In-memory computing device for performing signed MAC operations
By using symbol-dependent activation signals in the memory computing device to optimize the activation time of the memory unit, the problem of long refinement time of multiplication and accumulation operations in the prior art is solved, and the computing efficiency is improved.
Patent Information
- Application Number
- CN202410720512.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-06-03
- Filing Date
- 2024-06-05
- Publication Date
- 2024-12-06
AI Technical Summary
When existing memory computing devices perform signed multiplication and accumulation operations, the refinement time is long, resulting in a smaller number of operations that can be performed per second.
A in-memory computing device is designed, including multiple memory cell groups coupled to bit lines, each cell group is configured to store computational weights and provides symbol-dependent activation signals through row activation circuits, optimizing execution time of multiplication and accumulation operations .
通过优化激活信号的持续时间和结构设计,显著缩短了乘法和累加运算的细化时间,提高了每秒可执行的运算数量。
Smart Images

Figure CN119088340A_ABST
Abstract
Description
[0001] Priority claim
[0002] This application claims the benefit of priority from Italian Patent Application No. 102023000011343, filed on June 5, 2023, the contents of which are incorporated herein by reference in their entirety to the maximum extent permitted by law. Technical Field
[0003] Embodiments herein relate to an in-memory computing (IMC) device for performing multiply and accumulate (MAC) operations with low elaboration time. In addition, embodiments herein also relate to a corresponding method for controlling an IMC device. Background Art
[0004] As is known, in-memory computing (IMC) devices use a specific arrangement of memory cells of a memory array to perform analog refinement of data. For example, IMC devices are used to perform multiply and accumulate (MAC) operations, which are used, for example, to implement automatic learning algorithms such as neural networks.
[0005] The MAC operation provides the output vector Y = y 1 ,...,y M , as the input vector X = x 1 ,...,x N The result of multiplying the vector or matrix used to calculate the weights G, for example:
[0006] Right now:
[0007]
[0008] The IMC device will calculate the weight g ij Stored in cells of memory and perform multiplication and addition operations at the cell level.
[0009] In detail, for each value y of the output vector Y i , known IMC devices all generate currents indicative of corresponding MAC operations, i.e., And includes a readout circuit having a corresponding analog-to-digital converter (ADC) for discretizing the current.
[0010] IMC devices allow avoiding the need to transfer data back and forth between memory and refinement (eg, processing) units. Therefore, the performance of IMC devices is not limited by the bandwidth of data transfer between memory and refinement units, and have low energy consumption.
[0011] It is known that a signed MAC operation needs to be performed (ie, where the input value x 1, ..., x N and / or calculate the weight g ij The MAC operation can take on positive or negative values. Therefore, as a function of a particular combination of input values and calculation weights, the output value y 1 , ..., y M Positive or negative values can also be present.
[0012] It is recognized that known methods for performing signed MAC operations require long refinement times. Therefore, the number of signed MAC operations that can be performed per second is small.
[0013] There is a need in the art to overcome the above-mentioned shortcomings. Summary of the invention
[0014] In an embodiment, an in-memory computing (IMC) device performs a multiplication and accumulation (MAC) operation to provide at least one output signal in response to receiving an input signal indicating a plurality of input values, wherein each input value has a corresponding sign and a corresponding absolute value. The IMC device includes: a computation memory array including a plurality of memory cell groups coupled to bit lines, each memory cell group configured to store a corresponding computation weight and including a positive memory cell coupled to a corresponding positive word line and a negative memory cell coupled to a corresponding negative word line; and a row activation circuit configured to receive the input signal and, for each input value, provide a positive activation signal having a positive activation duration to the positive word line of the corresponding memory cell group and a negative activation signal having a negative activation duration to the negative word line of the corresponding memory cell group during an effective refinement interval, the sum of the positive activation duration and the negative activation duration being equal to the duration of the effective refinement interval, the effective refinement interval including a first refinement phase and a second refinement phase subsequent to the first refinement phase.
[0015] The row activation circuit is also configured to, for each input value: determine a positive theoretical duration of a positive activation signal and a negative theoretical duration of a negative activation signal, wherein the positive theoretical duration and the negative theoretical duration are each the algebraic sum of a reference value that does not depend on the input value and a variable value that varies according to the sign and absolute value of the input value, so that the sum of the corresponding positive theoretical duration and the corresponding negative theoretical duration is equal to the theoretical refinement duration; and during an effective refinement interval, first provide a positive activation signal and a negative activation signal whose theoretical duration is greater than or equal to the reference value, wherein the activation duration of one of the positive activation signal and the negative activation signal is equal to the sum of the effective reference value (which is equal to the reference value reduced by the reduction factor) and the variable value, and wherein the activation duration of the other of the positive activation signal and the negative activation signal is equal to the corresponding theoretical duration.
[0016] The first refinement phase of the effective refinement interval has a duration equal to the effective reference value.
[0017] The positive memory cells in the corresponding memory cell group are configured to be activated by the corresponding positive activation signal and have a positive cell current flow through them when activated, and the negative memory cells in the corresponding memory cell group are configured to be activated by the corresponding negative activation signal and have a negative cell current flow through them when activated. For each memory cell group, the positive cell current and the negative cell current are functions of the corresponding calculated weights stored by the memory cell group.
[0018] A bit line current flows through the bit line during an active refinement interval, the bit line current being a function of a sum of a positive cell current and a negative cell current of a group of memory cells coupled to the bit line. Column refinement circuitry coupled to the bit line samples the bit line current and provides at least one output signal in response thereto. The bit line current during the first refinement phase is sampled according to a reduction factor.
[0019] In an embodiment, a method for controlling an in-memory computing (IMC) device to perform a multiplication and accumulation (MAC) operation is presented. The method includes: providing a computational memory array, the computational memory array including a plurality of memory cell groups coupled to bit lines; storing a corresponding computational weight having a sign and an absolute value in each memory cell group, wherein each memory cell group includes a positive memory cell coupled to a corresponding positive word line and a negative memory cell coupled to a corresponding negative word line; activating the positive memory cell by a corresponding positive activation signal on the corresponding positive word line; activating the negative memory cell by a corresponding negative activation signal on the corresponding negative word line; and causing the corresponding positive memory cell of each memory cell group to flow a corresponding positive cell current in response to the corresponding positive activation signal; causing the corresponding negative memory cell of each memory cell group to flow a corresponding negative cell current in response to the corresponding negative activation signal.
[0020] For each memory cell group, the positive cell current and the negative cell current are a function of the corresponding calculated weights.
[0021] The method further includes, during an effective refinement interval, in response to an input signal indicating a plurality of input values (each input value having a corresponding sign and a corresponding absolute value), for each input value, providing a positive activation signal having a positive activation duration to a positive word line of a corresponding memory cell group and providing a negative activation signal having a negative activation duration to a negative word line of a corresponding memory cell group. Providing the positive activation signal and the negative activation signal includes: determining a positive theoretical duration of the positive activation signal and a negative theoretical duration of the negative activation signal; wherein the positive theoretical duration and the negative theoretical duration are each an algebraic sum of a reference value that does not depend on the input value and a variable value that varies according to the sign and the absolute value of the corresponding input value, so that the sum of the corresponding positive theoretical duration and the corresponding negative theoretical duration is equal to the theoretical refinement duration; and first providing the positive activation signal and the negative activation signal during the effective refinement interval, wherein one of the positive activation signal and the negative activation signal whose theoretical duration is greater than or equal to the reference value has an activation duration equal to the sum of the effective reference value (which is equal to the reference value reduced by the reduction factor) and the variable value, and the other of the positive activation signal and the negative activation signal has an activation duration equal to the corresponding theoretical duration.
[0022] The first refinement phase of the effective refinement interval has a duration equal to the effective reference value.
[0023] The method further includes causing a bit line current to flow through a bit line coupled to the group of memory cells during an active refinement interval, the bit line current being a function of a sum of a positive cell current and a negative cell current of the group of memory cells coupled to the bit line, and sampling the bit line current according to the reduction factor during a first refinement phase and providing at least one output signal in response thereto. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] For a better understanding of the invention, embodiments of the invention will now be described purely by way of non-limiting examples with reference to the accompanying drawings, in which:
[0025] Figure 1 A block diagram of an in-memory computing device is shown;
[0026] Figure 2 Shows Figure 1 A detailed circuit diagram of a memory cell group of the device;
[0027] Figure 3 Shows Figure 1 A block diagram of a row activation circuit of a device;
[0028] Figure 4 Shown by Figure 3 The circuit is implemented to control Figure 1 A flowchart of a method for activating a word line of a device;
[0029] Figure 5 Shows Figure 3 A detailed circuit diagram of a portion of the circuit;
[0030] Fig. 6A An example is shown by Figure 5 The waveform of the theoretical row activation signal determined by the circuit;
[0031] Figure 6B An example is shown in use from Fig. 6A The theoretical signal starts from Figure 5 The waveform of the valid row activation signal generated by the circuit;
[0032] Figure 7 Shows Figure 1 A block diagram of a digital detector of a device;
[0033] Figure 8 shows an example in use with Figure 1 a table of possible charge values associated with each memory cell group of the device;
[0034] Fig. 9 Shows Figure 1 Circuit diagram of the reference circuit of the equipment;
[0035] Fig.10 Shown is an example in use Figure 1 The waveform of the device;
[0036] Fig.11 Shows Figure 1 A detailed circuit diagram of a memory cell group of the device;
[0037] Fig.12 Shows Figure 1 A circuit diagram of a bias circuit of the device;
[0038] Fig.13 and Fig.14 Shows Figure 7 Detailed circuit diagram of the digital detector;
[0039] Fig.15 Shown is an example in use Fig.13 The waveform of the digital detector;
[0040] Fig.16 and Fig.17 Shows Figure 3 A detailed circuit diagram of the timer of the circuit; and
[0041] Fig.18 shows an example of an Figure 1 A table of possible charge values associated with each memory cell group of the device. DETAILED DESCRIPTION
[0042] Figure 1 An in-memory computing device (hereinafter referred to as an IMC device) 10 is shown, which includes a computing memory array (for simplicity, hereinafter also referred to as just a "memory array" or a "computing array") 12, a row activation unit or circuit 14, and an output refinement circuit, where the output refinement circuit includes a plurality of digital detectors 16 and a digital signal processor (DSP) 17.
[0043] Memory array 12 includes a plurality of memory cells 20 organized according to a matrix arrangement having M columns and K rows.
[0044] Next, the common memory cell in the plurality of memory cells 20 is denoted by reference numeral 20. i,k , M and k=1, . . . , K indicate columns and rows of universal memory cells in the memory array 12 , respectively.
[0045] In this embodiment, the memory cell 20 is of a non-volatile type.
[0046] The memory cells 20 are further organized to form a plurality of memory cell groups 22 (in Figure 1 Each group 22 is configured to store a corresponding calculated weight G ij .
[0047] In this embodiment, each calculated weight G ij Has sign and absolute value.
[0048] Calculate the weight G ij Can be used as weights for performing in-memory computations (particularly signed MAC operations).
[0049] In this embodiment, each memory cell group 22 is formed of a respective positive cell 23A and a respective negative cell 23B belonging to the plurality of memory cells 20 .
[0050] exist Figure 1 In the arrangement of FIG. 1 , the positive memory cells 23A and the negative memory cells 23B of each memory cell group 22 are arranged on the same column i and on two adjacent rows k, k+1 of the memory array 12 .
[0051] In practice, in this embodiment, the memory cell groups 22 form an array having M columns and N rows, where N=K / 2.
[0052] Hereinafter, a common memory cell group 22 among the plurality of memory cell groups 22 is denoted by reference numeral 22. ij , wherein the indexes i=1, ..., M and j=1, ..., N respectively indicate the general memory cell group 22 ijColumns and rows.
[0053] The memory cells 20 arranged in the same column of the memory array 12 are connected to each other via corresponding bit lines BL i , where i=1, . . . , M. The memory cells 20 arranged in the same row k of the memory array 12 are connected via corresponding word lines WL k are connected to each other, where k=1,…,K.
[0054] In detail, each memory cell group 22 i,j coupled to the positive word lines WL hereinafter respectively identified as j,+ and negative word line WL j,- Two adjacent word lines WL k , WL k+1 .
[0055] refer to Figure 1 The layout of the positive word line WL j,+ Coupled to the memory cell group 22 1,j ,...,twenty two M,j The positive cell 23A, while the negative word line WL j,- Coupled to the memory cell group 22 1,j ,...,twenty two M,j Negative unit 23B.
[0056] For example, the positive word line WL 1,+ Coupled to all memory cells 20 arranged in row k=1 and columns i=1 to i=M in the memory array 12. Negative word line WL 1,- Coupled to all memory cells 20 arranged in row k=2 and columns i=1 to i=M in the memory array 12 .
[0057] In practice, each memory cell group 22 i,j Coupled to the corresponding bit line BL i and two corresponding word lines WL j,+ , WL j,- .
[0058] The row activation unit 14 receives a plurality of input values x 1 , ..., x N The input vector X. The input value x 1 , ..., x N Each represents an input coefficient with sign and absolute value.
[0059] Input value x 1 , ..., x N Can be the input data of MAC operation.
[0060] For each input value x j(where j=1, ..., N), the row activation unit 14 is the corresponding positive word line WL j,+ Provide positive activation signal S j,+ and for the corresponding negative word line WL j,- Provide negative activation signal S j,- .
[0061] In practice, the row activation unit 14 sends a signal to the memory cell group 22. 1,j ,...,twenty two M,j The positive unit 23A provides a positive activation signal S j,+ , and to the memory cell group 22 1,j ,...,twenty two M,j The negative unit 23B provides a negative activation signal S j,- .
[0062] The row activation unit 14 is in the effective refinement interval T C Provide positive activation signal S during j,+ and negative activation signal S j,- , the effective refinement interval T C The row activation unit 14 refines the interval T* from the theoretical C Begin by determining, as described in detail below.
[0063] Positive activation signal S j,+ In the effective refinement interval T C The corresponding positive unit 23A is activated for the positive activation duration T j,+ .
[0064] Negative activation signal S j,- In the effective refinement interval T C The corresponding negative cell 23B is activated for the negative activation duration T j,- .
[0065] Positive activation duration T j,+ and negative activation duration T j,- is the corresponding input value x j is a function of the sign and absolute value of , as described in detail below.
[0066] In detail, the positive activation signal S j,+ and negative activation signal S j,- is a pulse, especially a rectangular pulse, the width (duration) of each pulse is equal to T j,+ and T j,- .
[0067] In this embodiment, the IMC device 10 further includes a bias circuit 18 that generates a bias voltage Vr and provides the bias voltage Vr to the bit line BL. 1 , ..., BLM .
