present invention are:

[0034] It accelerates the multi-value storage of large-scale resistive memory arrays, effectively improves the accuracy of multi-value storage modulation, and thus increases the number of distinguishable states of multi-value storage; it can be effectively used in the weight modulation process in the storage and computing integration process, improves the accuracy and speed of weight adjustment, and is used to accelerate the weight writing process for applications such as target detection and speech processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0036] Figure 1 It is a structural schematic diagram of an existing 1T1R cross structure array;

[0037] Figure 2 The measured relationship between the conductance and the control gate voltage of the 1T1R structure resistive memory device of the present invention is shown in FIG. 1 , wherein the gray points correspond to the measured points and the curve is the curve corresponding to the median;

[0038] Figure 3 A flowchart of an oscillation convergence programming method for a resistive random access memory according to the present invention;

[0039] Figure 4 A schematic diagram of a RRAM convergence process of the present invention;

[0040] Figure 5 It is a flow chart of the convergence process of the resistive random access memory according to the present invention;

[0041] Figure 6 This is a test result diagram of a multi-value storage technology based on a memristor according to an embodiment of the present invention;

[0042] Figure 7 This is a histogram of the error distribution of the memristor-based integrated weight mapping according to an embodiment of the present invention;

[0043] Figure 8A schematic diagram of the structure of a multi-value programming device based on a memristor array provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] It should be noted that, in the absence of conflict, the features in the following embodiments and implementations may be combined with each other.

[0046] like Figure 1 The figure shows the basic structure of a one-transistor-one-resistor (1T1R) cross array formed by existing RRAM in series, where WL (word line) represents the word line, which is connected to the gate of the MOS tube; BL (bit line) represents the bit line, which is connected to the RRAM set signal terminal; and SL (source line) represents the source line, which is connected to the source of the MOS tube.

[0047] During all operations, increasing the voltage means that a voltage with a larger amplitude is applied when a voltage pulse is applied next time, and decreasing the voltage means that a voltage with a smaller amplitude is applied when a voltage pulse is applied next time. It does not mean that a DC operating voltage is always applied.

[0048] like Figure 2 As shown, the device conductance obtained by the read operation after setting with different control gate voltages is expressed, indicating that there is a linear correspondence between the conductance of the device and the control gate voltage, and that the device can be efficiently programmed by the control gate voltage.

[0049] like Figure 3 As shown in FIG. 1 , a flowchart of an oscillation convergence programming method for a multi-valued resistive random access memory provided by an embodiment of the present invention is shown. Figure 3 As shown, the oscillation convergence programming method in the embodiment of the present invention may include the following steps:

[0050] Step 1: Get the RRAM target value g T (conductance or current), set initial parameters and configuration parameters, such as error range constraint t, reset voltage V RESET , initial gate voltage value V G , step value S, threshold value TH for adjusting step value, reduction amount Δ of step value, maximum gate voltage V Gmax , minimum gate voltage V Gmin etc., and use V GInitialize the resistive memory.

[0051] Step 2: Get the current value g of the resistive memory C , and judge whether it is within the target range (g T – t, g T + t), if it is within the target range, the programming is completed and the process ends.

[0052] Step 3: Determine the current value g of the RRAM C and whether it is in the range (g T – TH, g T + TH), if it is within this range, then reduce the step value S to S – Δ, and reduce the step value threshold TH; if the current value of the resistive memory g C Less than g T -t, then the gate voltage value V G Set to the sum of the current value and the gate voltage step value, that is, V G + S; if the current value of the resistive memory is g C Greater than g T + t, then a reset voltage V is applied to the resistive memory cell. RESET , perform a complete reset operation and set the gate voltage value V G Set to the difference between the current value and the gate voltage step value, that is, V G – S.

[0053] Step 4: Determine whether the current gate voltage value is greater than the maximum voltage V Gmax Or less than the minimum voltage V Gmin If so, exit the current programming process or initialize the gate voltage and start a new round of programming.

[0054] Step 5: Using the gate voltage value to perform a setting operation on the resistive memory cell.

[0055] Step 6: Repeat steps 2 to 5 until programming is completed and exit the programming process.

