10T-SRAM unit, dual-channel reading and content addressing logic circuit and chip thereof
By designing the CIM circuit with dual-channel reading and content addressing functions of the 10T-SRAM unit, the data transmission bandwidth and power consumption problems of traditional processor memory are solved, and efficient multi-bit content addressing and simplified circuit structure are realized, expanding Application scenarios and reduce energy consumption.
Patent Information
- Application Number
- CN202411210019.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-08-30
AI Technical Summary
The data transmission between processors and memory in the traditional von Neumann architecture brings bandwidth limitations and power consumption problems, and the existing CIM circuits are complex in content addressing, with large area occupancy and high power consumption, which limits the Its application in computing-intensive algorithms.
Design a 10T-SRAM unit, which includes dual-channel data reading and content addressing functions, consists of 2 PMOS tubes and 8 NMOS tubes, and configures circuit to realize data content query, providing two data readings that do not interfere with each other. Channel, simplifies the circuit structure to support multi-bit content addressing operations.
It realizes efficient multi-bit data matching in one cycle, significantly reduces the dynamic power consumption of the circuit, expands the application scenario, and optimizes the area and computing efficiency of the circuit and chip.
Smart Images

Figure CN119091943A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of integrated circuits, and in particular relates to a 10T-SRAM unit, a dual-channel read and content-addressed logic circuit designed based on the 10T-SRAM unit, and a corresponding CIM chip. Background Art
[0002] The rapid development of technologies such as artificial intelligence, edge computing, and neural networks has put higher demands on the speed and energy efficiency of processors. Traditional processors are designed based on the von Neumann architecture. The computing unit and the storage unit are separated. The calculation is performed in the arithmetic unit, and the instructions and data are stored in the memory. When performing calculations, the computer needs to read data from the memory. After the calculation is completed, the computer needs to write the results back to the memory, which causes a large amount of data to be transmitted back and forth between the computer and the memory. This data operation method will bring the following problems: (1) Data transmission requires higher and higher bandwidth for communication between the processor and the memory, but there are obvious limitations on the bandwidth that can be achieved in terms of physical bandwidth. (2) With the continuous development of semiconductor technology, the speed of the arithmetic unit has far exceeded the read and write speed of the storage unit, resulting in the "storage wall" problem. (3) The transmission of data between the computing unit and the storage unit generates a lot of power consumption. With the continuous advancement of technology, the power consumption problem has become more prominent, resulting in the "power consumption wall" problem.
[0003] In order to break through the storage bottleneck and power consumption bottleneck of the traditional von Neumann architecture, technicians have developed a new computing architecture and solution. Compute In Memory (CIM) directly uses storage units for calculations, integrates data and calculations, eliminates memory access operations from the architecture, and is expected to break through the computing power and power consumption bottlenecks. It is very suitable for fields with intensive algorithm access and calculation rules such as artificial intelligence and graph computing.
[0004] At present, in-memory arithmetic circuits can be used to perform logical operations such as multiplication and MAC in data processing tasks such as artificial intelligence, edge computing, and neural networks, and can also be used for content addressing. Content addressing operation is an operation that can complete the matching of input targets and storage contents in one search cycle. It has great applications in efficient machine learning models such as feature retrieval and memory calculations based on distance metrics. The content addressing unit design in traditional CIM circuits occupies a large area, has a single function, and a small application range, which limits its storage density for mapping computationally intensive algorithms. Summary of the invention
[0005] In order to solve the common shortcomings of existing CIM circuits supporting content addressing functions, such as complex circuits, large area occupation and high power consumption, the present invention provides a 10T-SRAM unit, and a dual-channel read and content-addressed logic circuit designed based on the 10T-SRAM unit and a corresponding CIM chip.
[0006] The technical solution provided by the present invention is:
[0007] A 10T-SRAM cell has dual-channel data reading and content addressing functions, and is composed of two PMOS tubes P1-P2 and eight NMOS tubes N1-N8. Among them, P1, P2, N1-N4 constitute a 6T storage cell and support data storage functions including data reading, writing, and retention. The remaining N5-N8 constitute a configuration circuit; the configuration circuit is used to realize content query of the data stored in the 6T storage cell and provide another data reading channel.