[0068] In the remaining description, for simplicity, it is considered that the bias circuit 18 is for all bit lines BL 1 , ..., BL M The same bias voltage Vr is provided. However, depending on the specific application, the bias circuit 18 can start from the bias voltage Vr to provide the bit line BL 1 , ..., BL M Bias voltages different from each other are provided.
[0069] The digital detectors 16 are analog-to-digital converter (ADC) circuits, each of which is coupled to a corresponding bit line BL. i And each digital detector provides a corresponding charge signal q i , the charge signal q i Indicates if the refinement interval is equal to the theoretical refinement interval T* C Then it will flow through the corresponding bit line BL i The current I BL,i , as referenced below Figure 7 Described in detail.
[0070] In detail, each digital detector 16 can provide an effective refinement interval T C The slave current I BL,i The corresponding charge signal q at the beginning of the integration i .
[0071] For example, the digital detector 16 may be formed by an integrating and sampling circuit known per se.
[0072] The digital processor 17 processes the charge signal q 1 , ..., q M , and in response provides an output signal Y MAC , the output signal Y MAC Indicates the input vector X and the calculated weight G ij The result of the MAC operation between them.
[0073] Output signal Y MAC Indicates multiple output values y 1 , ..., y M , where each bit line BL i An output value.
[0074] As described in detail below, the digital processor 17 can convert the charge signal q i With the reference charge signal q ref compares and provides a corresponding output value y in response to the comparison i The output value y iThe sign and absolute value of the corresponding charge signal q i With the reference charge signal q ref Comparison between.
[0075] Each memory cell 20 includes a storage element 25 and a selection element 26 .
[0076] Each memory cell 20 i,k The storage element 25 is a variable resistance element which can be programmed to have a transconductance g ik , the transconductance g ik is during programming of the memory array 12 according to the corresponding cell group 22 i,j The calculation weight G ij selected, as discussed in detail below.
[0077] In particular, the memory element 25 may be based on a phase change material (PCM), such as a chalcogenide.
[0078] Phase change materials have at least two phases, such as an amorphous phase and a crystalline phase, where each phase may have one or more different resistivity values.
[0079] Phase change materials can be transformed from one phase to another via heat transfer (eg, using a pulse of electric current).
[0080] The resistance of each storage element 25 associated with the corresponding phase state is used to distinguish between two or more logic states of the corresponding memory cell 20 .
[0081] For example, an amorphous phase may have a higher resistance (and therefore a lower transconductance) than a crystalline phase.
[0082] For example, a logic state “0” or a reset state may be associated with an amorphous phase of the storage element 25 . A logic state “1” or a set state may be associated with a crystalline phase of the storage element 25 .
[0083] In this embodiment, memory cells 20 are multi-level cells; that is, they can be programmed to represent a number of logic states greater than two.
[0084] In other words, each memory cell 20 i,k can be programmed so that the corresponding transconductance g i,k A value selected from three or more values may be presented.
[0085] Purely as an example, in the following, consider each memory cell 20 i,k The transconductance g i,k Three different transconductance values g can be selected 1 , g 2 and g 3 One of the1 <g 2 <g 3 .
[0086] The storage element 25 has a corresponding bit line BL coupled to the i A first terminal of a node 28 of the reference potential node 20 and a second terminal coupled to a reference potential node (here ground 29) via a selection element 26.
[0087] The selection element 26 is a switch, such as a BJT, a diode or a MOS transistor (here, an NMOS transistor), which is arranged in series with the corresponding storage element 25 and whose switching is controlled by the corresponding word line WL j,+ or WL j,- The activation signal S j,+ or S j,- control.
[0088] In this embodiment, the NMOS transistor forming the select element 26 has a source terminal coupled (here directly connected) to ground 29; a drain terminal coupled (here directly connected) to the second terminal of the storage element 25; and a drain terminal coupled (here directly connected) to the corresponding word line WL j,+ or WL j,- of the gate terminal.
[0089] In practice, the storage element 25 and the selection element 26 form a current path of the corresponding memory cell 20 .
[0090] refer to Figure 2 The general memory cell group 22 shown in detail in i,j , the selection element 26 of the positive unit 23A responds to the corresponding positive activation signal S j,+ The corresponding current path is closed by receiving the positive unit current I + i,j can be transferred from the common node 28 to the ground 29 for a period depending on the duration of the positive activation T j,+ The time interval, in particular, is equal to the positive activation duration T j,+ .
[0091] The selection element 26 of the negative cell 23B responds to the corresponding negative activation signal S j,- The corresponding current path is closed by receiving the negative unit current I- i,j can be transferred from the common node 28 to the ground 29 for a period of time depending on the negative activation duration T j,- The time interval, in particular, is equal to the negative activation duration T j,- .
[0092] For each memory cell group 22 i,j, depending on the application, the negative cell current I - i,j and positive unit current I + i,j According to the corresponding calculation weight G ij Make your selection.
[0093] In this embodiment, the common memory cell group 22 i,j The positive cell 23A and the negative cell 23B are programmed so that the positive cell current I + i,j and negative cell current I - i,j The linear combination of the total current I tot .
[0094] In particular, in this embodiment, I + i,j +I - i,j =I tot .
[0095] The user may select the total current I according to the specific application of the IMC device 10 and / or the specific characteristics of the memory array 12. tot .
[0096] In this embodiment, for all memory cell groups 22, the total current I tot are all the same.
[0097] Universal memory cell group 22 i,j The positive cell 23A and the negative cell 23B are further programmed so that the positive cell current I + i,j With negative cell current I - i,j The difference between them is the corresponding calculated weight G i,j A function of the sign and absolute value of .
[0098] In detail, it can be based on the total current I tot The reference current I ref To select the positive unit current I + i,j and negative cell current I - i,j .
[0099] In this embodiment, the reference current I ref Equal to the total current I tot half of I ref =I tot / 2.
[0100] In more detail, the positive cell current I+ i,j and negative cell current I - i,j , so that the positive unit current I + i,j With reference current I ref The difference between them is the corresponding calculated weight G i,j A function of the sign and absolute value of .
[0101] Positive unit current I + i,j With reference current I ref The sign of the difference between them depends on the calculated weight G i,j symbol.
[0102] Positive unit current I + i,j With reference current I ref The absolute value of the difference between i,j The absolute value of .
[0103] In this example:
[0104] - If G i,j >0, then I + i,j >I ref ;
[0105] - If G i,j =0, then I + i,j =I ref ;as well as
[0106] - If G i,j <0, then I + i,j ref .
[0107] Positive unit current I + i,j With reference current I ref The absolute value of the difference between |I + i,j -I ref | is the calculation weight G i,j The absolute value of the value |G j,j |function.
[0108] Absolute value |I + i,j -I ref | can be used to calculate the weight G i,j The absolute value of |G j,j |, such as an increasing monotonic function.
[0109] Purely by way of explanatory numerical example, assuming that the current I + i,j ,I - i,j can each assume a value equal to 0 μA, 1 μA or 2 μA, i.e., I tot =2μA and I ref =I tot / 2=1μA, then calculate the weight G i,j It can be expressed as:
[0110] - If G i,j = +1, then I + i,j =2μA and I - i,j =0μA;
[0111] - If G i,j =0, then I + i,j =1μA and I - i,j = 1 μA; and
[0112] - If G i,j = -1, then I + i,j =0μA and I - i,j =2μA.
[0113] In practice, for each memory cell group 22 i,j , the corresponding positive unit current I + i,j and the corresponding negative unit current I - i,j The difference between + i,j -I - i,j The symbol can be the corresponding calculation weight G ij is a function of the sign of , not its absolute value. In other words, if G ij >0, then I + i,j -I - i,j >0, if G ij <0, then I + i,j -I - i,j <0, and if G ij =0, then I + i,j -I - i,j =0.
[0114] The corresponding positive unit current I + i,j and the corresponding negative unit current I - i,j The difference between + i,j -I - i,j The absolute value of can be the corresponding calculation weight G ij is a function of the absolute value of , not its sign.
[0115] In detail, in this embodiment, for each memory cell group 22 i,j , the corresponding positive unit current I + i,j and the corresponding negative unit current I - i,j The difference between + i,j -I - i,j is the corresponding calculation weight G ij A monotonic function of , especially an increasing monotonic function.
[0116] The IMC device 10 may also include interface circuitry 30, including row decoding and selection circuitry, column decoding and selection circuitry, and read / write circuitry, which are useful and known per se for the operation of the IMC device 10. For example, the read / write circuitry may be used to program the transconductance value of the memory cell 20.
[0117] refer to Figure 3 , the row activation unit 14 includes a plurality of inputs to the time converter 46, each input value x 1 , ..., x N There is an input to the time converter 46, and the timing circuit, here the timing circuit includes a global timer or counter 45, an end comparator 47, an intermediate comparator 48 and an interval reduction circuit 49.
[0118] The timer 45 provides a timing signal TM, which is defined at the start time t start and the end time t end The effective refinement interval T C (For example, Figure 6B and Fig.10 ).
[0119] Effective refinement interval T C With duration T E1 The first effective refinement stage T E1 and has duration T E2The second effective refinement stage T E2 Formed so that T E1 +T E2 =T C .
[0120] In particular, in this embodiment, T E1 <T E2。
[0121] The second effective refinement stage T E2 Following the first effective refinement stage T E1 Afterwards, specifically, especially directly continuous therewith.
[0122] The row activation unit 14 receives (for example, from an external user of the IMC device 10) a determination start time t start And thus determine the effective refinement interval T C For example, the calculation start signal START_CNT can be a step logic signal, and its rising edge determines the starting time t start ( Fig.10 ).
[0123] The timing signal TM can be an analog signal or a digital signal. Here, it is an L-bit digital signal, which is also referred to as the timing signal TM in the following text and figures. <l:1>, which increases with time.
[0124] At the beginning of a new refinement operation, that is, at the starting time t start , the timer 45 can reset the timer signal TM to a starting value, for example, to zero.
[0125] The timer 45 increments the timing signal TM, for example, in a monotonic manner, in particular a strictly monotonic manner, with time increasing from the time t start Starting from the starting value at time t end The ending value at .
[0126] Timer 45 updates at a frequency f u Incremental timing signal TM, the update frequency f u It depends on the specific implementation of the timer 45 and can be selected by the user of the IMC device 10 according to the specific application and the specific implementation of the timer 45.
[0127] In practice, the timer signal TM is a counter signal.
[0128] The interval reduction circuit 49 receives: a refinement duration signal T_C, which indicates a theoretical refinement interval T* C , which has a theoretical refinement duration (again denoted by T* C and a reduction signal R_F indicating a theoretical refinement duration T* C The reduction factor RF.
[0129] Theoretical refinement interval T* C With duration T* E1 The first theoretical refinement stage T* E1 and has duration T* E2 The second theoretical refinement stage T* E2 Formed so that T* E1 +T* E2 =T* C .
[0130] In particular, in this embodiment, T* E1 =T* E2 =T* C / 2.
[0131] Theoretical refinement duration T* C It can be selected by the user according to the specific application; for example, it can be on the order of hundreds of nanoseconds or less than about 100 nanoseconds.
[0132] The reduction factor RF may be selected by the user depending on the specific application; for example, it may be an integer factor, in particular 2 or 4.
[0133] Effective refinement duration T C is the theoretical refinement duration T* C and a function of the reduction factor RF, as described in detail below.
[0134] Theoretical refinement duration T* C and the reduction factor RF can be used by the user of the IMC device 10 in the effective refinement interval T C The beginning of start ), for example, in a calibration or initialization step of the IMC device 10. Therefore, the effective refinement interval T C It is also fixed.
[0135] The interval reduction circuit 49 provides a maximum count signal MAX_CNT and an intermediate count signal T_MID starting from the thinning duration signal T_C and the reduction signal R_F.
[0136] In detail, in this embodiment, the reduction signal R_F indicates the first theoretical refinement stage T* E1 reduction.
[0137] In practice, the reduction circuit 49 refines the duration T* from the theoretical one according to the reduction factor RF. C Start to determine the effective refinement interval T C Duration T C .
[0138] In this embodiment, T E1 =T* E1 / RF=T* C / 2 / RF, and T E2 =T* E2 =T C * / 2, that is, T C =T* E1 / RF+T* E2 .
[0139] The maximum count signal MAX_CNT indicates the effective refinement interval T C The end time t end , that is, the effective refinement interval T C Duration T C .
[0140] The intermediate count signal T_MID indicates the effective refinement interval T C The separation time t M , so that t M -t start =T E1 .
[0141] The intermediate comparator 48 compares the timing signal TM with the intermediate count signal T_MID and provides a signal T_ME in response, such as a step logic signal ( Fig.10 ).
[0142] When the timing signal TM becomes equal to the intermediate count signal T_MID (separation time t M ), the signal T_ME switches, in particular switches to a high logic value here, thereby marking the first refinement stage T E1 The end and second refinement stage T E2 The beginning of Figure 6B and Fig.10 ).
[0143] The end comparator 47 compares the timing signal TM with the maximum count signal MAX_CNT and in response provides a refinement end signal STOP_CNT, such as a step logic signal.
[0144] When the timing signal TM becomes equal to the maximum count signal MAX_CNT, the count end signal STOP_CNT switches, in particular switches to a high logic value, thereby determining the end time t end ( Fig.10 ).
[0145] For example, the end comparator 47 and the middle comparator 48 may be bit verification circuits based on exclusive OR (XOR) logic gates.
[0146] The input to the time converter 46 each receives a corresponding input value x j And in the effective refinement interval T C Provided as the corresponding input value x j The positive activation signal S of the function j,+ and negative activation signal S j,- .
[0147] In detail, Figure 4 As shown in the flowchart of FIG. 1 , each input to the time converter 46 first determines the corresponding positive theoretical duration T* in step S01. j,+ and the corresponding negative theoretical duration T* j,- , so that the positive theoretical duration T* j,+ With negative theoretical duration T* j,- The sum is equal to the theoretical refinement duration T* C , that is, T* j,+ +T* j,- =T* C ; and make the positive theoretical duration T* j,+ With negative theoretical duration T j,- The difference between the corresponding input values x j A function of the sign and absolute value of .
[0148] Specifically, the positive theoretical duration T* j,+ and negative theoretical duration T* j,- is selected based on a reference duration that is less than the theoretical refinement duration T* C and here is equal to the first refinement stage T* E1 The duration (i.e., T* C / 2), so that the positive theoretical duration T* j,+ With reference duration T* E1 The difference between the input value x j The function f(x j ).
[0149] In practice, T* j,+ =T* E1 +f(x j ) and T* j,- =T* E1 -f(x j ).
[0150] In other words, the positive theoretical duration T* j,+ and negative theoretical duration T* j,- is constant and does not depend on the corresponding input value x j Reference value (T* E1 , which is equal to T* C / 2) and the variable value that depends on the corresponding input value (function f(x j ))’s algebraic sum is given.
