Method of operating a memristor array, data processing apparatus

The memristor array designed with a 2T2R structure and buffer circuit solves the problems of calculation accuracy and external circuit design difficulty caused by voltage drop in traditional memristor arrays, and achieves higher energy efficiency and computing power.

CN115862708BActive Publication Date: 2026-11-10TSINGHUA UNIVERSITY
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202211477580.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-11-23
Publication Date
2026-11-10
Estimated Expiration
2042-11-23

AI Technical Summary

Technical Problem

Traditional memristor arrays suffer from severe voltage drop (IR drop) during computation, which limits computational accuracy, increases the difficulty of external circuit design, and makes system performance a bottleneck.

Method used

The memristor unit adopts a 2T2R structure, and by operating the transistor in the saturation region and combining it with a buffer circuit design, voltage drop is avoided and calculation accuracy is improved.

Benefits of technology

It mitigates the impact of voltage drop on computational accuracy, reduces the design difficulty and power consumption of external circuits, and achieves higher energy efficiency and computing power.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115862708B_ABST
    Figure CN115862708B_ABST
Patent Text Reader

Abstract

A method for operating a memristor array and a data processing apparatus. The method for operating the memristor array includes a matrix-vector multiplication operation, which includes: applying input voltages corresponding to input data of each row of the memristor array to word lines of each row of the memristor array; applying source line voltages to source lines of each column of the memristor array, so that transistors in each memristor cell of each row of the memristor array are turned on and work in a saturation region; applying a first fixed voltage to bit lines of each row of the memristor array; and detecting output currents on the source lines of each column of the memristor array to obtain an output result. The method can alleviate the influence of voltage drop caused by large current in the memristor array on calculation accuracy, and reduce the design difficulty of external circuits.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] Embodiments of this disclosure relate to a method of operating a memristor array and a data processing apparatus. Background Technology

[0002] Memristor-based in-memory computing technology promises to break through the bottlenecks of the traditional von Neumann architecture, achieving device-level storage and computation integration, avoiding data transfer processes, and thus significantly saving latency and energy consumption. Therefore, it is considered the core of next-generation intelligent systems. However, each memristor unit in a memristor array adopts a 1T1R (1 transistor 1 memristor) basic unit structure, where the memristor is connected in series with one terminal of the transistor (e.g., the drain). For example, during forward computation, a control voltage is applied to the gate of the transistor to fully turn it on (conduct), placing it in the linear region, where the transistor is equivalent to a linear resistor. In operation, the entire memristor array is equivalent to a resistor network, in which there is a significant voltage drop (IR drop). Therefore, when using current-sensing quantization output, the above operation method places very high demands on the accuracy of external drivers and quantization circuits (especially clamping circuits). This ultimately limits the computational accuracy of traditional memristor-based in-memory computing, increases the difficulty of external circuit design, and makes its area, power consumption, and speed bottlenecks in system performance.

[0003] Therefore, based on the 1T1R architecture, a basic structure of 2T2R (2 transistors, 2 memristors) memristor cell has been proposed. This structure reduces accumulated current on the traces through local current differential calculation, alleviates voltage drop issues, and improves calculation accuracy. Other approaches have proposed using voltage-domain or charge-domain quantization sensing, but these methods may sacrifice speed and accuracy, and the size of the memristor array remains limited. Therefore, it is essential to propose and validate novel array cells to meet the demands of high computing power, increasing array size, and simplifying peripheral circuitry. Summary of the Invention

[0004] At least one embodiment of this disclosure provides an operation method for a memristor array, wherein the memristor array includes multiple rows and columns of memristor cells, word lines corresponding to each row of the memristor array, bit lines corresponding to each row of the memristor array, and source lines corresponding to each column of the memristor array. Each memristor cell includes a transistor and a memristor. The gate of the transistor is connected to the word line corresponding to the row where the memristor cell is located. The first terminal of the transistor is connected to the source line corresponding to the column where the memristor cell is located. The second terminal of the transistor is connected to the first terminal of the memristor. The second terminal of the memristor is connected to the bit line corresponding to the row where the memristor cell is located. The operation method includes a matrix-vector multiplication operation, which includes: applying the input voltage corresponding to the input data of each row in the memristor array to the word line of each row in the memristor array; applying the source line voltage to the source line of each column in the memristor array, so that the transistors in each memristor cell of each row in the memristor array are turned on and operate in the saturation region; applying a first fixed voltage to the bit line of each row in the memristor array; and detecting the output current on the source line of each column in the memristor array to obtain the output result.

