TWI868782B - In memory searching device - Google Patents
In memory searching device Download PDFInfo
- Publication number
- TWI868782B TWI868782B TW112125181A TW112125181A TWI868782B TW I868782 B TWI868782 B TW I868782B TW 112125181 A TW112125181 A TW 112125181A TW 112125181 A TW112125181 A TW 112125181A TW I868782 B TWI868782 B TW I868782B
- Authority
- TW
- Taiwan
- Prior art keywords
- data storage
- search
- memory
- data
- memory cell
- Prior art date
Links
- 238000013500 data storage Methods 0.000 claims abstract description 121
- 230000000875 corresponding effect Effects 0.000 claims description 30
- 230000002596 correlated effect Effects 0.000 claims description 4
- 210000004027 cell Anatomy 0.000 description 135
- 238000010586 diagram Methods 0.000 description 13
- 238000009826 distribution Methods 0.000 description 13
- 101000689199 Homo sapiens Src-like-adapter Proteins 0.000 description 6
- 102100024519 Src-like-adapter Human genes 0.000 description 6
- 101100049574 Human herpesvirus 6A (strain Uganda-1102) U5 gene Proteins 0.000 description 5
- 101150064834 ssl1 gene Proteins 0.000 description 5
- 101001121408 Homo sapiens L-amino-acid oxidase Proteins 0.000 description 4
- 102100026388 L-amino-acid oxidase Human genes 0.000 description 4
- 238000000034 method Methods 0.000 description 4
- 102100031885 General transcription and DNA repair factor IIH helicase subunit XPB Human genes 0.000 description 3
- 101000920748 Homo sapiens General transcription and DNA repair factor IIH helicase subunit XPB Proteins 0.000 description 3
- 102100028423 MAP6 domain-containing protein 1 Human genes 0.000 description 3
- 101710163760 MAP6 domain-containing protein 1 Proteins 0.000 description 3
- 101150020516 SLN1 gene Proteins 0.000 description 3
- 101000940558 Homo sapiens Chorionic somatomammotropin hormone-like 1 Proteins 0.000 description 1
- 101001122938 Homo sapiens Lysosomal protective protein Proteins 0.000 description 1
- 101100012902 Saccharomyces cerevisiae (strain ATCC 204508 / S288c) FIG2 gene Proteins 0.000 description 1
- 101100233916 Saccharomyces cerevisiae (strain ATCC 204508 / S288c) KAR5 gene Proteins 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 210000000352 storage cell Anatomy 0.000 description 1
Images
Landscapes
- Static Random-Access Memory (AREA)
- Semiconductor Memories (AREA)
- Read Only Memory (AREA)
Abstract
Description
本發明是有關於一種記憶體內搜尋裝置,且特別是有關於一種可提升搜尋效率的記憶體內搜尋裝置。 The present invention relates to an in-memory search device, and in particular to an in-memory search device capable of improving search efficiency.
隨著電子科技的進步,人工智慧的發展日漸成熟。隨之而來的,大量數據的搜尋需求被產生。據此,記憶體內搜尋裝置被推出。隨著所要搜尋的資料量的增加,如何應用記憶體內搜尋裝置,以針對大量的資料來執行資料搜尋動作,成為本領域技術人員的重要課題。 With the advancement of electronic technology, the development of artificial intelligence is becoming more mature. As a result, the need to search for large amounts of data has arisen. Accordingly, in-memory search devices have been introduced. As the amount of data to be searched increases, how to apply in-memory search devices to perform data search actions on large amounts of data has become an important topic for technicians in this field.
本發明提供一種記憶體內搜尋裝置,可針對記憶體進行分層式的資料搜尋,提升資料搜尋動作的效率。 The present invention provides a memory search device that can perform a hierarchical data search on the memory to improve the efficiency of the data search operation.
本發明的記憶體內搜尋裝置包括多個第一記憶胞串、控制器以及感測電路。第一記憶胞串共同耦接至第一共同位元線。各第一記憶胞串包括多個第一資料儲存層。第一資料儲存層分別包括多個第一記憶胞對,第一記憶胞對分別耦接至多個第一字元線 對。控制器耦接第一記憶胞串,選擇第一資料儲存層的其中之一為至少一選中資料儲存層,並提供搜尋資料至至少一選中資料儲存層對應的至少一選中字元線對。感測電路耦接第一共同位元線,感測第一共同位元線上的電流以產生搜尋結果。 The memory search device of the present invention includes a plurality of first memory cell strings, a controller, and a sensing circuit. The first memory cell strings are commonly coupled to a first common bit line. Each first memory cell string includes a plurality of first data storage layers. The first data storage layers respectively include a plurality of first memory cell pairs, and the first memory cell pairs are respectively coupled to a plurality of first word line pairs. The controller is coupled to the first memory cell strings, selects one of the first data storage layers as at least one selected data storage layer, and provides search data to at least one selected word line pair corresponding to at least one selected data storage layer. The sensing circuit is coupled to the first common bit line, and senses the current on the first common bit line to generate a search result.
基於上述,本發明的記憶體內搜尋裝置可將記憶體區分為多個資料儲存層。在進行資料搜尋動作時,可針對一層或多層的資料儲存層進行資料搜尋動作,可加速資料搜尋的效率。 Based on the above, the in-memory search device of the present invention can divide the memory into multiple data storage layers. When performing a data search action, the data search action can be performed on one or more data storage layers, which can speed up the efficiency of data search.
