Processing array device and computing storage unit using complementary XOR storage unit

The 3-port SRAM complementary XOR cell design addresses the limitations of SRAM cells by performing XOR operations and full adder calculations in a single clock cycle, improving computational efficiency.

CN112102864BActive Publication Date: 2024-12-06GSI TECHNOLOGY INC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010543431.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2019-06-18
Filing Date
2020-06-15
Publication Date
2024-12-06
Estimated Expiration
2040-06-15

AI Technical Summary

Technical Problem

Existing memory cells, particularly SRAM cells, are limited in their ability to perform certain logical functions like XOR operations, which are crucial for search operations, and require multiple clock cycles for full adder operations, hindering efficient computing performance.

Method used

A 3-port SRAM complementary XOR cell design that allows two logical calculations per clock cycle, enabling full adder operations in a single clock cycle through a processing array with separate read and write bit lines and specialized bitline read/write logic circuits.

Benefits of technology

Enables faster floating-point operations and efficient full adder calculations in a single clock cycle, reducing computational bottlenecks and enhancing processing efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112102864B_ABST
    Figure CN112102864B_ABST
Patent Text Reader

Abstract

The present application provides a processing array device and a computational storage unit using complementary XOR storage cells. The processing array and storage unit having multiple memories are capable of performing logic functions, including XOR or XNOR logic functions. The storage unit may have a read port, wherein digital data stored in the storage cells of the storage unit is isolated from a read bit line.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure generally relates to static random access memory cells that can be used for computing. Background Art

[0002] An array of memory cells such as dynamic random access memory (DRAM) cells, static random access memory (SRAM) cells, content addressable memory (CAM) cells, or non-volatile memory cells is a well-known mechanism for storing digital bits of data in various computer or processor-based devices. Various computer and processor-based devices may include computer systems, smart phone devices, consumer electronics, televisions, Internet switches and routers, etc. An array of memory cells is typically packaged in an integrated circuit, or may be packaged within an integrated circuit that also has a processing device within the integrated circuit. Different types of typical memory cells have different functions and characteristics that distinguish each type of memory cell. For example, DRAM cells require a longer time to access and will lose their data content unless they are refreshed regularly, but their manufacturing cost is relatively low due to the simple structure of each DRAM cell. On the other hand, SRAM cells have faster access times, will not lose their data content unless power is removed from the SRAM cell, and are relatively more expensive because each SRAM cell is more complex than a DRAM cell. CAM cells have the unique feature of being able to easily address the content in the cell and are more expensive to manufacture since each CAM cell requires more circuitry to implement the content addressing function.

[0003] Various computing devices that can be used to perform calculations on digital binary data are also well known. The computing devices may include microprocessors, CPUs, microcontrollers, etc. These computing devices are typically manufactured on integrated circuits, but may also be manufactured on integrated circuits that also have a certain amount of memory integrated into the integrated circuit. In these known integrated circuits having computing devices and memory, the computing devices perform calculations on digital binary data bits, while the memory is used to store various digital binary data, including, for example, instructions executed by the computing devices and data operated by the computing devices.

[0004] Recently, devices have been introduced that use memory arrays or storage cells to perform computational operations. In some of these devices, an array of processors for performing computations may be formed from the storage cells. These devices may be referred to as in-memory computational devices.

[0005] Big data operations are data processing operations in which large amounts of data must be processed. Machine learning uses artificial intelligence algorithms to analyze data and typically requires large amounts of data to perform. Big data operations and machine learning are also typically compute-intensive applications that often experience input / output issues due to bandwidth bottlenecks between the computing device and the memory where the data is stored. Since in-memory computing devices perform computations within the memory, the above-described in-memory computing devices can, for example, be used for these big data operations and machine learning applications, thereby eliminating bandwidth bottlenecks.

[0006] The SRAM cell can be configured to perform basic Boolean operations, such as AND, OR, NAND, and NOR. The SRAM cell can also support selective write operations. However, the SRAM cell cannot perform certain logical functions that can be expected. For example, since the XOR logical function is often used in the search operation when the search keyword needs to be compared with the content in the storage, it is expected that the XOR (XOR) logical function can be implemented.

[0007] One of the most basic operations of a graphics processing unit (GPU) is floating point arithmetic. Floating point arithmetic can be performed by a well-known full adder circuit. In a system described in co-pending and commonly owned U.S. patent application Ser. No. 15 / 708,181, filed Sep. 19, 2017, a full adder can be performed in 4 clock cycles. However, it is desirable to be able to perform a full adder in a single clock cycle, and the present disclosure is directed to this end. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 An embodiment of a complementary XOR cell with two read bit lines is shown;

[0009] Figure 2 Shows a number of Figure 1 The implementation of the first embodiment of the processing array of complementary XOR units shown;

[0010] Figure 3 Shown for Figure 1 Selective write truth table of complementary XOR 3-port SRAM cell;

[0011] Figure 4 and Figure 5 shows that a Figure 2 A truth table of a full adder implemented by the processing array of complementary XOR units shown;

[0012] Figure 6 Shows a number of Figure 1 An implementation of a second embodiment of a processing array of complementary XOR units and partitioning segments is shown;

[0013] Figure 7 Shown for Figure 6 An implementation of a first embodiment of bitline read / write logic for a processing array is shown;

[0014] Figure 8 Shown for Figure 6 An implementation of a second embodiment of bitline read / write logic for a processing array is shown;

[0015] Fig. 9 Shown for Figure 6 An implementation of a third embodiment of bitline read / write logic for a processing array is shown;

[0016] Fig.10 Shown for Figure 6 The fourth embodiment of the bit line read / write logic of the processing array shown in FIG. Figure 8 and Fig. 9 Embodiments in;

[0017] Fig.11 Shown for Figure 6 An implementation of a fifth embodiment of bitline read / write logic for a processing array is shown; and

[0018] Fig.12 Shown for Figure 6 An implementation of a sixth embodiment of bitline read / write logic for a processing array is shown. DETAILED DESCRIPTION

[0019] The present disclosure is particularly applicable to CMOS-implemented storage cells and processing arrays having multiple storage cells that are capable of performing two logical calculations that can be used to perform full adder operations in a single clock cycle, and the present disclosure will be described in this context. However, it will be understood that since the storage cells can be constructed using different processes and can have different circuit configurations compared to the circuit configurations disclosed below that perform two calculations and are therefore within the scope of the present disclosure, the storage cells and processing arrays have greater utility and are not limited to the embodiments disclosed below. For illustrative purposes, a 3-port SRAM complementary XOR cell is disclosed below and in the accompanying drawings. However, it should be understood that the SRAM computing cell and processing array can also be implemented with an SRAM cell having three or more ports, and the present disclosure is not limited to the 3-port SRAM cell disclosed below. It should also be understood that the construction of the SRAM cell having three or more ports may be slightly different from the 3-port SRAM shown in the figure, but those skilled in the art will understand how to construct the SRAM for those three or more ports disclosed below.