[0005] At least one embodiment of this disclosure provides a data processing apparatus, including: a word line driver, a source line driver, a bit line driver, an output processing circuit, and a memristor array. The memristor array includes multiple rows and columns of memristor cells, word lines corresponding to each row of the memristor array, bit lines corresponding to each row of the memristor array, and source lines corresponding to each column of the memristor array. Each memristor cell includes a transistor and a memristor. The gate of the transistor is connected to the word line corresponding to the row of the memristor cell. The first terminal of the transistor is connected to the source line corresponding to the column of the memristor cell. The second terminal of the transistor is connected to the first terminal of the memristor. The second terminal of the memristor is connected to the bit line corresponding to the row of the memristor cell. The word lines of each row are coupled to the word line driver, the source lines of each column are coupled to the source line processing circuit, and the bit lines of each row are coupled to the bit line processing circuit. The word line driver is configured to apply the input voltage corresponding to the input data of each row in the memristor array to the word line of each row in the memristor array; the source line driver is configured to apply the source line voltage to the source line of each column in the memristor array, so that the transistors in each memristor cell of each row in the memristor array are turned on and operate in the saturation region; the bit line driver is configured to apply a first fixed voltage to the bit line of each row in the memristor array; and the output processing circuit is configured to detect the output current on the source line of each column in the memristor array to obtain the output result. Attached Figure Description

[0006] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments will be briefly described below. Obviously, the drawings described below only relate to some embodiments of this disclosure and are not intended to limit this disclosure.

[0007] Figure 1A A schematic diagram of a memristor unit with a 1T1R structure is shown.

[0008] Figure 1B A schematic diagram of a memristor cell with a 2T2R structure is shown.

[0009] Figure 2 A schematic diagram of an exemplary memristor array structure is shown.

[0010] Figure 3 This illustrates the operation method of matrix-vector multiplication in a traditional memristor array.

[0011] Figure 4 A schematic diagram of a buffered memristor cell according to at least one embodiment of the present disclosure is shown.

[0012] Figure 5 The method of applying the operating voltage to a memristor cell of a 1T1R structure according to at least one embodiment of the present disclosure is shown.

[0013] Figure 6 The method of applying the operating voltage to a memristor cell of a 2T2R structure according to at least one embodiment of the present disclosure is shown.

[0014] Figure 7 The operation of matrix-vector multiplication is shown in an example memristor array according to this disclosure.

[0015] Figure 8 A schematic diagram of a memristor array with global bit lines according to an example of this disclosure is shown.

[0016] Figure 9 The operation method of setting up or initializing a memristor array during programming according to an example memristor array is shown.

[0017] Figure 10 The operation mode during the reset operation in the programming operation of an example memristor array according to this disclosure is shown.

[0018] Figure 11 A schematic diagram of a processing apparatus according to at least one embodiment of the present disclosure is shown. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the described embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0020] Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms “first,” “second,” and similar terms used in this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, terms such as “comprising” or “including” mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as “connected” or “linked” are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as “upper,” “lower,” “left,” and “right” are used only to indicate relative positional relationships, which may change accordingly when the absolute position of the described objects changes.

[0021] Memristors (e.g., resistive random access memory, phase-change memory, bridged memory, etc.) are non-volatile devices whose conductance can be adjusted by applying external stimuli. As a two-terminal device, memristors possess adjustable and non-volatile resistance, making them widely used in in-memory computing applications, such as for forward computation (inference) in artificial neural networks. According to Kirchhoff's current law and Ohm's law, arrays of memristors can be used to perform matrix-vector multiplication (multiply-accumulate) calculations in parallel, with both storage and computation occurring within the array's individual devices. Based on this computing architecture, in-memory computing can be achieved without requiring extensive data movement.

[0022] Figure 1A A schematic diagram of a memristor cell with a 1T1R structure is shown. (For example...) Figure 1A As shown, the memristor unit of the 1T1R structure includes a transistor M1 and a memristor R1.

[0023] For example, when transistor M1 is an N-type transistor, its gate is connected to the word line WL. For example, transistor M1 is turned on when the word line WL is high. The first terminal of transistor M1 can be the source and is configured to be connected to the source line SL. The second terminal of transistor M1 can be the drain and is configured to be connected to the second terminal of memristor R1. The first terminal of memristor R1 is connected to the bit line BL.

[0024] For example, when transistor M1 is a P-type transistor, its gate is connected to the word line WL. For example, transistor M1 is turned on when the word line WL inputs a low level. The first terminal of transistor M1 can be the drain and is configured to be connected to the source line SL. The second terminal of transistor M1 can be the source and is configured to be connected to the second terminal of memristor R1. The first terminal of memristor R1 is connected to the bit line BL.

[0025] It should be noted that the resistive random access memory structure can also be implemented in other ways, such as a structure in which the first terminal of memristor R1 is connected to a transistor and the second terminal is connected to the source line SL. The embodiments disclosed herein do not limit this.