100、500:記憶體內搜尋裝置 100, 500: Search device in memory
110、120、411~41(N)、511~5AN:記憶胞串 110, 120, 411~41(N), 511~5AN: memory cell string
130:控制器 130: Controller
140:感測器 140:Sensor
501:感測器及編碼器 501: Sensor and encoder
BL、BL1~BLA:共同位元線 BL, BL1~BLA: common bit line
Is1、IA、IB:電流 Is1, IA, IB: current
M1~MZ:記憶胞 M1~MZ: memory cells
MC1、MC2:記憶胞 MC1, MC2: memory cells
SD:搜尋資料 SD: Search data
SL1~SLA、SLB、SL11~SLN1、SLA1~SLAN、610:選中資料儲存層 SL1~SLA, SLB, SL11~SLN1, SLA1~SLAN, 610: Select the data storage layer
SSL1、SSL2、GSL、CSL:選擇信號 SSL1, SSL2, GSL, CSL: Select signal
SW1、SW2、SW(N):選擇開關 SW1, SW2, SW(N): switch selection
VA、VB、Vmax、Vmin:搜尋電壓 VA, VB, Vmax, Vmin: Search voltage
Vg1、Vg2:偏壓電壓 Vg1, Vg2: bias voltage
VH1~VH4:電壓 VH1~VH4: voltage
Vpass:通過電壓 Vpass: pass voltage
VT_A、VT_B:電壓曲線 VT_A, VT_B: voltage curve
VT1、VT4:電壓分布 VT1, VT4: voltage distribution
WL11~WL2(Z)、WL1_1~WL1_512、WL2_1~WL2_512、WL3_1~WL3_512、WL4_1~WL4_512、WL89_1~WL89_512、WL90_1~WL90_512、WL89_1~WL89_512、WL90_1~WL90_512:字元線 WL11~WL2(Z), WL1_1~WL1_512, WL2_1~WL2_512, WL3_1~WL3_512, WL4_1~WL4_512 , WL89_1~WL89_512, WL90_1~WL90_512, WL89_1~WL89_512, WL90_1~WL90_512: character line
X、Y、Z:軸向 X, Y, Z: axial direction
圖1繪示本發明一實施例的記憶體內搜尋裝置的示意圖。 FIG1 is a schematic diagram of an in-memory search device according to an embodiment of the present invention.
圖2繪示本發明圖1實施例的記憶體內搜尋裝置的另一實施方式的示意圖。 FIG2 is a schematic diagram showing another implementation of the in-memory search device of the embodiment of FIG1 of the present invention.
圖3繪示本發明圖1實施例的記憶體內搜尋裝置的另一實施方式的示意圖。 FIG3 is a schematic diagram showing another implementation of the in-memory search device of the embodiment of FIG1 of the present invention.
圖4A以及圖4B繪示本發明實施例的記憶體內搜尋裝置的資料搜尋動作的示意圖。 FIG. 4A and FIG. 4B are schematic diagrams showing the data search action of the in-memory search device of an embodiment of the present invention.
圖5A至圖5C繪示本發明另一實施例的記憶體內搜尋裝置,以及資料搜尋動作的實施方式的示意圖。 Figures 5A to 5C are schematic diagrams showing another embodiment of the in-memory search device and the implementation of the data search action of the present invention.
圖6A至圖6D繪示本發明實施例的記憶體內搜尋裝置中,選中資料儲存層的資料搜尋動作的多個執行方式的示意圖。 Figures 6A to 6D are schematic diagrams showing multiple execution modes of a data search action for selecting a data storage layer in an in-memory search device according to an embodiment of the present invention.
請參照圖1,圖1繪示本發明一實施例的記憶體內搜尋裝置的示意圖。記憶體內搜尋裝置100包括多個記憶胞串110、120、控制器130以及感測器140。記憶胞串110、120共同耦接至共同位元線BL。其中,記憶胞串110透過選擇開關SW1以耦接至共同位元線BL,記憶胞串120透過選擇開關SW2以耦接至共同位元線BL。選擇開關SW1以及SW2分別受控於選擇信號SSL1以及SSL2以被導通或斷開。記憶胞串110、120可包括多個記憶胞,以記憶胞串110為例,記憶胞串110包括相互串聯的多個記憶胞M1~MZ。記憶胞M1~MZ可以為反及式快閃記憶胞。
Please refer to FIG. 1 , which shows a schematic diagram of an in-memory search device according to an embodiment of the present invention. The in-
記憶胞串110、120上各具有串聯耦接的多個記憶胞。其中,記憶胞串110上的多個記憶胞分別耦接至多條字元線WL11~WL1(Z)。記憶胞串120上的多個記憶胞則分別耦接至多條字元線WL21~WL2(Z)。在本實施例中,在記憶胞串110、120的每一者上的多個記憶胞可兩兩分成一組。其中在單一記憶胞串110、120中的兩個記憶胞,可形成一資料儲存層。同樣以記憶胞串110為範例,記憶胞M1、M2可形成資料儲存層;記憶胞M3、M4可形成另一資料儲存層;...;記憶胞MZ-1、MZ則可形成圖示中的最後一資料儲存層。
Each of the
在本發明其他實施例中,在單一記憶胞串中,也可應用非相鄰的二快閃記憶胞來形成資料儲存層。本實施例圖1的做法僅只是說明用的範例,不用以限縮本發明的範疇。 In other embodiments of the present invention, two non-adjacent flash memory cells can also be used in a single memory cell string to form a data storage layer. The method shown in FIG. 1 of this embodiment is only an example for illustration and is not intended to limit the scope of the present invention.