[0020] In addition, although SRAM cells are used in the following examples, it should be understood that various different types of memory cells including DRAM, CAM, non-volatile memory cells and non-volatile storage devices can be used to implement the disclosed memory cells for computing and processing arrays using the same, and these embodiments using various types of memory cells are within the scope of the present disclosure.

[0021] Figure 1 1 shows an embodiment of a 3-port SRAM complementary XOR cell 100 that can have two read bit lines and generate two logic calculations per clock cycle. The 3-port SRAM cell 100 can include two cross-coupled inverters I31, I32 and Figure 1 The access transistors M31-M37 are coupled together to form a basic SRAM cell. The SRAM cell can be used as a storage latch and can have two read ports (including two read bit lines and transistors M31, M32, M36, M37) and a write port (including a write bit line and transistors M33, M34, M35) to form a 3-port SRAM. Figure 1 The input of the first inverter shown is connected to the output of the second inverter (marked as D), and the output of the first inverter (marked as Db) is coupled to the input of the second inverter, so the two inverters I31, I32 are cross-coupled. The cross-coupled inverters I31, I32 form a latch of the SRAM cell. Transistors M34 and M33 can have their respective gates connected to the write bit line and its complementary bit line (WBL, WBLb) respectively. The write word line carries a signal WE. The write word line WE is coupled to the gate of a transistor M35 that is part of the write access circuit of the SRAM cell.

[0022] Figure 1 The circuit in may also have a read word line RE, a complementary read word line REb, a read bit line RBL, a complementary read bit line RBLb, and a read port formed by transistors M31 and M32 coupled together, and another read port formed by transistors M36 and M37 coupled together. The read word line RE may be coupled to the gate of transistor M31 forming one read port, and the read bit line RBL is coupled to the drain terminal of transistor M31 to perform a first calculation during each clock cycle. The complementary read word line REb may be coupled to the gate of transistor M36 forming another read port, and the read bit line RBLb is coupled to the drain terminal of transistor M36 to perform a second calculation during each clock cycle. The gates of transistors M32 and M37 may be coupled to output terminals Db and D of cross-coupled inverters I31 and I32, respectively. The isolation circuit couples latch output terminals Db and D (at Figure 1) is isolated from the signal / voltage levels of RBL and RBLb) so that the Db and D signals are not susceptible to lower bit line levels caused by multiple "0" data stored in multiple cells compared to typical SRAM cells.

[0023] Figure 1 The cell 100 is shown with two read bit lines, RBL and RBLb. The two read bit lines allow two calculations to be performed per clock cycle (one calculation using each read bit line). Figure 2 The unit 100 in the illustrated processing array 200 allows for faster operation of floating point operations. For example, a processing array 200 having multiple units 100 may perform a full adder operation in a single clock cycle.

[0024] When the read port of the cell 100 is active, it can be operated so that RE or REb is high, and the REb signal / voltage level is the complement of the RE signal / voltage level. It can also be operated so that the RE or REb signal / voltage level is both high or both low. RBL is precharged high, and if the transistor pair M31, M32 are both turned on, RBL is discharged to 0. If any one of the M31, M32 transistors is turned off, since RBL is precharged high and not coupled to ground, RBL remains high at 1. RBLb is also precharged high, and if the transistor pair M36, M37 are both turned on, RBLb is discharged to 0. If any one of the M36, M37 transistors is turned off, since RBLb is precharged high and not coupled to ground, RBLb remains high at 1. The cell 100 can be used as a 3-port SRAM cell. The write operation is activated by WE, and data is written by switching WBL and WBLb. The read operation is activated by RE and REb, and the read data is accessed through RBL or RBLb. The unit 100 can also be used for calculations in which RBL and RBLb are also used for logical operations.

[0025] Figure 2 An embodiment of a processing array 200 is shown, which has Figure 1A plurality of SRAM cells (cell 00, ..., cell On and cell m0, ..., cell mn formed in an array) are shown that can perform two calculations using two read bit lines. The array shown is formed of M word lines (RE0, Reb0, WE0, ..., Rem, REbm, WEm) and N bit lines (WBLb0, WBL0, RBL0, RBL0b, ..., WBLbn, WBLn, RBLn, RBLnb). The processing array 200 may have a word line generator 202 that generates word line signals / voltage levels and a plurality of bit line read / write logic circuits 204 (BL read / write logic 0, ..., BL read / write logic n) that receive and process the bit line signals to generate the results of two Boolean logic functions / calculations performed at each clock cycle. In each clock cycle, the processing array 200 may have a read operation and a write operation, or a read operation or a write operation, or no operation, depending on the activation of RE, REb and WE and the operation of the BL read / write logic. The following disclosure of the BL read / write logic only shows the read operation and the write operation performed sequentially. However, the read bit line and the write bit line are separated, and the processing array 200 of the three-port SRAM cell 100 can perform the read operation and the write operation at the same time, and the present disclosure is not limited to the read operation and the write operation performed sequentially.

[0026] In a read operation, the WL generator 202 generates one or more RE or REb signals in a cycle, and RBL and / or RBLb form a Boolean function as described above, and the results of RBL and / or RBLb are sensed / read by the BL read / write logic 204. The BL read / write logic 204 processes the results of RBL and RBLb and sends the results back to its WBL / WBLb to be written / used in the cell; or sends the results to the adjacent BL read / write logic to be written / used in the adjacent cell; or sends the results outside the processing array. Alternatively, the BL read / write logic 204 can store the RBL results and / or RBLb results from its own bit line or from the adjacent bit line in a register or latch, so that the next cycle read / write logic can perform logic using the latched RBL and / or RBLb result data.

