Content addressable memory device and operating method thereof
The proposed CAM device improves computational accuracy and integration by calculating propagation delay time using both rising and falling edges, resolving asymmetry and power consumption issues in TD CAMs.
Patent Information
- Application Number
- KR1020250175341
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2025-05-30
- Filing Date
- 2025-11-19
- Publication Date
- 2026-11-11
- Estimated Expiration
- 2045-11-19
AI Technical Summary
Conventional content-addressable memory (CAM) devices face limitations in miniaturization due to requiring multiple transistors per memory cell, increased circuit complexity, and power consumption, with asymmetric search performance affecting calculation accuracy in Time Domain (TD) CAMs.
A content-addressable memory device that calculates propagation delay time using both the rising and falling edges of a voltage pulse, employing a bipolar memory transistor and a priority encoder to determine the address of the row with the shortest propagation delay time, thereby improving computational accuracy and integration.
Enhances computational accuracy and integration by reducing technology nodes, addressing the asymmetry in search performance and power consumption issues of conventional single-edge search operations.
Smart Images

Figure 112025129275029-PAT00008_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a content addressable memory (CAM) device and a method of operating the same. Background Technology
[0002] Content Addressable Memory (CAM) devices do not perform sequential memory operations dependent on addresses; instead, they conduct comparison operations between all data stored in memory and the data to be searched in parallel based on content. Accordingly, CAM devices are highly useful for high-speed, large-capacity data retrieval, which has recently become a hot topic in fields such as big data and artificial intelligence.
[0003] Generally, content-addressable memory devices require two or more transistors per memory cell, which limits the miniaturization of content-addressable memory devices. Additionally, content-addressable memory devices must perform non-linear operations to determine the address where the data to be searched is stored, which leads to increased circuit complexity and power consumption.
[0004] To address these issues, Time Domain (TD) CAM devices are being developed. A TD CAM is a type of in-memory searching (IMS) device that performs bitwise XNOR operations in parallel on data stored in memory (entry vector) and query data (query vector) to be searched. As the degree of mismatch between the two resulting data increases, the computing pulse propagation delay time at the output increases linearly. Based on this, the distance difference between the stored data and the query data is calculated. The linear signal margin of a TD CAM contrasts with the non-linear signal margin of conventional Voltage Domain (VD) CAMs, enabling more accurate calculations and improving compatibility with digital circuits, which can reduce the area of surrounding circuits and power consumption. The accurate and area / energy-efficient IMS operation of TD CAM is applied to AI applications such as few-shot learning and hyperdimensional computing, and aims to efficiently perform in-memory computing (IMC) by reducing communication with the processing chip by performing part of the computation on the memory chip and resolving the energy / performance bottleneck caused by communication between the memory chip and the processing chip in the existing Von Neumann architecture.
[0005] The inventor of the present patent application has presented an operation process of a TD CAM device that determines a row identical to a codeword input into a CAM memory based on propagation delay time while using a bipolar memory transistor through prior research (Korean Published Patent No. 10-2025-0083029 (Title of Invention: Content Addressing Memory Device and Method of Operation thereof).
[0006] FIGS. 1 to 4 are drawings for explaining the operation of a conventional TD type CAM device.
[0007] Figure 1 is a graph illustrating the operating characteristics of a bipolar memory transistor included in each memory cell of a TD-type CAM device.
[0008] As shown in FIG. 4 or the structural diagram of the memory cell described later, a plurality of memory cells are connected to a match line (ML), wherein each memory cell includes a bipolar memory transistor. The vertical axis, gate capacitance value (Cgg1), indicates the capacitance value between the gate terminal and the ground terminal of the memory transistor, and the horizontal axis represents the gate voltage (Vg1).
[0009] A memory transistor may have a capacitance value smaller than a threshold value in a specific gate voltage range and a capacitance value larger than the threshold value at other gate voltages. For example, the capacitance value (Cgg1) of the memory transistor may be smaller than or equal to the threshold value (Cr) when the gate voltage level (Vg1) is greater than or equal to a first voltage level (V1) and less than or equal to a second voltage level (V2). Otherwise (less than the first voltage level or greater than the second voltage level), the capacitance value (Cgg1) may be greater than the threshold value (Cr). In other words, the memory transistor may have an electrical characteristic (i.e., bipolarity) of having small capacitance only at a specific gate voltage. In this case, the specific gate voltage range may vary depending on the programmed state (the bit value being stored) of the memory transistor. That is, when the memory transistor stores a bit value '1', the capacitance value (Cgg1) of the memory transistor may be greater than or equal to the threshold value (Cr) at the first voltage level (V1) and lower than or equal to the second voltage level (V2) as described above. Conversely, when the memory transistor stores a bit value '0', the capacitance value (Cgg1) of the memory transistor may be less than or equal to the threshold value (Cr) at the third voltage level (V3) and lower than or equal to the fourth voltage level (V4). Therefore, by using a gate voltage level (Vg) in which the capacitance value (Cgg) is distinguished according to the bit value stored by the memory transistor, it is possible to determine whether the stored bit value and the codeword match. The gate voltage level (Vg) can be determined based on the first search line signal, the first input signal, and the first complementary search line signal.
[0010] Figure 2 is a graph illustrating the propagation delay time of a conventional TD CAM device.