在另一方面,控制器130耦接至記憶胞串110、120。在執行資料搜尋動作時,控制器130可選擇記憶胞串110、120中的資料儲存層的至少其中之一來做為一個或多個選中資料儲存層。例如,在圖1中,在執行資料搜尋動作時,控制器130可選擇記憶胞串110中的記憶胞M1、M2所形成的資料儲存層來做為選中資料儲存層SL1。並且,控制器130並選擇記憶胞串120中,對應選中資料儲存層SL1的二記憶胞以做為選中資料儲存層SL2。
On the other hand, the
此外,控制器130可提供搜尋資料SD至選中資料儲存層SL1上的選中字元線對(包括字元線WL11、WL12)以及選中資料儲存層SL2上的選中字元線對(包括字元線WL21以及WL22),以針對選中資料儲存層SL1、SL2來進行搜尋資料SD的資料搜尋動作。其中,當搜尋資料SD為數位資料時,控制器130可分別提供搜尋資料SD的多個位元至選中資料儲存層SL1上的字元線WL11、WL12以及選中資料儲存層SL2上的字元線WL21以及WL22。其中,當搜尋資料SD為類比資料時,控制器130可分別提供搜尋資料SD的多個類比成分電壓至字元線WL11、WL12、WL21以及WL22。
In addition, the
本發明實施例的控制器130可以應用數位電路來建構。或可應用本領域具通常知識者所熟知的記憶體控制器來實施,沒有特定的限制。
The
在上述的資料搜尋動作中,選中資料儲存層SL1、SL2中的記憶胞,可根據所具有的儲存資料,來與字元線WL11、WL12、
WL21以及WL22上所接收的搜尋資料SD進行比對,並藉以產生搜尋結果。在本實施例中,選中資料儲存層SL1、SL2中的記憶胞,根據所具有的儲存資料與搜尋資料SD的比對動作,可在共同位元線BL上產生電流。本實施例中,耦接至共同位元線BL的感測器140,可透過感測共同位元線BL上的電流來產生搜尋結果。值得一提的,在本實施例中,搜尋結果所代表的資料相似度,與共同位元線BL上的電流正相關。
In the above data search action, the memory cells in the selected data storage layers SL1 and SL2 can be compared with the search data SD received on the word lines WL11, WL12,
WL21 and WL22 according to the storage data they have, and the search results are generated. In this embodiment, the memory cells in the selected data storage layers SL1 and SL2 can generate a current on the common bit line BL according to the comparison action between the storage data they have and the search data SD. In this embodiment, the
以下並請參照圖2,圖2繪示本發明圖1實施例的記憶體內搜尋裝置的另一實施方式的示意圖。在圖2中,記憶胞串110中的兩個資料儲存層被選中以做為選中資料儲存層SL1、SL3。相對應的,記憶胞串120中的兩個資料儲存層被選中以做為選中資料儲存層SL2、SL4。在執行資料搜尋動作時,搜尋資料可被同步提供至選中資料儲存層SL1、SL3對應的二選中字元線對(包括字元線WL11、WL12、WL13、WL14)以及選中資料儲存層SL2、SL4對應的二選中字元線對(包括字元線WL21、WL22、WL23、WL24)。如此一來,相對於圖1的實施方式,相對大的尺寸資料的資料搜尋動作可被執行,並有效提升資料搜尋動作的頻寬。
Please refer to FIG. 2 below, which is a schematic diagram of another embodiment of the in-memory search device of the embodiment of FIG. 1 of the present invention. In FIG. 2, two data storage layers in the
值得一提的,在本實施方式中,選中資料儲存層SL1、SL3可以是相鄰設置的資料儲存層。在本發明其他實施方式中,選中資料儲存層也可以是非位置相鄰的二資料儲存層。圖2的實施方式僅只是說明用的範例,不用以限縮本發明的實施範疇。 It is worth mentioning that in this embodiment, the selected data storage layers SL1 and SL3 can be adjacent data storage layers. In other embodiments of the present invention, the selected data storage layers can also be two data storage layers that are not adjacent in position. The embodiment of Figure 2 is only an example for illustration and is not intended to limit the scope of implementation of the present invention.
以下請參照圖3,圖3繪示本發明圖1實施例的記憶體內
搜尋裝置的另一實施方式的示意圖。在圖3中,記憶胞串110中的所有的資料儲存層都可以同時被選中以做為選中資料儲存層SL1~SLA。相對應的記憶胞串120中的所有的資料儲存層都可以同時被選中以做為選中資料儲存層SL2~SLB。
Please refer to FIG. 3 below, which is a schematic diagram of another embodiment of the memory search device of the embodiment of FIG. 1 of the present invention. In FIG. 3, all data storage layers in the
在執行資料搜尋動作時,搜尋資料可被同步提供至選中資料儲存層SL1~SLA對應的選中字元線對(包括字元線WL11~WL1(Z))以及選中資料儲存層SL2~SLB對應的選中字元線對(包括字元線WL21~WL2(Z))。如此一來,相對於圖1、2的實施方式,更大的尺寸資料的資料搜尋動作可被執行,有效提升資料搜尋動作的頻寬。 When performing a data search operation, the search data can be synchronously provided to the selected word line pairs (including word lines WL11~WL1(Z)) corresponding to the selected data storage layers SL1~SLA and the selected word line pairs (including word lines WL21~WL2(Z)) corresponding to the selected data storage layers SL2~SLB. In this way, compared with the implementation methods of Figures 1 and 2, data search operations of larger size data can be performed, effectively improving the bandwidth of the data search operation.
