Tcam array based on capacitive antiferroelectric nonvolatile memory and search method thereof
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIDIAN UNIV HANGZHOU RES INST
- Filing Date
- 2026-07-10
- Publication Date
- 2026-08-07
AI Technical Summary
[0005]但是，现有基于电容型存储器的研究主要集中于普通存储或存内计算功能，尚缺乏一种适合于仅具有两种稳定电容态的反铁电电容器件、并能够同时支持三态存储与汉明距离并行计算的TCAM结构与方法
[0020]2.可将不匹配位直接映射为固定差分输出：通过设置全高电容态参考匹配线，使匹配位和存储“X”位的差分输出自然抵消，而每个不匹配位产生固定幅值的差分电荷或差分电压，因此整行输出能够直接并行表征汉明距离。
Smart Images
Figure CN122531431A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semiconductor storage and in-memory computing technology, and in particular to a TCAM array based on capacitive antiferroelectric nonvolatile memory and its search method. Background Technology
[0002] Content-addressable memory (CAM) and ternary content-addressable memory (TCAM) can perform parallel comparisons between input query data and stored data and quickly output matching results, which has important application value in network routing, data retrieval, pattern recognition, artificial intelligence acceleration and other fields.
[0003] Existing TCAM (Tie-Ahead Comparison) systems mostly employ static random access memory (SRAM) cells in conjunction with comparator circuits, which suffers from problems such as large cell area, high static power consumption, and limited array density. To reduce area and power consumption, TCAM schemes based on non-volatile devices have been proposed in recent years, such as constructing matching cells using resistive memory, phase-change memory, and ferroelectric transistors. However, these schemes typically rely on differences in resistive state conduction, threshold voltage differences, or current discharge paths for searching, and still face problems such as large device variability, significant read / write disturbances, complex external decision-making, and insufficient array consistency.
[0004] On the other hand, capacitive memories based on ferroelectric or antiferroelectric capacitors have the potential for non-destructive reading, low static power consumption, and suitability for cross-array integration because they can store information by utilizing differences in device capacitance states and can be read out using small-signal methods. In particular, antiferroelectric capacitors can form distinguishable high-capacitance states and low-capacitance states through different polarization paths, achieving a large capacitance window at low operating voltages. Therefore, they provide a new device foundation for constructing charge-matching and search arrays.
[0005] However, existing research on capacitive memory mainly focuses on general storage or in-memory computation functions, lacking a TCAM structure and method suitable for antiferroelectric capacitors with only two stable capacitance states, capable of simultaneously supporting three-state storage and parallel Hamming distance computation. In particular, when using high-capacitance and low-capacitance states to represent logical information, the low-capacitance state still has non-zero capacitance, leading to charge baseline bias during parallel array readout. Furthermore, ensuring that neither the stored "X" bit nor the retrieved "X" bit participates in distance accumulation during three-state search is also a key problem that needs to be solved.
[0006] Therefore, a new antiferroelectric capacitor array structure and search method are proposed. Without relying on the multi-level stable states of a single device, this method utilizes only the two-state capacitance difference between the high-capacitance state and the low-capacitance state to achieve three-state information encoding, parallel search based on Hamming distance, and approximate matching. Summary of the Invention
[0007] The purpose of this invention is to address the shortcomings of the prior art by providing a TCAM array based on a capacitive antiferroelectric nonvolatile memory and its search method.