[0011] First, the input signal (IS1) is a signal input to the match line (ML), and the output signal (OS1) represents a signal output by the reading process of the CAM memory.
[0012] As described, the input signal (IS1) may include a voltage pulse. That is, the input signal (IS1) is a voltage pulse in which a rising edge appears at a first time point (t1) and a falling edge appears at a second time point (t2). Similar to the input signal (IS1), the output signal (OS1) may include a rising edge and a falling edge similar to the input signal (IS1), but differs from the input signal (IS1) as it propagates through a match line (ML) connected to a plurality of electrical elements (e.g., memory cells). For example, the output signal (OS1) experiences a delay equal to the propagation delay time compared to the input signal (IS1). Specifically, the falling edge of the output signal (OS1) may appear at a third time point (t3), which is a propagation delay time (i) after the second time point (t2) in which the falling edge of the input signal (IS1) appears. This propagation delay time (i) may increase as the value of the capacitance loaded on the match line (ML) increases. That is, as the number of mismatch bits increases, the propagation delay time (i) may increase.
[0013] Figure 2(b) is a diagram illustrating the propagation delay time according to the number of mismatch bits.
[0014] First to fifth cases with different numbers of mismatched bits are described. For example, assuming that a memory cell group includes first to fourth memory cells (C11 to C14), each of the five possible cases that may occur in the memory cell group is described.
[0015] The first case refers to a case where the data stored in all memory cells of a memory cell group matches the codeword, the second case refers to a case where the number of memory cells that do not match the codeword among all memory cells is 1, the third case refers to a case where the number of memory cells that do not match the codeword among all memory cells is 2, the fourth case refers to a case where the number of memory cells that do not match the codeword among all memory cells is 3, and the fifth case refers to a case where the number of memory cells that do not match the codeword among all memory cells is 4. And, as the number of mismatch bits increases, the capacitance value loaded on the match line increases linearly.
[0016] As shown in (b), the input signal (IS1) includes a falling edge at the second time point (t2).
[0017] The larger the capacitance value loaded on the first match line, the longer the propagation delay time becomes. That is, as the number of mismatch bits increases, the capacitance value between the first match line (ML1) and the ground terminal increases, and accordingly, the propagation delay time increases.
[0018] For example, in the first case where the number of mismatch bits is 0, the propagation delay time is time i between the second time point (t2) and the third time point (t3). In the second case where the number of mismatch bits is 1, the propagation delay time is time 2i between the second time point (t2) and the fourth time point (t4). In the third case where the number of mismatch bits is 2, the propagation delay time is time 3i between the second time point (t2) and the fifth time point (t5). In the fourth case where the number of mismatch bits is 3, the propagation delay time is time 4i between the second time point (t2) and the sixth time point (t6). In the fifth case where the number of mismatch bits is 4, the propagation delay time is time 5i between the second time point (t2) and the seventh time point (t7). That is, for every increase of one in the number of mismatch bits, the propagation delay time increases linearly by i.
[0019] According to this configuration, the CAM device of the prior art can output the address of a memory cell array that matches a codeword based on the propagation delay time. However, in the invention, the propagation delay time is calculated based on the falling edge during the process of checking the number of mismatch bits between the entry vector and the query vector stored in the CAM memory, and the following problems are presented as a result.
[0020]
[19] Referring to Fig. 3, the search for bit 1 stored in the memory cell is performed by p-type operation at a low search voltage (Vsearch1), and the search for bit 0 stored in the memory cell is performed by n-type operation at a high search voltage (Vsearch0), which caused a problem of difference in search performance. As described below, the bipolar memory transistor includes an n-type doping region and a p-type doping region, and as their operating characteristics are asymmetric, a difference in search performance may occur.
[0021] [Mathematical Formula 1]
[0022]
[0023] According to Equation 1, the falling edge delay is the C×dV of the locus shown in blue on the bipolar CV curve illustrated in (a). ML It is determined by the value. For the operation searching for bit 1, the left locus (p-type mode) illustrated in (a) is traversed in the direction of the arrow, and for the operation searching for bit 0, the right locus (n-type mode) is traversed, so C×dV for both cases ML Differences in values occur, and asymmetry arises in which propagation delay times differ even for identical calculation results, which degrades the calculation accuracy of the TD CAM and the system.
[0024] The present invention proposes a content addressing memory device and a method of operation thereof that can solve these conventional technical problems. Prior art literature
[0025] Republic of Korea Published Patent No. 10-2025-0083029 (Title of Invention: Content Addressing Memory Device and Method of Operation thereof) The problem to be solved
[0026] The present invention aims to solve the problems of the aforementioned prior art by proposing a content addressing memory device of a new structure that calculates propagation delay time by utilizing both the rising edge and the falling edge of a voltage pulse, and a method of operation thereof.
[0027] However, the technical problem that this embodiment aims to solve is not limited to the technical problem described above, and other technical problems may exist. means of solving the problem
[0028] As a technical means for achieving the technical problem described above, a content addressing memory device according to the first aspect of the present invention comprises: a memory cell array including a plurality of memory cells each connected across a plurality of rows and a plurality of columns; a voltage generator that provides a plurality of input signals for each row of the memory cell array; a driving circuit that provides a plurality of search signals for each column of the memory cell array; and a priority encoder that calculates a propagation delay time based on signals output from each row of the memory cell array and outputs the address of the row corresponding to the shortest propagation delay time as a match address. In this case, each memory cell includes a bipolar memory transistor, and when searching for a first bit stored in each memory cell, the propagation delay time is calculated based on the falling edge of the input voltage pulse and the output voltage pulse, and when searching for a second bit stored in each memory cell, the propagation delay time is calculated based on the rising edge of the input voltage pulse and the output voltage pulse.
