Method for implementing universal content addressable memory on basis of n-type and p-type ferroelectric field effect transistors

By using the complementary characteristics of N-type and P-type FeFETs in content addressable memory, the functions of TCAM, MCAM and ACAM are realized, solving hardware overhead and complexity problems in existing designs, improving storage density and energy efficiency, and expanding Application scenarios.

WO2024244312A1PCT designated stage expired Publication Date: 2024-12-05BEIJING SUPERSTRING ACAD OF MEMORY TECH +1
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2023/130733
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-26
Filing Date
2023-11-09
Publication Date
2024-12-05

AI Technical Summary

Technical Problem

Existing content addressable memory designs based on emerging nonvolatile memory require additional hardware overhead to implement multi-value and analog matching operations, resulting in complex designs and high power consumption, limiting the expansion of storage density and application scenarios.

Method used

Adopting a general design based on N-type and P-type ferroelectric field effect transistors (FeFETs), the complementary characteristics of FeFETs are used to realize three-state content addressable memory (TCAM) and multi-value content addressable without additional hardware overhead. Memory (MCAM) and analog content addressable memory (ACAM) functions to achieve matching and search operations by programming different threshold voltage states.

Benefits of technology

It improves storage density and search energy efficiency, simplifies the operation process, and expands the application scenario of content addressable memory, and realizes multi-functional storage and matching operations without increasing hardware overhead.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2023130733_05122024_PF_FP_ABST
    Figure CN2023130733_05122024_PF_FP_ABST
Patent Text Reader

Abstract

A method for implementing a universal content addressable memory on the basis of N-type and P-type ferroelectric field effect transistors, which relates to the technical fields of novel storage and computation. According to the method, the complementary characteristics of an N-type FeFET and a P-type FeFET are utilized, so that it is possible to simultaneously achieve the functions of a TCAM, MACM and ACAM without extra hardware overhead. Moreover, a simpler search operation is provided, so that the storage density and search energy efficiency of a CAM are improved. When quantized into an MCAM that stores multi-level entry states, the CAM also has the ability to compress entry states, allowing the storage density of the CAM to be further improved, which is of great significance for CAM-based table lookup search operations.
Need to check novelty before this filing date? Find Prior Art

Description

Method for realizing universal content addressable memory based on N-type and P-type ferroelectric field effect transistors Technical Field

[0001] The present invention relates to the field of novel storage and computing technology, and in particular to a universal content addressable memory design based on N-type and P-type ferroelectric field effect transistors. Background Art

[0002] Content-addressable memory (CAM) can efficiently and parallelize search operations and is widely used in efficient machine learning models, such as routers, database searches, in-memory computing, and neuromorphic computing. CAM is a specialized type of memory designed for parallel search. In addition to the read and write operations of conventional memory, it can also perform unique search operations. CAM was originally used to accelerate table lookups in network routers, such as those related to packet forwarding and classification. Because CAM can complete the entire search operation in a single clock cycle, it significantly accelerates other hardware- or software-based search systems. Further developments have led to the development of ternary content-addressable memory (TCAM), which can store a masked "X" state. TCAM can perform both exact and fuzzy matching, improving table lookup efficiency. In the era of big data, CAM's ability to match an input vector (query) against all stored vectors (entries) in a single search cycle and perform feature retrieval based on a distance metric based on the degree of mismatch makes it highly attractive for edge machine learning tasks such as pattern matching and video and image processing.

[0003] CAM designs based on traditional static random access memory (SRAM) occupy a large unit area, limiting their storage density for mapping computationally intensive algorithms. The resulting large parasitic capacitance further increases search latency and power consumption. CAM designs based on various emerging non-volatile memories, such as resistive random access memory (RRAM), phase change memory (PCM), and ferroelectric field-effect transistors (FeFETs), have reduced unit area, search latency, and energy consumption. Furthermore, multi-value content addressable memory (MCAM) and analog content addressable memory (ACAM), designed by further leveraging the multi-value storage capabilities of emerging non-volatile memories, can not only improve CAM storage density but also, based on ACAM's unique range matching operation, be used in a wider range of application scenarios, such as decision trees and deep random forests.

[0004] However, various CAM designs based on emerging non-volatile memories require different hardware overheads. MCAM and ACAM designs require additional hardware overhead, such as extra control transistors or complex peripheral circuitry to match analog or multi-value inputs with the memory range. This also complicates programming and search methods. Therefore, achieving compact and streamlined TCAM, MCAM, and ACAM designs without requiring additional hardware overhead is crucial for further increasing CAM density and expanding functionality. Summary of the Invention

[0005] In response to the problems existing in the above-mentioned prior art, the present invention proposes a universal CAM design based on N-type and P-type FeFETs. By utilizing the complementary characteristics of N-type FeFETs and P-type FeFETs, the functions of TCAM, MACM and ACAM can be simultaneously realized without the need for additional hardware overhead. It also has a simpler search operation, improves the storage density and search energy efficiency of the CAM, and expands the application scenarios of the CAM.