[0026] The word line WL applies a voltage to the gate of transistor M1, thereby controlling its on or off state. When operating memristor R1, such as performing a set or reset operation, transistor M1 must first be turned on; that is, a turn-on voltage must be applied to the gate of transistor M1 via the word line WL. After transistor M1 is turned on, the resistance state of memristor R1 can be changed, for example, by applying voltages to the source line SL and the bit line BL. For example, a set voltage can be applied via the bit line BL to put memristor R1 in a low-resistance state; or a reset voltage can be applied via the source line SL to put memristor R1 in a high-resistance state. For example, the resistance value in the high-resistance state is more than 100 times, or even more than 1000 times, the resistance value in the low-resistance state.

[0027] In the embodiments of this disclosure, the operation of changing the memristor R1 from a high-resistance state to a low-resistance state is called a set operation; the operation of changing the memristor R1 from a low-resistance state to a high-resistance state is called a reset operation. Furthermore, the initialization process of the memristor refers to applying a high voltage to the memristor, thereby inducing the formation of conductive filaments inside the memristor. By applying a set voltage or a reset voltage between the electrodes of the memristor, the conductive filaments inside the memristor can be restored or broken. The initialization operation typically only needs to be performed once during the lifetime of the resistive random access memory.

[0028] Memristor R1 has a threshold voltage. When the input voltage amplitude is less than the threshold voltage of memristor R1, the resistance (or conductance) value of memristor R1 will not change. In this case, calculations can be performed using the resistance (or conductance) value of memristor R1 by inputting a voltage less than the threshold voltage; the resistance (or conductance) value of memristor R1 can be changed by inputting a voltage greater than the threshold voltage.

[0029] Figure 1B A schematic diagram of a memristor cell with a 2T2R structure is shown. (For example...) Figure 1BAs shown, the 2T2R memristor unit includes two transistors M1 and M2 and two memristors R1 and R2. The following explanation assumes that both transistors M1 and M2 are N-type transistors.

[0030] The gate of transistor M1 is connected to word line WL1. For example, transistor M1 is turned on when word line WL1 is high. The gate of transistor M2 is connected to word line WL2. For example, transistor M2 is turned on when word line WL2 is high. The first terminal of transistor M1 can be the source and is configured to be connected to the source line SL. The first terminal of transistor M2 can also be the source and is configured to be connected to the source line SL. The first terminals of transistors M1 and M2 are connected together to the source line SL. The second terminal of transistor M1 can be the drain and is configured to be connected to the second terminal (e.g., the negative terminal) of memristor R1. The first terminal (e.g., the positive terminal) of memristor R1 is connected to bit line BL1. The second terminal of transistor M2 can be the drain and is configured to be connected to the second terminal (e.g., the negative terminal) of memristor R2. The first terminal (e.g., the positive terminal) of memristor R2 is connected to bit line BL2.

[0031] It should be noted that the transistors M1 and M2 in the 2T2R structure memristor unit can both be P-type transistors, or one can be a P-type transistor and the other an N-type transistor. Different types of transistors have different control voltages, so the control voltage needs to be changed accordingly when applying it, which will not be elaborated here.

[0032] Figure 2 A schematic diagram of an exemplary memristor array structure is shown. Figure 2 As shown, the memristor array consists of multiple memristor cells forming an M-row, N-column array, where M and N are both positive integers (e.g., greater than 2, and can be 100 or more). Each memristor cell includes a switching element and one or more memristors. Figure 2 In the middle, WL <1> WL <2> ...WL <m>These represent the word lines for the first, second, ..., Mth rows, respectively. The control electrode (e.g., the gate of a transistor) of the switching element in the memristor cell circuit of each row is connected to the corresponding word line of that row; BL <1> BL <2> BL <m>These represent the bit lines of the first row, second row, ..., Mth row, respectively. The memristor in the memristor cell circuit of each row is connected to the corresponding bit line of that row; SL <1> SL <2> ……SL <n>These represent the source lines of the first column, the second column, ..., the Nth column, respectively. The source of the transistor in the memristor unit circuit of each column is connected to the source line corresponding to that column.

[0033] Figure 2 The memristor cells in a memristor array can be, for example, a 1T1R structure or a 2T2R structure. For a 2T2R structure, if the two transistors are of the same type, their gates can be connected to the same word line, while the two memristors are connected to two different bit lines; that is, each row of the array has one word line and two bit lines. If the two transistors are of different types, their gates can be connected to two different word lines, while the two memristors are connected to two different bit lines; that is, each row of the array has two word lines and two bit lines.

[0034] The embodiments of this disclosure do not limit the type or structure of the memristor device. It should be noted that the transistors used in the embodiments of this disclosure can be thin-film transistors or field-effect transistors (e.g., MOS transistors). The source and drain of the transistors used can be structurally symmetrical, so their source and drain can be structurally indistinguishable.