[0008] Among them, the circuit connection relationship of the configuration circuit is as follows:
[0009] The gates of N5 and N6 are connected to the storage nodes Q and QB in the 6T storage cell respectively; the gates of N7 and N8 are connected to the control signals SL and SR respectively; the drain of N5 is connected to the source of N7; the source of N8 is connected to the drain of N6; the sources of N5 and N6 are connected to the transfer signal line TL, and the drains of N7 and N8 are connected to the mark signal line ML.
[0010] In the solution of the present invention, the circuit connection relationship of the 6T storage unit is as follows:
[0011] The sources of P1 and P2 are connected to VDD; the gates of P2 and N2 are connected to the drains of P1 and N1 and serve as the storage node Q; the gates of P1 and N1 are connected to the drains of P2 and N2 and serve as the storage node QB; the sources of N1 and N2 are grounded; N3 serves as a transmission tube between the storage node Q and the bit line BL; N4 serves as a transmission tube between the storage node QB and the bit line BLB; the gates of N3 and N4 are connected to the word line WL.
[0012] As a further improvement of the present invention, the operation strategy of configuring the circuit to implement content addressing is as follows:
[0013] (1) Pre-store the content to be searched in the 6T storage unit:
[0014] When Q is high and QB is low, it means the stored data is "1". When Q is low and QB is high, it means the stored data is "0".
[0015] (2) Set SL and SR to high level, and then precharge the transfer signal line TL and the mark signal line ML to high level.
[0016] (3) After TL and ML are pre-charged, the coded query number is input through control signals SL and SR:
[0017] When SL is high and SR is low, it means the query number is "1". When SL is low and SR is high, it means the query number is "0".
[0018] (4) The transfer signal line TL is grounded, and the result of content addressing is determined according to the level change of the mark signal line ML:
[0019] If ML drops to a low level, the query number is the same as the stored data. If ML remains high, the query number is different from the stored data.
[0020] As a further improvement of the present invention, the 6T memory cell includes a channel 1 for reading data using the bit lines BL and BL, and the configuration circuit includes a channel 2 for reading data using the transfer signal line TL and the mark signal line ML;
[0021] The 10T-SRAM cell provided by the present invention allows data stored in the 6T storage cell to be read through channel 1 when performing a content addressing operation.
[0022] As a further improvement of the present invention, there are two strategies for realizing data reading in channel 1, one of which is as follows:
[0023] Precharge the bit lines BL and BLB to a high level, and then turn on the word line WL; the stored data is reflected in the level state of the bit lines BL and BLB:
[0024] If BL remains at a high level and BLB drops to a low level, it means that the stored data is "1". If BL drops to a low level and BLB remains at a high level, it means that the stored data is "0".
[0025] As a further improvement of the present invention, the strategy for implementing data reading in channel 2 is as follows:
[0026] After SL and SR are set to high level, the transfer signal line TL and the mark signal line ML are precharged to high level.
[0027] After the precharge is completed, SL is set to a high level and SR is set to a low level.
[0028] When the transfer signal line TL is discharged to a low level, the stored data is reflected in the voltage state of the mark signal line ML:
[0029] If ML remains at a high level, it means that the stored data is "0"; if ML drops to a low level, it means that the stored data is "1".
[0030] In the present invention, another strategy for implementing data reading in channel 2 is as follows:
[0031] After SL and SR are set to high level, the transfer signal line TL and the mark signal line ML are precharged to high level.
[0032] After the precharge is completed, SL is set to a low level and SR is set to a high level.
[0033] When the transfer signal line TL is discharged to a low level, the stored data is reflected in the voltage state of the mark signal line ML:
[0034] If ML remains at a high level, it means that the stored data is "1"; if ML drops to a low level, it means that the stored data is "0".
[0035] The present invention also includes a dual-channel read and content-addressed logic circuit, which is formed by arranging a plurality of 10T-SRAM cell arrays as described above.
[0036] Among them, each 10T-SRAM cell in the same row shares the same word line WL. The ML end of each 10T-SRAM cell in each row is connected to the TL end of the adjacent 10T-SRAM cell on the right; and the transfer signal line TL of the leftmost 10T-SRAM cell is used as an independent port, and the mark signal line ML of the rightmost 10T-SRAM cell is used as an independent port.
[0037] The 10T-SRAM cells in the same column share the same set of bit lines BL, BLB and signal lines corresponding to control signals SL and SR.