[0006] The technical solutions of the present invention are as follows:

[0007] A method for realizing a universal CAM based on complementary FeFET, characterized in that:

[0008] 1) The CAM cell consists of two parallel FeFETs, one of which is an N-type FeFET and the other is a P-type FeFET. The drains of the two FeFETs are connected as the CAM match line (ML), which is used to detect whether the search query matches the storage entry during the search operation. The sources of the two FeFETs are grounded. The gates of the two FeFETs are connected to two bit lines during programming and to the same search line (SL) during search to apply a search voltage corresponding to the search query.

[0009] 2) During the programming of the CAM storage entry, a certain programming voltage is applied to the gates of the N-type FeFET and the P-type FeFET respectively, programming them to different threshold voltage states. This is reflected in the transfer characteristic curve of the device as a shift along the gate voltage. The gate voltage corresponding to the minimum drain current is called V OFF For entry 0, program both FeFETs to V OFF0 State, for entry 1, program both FeFETs to V OFF1 state;

[0010] 3) In the search phase, for query 0, a voltage V is applied to SL SL0 =V OFF0 , for query 1, apply voltage V to SLSL1 =V OFF1 Therefore, only when the query matches the entry, both FeFETs in each CAM cell are in the off state with a very low drain current, indicating a match; otherwise, one FeFET is in the on state with a high drain current, indicating a mismatch. For entry X, the P-type FeFET is programmed to V OFF0 state, programming the N-type FeFET to V OFF1 state, then for any search query, both FeFETs are in the off state, indicating a match.

[0011] Furthermore, by programming the two FeFETs of the CAM cell to the same multiple V OFF state, which can be quantized into MCAM units, multiple V OFF The state represents a multi-level discrete storage entry, a quantified search query V SL Equal to the corresponding V OFF During the search, the corresponding search voltage V is applied to SL according to the search query. SL , only when the search query is consistent with the stored entry, V SL With V OFF Equal, the two FeFETs are in the off state. Further, the N-type FeFET and P-type FeFET of the MCAM unit are programmed to different V OFF state, then in two V OFF V between states SL The corresponding search queries are all matched, further realizing the function of entry compression.

[0012] Furthermore, by programming the threshold voltages of the N-type FeFET and the P-type FeFET of the CAM unit to V THN and V THP , where V THN >V THP , that is, the function of the ACAM unit can be realized. The matching range of the ACAM unit storage entry is [V THP , V THN ], the corresponding drain current is I Ref In the search phase, a certain voltage V is applied to SL SL , whose size corresponds to the search query, only when V SL Only within the matching range can the ML current of the ACAM unit be less than I Ref , indicating that the search query matches the storage entry range. SL Less than the lower boundary V of the matching range THPWhen the current of P-type FeFET is greater than I Ref , when V SL Greater than the upper boundary V of the matching range THN When the current of N-type FeFET is greater than I Ref , that is, the current of ML is greater than I Ref , indicating that the search query does not match the storage entry range, implementing the search operation of the ACAM unit.

[0013] Furthermore, several CAM cells form an array, and each row of CAM cells shares the ML. According to the input query vector, the corresponding analog search voltage is applied to all SL terminals at the same time. According to the different mismatches between the search query vector and the storage entry vector, each CAM cell will have a current that is positively correlated with the degree of mismatch of the bit. The ML of each row of the CAM array accumulates the current of each CAM cell. The magnitude of this current is positively correlated with the degree of mismatch between the search query and the storage entry vector. Based on this current magnitude, applications such as distance measurement can be implemented.

[0014] In summary, the complementary and symmetrical transfer characteristics of N-type FeFET and P-type FeFET can be used to implement TCAM, MCAM with entry compression, and ACAM functions without additional hardware overhead. Applications such as distance measurement can be further realized based on the CAM array.

[0015] The present invention proposes a universal CAM design based on N-type and P-type FeFETs, in which the ferroelectric material needs to adopt various HfO2-doped multi-domain ferroelectric materials such as HfO2 doped with Zr (HZO) and HfO2 doped with Al (HfAlO), and the device gate stack can be based on various structures such as MFMIS, MFIS, and MFS.