[0151] Function f(x j ) can depend on the input value x j The sign and / or absolute value of
[0152] Function f(x j ) can depend on the input value x j The sign and / or absolute value of
[0153] In particular, the function f(x j ) can depend only on the input value x j The symbol of the function f(x j ) can depend only on the input value x j The absolute value of .
[0154] According to one embodiment:
[0155] -If x j = 0, then f(x j )=0, that is, T* j,+ =T* j,- =T* E1 ;
[0156] -If x j >0, then f(x j )>0, that is, T* j,+ >T* E1 And T *j,- <T* E1 ;as well as
[0157] -If x j <0, then f(x j )<0, that is, T* j,+ <T* E1 And T* j,- >T* E1。
[0158] In detail, the function f(x j ) can be the input value x j A monotonic function of , such as an increasing monotonic function.
[0159] Absolute value |T* j,+ -T* E1 | can be combined with the input value x j The absolute value of |x j |Has linear or nonlinear correlation.
[0160] In this embodiment, the absolute value |T* j,+ -T* E1 | has a linear dependence, especially with the input value x j The absolute value of |x j | is proportional to, that is, |T* j,+ -T* E1 |=p·|x j |.
[0161] In addition, in this embodiment, the input value x j is positive and the input value x j When |T* is negative j,+ -T* E1 |with|x j |The scaling factor p is the same between them. However, |T* j,+ -T* E1 |with|x j| The relationship between x j >0, then the proportionality factor p + , and in x j <0 has a proportional factor p - , where p + With p - different.
[0162] Purely as an example, consider the theoretical refinement duration T* c = 128ns (and therefore, T E1 =T E2 =T* c / 2=64ns) and considering the input value x j Can be between -64 and +64:
[0163] -If x j = -64, then T* j,+ =0ns and T* j,- =128ns, that is, T* j,+ -T* c / 2 = -64;
[0164] -If x j = -32, then T* j,+ =32ns and T* j,- =96ns, that is, T* j,+ -T* c / 2 = -32;
[0165] -If x j = 0, then T* j,+ = 64ns and T* j,- =64ns, that is, T* j,+ -T* c / 2=0;
[0166] -If x j = +32, then T* j,+ =96ns and T* j,- =32ns, that is, T* j,+ -T* c / 2 = 32; and
[0167] -If x j = +64, then T* j,+ =128ns and T* j,- =0ns, that is, T* j,+ -T* c / 2=64.
[0168] Then, in step S02, the positive theoretical duration T* is input to the time converter 46. j,+ With negative theoretical duration T* j,- Make a comparison.
[0169] If the theoretical duration T* j,+ Greater than or equal to negative theoretical duration T* j,- (Branch "Yes" at the output of step S02), then in step S03, by reducing the positive theoretical duration T* according to the reduction factor RF j,+ To obtain the positive refinement duration T j,+ For example, in this embodiment, T j,+ =T* E1 / RF+f(x j ). Negative refinement duration T j,- Keep equal to the negative theoretical duration T* j,- That is, T j,- =T* j,- =T* E1 -f(x j ).
[0170] In this case, step S04, in the effective refinement interval T C During this period, the positive word line WL j,+ On the negative word line WL j,- was activated before.
[0171] If the theoretical duration T* j,+ Less than negative theoretical duration T* j,- (Branch “No” at the output of step S02), then in step S05, the negative theoretical duration T* is reduced by the reduction factor RF j,- To obtain the negative refinement duration T j,- For example, in this embodiment, T j,- =T* E1 / RF-f(x j ). Positive refinement duration T j,+ Keep equal to the positive theoretical duration T* j,+ That is, T j,+ =T* j,+ =T* E1 +f(x j ).
[0172] In this case, step S06, in the effective refinement interval T C During this period, the negative word line WL j,- On the positive line WL j,+ was activated before.
[0173] In other words, in the effective refinement interval T C During the positive activation signal S j,+ and negative activation signal S j,- In use, provided that its theoretical duration is greater than or equal to the reference value (here T* C / 2) is used as the first activation signal.
[0174] In addition, the activation duration of the first activation signal is determined by the effective reference value (here T* C / 2 / RF), which is equal to the sum of the variable values (f(x 1 )、f(x 2 )) and the reference value (here T*) reduced by the reduction factor RF C / 2). The activation duration of the other activation signal instead remains equal to the corresponding setpoint duration.
[0175] The following will refer to Figure 5 The embodiment of the general input to time converter 46 is described in detail.
[0176] The input to the time converter 46 includes a register 50 which receives an input value x j and provide an indication of the input value x j The sign (+ or -) of the input signal SGN and the input value x j The absolute value of the absolute value of the signal MOD. For example, if x j ≥0, then the sign signal SGN may assume the value "0", and if x j <0, then the value "1" is presented.
[0177] Logic block 51 receives the sign signal SGN, the absolute value signal MOD and the reduction signal R_F and in response provides a signal WL_DATA (here of digital type) configured to be proportional to the input value x j The sign and absolute value of the positive activation duration T j,+ and negative activation duration T j,- , as referenced Figure 4 discussed.
[0178] In practice, register 50 and logic block 51 are configured to implement reference Figure 4 The contents discussed at steps S01, S02, S03 and S05 of the flowchart.
[0179] The signal WL_DATA indicates that the timing signal TM is at time t start With t end One of the values presented between .
[0180] For example, the value indicated by signal WL_DATA may be obtained from a specific conversion table (or lookup table) implemented by logic block 51 and register 50 .
[0181] In detail, the switching signal WL_DATA indicates that the switching signal WL_DATA is arranged in the effective refinement interval T C Within (i.e., at the start time t start and the end time t end The switching time t s,j .
[0182] For example, in the following, for simplicity and without loss of generality, consider t start = 0. Therefore t end =T C .
[0183] In this embodiment, the logic block 51 is configured so that the signal WL_DATA indicates the positive theoretical duration T*. j,+ and negative theoretical duration T* j,- The switching time t associated with the larger of s,j .
[0184] For example, if x j >0, then the switching time t s,j According to the positive theoretical duration T* j,+ To choose, so that t s,j -t start =T j,+ .
[0185] Comparator 52 compares signal WL_DATA with timing signal TM and provides switching signal SWITCH_WL in response. Comparator 52 may be a bit verification circuit based on an XOR logic gate, for example.
[0186] For example, the switching signal SWITCH_WL may have a high logic value as long as the timing signal TM is different from the signal WL_DATA. s,j ), the comparator 52 switches the switching signal SWITCH_WL.
[0187] The set-reset latch 53 receives the switching signal SWITCH_WL at the set input terminal S and receives the count end signal STOP_CNT at the reset input terminal R.
[0188] The set-reset latch 54 receives the count start signal START_CNT at the set input terminal S and the count end signal STOP_CNT at the reset input terminal R, and provides a row enable signal EN_WL at the Q output terminal.
[0189] In practice, the row enable signal EN_WL can be used in the effective refinement interval T C Period (i.e., from the start time t start Until the end time t end ) has a high logic value.
[0190] XOR logic gate 55 has a first input terminal coupled to the Q output terminal of latch 53, and a second input terminal receiving thereon the sign signal SGN provided by register 50. XOR logic gate 55 provides a row selection signal SEL_WL at an output terminal.
[0191] The first AND logic gate 56A receives the selection signal SEL_WL and the enable signal EN_WL at the input terminal, and is coupled to the first row driver 58A (eg, a buffer) at the output terminal. The first row driver 58A outputs a signal to the corresponding word line WL. j,+ Provide positive activation signal S j,+ .
[0192] Inverter 57 receives the selection signal SEL_WL and provides a negated selection signal SEL_WL_N.
[0193] The second AND logic gate 56B receives the negated selection signal SEL_WL_N and the enable signal EN_WL at the input terminal, and is coupled to the second row driver 58B (eg, a buffer) at the output terminal. The second row driver 58B supplies the corresponding word line WL j,- Provide negative activation signal S j,- .
[0194] In practice, the effective refinement interval T C During the period, first (i.e., from the starting time t start Until the switching time t s,j ) asserts the activation signal associated with the longer theoretical duration; subsequently (i.e., from the switching instant t s,j Until the end time t end ) asserts another activation signal.
[0195] Purely as an example, Fig. 6A and Figure 6B shows that with the first input value x 1 The associated theoretical duration T* 1,+ , T* 1,- and, respectively, the effective duration T 1,+ 、T 1,- , and the second input value x 2 The associated theoretical duration T* 2,+ , T* 2,- and, respectively, the effective duration T 2,+ 、T 2,- , where x 1 >0, x 2 <0 and |x 2 |<|x 1 | in the example.
[0196] exist Fig. 6A , with the first and second input values x 1 、x 2 The associated theoretical activation signals are indicated by asterisks to distinguish them from Figure 6B The word lines are distinguished by the valid activation signal generated by the input to the time converter 46 for activating the corresponding word line.
[0197] refer to Fig. 6A , theoretical refinement interval T* C The first refinement stage T* E1 With the second refinement stage T* E2 have the same duration, i.e., T* E1 =T* E2 =T* M =T* C / 2 . Fig. 6A The theoretical separation time t* is identified by the dotted line in M Indicates the theoretical refinement interval T* C The first stage T* E1 With the second stage T* E2 The separation between.
[0198] Due to x 1 >0, so the positive theoretical duration T* 1,+ Greater than reference duration T* C / 2(T* 1,+ -T* C / 2>0).
[0199] In addition, due to T* 1,+ +T* 1,- =T* C , so the positive theoretical duration T* 1,+ Greater than negative theoretical duration T* 1,- In practice, T* 1,+ -T* 1,- >0.
[0200] Due to x 2 <0, so the positive theoretical duration T* 2,+ Less than the reference duration T C / 2(T* 2,+ -T* C / 2<0).
[0201] In addition, due to T* 2,+ +T* 2,- =T* C , so the positive theoretical duration T* 2,+ Less than negative theoretical duration T* 2,- In practice, T* 2,+ -T* 2,- <0.
[0202] exist Fig. 6A In the example, since |x 2 |<|x 1 |, hence |T* 2,- -T* C / 2|<|T* 1,+ -T* C / 2|, that is, T* 2,- <T* 1,+ .
[0203] refer to Figure 6B , effective refinement interval T C The first refinement stage T E1 With the second refinement stage T E2 have different durations from each other, especially T E1 <T E2 , for example, here RF = 2, that is, T E1 =T E2 / 2.
[0204] For the first input value x 1 , due to T 1,+ >T 1,- , so the positive activation signal S 1,+ The negative activation signal S 1,- Therefore, the positive activation duration T 1,+ Less than the positive theoretical duration T* 1,+ , especially here T 1,+ =T C / 2 / 2+f(x 1 ), and the negative activation duration T 1,- Equal to the negative theoretical duration T* 1,- .
[0205] For the second input value x 2 , due to T 2,+ <T 2,- , so the negative activation signal S 2,- In the positive activation signal S 2,+ Therefore, the negative activation duration T 2,- Less than negative theoretical duration T* 2,- , especially here T 2,- =T C / 2 / 2-f(x 2 ), while the positive activation duration T 2,+ Equal to the positive theoretical duration T* 2,+ .
[0206] refer to Figure 7 , and the bit line BL i The associated digital detector 16 comprises a charge packet detector 60 and a charge packet counter formed by a multiplexer circuit 61 and a counter circuit 62 .
[0207] The charge packet detector 60 is coupled to the bit line BL i , in the effective refinement interval T C During the period, the bit line current I BL,i The integration is performed and, as a response, when the effective refinement interval T C During the period, the bit line BL i The charge flowing in exceeds the threshold charge Q th When a packet detection signal CP is provided.
[0208] In practice, the packet detection signal CP indicates the current from the bit line I BL,i Start measuring the charge.
[0209] In this embodiment, the counter circuit 62 is formed by a series of flip-flops, here ten flip-flops 64.0, ..., 64.9, the charge signal q i There is one flip-flop for each bit of , and these flip-flops are cascaded to each other. The multiplexer circuit 61 provides a packet detection signal CP at the input of one of the flip-flops 64.0, ..., 64.9 according to the reduction signal R_F and the signal T_ME.
[0210] Flip-flops 64.0, ..., 64.9 each have a clock input (CK input), a data input (D input), a Q output and Output.
[0211] Each flip-flop 64.0, ..., 64.9 The outputs are coupled to the corresponding D inputs. The Q output of each flip-flop 64.0, ..., 64.9 constitutes the charge signal q i The corresponding bits qi(0),...,q i (9).
[0212] The CK input of flip-flops 64.1, ..., 64.9 is coupled to the CK input of the previous flip-flop in the sequence of flip-flops 64.0, ..., 64.9. Output.
[0213] In the effective refinement interval T C The first refinement stage T E1 During this time, the multiplexer circuit 61 provides the packet detection signal CP to the CK input of one of the flip-flops 64.0, ..., 64.9 according to the reduction factor RF. For example, in the illustrated embodiment, if RF=2, the multiplexer circuit 61 provides the packet detection signal CP to the CK input of the second flip-flop 64.1.
[0214] In the effective refinement interval T C The second refinement stage T E2 During this time, the multiplexer circuit 61 provides the packet detection signal CP to the CK input of the first flip-flop 64.0.
[0215] In other words, in the effective refinement interval T C The first refinement stage T E1 During this period, the digital detector 16 converts the charge signal q i Multiply by a factor equal to the factor used as input to the time converter 46 to reduce the theoretical refinement interval T* C The first refinement stage T* E1 duration.
[0216] Therefore, the charge signal q i Indicates the theoretical refinement interval T* C During the period, the bit line BL i The charge flowing in exceeds the threshold charge Q th The number of times, the threshold charge Q th At each activation signal S j,+ , S j,- The activation duration T j,+ 、T j,- Equal to the corresponding theoretical duration T* j,+ , T* j,- In the case of i The number of charge packets flowing in the
[0217] In use, the IMC device 10 may be used to provide an output vector Y MAC =y 1 , ..., y M , which indicates that the input vector X = x 1 , ..., x N and the weight G is calculated by ij MAC operation between the matrices formed.
[0218] As reference Figure 2 The commonly used unit group 22 i,j The positive unit 23A is in the effective refinement interval T C During this period, a cell current is absorbed, here from the corresponding node 28 to ground 29, which depends on the transconductance g of the corresponding storage element 25. i,k and the activation duration of the corresponding selection element 26, i.e., depending on the positive activation signal S j,+ The positive activation duration T j,+ .
[0219] In detail, in the effective refinement interval T C During this period, the current flowing through the positive cell 23A has a current equal to the positive cell current I + i,j The intensity of the intensity depends on the corresponding transconductance value g i,k , and has a positive activation duration T j,+ time duration.