[0008] The objective of this invention is achieved through the following technical solution: a TCAM array based on capacitive antiferroelectric nonvolatile memory, comprising: a cross array consisting of multiple columns of search lines, multiple columns of reverse search lines and multiple rows of data matching lines, and multiple memory cells disposed at the intersection points of the cross array;
[0009] Each storage cell includes a first antiferroelectric capacitor and a second antiferroelectric capacitor. One end of the first antiferroelectric capacitor is electrically connected to the search line of the corresponding column, and one end of the second antiferroelectric capacitor is electrically connected to the reverse search line of the corresponding column. The other ends of the first antiferroelectric capacitor and the second antiferroelectric capacitor are electrically connected to the corresponding data matching line. The first and second antiferroelectric capacitors each have two non-volatile programmable states: a high-capacitance state and a low-capacitance state. By combining them, each memory cell can have three different storage states. It also includes at least one reference matching line that is laid out in parallel with the data matching line, and the reference matching line is connected to a plurality of reference cells that are consistent with the structure of the storage cell; each reference cell includes a first reference antiferroelectric capacitor and a second reference antiferroelectric capacitor, and both the first reference antiferroelectric capacitor and the second reference antiferroelectric capacitor are programmed to a fixed capacitance state. In the search operation, the search line and the reverse search line are used to load the query signal corresponding to the query data, so that the storage unit generates charge coupling to the corresponding data matching line and the reference unit generates charge coupling to the reference matching line; by comparing the differential output of the data matching line and the reference matching line, the Hamming distance correlation matching result between the query data and the stored data of that row is obtained.
[0010] Preferably, the three storage states of the storage unit include storage bit "1", storage bit "0" and storage bit "X" respectively; When the storage bit of the storage cell is "1", the first antiferroelectric capacitor is in a high capacitance state and the second antiferroelectric capacitor is in a low capacitance state. When the storage bit of the storage cell is "0", the first antiferroelectric capacitor is in a low capacitance state and the second antiferroelectric capacitor is in a high capacitance state. When the storage bit of the storage cell is "X", both the first antiferroelectric capacitor and the second antiferroelectric capacitor are in a high capacitance state; the storage bit "X" is a non-interest bit and does not participate in the accumulation calculation of the Hamming distance.
[0011] Preferably, the query signal includes three query bits: "1", "0", and "X". When the query bit is "1", the corresponding search line receives a first voltage, and the corresponding reverse search line receives a second voltage. When the query bit is "0", the corresponding search line receives a second voltage, and the corresponding reverse search line receives a first voltage. When the query bit is "X", both the corresponding search line and the corresponding reverse search line receive the second voltage. The first voltage is greater than the second voltage; the first voltage is the search voltage. The second voltage is zero voltage.
[0012] Preferably, the first and second reference antiferroelectric capacitors in each reference cell corresponding to the reference matching line are simultaneously programmed to a high-capacitance state, so that the differential output of the matching bit between the data matching line and the reference matching line is zero or close to zero, and the non-matching bit generates a differential output of fixed polarity between the data matching line and the reference matching line.
[0013] Preferably, the differential output satisfies the following condition: for each valid mismatch bit, the corresponding data matching line generates a fixed charge difference or fixed voltage difference relative to the reference matching line, which is related to the difference between the high capacitance state and the low capacitance state and the search voltage amplitude.
[0014] Preferably, the data matching line and the reference matching line are pre-charged to the same initial potential before the search operation is performed, and after the query signal is applied, the differential voltage or differential charge between the data matching line and the reference matching line is read by the differential sensing circuit; the differential sensing circuit is used to output an analog signal proportional to the number of valid mismatch bits in a row of memory cells to characterize the Hamming distance between the query data and the stored data.
[0015] The search method for TCAM arrays based on capacitive antiferroelectric nonvolatile memory includes the following specific steps: S1: Construct a TCAM array including storage cells, reference cells, search lines, reverse search lines, data matching lines, and reference matching lines; and program the first and second antiferroelectric capacitors in all reference cells to a high capacitance state; S2: Based on the data to be stored, program the capacitance states of the first antiferroelectric capacitor and the second antiferroelectric capacitor in the storage cell to complete the non-volatile storage of the data. S3: Before performing the search operation, precharge the data matching line and the reference matching line to the same initial potential; S4: Based on the query data, apply corresponding query signals to the search line and the reverse search line respectively, so that each memory cell and each reference cell generate charge coupling to the corresponding data matching line and the reference matching line. S5: Read the differential output between the data matching line and the reference matching line; S6: Obtain the Hamming distance correlation results between the queried data and the stored data based on the difference output.