[0029] Additionally, a method of operation of a content-addressing memory device according to a second aspect of the present invention comprises: a step of calculating a first propagation delay time based on the falling edge of an output signal of a match line while applying a first search voltage to a memory transistor at a position corresponding to a first bit included in an input codeword and deactivating the remaining memory transistors; a step of calculating a second propagation delay time based on the rising edge of an output signal of a match line while applying a second search voltage to a memory transistor at a position corresponding to a second bit included in an input codeword and deactivating the remaining memory transistors; and a step of outputting the address of the row with the shortest sum of the first propagation delay time and the second propagation delay time as a match address. Effects of the invention
[0030] According to the configuration of the present invention, it provides the advantages of reduced technology nodes (high integration) compared to conventional single-edge search operations and improved computational accuracy in the implementation of large-area arrays for large-scale data processing. Brief explanation of the drawing
[0031] FIGS. 1 to 4 are drawings for explaining the operation of a conventional TD type CAM device. FIG. 5 illustrates a content addressing memory device according to one embodiment of the present invention. FIG. 6 illustrates in detail a first memory cell group of a content addressing memory device according to one embodiment of the present invention. Figure 7 illustrates a bipolar memory transistor included in each memory cell. FIGS. 8 and 9 are drawings for explaining the operation of another CAM device in one embodiment of the present invention. FIG. 10 is a flowchart illustrating the operation method of a CAM device according to one embodiment of the present invention. FIGS. 11 and FIGS. 12 illustrate the experimental results of a dual-edge-based search operation of a CAM device according to one embodiment of the present invention. Specific details for implementing the invention
[0032] Embodiments of the present invention are described below in detail with reference to the attached drawings so that those skilled in the art can easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.
[0033] Throughout the specification, when a part is described as being "connected" to another part, this includes not only cases where they are "directly connected," but also cases where they are "electrically connected" with other components interposed between them. Furthermore, when a part is described as "including" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.
[0034] Throughout this specification, when a component is described as being located “on” another component, this includes not only cases where a component is in contact with another component, but also cases where another component exists between the two components.
[0035] Preferred embodiments of the present invention will be described in detail below with reference to the attached drawings and the contents described below. However, the present invention is not limited to the embodiments described herein and may be embodied in other forms. Throughout the specification, the same reference numerals indicate the same components.
[0036] FIG. 5 illustrates a content addressing memory device according to one embodiment of the present invention.
[0037] Referring to FIG. 5, a Content Addressable Memory (CAM; hereinafter referred to as a CAM device, 1000) may include a voltage generator (1100), a driver circuit (1200), a memory cell array (1300), and a priority encoder (1400). The memory cell array (1300) may include a plurality of rows and a plurality of columns, and may include a plurality of memory cells connected to each row and column. For example, the memory cell array (1300) may include first to N rows and first to M columns. In this case, N and M may each be any natural number. In FIG. 1, for convenience of explanation, it is illustrated as including first to fourth rows and first to fourth columns, but the number of each row and column may be varied in various ways.
[0038] In some embodiments, the CAM device (1000) may communicate with a processor (not shown). The processor may control a voltage generator (1100) and a driver circuit (1200). The processor may provide a codeword (or search data, input data) to the driver circuit (1200). The codeword is data to be searched among the data stored in the memory cell array (1300). Hereinafter, the term codeword is used interchangeably with search data and input data. For example, a codeword of length M may include first to M codes. In FIG. 1, for convenience of explanation, the processor is described as providing a codeword including first to fourth codes to the driver circuit (1200).
[0039] The voltage generator (1100) can generate first to N input signals under the control of the processor. For example, the voltage generator (1100) can provide first to fourth input signals (IS1 to IS4) to first to fourth memory cell groups (1300_1 to 1300_4) respectively through first to fourth match lines (ML1 to ML4). In some embodiments, each of the first to fourth input signals (IS1 to IS4) may include a voltage pulse. The driver circuit (1200) can receive a codeword from the processor. Based on the codeword, the driver circuit (1200) can generate first to fourth search signals and first to fourth complementary search signals corresponding to the codeword. The driver circuit (1200) can provide the first to fourth search signals to first to fourth columns respectively through first to fourth search lines (SL1 to SL4). The driver circuit (1200) can provide the first to fourth complementary search signals to the first to fourth columns, respectively, through the first to fourth complementary search lines (SL1'~SL4').
[0040] The first memory cell group (1300_1) may include first to fourth memory cells (C11–C14). The second memory cell group (1300_2) may include fifth to eighth memory cells (C21–C24). The third memory cell group (1300_3) may include ninth to twelfth memory cells (C31–C34). The fourth memory cell group (1300_4) may include thirteenth to sixteenth memory cells (C41–C44). Here, the number of memory cells illustrated is merely exemplary, and the number of memory cells may be determined according to various codeword sizes, such as 32 bits to 256 bits. Additionally, the number of search lines illustrated may change depending on the codeword size.
