Resistive random access memory ReRAM data judgment method based on sneak path correlation

Abstract

The invention provides a resistive random access memory ReRAM data judgment method based on sneak path correlation. The method comprises the implementation steps that parameters are initialized; calculating a reading resistance value of the ReRAM storage unit; classifying the ReRAM storage units based on the sneak path correlation; calculating a log-likelihood ratio and a judgment threshold of each type of storage unit; and obtaining a ReRAM data judgment result. The storage units are classified according to the sneak path correlation values of the row and the column where each storage unit is located, and the likelihood ratio function of each category of storage units is calculated according to the conditional probability distribution of the resistance value read by the storage data of each storage unit under different value conditions. According to the method for calculating the likelihood ratio function by using the sneak path correlation, the influence of the difference of interference suffered by different storage units on the bit error rate is fully considered, and the judgment precision is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the field of memories and relates to a resistive random access memory (ReRAM) data decision method based on sneak path correlation. Background Art

[0002] A resistive random access memory (ReRAM) is a non-volatile memory that realizes data storage based on the change of resistance state. Its working principle is to apply different voltages to make the storage unit switch between the high resistance state and the low resistance state, thereby completing the writing of binary data "0" or "1". During the data reading process, a relatively small voltage is usually applied to the storage unit, and the current value flowing through the unit is measured and its resistance value is calculated, and then it is judged whether the data stored in the storage unit is "0" or "1". However, during the actual reading process, due to the influence of sneak paths and noise interference, the detected resistance value may deviate, resulting in data misreading, and this deviation will seriously affect the data accuracy of the memory.

[0003] ReRAM data decision refers to reducing the influence of sneak paths and noise interference on the detected resistance value through a specific algorithm or method, so as to accurately identify the original data "0" or "1" in the storage unit and ensure the reliability of data reading. For example, in the paper "Performance limit and coding schemes for resistive random-access memory channels" published by Song Guanghui et al. in the IEEE Transactions on Communications journal in April 2021, a ReRAM data decision method based on probability estimation and log-likelihood ratio detection was proposed. This method first estimates the proportion of storage units affected by sneak paths in ReRAM through hard decision on the read resistance value, and takes it as the probability of sneak path interference occurrence, then assumes that all storage units in ReRAM are affected by sneak path interference with this probability, and calculates the log-likelihood ratio of each storage unit based on the probability distribution of the read resistance value and the prior probability of the data stored in the storage unit, and finally makes a decision on the data stored in the storage unit according to the log-likelihood ratio. This method has a relatively high decision accuracy, but since the log-likelihood ratio of each storage unit is calculated based on the probability of the assumption that the storage unit is affected by sneak path interference, the difference in interference received by different storage units in practice is not fully considered, resulting in a still relatively low decision accuracy. Summary of the Invention

[0004] The purpose of the present invention is to overcome the defects existing in the above-mentioned prior art, and propose a resistive random access memory (ReRAM) data decision method based on the correlation between sneak paths, aiming to improve the data decision accuracy.

[0005] To achieve the above object, the technical solution adopted by the present invention includes the following steps:

[0006] (1) Initialize parameters:

[0007] Initialize the ReRAM to include M×N memory cells arranged periodically. The memory cell c at the m-th row and n-th column m,n stores binary data and the reference resistance value are x m,n and R m,n respectively, where M≥16 and N≥16;

[0008] (2) Calculate the read resistance value of the ReRAM memory cell:

[0009] According to the reference resistance value R m,n of the memory cell c m,n calculate the read resistance value y m,n of c m,n ;

[0010] (3) Classify the ReRAM memory cells based on sneak path correlation:

[0011] Calculate the sneak path correlation values L m,n and L m,n of the row and column where the memory cell c m is located through the read resistance value y n , and classify the M×N memory cells included in the ReRAM through L m and L n to obtain type A memory cells where both the row and column are affected by the sneak path, type B memory cells where neither the row nor the column is affected by the sneak path, and type C memory cells where the row is affected by the sneak path and the column is not affected by the sneak path or the row is not affected by the sneak path and the column is affected by the sneak path;

[0012] (4) Calculate the log-likelihood ratio and decision threshold of each type of memory cell:

[0013] Calculate the log-likelihood ratio of each type of memory cell through the conditional probability distribution of the read resistance value y m,n under different value conditions of x m,n , and calculate the decision threshold of each type of memory cell through the prior probability of different values of x m,n ;

[0014] (5) Obtain the ReRAM data decision result:

[0015] Perform a state decision on the resistance state of the data stored in the corresponding type of memory cell through the log-likelihood ratio and decision threshold of type A, B, and C memory cells to obtain the decision data

[0016] Compared with the prior art, the present invention has the following advantages:

[0017] The present invention classifies storage units according to the sneak path correlation values of the rows and columns where each storage unit is located, and calculates the log-likelihood ratio of each class of storage units through the conditional probability distribution of the read resistance values when the stored data of each storage unit takes different values. This method of calculating the log-likelihood ratio using the sneak path correlation fully considers the differences in interference suffered by different storage units, avoids the influence of the high bit error rate caused by calculating the likelihood ratio based on the assumed probability of the storage unit being interfered by the sneak path in the prior art on the decision result, and effectively improves the decision accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flowchart for the implementation of the present invention;

[0019] Figure 2 is a simulation comparison diagram of the decision accuracy between the present invention and the prior art. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0021] Referring to Figure 1 , the present invention includes the following steps:

[0022] Step 1) Initialize parameters:

[0023] Initialize a ReRAM including M×N storage units arranged periodically. The storage unit c at the m-th row and n-th column m,n stores binary data and the reference resistance value are x m,n , R m,n respectively, where M≥16, N≥16. In this embodiment, M = N = 32.

[0024] x m,n and R m,n are related as follows:

[0025] When x m,n = 0, R m,n = R0 = 1000Ω, indicating that c m,n is in the high resistance state. When x m,n = 1, R m,n = R1 = 100Ω, indicating that c m,n is in the low resistance state.

[0026] Step 2) Calculate the read resistance value of the ReRAM storage unit:

[0027] During the data reading process, a relatively small voltage is usually applied to the storage cell, and the resistance value of the cell is calculated by measuring the current flowing through the cell. However, during actual reading, the resistance value of the storage cell may be affected by two types of interference: sneak path interference and Gaussian noise interference. Sneak path interference refers to the occurrence of an additional current path during the measurement process, resulting in an increase in the measured current value, thereby making the read resistance value smaller than the reference resistance value. This interference can be equivalent to a resistor connected in parallel with the storage cell resistance.

[0028] According to the reference resistance value R of the storage cell c m,n calculate the read resistance value y of c m,n as follows: m,n m,n :

[0029]

[0030] where z m,n represents a Gaussian noise random variable, denotes that z m,n follows a Gaussian distribution with a mean of 0 and a variance of σ 2 σ represents the noise standard deviation of z m,n e m,n represents a sneak path interference indicator, e m,n = 1 indicates that c m,n is affected by sneak path interference, e m,n = 0 indicates that c m,n is not affected by sneak path interference, R0′ represents the resistance value of the high-resistance state cell after being affected by sneak path interference during the read resistance value process, R s = 250Ω represents the equivalent resistance value introduced by sneak path interference.

[0031] It can be seen from the calculation formula of the read resistance value y m,n that the low-resistance state cell, i.e., the cell storing data x m,n = 1, is only affected by Gaussian noise, while sneak path interference is only likely to occur in the high-resistance state cell, i.e., the cell storing data x m,n = 0.

[0032] Step 3) Classify ReRAM storage cells based on sneak path correlation:

[0033] ​The prior art usually estimates the proportion of storage cells in ReRAM affected by sneak path interference by making a hard decision on the read resistance value, and uses this proportion as the probability of sneak path interference. Then it is assumed that all storage cells in ReRAM are affected by sneak path interference with this probability. However, this average assumption based on the probability of sneak path occurrence will lead to unstable judgment results: for storage cells whose probability of sneak path interference is close to the average value, the data judgment result is relatively accurate; but for storage cells whose probability of sneak path interference is significantly higher or lower than the average value, the judgment result may have a large error. In fact, there are differences in the sneak path interference suffered by different storage cells. The present invention takes this difference into account. By calculating the sneak path correlation values ​​of the rows and columns where the storage cells are located, the M×N storage cells in ReRAM are divided into storage cells whose rows and columns are affected by the sneak path; storage cells whose rows and columns are not affected by the sneak path; storage cells whose rows are affected by the sneak path but whose columns are not affected by the sneak path, or whose rows are not affected by the sneak path but whose columns are affected by the sneak path. The specific steps are as follows:

[0034] Step 3a) Read the resistance value y m,n Computational storage unit c m,n The correlation value L of the latent path of the row and column m , L n , the calculation formulas are:

[0035]

[0036] in, Representation unit c m,n Read the resistance value y m,n The mixed probability density function, e (·) Represents an exponential function with base e. Read the resistance value y m,n There are three possible values ​​for R1+z m,n 、R0+z m,n 、R0′+z m,n , when the stealth path interference occurs, its mixed probability density function is The prior probabilities of the three values ​​are q=0.5, (1-q) 2 , q(1-q), and the mixed probability density function when there is no stealth path interference is: The prior probabilities of the three values ​​are q, 1-q, and 0 respectively.

[0037] The above expression shows that when L m >0、L n >0, the mth row and the nth column are the rows and columns affected by the sneak path, otherwise they are the rows and columns not affected by the sneak path.