[0016] The technical effects of the present invention are as follows:

[0017] 1. The universal CAM design based on N-type and P-type FeFETs proposed in this invention utilizes the complementary and symmetrical transfer characteristics of N-type FeFET and P-type FeFET to implement TCAM units, and implements MCAM units and ACAM units without additional hardware overhead, thereby improving the storage density and search energy efficiency of CAM and expanding the application scenarios of CAM.

[0018] 2. The universal CAM design based on N-type and P-type FeFETs proposed in the present invention has the ability to compress entry states when quantized into an MCAM that stores multi-level entry states, so that it can further improve the storage density of the CAM, which is of great significance for CAM-based table lookup search operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] FIG1 is a schematic diagram of a unit structure of a universal CAM design based on N-type and P-type FeFETs of the present invention;

[0020] FIG2 is a functional schematic diagram of a universal CAM based on N-type and P-type FeFETs when used in a TCAM according to the present invention;

[0021] FIG3 is a functional schematic diagram of a universal CAM based on N-type and P-type FeFETs when used in an MCAM according to the present invention;

[0022] FIG4 is a functional schematic diagram of the universal CAM based on N-type and P-type FeFETs of the present invention when used as an ACAM. Implementation Method

[0023] The present invention will be further clearly and completely described below through specific embodiments in conjunction with the accompanying drawings.

[0024] The schematic diagram of the unit structure of the universal CAM design based on N-type and P-type FeFETs of the present invention is shown in Figure 1. The CAM unit consists of an N-type FeFET and a P-type FeFET connected in parallel. The drains of the two parallel FeFETs are connected to ML, which is used to detect whether the query and entry match during the search operation. The sources of the two parallel FeFETs are grounded. The gates of the two FeFETs are connected to two bit lines during the programming operation to apply the programming voltage; the gates of the two FeFETs are connected to the same SL during the search operation to apply the search voltage corresponding to the query. Figure 2 is a functional schematic diagram of the universal CAM based on N-type and P-type FeFETs of this embodiment when used for TCAM. During the programming stage, a certain programming voltage is applied to the N-type FeFET and the P-type FeFET respectively. For entry 0, the two FeFETs are programmed to V OFF0 State, for entry 1, program both FeFETs to V OFF1 State, for entry X, program the P-type FeFET to V OFF0 state, programming the N-type FeFET to V OFF1 In the search phase, for query 0, a voltage V is applied to SL. SL0 =V OFF0 , for query 1, apply voltage V to SL SL1 =V OFF1 Therefore, only when the search query and the storage entry are consistent, both FeFETs are in the off state, indicating a match.

[0025] Figure 3 is a functional schematic diagram of a universal CAM based on N-type and P-type FeFETs used in MCAM in this embodiment. Taking a 2-bit MCAM as an example, the CAM unit is programmed into four non-overlapping matching ranges, representing the entries "00", "01", "10", and "11" of the 2-bit MCAM unit, respectively. The search query is the center voltage value of the corresponding entry storage range. During the search, the corresponding search voltage is applied according to the query. Only when the search query matches the storage entry, V SL Only when it is within the matching range does it match, otherwise V SL It will be outside the matching range, and a large current will be detected at the ML end, indicating a mismatch. Furthermore, the MCAM cell entries "00" and "01" can be programmed as a matching range. This storage state then represents entry "0X," meaning that entry "0X" can match both search queries "00" and "01." Similarly, the MCAM cell entries "10" and "11" can be programmed as a matching range. This storage state then represents entry "1X," meaning that entry "1X" can match both search queries "10" and "11." Furthermore, "0X" and "10" can be programmed as a matching range and, similar to entry X in a TCAM, can be programmed to the "XX" state, indicating a match for all search queries.

[0026] FIG4 is a functional schematic diagram of a universal CAM based on N-type and P-type FeFETs when used as an ACAM in this embodiment. During the programming phase, a certain programming voltage is applied to the N-type FeFET and the P-type FeFET, respectively, and the threshold voltage of the N-type FeFET is programmed to V THN , programming the threshold voltage of the P-type FeFET to V THP , it is necessary to ensure V THP <V THN , indicating that the ACAM unit is programmed to match the range [V THP , V THN ] state, the corresponding drain current is I Ref In the search phase, a certain voltage V is applied to SL SL , whose size corresponds to the search query, only when V SL Only within the matching range can the ML current of the ACAM unit be less than I Ref , indicating that the search query matches the storage entry range. SL Less than the lower boundary V of the matching range THP When the current of P-type FeFET is greater than I Ref , when V SLGreater than the upper boundary V of the matching range THN When the current of N-type FeFET is greater than I Ref , that is, the current of ML is greater than I Ref , indicating that the search query does not match the storage entry range.