以下請參照圖4A以及圖4B,圖4A以及圖4B繪示本發明實施例的記憶體內搜尋裝置的資料搜尋動作的示意圖。在圖4A以及圖4B中,記憶體內搜尋裝置400包括多個記憶胞串411~41(N)。記憶胞串411~41(N)分別透過選擇開關SW1~SW(N)以共同耦接至共同位元線BL。選擇開關SW1~SW(N)分別受控於選擇信號SSL1~SSLN。在執行資料搜尋動作時,選擇開關SW1~SW(N)可分別根據選擇信號SSL1~SSLN而被同時導通。
Please refer to FIG. 4A and FIG. 4B below, which are schematic diagrams of the data search action of the in-memory search device of the embodiment of the present invention. In FIG. 4A and FIG. 4B, the in-
在本實施例中,如圖4A所示,在進行資料搜尋動作的第一間區間中,記憶胞串411~41(N)中,相對應的一個資料儲存層同時被設定為選中資料儲存層SL11~SLN1。控制器並可提供搜尋資料的多個成分至選中資料儲存層SL11~SLN1上的選中字元線對(包括字元線WL1_1~WL1_512以及字元線WL2_1~WL2_512)。
如此一來,記憶體內搜尋裝置400可完成第一層的資料搜尋動作。
In this embodiment, as shown in FIG. 4A , in the first interval of the data search operation, a corresponding data storage layer in the memory cell strings 411 to 41 (N) is simultaneously set as the selected data storage layer SL11 to SLN1. The controller can also provide multiple components of the search data to the selected word line pair (including word lines WL1_1 to WL1_512 and word lines WL2_1 to WL2_512) on the selected data storage layer SL11 to SLN1.
In this way, the in-
附帶一提的,在資料搜尋動作中,記憶體內搜尋裝置400可提供通過電壓Vpass至未被選中的資料儲存層的多個字元線上。此外,各個記憶胞串411~41(N)中並具有受控於選擇信號GSL、CSL的多個開關。
Incidentally, during the data search operation, the
接著,在圖4B中,在進行資料搜尋動作的第二間區間中,記憶胞串411~41(N)中,相對應的另一個資料儲存層同時被設定為選中資料儲存層SL12~SLN2。控制器並可提供搜尋資料的多個成分至選中資料儲存層SL12~SLN2上的選中字元線對(包括字元線WL3_1~WL3_512以及字元線WL4_1~WL4_512)。如此一來,記憶體內搜尋裝置400可完成第二層的資料搜尋動作。
Next, in FIG. 4B , in the second interval of the data search operation, another corresponding data storage layer in the
根據圖4A以及圖4B的說明加以類推,記憶體內搜尋裝置400可依據時間順序,逐一選擇記憶胞串411~41(N)中的各個資料儲存層來做為選中資料儲存層,並據以執行資料搜尋動作。
By analogy with the description of FIG. 4A and FIG. 4B , the in-
值得一提的,在本發明其他實施例中,在資料搜尋動作中同一時間區間中,對應各記憶胞串,也可選中多個的資料儲存層以做為選中資料儲存層,並同步的進行資料搜尋動作。如此一來,可有效提升資料搜尋動作的效率。 It is worth mentioning that in other embodiments of the present invention, in the same time period during the data search action, corresponding to each memory cell string, multiple data storage layers can also be selected as the selected data storage layer, and the data search action can be performed synchronously. In this way, the efficiency of the data search action can be effectively improved.
以下請參照圖5A至圖5C,圖5A至圖5C繪示本發明另一實施例的記憶體內搜尋裝置,以及資料搜尋動作的實施方式的示意圖。在圖5A中,記憶體內搜尋裝置500包括多個記憶胞串511~5AN以及感測器及編碼器501。記憶胞串511~51N透過所具
有的選擇開關以共同耦接至共同位元線BL1;記憶胞串521~52N透過所具有的選擇開關以共同耦接至共同位元線BL2;...;記憶胞串5A1~5AN透過所具有的選擇開關以共同耦接至共同位元線BLA。在本實施例中,記憶胞串511~5AN可排列成一立方體(cube)的結構。其中,記憶體內搜尋裝置500中,各共同位元線BL1~BLA沿Y軸方向延伸,各字元線沿X軸方向延伸,各個記憶胞串511~5AN的多個記憶胞則沿Z軸方向串聯耦接。此外,各個記憶胞串511~5AN中並具有受控於選擇信號GSL、CSL的多個開關
Please refer to FIG. 5A to FIG. 5C below, which are schematic diagrams showing an in-memory search device and an implementation method of a data search action according to another embodiment of the present invention. In FIG. 5A, the in-
感測器及編碼器501耦接至共同位元線BL1~BLA。其中感測器及編碼器501中的感測器用以感測共同位元線BL1~BLA上的電流以產生搜尋結果。感測器及編碼器501中的編碼器用以針對搜尋結果進行編碼以產生編碼後搜尋結果。
The sensor and
附帶一提的,感測器及編碼器501中的感測器可為感測放大器,並可應用本領域具通常知識者熟知的感測放大器電路來實施。感測器及編碼器501中的編碼器可以為一數位電路,並可設置在頁緩衝器(page buffer)中。
Incidentally, the sensor in the sensor and
在記憶體內搜尋裝置500執行資料搜尋動作時,在單一時間區間中,記憶體內搜尋裝置500可設定各個記憶胞串511~5AN中,位置相對應的一資料儲存層為選中資料儲存層。在本實施例中,記憶體內搜尋裝置500可針對記憶胞串511~51N分別設定選中資料儲存層SL11~SL1N;針對記憶胞串521~52N分別設定選中資料儲存層SL21~SL2N;...;以及針對記憶胞串5A1~5AN分別設
定選中資料儲存層SLA1~SLAN。
When the in-
在本實施例中,基於位置上的相互對應,選中資料儲存層SL11~SL1N分別與選中資料儲存層SL21~SL2N耦接相對的選中字元線對,選中資料儲存層SL21~SL2N也分別與選中資料儲存層SLA1~SLAN耦接相對的選中字元線對。 In this embodiment, based on the mutual correspondence in position, the selected data storage layers SL11~SL1N are respectively coupled to the selected word line pairs corresponding to the selected data storage layers SL21~SL2N, and the selected data storage layers SL21~SL2N are also respectively coupled to the selected word line pairs corresponding to the selected data storage layers SLA1~SLAN.