[0027] In a write operation, the WL generator 202 generates one or more WE signals for the cell to be written, and the BL read / write logic 204 processes the write data, either from its own RBL or RBLb line, or from an adjacent RBL or RBLb line, or from data outside the processing array. The ability of the BL read / write logic 204 to process data from adjacent bit lines means that data can be transferred from one bit line to an adjacent bit line, and one or more or all bit lines in the processing array can be transferred simultaneously. The BL read / write logic 204 can also decide not to write for a selective write operation based on the RBL result and / or the RBLb result. For example, if RBL=1, the WBL can be written with data. If RBL=0, no write is performed.

[0028] Each BL read / write logic 204 may have one or more Boolean logic circuits that use the read bit line as an input to perform various Boolean operations. For example, each BL read / write logic 204 may include one or more AND circuits and OR circuits implemented in various known ways, and the present disclosure is not limited to a specific implementation of AND circuits and OR circuits in the BL read / write logic 204.

[0029] use Figure 1 Unit 100 and Figure 2 For the processing array 200 shown in FIG. , the calculation performed can be written as:

[0030] RBL=AND(D1,D2,..,Dn) (EQ1)

[0031] Where D1, D2, ..., Dn are the "n" data stored in the cell when the RE signal is turned on / activated

[0032] RBLb=AND(Dbi1,Dbi2,..,Dbin) (EQ2)

[0033] Where Dbi1, Dbi2, ..., Dbin are the "n" data stored in the unit where the REb signal is turned on / activated

[0034] RBLb=NOR(Di1, Di2,...,Din) (EQ3)

[0035] Where Di1, Di2, ..., Din are the "n" data stored in the cell when the REb signal is turned on / activated

[0036] The RE and REb signals of a cell may be both on, one on and one off, or both off. Therefore, Dm and Dbim (where m=1 to n) in EQ1 and EQ2 may be true data and complement data of the same cell, or may not be so.

[0037] In the BL read / write logic 204, the AND gates may form the following equation:

[0038] Y1=AND(RBL,RBLb), then

[0039] Y1=AND(D1,D2,...,Dn,Dbil,Dbi2,...Dbim) (EQ4)

[0040] Wherein D1, D2, ... Dn are "n" data stored in the cells where RE is turned on, and Dbi1, Dbi2, ... Dbim are "m" data stored in the cells where Reb is turned on.

[0041] If RE and REb of the same cell are complementary signals, meaning they are not on at the same time, then Y1 behaves as if RBL and RBLb are connected together and can be shown as

[0042] Y1=AND(XNOR(RE1,D1),XNOR(RE2,D2),...,XNOR(REn,Dn)) (EQ5)

[0043] Alternatively, if RBL and RBL are connected together by connecting MOS transistors or by hard wiring, we can also realize Y1 of EQ5.

[0044] Y1 performs identically to the RBL line of the XOR cell (disclosed in U.S. patent application Ser. No. 15 / 709,399, filed Sept. 19, 2017, which is incorporated herein by reference), and the details regarding the operation of Y1 occur in the same manner as previously disclosed and will not be further described here.

[0045] In the BL read / write logic 204, the OR gate may form the following equation:

[0046] Y2=OR(RBL,RBLb), then

[0047] Y2=OR(AND(D1,D2,...,Dn),AND(Dbi1,Dbi2,...,Dbin)) (EQ6)

[0048] The cell x or ix may be the same cell. In other words, the RE and REb of the cell may be turned on at the same time.

[0049] If RE and REb of cells 1 and 2 are turned on, Y2 can be expressed as

[0050] Y2=OR(AND(D1,D2),AND(Db1,Db2))=XNOR(D1,D2) (EQ7)

[0051] Y2 performs the XOR function of units 1 and 2 on the same bit line.

[0052] Figure 3 Shows Figure 1 If WE is 0, no write operation is performed (e.g. Figure 3 1). If WE is 1, the storage node D and the complementary Db are written by WBL and WBLb. If WBL=1 and WBLb=0, then D=1 and Db=0. If WBL=0 and WBLb=1, then D=0 and Db=1. If both WBL and WBLb are 0, no write is performed. Therefore, for WE=1, the cell can perform a select write function with WBL=WBLb=0.

[0053] When performing selective writing, it is usually desirable to write data "1" on some cells and data "0" on other cells on the same bit line. Figure 1 For the 3-port XOR cell shown, this can be achieved by spending 2 cycles writing data, where data "1" is written in one cycle and data "0" is written in the other cycle.

[0054] Figure 4 and Figure 5 shows the full adder truth table for a full adder, which can be used Figure 2 The invention can be realized by using a processing array with complementary XOR units shown in FIG. Ain and Bin are 2 full adder inputs, and Cin is the carry input. Bout is the sum output, and Cout is the carry output. Cinb is the inversion of Cin, and Coutb is the inversion of Cout. As shown in the truth table, two or more of the three inputs (Ain, Bin, and Cin) being logic "1" will generate Cout as logic "1", and an odd number of the three inputs being logic "1" (one of the inputs is "1" or three of the inputs are "1") will generate Bout as logic "1".

[0055] Using the above processing array 200 and unit 100, the three inputs of the full adder can be stored in three units along the same bit line, so after the logic operation, the processing array can generate sum and carry outputs in two units. In this exemplary embodiment, the unit storing the Bin input is shared as the sum output Bout, and the unit storing the Cin unit can be shared as the carry output Cout. The processing array can further generate the Coutb and Cinb signals shown in the truth table because they are simply the inverted signals of Cout and Cin.

[0056] exist Figure 4 In states 1, 3, 6, and 8 shown, Bout = Bin and Coutb = Cinb. This means that if the input is a combination of the 4 values ​​shown in states 1, 3, 6, and 8, no calculations need to be done and the processing array will be able to generate the output without any logic operations. Figure 5 As shown, there is a subset of states (2, 4, 5, and 7) where the output is different from the input (called the "post-operation state"). Figure 5 As shown, in all post-operation states, Ain and Cinb are either both 0 or both 1. In states where the output does not need to be changed, the values ​​of Ain and Cinb are different. Therefore, the circuit in the BL read / write logic 204 can perform an operation of XNOR (Ain, Cinb) = 1, and then the output of the full adder needs to be changed. Figure 5 It is also shown that for all post-operation states, Bout has the same value as Coutb, and the values ​​of Bout and Coutb are the inverse of Bin. Therefore, the circuit in the BL read / write logic 204 can perform the following operations: if XNOR(Ain, Cinb)=1, then Bout=Cinb=NOT(Bin); and if XNOR(Ain, Cinb)=0, then Bout and Cinb are unchanged.