[0016] Preferably, the specific method of step S1 is as follows: a cross array is formed by multiple column search lines, multiple column reverse search lines, and multiple row data matching lines. A storage unit is set at the intersection of the cross array. Each storage unit includes a first antiferroelectric capacitor and a second antiferroelectric capacitor. One end of the first antiferroelectric capacitor is connected to the search line of the corresponding column, and one end of the second antiferroelectric capacitor is connected to the reverse search line of the corresponding column. The other ends of the first antiferroelectric capacitor and the second antiferroelectric capacitor are connected together to the data matching line of the corresponding row. At least one reference matching line is laid parallel to the data matching line, and a reference unit with the same structure as the storage unit is set on the reference matching line.
[0017] Preferably, in step S6, the differential output is a differential charge signal or differential voltage signal between the data matching line and the reference matching line; the amplitude of the differential output is linearly positively correlated with the number of valid mismatch bits in the data row.
[0018] Preferably, in step S6, the Hamming distance correlation result is compared with a preset threshold. When the Hamming distance correlation result does not exceed the preset threshold, it is determined to be a valid search hit, thereby achieving approximate matching or fuzzy search.
[0019] The objective of this invention is achieved through the following technical solution: 1. Three-state storage and distance calculation can be achieved using only two-state antiferroelectric capacitors: This invention does not require a single antiferroelectric capacitor to have multiple stable capacitance states. It only relies on two states, high capacitance state and low capacitance state, and can achieve the three-state encoding of logic "1", "0" and "X" through differential combination of two capacitors, which reduces the difficulty of device implementation.
[0020] 2. Mismatched bits can be directly mapped to fixed differential outputs: By setting a full-high capacitance state reference matching line, the differential outputs of the matched bits and the stored "X" bits are naturally canceled out, while each mismatched bit generates a differential charge or differential voltage of a fixed amplitude. Therefore, the entire row output can directly represent the Hamming distance in parallel.
[0021] 3. Simultaneous shielding of storage "X" bit and query "X" bit: By encoding storage "X" as two antiferroelectric capacitors in a high capacitance state, and defining query "X" as zero voltage input to both search line and reverse search line, storage "X" bit and query "X" bit can be excluded from distance accumulation, thus supporting the three-state search mechanism of TCAM array.
[0022] 4. Suitable for parallel computation in the charge domain, with low power consumption and good array scalability: This invention adopts a charge-type readout method based on capacitive coupling, which does not rely on the traditional discharge-type current path, has low static power consumption, and is suitable for building high-density crossbar arrays, making it easy to realize parallel search and in-memory distance calculation.
[0023] 5. Supports exact and approximate matching: The differential output of this invention can not only be used to determine a perfect match, but also to achieve approximate matching and fuzzy search through threshold comparison. Therefore, it can be used as a TCAM or as a similarity search engine based on Hamming distance.
[0024] 6. Compatible with the non-destructive read characteristics of antiferroelectric capacitors: This invention is based on the difference in capacitance state of antiferroelectric capacitors for readout, which is suitable for non-destructive readout using small signal search voltage, thereby improving array read durability and system reliability. Attached Figure Description
[0025] Figure 1 This is a schematic diagram of the TCAM array of the present invention.
[0026] Figure 2 This is a schematic diagram of the storage unit of the present invention.
[0027] Figure 3 This is a schematic diagram illustrating different states during the search and addressing process of the present invention.
[0028] Figure 4 This diagram illustrates the writing and reading processes for the two states of the antiferroelectric capacitor: HCS (high capacitance state) and LCS (low capacitance state).
[0029] Figure 5 This is a schematic diagram of a differential sensing circuit.
[0030] Figure 6 This is the truth table for the data row cells.
[0031] Figure 7 For reference row cell truth table.
[0032] Figure 8 A truth table of the charge amount after deducting the reference row cell from the data row cell.
[0033] Figure 9This diagram illustrates the writing and reading processes for the two states of the antiferroelectric capacitor: HCS (high capacitance state) and LCS (low capacitance state).
[0034] Figure 10 This is a schematic diagram of the Hamming distance output by the TCAM array of the present invention. Detailed Implementation
[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.
[0036] Those skilled in the art should understand that, in the disclosure of this invention, the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the above terms should not be construed as limiting this invention.
[0037] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.