[0220] Universal unit group 22 i,j The negative cell 23B is in the effective refinement interval T C During this period, a cell current is absorbed, here from the corresponding node 28 to ground 29, which depends on the transconductance g of the corresponding storage element 25. i,k+1 and the activation duration of the corresponding selection element 26, i.e., depending on the negative activation signal S j,- The negative activation duration T j,- .
[0221] In detail, in the effective refinement interval T C During this period, the current flowing through the negative cell 23B has a current equal to the negative cell current I - i,j The intensity of the intensity depends on the corresponding transconductance value g i,k+1 , and has a negative activation duration T j,- time duration.
[0222] In this embodiment, the bias voltage Vr is applied to all bit lines BL. 1 , ..., BL M are the same; therefore, the current intensity I + i,j and I - i,j It depends only on the transconductance of the positive memory cell 23A and the negative memory cell 23B, respectively.
[0223] In the effective refinement interval T C During this period, reference unit group 22 i,j , positive unit 23A has a charge displacement Q + i,j Contributes to the charge displacement Q + i,j is the current intensity I + i,j With positive activation duration T j,+ The negative unit 23B is a function of the product of - i,j Contributes to the charge displacement Q - i,j is the current intensity I - i,j With negative activation duration T j,- The function of the product of .
[0224] In practice, the effective refinement interval T C During this period, each unit group 22 i,j Both charge displacement Q i,j =I + i,j ·T j,+ +I - i,j ·T j,- Have contributed.
[0225] Since the positive activation duration T j,+ and negative activation duration T j,- Depends on the corresponding input value x j , and the current intensity I + i,j ,I - i,j Depends on the calculation weight G ij , so with unit group 22 i,j The associated charge Q i,j is the product G ij ·x j function.
[0226] Purely as an example, Figure 8 The table provides an example where the theoretical refinement duration is T* C =128ns, input value x j can assume values between -64 and +64, and the current intensity I + i,j ,I - i,j It can be 0μA, 1μA or 2μA.
[0227] As can be seen from the table in FIG6 , if the theoretical duration T* j,+ =T* j,- =T* ref =T* C / 2 and / or I + i,j =I - i,j =I ref =I tot / 2, then unit group 22 i,j In the theoretical refinement interval T* C The displacement of the theoretical charge at the end is Q* i,j Contribution, which is equal to the reference value Q ref , in the example shown, the reference value Q ref Equal to 128.
[0228] In practice, when the corresponding activation duration T j,+ 、T j,- Equal to the reference duration (here T* C / 2), and / or when the unit current I + i,j ,I - i,j Equal to the reference current (here I tot / 2), unit group 22 i,j The corresponding bit line BL i The contribution is equal to the reference value Q ref The charge displacement.
[0229] In other words, if x j = 0 and / or if G ij =0, then unit group 22 i,j Equalizing contribution to Q ref The charge displacement.
[0230] In x j >0 and G i,j >0, or x j <0 and G i,j <0, Q* i,j >Q ref .
[0231] In x j >0 and G i,j <0, or x j <0 and G i,j >0, Q* i,j ref .
[0232] In practice, by comparing the charge Q* i,j With reference charge Q ref , we can determine the operation G ij ·x j symbol.
[0233] In addition, the charge Q* i,j With reference charge Q ref The absolute value of the difference between the indicated operation G ij ·x j The absolute value of .
[0234] Indeed, by considering, for example, the first row of the table, where G i,j =1, as x j The absolute value of the charge Q* increases i,j presents a higher value.
[0235] In detail, as the absolute value |x j |Increase, absolute value of difference|Q* i,j -Q ref |Increase, especially here in a proportional manner.
[0236] In addition, the same charge Q* i,j Value and sum operation G ij ·x j The same result of G ij and x j ) are associated with different combinations of ij = -1 and x j = -32, or G ij = +1 and x j =+32).
[0237] For each bit line BL i , the corresponding bit line current I BL,i By coupling to the bit line BL i The sum of the currents of all cells 20 is given by C At the end, it flows into the bit line BL i The total charge Q tot,i Equal to the coupling to the bit line BL i All unit groups 22 i,1 ,...,twenty two i,N The charge contribution Q i,1 , ..., Q i,N the sum; that is,
[0238] In practice, the total charge Q tot,i Indicative operation G i1 ·x 1 +G i2 ·x 2 +...+G iN ·x N .
[0239] Therefore, with the bit line BL i Associated is the total reference charge Q tot,i,ref , which is equal to the reference value Q associated with each cell group 22 ref In other words, Q tot,i,ref =N·Q ref , where N is the bit line BL coupled to i The number of unit groups 22.
[0240] As reference Figure 3-5 As discussed, for each input value x j , depending on x j The sign of the corresponding positive activation duration T j,+ or the corresponding negative activation duration T j,- is smaller than the corresponding theoretical duration, since the duration T* of the first refinement stage E1 (equal to the reference duration T* C / 2) is reduced by the reduction factor RF. For example, if x j >0, then T j,+ =T* C / 2 / RF+f(x j ). Therefore, in the effective refinement interval T C The first refinement stage T E1 During this period, the memory cell group 22 i,j The effective charge displaced is Q i,j is less than the positive activation signal S j,+ With a theoretical duration T* j,+ The activation duration T j,+ Case memory cell group 22 i,j The theoretical charge Q* of the displacement i,j .
[0241] However, in the first refinement stage T E1 During this time, the digital detector 16 multiplies the detected charge by a factor RF. This enables compensation of the first refinement stage T E1 With the theoretical duration T* E1 Therefore, the charge signal q provided by the digital detector 16 is i Indicates if the positive activation signal S 1,+ , ..., S N,+ Each has a theoretical duration T* j,+ The activation duration T j,+ Then the bit line BL i The charge flowing in.
[0242] DSP 17 converts the charge signal q i With the reference charge signal q ref For comparison, the reference charge signal q ref Indicates the total reference charge Q tot,i,ref .
[0243] In response to q i With q ref By comparing the MAC operation G i1 ·x 1 +G i2 ·x 2 +...+G iN ·x N The sign and absolute value of the result of the above operation are determined, and a corresponding output signal y indicating the result is provided at the output terminal. i .
[0244] In detail, the DSP 17 may i With q ref The output signal y is determined by the difference between i The sign and absolute value of .
[0245] Output signal y i The sign of the difference q i -q ref Starting from the previous discussion, in this embodiment, the following conclusions are drawn:
[0246] -If q i =q ref , then y i =0;
[0247] -If q i >q ref , then y i >0; and
[0248] -If q i ref , then y i <0.
[0249] Based on the input value x 1 , ..., x N Mapped to the corresponding activation duration T 1,+ , ..., T N,+ and T 1,- , ..., T N,- and for calculating the weight G i1 , ..., G iN Mapped to the corresponding current value I + i,1 , ..., i + i,N and I - i,1 , ..., i - i,N The specific transfer function of i The absolute value of the difference q i -q ref The absolute value of .
[0250] For example, the DSP 17 may use a specific conversion table previously stored in the IMC device 10 to convert the difference q i -q ref The absolute value of the output signal y i a specific value.
[0251] In practice, the IMC device 10 allows for use for a duration shorter than the theoretical interval T* C The effective refinement interval T C To obtain the instruction MAC operation G i1 ·x 1 +G i2 ·x 2 +...+G iN ·x N Thus, the IMC device allows to obtain a high number of operations per second, which is higher than that obtained using the theoretical refinement interval T* C The corresponding number of IMC devices.
[0252] For example, consider an example where the duration of the theoretical refinement interval is T* C =128ns, where T* E1 =T* E2 =64ns, and where the reduction factor RF = 4, the duration T is obtained C = 80ns effective refinement interval, where T E1 =T* E1 / 4=16ns and T E2 =T* E2 =64ns. Therefore, the effective refinement interval is approximately 37% shorter than the theoretical refinement interval. Therefore, the number of operations that can be performed per second by the IMC device 10 is 60% higher than the case where the effective refinement interval is equal to the theoretical refinement interval.
[0253] For each memory cell group 22 i,j , total current I tot can be a linear combination, especially the positive unit current I + i,j and negative cell current I - i,j The sum of the positive unit current I + i,j With negative cell current I - i,j The difference between them can be used to calculate the weight G ij The fact that each memory cell group 22 i,j Able to map the corresponding stored calculation weights G ij The sign and absolute value of .
[0254] Furthermore, for each input value x j , respectively from the positive theoretical duration T* j,+ and negative theoretical duration T* j,- Start to determine the corresponding positive activation duration T j,+ and negative activation duration T j,- , which makes it possible to map the corresponding input value x j Both the sign and the absolute value of .
[0255] In practice, the IMC device 10 allows for a single effective refinement interval T C Get the instruction MAC operation The sign and absolute value of the result, where the positive word line WL 1,+ , ..., W.L. N,+ and negative word line WL 1,- , ..., W.L. N,- All are activated.
[0256] In other words, the instruction operation The sign and absolute value of the output signal y i can be obtained in a single MAC refinement operation of the IMC device 10, that is, at the starting value (at time t start ) and the final value (at time t end ) only one sweep of the timing signal TM is used between the two.
[0257] Therefore, with two different effective refinement intervals T C Compared with the solution that performs positive refinement and negative refinement in , the IMC device 10 can perform signed MAC operations per second. The number of trillion operations per second (TOPS) can be doubled.
[0258] In particular, Figure 5 The embodiment of the input to the time converter 46 allows each bit line BL i The current is distributed to the effective refinement interval T C Two different subintervals (T j,+ and T j,- ). This makes it possible to simplify the design of the digitizer 16 and improve its accuracy.
[0259] According to one embodiment, the IMC device 10 may include a reference circuit 70 ( Fig. 9 ), the reference circuit 70 is configured to generate a reference charge signal q ref DSP 17 uses the reference charge signal q ref To determine the output signal y i The sign and absolute value of .
[0260] The reference circuit 70 includes a reference array 71 formed by one or more reference memory cells 72 and at least one digitizer (also indicated here by 16).
[0261] Memory unit 72 can be connected to the reference Figure 1 The memory cell 20 depicted is identical; namely, formed by a storage element 25 and a selection element 26 .
[0262] In the illustrated embodiment, the reference array 71 is composed of reference bit lines BL coupled to K memory cells 72. ref form.
[0263] The memory cells 72 are organized so as to form N reference memory cell groups 74, such as for Figure 1 The memory cell group 22 is discussed.
[0264] The reference memory cell groups 74 are also each denoted by the reference numeral 74 ref,j Specify, where j=1,...,N.
[0265] Each reference memory cell group 74 ref,j It is formed by positive cell 72A and negative cell 72B, which are respectively coupled to positive word line WL refj,+ and negative word line WL refj,- , such as for Figure 1 The memory cell group 22 is discussed.
[0266] Word Line WL refj,+ , WL refj, - receiving a corresponding reference activation signal S in use refj,+ , S refj,- , these signals can be Figure 4 The same reference input to the time converter 77 is generated as the input to the time converter 46.
[0267] The reference circuit 70 is configured to, in use, C The reference bit line BL ref The charge displacement in the bit line BL indicates i The MAC refinement operation has zero result (y i = 0), for example, where all input values x 1 , ..., x N are all equal to zero and / or all calculated weights G i,1 , ..., G i,N All are equal to zero.
[0268] In this embodiment, all positive reference activation signals S refj,+ With a first refinement stage T E1 Duration T E1 activation duration, and all negative reference activation signals S refj,- With a second refinement stage T E2 Duration T E2 activation duration.
[0269] However, the signal S refj,+ , S refj,- The duration can be equal to T E2 and T E1 .
[0270] Furthermore, all reference groups 74 are programmed so that the corresponding stored calculation weights G i1 , ..., G iN corresponds to a zero calculation weight, that is, such that the current intensity used in the corresponding memory cell is I + ref,j =I - ref,j =I ref .
[0271] However, during the initialization phase, the reference activation signal S refj,+ , S refj,- Different combinations of activation duration and / or reference current I + ref,j ,I - ref,j Different combinations of make the effective refinement interval T C During the period, the reference bit line BL ref The charge shifted in corresponds to a result of the MAC refinement operation of array 12 that is equal to zero.
[0272] Fig.10 Purely by way of example, the effective refinement interval T C The timing signal TM within the reference circuit 70, the reference charge signal q generated by the reference circuit 70 ref as well as Figure 1 The two common charge signals q shown in i1 ,q i2 .
[0273] In detail, the charge signal q i1 is coupled to the bit line BL i1 The digital detector 16 provides a signal, and the charge signal q i2 is coupled to the bit line BL i2 The digital detector 16 provides a signal.
[0274] exist Fig.10 For simplicity, the signals TM and q are shown. ref ,q i1 and the corresponding simulation trends of qi2 to highlight the C Its possible evolution.
[0275] Furthermore, purely as an example, Fig.10 signal q in i1 ,q i2 ,q ref It has a slope-like trend. In the effective refinement interval T C However, the effective trend depends on the specific input values and calculation weights.
[0276] In the example shown, the charge signal q i2 Greater than the reference charge signal q ref Therefore, from the charge signal q i2 Initially, the DSP 17 provides a corresponding output signal y indicating a value greater than zero. i2 .
[0277] Charge signal q i1 Below the reference charge signal q ref Therefore, the corresponding output signal y provided by DSP 17 is i1 Indicates a value less than zero.
[0278] In addition, q i1 With q ref The difference Δq i1 The absolute value of is less than q i2 With q ref The difference Δq i2 Therefore, the output signal y i1 Indicated by the output signal y i2 In other words, the MAC operation associated with column i1 is has a MAC operation associated with column i2 lower than The absolute value of .
[0279] Reference charge signal q ref The fact that the reference memory cell 72 is used by the reference circuit 70 allows tracking of the reference charge value Q ref Any possible modification of , for example due to a drift of the transconductance value of the storage element 25 over time.
[0280] Thus, the IMC device 10 can provide accurate results at the output even if the memory array 12 deviates from the behavior established during the initialization phase.
[0281] In addition, when calculating the weight G ij To the corresponding current value I + i,j and I - i,j The mapping and / or input value x j To the corresponding theoretical duration T* j,+ and T* j,- In the case where the mapping is not symmetric with respect to zero, the reference charge signal q ref The possibility of indicating the charge value can be useful.
[0282] Fig.11 Shows Figure 1 The general cell group (herein denoted by reference numeral 80) of the memory array 12 i,j denoting different embodiments of the present invention.
[0283] Memory cell group 80 i,j It consists of four cells, including two positive cells 82A', 82A" and two negative cells 82B', 82B". In practice, the positive cells 82A', 82A" form a positive subset of cells, while the negative cells 82B', 82B" form a negative subset of cells.