在另一方面,在資料搜尋動作中,搜尋資料SD可被拆分為多個部分,並分別被提供至選中資料儲存層SL11~SLAN所耦接的多個選中字元線對上。上述的多個選中字元線對包括字元線WL89_1~WL89_512以及字元線WL90_1~WL90_512。其中,字元線WL89_1~WL89_512分別與字元線WL90_1~WL90_512形成多個字元線對。 On the other hand, in the data search action, the search data SD can be split into multiple parts and provided to multiple selected word line pairs coupled to the selected data storage layers SL11~SLAN. The multiple selected word line pairs include word lines WL89_1~WL89_512 and word lines WL90_1~WL90_512. Among them, word lines WL89_1~WL89_512 form multiple word line pairs with word lines WL90_1~WL90_512 respectively.
此外,未被選中的資料儲存層的字元線則接收通過電壓Vpass。 In addition, the word lines of the unselected data storage layer receive the pass voltage Vpass.
在圖5A的資料搜尋動作中,各個記憶胞串511~5AN的選中資料儲存層的數量為1(n=1)。並且,在執行搜動作中,記憶體內搜尋裝置500可透過控制器,在不同的時間區間中,依據時間順序,依序設定各個記憶胞串511~5AN的一個資料儲存層來獲得選中資料儲存層,再透過傳送搜尋資料SD至選中資料儲存層的多個選中字元線對,以執行資料搜尋動作。
In the data search action of FIG. 5A , the number of selected data storage layers of each
在圖5B的實施方式中,記憶體內搜尋裝置500可在資料搜尋動作的同一時間區間中,設定各個記憶胞串511~5AN的選中資料儲存層的數量為2(n=2)。並透過將搜尋資料SD拆分為多個
成分,並分別提供至多個選中資料儲存層對應的多個選中字元線對,以進行資料搜尋動作。在本實施例中,選中字元線對包括字元線WL1_1~WL4_1、WL1_2~WL4_2、...、WL1_512~WL4_512。
In the implementation of FIG. 5B , the in-
在執行搜動作中,記憶體內搜尋裝置500可透過控制器,在不同的時間區間中,依據時間順序,依序設定各個記憶胞串511~5AN的兩個資料儲存層來分別獲得兩個選中資料儲存層,再透過傳送搜尋資料SD至選中資料儲存層的多個選中字元線對,以執行資料搜尋動作。
In executing the search operation, the in-
在圖5C的實施方式中,記憶體內搜尋裝置500可在資料搜尋動作中,同時設定各記憶胞串511~5AN所有的資料儲存層為選中資料儲存層。其中,選中資料儲存層的數量n=Z/2,Z為各記憶胞串511~5AN中記憶胞的數量。並且,透過將搜尋資料SD拆分為多個成分,並分別提供至多個選中資料儲存層對應的多個選中字元線對,以進行資料搜尋動作。如此一來,透過一次性的搜尋動作,就可完成對整個記憶體內搜尋裝置500的所有儲存資料的資料搜尋動作。
In the implementation of FIG. 5C , the in-
根據上述的說明,可以得知,本發明的記憶體內搜尋裝置500可根據時記的需求,透過設定選中資料儲存層的數量n,來進行資料搜尋動作,可有效提升資料搜尋動作的效率。
According to the above description, it can be known that the in-
以下請參照圖6A至圖6D,圖6A至圖6D繪示本發明實施例的記憶體內搜尋裝置中,選中資料儲存層的資料搜尋動作的多個執行方式的示意圖。在圖6A中,本發明實施例的選中資料儲
存層610包括記憶胞MC1以及MC2。記憶胞MC1以及MC2可分別為單階儲存胞元(single-level cell,SLC)、多階儲存胞元(multiple-level cell,MLC)、三階儲存胞元(triple-level cell,TLC)、四階儲存胞元(quad-level cell,QLC)、或儲存類比資料的儲存胞元。記憶胞MC1以及MC2的字元線可分別接收根據搜尋資料所產生的偏壓電壓Vg1、Vg2,並根據搜尋結果在共同位元線BL1上產生反應搜尋結果的電流Is1。
Please refer to FIG. 6A to FIG. 6D below, which are schematic diagrams showing multiple execution modes of the data search action of the selected data storage layer in the memory search device of the embodiment of the present invention. In FIG. 6A, the selected
在圖6B中,選中資料儲存層610可具有儲存資料DAT。儲存資料DAT包括第一位元D1以及第二位元D2。儲存資料DAT可根據記憶胞MC1以及MC2的臨界電壓Vth1、Vth2的電壓分布VT1至VT4來進行編碼,如下表所示:
在進行資料搜尋動作時,記憶胞MC1以及MC2的閘極透過對應的字元線所分別接收的電壓可以是電壓VH1~VH4的其中之一。記憶胞MC1以及MC2所接收的電壓根據對應的搜尋資料來產生,其中以兩位元的搜尋資料為範例,搜尋資料與電壓VH1~VH4的對應關係如表2:
在進行資料搜尋動作時,記憶胞MC1以及MC2可根據閘極所接收的電壓以及所具有的儲存資料間的電壓差,來在位元線BL1上產生電流Is1。 When performing a data search, memory cells MC1 and MC2 can generate a current Is1 on the bit line BL1 according to the voltage received by the gate and the voltage difference between the stored data.