[0038] The N 10T-SRAM units in the same row of the dual-channel read and content-addressed logic circuit provided by the present invention together constitute a basic unit for implementing N-bit data content addressing.
[0039] As a further improvement of the present invention, the operation strategy of the dual-channel read and content addressing logic circuit to implement multi-bit content addressing is as follows:
[0040] (1) The multi-bit number to be searched is pre-stored bit by bit in each 10T-SRAM cell in the same row through the bit lines BL and BLB.
[0041] (2) Set the SL and SR corresponding to each 10T-SRAM cell to a high level, and then precharge the transfer signal line TL and the mark signal line ML to a high level.
[0042] (3) After TL and ML precharge is completed, the query number is input bit by bit into each 10T-SRAM cell in the same row through control signals SL and SR.
[0043] (4) The transfer signal line TL is grounded, and the result of content addressing is determined according to the level change of the mark signal line ML:
[0044] If ML drops to a low level, it indicates that the query number is the same as the stored data; if ML remains at a high level, it indicates that the query number is different from the stored data.
[0045] The present invention also includes a CIM chip, which includes an SRAM array and a peripheral circuit that cooperates with the SRAM array to realize data storage function and content addressing function. The SRAM array adopts the aforementioned dual-channel reading and content addressing logic circuit.
[0046] The technical solution provided by the present invention has the following beneficial effects:
[0047] The present invention designs a novel 10T-SRAM unit, and based on it designs an in-memory operation circuit that can realize dual-channel data reading and content addressing operations. The CIM circuit provided by the present invention can not only realize basic read and write functions, but also has two channels for reading data that do not interfere with each other. The data reading path provided by the configuration circuit can effectively overcome the read corruption problem.
[0048] The circuit structure of the present invention is simple, but can realize complex multi-bit content addressing functions, and can achieve matching of multiple multi-bit data within one cycle, with high query efficiency and can significantly reduce the dynamic power consumption of the circuit.
[0049] The circuit provided by the present invention provides dual functions, which can not only realize the read and write operations of the storage unit, but also realize the addressing function of the storage unit content, and can use channel 1 to read the storage content while performing content addressing, so the application scenarios are more extensive.
[0050] The circuit design of the present invention simplifies the calculation process, can effectively improve the operation efficiency and speed, reduce the energy consumed in the transmission process, and optimize the area of the circuit and chip. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Figure 1 This is a circuit diagram of a 10T-SRAM unit provided in Embodiment 1 of the present invention.
[0052] Figure 2 Schematic diagram of the circuit that performs the task of matching the query number "1" with the stored data "1" for a 10T-SRAM cell.
[0053] Figure 3 Schematic diagram of the circuit that performs the task of matching the query number "1" with the stored data "0" for a 10T-SRAM cell.
[0054] Figure 4Schematic diagram of the circuit that performs the task of matching the query number "0" with the stored data "1" for a 10T-SRAM cell.
[0055] Figure 5 Schematic diagram of the circuit that performs the task of matching the query number "0" with the stored data "0" for a 10T-SRAM cell.
[0056] Figure 6 It is a circuit diagram of a dual-channel read and content-addressed logic circuit provided in Embodiment 2 of the present invention.
[0057] Figure 7 Schematic diagram of the dual-channel read and content-addressed logic circuit performing the content-addressing task of a 5-bit number.
[0058] Figure 8 This is a schematic diagram of the CIM circuit provided in Example 3 of the present invention. DETAILED DESCRIPTION
[0059] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0060] Example 1
[0061] This embodiment provides a 10T-SRAM cell, which has dual-channel data reading and content addressing functions, such as Figure 1 As shown, the 10T-SRAM unit is composed of 2 PMOS tubes P1-P2 and 8 NMOS tubes N1-N8. Among them, P1, P2, N1-N4 constitute a 6T storage unit and support the data storage function including data reading, writing and retention. The remaining N5-N8 constitute a configuration circuit; the configuration circuit is used to realize the content query of the data stored in the 6T storage unit and provide another data reading channel.
[0062] Among them, the circuit connection relationship of the 6T storage unit is as follows:
[0063] The sources of P1 and P2 are connected to VDD; the gates of P2 and N2 are connected to the drains of P1 and N1 and serve as the storage node Q; the gates of P1 and N1 are connected to the drains of P2 and N2 and serve as the storage node QB; the sources of N1 and N2 are grounded; N3 serves as a transmission tube between the storage node Q and the bit line BL; N4 serves as a transmission tube between the storage node QB and the bit line BLB; the gates of N3 and N4 are connected to the word line WL.