[0035] For example, the weight matrix of a neural network layer can be mapped to the conductance matrix of a memristor array, the data to be processed is mapped to the input voltage of the array word lines, and the processing result is represented by the output current of the array source lines. By inputting a voltage difference less than the threshold voltage of each memristor in the memristor array, forward computation (e.g., matrix-vector multiplication) can be performed using the conductance of the memristors.

[0036] Figure 3 This illustrates the operation of matrix-vector multiplication in a traditional memristor array. For example... Figure 3 As shown, memristor cells in the same row with a 1T1R structure share the word line WL and the bit line BL, while memristor cells in the same column share the source line SL. To perform matrix-vector multiplication, firstly, the values ​​of the matrix elements (e.g., weights in a neural network) are mapped to the resistance values ​​of the corresponding memristors in the memristor array. Secondly, each data element of the vector, after digital-to-analog conversion, is applied as an input voltage to each bit line. A transistor turn-on voltage is applied to each word line, causing all transistors to conduct simultaneously and operate in the linear region. A fixed voltage (e.g., ground voltage) is applied to each source line.

[0037] Therefore, a voltage difference (i.e., the input voltage applied to the corresponding row's shared bit line BL minus the fixed voltage applied to the corresponding source line SL) is applied to each memristor in the memristor array, thereby generating current in the memristors. The current generated by a row of memristor cells converges on the corresponding source line SL as the output current. This matrix-vector multiplication calculation is performed as follows. According to Kirchhoff's laws, the output current of the memristor array can be obtained using the following formula:

[0038]

[0039] Among them, V i G is the input voltage on bit line BL in the i-th row of the memristor array. ij Let I be the conductance value of the memristor in the i-th row and j-th column of the memristor array. j V is the output current on the j-th column source line SL of the memristor array. sl A fixed voltage is applied to the source line SL of the memristor array during the calculation.

[0040] In the above operation, during the forward calculation of the 1T1R memristor cell, a gate voltage is applied to the transistor's gate to make the transistor operate in the linear region, thus making the transistor's on-resistance negligible relative to the memristor's resistance. At this time, a voltage is applied to both the source line SL and the bit line BL, creating a voltage difference between them. This voltage difference is directly applied to the memristor, generating current output on the source line SL. During this forward calculation, both the driver providing the source line voltage and the driver providing the bit line voltage perceive a low-impedance load.

[0041] As mentioned above, current computing methods for memristor arrays used to achieve in-memory computing inevitably suffer from significant voltage drop (IR drop) effects, leading to accuracy issues and huge external circuit overhead due to limitations in drive capability and accuracy.

[0042] To address the aforementioned problems, at least one embodiment of this disclosure provides an operation method for a memristor array and a corresponding data processing apparatus. The memristor array includes multiple rows and columns of memristor cells, word lines corresponding to each row of the memristor array, bit lines corresponding to each row of the memristor array, and source lines corresponding to each column of the memristor array. Each memristor cell includes a transistor and a memristor. The gate of the transistor is connected to the word line corresponding to the row where the memristor cell is located. The first terminal of the transistor is connected to the source line corresponding to the column where the memristor cell is located. The second terminal of the transistor is connected to the first terminal of the memristor. The second terminal of the memristor is connected to the bit line corresponding to the row where the memristor cell is located.

[0043] For example, the operation method includes performing matrix-vector multiplication operations using a memristor array, or programming operations on the memristors of the memristor cells in the memristor array, the programming operations including setting operations or reset operations.

[0044] Figure 4 A schematic diagram of a novel buffered memristor cell according to at least one embodiment of the present disclosure is shown. This memristor cell can be used for in-memory computing operations, such as matrix-vector multiplication. Figure 1A The 1T1R unit shown differs from the method of directly applying an operating voltage across the memristor, such as... Figure 4 As shown, in at least one embodiment of this disclosure, the input voltage is applied to the memristor through a buffer circuit, and the current flowing through the memristor is also output through a buffer circuit. The input buffer circuit is used to convert the low input impedance of the memristor unit to a high input impedance, and the output buffer circuit is used to suppress the effect of voltage changes at the current output terminal on the current.

[0045] In at least one example, the memristor cell of this disclosure embodiment is structurally similar to... Figure 1A The 1T1R structure memristor cells shown are identical, each including a memristor and a transistor (e.g., a MOS transistor). However, in various embodiments of this disclosure, the method of applying the operating voltage to the memristor cell of the 1T1R structure is different from that shown. Figure 3 The current memristor cells shown are applied in different ways. The embodiments of this disclosure are based on putting the transistors in different working states and using different working mechanisms to avoid problems such as voltage drop and accuracy reduction caused by the current way of applying the operating voltage.