[0056] Among them, the maximum gate voltage is greater than the initial gate voltage, and the initial gate voltage is greater than the minimum gate voltage, that is, V Gmax >V G >V Gmin In general, the threshold TH of the initial adjustment step value is greater than the preset error range constraint t, that is, TH>t. At the same time, it is also compatible with the situation where the threshold of the adjustment step value is less than the preset error range.

[0057] Specifically, step three is implemented through the following sub-steps:

[0058] (3.1) If the current value of the RRAM is within the range (g T– TH, g T + TH), the step value is reduced to S-Δ, so as to make more precise adjustment in a narrower conductance range and reduce the step value threshold TH at the same time.

[0059] (3.2) The current value of the RRAM is greater than the target value plus the error range, i.e., g C > T +t, a complete reset operation is performed on the resistive memory cell, and the gate voltage value is set to the difference between the current value and the gate voltage step value.

[0060] (3.3) The current value of the RRAM is less than the target value minus the error range, i.e., g C <g T – t, the gate voltage value is set to the sum of the current value and the gate voltage step value.

[0061] Optionally, in one embodiment of the present application, it is necessary to set the initial parameters and configuration parameters in the programming process according to the actual characteristics of the device. Such parameters are configured by engineering personnel based on actual conditions and are not specifically limited.

[0062] Optionally, in one embodiment of the present application, a matrix with positive and negative values ​​in actual applications is mapped to a non-negative matrix in a specific corresponding manner, and then mapped to the conductivity (or current) of the resistive random access memory, and the corresponding write conductivity matrix parameters are obtained.

[0063] Optionally, during the set operation, a gate voltage is applied to the word line corresponding to the resistive memory, and a transistor saturation drain voltage is applied to the bit line, so that the MOS tube operates in the saturation region at this time, and the current limiting effect of the MOS tube is used to more effectively control the resistive memory.

[0064] Example 1: Example schematic diagram as shown Figure 5 The flowchart of the RRAM convergence process shown in FIG. Figure 4 Schematic diagram of the convergence process of the resistive random access memory. In the adjustment process of the present invention, it is assumed that the write device uses conductance as the measurement standard and the target conductance is g T , this embodiment is mainly divided into the following steps:

[0065] Step 1: According to the characteristics of the RRAM device and array, set the initial parameters of the device, such as the error range t and the initial gate voltage value V G , step value S, threshold value TH for adjusting step value, reduction amount Δ of step value, maximum gate voltage V Gmax , minimum gate voltage V Gmin , and initialize the device, which is recorded as time t0, and read the device conductivity as g0. The relationship between g0 and the target range is determined, and it can be obtained that g0 is less than the lower limit of the target range g T-t, and g0 is not within the range of the threshold constraint. At this time, keep the step value unchanged, adjust the threshold of the step value unchanged, and increase the gate voltage according to the step value.

[0066] Optionally, in the initialization process in step 1, the device may be reset to the lowest conductance of the device first, and then a set operation may be performed under the constraint of the initial value of the gate voltage to obtain the initial conductance.

[0067] Step 2: Under the existing gate operating voltage, the device is set; after the set operation, this time is recorded as time t1, and the device conductance g1 is read. After judgment, it can be known that the conductance g1 is less than the lower limit of the target range g T -t, and it is not within the range of the threshold constraint, at this time, the step value and the threshold of the adjustment step value are kept unchanged, and the gate voltage is increased according to the step value.

[0068] Step 3: Under the existing gate operating voltage, the device is set; after the set operation, this time is recorded as time t2, and the device conductance g2 is read. After judgment, the g2 conductance is greater than the upper limit of the target range g T +t, and is not within the threshold constraint range, at this time, the step value and the threshold of the adjustment step value are still kept unchanged, and the resistive memory is completely reset, and the gate voltage is reduced according to the step value.

[0069] Step 4: Under the existing gate operating voltage, the device is set; after the set operation, at time t3, the device conductance g3 is read. After judgment, the conductance g3 is greater than the upper limit of the target range g T +t, and is not within the threshold constraint range, at this time, the step value and the threshold of the adjustment step value are still kept unchanged, and the resistive memory is completely reset, and the gate voltage is reduced according to the step value.