[0038] like Figures 1 to 10 As shown, the TCAM array based on capacitive antiferroelectric nonvolatile memory includes: a cross array consisting of multiple search lines, multiple reverse search lines and multiple rows of data matching lines, and multiple memory cells disposed at the intersections of the cross array; Each storage cell includes a first antiferroelectric capacitor and a second antiferroelectric capacitor. One end of the first antiferroelectric capacitor is electrically connected to the search line of the corresponding column, and one end of the second antiferroelectric capacitor is electrically connected to the reverse search line of the corresponding column. The other ends of the first antiferroelectric capacitor and the second antiferroelectric capacitor are electrically connected to the corresponding data matching line. The first and second antiferroelectric capacitors each have two non-volatile programmable states: a high-capacitance state and a low-capacitance state. By combining them, each memory cell can have three different storage states. It also includes at least one reference matching line that is laid out in parallel with the data matching line, and the reference matching line is connected to a plurality of reference cells that are consistent with the structure of the storage cell; each reference cell includes a first reference antiferroelectric capacitor and a second reference antiferroelectric capacitor, and both the first reference antiferroelectric capacitor and the second reference antiferroelectric capacitor are programmed to a fixed capacitance state. In the search operation, the search line and the reverse search line are used to load the query signal corresponding to the query data, so that the storage unit generates charge coupling to the corresponding data matching line and the reference unit generates charge coupling to the reference matching line; by comparing the differential output of the data matching line and the reference matching line, the Hamming distance correlation matching result between the query data and the stored data of that row is obtained.
[0039] The storage unit has three storage states: storage bit "1", storage bit "0", and storage bit "X". When the storage bit is "1", the first antiferroelectric capacitor is in a high capacitance state and the second antiferroelectric capacitor is in a low capacitance state. When the storage bit is "0", the first antiferroelectric capacitor is in a low capacitance state and the second antiferroelectric capacitor is in a high capacitance state. When the storage bit is "X", both the first and second antiferroelectric capacitors are in a high capacitance state. The storage bit "X" is not of interest and is not involved in the accumulation calculation of the Hamming distance.
[0040] In this application, the memory cell adopts a crossbar structure without gated transistors, and is composed of a first antiferroelectric capacitor and a second antiferroelectric capacitor.
[0041] Antiferroelectric capacitors utilize the built-in electric field / internal bias electric field within the antiferroelectric material to regulate the polarization state and capacitance state of the antiferroelectric capacitor, keeping it stably in a high capacitance state (HCS) or a low capacitance state (LCS), thereby preserving data for a long time without applying power and achieving non-volatile storage.
[0042] The writing and reading processes for the antiferroelectric capacitance HCS (high capacitance state) and LCS (low capacitance state) states can be found in the appendix. Figure 9 .
[0043] The first antiferroelectric capacitor is denoted as One end of it is connected to the search line ; The second antiferroelectric capacitance is denoted as One end of it is connected to the reverse search line. ; The other ends of the first and second antiferroelectric capacitors are connected together to the data matching line. .
[0044] Specifically, both the first and second antiferroelectric capacitors can be programmed as either a high-capacitance state (HCS) or a low-capacitance state (LCS). To achieve tri-state storage of the memory cell, the following encoding method is used: When the storage bit is "1", the storage cell is programmed to ; When the storage bit is "0", the storage cell is programmed as ; When storing bit "X", the unit is programmed to .
[0045] The storage bit "X" indicates that this bit is not of interest and should not contribute to the distance in subsequent search and distance calculation processes.
[0046] The query signal includes three query bits: "1", "0", and "X". When the query bit is "1", the corresponding search line receives a first voltage, and the corresponding reverse search line receives a second voltage. When the query bit is "0", the corresponding search line receives a second voltage, and the corresponding reverse search line receives a first voltage. When the query bit is "X", both the corresponding search line and the corresponding reverse search line receive the second voltage. The first voltage is greater than the second voltage; the first voltage is the search voltage. The second voltage is zero voltage.
[0047] Specifically, the query vector is obtained through the search line. With reverse search line Input. To ensure that the output directly represents the Hamming distance in the significant bits, the following search input encoding is used: When the query bit is "1", the input is: ; When the query bit is "0", the input is: ; When the query bit is "X", the input is: .