[0046] Figure 5 The method of applying the operating voltage to a memristor cell of a 1T1R structure according to at least one embodiment of the present disclosure is shown. For example... Figure 5 As shown, in the operating voltage application method of a 1T1R memristor cell, by controlling the control voltage (word line voltage) applied to the transistor gate and the source line voltage applied to the source line SL within a certain range, and by grounding the bit line voltage BL, the transistor in the memristor cell operates in the saturation region. A 1T1R memristor cell with the transistor operating in the saturation region can be called a source follower cell. In the source follower cell, the transistor gate serves as the voltage input terminal. The input voltage is buffered by the source follower structure before being applied to the memristor, and the current through the memristor flows through the transistor to the source line. In this process, the transistor in the source follower cell acts as an input voltage buffer circuit, ensuring the input voltage is applied to the transistor gate and then buffered to the memristor, thereby increasing the impedance at the voltage input terminal. On the other hand, the transistor also acts as an output current buffer, operating in the saturation region to suppress the influence of source line voltage changes on the output current.

[0047] for Figure 5 With the operating voltage applied as shown, the transistor in the source follower unit operates in the saturation region during the forward calculation. At this time, the current through the source follower unit basically satisfies the following formula:

[0048] I = f(V) * g(G)

[0049] f(V)=a(V+b)

[0050] g(G) = G + c

[0051] Where V is the gate voltage of the source follower cell, G is the conductance of the memristor in the source follower cell, a is the voltage transfer coefficient of the source follower, and b and c are coefficients related to the selected gate voltage and memristor conductance range, which are constants in the calculation process. These coefficients a, b, and c can be obtained through measurement or experience. As shown in the above equation, the current of the source follower cell exhibits a linear relationship with the gate voltage and conductance of the memristor during the forward calculation.

[0052] In another example of this disclosure, a 2T2R memristor cell (source follower cell) is constructed based on the above-mentioned 1T1R structure memristor cell (i.e., the above-mentioned source follower cell).

[0053] Figure 6 The method of applying the operating voltage to a memristor cell with a 2T2R structure according to at least one embodiment of the present disclosure is shown. For example... Figure 6 As shown, an exemplary 2T2R memristor cell includes one NMOS transistor and one PMOS transistor. The first terminals (e.g., drains) of these two transistors are electrically connected to each other and to their respective source lines SL. The gates of the two transistors are connected to two different word lines. The first terminals of the two memristors are electrically connected to the second terminals (e.g., sources) of the two transistors, and the second terminals of the two memristors are connected to two different bit lines. This allows the two word line voltages required to turn on the two transistors to be different from each other. The current through the NMOS transistor and the current through the PMOS transistor are differentiated and then converged on the source line SL, which can be used to implement positive and negative weights and can also reduce the operating current on the source line SL accordingly.

[0054] The following description uses a 1T1R memristor cell as an example, wherein the transistors in the memristor cell are NMOS transistors. The embodiments disclosed herein are not limited to this case; the specific operation can be adjusted accordingly based on the type and number of transistors.

[0055] In the embodiments of this disclosure, the operation method for the memristor array includes a matrix-vector multiplication operation, which includes: applying the input voltage corresponding to the input data of each row in the memristor array to the word line of each row in the memristor array; applying the source line voltage to the source line of each column in the memristor array, so that the transistors in each memristor cell of each row in the memristor array are turned on and operate in the saturation region; applying a first fixed voltage to the bit line of each row in the memristor array; and detecting the output current on the source line of each column in the memristor array to obtain the output result.

[0056] Figure 7 This illustration shows the operation of matrix-vector multiplication in an example memristor array according to this disclosure. In this example, in a 1T1R structure memristor array, memristor cells in the same row share a word line WL and a bit line BL, while memristor cells in the same column share a source line SL. For matrix-vector multiplication, firstly, the matrix cell values ​​(weight values) are mapped to the resistance values ​​of the corresponding memristors in the memristor array; secondly, each data element of the vector, after digital-to-analog conversion, is applied as an input voltage to each word line, causing the transistors to conduct. A source line voltage is applied to each source line to make the transistors operate in the saturation region, and a fixed voltage (an example of a first fixed voltage, such as ground voltage) is applied to each bit line. Current flows through the memristors, and the current generated by a column of memristor cells converges on the corresponding source line SL as the output current.

[0057]

[0058] f(V i )=a(V i +b)

[0059] g(G ij ) = G ij +c

[0060] Among them, V i G is the word line voltage (input voltage) applied to the word line of the i-th row in the memristor array. ij Let I be the conductance value of the memristor in the i-th row and j-th column of the memristor array. j Let be the output current on the source line SL of the j-th column of the memristor array. In this calculation, the source line voltage applied to each source line SL can be the same, and the voltage magnitude should be chosen such that each crystal operates in the saturation region.