接著請參照圖6C,以上述的電壓分布VT1~VT4分別為0V至1V、1.5V至2.5V、3V至4V、4.5V至5.5V,電壓VH1~VH4分別為5.5V、7V、8.5V以及10V為範例。其中電壓分布VT1~VT4的數量峰值分別對應電壓0.5V、2V、3.5V以及5V。 Next, please refer to Figure 6C, taking the above voltage distributions VT1~VT4 as 0V to 1V, 1.5V to 2.5V, 3V to 4V, 4.5V to 5.5V, and voltages VH1~VH4 as 5.5V, 7V, 8.5V, and 10V, as an example. The peak values of the voltage distributions VT1~VT4 correspond to voltages of 0.5V, 2V, 3.5V, and 5V, respectively.
在執行搜尋動作時,當搜尋資料為[0 0]時,記憶胞MC1、MC2的閘極分別接收的偏壓電壓Vg1、Vg2分別等於VH1、VH4。若此時記憶胞MC1、MC2具有的儲存資料為[0 0],則記憶胞MC1的臨界電壓Vth1在電壓分布VT1,記憶胞MC2的臨界電壓Vth2在電壓分布VT4。此時,記憶胞MC1上,偏壓電壓Vg1與臨界電壓Vth1間的差值可為5.5V-1V=4.5V。以每1.5V為一個層級L來計算,記憶胞MC1上偏壓電壓Vg1與臨界電壓Vth1間的差值可為3L。在記憶胞MC2上,偏壓電壓Vg2與臨界電壓Vth2間的差值可為10V-5.5V=4.5V。記憶胞MC2上偏壓電壓Vg2與臨界電壓Vth2間的差值可為3L。 When performing a search operation, when the search data is [0 0], the bias voltages Vg1 and Vg2 received by the gates of memory cells MC1 and MC2 are respectively equal to VH1 and VH4. If the storage data of memory cells MC1 and MC2 is [0 0] at this time, the critical voltage Vth1 of memory cell MC1 is in voltage distribution VT1, and the critical voltage Vth2 of memory cell MC2 is in voltage distribution VT4. At this time, on memory cell MC1, the difference between the bias voltage Vg1 and the critical voltage Vth1 can be 5.5V-1V=4.5V. Calculated with each 1.5V as a level L, the difference between the bias voltage Vg1 and the critical voltage Vth1 on the memory cell MC1 can be 3L. On the memory cell MC2, the difference between the bias voltage Vg2 and the critical voltage Vth2 can be 10V-5.5V=4.5V. The difference between the bias voltage Vg2 and the critical voltage Vth2 on the memory cell MC2 can be 3L.
也就是說,在此條件下,當搜尋資料與儲存資料相匹配時,記憶胞MC1以及MC2可產生對應3L的電流Is1。 That is to say, under this condition, when the search data matches the storage data, the memory cells MC1 and MC2 can generate a current Is1 corresponding to 3L.
在執行搜尋動作時,當搜尋資料為[0 0]時,記憶胞MC1、MC2的閘極分別接收的偏壓電壓Vg1、Vg2分別等於VH1、VH4。若此時記憶胞MC1、MC2具有的儲存資料為[0 1],則記憶胞MC1的臨界電壓Vth1在電壓分布VT2,記憶胞MC2的臨界電壓Vth2在電壓分布VT3。記憶胞MC1的偏壓電壓Vg1(=VH1)與臨界電壓Vth1(=VT2)間的差值可為2L。記憶胞MC2的偏壓電壓Vg2(=VH4)與臨界電壓Vth2(=VT3)間的差值可為4L。 When performing a search operation, when the search data is [0 0], the bias voltages Vg1 and Vg2 received by the gates of memory cells MC1 and MC2 are respectively equal to VH1 and VH4. If the storage data of memory cells MC1 and MC2 is [0 1] at this time, the critical voltage Vth1 of memory cell MC1 is in voltage distribution VT2, and the critical voltage Vth2 of memory cell MC2 is in voltage distribution VT3. The difference between the bias voltage Vg1 (=VH1) and the critical voltage Vth1 (=VT2) of memory cell MC1 can be 2L. The difference between the bias voltage Vg2 (=VH4) and the critical voltage Vth2 (=VT3) of the memory cell MC2 can be 4L.