[0284] Units 82A', 82A", 82B', 82B" and Figure 1 The units 20 are identical and each unit comprises a corresponding storage element (here represented by 85 ) and a selection element 26 .
[0285] The common node 28 of each cell 82A', 82A", 82B', 82B" is coupled (directly connected here) to the bit line BL i .
[0286] The selection elements 26 of the positive cells 82A′, 82A″ are coupled to the positive word line WL j,+ The selection elements 26 of the negative cells 82B', 82B" are coupled to the negative word line WL j,- .
[0287] The storage elements 85 are each a variable resistive element, in particular here also based on a phase change material (PCM).
[0288] In this embodiment, each storage element 85 can be programmed to represent two different logic states, such as '0' and '1'; that is, they can be programmed to have two different transconductance values g 1 , g 2 , where g 1 <g 2 , which is related to the two corresponding current values I 1 and I 2 correspond.
[0289] For example, for Vr = 100mV, if the corresponding transconductance is equal to g 1 , then each unit 82A', 82A", 82B', 82B" can flow a current I 1 = 0μA, or if the corresponding transconductance is equal to g 2 , then the current I 2 =1μA.
[0290] The positive cells 82A′, 82A″ are arranged in parallel with each other from an electrical point of view and are coupled to the positive word line WL j,+ Therefore, the positive cells 82A', 82A" operate as a single positive cell, which can be programmed to operate from the bit line BL i Sinking a positive cell current I with one of the following three values + i,j :I 1 +I 1 ,I 1 +I 2 or I 2 +I 2 .
[0291] Likewise, negative cells 82B′, 82B″ are arranged in parallel with each other from an electrical perspective and are coupled to the same negative word line WL j,- Therefore, negative cells 82B', 82B" operate as a single negative cell, which can be programmed to i Sinking current I with one of the following three values - i,j :I 1 +I 1 ,I 1 +I 2 or I 2 +I 2 .
[0292] Therefore, from an electrical point of view, the unit group 80 i,j Overall behavior and targeting Figure 1 Memory cell group 22 i,j Same behavior as described.
[0293] Therefore, it can be used with Figure 1 Unit Group 22 i,j The same method as discussed above will be used to group 80 i,j Used to represent the corresponding calculated weight G with sign and absolute value ij .
[0294] Fig.12 An embodiment of a bias circuit 18 for an IMC device 10 is shown.
[0295] The bias circuit 18 is biased by the bias current I b Start generating bias voltage Vr and provide bias voltage Vr to Figure 1 The bit lines BL of the memory array 12 1 , ..., BL M .
[0296] In this embodiment, bias circuit 18 biases all bit lines BL 1 ,…,BL M However, the bias circuit 18 may provide a bias voltage Vr to the bit line BL starting from the bias voltage Vr according to the specific application. 1 ,…,BL M Bias voltages different from each other are provided.
[0297] The bias circuit 18 includes a circuit for generating a bias current I b A current source 90 having an input node 92 and a bias impedance Z b The reference network is 91.
[0298] In this embodiment, the current source 90 is a controllable current source that receives an external signal EXT (eg, from a user of the IMC device 10 ) indicating the bias current I b expected value.
[0299] The bias network 91 receives a bias current I at an input node 92. b The input node 92 is in the bias current I b and bias impedance Z b The voltage as a function of .
[0300] The bias circuit 18 further comprises a voltage distribution circuit, here formed by an operational amplifier 95 having an output terminal 96 providing a bias voltage Vr.
[0301] The operational amplifier 95 has a non-inverting input terminal coupled to an input node 92 of the bias network 91. The operational amplifier 95 has an inverting input terminal coupled (particularly here directly coupled) to an output terminal 96 of the operational amplifier 95.
[0302] Depending on the specific application, the output terminal 96 of the amplifier 95 is coupled to the bit line BL, for example, directly or via a dedicated circuit. 1 , ..., BL M .
[0303] In practice, in this embodiment, the voltage at the input node 92 of the bias network 91 constitutes the bias voltage Vr.
[0304] In detail, in this embodiment, the bias network 91 is formed by a bias memory array (hence also denoted by 91 hereinafter) having a total bias transconductance value g b And comprises one or more memory cells, here a plurality of memory cells 97 of the non-volatile type.
[0305] Bias array 91 may be part of memory array 12 or may be a separate memory array.
[0306] The element 97 of the bias array 91 has the same Figure 1 The memory cells 20 of the memory array 12 have the same circuit configuration.
[0307] In detail, the cells 97 of the bias array 91 each comprise a storage element 98 and a selection element 99 , in particular identical to the storage element 25 and the selection element 26 , respectively, of the memory cell 20 .
[0308] In practice, the storage element 98 is based on the same technology as used to obtain the storage element 25. For example, if the storage element 25 is based on a PCM material, then the storage element 98 is also based on a PCM material, in particular the same PCM material.
[0309] Depending on the specific application, cells 97 of bias array 91 are programmed to store respective transconductance values that may be equal to or different from each other.
[0310] The number of cells 97 and corresponding transconductance values of bias array 91 may be selected at the design stage such that bias array 91 constitutes a statistically significant sample of memory array 12 .
[0311] In practice, from a statistical point of view, the total bias transconductance g of the bias array 91 is b represents the total transconductance of the memory array 12.
[0312] For example, the total transconductance of the memory array 12 may be equal to the total transconductance of all memory cells 20 at the effective refinement interval T C If the periods are activated simultaneously, the memory array 12 will have a transconductance of .
[0313] For example, the number of memory cells 97 may be higher than one hundred.
[0314] For example, memory cell 97 may be programmed so that the total bias transconductance of bias array 91 is g b is equal to the average value of the total transconductance of the memory array 12.
[0315] For example, the mean value may represent an average transconductance value of the memory array 12 during use, such as measured during a calibration or initialization phase of the IMC device 10 .
[0316] The storage element 98 has a first terminal coupled to the input node 92 of the bias network 91 and a second terminal coupled to a reference potential node (here, ground 29 ) via a selection element 99 .
[0317] In practice, all memory cells 97 share the same bias bit line BL bias .
[0318] The selection elements 99 are each formed by a respective switch, for example a BJT, a diode or a MOSFET, here an NMOS transistor, in particular identical to the selection element 26 of the memory cell 20 and arranged in series with the respective storage element 98 .
[0319] In this embodiment, the selection elements 99 of all cells 97 are coupled to the same word line WL_B so as to be controlled by the same bias activation signal, which may be generated, for example, by the word line activation unit 14 or other components of the IMC device 10 (not shown here).
[0320] However, the selection elements 99 may each be controlled by respective activation signals that are different from one another, for example depending on which memory cells are to be activated during a calculation performed by the IMC device 10 .
[0321] Bias impedance Z b is a function of the transconductance value stored by memory cell 97 and the duration of the bias activation signal provided to bias word line WL_B.
[0322] The bias voltage Vr is determined by the bias current I b The fact that the generator starts to generate makes it possible to change the bias impedance Z b and / or bias current I b To adjust the bias voltage Vr.
[0323] During the lifetime of the IMC device 10, the IMC device 10 is subject to temperature variations, which may affect the value of the transconductance of the memory cell 20. Furthermore, the value of the transconductance of the memory cell 20 is subject to drift; for example, in case the memory cell 20 is a PCM memory cell, the storage element 25 is subject to aging phenomena (e.g., amorphization).
[0324] This deviation of the transconductance from the programmed value causes the positive current I + i,j and negative current I - i,j relative to the programmed value, resulting in a change in the calculated weight G ij The mapping of i An error occurred.
[0325] However, temperature variations or drifts that affect memory cell 20 also affect the value of the transconductance of memory cell 97 of bias array 91 because memory cell 97 represents a statistically significant sample of memory cells 20 .
[0326] Therefore, the transconductance g of the bias array 91 is b The overall value of will also be affected by such deviations or drifts.
[0327] Total transconductance g b The change in bias impedance Z b The change of Vr and thus means the change of the voltage at the input node 92 of the reference network 91. Therefore, the bias voltage Vr may also change according to the temperature and the drift phenomenon of the memory cell 20.
[0328] In detail, the variation of the bias voltage Vr allows the variation of the transconductance value of the memory cell 20 to be compensated.
[0329] For example, if memory cell 20 experiences a drift that causes the corresponding transconductance value to increase, then current I + i,j ,I - i,j Therefore, the corresponding bit line current I BL,i However, at the same time, the bias transconductance g of the bias array 91 b The total value also increases.
[0330] Transconductance g b The increase in bias impedance Z b and therefore means a reduction in the bias voltage Vr.
[0331] The reduction of bias voltage Vr involves current I + i,j ,I - i,j The decrease in the bit line current I BL,i of reduction.
[0332] in other words, Fig.10 The bias circuit 18 enables compensation of the current I + i,j ,I - i,j The possible variation of the drift of the transconductance value of the memory cell 20 is thus reduced. In this way, the current I + i,j ,I - i,j is kept equal to the corresponding design value, thereby preventing errors that may occur during the execution of the MAC operation by the IMC device 10.
[0333] Fig.13 Shown in detail Figure 7 An embodiment of the digital detector 16, in particular the corresponding charge packet detector 60, is also referred to as an “integration stage” in the following.
[0334] The following reference is made to the plurality of bit lines BL coupled to 1 , ..., BL M Any bit line BL in i The digital detector 16 will be described in detail below.
[0335] In this embodiment, the integrating stage 60 includes a current mirror 115 which converts the bit line BL i The bit line current I BL,i Mirrored into the input node 116 of the corresponding integrating stage 111 .
[0336] The current mirror 115 has a current ratio of 1:k, so that the mirrored bit line current k·I BL,i flows through the input node 116 of the integrating stage 60 .
[0337] In detail, the current mirror 115 has a current mirror coupled to the bit line BL. i A first branch, here formed by a corresponding PMOS transistor 117 , and a second branch, here formed by a corresponding PMOS transistor 118 , are coupled to the corresponding integrating stage 60 .
[0338] The sources of PMOS transistors 117 and 118 are coupled to a supply node 120, where the voltage is V DD , and the gates of the PMOS transistors 117, 118 are coupled to each other and to the drain of the PMOS transistor 117. The drain of the PMOS transistor 118 is coupled (specifically directly connected here) to the input node 116 of the integration stage 60.
[0339] The integration stage 60 includes a first integration circuit 121 , a second integration circuit 122 , and a switch circuit 123 coupled between the first integration circuit 121 and the second integration circuit 122 .
[0340] The first and second integrator circuits 121, 122 are coupled to the input node 116 to receive the mirrored bit line current k·I BL,i .
[0341] The first integrator circuit 121 includes a first inverter 124 having an output terminal 125, a capacitor C coupled to the output terminal 125 of the first inverter 124, and a capacitor C coupled to the output terminal 125 of the first inverter 124. A A capacitor 127 is provided, and a second inverter 128 whose input terminal is coupled to the output terminal 125 of the first inverter 124.
[0342] The first inverter 124 has a phase shifter coupled to the integrating stage 60 ( Fig.14 ) and receives a first control signal IN at the input terminal. A .
[0343] In practice, the first inverter 124 is provided by mirroring the bit line current k·I BL,i Bias.
[0344] Capacitor 127 has a first terminal coupled to the output node 125 of first inverter 124 and a second terminal coupled to a reference node (here, ground).
[0345] The output node 125 of the first inverter 124 is at the first integrated voltage drop V dropped across the capacitor 127 A .
[0346] The second inverter 128 has a first sampling threshold (hereinafter defined as “first threshold V th1 ”), receiving a first integrated voltage V at the input terminal A and provides a first threshold V at the output th1 and the first integrated voltage V A A first switching signal S1 is a function of .
[0347] In detail, the first switch signal S1 is when the first integrated voltage V A Below the first threshold V th1 has a high logic value and when the first integrated voltage V A Above the first threshold V th1 A logic signal with a low logic value when
[0348] The second integrator circuit 122 includes a first inverter 130 having an output terminal 131, a capacitor C coupled to the output terminal 131 of the first inverter 130, and a capacitor C coupled to the output terminal 131 of the first inverter 130. B A capacitor 132 , and a second inverter 133 whose input terminal is coupled to the output terminal 131 of the first inverter 130 .
[0349] The first inverter 130 has a supply node coupled to the input node 116 of the integrating stage 60 and receives the second control signal IN at an input terminal. B .
[0350] In practice, the first inverter 130 is provided by mirroring the bit line current k·I BL,i Bias.
[0351] Capacitor 132 has a first terminal coupled to the output node 131 of first inverter 130 and a second terminal coupled to a reference node (here, ground).
[0352] The output node 131 of the first inverter 130 is at the second integrated voltage V dropped across the capacitor 131. B .
[0353] The second inverter 133 has a second sampling threshold V th2 (hereinafter defined as the second threshold V th2 ), receiving a second integrated voltage V at the input terminal B , and provides at the output a second threshold value V th2 and the second integrated voltage V B The second switching signal S2 is a function of .
[0354] In detail, the second switch signal S2 is when the second integrated voltage V B Below the second threshold V th2 has a high logic value and when the second integrated voltage V B Above the second threshold V th2 A logic signal with a low logic value when
[0355] In this embodiment, the first threshold V th1 Equal to the second threshold V th2 However, depending on the specific application, the first threshold V th2 can be compared with the second threshold V th2 different.
[0356] The switch circuit 123 is a latch formed by two inverters 135 , 136 arranged in a ring configuration, a first switch 137 controlled by a first switch signal S1 , and a second switch 138 controlled by a second switch signal S2 .
[0357] The switch circuit 123 has a first node 140 coupled to the input of the inverter 136 and the output of the inverter 135 , and a second node 141 coupled to the output of the inverter 136 and the input of the inverter 135 .
[0358] The first node 140 provides a first control signal IN A The second node 141 provides a second control signal IN B .
[0359] The first switch 137 is coupled to the first node 140 and the voltage V′ DD and the second switch 138 is coupled between the second node 141 and the voltage V' DD between nodes.
[0360] Voltage V' DD can be equal to the voltage V DD Or different from it. For example, if the voltage V' DD With voltage V DD different, especially less than the voltage V DD , the digital detector 16 may then include a voltage scaling circuit, such as a transistor (not shown here), whose source and drain terminals are coupled between the supply node 120 and the input node 116 of the integrating stage 60 .
[0361] In this embodiment, the switch circuit 123 also receives an enable signal EN, which controls the activation of the switch circuit 123. For example, the enable signal EN can be used to keep the switch circuit 123 closed when it is not in use, thereby optimizing energy consumption. In addition, the enable signal EN can be used to set the switch circuit 123 to a defined state, such as when the IMC device 10 is turned on.