[0061] In the above calculation process, since each word line WL is connected to the gate of the transistor in each memristor cell within that row, no current flows through the word line WL during the forward calculation. Therefore, no voltage drop (IR drop) occurs between different positions on the word line WL, and the word line voltage of each cell does not deviate from the voltage provided by the external word line driver. Furthermore, for the word line driver design, the word line, as a voltage input terminal, only provides a capacitive load (high impedance load) to the word line driver, thus significantly reducing the power consumption and area of ​​the word line driver design.

[0062] Furthermore, the current from each column of memristor cells converges on the corresponding source line SL as the output current. Therefore, due to the relatively large current and relatively long trace on the source line SL, a voltage drop exists between different locations. However, the impact of this voltage drop on the output current is greatly reduced because the transistors operate in the saturation region. Similarly, since all transistors in the memristor cells operate in the saturation region, the output current of each column is minimally affected by fluctuations in the source line voltage applied to the source line SL. Precise clamping of the source line voltage is not required, which reduces the area and power consumption of the external clamping circuitry.

[0063] In another example, to optimize memristor array operation, the memristor array can also short all bit lines BL to form one or more global bit lines GBL, which can further reduce the voltage drop on the bit lines BL.

[0064] Figure 8 A schematic diagram of a memristor array with global bit lines according to an example of this disclosure is shown. This memristor array structure can also... Figure 7 The matrix-vector multiplication is performed in the manner shown (as well as the programming operations described later).

[0065] For example, the matrix-vector multiplication operation described above also includes performing digital-to-analog conversion on the input data of each row in the memristor array to obtain the corresponding input voltage.

[0066] For example, in the matrix-vector multiplication described above, the output current of each column in the memristor array is converted from analog to digital to obtain the output result in digital form.

[0067] As described above, in the embodiments of this disclosure, in order to perform matrix-vector multiplication calculations using a memristor array, it is necessary to first map the values ​​of the matrix elements (e.g., the weight values ​​of a neural network) to the resistance values ​​of the memristors in the memristor array, that is, to program the memristor array.

[0068] By inputting an operating voltage greater than the threshold voltage of each memristor in the memristor array, the conductance value of the memristor can be changed so that the conductance value of the memristor corresponds to the weight value to be mapped. This operation is called programming. Programming operations include setting operations and reset operations.

[0069] For example, the setting operation includes: applying a first control voltage to the word line of the row containing the memristor cell to be set so that the transistor of the memristor cell to be set is turned on, and applying a second control voltage to other word lines in the memristor array so that the transistors of the memristor cells in other rows are turned off; applying a corresponding setting voltage to the bit line of the row containing the memristor cell to be set, and applying a first fixed voltage to other bit lines of the memristor array; applying a first fixed voltage to the source line of the column containing the memristor cell to be set, and applying setting voltage to other source lines of the memristor array.

[0070] For example, the second control voltage is the same as the first operating voltage, for example, both are ground voltage (GND).

[0071] Figure 9 The present disclosure illustrates the operation of a set or forming operation during programming a memristor array according to an example. In the memristor array of the 1T1R structure memristor cells in this example, the 1T1R structure memristor cells in the same row share the word line WL and the bit line BL, and the memristor cells in the same column share the source line SL.

[0072] like Figure 9 As shown, during the setup / initialization operation, a high level is applied to the word line WL corresponding to the row containing the selected memristor cell to be set / initialized, turning on the transistor connected to that word line. The source line SL of the column containing the selected memristor cell is set to a fixed voltage (low level, e.g., ground). Simultaneously, the bit line BL or global bit line of the row containing that memristor cell is set to the required operating voltage, which is a high level, for example, the value corresponding to the setup or initialization voltage required for the setup / initialization operation. This allows the setup / initialization operation for that memristor cell to be performed. Additionally, the source lines SL of the other unselected memristor cells in the same row as the selected memristor cell are also set to the same high level as the bit line BL (thus, the voltage difference across the memristors in these memristor cells is 0), and the word lines WL of other rows in the memristor array are set to a low level, turning off the transistors connected to these word lines. This ensures that only the selected memristor cell performs the setup / initialization operation, while the other memristor cells do not.

[0073] For example, the reset operation includes: applying a first control voltage to the word line of the row containing the memristor cell to turn on the transistor of the memristor cell, and applying a second control voltage to the other word lines of the memristor array to turn off the transistors of the memristor cells in other rows; applying a first fixed voltage to the bit lines of each row in the memristor array; applying a reset voltage to the source line of the column containing the memristor cell; and applying a first operating voltage to the other source lines of the memristor array.

[0074] For example, the second control voltage is the same as the first operating voltage, for example, both are ground voltage (GND).

[0075] Figure 10 This illustration shows the operation during a reset operation in a programming operation of an example memristor array according to this disclosure. The memristor array corresponds to... Figure 8 The memristor array shown.