[0064] The circuit connection relationship of the configuration circuit is as follows:
[0065] The gates of N5 and N6 are connected to the storage nodes Q and QB in the 6T storage cell respectively; the gates of N7 and N8 are connected to the control signals SL and SR respectively; the drain of N5 is connected to the source of N7; the source of N8 is connected to the drain of N6; the sources of N5 and N6 are connected to the transfer signal line TL, and the drains of N7 and N8 are connected to the mark signal line ML.
[0066] In the solution of this embodiment, the strategy for implementing data writing in the 6T storage unit is:
[0067] In the initial state, WL is at a low level; at this time, the bit lines BL and BLB are set according to the data to be written: when the data to be written is "1", BL is set to a high level and BLB is set to a low level; when the data to be written is "0", BL is set to a low level and BLB is set to a high level.
[0068] Next, WL is set to a high level, and transmission tubes N3 and N4 are turned on; at this time, charge sharing is performed between the storage node and the corresponding bit line. When the data to be written is "1", the bit line BL pair charges the storage node Q, pulls Q to a high level, and the storage node QB is discharged through the bit line BLB, and QB is pulled down to a low level, thereby realizing the write operation of data "1". When the data to be written is "0", the bit line BLB pair charges the storage node QB, pulls QB to a high level, and the storage node Q is discharged through the bit line BL, and Q is pulled down to a low level, thereby realizing the write operation of data "0".
[0069] After completing the data writing, WL is set to a low level, and the transmission tubes N3 and N4 are turned off. At this time, the circuit is in a holding state, and the latch structure composed of P1, P2, N1, and N2 can stably hold the written data.
[0070] In the solution of this embodiment, the 6T memory cell includes a channel 1 for reading data using the bit lines BL and BL. The strategy for the 6T memory cell to read data using the channel 1 is:
[0071] Precharge the bit lines BL and BLB to a high level, and then turn on the word line WL; the stored data is reflected in the level state of the bit lines BL and BLB:
[0072] Specifically, when the data stored in the 6T storage cell is "1", Q is at a high level and QB is at a low level; at this time, the storage node Q has the same potential as the directly connected bit line BL, and BL still maintains a high level; and the potential of the storage node QB is lower than the bit line BLB, so the charge sharing between the two causes the bit line BLB to drop to a low level.
[0073] On the contrary, when the data stored in the 6T memory cell is "0", Q is at a low level and QB is at a high level; at this time, the storage node QB has the same potential as the directly connected bit line BLB, and BLB still maintains a high level; and the potential of the storage node Q is lower than that of the bit line BL, so the charge sharing between the two causes the bit line BL to drop to a low level.
[0074] In summary: in the data reading stage, if BL remains at a high level and BLB drops to a low level, it means that the stored data is "1". If BL drops to a low level and BLB remains at a high level, it means that the stored data is "0".
[0075] In the 10T-SRAM cell provided in this embodiment, content addressing can be implemented by configuring the circuit part, that is, querying whether the data stored in the storage cell is the same as the specified data. Specifically, the operation strategy of the 10T-SRAM cell of this embodiment to implement content addressing is as follows:
[0076] (1) Based on the data storage function of the 6T storage unit, the content to be searched is pre-stored in the 6T storage unit:
[0077] In each 6T memory cell, when Q is high and QB is low, it indicates that the stored data is "1". When Q is low and QB is high, it indicates that the stored data is "0".
[0078] (2) Set SL and SR to high level, and then precharge the transfer signal line TL and the mark signal line ML to high level.
[0079] (3) After TL and ML are pre-charged, the coded query number is input through control signals SL and SR:
[0080] When SL is high and SR is low, it means the input query number is "1". When SL is low and SR is high, it means the input query number is "0".
[0081] (4) The transfer signal line TL is grounded, and the result of content addressing is determined according to the level change of the mark signal line ML:
[0082] If ML drops to a low level, the query number is the same as the stored data. If ML remains high, the query number is different from the stored data.