[0048] in, To search for the voltage amplitude, i.e., the first voltage.
[0049] In this input method: A query bit of "1" only activates the positive branch; A query bit of "0" only activates the reverse branch; The query bit "X" does not apply any excitation to this bit, so that this bit does not generate charge contribution in either the data match line or the reference match line, thus achieving shielding of this bit.
[0050] Therefore, searching for "X" indicates that the query bit is not involved in the Hamming distance calculation, rather than being a forced match level in the traditional sense.
[0051] To eliminate the non-zero baseline caused by the low capacitance state and to directly map the matching result to the number of mismatched bits, this embodiment further sets a reference matching line in the array. .
[0052] Reference Matching Line Set in parallel with the data matching line (data row), each corresponding reference cell also consists of two antiferroelectric capacitors, namely a first reference antiferroelectric capacitor and a second reference antiferroelectric capacitor, one end of which is connected to the search line of the corresponding column. and reverse search line The other ends of both are connected to the reference matching line. .
[0053] Furthermore, the first and second reference antiferroelectric capacitors in each reference cell corresponding to the reference matching line are simultaneously programmed to a high-capacitance state, so that the differential output of the matching bit between the data matching line and the reference matching line is zero or close to zero, and the non-matching bit generates a differential output of fixed polarity between the data matching line and the reference matching line.
[0054] The first reference antiferroelectric capacitance is denoted as The second reference antiferroelectric capacitance is denoted as In the reference cell, the first and second reference antiferroelectric capacitors are simultaneously programmed to a high-capacitance state, i.e. ; The reason for using the first and second reference antiferroelectric capacitors as full-capacitance state references is as follows: 1. It allows the matching bit and the storage bit "X" to cancel each other out naturally in the differential output; 2. It can generate a negative charge difference of fixed amplitude in the differential output for each mismatched bit; 3. Its output format directly corresponds to the number of mismatched bits, i.e., the Hamming distance; 4. When used as a reference, it has a large overall coupled charge, which is beneficial to improving reference stability and readout signal-to-noise ratio.
[0055] The differential output satisfies the following condition: for each valid mismatch bit, the corresponding data matching line generates a fixed charge difference or fixed voltage difference relative to the reference matching line, which is related to the difference between the high capacitance state and the low capacitance state and the search voltage amplitude.
[0056] The data matching line and the reference matching line are pre-charged to the same initial potential before the search operation is performed. After the query signal is applied, the differential voltage or differential charge between the data matching line and the reference matching line is read by a differential sensing circuit. The differential sensing circuit outputs an analog signal proportional to the number of valid mismatch bits in a row of memory cells to characterize the Hamming distance between the query data and the stored data. The differential sensing circuit is as follows: Figure 5 As shown.
[0057] Specifically, let the difference between the high capacitance state and the low capacitance state be... for: .
[0058] When the data bits match the query bits, the specific situation is as follows: When a certain storage cell has a storage bit of "1", that is And the query input is "1", that is hour: Then the storage cell contributes charge to the data matching line. for: ; The reference cell contributes charge to the reference matching line. for: ; Therefore, under the current conditions, the differential output... for: ; When the differential output is 0, the data bit matches the query bit, and this matching bit does not contribute to the Hamming distance.
[0059] Similarly, when storing "0" and searching for "0", the difference output is also 0.
[0060] When the data bits match the query bits, the specific situation is as follows: When a certain storage cell is stored as "1", that is... The query input is "0", which means... hour: Then the storage cell contributes charge to the data matching line. for: ; The reference cell contributes charge to the reference matching line. for: ; then: ; That is, each mismatched bit generates a fixed negative differential charge.
[0061] Similarly, storing "0" and searching for "1" will also generate negative differential charge of the same magnitude.
[0062] When the storage bit of a memory cell is "X", the memory cell is programmed as When the query input is "1" or "0", the contribution of its storage unit to the data matching line is the same as that of the reference unit. Therefore: ; The differential output is 0, which means that the storage bit "X" does not contribute to the Hamming distance.