[0057] Figure 6 1 shows an implementation of a second embodiment of a processing array 600 having Figure 1 The processing array has the same word line generator 602, which generates control signals (RE0, ..., Rem, Reb0, ..., REbm and WE0, ..., WEm), and is combined with the above Figure 2 The same operation is described above. In this embodiment, each bit line has two segments. Segment 1 has RBLs1 and RBLs1b read bit lines (RBL0s1, ..., RBLns1 and RBL0s1b, ..., RBLns1b) to which multiple cells ( Figure 6 , . . . , cell 0n in the example in FIG. 5 ), which are all connected to the BL read / write circuit 604. Segment 2 has only RBLs2 lines (RBL0s2, . . . , RBLns2) which have multiple cells ( Figure 6 , cells m0, ..., cells mn in the example in ), which are all connected to another input of the BL read / write circuit 604. For illustration purposes, RBLb in segment 2 is hardwired to RBL. As previously mentioned, RBLb in segment 2 can also be connected to RBL through a transistor. In this example, WBL and WBLb are shared between the two segments. For example, to make a 16-bit adder, we need 16 bits of Ain and 1 bit of Bin input to store Cin / Cout. Then, in this configuration, there may be 16 or more cells on Rbin2 for Bin and 17 or more cells on RBLs1, with 16 cells for Ain plus 1 additional cell for Cin. As Figure 6 As shown, each BL read / write logic 604 has three inputs from the RBL (e.g., RBL0s1, RBL0s1b, and RBL0s2), which can perform a full adder operation with three inputs in one cycle in many different implementations. Figure 7-Figure 12 It is shown in Figure 6 6. Six non-limiting examples of circuits of the BL read / write circuit 604 in FIG. 6 that perform full adder operations and / or search operations in one clock cycle.

[0058] Figure 7 Shown for Figure 6 1 is an implementation of a first embodiment of the BL read / write logic 604 of the processing array shown. In this implementation, Ain and Cinb are stored in cells in segment 1 along the same bit line, and Bin is stored in segment 2. During a read operation, the states of these cells are as follows:

[0059] 1. When reading, the RE and REb of the Ain and Cinb units are both 1, and the result is:

[0060] a.RBLs1=AND(Ain,Cinb) (EQ8)

[0061] b.RBLs1b=AND(Ainb,Cin) (EQ9)

[0062] 2. When reading, the RE of the Bin unit is 1 and the REb of the Bin unit is 0, the result is:

[0063] a.RBLs2=Bin (EQ10)

[0064] As described in EQ 7 and from EQ 7-9, the OR gate 700 can be represented as

[0065] Y3=OR(RBLs1,RBLs1b)=OR(AND(Ain,Cinb),AND(Ainb,Cin))=XNOR(Ain,Cinb)(EQ11)

[0066] The BL read / write logic 604 may further include an inverter 702, whose input is connected to the Bin signal (RBLs2 line) and whose output (inverted Bin signal) is connected to the input of a first AND gate 704. Another input of the AND gate 704 is connected to the Y3 signal. The BL read / write logic 604 may further include a second AND gate 706, whose input is connected to the Bin signal and the Y3 signal. The outputs of the two AND gates are the WBL signal and the WBLb signal.

[0067] based on Figure 7 The above logic in:

[0068] If Y3 = 1, then Bout = Coutb = NOT (Bin) (EQ12)

[0069] If Y3 = 0, Bout and Coutb remain unchanged (EQ13)

[0070] In this implementation, Bin and Cinb need to be selectively written, and if Y3=1, they are written as NOT(Bin). If Y3=0, no write operation is performed. After the selective writing, Bin and Cinb become Bout and Coutb respectively:

[0071] WBL=AND(Y3, NOT(Bin)) (EQ14)

[0072] WBLb=AND(Y3,Bin) (EQ15)

[0073] During writing, WE of Bin and Cinb is 1. (EQ16)

[0074] use Figure 7 With the above BL read / write logic in place, the full adder operation can be performed in a single clock cycle with Figure 5 In a single clock cycle, RE may be a pulse in the first half of the clock cycle to generate WBL and WBLb, and WE may be another non-overlapping pulse of RE in the second half of the clock cycle to perform a write to the Bout and Coutb cells.

[0075] Figure 7The full adder circuit is demonstrated by storing Ain and Cinb on segment 1 and Bin on segment 2. Alternatively, the full adder circuit can be demonstrated by storing Ain and Cin on segment 1 and Bin on segment 2. Figure 7 A small modification of the circuit in can be used to make a full adder circuit. OR gate 700 can be changed to a NOR gate so that Y3 is still defined by EQ11. WBL and WBLb of segment 1 and segment 2 need to be connected to BL read / write logic 604 respectively. WBL and WBLb of segment 2 are connected to AND gates 704 and 706 respectively to write the Bout value. WBL and WBLb of segment 1 are connected to AND gates 706 and 704 respectively to write the Cout value. The implementation discussed below is also demonstrated with Ain, Cinb and Bin stored in the processing array, but other combinations of storing different input polarities can be easily implemented by those skilled in the art.

[0076] Figure 8 Shown for Figure 6 1 is an implementation of a second embodiment of the BL read / write logic 604 of the processing array shown. In this implementation, Ain and Cinb are stored in cells in segment 1 along the same bit line. Bin is stored in segment 2. During a read operation, the states of these cells are as follows:

[0077] 1. For Ain, when reading, RE=1, REb=0. For Cinb, when reading, RE=0, REb=1. Therefore

[0078] a.RBLs1=Ain (EQ17)

[0079] b.RBLs1b=NOT(Cinb) (EQ18)

[0080] 2. When reading, RE of Bin is 1 and REb of Bin is 0.

[0081] a.RBLs2=Bin (EQ19)

[0082] Y4 is the XOR function of RBLs1 and RBLs1b (via XOR logic gate 800), so

[0083] Y4=XOR(RBLs1, RBLs1b)=XOR(Ain, NOT(Cinb))=XNOR(Ain, Cinb) (EQ20)

[0084] EQ20 shows the same result as EQ12, resulting in Y3=Y4, so Figure 8 WBL and WBLb have the Figure 7, and has the same inverter 802 and two AND gates 804, 806 that perform in the same manner as described above. Figure 8 A full adder operation is performed in .