[0076] like Figure 10 As shown, during a reset operation, the word line WL corresponding to the row containing the selected memristor cell to be reset is set to a high level to turn on the transistor connected to that word line. The source line SL of the column containing the selected memristor cell to be reset is set to the reset voltage (high level). Simultaneously, the bit line or global bit line of the row containing that memristor cell is set to a fixed voltage (low level, e.g., ground). This achieves the reset operation for that memristor cell. Furthermore, the source lines SL of the columns containing other unselected memristors in the row containing the selected memristor cell to the same low level as the bit line BL (e.g., ground) (thus the voltage difference across the memristors in these memristors is 0), and the word lines WL of other rows are set to a low level to turn off the transistors connected to these word lines. This ensures that only the selected memristor cell is reset, while other memristors are not reset.

[0077] Embodiments of this disclosure also provide a data processing apparatus based on the memristor and the corresponding operating method described above.

[0078] Figure 11 A schematic diagram of a data processing apparatus provided according to at least one embodiment of the present disclosure is shown. Figure 11 As shown, the data processing device includes word line drivers, source line drivers, bit line drivers, output processing circuitry, and a memristor array.

[0079] The memristor array can be, for example, a... Figures 7-8 The memristor array is shown. The word line driver is configured to apply the input voltage corresponding to the input data of each row in the memristor array to the word line of each row in the memristor array; the source line driver is configured to apply the source line voltage to the source line of each column in the memristor array, so that the transistors in each memristor cell of each row in the memristor array are turned on and operate in the saturation region; the bit line driver is configured to apply a first fixed voltage to the bit line corresponding to each row in the memristor array; the output processing circuit is configured to detect the output current on the source line of each column in the memristor array to obtain the output result.

[0080] For example, the data processing device further includes an input circuit configured to perform digital-to-analog conversion (DAC) on the input data of each row in the memristor array to obtain a corresponding input voltage, and to provide the corresponding input voltage to the word line driver.

[0081] For example, in the data processing device described above, the output processing circuit is further configured to receive the output current of each column in the memristor array and perform analog-to-digital conversion (ADC) on the output current to obtain a digital output result.

[0082] The aforementioned drivers, processing circuits, and input circuits can be implemented using digital circuits, analog circuits, or any combination thereof.

[0083] As described above, each memristor cell includes a single transistor and a single memristor, and the transistor and memristor are connected to each other and to corresponding word lines, bit lines and source lines as described above; or, each memristor cell includes two transistors and two memristors, the two transistors and two memristors are respectively connected to each other as described above, and the first terminals of the two transistors are electrically connected to each other and to the source line corresponding to the row where the memristor cell is located, the gates of the two transistors are respectively connected to two different word lines corresponding to the row where the memristor cell is located, and the second terminals of the two memristors are electrically connected to two different bit lines corresponding to the row where the memristor cell is located.

[0084] The novel in-memory computing structure based on memristor arrays proposed in at least one embodiment of this disclosure, namely the memristor in-memory computing circuit structure based on buffered cell structure, can alleviate the impact of voltage drop (IRdrop) caused by large current in memristor array on computing accuracy and reduce the design difficulty of external circuits. It eliminates the static current load of the driver during computing and greatly reduces the clamping accuracy requirements of the quantization module, enabling the memristor in-memory computing system to achieve higher energy efficiency and computing power.

[0085] For example, at least one embodiment of this disclosure also provides an electronic device that includes the aforementioned data processing device. For example, the electronic device can be any device such as a mobile phone, tablet computer, laptop computer, e-book, game console, television, digital photo frame, navigator, home appliance, communication base station, industrial controller, server, etc., or it can be any combination of data processing device and hardware. The embodiments of this disclosure do not limit this.

[0086] The following points should be noted regarding this disclosure:

[0087] (1) The accompanying drawings of the embodiments of this disclosure only involve the structures involved in the embodiments of this disclosure. Other structures can be referred to the general design.

[0088] (2) Where there is no conflict, the embodiments of this disclosure and the features in the embodiments can be combined with each other to obtain new embodiments.

[0089] The above are merely exemplary embodiments of this disclosure and are not intended to limit the scope of protection of this disclosure, which is determined by the appended claims.< / n> < / m> < / m>

Claims

1. A method of operating a memristor array, wherein, The memristor array includes multiple rows and columns of memristor cells, word lines corresponding to each row of the memristor array, bit lines corresponding to each row of the memristor array, and source lines corresponding to each column of the memristor array. Each memristor cell includes a transistor and a memristor. The gate of the transistor is connected to the word line corresponding to the row of the memristor cell, the first terminal of the transistor is connected to the source line corresponding to the column of the memristor cell, the second terminal of the transistor is connected to the first terminal of the memristor, and the second terminal of the memristor is connected to the bit line corresponding to the row of the memristor cell. The operation method includes matrix-vector multiplication, wherein the matrix-vector multiplication includes: The input voltage corresponding to the input data of each row in the memristor array is applied to the gate of the transistor corresponding to the word line of each row in the memristor array. A source line voltage is applied to the source line of each column in the memristor array so that the transistors in each memristor cell of each row in the memristor array are turned on and operate in the saturation region. A first fixed voltage is applied to the bit lines of each row in the memristor array; The output current on the source lines of each column in the memristor array is detected to obtain the output result.