[0038] Step 3b) Classify the M×N memory cells included in the ReRAM through L m and L n The specific classification method is as follows:

[0039] When L m > 0 and L n > 0, c m,n is class A memory cells affected by sneak paths. When reading the resistance value, this type of memory cells will be interfered by sneak paths with a high probability;

[0040] When L m < 0 and L n < 0, c m,n is class B memory cells not affected by sneak paths. When reading the resistance value, this type of memory cells will not be interfered by sneak paths;

[0041] When L m > 0 and L n < 0 or L m < 0 and L n > 0, c m,n is class C memory cells where the row is affected by sneak paths while the column is not affected by sneak paths or the row is not affected by sneak paths while the column is affected by sneak paths. When reading the resistance value, this type of memory cells will be interfered by sneak paths with a low probability.

[0042] Through the above classification method, the sneak path interference characteristics of different memory cells in the ReRAM can be more accurately identified, thus providing a more reliable basis for subsequent data decision-making.

[0043] Step 4) Calculate the log-likelihood ratio and decision threshold of each category of memory cells:

[0044] Since different categories of memory cells are affected by sneak paths to different degrees and their read resistance value distribution characteristics also vary, it is necessary to calculate the log-likelihood ratio and decision threshold for each category of memory cells separately to ensure the accuracy of data decision-making. When c m,n belongs to class A memory cells, its read resistance value y m,n has two values: R1 + z m,n , R0'+ z m,n , and the stored data corresponding to these two values are x m,n = 1, x m,n = 0, and their prior probabilities are q and 1 - q respectively; when c m,n belongs to class B memory cells, its read resistance value y m,n has two values: R1 + z m,n , R0 + z m,n , and the stored data corresponding to these two values are x m,n= 1, x m,n = 0, and their prior probabilities are q and 1 - q respectively; when c m,n belongs to the C - type storage unit, its read resistance value y m,n has two values: R1 + z m,n and R0'+ z m,n , and the stored data corresponding to these two values is x m,n = 1, x m,n = 0, and their prior probabilities are 1 - α and α = 0.0015 respectively;

[0045] Through x m,n Under different value conditions of the read resistance value y m,n Calculate the log - likelihood ratio of each category of storage unit through the conditional probability distribution, and at the same time calculate the decision threshold of each category of storage unit through the prior probabilities of different values of x m,n The specific steps are as follows:

[0046] Step 4a) Calculate the log - likelihood ratio of each category of storage unit. Among them, the log - likelihood ratio Λ1 of the A - type and C - type storage units, and the log - likelihood ratio Λ2 of the B - type storage units. The calculation formulas are as follows:

[0047]

[0048] Among them, represents the conditional probability distribution of the read resistance value y m,n under the condition of x m,n = 0, represents the conditional probability distribution of the read resistance value y m,n under the condition of x m,n = 1.

[0049] Step 4b) Calculate the decision threshold of each category of storage unit. Among them, the decision threshold η1 of the A - type and B - type storage units, and the decision threshold η2 of the C - type storage units. The calculation formulas are as follows:

[0050]

[0051] Among them, q represents the prior probability of x m,n = 1 for the A - type and B - type storage units, and α represents the prior probability of x m,n = 0 in the C - type storage unit.

[0052] Step 5) Obtain the ReRAM data decision result:

[0053] Perform state decision on the resistance states of the data stored in the corresponding category of storage units through the log - likelihood ratio and decision threshold of the A, B, and C - type storage units to obtain the decision data The specific decision method is:

[0054] Determine whether Λ1≥lnη1 holds. If so, obtain the decision data of type A storage units Otherwise,

[0055] Determine whether Λ2≥lnη1 holds. If so, obtain the decision data of type B storage units Otherwise,

[0056]

[0057] Determine whether Λ1≥lnη2 holds. If so, obtain the decision data of type C storage units Otherwise,

[0058]

[0059] Next, in combination with the simulation results, the technical effects of the present invention will be further described:

[0060] 1. Experimental conditions and content:

[0061] Using the simulation software Visual Studio, under the conditions of M = N = 32 and the probability of selector failure being 0.001, a comparative simulation of the decision accuracy between the present invention and the existing ReRAM data decision method based on probability estimation and log-likelihood ratio detection is carried out. The results are as Figure 2 shown. The calculation formula of the bit error rate BER in the figure is:

[0062]

[0063] where, represents the indicator function. When the condition holds, the value is 1, and when the condition does not hold, the value is 0.