[0041] Each of the first to sixteenth memory cells (C11 to C44) may include a memory transistor. In some embodiments, the transistor included in each of the first to sixteenth memory cells (C11 to C44) may have an electrical characteristic (i.e., bipolarity) having a capacitance lower than a threshold value in a specific gate voltage level range (e.g., above a first voltage level and below a second voltage level) and a capacitance higher than a threshold value otherwise.
[0042] In some embodiments, the memory transistor included in each of the first to 16 memory cells (C11 to C44) may be a ferroelectric tunnel field-effect transistor (FeTFET). A more detailed description of the FeTFET device will be given later with reference to FIG. 7. In addition to the FeTFET device, various types of bipolar transistors having symmetric CV characteristics may be used as memory transistors for each memory cell.
[0043] In some embodiments, each of the first to sixteenth memory cells (C11 to C44) may store 1 bit or 2 bits. However, the scope of the present disclosure is not limited thereto, and the number of bits stored in the memory cells may increase or decrease depending on the implementation. For convenience of explanation, it is assumed below that each of the first to sixteenth memory cells (C11 to C44) stores 1 bit. For example, each of the first to sixteenth memory cells (C11 to C44) may store one of a bit value '1' and a bit value '0'.
[0044] In some embodiments, the CAM device (1000) may include first to fourth resistor elements (not shown) respectively connected between a voltage generator (1100) and a memory cell array (1300). Each resistor element may transmit a first input signal (IS1) received from the voltage generator (1100) to a first match line (ML1). The CAM device (1000) may improve the resolution of the priority encoder (1400) by adjusting the resistance values of the resistor elements. The resolution may correspond to the difference between the propagation delay times of the match lines (ML1 to ML4).
[0045] The priority encoder (1400) calculates the propagation delay time based on the signals output from each row of the memory cell array (1300) and outputs the address of the row corresponding to the shortest propagation delay time as the match address. The priority encoder (1400) can output the address of the memory cell array having data that matches the codeword as the match address (ML_add) based on the first to fourth output signals (OS1 to OS4) output from the first to fourth memory cell groups (1300_1 to 1300_4) of the memory cell array (1300), respectively. For example, the first to fourth output signals (OS1 to OS4) can each refer to the electrical signals that the first to fourth input signals (IS1 to IS4) have propagated through the match line and reached the priority encoder (1400). The first to fourth output signals (OS1 to OS4) may each include voltage pulses corresponding to the first to fourth input signals (IS1 to IS4).
[0046] For example, the priority encoder (1400) can determine a row among the first to fourth rows that matches a codeword based on the first to fourth output signals (OS1 to OS4). Specifically, it can determine a memory cell group among the first to fourth memory cell groups (1300_1 to 1300_4) in which the stored data (stored bits) matches a codeword. If there are multiple matching rows, the priority encoder (1400) can output the address of one row according to a priority algorithm.
[0047] As previously described in the background art, the priority encoder (1400) can determine the row matching the codeword based on the propagation delay time. However, unlike the prior art which uses only the falling edge, the present invention is characterized by calculating the propagation delay time by using both the rising edge and the falling edge of the voltage pulse.
[0048] The priority encoder (1400) can calculate a first propagation delay time based on the time difference between the first input signal (IS1) and the first output signal (OS1). Specifically, the length of the time interval between the first time point of the first input signal (IS1) and the second time point of the first output signal (OS1) can be set as the first propagation delay time. The first time point may be the time point when the rising edge of the first input signal (IS1) occurs, and the second time point may be the time point when the rising edge of the first output signal (OS1) occurs in response to the rising edge. Alternatively, the first time point may be the time point when the falling edge of the first input signal (IS1) occurs, and the second time point may be the time point when the falling edge of the first output signal (OS1) occurs in response to the falling edge. Similarly, the priority encoder (1400) can calculate the second to fourth propagation delay times based on the second to fourth input signals (IS2 to IS4) and the second to fourth output signals (OS2 to OS4).
[0049] Each of the first to fourth propagation delay times becomes longer as the number of bits (number of mismatched bits) in which the data stored by the corresponding row is inconsistent with the codeword increases. Accordingly, the priority encoder (1400) can determine the number of mismatched (or mismatched) bits between the data and the codeword stored by each of the first to fourth memory cell groups (1300_1 to 1300_4) based on the first to fourth propagation delay times. For example, the priority encoder (1400) can determine the number of mismatched bits between the data and the codeword stored by the first memory cell group (1300_1) based on the first propagation delay signal.
[0050] In some embodiments, the priority encoder (1400) may determine that the number of mismatch bits increases as the propagation delay time increases. For example, as the number of mismatch bits increases, the propagation delay time may increase linearly.
[0051] In some embodiments, the memory cell array (1300) may further include first to fourth input inverter circuits (not shown) respectively connected between the voltage generator (1100) and the first to fourth memory cell groups (1300_1 to 1300_4). The first to fourth input inverter circuits may each receive first to fourth input signals (IS1 to IS4) and may provide inverted input signals to the memory cell groups (1300_1 to 1300_4) respectively through the first to fourth match lines (ML1 to ML4). Since each of the first to fourth input inverter circuits is connected to the VDD voltage, the VDD voltage may each be provided to the first to fourth match lines (ML1 to ML4).