[0083] The following describes in detail the process and principle of implementing content addressing in the 10T-SRAM unit provided in this embodiment for different query numbers and storage data situations:
[0084] Specifically, Figure 2As shown in the figure, when the query number is "1" and the stored data is "1", Q in the 6T storage unit is high and QB is low; SL in the configuration circuit is high and SR is low. At this time, N5 is turned on, N6 is turned off, N7 is turned on, and N8 is turned off. In this state, a discharge path is formed between TL and ML through N5 and N7, and ML eventually drops to a low level, indicating that the query number and the stored data match successfully.
[0085] like Figure 3 As shown in the figure, when the query number is "1" and the target number is "0", Q in the 6T storage unit is high and QB is low; SL in the configuration circuit is low and SR is high. At this time, N5 is turned on, N6 is turned off, N7 is turned off, and N8 is turned on. In this state, a discharge path cannot be formed between TL and ML, and ML remains high, which means that the query number and the stored data fail to match.
[0086] like Figure 4 As shown in the figure, when the query number is "0" and the target number is "1", Q in the 6T storage unit is low and QB is high; SL in the configuration circuit is high and SR is low. At this time, N5 is off, N6 is on, N7 is on, and N8 is off. In this state, a discharge path cannot be formed between TL and ML, and ML remains high, which means that the query number and the stored data fail to match.
[0087] like Figure 5 As shown in the figure, when the query number is "0" and the target number is "0", Q in the 6T storage unit is low and QB is high; SL in the configuration circuit is low and SR is high. At this time, N5 is off, N6 is on, N7 is off, and N8 is on. In this state, a discharge path is formed between TL and ML through N6 and N8, and ML eventually drops to a low level, indicating that the query number and the stored data match successfully.
[0088] The truth table corresponding to the above data matching process is shown in the following table:
[0089] Table 1: Truth table for content-addressed operations in 10T-SRAM cells
[0090]
[0091] In summary, in the configuration circuit of the 10T-SRAM unit designed in this embodiment, if the query number and the storage data are the same, ML and TL will be directly connected and form a discharge path. If the query number and the storage data are different, ML and TL will not be connected, and thus the discharge path cannot be formed. The existence or non-existence of the connection path will ultimately affect the final level state of ML. Therefore, according to the change of the ML level state, it can be determined whether the query number and the storage data are matched successfully, thereby completing the content addressing task.
[0092] It should be noted that: in the 10T-SRAM cell of this embodiment, the data reading and content addressing operations are performed by different objects, and the operation process of the configuration circuit will not affect the data stored in the 6T storage cell, so the circuit supports reading the data stored in the 6T storage cell through channel 1 while performing the content addressing operation. This helps to improve the circuit's operational efficiency in certain computing tasks.
[0093] In addition, technicians can also use the configuration circuit in the 10-SRAM of this embodiment to realize data reading. The principle of this circuit to perform such data reading operation is: fix the query number to "1" or "0" in the data reading stage, and then quantify the matching result between the storage data and the query number, and use the quantization result as the result of data reading. For example, when the query number is fixed to "1", if the storage data matches the query number successfully, it means that the storage data read is "1", and if the storage data fails to match the query number, it means that the storage data read is "0". On the contrary, when the query number is fixed to "0", if the storage data matches the query number successfully, it means that the storage data read is "0", and if the storage data fails to match the query number, it means that the storage data read is "1". That is: the configuration circuit in the 10-SRAM provides a channel two for realizing the read operation of the data stored in the 6T storage unit.
[0094] In detail, the strategy for implementing data reading in channel 2 in this embodiment is as follows:
[0095] After SL and SR are set to high level, the transfer signal line TL and the mark signal line ML are precharged to high level.
[0096] After the precharge is completed, SL is set to a high level and SR is set to a low level.
[0097] When the transfer signal line TL is discharged to a low level, the stored data is reflected in the voltage state of the mark signal line ML:
[0098] If ML remains at a high level, it means that the stored data is "0"; if ML drops to a low level, it means that the stored data is "1".
[0099] Alternatively, the data reading strategy for channel 2 is as follows:
[0100] After SL and SR are set to high level, the transfer signal line TL and the mark signal line ML are precharged to high level.
[0101] After the precharge is completed, SL is set to a low level and SR is set to a high level.
[0102] When the transfer signal line TL is discharged to a low level, the stored data is reflected in the voltage state of the mark signal line ML:
[0103] If ML remains at a high level, it means that the stored data is "1"; if ML drops to a low level, it means that the stored data is "0".