[0362] The first and second control signals IN A , IN B Can constitute the packet detection signal CP ( Figure 7 ), and thus may be provided at the input of the multiplexer circuit 61, as referenced Figure 7 For example, the signal IN A , IN B Can be used alternately and successively as packet detection signals.
[0363] refer to Fig.14 The first inverter 125 of the first integration circuit 121 is a CMOS inverter formed by a series circuit of a PMOS transistor 150 and an NMOS transistor 151 coupled to each other at an output node 125. The PMOS transistor 150 and the NMOS transistor 151 receive a first control signal IN at respective gate terminals. A .
[0364] The source of the PMOS transistor 150 is coupled to the input node 116 of the integrating stage 60 .
[0365] The second inverter 128 of the first integrating circuit 121 is a CMOS inverter formed by a series circuit of a PMOS transistor 152 and an NMOS transistor 153 coupled to each other at a node 154 where the first switching signal S1 is provided.
[0366] The first threshold V of the second inverter 128 th1 is the switching threshold of the second inverter 128 and therefore depends on the characteristics, such as threshold or on-resistance, of the PMOS transistor 72 and the NMOS transistor 73. In practice, the switching threshold may be the input voltage at which the output of the inverter has a high logic value or otherwise the input voltage at which the output of the inverter has a low logic value.
[0367] For example, the switching threshold of the second inverter 128 may be defined as the corresponding input voltage (ie, the first integrated voltage V A ) is equal to the operating point of the corresponding output voltage (ie, the first switching signal S1).
[0368] The first inverter 130 of the second integration circuit 122 is a CMOS inverter formed by a series circuit of a PMOS transistor 155 and an NMOS transistor 156 coupled to each other at an output node 131. The PMOS transistor 155 and the NMOS transistor 156 receive a second control signal IN at their respective gate terminals. B .
[0369] The source of the PMOS transistor 155 is coupled to the input node 116 of the integrating stage 60 .
[0370] The second inverter 133 of the second integrating circuit 122 is a CMOS inverter formed by a series circuit of a PMOS transistor 157 and an NMOS transistor 158 coupled to each other at a node 159 where the second switching signal S2 is provided.
[0371] The second threshold V of the second inverter 133 th2 is the switching threshold of the second inverter 133; that is, depends on the characteristics of the PMOS transistor 157 and the NMOS transistor 158. For example, the switching threshold depends on the gate-source voltage that enables current to flow through the source-drain path of the PMOS transistor 157 and the NMOS transistor 158.
[0372] like Fig.14 As shown in the detailed embodiment of the present invention, the first inverter 135 and the second inverter 136 of the switch circuit 123 are cross-coupled CMOS inverters, each of which includes a voltage V' at a power supply node (here DD ) and the ground and a corresponding PMOS transistor 160 and a corresponding NMOS transistor 161 are coupled in series with each other.
[0373] In addition, the first and second inverters 135, 136 of the switch circuit 123 each include a corresponding enable switch, which is a PMOS transistor 162, which is coupled to the voltage V'. DD between the power supply node of the inverter and the PMOS transistor 160 of the corresponding inverter.
[0374] The PMOS transistor 162 is controlled by an enable signal EN.
[0375] In use, the bit line BL i The bit line current I BL,i is mirrored into the integration stage 113 of the corresponding digital detector 16 .
[0376] Fig.15 Shows Fig.13 The first control signal IN of the digital detector 16 A , the first integrated voltage V A and the second integrated voltage V B Example of trends over time.
[0377] For t 0 <t<t 1 , the first integrated voltage V A Below the first threshold V th,1 Therefore, the PMOS transistor 152 of the second inverter 128 is turned on, and the NMOS transistor 153 of the second inverter 128 is turned off. Therefore, the first switch signal S1 (not shown here) has a high value, and the first switch 137 is turned on. The first control signal IN A Has a low value.
[0378] Therefore, referring to the first inverter 124 of the first integrating circuit 121, for t 0 <t<t 1 , the PMOS transistor 150 is turned on, and the NMOS transistor 151 is turned off.
[0379] At the same time, for t 0 <t<t 1 , the second control signal IN B has a high value, therefore, referring to the first inverter 130 of the second integrator circuit 122, for t 0 <t<t 1 , the PMOS transistor 155 is turned off, and the NMOS transistor 156 is turned on.
[0380] Therefore, the mirrored bit line current k·I BL,i The voltage flows out from the input node 116 and flows only through the first inverter 124 of the first integrating circuit 121 , but does not flow through the first inverter 130 of the second integrating circuit 122 .
[0381] In detail, the mirrored bit line current k·I BL,i Passes through PMOS transistor 150 and charges capacitor 127. Therefore, the first integrated voltage V A For t 0 <t<t 1 Increases over time.
[0382] In detail, Fig.15 In the example, the first integrated voltage V A For t 0 <t<t 1 increases linearly with time; however, the first integrated voltage V A The evolution of depends on the time interval t 0 <t<t 1 Internal bit line current I BL,i specific evolution.
[0383] When the first integrated voltage V A becomes equal to the first threshold voltage V th,1 , the NMOS transistor 153 of the second inverter 128 is turned on and the PMOS transistor 152 is turned off.
[0384] In this embodiment, the first control signal IN A At time t 2 Shows high value.
[0385] Time t 1 With t 2 The time delay between φ and φ may correspond, for example, to a propagation delay of the second inverter 128 of the first integration circuit 121 and / or a switching time of the first switch 137 .
[0386] For t 1 <t<t 2 , the mirrored bit line current I BL,i The capacitor 127 continues to be charged; therefore, the first integrated voltage V A Increases until the maximum value (time t 2 ).
[0387] At time t 2 , when the first control signal IN A When the second control signal IN is high, B (not shown here) presents a low value (the inverter 136 of the switch circuit 123 receives the first control signal IN at the input terminal A ).
[0388] When the first control signal IN A When the second control signal IN has a high value, the PMOS transistor 150 and the NMOS transistor 151 of the first inverter 124 of the first integration circuit 121 are turned off and on, respectively. B With a low value, the PMOS transistor 155 and the NMOS transistor 156 of the first inverter 130 of the second integration circuit 122 are turned on and off, respectively.
[0389] Therefore, for t>t 2 , the mirrored bit line current k·I BL,i The voltage flows out from the input node 116 and flows only through the first inverter 130 of the second integrating circuit 122 , but does not flow through the first inverter 124 of the first integrating circuit 121 .
[0390] In detail, the mirrored bit line current k·I BL,i Flows through the PMOS transistor 155 and charges the capacitor 132 of the second integration circuit 122. Therefore, the second integration voltage V B From time t 2 Start to increase over time.
[0391] In detail, Fig.15 In the example, the second integrated voltage V B For t>t 2 increases linearly with time; however, the second integrated voltage V B The evolution of the bit line current I BL,i specific evolution.
[0392] When the first control signal IN A When the first integration circuit 121 has a high value, the capacitor 127 is discharged through the NMOS transistor 151 of the first inverter 124, and the first integration voltage V A Therefore it decreases to zero.
[0393] When the second integrated voltage V B becomes equal to the second threshold voltage V th,2 Time (time t 3 ), the NMOS transistor 158 of the second inverter 133 is turned on, and the PMOS transistor 157 is turned off.
[0394] Therefore, at time t4, the same as above for time t 2 The first control signal IN A In a similar manner to that discussed above, the second control signal IN B Shows high value.
[0395] In detail, in response to the second integrated voltage V B Reaching the second threshold V th,2 , the second switch signal S2 switches to a low value and the second switch 138 is closed, so that the second node 141 is at a voltage V' DD , and therefore the second control signal IN B Shows high value.
[0396] Time t 3 With t 4 The time delay between φ and φ may correspond, for example, to a propagation delay of the second inverter 133 of the second integration circuit 122 and / or a switching time of the second switch 138 .
[0397] For t 3 <t<t 4 , the mirrored bit line current k·I BL,i The capacitor 132 of the second integration circuit 122 continues to be charged; therefore, the second integration voltage VB increases until the maximum value (time t 4 ).
[0398] For t 2 <t<t 4 , the switch circuit 123 maintains the first control signal IN A is high, and the second control signal IN B Is a low value.
[0399] At time t 4 , in response to the second control signal IN B The fact that the first control signal IN A Again, low values.
[0400] In response to the first control signal IN A The fact that the mirrored bit line current k·I BL,i The capacitor 127 of the first integrating circuit 121 is charged again until time t 6 , similar to the case for t 1 <t<t 2 The situation under discussion.
[0401] Therefore, from time t 6 Until time t 7 , the mirrored bit line current k·I BL,i The capacitor 132 of the second integrating circuit 122 is charged until time t 7 , similar to the case for t 2 <t<t 4 The situation under discussion.
[0402] In practice, Figure 7 The counter circuit can be configured to respond to the first control signal IN A and / or the second control signal IN B The number of switching events is counted to count the bit line I BL,i The number of associated charge packets.
[0403] In practice, Fig.13 and Fig.14 The digital detectors 16 can each integrate the corresponding bit lines BL i The bit line current I BL,i , performing multiple successive sampling iterations. In each sampling iteration, for example, the bit line current k·I of the reference mirror BL,i The sampling iteration of the first integration circuit 121 generates the bit line current k·I as a mirror image at the integration stage 60. BL,i The time-integrated first integrated voltage V A , the first integrated voltage V A With the first threshold V th,1 The comparison is performed, and in response to the first integrated voltage VA reaching the first threshold value V th,1 , reset the first integration voltage V A , in particular by switching the first control signal IN A The counter circuit 62 can respond to the first integrated voltage V A Reaching the first threshold V th,1 And update the corresponding charge signal q i .
[0404] In other words, each of the digital detectors 16 converts the bit line current I BL,i Converts into multiple charge packets and counts the charge packets to sample the corresponding bit line current I BL,i , where each charge packet corresponds to a charge accumulated on capacitors 127 , 132 , which causes switching of second inverters 128 , 133 .
[0405] Therefore, the bit line current is integrated into the capacitor C tot Compared with the case of a single capacitor, the capacitance of the capacitors 127 and 132 can be reduced. In detail, the capacitance of the capacitors 127 and 132 can be smaller than the capacitance C tot Low 2 F times, where F is the charge signal q i The number of digits.
[0406] Therefore, the digital detector 16 has a low die footprint, and thus the IMC device 10 may exhibit a low production cost.
[0407] In addition, when the bit line current I BL,i Flows through the corresponding bit line BL i When (i.e., at the effective refinement interval T C During this period), the digital detectors 16 each begin to discretize the corresponding bit line current I BL,i Therefore, the charge signal q i The effective refinement interval T C Immediately after completion, be ready for subsequent refinement operations.
[0408] For example, the digital detectors 16 may each detect the corresponding bit line current I BL,i Sampling is performed until the count end signal STOP_CNT indicates the refinement interval T C The end of.
[0409] In addition, according to the embodiment shown, the switch circuit 123 responds to the first integrated signal V A Reaching the first threshold V th,1 The first integration circuit 121 is disabled and the second integration circuit 122 is enabled, and in response to the second integration signal V B Reaching the second threshold V th,2 The first integration circuit 121 is enabled and the second integration circuit 122 is disabled.
[0410] This allows the bit line current I BL,i The first integration circuit 121 and the second integration circuit 122 are alternately sampled, so that the bit line current I BL,i Capacitor 127 can be charged while capacitor 132 is discharging, and capacitor 132 can be charged while capacitor 127 is discharging. By doing so, no charge is lost during sampling, and digital detector 16 can achieve bit line current I BL,i High accuracy of integration.
[0411] Reference again Fig.14 The set stage 60 may also include a first stop switch and a second stop switch, here a first NMOS transistor 171 and a second NMOS transistor 172, indicated by dashed lines, which prevent the corresponding digital detector 16 from responding to the corresponding bit line current I BL,i Take samples.
[0412] In detail, the first and second NMOS transistors 171, 172 have drain terminals coupled to the output nodes 127, 131 of the first integration circuit 121 and the second integration circuit 122, respectively, and source terminals coupled to a reference potential line (here, ground). The first and second NMOS transistors 171, 172 receive the count end signal STOP_CNT at their respective gate terminals.
[0413] In the refinement interval T C When finished, the count end signal STOP_CNT switches to a high logic value, thereby turning on the first and second NMOS transistors 171 , 172 and shorting the output nodes 127 , 131 of the first and second integration circuits 121 , 122 to ground.
[0414] Therefore, the first and second integration circuits 121 and 122 stop integrating the bit line current BL i integral.
[0415] According to one embodiment, Figure 3 The timer 45 may be formed by a current-controlled oscillator, such as a ring oscillator formed by an odd number of inverters or a Fig.16 The oscillator shown in FIG. 1 has an oscillation frequency f o In this case, the timer 45 receives the supply current. The oscillation frequency f o Depends on the supply current.
[0416] The timer 45 updates the timing signal TM at an update frequency f u Depends on the oscillation frequency f o , and thus depends on the supply current.
[0417] Therefore, the refinement time T C , that is, the timing signal TM changes from an initial value (e.g., zero) to a value at time t start )) reaches the final value MAX_CNT (final moment t end ) also depends on the supply current of timer 45.
[0418] In practice, the desired refinement duration T can be set by adjusting the supply current of the timer 45. C .
[0419] like Figure 3 As indicated by the dashed arrow in FIG. 1 , the supply current of the timer 45 can be Fig.10 The bias circuit 18 generates a bias current I b In this way, the supply current of the timer 45 can be controlled by the external signal EXT.
[0420] In practice, the refinement duration T of the IMC device 10 is C It can be adjusted by the user of the IMC device 10.
[0421] In addition, due to the bias current I b The update frequency f of the timer signal TM is therefore 1 / 200 MHz, which is generated by the current source 90 and is therefore not affected by the drift of the value transconductance of the bias array 91. u And thus refine the time T C It is also independent of the drift of the memory cell 97 .
[0422] Fig.16 and 17 An embodiment of a current-controlled timer 45 is shown in detail.
[0423] In detail, the timer 45 includes a circuit for generating an oscillator current I from a supply current. OSC Current mirror 180, where the supply current is equal to Fig.12 The bias current I b , and the oscillator current I OSC A counting portion 181 is provided which provides a timer signal TM.
[0424] The current mirror 180 has a mirror ratio of 1:p, so the oscillator current I OSC will be p·I REF .
[0425] In detail, the current mirror 180 has a first branch (here formed by a corresponding PMOS transistor 183 , coupled to the current source 32 ) and a second branch (here formed by a corresponding PMOS transistor 184 , coupled to the counting part 181 ).
[0426] The sources of PMOS transistors 183, 184 are coupled to a supply node 185 (which is at a voltage V DD ), and the gates of the PMOS transistors 183 and 184 are coupled together and coupled to the drain of the PMOS transistor 183. The drain of the PMOS transistor 184 is coupled (specifically directly connected here) to the input node 187 of the counting part 181.