[0085] Figure 8 Embodiments of circuit 604 in may also combine additional inputs to the full adder. For example, it may be desirable to have a full adder with inputs such as X, Bin, and Cin, where X is an AND function of Ain and W. Ain and W may be 2 values ​​stored in 2 cells along the same RBL, RBLs1, and X is formed by turning on the read word lines (RE) of the two cells storing Ain and W, as shown in EQ1. This functionality is particularly useful when implementing a multiplier circuit when the basic cell is a full adder with an AND function of 2 inputs as the multiplicand. One implementation method is to store W in an additional cell in segment 1 along the same bit line as Ain and Cinb. Then in a read operation, the state of the cell is as follows:

[0086] 1. For Ain and W, when reading, RE = 1, REb = 0. For Cinb, when reading, RE = 0, REb = 1. Therefore

[0087] a.RBLs1=AND(Ain,W)=X (EQ21)

[0088] b.RBLs1b=NOT(Cinb) (EQ22)

[0089] 2. When reading, RE of Bin is 1 and REb of Bin is 0.

[0090] a.RBLs2=Bin (EQ23)

[0091] Y4 is the XOR function of RBLs1 and RBLs1b, so

[0092] Y4=XOR(RBLs1, RBLs1b)=XOR(X, NOT(Cinb0)=XNOR(X, Cinb) (EQ24)

[0093] EQ24 shows the same results as EQ20, but with an additional AND function as the full adder input. Another approach is to use signal W as the read word line for the Ain cell. In a read operation, the state of the cell is as follows:

[0094] 1. For Ain, when reading, RE=0, REb=W. For Cinb, when reading, RE=1, REb=0. Therefore

[0095] a.RBLs1=Cinb (EQ25)

[0096] b.RBLs1b=NAND(W,Ain)=NOT(X) (EQ26)

[0097] 2. When reading, RE of Bin is 1 and REb of Bin is 0.

[0098] a.RBLs2=Bin (EQ27)

[0099] Y4 is the XOR function of RBLs1 and RBLs1b, so

[0100] Y4=XOR(RBLs1, RBLs1b)=XOR(Cinb, NOT(X))=XNOR(X, Cinb) (EQ28)

[0101] EQ28 shows the same results as EQ20, with an additional AND function as the full adder input.

[0102] As described in EQ4, the AND function of RBLs1 and RBLs1b generates the following equation:

[0103] Y1=AND(XNOR(RE1,D1),XNOR(RE2,D2),...,XNOR(REn,Dn)) (EQ4)

[0104] Where REi, REbi, i=1 to n, as complementary signals; REi=1 means REi=1, REbi=0; REi=0 means REi=0, REbi=1. Di is the data stored in the cell i with the read word line REi. Y1 is also the comparison result of RE1 to REn with D1 to Dn. If RE1 to REn as the search keyword matches D1 to Dn, then Y1=1. If any one of RE1 to REn does not match the corresponding D1 to Dn, then Y1=0.

[0105] Similarly, RBLs2 is a wired AND function of RBLs2 and RBLs2b, and RBLs2 is the comparison result of the activated REi and REib of RBLs2.

[0106] Fig. 9 Shown for Figure 6 6 is a third embodiment of the BL read / write logic 604 of the processing array. Fig. 9 middle, Figure 7 OR logic gate 700 or Figure 8 The XOR logic gate 800 in is replaced by an AND logic gate 900, and the inputs of the AND logic gate 900 are the RBLs1, RBLs1b and RBLs2 lines. Fig. 9 In the example, signal Y5 is as follows:

[0107] Y5=AND(RBLs1, RBLs1b, RBLs2) EQ(29)

[0108] Y5 is the comparison result of RBL segment 1 and RBL segment 2. If the search key consists of the read word lines in segment 1 and segment 2, Y5 will produce a combined result. If the search key consists of only RE and REb of segment 1, and neither RE nor REb of segment 2 is active, then RBLs2 remains at 1 because RBLs2 is precharged to 1 and Y5 produces the comparison result of segment 1. Similarly, if the search key consists of only RE and REb of segment 2, and neither RE nor REb of segment 1 is active, then RBLs1 remains at 1 because RBLs1 is precharged to 1 and Y5 produces the comparison result of segment 2. In summary, Y5 produces a comparison result that RE and Reb are turned on in segment 1 or segment 2 or both.

[0109] exist Fig. 9 In the embodiment, the WDb signal can be a data line connected in parallel with RE to feed to the BL read / write logic 604, or can be fed from the adjacent BL read / write logic. If Y5 is 1, WBL=NOT(WDb)=WD and WBLb=WDb to perform selective writing to store WD in the cell selected by WE during writing. If Y5 is 0, writing will not be performed even if the selected WE is turned on. The above logic is implemented by an inverter 902 connected to the WDb signal, an AND logic gate 904 whose inputs are the Y5 signal and the WD signal and whose output is fed to the WBL line, and a second AND gate 906 whose inputs are the Y5 signal and the WDb signal and whose output is fed to the WBLb line.

[0110] Fig.10 Shown for Figure 6 , which incorporates the fourth embodiment of the BL read / write logic 604 of the processing array shown in FIG. Figure 8 and 9 The circuit in the embodiments of the present invention can perform full adder operations and / or search operations. Fig.10 The circuits 800, 802, 804, 806 and 900 in are the same as described above. In addition, Fig.10 The signal Y4 in Figure 8 same as Y4 in , as a selective write control signal for full adder operation, and Fig.10 The signal Y5 in Fig. 9The BL read / write logic 604 can receive a global multiplexer (Mux) signal in parallel with the word line, which uses a 2:1 multiplexer 1007 to select the Y4 signal or the Y5 signal to generate a Y6 signal as a selective write control signal. The second multiplexer 1008 (also a 2:1 multiplexer) selects RBLs2 for full adder operation or WDb for search operation to generate a Y7 signal as the write data to be written in the selective write operation. In this way, Fig.10 It consists of circuits that perform two main operations: full adder and search.