[0104] In the above two strategies of using channel 2 for data reading operation, the quantization logic of the former is the same as the quantization logic of channel 1, and thus can be implemented using the same ADC circuit. Therefore, this quantization logic is relatively better. Considering that the configuration circuit will not affect the data stored in the storage unit, the data reading operation of channel 2 provided in this embodiment does not actually conflict with the data reading operation of channel 1, and can be operated in parallel.
[0105] Example 2
[0106] On the basis of the solution of embodiment 1, this embodiment further provides a dual-channel read and content-addressed logic circuit, which is composed of a plurality of 10T-SRAM cell arrays as those of embodiment 1.
[0107] Among them, Figure 6 As shown, each 10T-SRAM cell in the same row shares the same word line WL. The ML end of each 10T-SRAM cell in each row is connected to the TL end of the adjacent 10T-SRAM cell on the right; and the transfer signal line TL of the leftmost 10T-SRAM cell is used as an independent port, and the mark signal line ML of the rightmost 10T-SRAM cell is used as an independent port.
[0108] The 10T-SRAM cells in the same column share the same set of bit lines BL, BLB and signal lines corresponding to control signals SL and SR.
[0109] In combination with the solution of Example 1, it can be known that each 10T-SRAM unit can realize numerical matching of a single-bit query number and a single-bit storage data. When the match is successful, TL and ML are connected; when the match fails, TL and ML cannot be connected. Figure 6 Under this circuit design, after the ML and TL of adjacent 10T-SRAM cells in the same row are directly connected, if the corresponding two single-bit numbers in each 10T-SRAM cell in the same row are successfully matched, the TL end of the first 10T-SRAM cell in the row can be connected to the ML end of the last 10T-SRAM cell. If the corresponding two single-bit numbers in any 10T-SRAM cell in the same row fail to match, the TL end of the first 10T-SRAM cell in the row cannot be connected to the ML end of the last 10T-SRAM cell.
[0110] Based on the above circuit characteristics, the N 10T-SRAM cells in the same row of the dual-channel read and content-addressed logic circuit provided in this embodiment together constitute a basic unit for implementing N-bit data content addressing.
[0111] When the circuit implements a multi-bit content addressing operation, it only needs to input the storage data bit by bit into the 6T storage unit part of each 10T-SRAM unit in the same row, and then input the query number bit by bit into the configuration circuit in each 10T-SRAM unit;
[0112] In detail, the dual-channel read and content addressing logic circuit provided in this embodiment implements the operation strategy of multi-bit content addressing as follows:
[0113] (1) The multi-bit number to be searched is pre-stored bit by bit in each 10T-SRAM cell in the same row through the bit lines BL and BLB.
[0114] For example, if the pre-stored data in the four 10T-SRAM cells in a row are A1, A2, A3, and A4, they can be used to represent a 4-bit number. If the stored data is input in order from high to low, the pre-stored 4-bit number is "A1A2A3A4".
[0115] (2) Set the SL and SR corresponding to each 10T-SRAM cell to a high level, and then precharge the transfer signal line TL and the mark signal line ML to a high level.
[0116] (3) After TL and ML precharge is completed, the query number is input bit by bit into each 10T-SRAM cell in the same row through control signals SL and SR.
[0117] Assuming that a 4-bit query number is "B1B2B3B4", it should be decomposed into B1, B2, B3, B4, and the four single-bit numbers should be input into each 10T-SRAM cell in the corresponding row in the order of bit weight. Specifically, B1 is input into the 10T-SRAM cell corresponding to A1, B2 is input into the 10T-SRAM cell corresponding to A2, and so on.
[0118] (4) The transfer signal line TL is grounded, and the result of content addressing is determined according to the level change of the mark signal line ML.
[0119] According to the aforementioned circuit principle, if the query number and the storage result are exactly the same, the TL and ML of each 10T-SRAM unit will be connected and form a discharge path, and the level of ML will drop, indicating that the two multi-bit numbers match successfully. When any bit in the query number and the target number is different, the 10T-SRAM unit at the corresponding bit will fail to match and disconnect, which will cause the TL end and ML end to be unable to connect, and the discharge path cannot be formed. The ML end remains at a high level, indicating that the two multi-bit numbers fail to match.