[0070] Step 5: Under the existing gate operating voltage, the device is set; after the set operation, at time t4, the device conductance g4 is read. After judgment, the conductance g4 is less than the lower limit of the target range g T -t, and within the threshold constraint range (g T – TH,g T + TH), at this time, reduce the step value (S = S-Δ) and the step value threshold, and increase the gate voltage according to the current step value.

[0071] Step 6: Under the existing gate operating voltage, the device is set, which is recorded as time t5. The device conductance g5 is read. After judgment, g5 is greater than the upper limit of the target range g T +t, and is not within the threshold constraint range, at this time, the step value and the threshold of the adjustment step value are still kept unchanged, and the resistive memory is completely reset, and the gate voltage is reduced according to the step value.

[0072] Step 7: Under the existing operating gate voltage, the device is set, which is recorded as time t6. The device conductance g6 is read. After judgment, g6 reaches the target range and programming is successful.

[0073] In order to fully explain the details of each step, the above embodiment is a relatively extreme case, making full use of Figure 3 Each link in the programming scheme diagram.

[0074] Embodiment 2: Figure 6 The actual test results of a multi-value storage embodiment are shown. This embodiment tests the storage characteristics of 4-bit data. The embodiment is described in detail below. This embodiment includes the following steps:

[0075] Step 1: Map the binary data to the conductance range of the memristor (G L , G H ), set the number of data bits N, and the conductivity range is In this embodiment, the conductivity interval used is (40 μS, 190 μS). Since the stored data is 4 bits, every 10 μS is used as a storage interval. At this time, 35 μS-45 μS corresponds to data 0000, 45 μS-55 μS corresponds to data 0001…

[0076] Step 2: Correspond the binary data to be stored to the corresponding conductivity range, or to the middle value of the conductivity range.

[0077] Step 3: Set the constraints of the programming scheme according to the actual situation, set the target conductance to the middle value of the corresponding conductance range, set the error range to be slightly less than half of the conductance range, and then use the memristor oscillation convergence programming method to gradually approach the conductance range, so that the data can be efficiently stored in the storage range.

[0078] Through this method, this embodiment successfully stored 1600 device unit data, where each data corresponds to 100 device units. When setting a smaller conductivity range constraint and a finer-grained adjustment step, data storage with a higher number of bits can be achieved. This embodiment proves that the oscillation convergence programming method can be effectively applied to multi-value data storage of memristors and further used for computing operations.

[0079] Example 3: Figure 7 The error test results of a storage-computation integrated parameter matrix programming are shown. This embodiment is used to store simulation domain weights. The embodiment is described in detail below. This embodiment includes the following steps:

[0080] Step 1: Map the parameter matrix to the array conductance interval (GL , G H ), obtain the target conductance matrix, and set the device programming error range according to specific requirements;

[0081] Step 2: Using the RRAM oscillation convergence programming method to write the target conductivity matrix into the RRAM array;

[0082] Step 3: After successful programming, the actual conductance value of the current array is obtained and compared with the target conductance matrix.

[0083] Optionally, the input signal is programmed into the form of a voltage pulse, the input array performs matrix calculation, and the output current is converted into an output value through an ADC.

[0084] Optionally, the ADC output value is compared with the actual output value to obtain corresponding parameters, and the ADC output value is converted into array output.

[0085] Optionally, the array output value is compared with the expected output value to determine the programming error.

[0086] In this embodiment, the error range is set to 1 μS. Through the programming method, the device conductance range after programming is constrained to the range of (-1 μS, 1 μS). According to the actual application requirements, the device conductance can be constrained to a narrower or wider conductance range. In this embodiment, the target conductance matrix is ​​programmed and the programming error is statistically analyzed to prove that the oscillation convergence programming method can be used for accurate programming of memristors.