[0111] Fig.11 Shown for Figure 6 , which incorporates the fifth embodiment of the BL read / write logic 604 of the processing array shown in FIG. Figure 8 and 9 The circuit in the embodiment in can perform more complete adder operations and / or search operations. Fig.11 The circuits 800, 802, 804, 806 and 900 in are the same as described above. In addition, Fig.11 The signal Y4 in Figure 8 same as Y4 in , as a selective write control signal for full adder operation, and Fig.11 The signal Y5 in Fig. 9 The same as Y5 in, serving as a selective write control signal for search operations. Fig.11 Added an additional feature to Fig.10 The circuit shown. Fig.11 The circuit in includes a new input W2 to multiplexer 1107 (now a 3:1 multiplexer with inputs Y4 (mux[0]), Y5 (mux[1]), and W2 (mux[2])). W2 is added as another write control signal to perform a write function on input 2 of write data multiplexer 1108. Y5 is fed into write data multiplexer 1108 as input 2. Therefore if Mux[2] is selected and Mux[1]=Mux[0]=0, multiplexer 1107 selects W2 as the selective write control signal. If W2=1, WBLb=NOT(WBL), then the write data to be written on the WBL of the cell is shown as:

[0112] WBL=NOT(Y7)=NOT(Y5)=NAND(RBLs1, RBLs1b, RBLs2) EQ

[30]

[0113] If RBLs2 is not active, meaning that both RE and REb on segment 2 are not selected and RBLs2 is precharged high and remains high, then EQ

[30] may be shown as EQ

[31] . Similarly, if RBLs1b or RBLs1 is not active, then it may be shown as EQ

[32] or EQ

[33] , respectively.

[0114] If RBLs2 is not active, then WBL = NAND (RBLs1, RBLs1b) EQ

[31]

[0115] If RBLs1b is not active, then WBL=NAND(RBLs1, RBLs2) EQ

[32]

[0116] If RBLs1 is not active, then WBL = NAND (RBLs1b, RBLs2) EQ

[33]

[0117] In addition, if only one of the RBLs in the above equation is active, the equations are simplified to EQ

[34] to EQ

[37] WBL = NOT(RBLs1) = NOT(AND(D1, D2, ..., Dn)) = NAND(D1, D2, ..., Dn)

[0118] D1, D2, ..., Dn are the data in the cells where RE=1 when reading RBLs1. EQ

[34]

[0119] WBL=NOT(RBLs1b)=NOT(AND(D1b,D2b,...,Dnb))=NAND(D1b,D2b,...,Dnb)=OR(D1,D2,...Dn)

[0120] D1, D2, ..., Dn are the data in the cells where REb = 1 when reading RBLs1b. EQ

[35]

[0121] WBL=NOT(RBLs2)=NOT(AND(D1, D2,...,Dn))=NAND(D1, D2,...Dn)

[0122] D1, D2, ..., Dn are the data in the cells where RE=1 when reading RBLs2. EQ

[36]

[0123] WBL=NOT(RBLs2)=NOT(AND(D1b,D2b,...,Dnb))=NAND(D1b,D2b,...,Dnb)=OR(D1,D2,...Dn)

[0124] D1, D2, ..., Dn are the data in the cells where REb = 1 when reading RBLs2. EQ

[37]

[0125] In conclusion, Fig.11 The circuit shown can implement the following three basic functions:

[0126] 1.Mux[0]: full adder operation.

[0127] 2.Mux[1]: Search operation.

[0128] 3.Mux[2]: Combinatorial Boolean operation of RBLs1, RBLs1b and RBLs2.

[0129] Fig.12 Shown for Figure 6 The sixth embodiment of the BL read / write logic 604 of the processing array shown in FIG. 6 has the same features as those described above for Figure 8 , Fig. 9 and Fig.11 The same circuits 800-806, 900, 1107 and 1108 perform the same functions as described above. Similarly, the Y4, Y5, Y6 and Y7 signals are the same as those described above for Fig.11 The signals described above are the same. In this embodiment, Y7 of the adjacent bit line of the processing array can be fed into the data multiplexer 1209 as an additional data input for the shift operation. Y7 of the adjacent bit line, data outside the processing array, can also be fed into the data multiplexer 1209 as an additional input to transfer external data to the processing array. The write data multiplexer 1209 multiplexes the data of the adjacent BL read / write logic and the external data into the processing array to be written to WBL and WBLb.

[0130] For the purpose of explanation, the foregoing description has been described with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. In view of the above teachings, many modifications and variations are possible. The embodiments are selected and described in order to best explain the principles of the present disclosure and its practical application, so that others skilled in the art can best utilize the present disclosure and various embodiments with various modifications suitable for the intended specific use.

[0131] The systems and methods disclosed herein may be implemented via one or more components, systems, servers, devices, other subcomponents, or may be distributed among these elements. When implemented as a system, such a system may include and / or involve components such as software modules, general-purpose CPUs, RAM, etc. found in general-purpose computers, among others. In embodiments where the innovation resides on a server, such a server may include or involve components such as CPUs, RAM, etc., such as those found in general-purpose computers.

[0132] In addition, the systems and methods herein can be implemented by implementations having completely different software, hardware and / or firmware components other than those described above. With respect to other components (e.g., software, processing components, etc.) and / or computer-readable media related to or embodying the present invention, for example, the innovative aspects herein can be implemented consistently with many general or special-purpose computing systems or configurations. Various exemplary computing systems, environments, and / or configurations suitable for use with the innovations herein may include, but are not limited to: software or other components within or embodied on a personal computer, server, or server computing device (such as a routing / connection component, a handheld or laptop device, a multiprocessor system, a microprocessor-based system, a set-top box, a consumer electronic device, a network PC, other existing computer platforms, a distributed computing environment including one or more of the above systems or devices, etc.).

[0133] In some cases, for example, aspects of the systems and methods may be implemented or implemented by logic and / or logic instructions including program modules that are executed in association with such components or circuits. Typically, program modules may include routines, programs, objects, components, data structures, etc. that implement specific tasks or implement specific instructions therein. The present invention may also be practiced in the context of distributed software, computer or circuit arrangements, wherein the circuits are connected by communication buses, circuits or links. In a distributed arrangement, control / instructions may be performed from local and remote computer storage media (including memory storage devices).