[0427] The counting part 181 of the timer 45 includes an integration stage 190 and a counter stage 195 coupled to the integration stage 190 and providing a timer signal TM. The integration stage 190 is formed by a first integration circuit 191, a second integration circuit 192 and a switching circuit 193 coupled between the first and second integration circuits 191 and 192.
[0428] First and second integrator circuits 191, 192 are coupled to input node 187 to receive the oscillator current I OSC .
[0429] The first integrator circuit 191 includes a first inverter 197 having an output terminal 198, a first integrator circuit 198 having a capacitor C′ coupled to the output terminal 198 of the first inverter 197, and a first integrator circuit 199 having a capacitor C′ coupled to the output terminal 198 of the first invertor 197. A A capacitor 199 is provided, and a second inverter 200 whose input terminal is coupled to the output terminal 198 of the first inverter 197.
[0430] The first inverter 197 has a coupling to the counting portion 181 ( Fig.14 ) is a power supply node of an input node 187 and receives a first oscillator control signal OS at an input terminal A .
[0431] In practice, the first inverter 197 is driven by the oscillator current I OSC Bias.
[0432] Capacitor 199 has a first terminal coupled to the output node 198 of the first inverter 197 and a second terminal coupled to a reference potential node (here, ground).
[0433] The output node 198 of the first inverter 197 is at the first oscillator integrated voltage V′ dropped across the capacitor 199. A .
[0434] The second inverter 200 has a first oscillator threshold V' th1 (hereinafter referred to as the first threshold V' th1 ), receiving the first oscillator integrated voltage V' at the input terminal A And at the output end according to the first threshold V' th1 and the first oscillator integrated voltage V' A A first oscillator switching signal S'1 is provided.
[0435] In detail, when the first oscillator integrates the voltage V' A Below the first threshold V' th1 When the first oscillator integrated voltage V' A Above the first threshold V' th1 When , the first oscillator switch signal S'1 has a low logic value.
[0436] The second integrator circuit 192 includes a first inverter 202 having an output terminal 203, a capacitor C′ coupled to the output terminal 203 of the first inverter 202, and a capacitor C′ coupled to the output terminal 203 of the first inverter 202. B A capacitor 204 is provided, and a second inverter 205 whose input terminal is coupled to the output terminal 203 of the first inverter 202.
[0437] The first inverter 202 has a circuit coupled to the counting portion 181 ( Fig.14 ) and receives the second oscillator control signal OS at the input terminal. B .
[0438] In practice, the first inverter 202 is driven by the oscillator current I OSC Bias.
[0439] The capacitor 204 has a first terminal coupled to the output node 203 of the first inverter 202 and a second terminal coupled to a reference potential node (here, ground).
[0440] The output node 203 of the first inverter 202 is at the second oscillator integrated voltage V′ dropped across the capacitor 204 B .
[0441] The second inverter 205 has a second oscillator threshold V' th2 (hereinafter referred to as the second threshold V' th2 ), receiving the second oscillator integrated voltage V' at the input terminal B And at the output end according to the second threshold V' th2 and the second oscillator integrated voltage V B A second oscillator switching signal S'2 is provided.
[0442] In detail, when the second oscillator integrates the voltage V B Below the second threshold V' th2 When the second oscillator integrated voltage V' B Above the second threshold V' th2 When , the second oscillator switch signal S'2 has a low logic value.
[0443] In this embodiment, the first threshold V′ of the second inverter 200 is th1 is equal to the first threshold V of the second inverter 128 of the digital detector 16 th1 ( Fig.13 and Fig.14 ). The second threshold value V′ of the second inverter 205 th2 Equal to digital detector 16 ( Fig.13 and Fig.14 ) of the second inverter 133 th2 .
[0444] Reference again Fig.16 The switch circuit 193 is a latch formed by two inverters 208, 209 arranged in a ring configuration, a first switch 210 controlled by a first oscillator switch signal S'1, and a second switch 211 controlled by a second oscillator switch signal S'2.
[0445] The switch circuit 193 has a first node 213 coupled to the input of the inverter 209 and the output of the inverter 208 , and a second node 214 coupled to the output of the inverter 209 and the input of the inverter 208 .
[0446] The first node 213 provides a first oscillator control signal OS A The second node 214 provides a second oscillator control signal OS B .
[0447] The first switch 210 is coupled to a first node 213 and a voltage V′ DD and the second switch 211 is coupled to the second node 214 and the voltage V' DD between nodes.
[0448] According to one embodiment, the voltage V′ of the counting portion 181 of the timer 45 is DD can be equal to the voltage V' of the integrating stage 60 of the digital detector 16 DD ( Fig.13 ).
[0449] exist Fig.16 In an embodiment of the present invention, the switch circuit 193 also receives an oscillator enable signal EN', which controls the activation of the switch circuit 193. For example, the oscillator enable signal EN' can be used to keep the switch circuit 193 closed when it is not in use, thereby optimizing energy consumption. In addition, the enable signal EN' can be used to set the switch circuit 193 to a clear state, such as when the IMC device 10 is turned on.
[0450] The counter stage 195 is coupled to a first node 213 and a second node 214 of the switch circuit 193 .
[0451] In detail, the charge counter stage 195 includes an inverter 216 having an input coupled to the second node 214, and a counter including an inverter 217 having an input coupled to the first node 213 and a plurality of D flip-flops 218 including a first flip-flop 218.2, a second flip-flop 218.3 and a last flip-flop 218.L, where L is the timer signal TM <l:1>The number of digits.
[0452] In practice, the counter of the charge counter stage 195 has L−1 flip-flops 218 .
[0453] The output of inverter 216 provides the first bit TM(1), ie, the least significant bit, of the timer signal TM.
[0454] The flip-flops 218 are cascaded to each other, sequentially from the first flip-flop 218.2 to the last flip-flop 218.L.
[0455] The flip-flops 218 each have a clock input (CK input), a data input (D input), a Q output, and a Output.
[0456] The CK input of the first flip-flop 218.2 is coupled to the output of the inverter 217. The output is sent back to the D input of the first flip-flop 218.2. The Q output of the first flip-flop 218.2 constitutes the timer signal TM <l:1>The second TM(2).
[0457] The CK input of the second flip-flop 218.3 is coupled to the CK input of the first flip-flop 218.2. The second flip-flop 218.3 The output is fed back to the D input of the second flip-flop 218.3. The Q output of the second flip-flop 218.3 constitutes the timer signal TM <l:1>The third TM(3).
[0458] The description for the second flip-flop 218 . 3 applies, mutatis mutandis, to all subsequent flip-flops (not shown here) up to the L−1th flip-flop (also not shown).
[0459] Finally, the CK input of the last flip-flop 218.L is coupled to the L-1th flip-flop Output. The last flip-flop 161.L The output is sent back to the D input of the last flip-flop 218.L. The Q output of the last flip-flop 218.L constitutes the timer signal TM <l:1>The most significant bit TM(L).
[0460] refer to Fig.17 Detailed implementation of the integration stage 190 shown in FIG. 1 , the first inverter 197 of the first integration circuit 191 is a CMOS inverter formed by a series circuit of a PMOS transistor 220 and an NMOS transistor 221 coupled to each other at an output node 198. The PMOS transistor 220 and the NMOS transistor 221 receive the first oscillator control signal OS at their respective gate terminals. A .
[0461] The source of PMOS transistor 220 is coupled to input node 187 of integrating stage 190 of timer 45 .
[0462] The second inverter 200 of the first integrating circuit 191 is a CMOS inverter formed by a series circuit of a PMOS transistor 222 and an NMOS transistor 223 coupled to each other at a node 224 providing the first oscillator switching signal S′ 1 .
[0463] The first threshold V′ of the second inverter 200 th1 is the switching threshold of the second inverter 200 and thus depends on the characteristics, such as threshold or on-resistance, of the PMOS and NMOS transistors 222, 223. In practice, the switching threshold may be the input voltage at which the output of the inverter has a high logic value or otherwise the input voltage at which the output of the inverter has a low logic value.
[0464] For example, the switching threshold of the second inverter 200 can be defined as the corresponding input voltage (ie, the first oscillator integrated voltage V') A ) is equal to the operating point of the corresponding output voltage (ie, the first oscillator switching signal S'1).
[0465] The first inverter 202 of the second integrating circuit 192 is a CMOS inverter formed by a series circuit of a PMOS transistor 225 and an NMOS transistor 226 coupled to each other at an output node 203. The PMOS transistor 225 and the NMOS transistor 226 receive the second oscillator control signal OS at their respective gate terminals. B .
[0466] The source of PMOS transistor 225 is coupled to input node 187 of count portion 190 of timer 45 .
[0467] The second inverter 205 of the second integrating circuit 192 is a CMOS inverter formed by a series circuit of a PMOS transistor 227 and an NMOS transistor 228 coupled to each other at a node 229 providing a second oscillator switching signal S'2.
[0468] The second threshold V′ of the second inverter 205 th2 is the switching threshold of the second inverter 205 and thus depends on the characteristics, such as threshold or on-resistance, of the PMOS transistor 227 and the NMOS transistor 228. In practice, the switching threshold may be the input voltage at which the output of the inverter has a high logic value or otherwise the input voltage at which the output of the inverter has a low logic value.
[0469] For example, the switching threshold of the second inverter 205 can be defined as the corresponding input voltage (ie, the second oscillator integrated voltage V') B ) is equal to the operating point of the corresponding output voltage (ie, the second oscillator switching signal S'2).
[0470] like Fig.17 As shown in the detailed embodiment of the switch circuit 193, the first inverter 208 and the second inverter 209 are cross-coupled CMOS inverters, each of which includes a supply node (here at a voltage V') DD ) and the ground are coupled in series with each other.
[0471] In addition, the first and second inverters 208, 209 of the switch circuit 193 each include a corresponding enable switch, which is a PMOS transistor 232, which is coupled to the voltage V'. DD between the power supply node of the inverter and the PMOS transistor 230 of the corresponding inverter.
[0472] The PMOS transistor 232 is controlled by the oscillator enable signal EN′.
[0473] In this embodiment, reference Fig.17 The timer 45 includes a first stop switch and a second stop switch, here a first NMOS transistor 234 and a second NMOS transistor 235 , configured to interrupt the updating of the timer signal TM by the timer 45 .
[0474] In detail, the first and second NMOS transistors 234, 235 have drain terminals coupled to the output nodes 198, 203 of the first and second integration circuits 191, 192, respectively, and source terminals coupled to a reference (here, ground). The first and second NMOS transistors 234, 235 receive the count end signal STOP_CNT at their respective gate terminals.
[0475] When the end comparator 47 ( Figure 3 ) switches the count end signal STOP_CNT to a high logic value, the first and second NMOS transistors 234 and 235 are turned on, thereby short-circuiting the output nodes 198 and 203 of the first integration circuit 191 and the second integration circuit 192 to the ground.
[0476] Therefore, the timer 45 stops updating the timer signal TM.
[0477] In practice, in this embodiment, the timer 45, and in particular the corresponding integration stage 181, has the same Fig.13 The circuit diagram of any one of the digital detectors 16 is identical to the circuit diagram, in particular the corresponding integration stage 60.
[0478] Therefore, by comparing with any one of the digital detectors 16 according to the corresponding bit line current I BL,i Generate charge signal q i In the same way, the timer 45 is based on the oscillator current I OSC Generates timer signal TM <l:1>.
[0479] Thus, the timer 45 can generate the timer signal TM by executing a plurality of successive timing iterations. <l:1>In each timing iteration, for example, the oscillator current I OSC The integrator stage 190 generates a current I OSC The time-integrated first oscillator integrated voltage V' A , the first oscillator integrated voltage V' A With the first threshold V' th,1 is compared, and in response to the first oscillator integrated voltage V' A Reaching the first threshold V' th,1 Reset the first oscillator integrated voltage V' A , in particular, the first oscillator control signal OS is switched here A The counter stage 195 is responsive to the first oscillator integrated voltage V' A Reaching the first threshold V' th,1 The update timer signal TM <l:1>.
[0480] In this embodiment, the least significant bit of the timer signal TM is the second oscillator control signal OS B The value of .
[0481] In other words, the timer 45 can be switched by setting the oscillator current I OSC The oscillator current I is converted into a plurality of charge packets and counted. OSC Sampling is performed, wherein each charge packet corresponds to a charge accumulated on the capacitor 199 , 204 , which charge causes the switching of the second inverter 200 , 205 .
[0482] Therefore, the timer signal TM <l:1>The update frequency f u The first oscillator control signal OS A The frequency of switching events is given by (similar to reference Fig.13 For the first control signal IN A Therefore, the update frequency f u Depends on the oscillator current I OSC The value of, that is, depends on the bias current I b and the mirror factor p of the current mirror 180, the capacitance C' A , C' B and the first and second thresholds V' of the second inverters 200, 205 th,1 、V' th,2 .
[0483] In practice, the integrating stage 60 of the timer 45 behaves like a current controlled oscillator.
[0484] In use, the timer 45 and the digital detector 16 each extract the current I BL,i Generate the corresponding charge signal q i In the same way, the oscillator current I OSC Generates timer signal TM <l:1>The fact that the corresponding integrator circuits 60, 181 have the same circuit diagram makes it possible to <l:1>With the charge signal q 1 , ..., q M A strong correlation is obtained between them.
[0485] Therefore, global changes that may affect the IMC device 10, such as the supply voltage V DD 、V' DD Drift and / or temperature variations of the IMC device 10 are compensated by the timer 45 and the digital detector 16 without affecting the accuracy of the MAC operation performed by the IMC device 10 in this way.
[0486] By changing the oscillator current I OSC , for example by changing the bias current I via an external signal EXT b , can modify the timer signal TM <l:1>The update frequency f u , thus modifying the theoretical (and effective) refinement time of the IMC device 10.
[0487] In fact, for example, the oscillator current I OSC The increase in the oscillator integrated voltage V' A 、V' B ( Fig.13 and Fig.14 ) increases faster; therefore, the first and second oscillator control signals OSA, OSB switch faster, thereby also increasing the timer signal TM <l:1>The update frequency f u .
[0488] Finally, it is clear that modifications and variations may be made to what has been described and illustrated herein without departing from the scope of the present invention as defined in the appended claims.
[0489] For example, the theoretical duration *T j,+ With negative theoretical duration T* j,- The difference between the two may be a different corresponding input value x than described and illustrated above. j A function of the sign and absolute value of , for example, a linear function that is different from direct or inverse proportionality, or a nonlinear function.
[0490] For example, the positive current I + i,j With negative current I - i,j The difference between them may be a corresponding calculated weight G different from that described and illustrated above. ij A function of the sign and absolute value of , for example, a linear function that is different from direct or inverse proportionality, or a nonlinear function.