[0120] That is, if ML drops to a low level, it indicates that the query number is the same as the stored data; if ML remains at a high level, it indicates that the query number is different from the stored data.
[0121] Specifically, Figure 7 A typical content addressing process of a 5-bit number is shown. In the figure, the query number of content addressing is "11011" from MSB to LSB, the storage content in the first row of the array is "11010", and the storage content in the second row is "11011".
[0122] Before the content addressing operation, the transmission tubes between the cell columns are first turned on, that is, SL1~SL5 and SR1~SR5 in the configuration circuit are all set to VDD; then ML1 and ML2 in the configuration circuit are precharged to VDD, and TL1 and TL2 are also precharged to VDD.
[0123] When a content-addressed operation begins, the signal lines for each column in the configuration block are set as follows:
[0124] SL1=1, SR1=0, SL2=1, SR2=0, SL3=0, SR3=1, SL4=1, SR4=0, SL5=1, SR5=0;
[0125] Then discharge TL1 and TL2 in the configuration circuit to VSS. At this time, in the first row of cells, the storage content of the 1st to 4th column cells matches the target value, so the transfer signal TL5 of the 5th column cell is discharged to VSS. However, since the storage data of the fifth column cell does not match the query number, a discharge path cannot be formed. Therefore, the result of the flag signal ML1 remains at VDD, indicating that the data match fails. In the second row of cells, the storage data of the 1st to 5th column cells all match the query number, and the flag signal ML2 is discharged to VSS through the 1st to 5th column cells, indicating that the data match is successful.
[0126] Example 3
[0127] Based on the solution of embodiment 2, this embodiment further provides a CIM chip, such as Figure 8As shown, it includes an SRAM array, and peripheral circuits that cooperate with the SRAM array to realize data storage functions and content addressing functions. The SRAM array adopts a dual-channel read and content addressing logic circuit as in Example 2. Each SRAM cell in the CIM of this embodiment can realize complete data read, write, and hold operations, and the CIM circuit provides two independent data read channels for each SRAM cell. In addition, the CIM circuit can also realize single-bit or multi-bit content addressing; wherein each SRAM cell serves as an execution object of a single-bit content addressing operation, and multiple SRAM cells in the SRAM array together constitute an execution object of a multi-bit content addressing operation.
[0128] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0129] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.
Claims
1. A 10T-SRAM cell having dual-channel data reading and content addressing functions, characterized in that: It is composed of 2 PMOS tubes P1-P2 and 8 NMOS tubes N1-N8; among them, P1, P2, N1-N4 constitute a 6T storage unit and support the data storage function including data reading, writing and retention; the remaining N5-N8 constitute a configuration circuit, which is used to realize content query of the data stored in the 6T storage unit and provide another data reading channel; Wherein, the circuit connection relationship of the configuration circuit is as follows: The gates of N5 and N6 are connected to the storage nodes Q and QB in the 6T storage cell respectively; the gates of N7 and N8 are connected to the control signals SL and SR respectively; the drain of N5 is connected to the source of N7; the source of N8 is connected to the drain of N6; the sources of N5 and N6 are connected to the transfer signal line TL, and the drains of N7 and N8 are connected to the mark signal line ML.
2. The 10T-SRAM cell according to claim 1, wherein: The circuit connection relationship of the 6T storage unit is as follows: The sources of P1 and P2 are connected to VDD; the gates of P2 and N2 are connected to the drains of P1 and N1 and serve as the storage node Q; the gates of P1 and N1 are connected to the drains of P2 and N2 and serve as the storage node QB; the sources of N1 and N2 are grounded; N3 serves as a transmission tube between the storage node Q and the bit line BL; N4 serves as a transmission tube between the storage node QB and the bit line BLB; the gates of N3 and N4 are connected to the word line WL.
3. The 10T-SRAM cell according to claim 2, wherein: The operation strategy of configuring the circuit to implement content addressing is as follows: (1) Pre-store the content to be searched in the 6T storage unit; When Q is high and QB is low, it means the stored data is "1"; when Q is low and QB is high, it means the stored data is "0"; (2) Set SL and SR to high level, and then precharge the transfer signal line TL and the mark signal line ML to high level; (3) After TL and ML are pre-charged, the coded query number is input through control signals SL and SR: When SL is high and SR is low, it means the query number is "1"; when SL is low and SR is high, it means the query number is "0"; (4) The transfer signal line TL is grounded, and the result of content addressing is determined according to the level change of the mark signal line ML: If ML drops to a low level, it indicates that the query number is the same as the stored data; if ML remains at a high level, it indicates that the query number is different from the stored data.