2. The operating method according to claim 1, wherein, Applying the first fixed voltage to the bit lines of each row in the memristor array includes: The bit lines of each row in the memristor array are shorted to each other to apply the first fixed voltage.

3. The operating method according to claim 1, wherein, The matrix-vector multiplication operation also includes: The input data of each row in the memristor array is converted from digital to analog to obtain the corresponding input voltage.

4. The operating method according to claim 1, wherein, The output result is obtained by detecting the output current on the source lines of each column in the memristor array, including: The output current of each column in the memristor array is converted from analog to digital to obtain the output result in digital form.

5. The operating method according to claim 1, wherein, The operation method further includes programming operations, wherein the programming operations include setting and resetting operations for the memristor of the set memristor unit.

6. The operating method according to claim 5, wherein, The settings operation includes: A first control voltage is applied to the word line of the row containing the memristor cell to turn on the transistor of the memristor cell, while a second control voltage is applied to the other word lines in the memristor array to turn off the transistors of the memristor cells in the other rows. A corresponding setting voltage is applied to the bit line of the row where the set memristor cell is located, and a first operating voltage is applied to the other bit lines of the memristor array. The first operating voltage is applied to the source line of the column containing the configured memristor cell, and the setting voltage is applied to the other source lines of the memristor array.

7. The operating method according to claim 5, wherein, The reset operation includes: A first control voltage is applied to the word line of the row containing the memristor cell to turn on the transistor of the memristor cell, while a second control voltage is applied to the other word lines of the memristor array to turn off the transistors of the memristor cells in other rows. A first operating voltage is applied to the bit lines of each row in the memristor array; A reset voltage is applied to the source line of the column containing the configured memristor cell, and the first operating voltage is applied to the other source lines of the memristor array.

8. The operating method according to claim 6 or 7, wherein, The second control voltage is the same as the first operating voltage.

9. A data processing apparatus, comprising: Word line driver, source line driver, bit line driver, and output processing circuitry; as well as A memristor array, comprising multiple rows and columns of memristor cells, word lines corresponding to each row of the memristor array, bit lines corresponding to each row of the memristor array, and source lines corresponding to each column of the memristor array. Each memristor cell includes a transistor and a memristor. The gate of the transistor is connected to the word line corresponding to the row of the memristor cell. The first terminal of the transistor is connected to the source line corresponding to the column of the memristor cell. The second terminal of the transistor is connected to the first terminal of the memristor. The second terminal of the memristor is connected to the bit line corresponding to the row of the memristor cell. The word lines of each row are coupled to a word line driver. The source lines of each column are coupled to a source line processing circuit. The bit lines of each row are coupled to a bit line processing circuit. in, The word line driver is configured to apply the input voltage corresponding to the input data of each row in the memristor array to the gate of the transistor corresponding to the word line of each row in the memristor array. The source line driver is configured to apply a source line voltage to the source lines of each column in the memristor array, so that the transistors in each memristor cell of each row in the memristor array are turned on and operate in the saturation region. The bit line driver is configured to apply a first fixed voltage to the bit lines of each row in the memristor array. The output processing circuit is configured to detect the output current on the source lines of each column in the memristor array to obtain the output result.

10. The data processing apparatus according to claim 9, further comprising an input circuit, wherein, The input circuit is configured to perform digital-to-analog conversion on the input data of each row in the memristor array to obtain the corresponding input voltage, and to provide the corresponding input voltage to the word line driver.

11. The data processing apparatus according to claim 9, wherein, The output processing circuit is further configured to receive the output current of each column in the memristor array and perform analog-to-digital conversion on the output current to obtain the output result in digital form.

12. The data processing apparatus according to claim 9, wherein, Each of the memristor units comprises a single transistor and a single memristor, or, Each memristor cell includes two transistors and two memristors. The first terminals of the two transistors are electrically connected to each other and to the source line corresponding to the row of the memristor cell. The gates of the two transistors are respectively connected to two different word lines corresponding to the row of the memristor cell. The second terminals of the two memristors are electrically connected to two different bit lines corresponding to the row of the memristor cell.

Citation Information

Patent Citations

  • 2T1R array based on memristors and parallel operation method and algorithm thereof

    CN113222128A