[0063] When the query bit is "X", the query input is: At this time, neither the memory cell nor the reference cell injects charge into the matching line, therefore: ; This means that the query bit "X" is completely masked and does not contribute to the Hamming distance.
[0064] Furthermore, suppose a certain data row contains a total of Each storage unit (i.e., connected on the same data matching line) (number of storage cells), of which the number of valid mismatch bits is 1. The total differential charge output for this data row is: ; An equivalent load capacitor is connected to the rear end of the data matching line via a differential sensing circuit. Then its differential output voltage for: ; therefore: When the storage cells of the data row are fully matched, the differential output voltage... ; When a mismatch bit occurs, the differential output voltage ; When multiple mismatched bits occur, the output voltage amplitude is proportional to the number of mismatched bits.
[0065] Based on the above principle, the Hamming distance between the entire query word and the stored word can be accumulated in parallel, so that the difference output result will be proportional to the number of mismatched bits.
[0066] The present invention also provides a TCAM array based on a capacitive antiferroelectric nonvolatile memory, comprising the following specific steps: S1: Construct a TCAM array including storage cells, reference cells, search lines, reverse search lines, data matching lines, and reference matching lines; and program the first and second antiferroelectric capacitors in all reference cells to a high-capacitance state.
[0067] The specific method of this step is as follows: A cross array is formed by multiple column search lines, multiple column reverse search lines, and multiple row data matching lines. A storage unit is set at the intersection of the cross array. Each storage unit contains a first antiferroelectric capacitor and a second antiferroelectric capacitor. One end of the first antiferroelectric capacitor is connected to the search line of the corresponding column, and one end of the second antiferroelectric capacitor is connected to the reverse search line of the corresponding column. The other ends of the first antiferroelectric capacitor and the second antiferroelectric capacitor are connected together to the data matching line of the corresponding row. At least one reference matching line is laid parallel to the data matching line, and a reference unit with the same structure as the storage unit is set on the reference matching line.
[0068] S2: Based on the data to be stored, program the capacitance states of the first antiferroelectric capacitor and the second antiferroelectric capacitor in the storage cell to complete the non-volatile storage of the data.
[0069] S3: Before performing the search operation, precharge the data matching line and the reference matching line to the same initial potential.
[0070] In this step, the data matching line is... Matching line with reference Precharged to the same initial potential .
[0071] S4: Based on the query data, apply corresponding query signals to the search line and the reverse search line respectively, so that each memory cell and each reference cell generate charge coupling to the corresponding data matching line and reference matching line.
[0072] In this step, based on the value of each bit in the query word, corresponding voltages are applied to the corresponding search line and the reverse search line, as follows: Query bit "1": ; Query bit "0": ; Query bit "X": .
[0073] S5: Read the differential output between the data matching line and the reference matching line; All columns are capacitively coupled simultaneously, causing the storage cells to be connected to the data matching lines. Injecting charge, the reference cell to the reference matching line Injecting charge.
[0074] S6: Obtain the Hamming distance correlation results between the queried data and the stored data based on the difference output.
[0075] The differential output is a differential charge signal or differential voltage signal between the data matching line and the reference matching line; the amplitude of the differential output is linearly positively correlated with the number of valid mismatch bits in the data row.
[0076] It is worth mentioning that the TCAM array of the present invention has multiple rows of data matching lines and corresponding reference matching lines, so the array can support multi-row parallel search / matching and Hamming distance output.
[0077] The differential output voltage read by the differential sensing circuit is: ; The matching result is determined based on the difference output: when A value close to zero indicates a complete match; when When the value increases with the number of mismatched bits, it can directly represent the Hamming distance; Furthermore, the Hamming distance correlation results are compared with a preset threshold. When the Hamming distance correlation results do not exceed the preset threshold, it is determined to be a valid search hit, thereby achieving approximate matching or fuzzy search.
[0078] Compared with existing technologies, the present invention has the following advantages: 1. Three-state storage and distance calculation can be achieved using only two-state antiferroelectric capacitors: This invention does not require a single antiferroelectric capacitor to have multiple stable capacitance states. It only relies on two states, high capacitance state and low capacitance state, and can achieve the three-state encoding of logic "1", "0" and "X" through differential combination of two capacitors, which reduces the difficulty of device implementation.