4. The 10T-SRAM cell according to claim 3, wherein: The 6T memory cell includes a channel 1 for reading data using the bit lines BL and BL, and the configuration circuit includes a channel 2 for reading data using the transfer signal line TL and the mark signal line ML; The 10T-SRAM cell allows data stored in the 6T memory cell to be read through channel one when performing a content addressing operation.
5. The 10T-SRAM cell according to claim 4, wherein: The strategy for implementing data reading in channel 1 is as follows: Precharge the bit lines BL and BLB to a high level, and then turn on the word line WL; the stored data is reflected in the level state of the bit lines BL and BLB: If BL remains at a high level and BLB drops to a low level, it means that the stored data is "1"; if BL drops to a low level and BLB remains at a high level, it means that the stored data is "0".
6. The 10T-SRAM cell according to claim 4, wherein: The strategy for data reading in channel 2 is as follows: After SL and SR are set to high level, the transfer signal line TL and the mark signal line ML are precharged to high level; After the precharge is completed, SL is set to a high level and SR is set to a low level; When the transfer signal line TL is discharged to a low level, the stored data is reflected in the voltage state of the mark signal line ML: If ML remains at a high level, it means that the stored data is "0"; if ML drops to a low level, it means that the stored data is "1".
7. The 10T-SRAM cell according to claim 4, wherein: The strategy for data reading in channel 2 is as follows: After SL and SR are set to high level, the transfer signal line TL and the mark signal line ML are precharged to high level; After the precharge is completed, SL is set to a low level and SR is set to a high level; When the transfer signal line TL is discharged to a low level, the stored data is reflected in the voltage state of the mark signal line ML: If ML remains at a high level, it means that the stored data is "1"; if ML drops to a low level, it means that the stored data is "0".
8. A dual-channel read and content-addressed logic circuit, characterized in that: It is composed of a plurality of 10T-SRAM cell arrays as described in any one of claims 1 to 7; Among them, each 10T-SRAM cell in the same row shares the same word line WL; the ML end of each 10T-SRAM cell in each row is connected to the TL end of the adjacent 10T-SRAM cell on the right; and the transfer signal line TL of the leftmost 10T-SRAM cell is used as an independent port, and the mark signal line ML of the rightmost 10T-SRAM cell is used as an independent port; Each 10T-SRAM cell in the same column shares the same set of bit lines BL, BLB and control signals SL and SR; The N 10T-SRAM units in the same row of the dual-channel read and content-addressed logic circuit together constitute a basic unit for implementing N-bit data content addressing.
9. The dual-channel read and content-addressed logic circuit of claim 8, wherein: The dual-channel read and content addressing logic circuit implements the operation strategy of multi-bit content addressing as follows: (1) The multi-bit number to be searched is pre-stored bit by bit in each 10T-SRAM cell in the same row through the bit lines BL and BLB; (2) Set the SL and SR corresponding to each 10T-SRAM cell to a high level, and then precharge the transfer signal line TL and the mark signal line ML to a high level; (3) After TL and ML precharge is completed, the query number is input bit by bit into each 10T-SRAM cell in the same row through control signals SL and SR; (4) The transfer signal line TL is grounded, and the result of content addressing is determined according to the level change of the mark signal line ML: If ML drops to a low level, it indicates that the query number is the same as the stored data; if ML remains at a high level, it indicates that the query number is different from the stored data.
10. A CIM chip, characterized in that: It includes an SRAM array and a peripheral circuit that cooperates with the SRAM array to realize data storage function and content addressing function; the SRAM array adopts the dual-channel reading and content addressing logic circuit as described in claim 8 or 9.
Citation Information
Patent Citations
In-memory computing device of double-6T SRAM structure
CN111816231A
Memory circuit and in-memory computing circuit with BCAM addressing and logical operation functions
CN115810374A
10T-SRAM (10 T-Static Random Access Memory)-based multiplication and multiply-accumulate
CN117608519A
In-memory calculation and CAM circuit based on 10T SRAM unit
CN118016123A
Storage and calculation integrated unit structure
CN118412023A