基於記憶胞MC1、MC2是相互串連的,因此選中資料儲存層610可產生對應2L的電流Is1。
Since memory cells MC1 and MC2 are connected in series, selecting
在執行搜尋動作時,當搜尋資料為[0 0]時,記憶胞MC1、
MC2的閘極分別接收的偏壓電壓Vg1、Vg2分別等於VH1、VH4。若此時記憶胞MC1、MC2具有的儲存資料為[1 0],則記憶胞MC1的臨界電壓Vth1在電壓分布VT3,記憶胞MC2的臨界電壓Vth2在電壓分布VT2。記憶胞MC1的偏壓電壓Vg1(=VH1)與臨界電壓Vth1(=VT3)間的差值可為1L。記憶胞MC2的偏壓電壓Vg2(=VH4)與臨界電壓Vth2(=VT2)間的差值可為5L。基於記憶胞MC1、MC2是相互串連的,因此選中資料儲存層610可產生對應1L的電流Is1。
When performing a search operation, when the search data is [0 0], the bias voltages Vg1 and Vg2 received by the gates of memory cells MC1 and MC2 are respectively equal to VH1 and VH4. If the storage data of memory cells MC1 and MC2 is [1 0] at this time, the critical voltage Vth1 of memory cell MC1 is in voltage distribution VT3, and the critical voltage Vth2 of memory cell MC2 is in voltage distribution VT2. The difference between the bias voltage Vg1 (=VH1) and the critical voltage Vth1 (=VT3) of memory cell MC1 can be 1L. The difference between the bias voltage Vg2 (=VH4) and the critical voltage Vth2 (=VT2) of the memory cell MC2 can be 5L. Since the memory cells MC1 and MC2 are connected in series, the selected
在執行搜尋動作時,當搜尋資料為[0 0]時,記憶胞MC1、MC2的閘極分別接收的偏壓電壓Vg1、Vg2分別等於VH1、VH4。若此時記憶胞MC1、MC2具有的儲存資料為[1 1],則記憶胞MC1的臨界電壓Vth1在電壓分布VT4,記憶胞MC2的臨界電壓Vth2在電壓分布VT1。記憶胞MC1的偏壓電壓Vg1(=VH1)與臨界電壓Vth1(=VT4)間的差值可為0L。記憶胞MC2的偏壓電壓Vg2(=VH4)與臨界電壓Vth2(=VT1)間的差值可為6L。基於記憶胞MC1、MC2是相互串連的,因此選中資料儲存層610可產生對應0L的電流Is1。
When performing a search operation, when the search data is [0 0], the bias voltages Vg1 and Vg2 received by the gates of memory cells MC1 and MC2 are respectively equal to VH1 and VH4. If the storage data of memory cells MC1 and MC2 is [1 1] at this time, the critical voltage Vth1 of memory cell MC1 is in voltage distribution VT4, and the critical voltage Vth2 of memory cell MC2 is in voltage distribution VT1. The difference between the bias voltage Vg1 (=VH1) and the critical voltage Vth1 (=VT4) of memory cell MC1 can be 0L. The difference between the bias voltage Vg2 (=VH4) and the critical voltage Vth2 (=VT1) of the memory cell MC2 can be 6L. Since the memory cells MC1 and MC2 are connected in series, selecting the
由上述的說明不難得知,資料儲存層610所產生的電流Is1的大小,與搜尋資料和儲存資料間的相似度正相關。
From the above description, it is not difficult to know that the size of the current Is1 generated by the
值得一提的,只要適度的調整電壓分布的數量以及記憶胞MC1、MC2所接收的偏壓電壓的數量,本發明實施例選中資料儲存層610也可用以儲存多種不同階數的儲存資料,並執行相
對應階數的搜尋資料的搜尋動作。
It is worth mentioning that, as long as the number of voltage distributions and the number of bias voltages received by memory cells MC1 and MC2 are appropriately adjusted, the
在圖6D中,選中資料儲存層610中的記憶胞MC1、MC2也可用於記錄類比的儲存資料。並針對類比的搜尋資料進行資料搜尋動作。其中,記憶胞MC1的臨界電壓可以為電壓曲線VT_A,而記憶胞MC2的臨界電壓則可以為電壓曲線VT_B。其中,記憶胞MC1、MC2的可分別獨立被程式化以使臨界電壓分別對應電壓曲線VT_A、VT_B。
In FIG. 6D , the memory cells MC1 and MC2 in the
此外,電流IA、IB分別表示記憶胞MC1以及MC2的通道電流。 In addition, currents IA and IB represent the channel currents of memory cells MC1 and MC2, respectively.
在本實施例中,搜尋資料可以為搜尋電壓VA以及VB,其中搜尋電壓VA以及VB具有一關係式:VB=Vmax+Vmin-VA。其中Vmax、Vmin分別為搜尋動作中最大的類比搜尋電壓以及最小的類比搜尋電壓。在本實施例中,最大的類比搜尋電壓Vmax以及最小的類比搜尋電壓Vmin均為常數,例如分別為9V以及0V。 In this embodiment, the search data may be search voltages VA and VB, wherein the search voltages VA and VB have a relationship: VB=Vmax+Vmin-VA. Vmax and Vmin are respectively the maximum analog search voltage and the minimum analog search voltage in the search action. In this embodiment, the maximum analog search voltage Vmax and the minimum analog search voltage Vmin are both constants, such as 9V and 0V respectively.
當搜尋電壓VA等於0.7V且搜尋電壓VB等於8.3V時,記憶胞MC1不提供電流但記憶胞MC2可提供電流。當搜尋電壓VA等於4.05V且搜尋電壓VB等於4.95V時,記憶胞MC1與記憶胞MC2均不提供電流。當搜尋電壓VA等於8.52V且搜尋電壓VB等於0.48V時,記憶胞MC1提供電流但記憶胞MC2不提供電流。 When the search voltage VA is equal to 0.7V and the search voltage VB is equal to 8.3V, the memory cell MC1 does not provide current but the memory cell MC2 can provide current. When the search voltage VA is equal to 4.05V and the search voltage VB is equal to 4.95V, neither the memory cell MC1 nor the memory cell MC2 provides current. When the search voltage VA is equal to 8.52V and the search voltage VB is equal to 0.48V, the memory cell MC1 provides current but the memory cell MC2 does not provide current.
如此一來,選中資料儲存層610中的記憶胞MC1、MC2所產生的電流Is1的大小,可以做為所產生搜尋結果中,儲存資料
與搜尋資料間相似度的高低的依據。同樣的,電流Is1的大小與搜尋結果中儲存資料與搜尋資料間相似度正相關。
In this way, the magnitude of the current Is1 generated by the memory cells MC1 and MC2 in the selected
附帶一提的,在本發明其他實施例中,記憶胞MC1的臨界電壓與記憶胞MC2的臨界電壓的電壓曲線也可以相互交換。也就是說,記憶胞MC1的臨界電壓可具有電壓曲線VT_B,而記憶胞MC2的臨界電壓則可具有電壓曲線VT_A。 Incidentally, in other embodiments of the present invention, the voltage curves of the critical voltage of the memory cell MC1 and the critical voltage of the memory cell MC2 can also be interchanged. That is, the critical voltage of the memory cell MC1 can have a voltage curve VT_B, and the critical voltage of the memory cell MC2 can have a voltage curve VT_A.