[0079] 2. Mismatched bits can be directly mapped to fixed differential outputs: By setting a full-high capacitance state reference matching line, the differential outputs of the matched bits and the stored "X" bits are naturally canceled out, while each mismatched bit generates a differential charge or differential voltage of a fixed amplitude. Therefore, the entire row output can directly represent the Hamming distance in parallel.
[0080] 3. Simultaneous shielding of storage "X" bit and query "X" bit: By encoding storage "X" as two antiferroelectric capacitors in a high capacitance state, and defining query "X" as zero voltage input to both search line and reverse search line, storage "X" bit and query "X" bit can be excluded from distance accumulation, thus supporting the three-state search mechanism of TCAM array.
[0081] 4. Suitable for parallel computation in the charge domain, with low power consumption and good array scalability: This invention adopts a charge-type readout method based on capacitive coupling, which does not rely on the traditional discharge-type current path, has low static power consumption, and is suitable for building high-density crossbar arrays, making it easy to realize parallel search and in-memory distance calculation.
[0082] 5. Supports exact and approximate matching: The differential output of this invention can not only be used to determine a perfect match, but also to achieve approximate matching and fuzzy search through threshold comparison. Therefore, it can be used as a TCAM or as a similarity search engine based on Hamming distance.
[0083] 6. Compatible with the non-destructive read characteristics of antiferroelectric capacitors: This invention is based on the difference in capacitance state of antiferroelectric capacitors for readout, which is suitable for non-destructive readout using small signal search voltage, thereby improving array read durability and system reliability.
[0084] This invention is not limited to the preferred embodiments described above. Anyone can derive other products in various forms under the guidance of this invention. However, regardless of any changes in shape or structure, any technical solution that is the same as or similar to this application falls within the protection scope of this invention.
Claims
1. A TCAM array based on capacitive antiferroelectric nonvolatile memory, characterized in that, include: A cross array consisting of multiple search lines, multiple reverse search lines and multiple rows of data matching lines, and multiple storage units located at the intersections of the cross array; Each storage cell includes a first antiferroelectric capacitor and a second antiferroelectric capacitor. One end of the first antiferroelectric capacitor is electrically connected to the search line of the corresponding column, and one end of the second antiferroelectric capacitor is electrically connected to the reverse search line of the corresponding column. The other ends of the first antiferroelectric capacitor and the second antiferroelectric capacitor are electrically connected to the corresponding data matching line. The first and second antiferroelectric capacitors each have two non-volatile programmable states: a high-capacitance state and a low-capacitance state. By combining them, each memory cell can have three different storage states. It also includes at least one reference matching line that is laid out in parallel with the data matching line, and the reference matching line is connected to a plurality of reference cells that are consistent with the structure of the storage cell; each reference cell includes a first reference antiferroelectric capacitor and a second reference antiferroelectric capacitor, and both the first reference antiferroelectric capacitor and the second reference antiferroelectric capacitor are programmed to a fixed capacitance state. In the search operation, the search line and the reverse search line are used to load the query signal corresponding to the query data, so that the storage unit generates charge coupling to the corresponding data matching line and the reference unit generates charge coupling to the reference matching line; by comparing the differential output of the data matching line and the reference matching line, the Hamming distance correlation matching result between the query data and the stored data of that row is obtained.
2. The TCAM array based on capacitive antiferroelectric nonvolatile memory according to claim 1, characterized in that, The three storage states of a storage cell include storage bit "1", storage bit "0", and storage bit "X". When the storage bit of the memory cell is "1", the first antiferroelectric capacitor is in a high capacitance state and the second antiferroelectric capacitor is in a low capacitance state. When the storage bit of the storage cell is "0", the first antiferroelectric capacitor is in a low capacitance state and the second antiferroelectric capacitor is in a high capacitance state. When the storage bit of the storage cell is "X", both the first antiferroelectric capacitor and the second antiferroelectric capacitor are in a high capacitance state; the storage bit "X" is a non-interest bit and does not participate in the accumulation calculation of the Hamming distance.