[0134] The software, circuits and components herein may also include and / or utilize one or more types of computer-readable media. Computer-readable media may be any available media that resides on such circuits and / or computing components, is associated with or can be accessed by such circuits and / or computing components. As an example and not limitation, computer-readable media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical storage, tape, disk storage or other magnetic storage device or any other medium that can be used to store required information and can be accessed by computing components. Communication media may include computer-readable instructions, data structures, program modules and / or other components. In addition, communication media may include wired media such as wired networks or direct wired connections, however, any such type of media herein does not include transient media. Any of the above combinations are also included within the scope of computer-readable media.

[0135] In this specification, the term component, module, device, etc. may refer to any type of logic or functional software element, circuit, block and / or process that can be implemented in various ways. For example, the functions of various circuits and / or blocks can be combined with each other into any other number of modules. Each module can even be implemented as a software program stored in a tangible memory (e.g., random access memory, read-only memory, CD-ROM memory, hard disk drive, etc.) to be read by a central processing unit to implement the innovative functions of this article. Alternatively, a module may include programming instructions transmitted to a general-purpose computer or processing / graphic hardware by a transmission carrier. In addition, a module may be implemented as a hardware logic circuit that implements the functions included in the innovation of this article. Finally, a module may be implemented using a dedicated instruction (SIMD instruction), a field programmable logic array, or any combination thereof that provides desired performance and cost.

[0136] As disclosed herein, features consistent with the present disclosure can be implemented via computer hardware, software and / or firmware. For example, the systems and methods disclosed herein can be embodied in various forms, including, for example, data processors, such as computers, digital electronic circuits, firmware, software, or combinations thereof that also include databases. In addition, although some of the disclosed embodiments describe specific hardware components, systems and methods consistent with the innovations herein can be implemented in any combination of hardware, software, and / or firmware. In addition, the above-mentioned features and other aspects and principles of the innovations herein can be implemented in various environments. Such environments and related applications can be specially constructed to perform various routines, processes, and / or operations according to the present invention, or they can include general-purpose computers or computing platforms that are selectively activated or reconfigured by code to provide necessary functions. The processes disclosed herein are not inherently related to any particular computer, network, structure, environment, or other device, and can be implemented by appropriate combinations of hardware, software, and / or firmware. For example, various general-purpose machines can be used with programs written according to the teachings of the present invention, or it may be more convenient to construct special-purpose devices or systems to perform the required methods and techniques.

[0137] Various aspects of the methods and systems described herein, such as logic, may also be implemented as functions programmed into various circuits including programmable logic devices ("PLDs"), such as field programmable gate arrays ("FPGAs"), programmable array logic ("PAL") devices, electrically programmable logic and memory devices, and standard cell-based devices, and application specific integrated circuits. Some other possibilities for implementation aspects include: memory devices, microcontrollers with memory (such as EEPROM), embedded microprocessors, firmware, software, etc. In addition, various aspects may be embodied in microprocessors with software-based circuit simulation, discrete logic (sequential and combinational logic), custom devices, fuzzy (neural) logic, quantum devices, and hybrid devices of any of the above device types. The underlying device technology may be provided in a variety of component types, such as metal oxide semiconductor field effect transistor ("MOSFET") technology such as complementary metal oxide semiconductor ("CMOS"), bipolar technology such as emitter coupled logic ("ECL"), polymer technology (e.g., silicon conjugated polymers and metal conjugated polymer-metal structures), analog and digital hybrids, and the like.

[0138] It should also be noted that the various logics and / or functions disclosed herein can be enabled using hardware, firmware, and / or any number of combinations of data and / or instructions embodied in various machine-readable media or computer-readable media, in terms of their behavior, register transfer, logic components, and / or other features. Computer-readable media that can embody such formatted data and / or instructions include, but are not limited to, various forms of non-volatile storage media (e.g., optical, magnetic, or semiconductor storage media), although again, transient media are not included. Unless the context clearly requires otherwise, throughout the description, the words "including", "comprising", "including", etc. should be understood as inclusive, rather than exclusive or exhaustive; that is, in a sense "including but not limited to". Words using the singular or plural also include the plural or singular, respectively. In addition, the words "herein", "below", "above", "below" and words of similar meaning refer to the present application as a whole, rather than to any particular part of the present application. When the word "or" is used to refer to a list of two or more items, the word covers all of the following interpretations of the word: any item in the list, all items in the list, and any combination of items in the list.

[0139] Although certain currently preferred embodiments of the present invention have been specifically described herein, it will be apparent to those skilled in the art that variations and modifications may be made to the various embodiments shown and described herein without departing from the spirit and scope of the present invention. Therefore, it is intended that the present invention be limited only to the extent required by applicable legal rules.

[0140] Although the foregoing has been referred to specific embodiments of the present disclosure, those skilled in the art will appreciate that changes may be made to the embodiments without departing from the principles and spirit of the present disclosure as defined by the appended claims.

Claims

1. A memory computing unit, comprising: A storage unit having data and complementary data; a read bit line and a complementary read bit line connected to the storage cell, the read bit line and the complementary read bit line providing read access to data and complementary data of the storage cell; an isolation circuit connected between the read bit line and the complementary read bit line and the storage unit, the isolation circuit isolating the storage unit from a signal on any one of the read bit line and the complementary read bit line, the isolation circuit being connected to a read word line and a complementary read word line; as well as Wherein, when the memory computing unit is connected to the same read bit line or the complementary read bit line together with another memory computing unit, each of the read bit line and the complementary read bit line is capable of performing a logic function.

2. The memory computing unit of claim 1, further comprising a write bit line connected to the memory computing unit, wherein data is written into the storage unit.

3. The memory computing unit of claim 2 further comprising a write port device that isolates the storage cell from the write bit line.

4. The memory computing unit of claim 2, capable of performing selective write operations.

5. The memory computing unit according to claim 1, wherein: The memory computing unit is a static random access memory unit.

6. The memory computing unit according to claim 5, wherein: The static random access memory unit is a three-port static random access memory unit.

7. The memory computing unit according to claim 1, wherein: The memory computing unit is a non-volatile memory.

8. The memory computing unit according to claim 7, wherein: The nonvolatile memory is one of a nonvolatile storage unit and a nonvolatile storage device.

9. The memory computing unit according to claim 1, wherein: The isolation circuit also includes a first set of transistors connected between the complementary read bit lines and a data storage point of the storage cell.