綜上所述,本發明透過使記憶體內搜尋裝置設定記憶胞串中一個或多個資料儲存層來做為一個或多個選中資料儲存層。 In summary, the present invention enables the in-memory search device to set one or more data storage layers in the memory cell string as one or more selected data storage layers.
並同時提供搜尋資料的多個部分至一個或多個選中資料儲存層的選中字元線對上,可針對一個或多個選中資料儲存層同步執行資料搜尋動作。如此一來,記憶體內的資料搜尋動作的速率可以有效的增加,提升資料搜尋動作的效能。 And provide multiple parts of the search data to the selected character line pairs of one or more selected data storage layers at the same time, and perform data search actions on one or more selected data storage layers synchronously. In this way, the rate of data search actions in the memory can be effectively increased, improving the performance of data search actions.
100:記憶體內搜尋裝置 100: Search device in memory
110、120:記憶胞串 110, 120: memory cell string
130:控制器 130: Controller
140:感測器 140:Sensor
BL:共同位元線 BL: Common bit line
M1~MZ:記憶胞 M1~MZ: memory cells
SD:搜尋資料 SD: Search data
SL1、SL2:選中資料儲存層 SL1, SL2: Select the data storage layer
SSL1、SSL2:選擇信號 SSL1, SSL2: Select signal
SW1、SW2:選擇開關 SW1, SW2: switch selection
WL11~WL2(Z):字元線 WL11~WL2(Z): character line
Claims (14)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW112125181A TWI868782B (en) | 2023-07-06 | 2023-07-06 | In memory searching device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW112125181A TWI868782B (en) | 2023-07-06 | 2023-07-06 | In memory searching device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| TWI868782B true TWI868782B (en) | 2025-01-01 |
| TW202503748A TW202503748A (en) | 2025-01-16 |
Family
ID=95152137
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| TW112125181A TWI868782B (en) | 2023-07-06 | 2023-07-06 | In memory searching device |
Country Status (1)
| Country | Link |
|---|---|
| TW (1) | TWI868782B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090190404A1 (en) * | 2008-01-25 | 2009-07-30 | Roohparvar Frankie F | Nand flash content addressable memory |
| TW201104843A (en) * | 2009-05-07 | 2011-02-01 | Aplus Flash Technology Inc | A NAND based NMOS NOR flash memory cell, a NAND based NMOS NOR flash memory array, and a method of forming a NAND based NMOS NOR flash memory array |
| WO2020219122A1 (en) * | 2019-04-22 | 2020-10-29 | Western Digital Technologies, Inc. | Cam storage schemes and cam read operations for detecting matching keys with bit errors |
-
2023
- 2023-07-06 TW TW112125181A patent/TWI868782B/en active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090190404A1 (en) * | 2008-01-25 | 2009-07-30 | Roohparvar Frankie F | Nand flash content addressable memory |
| TW201104843A (en) * | 2009-05-07 | 2011-02-01 | Aplus Flash Technology Inc | A NAND based NMOS NOR flash memory cell, a NAND based NMOS NOR flash memory array, and a method of forming a NAND based NMOS NOR flash memory array |
| WO2020219122A1 (en) * | 2019-04-22 | 2020-10-29 | Western Digital Technologies, Inc. | Cam storage schemes and cam read operations for detecting matching keys with bit errors |
| US10910057B2 (en) * | 2019-04-22 | 2021-02-02 | Western Digital Technologies, Inc. | CAM storage schemes and CAM read operations for detecting matching keys with bit errors |
Also Published As
| Publication number | Publication date |
|---|---|
| TW202503748A (en) | 2025-01-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US7751240B2 (en) | Memory device with negative thresholds | |
| US9230666B2 (en) | Drain select gate voltage management | |
| CN113196401B (en) | Memory device that compensates for changes in programming speed due to block oxide thinning | |
| US9536603B2 (en) | Methods and apparatuses for determining threshold voltage shift | |
| US9543000B2 (en) | Determining soft data for combinations of memory cells | |
| US10325657B2 (en) | Non-volatile memory devices and methods of programming the same | |
| US20140313828A1 (en) | Sharing support circuitry in a memory | |
| CN103339677A (en) | Outputting a particular data quantization from memory | |
| US9105311B2 (en) | Inter-word-line programming in arrays of analog memory cells | |
| US12400706B2 (en) | In memory searching device | |
| TWI868782B (en) | In memory searching device | |
| KR101559336B1 (en) | Method of operating a memory device and a memory device therefor | |
| TWI860617B (en) | Memory device for computing in-memory | |
| KR20080029758A (en) | Flash memory device and program method thereof | |
| US9230680B2 (en) | Applications for inter-word-line programming | |
| JP7480391B2 (en) | Storage for in-memory computing | |
| Tseng et al. | An analog in-memory-search solution based on 3D-NAND flash memory for brain-inspired computing | |
| CN112017701B (en) | Threshold voltage adjusting device and threshold voltage adjusting method | |
| TWI827137B (en) | Memory device and data search method for in-memory search | |
| CN119229929B (en) | An analog domain content addressable memory system | |
| US12451173B2 (en) | Method for NAND flash memory to complete convolution operation of multi-bit data | |
| US20240274165A1 (en) | Memory device for performing in-memory-search and operating method thereof | |
| CN117409830A (en) | In-memory computing memory device and solid state drive module | |
| TW202433470A (en) | Hybrid ims cam cell, memory device and data search method | |
| JP2012203945A (en) | Semiconductor memory |