3. The TCAM array based on capacitive antiferroelectric nonvolatile memory according to claim 1, characterized in that, The query signal includes three query bits: "1", "0", and "X". When the query bit is "1", the corresponding search line receives a first voltage, and the corresponding reverse search line receives a second voltage. When the query bit is "0", the corresponding search line receives a second voltage, and the corresponding reverse search line receives a first voltage. When the query bit is "X", both the corresponding search line and the corresponding reverse search line receive the second voltage. The first voltage is greater than the second voltage; the first voltage is the search voltage. The second voltage is zero voltage.
4. The TCAM array based on capacitive antiferroelectric nonvolatile memory according to claim 1, characterized in that, The first and second reference antiferroelectric capacitors in each reference cell corresponding to the reference matching line are simultaneously programmed to a high capacitance state, so that the differential output of the matching bit between the data matching line and the reference matching line is zero or close to zero, and the non-matching bit generates a differential output of fixed polarity between the data matching line and the reference matching line.
5. The TCAM array based on capacitive antiferroelectric nonvolatile memory according to claim 1, characterized in that, The differential output satisfies the following condition: for each valid mismatch bit, the corresponding data matching line generates a fixed charge difference or fixed voltage difference relative to the reference matching line, which is related to the difference between the high capacitance state and the low capacitance state and the search voltage amplitude.
6. The TCAM array based on capacitive antiferroelectric nonvolatile memory according to claim 1, characterized in that, The data matching line and the reference matching line are pre-charged to the same initial potential before the search operation is performed, and after the query signal is applied, the differential voltage or differential charge between the data matching line and the reference matching line is read by the differential sensing circuit. The differential sensing circuit is used to output an analog signal proportional to the number of valid mismatch bits in a row of memory cells to characterize the Hamming distance between the query data and the stored data.
7. A search method for a TCAM array based on a capacitive antiferroelectric nonvolatile memory according to claim 1, characterized in that, The specific steps include the following: S1: Construct a TCAM array including storage cells, reference cells, search lines, reverse search lines, data matching lines, and reference matching lines; and program the first and second antiferroelectric capacitors in all reference cells to a high capacitance state; S2: Based on the data to be stored, program the capacitance states of the first antiferroelectric capacitor and the second antiferroelectric capacitor in the storage cell to complete the non-volatile storage of the data. S3: Before performing the search operation, precharge the data matching line and the reference matching line to the same initial potential; S4: Based on the query data, apply corresponding query signals to the search line and the reverse search line respectively, so that each memory cell and each reference cell generate charge coupling to the corresponding data matching line and the reference matching line. S5: Read the differential output between the data matching line and the reference matching line; S6: Obtain the Hamming distance correlation results between the queried data and the stored data based on the difference output.
8. The search method for a TCAM array based on a capacitive antiferroelectric nonvolatile memory according to claim 7, characterized in that, The specific method of step S1 is as follows: a cross array is formed by multiple column search lines, multiple column reverse search lines and multiple row data matching lines. A storage unit is set at the intersection of the cross array. Each storage unit contains a first antiferroelectric capacitor and a second antiferroelectric capacitor. One end of the first antiferroelectric capacitor is connected to the search line of the corresponding column, one end of the second antiferroelectric capacitor is connected to the reverse search line of the corresponding column, and the other ends of the first antiferroelectric capacitor and the second antiferroelectric capacitor are connected together to the data matching line of the corresponding row. At least one reference matching line is laid out parallel to the data matching line, and a reference cell with the same structure as the storage cell is set on the reference matching line.
9. The search method for a TCAM array based on a capacitive antiferroelectric nonvolatile memory according to claim 7, characterized in that, In step S6, the differential output is a differential charge signal or differential voltage signal between the data matching line and the reference matching line; the amplitude of the differential output is linearly positively correlated with the number of valid mismatch bits in the data row.
10. The search method for a TCAM array based on a capacitive antiferroelectric nonvolatile memory according to claim 7, characterized in that, In step S6, the Hamming distance correlation result is compared with a preset threshold. When the Hamming distance correlation result does not exceed the preset threshold, it is determined to be a valid search hit, thereby achieving approximate matching or fuzzy search.