10. The memory computing unit according to claim 9, wherein: The first set of transistors also includes: a first transistor having a gate connected to the complementary read word line; and a second transistor having a gate connected to a data storage point of the storage unit.

11. The memory computing unit according to claim 9, wherein: The isolation circuit also includes a second set of transistors connected between the read bit line and a complementary data storage point of the storage cell.

12. The memory computing unit according to claim 11, wherein: The second set of transistors also includes: a third transistor having a gate connected to the read word line; and a fourth transistor having a gate connected to a complementary data storage point of the storage unit.

13. The memory computing unit according to claim 5, wherein: The storage unit also includes a first inverter and a second inverter cross-coupled to the first inverter.

14. A processing array comprising: a plurality of read bit lines; A memory cell array having a plurality of rows and columns of memory cells, each memory cell having a storage unit, the plurality of read bit lines being connected to the plurality of columns of memory cells in the memory cell array; The memory cell array has a first section and a second section, the first section includes at least two rows of memory cells, the second section includes one or more rows of memory cells, each column of memory cells in the first section is connected to at least two read bit lines, and each column of memory cells in the second section is connected to at least one read bit line; a plurality of bit line read / write circuits located between a first segment and a second segment in the memory cell array, each bit line read / write circuit having inputs from the at least two read bit lines of the first segment and inputs from the at least one read bit line of the second segment; as well as Wherein, when two or more memory cells are connected to the read bit line, each read bit line connected to the memory cells in the first segment is capable of performing a logic function.

15. The processing array of claim 14, wherein: A first input and a carry input of a full adder are stored along the same read bit line of the first segment, and wherein a second input of the full adder is stored along the same read bit line of the second segment.

16. The processing array of claim 15, wherein: The first input is a signal, the carry input is an inverted signal, and the second input is a signal.

17. The processing array of claim 16, wherein: Each bit line read / write circuit also includes: an OR gate that performs a logical OR on the at least two read bit lines of the first segment and generates an output; an inverter that inverts the second input of the full adder to generate an inverted second input; a first AND gate that performs a logical AND on the output of the OR gate and the inverted second input to generate a first output; and a second AND gate that performs a logical AND on the output of the OR gate and the second input of the full adder to generate a second output.

18. The processing array of claim 17, further comprising: a plurality of write bit lines, wherein each write bit line is connected to a selected column of memory cells and the bit line read / write circuit connected to the selected column of memory cells; and a plurality of complementary write bit lines, wherein each complementary write bit line is connected to the selected column memory cell and a bit line read / write circuit connected to the selected column memory cell.

19. The processing array of claim 18, wherein: The first output is connected to the write bit line, which is connected to the bit line read / write circuit, and the second output is connected to the complementary write bit line, which is connected to the bit line read / write circuit.

20. The processing array of claim 17, wherein: The at least two read bit lines of the first segment also include a first read bit line and a second read bit line, wherein the first read bit line carries a signal corresponding to the logical AND of the first input of the full adder and the inverted carry input of the full adder, and the second read bit line carries a signal corresponding to the logical AND of the inverted first input of the full adder and the carry input of the full adder.

21. The processing array of claim 16, wherein: Each bit line read / write circuit also includes: an XOR gate that performs a logical XOR on the at least two read bit lines of the first segment and generates an output; an inverter that inverts the second input of the full adder; a first AND gate that performs a logical AND on the output of the XOR gate and the inverted second input of the full adder to generate a first output; and a second AND gate that performs a logical AND on the output of the XOR gate and the second input of the full adder to generate a second output.

22. The processing array of claim 21 further comprising: a plurality of write bit lines, wherein each write bit line is connected to a selected column of memory cells and the bit line read / write circuit is connected to the selected column of memory cells; and a plurality of complementary write bit lines, wherein each complementary write bit line is connected to the selected column memory cell and the bit line read / write circuit connected to the selected column memory cell.

23. The processing array of claim 22, wherein: The first output is connected to the write bit line, which is connected to the bit line read / write circuit, and the second output is connected to the complementary write bit line, which is connected to the bit line read / write circuit.

24. The processing array of claim 21, wherein: The at least two read bit lines of the first segment also include a first read bit line and a second read bit line, wherein the first read bit line carries a signal corresponding to the first input of the full adder, and the second read bit line carries a signal corresponding to the logical negation of the inverted carry input of the full adder.

25. The processing array of claim 16, wherein: Each bit line read / write circuit also includes: an AND gate, which performs a logical AND on the at least two read bit lines of the first segment and the read bit line of the second segment and generates an output; an inverter, which inverts a write data signal; a first AND gate, which performs a logical AND on the output of the AND gate and the inverted write data signal to generate a first output; and a second AND gate, which performs a logical AND on the output of the AND gate and the write data signal to generate a second output.

26. The processing array of claim 25, wherein: A search operation is performed on the first segment and the second segment.

27. The processing array of claim 25, wherein: The first output is connected to a write bit line connected to the bit line read / write circuit, and the second output is connected to a complementary write bit line connected to the bit line read / write circuit.

28. The processing array of claim 16, wherein: Each bit line read / write circuit also includes: a logic gate that generates a selective write control signal for a full adder operation; a logic gate that generates a selective write control signal for a search operation; a first multiplexer that selects one of the selective write control signal for the full adder operation and the selective write control signal for the search operation; and a second multiplexer that generates write data for the selective write operation.

29. The processing array of claim 16, wherein: Each bit line read / write circuit also includes: a logic gate that generates a selective write control signal for a full adder operation; a logic gate that generates a selective write control signal for a search operation; a first multiplexer that selects one of the selective write control signal for the full adder operation, the selective write control signal for the search operation, and the write data control signal; and a second multiplexer that generates write data for the selective write operation.

30. The processing array of claim 16, wherein: Each bit line read / write circuit also includes: a logic gate that generates a selective write control signal for a full adder operation; a logic gate that generates a selective write control signal for a search operation; a first multiplexer that selects one of the selective write control signal for the full adder operation, the selective write control signal for the search operation, and the write data control signal; a second multiplexer that generates write data for the selective write operation; and a third multiplexer that selects one of the write data generated from the second multiplexer, the write data signal generated from a combined Boolean operation of the two read bit lines of the first segment and the read bit line of the second segment, and a write data signal outside the processing array.