[0052] For example, the first inverter circuit may receive a first input signal (IS1), invert the first input signal (IS1), and provide the inverted first input signal to the first memory cell group (1300_1) through the first match line (ML1). The second to fourth inverter circuits may each operate in a similar manner.
[0053] In some embodiments, the memory cell array (1300) may further include first to fourth output inverter circuits (not shown) respectively connected between first to fourth memory cell groups (1300_1 to 1300_4) and a priority encoder (1400). The first to fourth output inverter circuits may each be connected to first to fourth match lines (ML1 to ML4) and may each provide first to fourth output signals (OS1 to OS4) to the priority encoder (1400). Since each of the first to fourth output inverter circuits is connected to a VDD voltage, they may each provide a VDD voltage to the first to fourth match lines (ML1 to ML4).
[0054] For example, the first output inverter circuit may be connected to the first match line (ML1) to receive the propagated first input signal (IS1) (or, if there is a first input inverter, the inverted first input signal) and provide the first output signal (OS1) to the priority encoder (1400). Each of the second to fourth output inverter circuits may operate in a similar manner.
[0055] In some embodiments, the memory cell array (1300) may further include first to fourth time-to-digital converter (TDC) circuits (not shown). The first to fourth TDC circuits may each be connected between the first to fourth memory cell groups (1300_1 to 1300_4) and the priority encoder (1400). As described above, if the first to fourth output inverter circuits are included, the first to fourth TDC circuits may be connected between the first to fourth output inverter circuits and the priority encoder (1400).
[0056] For example, the first TDC circuit may be connected between the first output inverter circuit and the priority encoder (1400). The first TDC circuit may receive a first output signal (OS1) from the first output inverter circuit. Based on the first input signal (IS1) and the first output signal (OS1), the first TDC circuit may convert the propagation delay time into digital data. The first TDC circuit may provide the digital data corresponding to the propagation delay time to the priority encoder (1400). Based on the digital data received from each of the first to fourth TDC circuits, the priority encoder (1400) may output the address of a match line that matches the codeword.
[0057] As another example, the first to fourth TDC circuits may be included in the priority encoder (1400). The driver circuit (1200) may provide a first search signal and a first complementary search signal corresponding to the first code '1' of the codeword to the first search line (SL1) and the first complementary search line (SL1'), respectively. Similarly, the driver circuit (1200) may provide a second search signal and a second complementary search signal corresponding to the second code '0' of the codeword to the second search line (SL2) and the second complementary search line (SL2'), respectively; provide a third search signal and a third complementary search signal corresponding to the third code '1' to the third search line (SL3) and the third complementary search line (SL3'), respectively; and provide a fourth search signal and a fourth complementary search signal corresponding to the fourth code '0' to the fourth search line (SL4) and the fourth complementary search line (SL4'), respectively. A more detailed explanation regarding this will be described later with reference to FIG. 4.
[0058] The first, fifth, ninth, and thirteenth memory cells (C11, C21, C31, C41) can each receive the first code '1' of the codeword through the first search line (SL1) and the first complementary search line (SL1'). The second, sixth, tenth, and fourteenth memory cells (C12, C22, C32, C42) can each receive the second code '0' through the second search line (SL2) and the second complementary search line (SL2'). The third, seventh, eleventh, and fifteenth memory cells (C13, C23, C33, C43) can each receive the third code '1' through the third search line (SL3) and the third complementary search line (SL3'). The 4th, 8th, 12th, and 16th memory cells (C14, C24, C34, C44) can each receive the 4th code '0' through the 4th search line (SL4) and the 4th complementary search line (SL4').
[0059] In some embodiments, each of the first to sixteenth memory cells (C11 to C44) may include an electrical element having a programmed state corresponding to a bit value being stored. The programmed state may refer to one of a first bit value and a second bit value. In other words, each of the first to sixteenth memory cells (C11 to C44) may store (or program) one of a first bit value and a second bit value. Each of the first to sixteenth memory cells (C11 to C44) may have a capacitance lower than a threshold value if the bit being stored corresponds to a codeword. For example, the capacitance may refer to the capacitance between the gate terminal and the ground terminal of a memory transistor.
[0060] For each of the first to fourth memory cell groups (1300_1 to 1300_4), the capacitance value between the match line and the ground terminal may increase as the number of mismatch bits increases. Accordingly, for each of the first to fourth memory cell groups (1300_1 to 1300_4), the propagation delay time may increase as the number of mismatch bits increases.
[0061] The priority encoder (1400) can output an address corresponding to the first memory cell group (1300_1) which has the smallest number of mismatch bits among the first to fourth memory cell groups (1300_1 to 1300_4) (e.g., storing data identical to the codeword '1 0 1 0') as a match address (ML_add).
[0062] In some embodiments, a CAM system including the CAM device (1000) of FIG. 5 and a processor controlling the same may be provided. For example, the CAM system may be implemented as an electronic system such as a PC, tablet, smartphone, data center, server device, etc.
[0063] FIG. 6 illustrates in detail a first memory cell group of a content addressing memory device according to one embodiment of the present invention, and FIG. 7 illustrates a bipolar memory transistor included in each memory cell.