[0027] This embodiment fully and in detail explains how a universal CAM cell based on N-type and P-type FeFETs can implement TCAM, MCAM, and ACAM functions without requiring additional hardware overhead, thus expanding the application scenarios of CAM.

[0028] Finally, it should be noted that the purpose of disclosing the embodiments is to facilitate a further understanding of the present invention. However, those skilled in the art will appreciate that various substitutions and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the present invention should not be limited to the contents disclosed in the embodiments, and the scope of protection claimed by the present invention shall be determined by the scope defined in the claims.

Claims

1. A method for realizing a universal content addressable memory based on N-type and P-type ferroelectric field effect transistors, the steps comprising: 1) The content addressable memory CAM unit consists of two parallel FeFETs, one of which is an N-type FeFET and the other is a P-type FeFET. The drains of the two FeFETs are connected as the match line of the CAM unit to detect whether the search query matches the storage entry in the search operation. The sources of the two FeFETs are grounded together. The gates of the two FeFETs are connected to two bit lines respectively during programming and to the same search line during searching to apply a search voltage corresponding to the search query. 2) In the stage of programming the CAM cell storage entry, a certain programming voltage is applied to the gate of the N-type FeFET and the P-type FeFET respectively, programming them to different threshold voltage states, which is reflected in the transfer characteristic curve of the device as a shift along the gate voltage. The gate voltage corresponding to the minimum drain current is called V OFF , for entry 0, program both FeFETs to V OFF0 state, for entry 1, program both FeFETs to V OFF1 state, for entry X, the P-type FeFET is programmed to V OFF0 state, programming the N-type FeFET to V OFF1 state; 3) In the search phase, for query 0, a voltage V is applied to SL SL0 =V OFF0 , for query 1, apply voltage V to SL SL1 =V OFF1 When the query matches the entry, the two FeFETs of the CAM unit are in the off state, indicating that match; otherwise, one FeFET will be in the on state, indicating a mismatch.

2. The method according to claim 1, characterized in that The two FeFETs of the CAM cell are programmed to the same multiple V OFF state, that is, quantized into MCAM units, multiple V OFF The state represents multiple levels of discrete storage entries, and the V of the quantized search query SL Equal to the corresponding V OFF When searching, a corresponding search voltage V is applied to SL according to the search query. SL , V SL With V OFF are equal, and both FeFETs are in the off state.

3. The method according to claim 2, characterized in that Programming the N-type FeFET and P-type FeFET of the MCAM cell to different V OFF state, then in two V OFF V between states SL The corresponding search queries are all matched, realizing the function of entry compression.

4. The method according to claim 1, characterized in that The threshold voltages of the N-type FeFET and the P-type FeFET of the CAM cell are respectively programmed to V THN and V THP , where V THN >V THP , realize the function of ACAM unit, the matching range of ACAM unit storage entry is [V THP , V THN ], and its corresponding drain current is I Ref , in the search phase, a certain voltage V is applied to SL SL , whose size corresponds to the search query. When V SL Within the matching range, the ML current of the ACAM cell is less than I Ref , indicating that the search query matches the storage entry range; when V SL Less than the lower boundary V of the matching range THP When the current of P-type FeFET is greater than I Ref , when V SL Greater than the upper boundary V of the matching range THN When the current of N-type FeFET is greater than I Ref , that is, the current of ML is greater than I Ref , indicating that the search query does not match the storage entry range.

5. The method according to claim 1, characterized in that Several CAM cells form an array, and each row of CAM cells shares ML. According to the input query vector, corresponding analog search voltages are applied to all SL terminals at the same time. According to the different mismatches between the search query vector and the storage entry vector, each CAM cell will have a current that is positively correlated with the degree of mismatch of the bit. Each row of ML in the array accumulates the current of each CAM cell, and the magnitude of the current is positively correlated with the degree of mismatch between the search query and the storage entry vector. Distance measurement applications are implemented based on the magnitude of the current.

Citation Information

Patent Citations

  • High-energy-efficiency TCAM based on FeFET structure, and operation method thereof

    CN113053434A

  • Method for realizing analog and multi-valued content addressable memory based on pulse width modulation ferroelectric field effect transistor

    CN116110450A

  • Method for realizing universal content addressable memory based on N-type and P-type ferroelectric field effect transistors

    CN116580733A

  • NOR Type Content Addressable Memory Cell Based on Ferroelectric Element and Content Addressable Memory Including Same

    KR102481452B1