[0064] 2. Analysis of experimental results:

[0065] Referring to Figure 2 , it can be seen from the figure that as the abscissa noise standard deviation σ decreases from 80 to 20, the bit error rate BER of the present invention decreases from 3.435e-02 to 7.585e-04, while the bit error rate BER of the existing technology decreases from 5.683e-02 to 1.350e-03. The bit error rate BER of the present invention is always lower than that of the existing technology, indicating that the present invention classifies storage units by utilizing sneak path correlation, calculates the log-likelihood ratio and decision threshold for different types of storage units respectively, fully considers the read resistance value distribution characteristics of different types of storage units, thereby avoiding the defects of the log-likelihood ratio decision method based on the assumed average sneak path interference probability in the existing technology and making the decision accuracy higher.

Claims

1. A method for determining data of a resistive random access memory (ReRAM) based on sneak path correlation, characterized in that: The steps include: (1) Initialization parameters: The initialization ReRAM includes M×N storage cells arranged periodically, and the storage cell c in the mth row and nth column m,n The stored binary data and reference resistance values ​​are x m,n , R m,n , where M≥16, N≥16; (2) Calculate the read resistance value of the ReRAM memory cell: According to the storage unit c m,n The reference resistance value R m,n Calculate c m,n The read resistance value y m,n ; (3) Classification of ReRAM memory cells based on sneak path correlation: By reading the resistance value y m,n Computational storage unit c m,n The correlation value L of the latent path of the row and column m , L n , and through L m and L n Classifying the M×N storage cells included in the ReRAM to obtain A-type storage cells whose rows and columns are both affected by the sneak path, B-type storage cells whose rows and columns are not affected by the sneak path, and C-type storage cells whose rows are affected by the sneak path but whose columns are not affected by the sneak path, or whose rows are not affected by the sneak path but whose columns are affected by the sneak path; (4) Calculate the log-likelihood ratio and decision threshold of each category storage unit: By x m,n Read the resistance value y under different value conditions m,n The conditional probability distribution of the log-likelihood ratio of each category storage unit is calculated, and x m,n The decision threshold of each category storage unit is calculated based on the prior probability of different values; (5) Obtaining the ReRAM data judgment result: The resistance state of the data stored in the corresponding storage unit is judged by the log-likelihood ratio and decision threshold of the A, B, and C storage units to obtain the decision data.

2. The method according to claim 1, characterized in that The storage unit c in the mth row and nth column described in step (2) m,n , which stores binary data x m,n With the reference resistance R m,n The relationship is: When x m,n = 0 when R m,n =R0, indicating c m,n In high impedance state, when x m,n =1 when R m,n =R1, indicating c m,n In low impedance state.

3. The method according to claim 2, characterized in that c described in step (2) m,n The read resistance value y m,n , the calculation formula is: Among them, z m,n represents a Gaussian noise random variable, The mean is 0 and the variance is σ 2 Gaussian distribution, σ represents z m,n The standard deviation of the noise, e m,n represents the sneak path interference indicator, R0′ represents the resistance value of the high-impedance unit affected by the sneak path interference, R s Indicates the equivalent resistance value introduced by sneak path interference.

4. The method according to claim 3, characterized in that c described in step (3) m,n The correlation value L of the latent path of the row and column m , L n , the calculation formulas are: in, Representation unit c m,n Read the mixed probability density function of the resistance value, q represents the data x m,n =1, e( · ) represents an exponential function with base e.

5. The method according to claim 4, characterized in that The M×N storage cells included in the ReRAM described in step (3) are divided into three categories, and the specific classification method is: When L m >0 and L n >0, c m,n It is a Class A storage unit affected by the stealth path; When L m <0 and L n When <0, c m,n It is a Class B storage unit that is not affected by the sneak path; When L m >0 and L n <0 or L m <0 and L n >0, c m,n The C-type storage unit is a storage unit in which the row is affected by the sneak path but the column is not affected by the sneak path, or the row is not affected by the sneak path but the column is affected by the sneak path.

6. The method according to claim 4, characterized in that The log-likelihood ratios of each category of storage units described in step (4), wherein the log-likelihood ratios Λ1 of category A and category C storage units, and the log-likelihood ratios Λ2 of category B storage units are calculated by: in, Respectively represent x m,n =0, x m,n =1, read the resistance value y m,n The conditional probability distribution of .

7. The method according to claim 6, characterized in that The decision thresholds of each category of storage units described in step (4), wherein the decision thresholds η1 of category A and category B storage units, and the decision threshold η2 of category C storage units are calculated by: Among them, α represents the x in the C-type storage unit m,n =0 prior probability.

8. The method according to claim 7, characterized in that The specific method of obtaining the ReRAM data judgment result described in step (5) is as follows: Determine whether Λ1≥lnη1 is true. If so, get the judgment data of the A-type storage unit. otherwise, Determine whether Λ2≥lnη1 is true. If so, get the judgment data of the B-type storage unit otherwise, Determine whether Λ1≥lnη2 is true. If so, get the judgment data of the C-type storage unit otherwise,

Images