[0491] For example, to calculate the weight G ij In this case, the positive current I + i,j With negative current I - i,j The difference between them can be the corresponding calculation weight G ij is a function of the absolute value of .
[0492] For example, the input to the time converter 46 may be configured such that the difference T* j,+ -T* E1 Symbols and references Figure 4 and Figure 5 The sign of the description is opposite; that is, it can be j >0, then T* j,+ <T* E1 , and if x j <0, then T* j,+ >T* E1 In other words, it can be that if x j >0, then T* j,+ -T* j,- <0, and if x j >0, then T* j,+ -T* j,- >0.
[0493] For example, the cell group can be programmed so that the difference I + i,j -I ref The sign of G is opposite to that described above; that is, it can be ij >0, then I + i,j ref , and if G ij <0, then it is I + i,j >I ref , so if G ij >0, then I + i,j -I - i,j <0, and if G ij <0, then I + i,j -I - i,j >0.
[0494] For example, refer to Figure 4 At step S02, each input to the time converter 46 can verify the theoretical duration T* j,+ Is it only greater than (not greater than or equal to) the negative theoretical duration T* j,- In other words, in this case, if x j = 0 and therefore T* j,+ =T* j,- , then the input is sent to the time converter 46 to continue executing step S05 (instead of step S03).
[0495] For example, the memory cell 20 may be a multi-level memory cell, the number of levels of which is greater than three. Fig.16 The table shows that in the case where the cell 20 can be programmed in one of five different transconductance levels, the refinement interval T C Period and Unit Group 22 i,j The associated charge value Q ij .
[0496] exist Fig.18 In the example of FIG. 1 , the cell 20 can be programmed so that the corresponding current I + i,j ,I - i,j Five values equal to 0 μA, 1 μA, 2 μA, 3 μA or 4 μA can be assumed respectively.
[0497] In this example, we further derive I tot =4μA and I ref =I tot / 2=2μA.
[0498] In addition, also in this example, the input value x j Can present values between -64 and +64, T* C =128ns and T E1 =T* C / 2.
[0499] As from Fig.18 As can be seen in the example, for I + i,j =I - i,j =I - i,j =I ref and / or T* j,+ =T* j,- =T E1 , charge Q* i,j Presented with reference charge Q ref Same same value.
[0500] Product G ij ·x j Combinations with positive results and Q* i,j >Q ref Correspondingly, the product G ij ·x j Combinations with negative results and Q* i,j ref correspond.
[0501] For example, depending on the specific implementation of the memory cell and the corresponding calculation weight G to be mapped ij , each memory cell group of the computing array 12 may include a positive subset of cells and a negative subset of cells, wherein the number of cells in the positive subset of cells and the number of cells in the negative subset of cells is greater than one ( Figure 1 and Figure 2 ) or two ( Fig.11 ).
[0502] For example, the memory cells 20, 72, 97 may be resistive memory cells not based on PCM material but on a different technology; for example, they may be magnetoresistive memory cells (MRAM), resistive memory cells (RRAM) or static memory cells (SRAM).
[0503] The row activation unit 14 can also receive the address signal ADR ( Figure 3 The dashed arrow in the figure indicates that the effective refinement interval T C For example, the address signal ADR may be used to activate only some of the plurality of word lines during use, for example if the number of values of the input vector X is less than N.
[0504] For example, refer to Figure 1 , unit group 22 i,j The positive cell 23A and the negative cell 23B in the memory array 12 may be arranged in two rows that are not adjacent to each other in the memory array 12. i,j The positive cell 23A and the negative cell 23B in the memory array 12 may be arranged on two different columns of the memory array 12, and when in use, the two columns may be coupled together via a specific column selection circuit known per se.
[0505] For example, the IMC device 10 may include a number of digital detectors 16 that is less than the number of columns M of the array 12. In this case, the corresponding current I BL,1 , ..., I BL,M The starting charge signal q 1 , ..., q M The generation of can be controlled by a specially provided multiplexer circuit known per se.
[0506] For example, the bias circuit 18 may be configured to bias each column of the array 12 or each cell group 22. i,j Different bias voltages are generated so that each cell group 22 i,j The positive unit current I + i,j and negative cell current I - i,j Not only can the adjustment be made according to the corresponding programmed transconductance value, but it can also be adjusted according to the corresponding bias voltage received.
[0507] Finally, the different embodiments described above can be combined to provide further solutions.
Claims
1. An in-memory compute (IMC) device for performing a multiply and accumulate (MAC) operation, configured to receive an input signal indicating a plurality of input values and to provide at least one output signal, wherein each input value has a corresponding sign and a corresponding absolute value, the device comprising: a computational memory array comprising a plurality of memory cell groups coupled to the bit lines, each memory cell group configured to store a respective computational weight and comprising a positive memory cell coupled to a respective positive word line and a negative memory cell coupled to a respective negative word line; as well as a row activation circuit configured to receive an input signal and, for each input value, provide a positive activation signal having a positive activation duration to a positive word line of a corresponding memory cell group and provide a negative activation signal having a negative activation duration to a negative word line of a corresponding memory cell group during an active refinement interval, the sum of the positive activation duration and the negative activation duration being equal to a duration of the active refinement interval, the active refinement interval comprising a first refinement phase and a second refinement phase subsequent to the first refinement phase; The row activation circuit is also configured as follows: determining a positive theoretical duration of the positive activation signal and a negative theoretical duration of the negative activation signal, wherein the positive theoretical duration and the negative theoretical duration are each an algebraic sum of a reference value that is independent of the input value and a variable value that varies according to the sign and the absolute value of the input value, such that the sum of the respective positive theoretical duration and the respective negative theoretical duration is equal to the theoretical refinement duration; as well as first providing a positive activation signal and a negative activation signal whose theoretical duration is greater than or equal to a reference value during an effective refinement interval, wherein the activation duration of one of the positive activation signal and the negative activation signal is equal to the sum of an effective reference value and the variable value, the effective reference value being equal to the reference value reduced by a reduction factor, and wherein the activation duration of the other of the positive activation signal and the negative activation signal is equal to the corresponding theoretical duration; A first refinement phase of the effective refinement interval has a duration equal to the effective reference value; The positive memory cells in the corresponding memory cell group are configured to be activated by the corresponding positive activation signal and to have a positive cell current flow through the positive memory cells when activated; A negative memory cell in a corresponding memory cell group is configured to be activated by a corresponding negative activation signal and to have a negative cell current flow through the negative memory cell when activated; wherein, for each memory cell group, the positive cell current and the negative cell current are functions of corresponding calculated weights stored by the memory cell group; wherein a bit line current flows through the bit line during the active refinement interval, the current being a function of a sum of a positive cell current and a negative cell current of a group of memory cells coupled to the bit line; A column thinning circuit is coupled to the bit line and configured to sample the bit line current and provide the at least one output signal in response thereto, wherein the bit line current during the first thinning phase is sampled according to a reduction factor.
2. The IMC device of claim 1, wherein the reference value is equal to half of the theoretical refinement interval duration.
3. The IMC device according to claim 1, wherein: For each input value, the sign of the variable value is a function of the sign and / or absolute value of the corresponding input value.
4. The IMC device according to claim 1, wherein: For each input value, the sign of the variable value is a function of the sign of the corresponding input value, but not the absolute value of the corresponding input value.
5. The IMC device according to claim 1, wherein: For each input value, the absolute value of the variable value is a function of the sign and / or absolute value of the corresponding input value.
6. The IMC device according to claim 1, wherein: For each input value, the absolute value of the variable value is a function of the absolute value of the corresponding input value, but not the sign of the corresponding input value.
7. The IMC device according to claim 1, wherein: For each input value, the variable value is a monotonic function of the corresponding input value.
8. The IMC device according to claim 1, wherein: For each input value, the variable value is proportional to the corresponding input value.
9. The IMC device according to claim 1, wherein: For each input value, when the input value is equal to zero, the positive theoretical duration is equal to the negative theoretical duration.
10. The IMC device of claim 1, wherein the row activation circuit comprises: A timing circuit configured to provide a timing signal defining a valid refinement interval between a start time and an end time; as well as A plurality of input-to-time converters, one for each input value, each input-to-time converter configured to generate a signal indicative of one of a corresponding positive activation duration and a corresponding negative activation duration, compare the signal to the timing signal, and provide a corresponding positive activation signal and a corresponding negative activation signal in response.
11. The device according to claim 10, wherein: For each input value, the corresponding input-to-time converter is configured to provide the one of the positive activation signal and the negative activation signal from the start time of the valid refinement interval until the corresponding switching time included between the start time and the end time of the valid refinement interval, and to provide the other of the positive activation signal and the negative activation signal from the corresponding switching time until the end time of the valid refinement interval.
12. The IMC device of claim 1, wherein the column thinning circuit comprises: a digital detector configured to generate a measured charge signal indicative of an integral of a bit line current during an active refinement interval and to provide a charge signal in response thereto; Wherein the digital detector is further configured to multiply the measured charge signal by the reduction factor during the first refinement phase, so that the charge signal is indicative of the charge that would flow in the bit line during a theoretical refinement interval.
13. The IMC device of claim 12, wherein the column refinement circuit is further configured to compare the charge signal to a reference charge signal, and in response thereto determine the sign and absolute value of the result of the multiplication and accumulation operation, the reference charge signal being indicative of the charge that would flow in the bit line during a theoretical refinement interval if all input values were zero and / or if all calculated weights of the group of memory cells coupled to the bit line were zero.
14. The IMC device of claim 13, further comprising a reference circuit configured to generate a reference charge signal, comprising: a reference memory array comprising at least one group of memory cells coupled to a reference bit line and configured to receive at least one reference activation signal having a reference activation duration during an active refinement interval; wherein the reference current flowing through the reference bit line during the effective refinement interval corresponds to a charge that would flow in the bit line of the computational memory array during a theoretical refinement interval if all input values were zero and / or all computational weights of the memory cell groups coupled to the bit line of the computational memory array were zero; and A refinement circuit is configured to sample the reference bit current during a refinement interval and provide a reference charge signal in response thereto.
15. The IMC device of claim 1, wherein the positive memory cells and the negative memory cells of each memory cell group each comprise a corresponding multi-level memory cell configured to be programmed to one of three or more transconductance levels.
16. The IMC device of claim 1, wherein the positive memory cells and the negative memory cells of each memory cell group each comprise a respective two-level memory cell configured to be programmed to one of two transconductance levels.
17. The IMC device of claim 1, wherein the positive memory cells and the negative memory cells of each memory cell group comprise nonvolatile phase change memory cells, and wherein each nonvolatile phase change memory cell has a current path including a storage element and a selection element and extending between a common node and a reference potential node; The selection elements of the positive memory cell and the negative memory cell are configured to selectively close corresponding current paths in response to receiving a positive activation signal and a negative activation signal, respectively.
18. A method for controlling an in-memory compute IMC device to perform a multiply and accumulate MAC operation using a compute memory array, the compute memory array comprising a plurality of memory cell groups coupled to bit lines, the method comprising: storing a corresponding calculated weight having a sign and an absolute value in each memory cell group, wherein each memory cell group includes a positive memory cell coupled to a corresponding positive word line and a negative memory cell coupled to a corresponding negative word line; activating a positive memory cell by a corresponding positive activation signal on a corresponding positive word line; activating the negative memory cell by a corresponding negative activation signal on the corresponding negative word line; flowing a respective positive cell current through a respective positive memory cell of each memory cell group in response to a respective positive activation signal; flowing a respective negative cell current through a respective negative memory cell of each memory cell group in response to a respective negative activation signal; wherein, for each memory cell group, the positive cell current and the negative cell current are functions of the corresponding calculated weights; During an active refinement interval, in response to an input signal indicative of a plurality of input values, for each input value, providing a positive activation signal having a positive activation duration to a positive word line of a corresponding memory cell group and providing a negative activation signal having a negative activation duration to a negative word line of a corresponding memory cell group, wherein each input value has a corresponding sign and a corresponding absolute value; The positive activation signal and the negative activation signal are provided as follows: determining a positive theoretical duration of a positive activation signal and a negative theoretical duration of a negative activation signal; wherein the positive theoretical duration and the negative theoretical duration are each the algebraic sum of a reference value that is independent of the input value and a variable value that varies according to the sign and the absolute value of the respective input value, such that the sum of the respective positive theoretical duration and the respective negative theoretical duration is equal to the theoretical refinement duration; and first providing a positive activation signal and a negative activation signal during an effective refinement interval, wherein one of the positive activation signal and the negative activation signal whose theoretical duration is greater than or equal to a reference value has an activation duration equal to a sum of an effective reference value and a variable value, the effective reference value being equal to the reference value reduced by a reduction factor, and the other of the positive activation signal and the negative activation signal has an activation duration equal to the corresponding theoretical duration; wherein a first refinement phase of the effective refinement interval has a duration equal to the effective reference value; causing a bit line current to flow through a bit line coupled to the group of memory cells during an active refinement interval, the bit line current being a function of a sum of a positive cell current and a negative cell current of the group of memory cells coupled to the bit line; and The bit line current is sampled during the first refinement phase according to the reduction factor and at least one output signal is provided in response thereto.
19. A control method according to claim 18, wherein each calculation weight has a sign and an absolute value, each memory cell group is programmed so that the linear sum of the corresponding positive cell current and the corresponding negative cell current is equal to the total current, and the difference between the corresponding positive cell current and the corresponding negative cell current is a function of the sign and absolute value of the corresponding calculation weight.
20. The control method according to claim 19, wherein: For each memory cell group, the sign of the difference between the corresponding positive cell current and the corresponding negative cell current is a function of the sign and / or absolute value of the corresponding calculation weight.
21. The control method according to claim 19, wherein: For each memory cell group, the sign of the difference between the corresponding positive cell current and the corresponding negative cell current is a function of the sign of the corresponding calculation weight, but not a function of the absolute value of the corresponding calculation weight.
22. The control method according to claim 19, wherein: For each memory cell group, the absolute value of the difference between the corresponding positive cell current and the corresponding negative cell current is a function of the sign and / or absolute value of the corresponding calculation weight.
23. The control method according to claim 19, wherein: For each memory cell group, the absolute value of the difference between the corresponding positive cell current and the corresponding negative cell current is a function of the absolute value of the corresponding calculation weight, but not a function of the sign of the corresponding calculation weight.
24. The control method according to claim 19, wherein: For each memory cell group, the difference between the corresponding positive cell current and the corresponding negative cell current is a monotonic function of the corresponding calculation weight.
25. The control method according to claim 19, wherein: For each memory cell group, when the corresponding calculated weight is equal to zero, the corresponding positive cell current is equal to the corresponding negative cell current.