[0087] The schematic diagram of the device block corresponding to the aforementioned multi-valued resistive memory oscillation convergence programming method is shown in FIG. Figure 8 As shown, a resistive memory oscillation convergence programming device includes the following modules:

[0088] Signal generation module: according to the resistive memory device and array characteristics, the initial parameters required in the programming process of the resistive memory are pre-set, and part of the initial parameters are used to generate control signals to initialize the resistive memory; the initial parameters are target value, error range constraint, reset voltage, gate voltage initial value, step value, adjustment step value threshold and gate voltage range constraint; the control signal determines whether to use the existing gate operation voltage to set the device and generate a read pulse signal;

[0089] Specifically, the signal generation module generates a pulse signal according to the control module, and acts on the resistive memory array, thereby generating a read and write operation.

[0090] Current detection module: when the signal generation module applies the RRAM read signal, the current output by the array or device is measured and then input into the decision module for judgment;

[0091] Specifically, the control signal means a reset and a set voltage, and the detection voltage means a read pulse signal.

[0092] Judgment module: obtain the current measured by the current detection module to obtain the current value of the resistive memory, and judge whether the current value is within the target range. If it is within the target range, the programming is completed and the process ends; judge whether the current value of the resistive memory is within the range of the step value threshold constraint, so as to adjust the size of the step value and the step value threshold; judge the size relationship between the current value of the resistive memory and the target range, decide whether to generate a signal to initialize the conductivity, and adjust the gate control voltage value; judge whether the current gate voltage is within the gate voltage constraint range, if not within the gate voltage constraint range, then exit the current programming process or initialize the gate voltage and start a new round of programming;

[0093] Control module: adjusts various initial parameters in the signal generation module according to the judgment result of the judgment module, generates a gate voltage value signal to perform a setting operation on the resistive memory unit;

[0094] Oscillation convergence module: Repeat the judgment module to the setting operation module to continue adjusting various parameters until the programming is completed, that is, the oscillation convergence of the multi-valued resistive memory is achieved. The control module can be a computing system for controlling code programming or a hardware state machine, which can also be implemented by combining software and hardware.

[0095] Optionally, the control module and the judgment module both serve as control modules to implement signal judgment and control signal output.

[0096] According to the device proposed in this embodiment, the signal can be controlled, which can not only be used for the storage and reading process of the resistive memory array, but also support the calculation process of the resistive memory array.

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An oscillation convergence programming method for a multi-valued resistive random access memory, characterized in that: The method comprises the following steps: (1) Presetting the initial parameters required in the programming process of the resistive memory according to the resistive memory device and array characteristics, and using part of the initial parameters to generate a control signal to initialize the resistive memory; and the control signal determines whether to use the existing gate operating voltage to set the device; and generates a read pulse signal; (2) When a read pulse signal of the resistive random access memory is applied, the current output by the array or device is measured and then input into the subsequent decision step for judgment; (3) Obtaining the current value of the resistive random access memory from the current measured in step (2), and determining whether the current value is within the target range. If it is within the target range, the programming is completed and the process ends; (4) Determine whether the current value of the resistive memory is within the range of the step value threshold constraint, thereby adjusting the size of the step value and the step value threshold; determine the size relationship between the current value of the resistive memory and the target range, decide whether to generate a signal to adjust the initialization conductance, and adjust the gate control voltage value; (5) Determine whether the current gate voltage is within the gate voltage constraint range. If it is not within the gate voltage constraint range, exit the current programming process or initialize the gate voltage and start a new round of programming; (6) adjusting the initial parameters in step (1) according to the judgment results of steps (3) to (5) to generate a gate voltage value signal to perform a setting operation on the resistive memory cell; (7) Repeat steps (3) to (6) to continue adjusting various parameters until programming is completed, that is, the oscillation convergence of the multi-valued resistive random access memory is achieved.

2. The oscillation convergence programming method of a multi-valued resistive random access memory according to claim 1, characterized in that: The initialization in step (1) specifically includes: resetting the resistive memory to a minimum conductance value; and performing a set operation under a voltage pulse constraint with an initial gate voltage value to set the resistive memory to a specific value.

3. The oscillation convergence programming method of a multi-valued resistive random access memory according to claim 1, characterized in that: The initial parameters of step (1) are target value, error range constraint, reset voltage, gate voltage initial value, step value, adjustment step value threshold and gate voltage range constraint.