[0064] The remaining memory cell groups (1300_2 to 1300_4) may be identical or similar to the first memory cell group (1300_1).
[0065] The first memory cell group (1300_1) can receive a first input signal (IS1) from the voltage generator (1100) through the first match line (ML1). The first memory cell group (1300_1) can output a first output signal (OS1) to the priority encoder (1400) through the first match line (ML1).
[0066] Each of the first to fourth memory cells (C11 to C14) may include first to fourth memory transistors (MT11 to MT14). The first memory transistor (MT11) may include a first source terminal connected to a first complementary search line (SL1'), a first gate terminal connected to a first match line (ML1), a first drain terminal connected to a first search line (SL1), and a first electrical element (EE1). Likewise, the second memory transistor (MT12) may include a second source terminal connected to a second complementary search line (SL2'), a second gate terminal connected to a first match line (ML1), a second drain terminal connected to a second search line (SL2), and a second electrical element (EE2).
[0067] Each of the first to fourth electrical elements (EE1 to EE4) can be programmed to store one of a first bit value and a second bit value. For example, the first electrical element (EE1) may correspond to a state of storing one of a bit value '1' and a bit value '0'. The other electrical elements (EE2 to EE4) are similar to the first electrical element (EE1).
[0068] In some embodiments, each of the first to fourth memory transistors (MT11 to MT14) may be a FeTFET as shown in FIG. 7, and each of the first to fourth electrical elements (EE1 to EE4) may refer to a ferroelectric layer of the FeTFET.
[0069] Referring to FIG. 7, the FeTFET device may include a P-doped region and an N-doped region formed at a certain distance apart on a semiconductor substrate, and a channel region formed between them. Additionally, it may include an insulating layer (SiO2) formed on the channel region, a ferroelectric layer formed on top of it, and a gate electrode formed on top of it. Furthermore, the P-doped region and the N-doped region may each operate as a source or a drain.
[0070] The channel region may be formed as an intrinsic region that is weakly doped with P-type impurities (P-region) or undoped, compared to the P-doped region. In one embodiment, the degree of doping in each doped region may be adjusted to improve turn-on operation and ambipolar operation.
[0071] The ferroelectric layer can be formed by including at least one of PZT (Pb(Zr,Ti)O3), SBT (SrBi2Ta2O9), SBTN (SrBi2(Ta,Nb)O9), BLT (Bix,La1- x)4Ti3O12), and BST (BaxSr(1-x)TiO3), hafnium oxide (HfO2), silicon-doped hafnium oxide (HfO2 / Si), aluminum-doped hafnium oxide (HfO2 / Al), zirconium-doped hafnium oxide (HfO2 / Zr), or a combination thereof.
[0072] A memory transistor of this structure can be programmed (or stored) with a bit value of '1' or '0'. For example, if a positive voltage is applied to the gate electrode and a ground voltage or low voltage is applied to the source and drain electrodes, the ferroelectric layer becomes down-polarized, and electrons can accumulate in the upper part of the channel region. Through this, the memory cell can be programmed with bit '1'. Conversely, if a negative voltage is applied to the gate electrode and a ground voltage or low voltage is applied to the source and drain electrodes, the ferroelectric layer becomes up-polarized, and holes can accumulate in the upper part of the channel region. Through this, the memory cell can be programmed with bit '0'.
[0073] In some embodiments, the capacitance value of the first memory transistor (MT11) may be smaller than a threshold value when the stored bit value corresponds to the first code of the codeword. Conversely, the capacitance value may be larger than a threshold value when the stored bit does not match the codeword. The driver circuit (1200) may provide a first search signal and a first complementary search signal corresponding to the first code of the codeword to the first memory cell group (1300_1), respectively, through the first search line (SL1) and the first complementary search line (SL1'). Similarly, when each of the second to fourth memory transistors (MT12 to MT14) stores a bit that matches the codeword, the capacitance value of the memory transistor may be smaller than a threshold value.
[0074] FIGS. 8 and 9 are drawings for explaining the operation of another CAM device in one embodiment of the present invention.
[0075] In order to resolve the problem arising from the conventional invention calculating the propagation delay time based only on the falling edge, the present invention is characterized by calculating the propagation delay time using a dual-edge method that considers both the rising edge and the falling edge.
[0076] For example, the propagation delay time can be calculated based on the rising edge for bit 1 stored in the memory transistor and based on the falling edge for bit 0 stored in the memory transistor. Or, conversely, the propagation delay time can be calculated based on the falling edge for bit 1 stored in the memory transistor and based on the rising edge for bit 0 stored in the memory transistor.
[0077] FIGS. 8 and 9 show a first search voltage (V) corresponding to bit 1 stored in a memory transistor. search1 ), or the second search voltage (V) corresponding to bit 0 search0 It illustrates that when a CAM operation is performed, if the bit stored in the memory transistor matches the bit being searched when the ) is applied to the search line (SL), a low capacitance (C) is output, and if it does not match, a high capacitance (C) is output. At this time, the voltage input to the match line is applied in the form of a voltage pulse as described in FIG. 2, and is applied to the gate of each memory transistor, and the propagation delay time can be determined by the capacitance value determined by the bit stored in each memory transistor. At this time, the propagation delay time is calculated based on the falling edge of the input voltage pulse and the falling edge of the output voltage pulse, or based on the rising edge of the input voltage pulse and the rising edge of the output voltage pulse.
[0078] First, referring to FIG. 8, the bit 1 search operation performed by the p-channel mode of the memory transistor utilizes the propagation delay time of the falling edge of the pulse trajectory shown in blue in the graph shown in (a). Then, as shown in (b), when searching for bit 1, a first search voltage (V) is applied to the search line (SL) of the memory transistor to be searched. search1 By applying ), the corresponding memory transistor is activated, and memory transistors that are not the search target are deactivated. For example, to deactivate memory transistors, an inhibit voltage is applied to the search line (SL) or a floating state is maintained. Then, a propagation delay time is calculated based on the falling edge of the input voltage pulse applied to the match line and the falling edge of the output voltage pulse output from the match line.
[0079] Additionally, referring to FIG. 9, the bit 0 search operation performed by the n-channel mode of the memory transistor utilizes the propagation delay time of the rising edge of the pulse trajectory shown in red in the graph shown in (a). And, as shown in (b), when searching for bit 0, a second search voltage (V) is applied to the search line (SL) of the memory transistor to be searched. search0 By applying ), the corresponding memory transistor is activated, and memory transistors that are not the search target are deactivated. For example, to deactivate memory transistors, an inhibit voltage is applied to the search line (SL) or a floating state is maintained. Then, a propagation delay time is calculated based on the rising edge of the input voltage pulse applied to the match line and the rising edge of the output voltage pulse output from the match line.
[0080] At this time, the activation status of each memory transistor is determined based on the codeword. For example, if the codeword is '1 0 1 0', first, to search for bit 1, the memory transistor connected to the search line corresponding to bit 1 is activated, and the remaining memory transistors are deactivated. Next, to search for bit 0, the memory transistor connected to the search line corresponding to bit 0 is activated, and the remaining memory transistors are deactivated.
[0081] Thus, unlike the prior art, a first search voltage (V) for searching for bit 1 search1 When ) is applied, the propagation delay time is calculated based on the falling edge of the voltage pulse, and a second search voltage (V) for searching for bit 0 is used. search0 When applying ), the propagation delay time is calculated based on the rising edge of the voltage pulse, so the asymmetry problem of the aforementioned conventional invention can be resolved.
[0082] In particular, if we examine Equation 2, which represents the delay on the falling edge, and Equation 3, which represents the delay on the rising edge, respectively C G *V G It can be confirmed that the amount of change is the same.
[0083] [Mathematical Formula 2]
[0084]
[0085] [Mathematical Formula 3]
[0086]
[0087] In this way, the priority encoder (1400) calculates the propagation delay time of the output signal based on the falling edge of the input voltage pulse and the output voltage pulse during the search for the first bit, calculates the first mismatch bit number from the calculated propagation delay time, calculates the propagation delay time of the output signal based on the rising edge of the input voltage pulse and the output voltage pulse during the search for the second bit, calculates the second mismatch bit number from the calculated propagation delay time, and finally determines the match address based on the sum of the first mismatch bit number and the second mismatch bit number. That is, the address of the row with the smallest sum of the first mismatch bit number and the second mismatch bit number can be determined as the match address.
[0088] At this time, the first bit is "1", the second bit is "0", and the first search voltage (V search1 ) is the second search voltage (V search2 It can be set to be smaller than ).
[0089] Also, as another embodiment, the first bit is "0" and the second bit is "1", and the first search voltage (V search0 ) is the second search voltage (V search1 It can be set to be smaller than ).
[0090] FIG. 10 is a flowchart illustrating the operation method of a CAM device according to one embodiment of the present invention.
[0091] First, a first search voltage is applied to the memory transistor at the position corresponding to the first bit included in the input codeword, and the remaining memory transistors are disabled, and a first propagation delay time is calculated based on the falling edge of the output signal of the match line (S110).
[0092] For example, as described in FIG. 8, to search for bit 1, a first search voltage (V) is applied to the search line (SL) of the memory transistor corresponding to the location of bit 1 in the codeword.search1 ) is applied, and the rest are disabled. Then, the first propagation delay time is calculated based on the difference between the falling edge of the input voltage pulse and the output voltage pulse of the match line.
[0093] Next, a second search voltage is applied to the memory transistor at the position corresponding to the second bit included in the input codeword, and with the remaining memory transistors disabled, a second propagation delay time is calculated based on the rising edge of the output signal of the match line (S120).
[0094] For example, as described in FIG. 9, to search for bit 0, a second search voltage (V) is applied to the search line (SL) of the memory transistor corresponding to the location of bit 0 in the codeword. search0 ) is applied, and the rest are disabled. Then, a second propagation delay time is calculated based on the difference between the rising edge of the input voltage pulse and the output voltage pulse of the match line.
[0095] Next, a match address is output based on the sum of the first propagation delay time and the second propagation delay time (S130).
[0096] For example, the address of the row with the shortest sum of the first propagation delay time and the second propagation delay time can be output as the match address.
[0097] In addition, a first mismatch bit count can be calculated from a first propagation delay time and a second mismatch bit count can be calculated from a second propagation delay time, and then a match address can be determined based on the smallest value of the sum of the first mismatch bit count and the second mismatch bit count.