4. The oscillation convergence programming method of a multi-valued resistive random access memory according to claim 1, characterized in that: The step (4) is implemented by the following sub-steps: (4.1) If the current value of the resistive memory is within the step value threshold constraint range, the step value is reduced, thereby adjusting within the conductance range and reducing the step value threshold constraint; (4.2) If the current value of the resistive memory is greater than the upper limit of the target range, the resistive memory unit is reset and the gate voltage value is set to the difference between the current value and the gate voltage step value; the upper limit of the target range is the target value plus the error range constraint; (4.3) If the current value of the resistive memory is less than the lower limit of the target range, the gate voltage value is set to the sum of the current value and the gate voltage step value; the lower limit of the target range is the target value minus the error range constraint.

5. The oscillation convergence programming method of a multi-valued resistive random access memory according to claim 1, characterized in that: The process of the setting operation in step (6) is to apply a gate voltage to the word line corresponding to the resistive memory and apply a saturated drain voltage of the transistor to the bit line, so that the MOS tube works in the saturation region at this time, and the resistive memory is regulated by the current limiting of the MOS tube.

6. The oscillation convergence programming method of a multi-valued resistive random access memory according to claim 1, characterized in that: In step (7), various parameters are continuously adjusted, during which the maximum gate voltage V Gmax Constrained to be greater than the initial gate voltage V G , initial gate voltage value V G Greater than the minimum gate voltage constraint V Gmin , that is, V Gmax > V G > V Gmin ; Secondly, the threshold TH of the initial adjustment step value is greater than the preset error range constraint t, that is, TH>t, and it is also compatible with the situation where the threshold TH of the adjustment step value is less than the preset error range constraint t.

7. The oscillation convergence programming method of a multi-valued resistive random access memory according to claim 1, characterized in that: The resistive memory cells are cross-arranged to form a resistive memory storage and computing integrated array, which includes a plurality of word lines, bit lines and source lines. Each intersection of a bit line and a source line is a basic storage unit, which includes a transistor and a resistive memory, wherein the gate of the transistor is connected to the word line.

8. The oscillation convergence programming method of a multi-valued resistive random access memory according to claim 1, characterized in that: The initial parameters and configuration parameters in the programming process need to be set according to the actual characteristics of the device. Such parameters are configured by the engineering personnel according to the actual situation and are not specifically restricted.

9. The oscillation convergence programming method of a multi-valued resistive random access memory according to claim 1, characterized in that: The matrix with positive and negative values ​​in practical applications is mapped to a non-negative matrix in a specific corresponding manner, and then mapped to the conductance or current of the resistive memory, and the corresponding write conductance matrix parameters are obtained.

10. An oscillation convergence programming device for a multi-valued resistive random access memory, characterized in that: include: Signal generation module: according to the resistive memory device and array characteristics, the initial parameters required in the programming process of the resistive memory are pre-set, and part of the initial parameters are used to generate control signals to initialize the resistive memory; the initial parameters are target value, error range constraint, reset voltage, gate voltage initial value, step value, adjustment step value threshold and gate voltage range constraint; the control signal determines whether to use the existing gate operation voltage to set the device; generating a read pulse signal; Current detection module: when the signal generation module applies the RRAM read signal, the current output by the array or device is measured and then input into the decision module for judgment; Judgment module: obtain the current measured by the current detection module to obtain the current value of the resistive memory, and judge whether the current value is within the target range. If it is within the target range, the programming is completed and the process ends; judge whether the current value of the resistive memory is within the range of the step value threshold constraint, so as to adjust the size of the step value and the step value threshold; judge the size relationship between the current value of the resistive memory and the target range, decide whether to generate a signal to initialize the conductivity, and adjust the gate control voltage value; judge whether the current gate voltage is within the gate voltage constraint range, if not within the gate voltage constraint range, then exit the current programming process or initialize the gate voltage and start a new round of programming; Control module: adjusts various initial parameters in the signal generation module according to the judgment result of the judgment module, generates a gate voltage value signal to perform a setting operation on the resistive memory unit; Oscillation convergence module: Repeat the judgment module to the setting operation module to continue adjusting various parameters until the programming is completed, that is, the oscillation convergence of the multi-valued resistive memory is achieved.