[0098] FIGS. 11 and FIGS. 12 illustrate the experimental results of a dual-edge-based search operation of a CAM device according to one embodiment of the present invention.
[0099] First, looking at graphs (a) and (c) of Fig. 11, it can be seen that the propagation delay time of the output pulse for the search operation changes linearly with the number of mismatch bits.
[0100] In addition, looking at graphs (b) and (d) of Fig. 11, it can be seen that the falling edge delay or rising edge delay of the output pulse regarding the search operation changes linearly according to the number of mismatch bits.
[0101] In addition, Figures 12 (a) and (c) illustrate the propagation delay time in single-edge operation, and (b) and (d) illustrate the propagation delay time in double-edge operation. It can be seen that the distribution of propagation delay time in double-edge operation is significantly improved compared to single-edge operation, thereby enhancing the sensing margin and computational reliability.
[0102] According to these experimental results, in conventional single-edge operation, computational accuracy is expected to decrease significantly as device integration increases and memory cell array capacity increases, thereby limiting scalability. However, the dual-edge search operation of the present invention guarantees computational accuracy even in situations where device / circuit integration is high and the capacity of information stored and processed is high, thus enabling significantly improved scalability.
[0103] One embodiment of the present invention may also be implemented in the form of a non-transient recording medium comprising computer-executable instructions, such as program modules executed by a computer. A computer-readable medium may be any available medium accessible by a computer and includes both volatile and non-volatile media, and both removable and non-removable media. Additionally, a computer-readable medium may include all computer storage media. A computer storage medium includes both volatile and non-volatile, removable and non-removable media implemented by any method or technique for storing information, such as computer-readable instructions, data structures, program modules, or other data.
[0104] Although the method and system of the present invention have been described in relation to specific embodiments, some or all of their components or operations may be implemented using a computer system having a general-purpose hardware architecture.
[0105] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will understand that other specific forms can be easily modified without altering the technical spirit or essential features of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form.
[0106] The scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts thereof should be interpreted as being included within the scope of the present invention. Explanation of the symbols
[0107] 1000: Content-addressed memory device 1100: Voltage generator 1200: Driving circuit 1300: Memory cell array 1400: Priority Encoder
Claims
Claim 1 A content addressable memory (CAM) device comprising: a memory cell array including a plurality of memory cells each connected across a plurality of rows and a plurality of columns; a voltage generator each providing a plurality of input signals including an input voltage pulse for each row of the memory cell array; a driving circuit each providing a plurality of search signals for each column of the memory cell array; and a priority encoder that calculates a propagation delay time based on an output signal including an output voltage pulse output from each row of the memory cell array and outputs the address of the row corresponding to the shortest propagation delay time as a match address, wherein each memory cell includes a bipolar memory transistor, and when searching for a first bit stored in each memory cell, the propagation delay time is calculated based on the falling edge of the input voltage pulse and the output voltage pulse, and when searching for a second bit stored in each memory cell, the propagation delay time is calculated based on the rising edge of the input voltage pulse and the output voltage pulse. Claim 2 A content addressing memory device according to claim 1, wherein the driving circuit applies a first search voltage through a search line to a memory cell that is the search target of a first bit and performs deactivation processing on the remaining memory cells, applies a second search voltage through a search line to a memory cell that is the search target of a second bit and performs deactivation processing on the remaining memory cells. Claim 3 A content addressing memory device according to claim 1, wherein the priority encoder calculates the propagation delay time of an output signal output through a match line based on the falling edge of an input voltage pulse and an output voltage pulse when searching for the first bit, calculates the first mismatch bit number from the calculated propagation delay time, calculates the propagation delay time of an output signal output through a match line based on the rising edge of an input voltage pulse and an output voltage pulse when searching for the second bit, calculates the second mismatch bit number from the calculated propagation delay time, and determines the match address based on the sum of the first mismatch bit number and the second mismatch bit number. Claim 4 A content-addressing memory device according to claim 1, wherein the bipolar memory transistor is a FeTFET (ferroelectric tunnel field-effect transistor). Claim 5 A method of operation of a content addressing memory device comprising: (a) a step of applying a first search voltage to a memory transistor at a position corresponding to a first bit included in an input codeword and deactivating the remaining memory transistors, and calculating a first propagation delay time based on the falling edge of the output signal of a match line; (b) a step of applying a second search voltage to a memory transistor at a position corresponding to a second bit included in an input codeword and deactivating the remaining memory transistors, and calculating a second propagation delay time based on the rising edge of the output signal of a match line; and (c) a step of outputting a match address based on the sum of the first propagation delay time and the second propagation delay time. Claim 6 A method of operation of a content addressing memory device, wherein, in claim 5, the step of calculating the first propagation delay time calculates the propagation delay time based on the difference between the falling edges of each input voltage pulse and the output voltage pulse, and the step of calculating the second propagation delay time calculates the propagation delay time based on the difference between the rising edges of each input voltage pulse and the output voltage pulse. Claim 7 A method of operation of a content addressing memory device, wherein in claim 5, step (c) calculates a first mismatch bit number from the calculated first propagation delay time, calculates a second mismatch bit number from the calculated second propagation delay time, and determines the match address based on the sum of the first mismatch bit number and the second mismatch bit number.
Citation Information
Patent Citations
Content addressable memory device and operating method thereof
KR1020230029536A
