An approximate search TCAM matching system based on current mode measurement

Abstract

The present invention discloses an approximate search TCAM matching system based on current mode measurement, including a TCAM array and a sense amplifier, wherein the TCAM array is used as a front-stage circuit, and the sense amplifier is used as a back-stage circuit, which is used to receive the analog current signal output by each row of TCAM cells in the TCAM array and convert it into a digital pulse equal to the mismatch degree. The present invention makes full use of the matching discharge characteristics of TCAM to realize the design of a new TCAM approximate search array and measurement circuit, realizes the measurement of the number of search mismatch bits, and obtains the scalability of the storage array.

Description

Technical Field

[0001] The present invention relates to the field of storage and computing integration, and in particular to an approximate search TCAM matching system based on current mode measurement. The present invention considers using a current mode measurement method for a low-power and high-performance TCAM design with an approximate search function. Background Art

[0002] In the era of big data, data interaction between memory and processor has become a shortcoming of computer architecture performance. As one of the most promising solutions, storage and computing integration has made great progress in parallel search and other aspects. Ternary content addressable memory (TCAM) has attracted widespread attention for its high parallel addressing and partial keyword matching function. The unique function of TCAM makes it widely used in routers, data mining, artificial intelligence models and other fields.

[0003] In terms of devices, TCAM based on CMOSMOS tubes has problems such as large area and low energy efficiency. To this end, new non-volatile devices have been introduced into the storage field, including ferroelectric field effect transistors (FeFETs) and magnetic tunnel junctions. The basic TCAM storage unit structure based on FeFET has been successfully applied to scenarios such as few-sample learning hardware architecture.

[0004] In terms of function, the precise search mode of traditional TCAM requires that the search data and storage content are completely matched. When faced with massive data generated by users, it mostly outputs mismatch signals. The high-parallel hardware search module brings limited energy efficiency improvement to the system, but instead brings a lot of energy consumption and waste of search time. Approximate search outputs the storage content closest to the search data. By allowing output deviations within a reasonable fault tolerance range, it can greatly reduce search time and power consumption, and alleviate the contradiction between the hardware scale cost of CAM that can only accurately search limited data and massive data query search operations. How to obtain a quantitative comparison result of the mismatch between the search data and the storage content is a technical problem that has not yet been completely solved. Summary of the invention

[0005] The purpose of the present invention is to provide an approximate search TCAM matching system based on current mode measurement to obtain a lower energy-delay product in order to solve the problems of low working efficiency, large energy consumption and delay of the existing precise search TCAM.

[0006] The objective of the present invention is achieved through the following technical solutions:

[0007] The present invention provides an approximate search TCAM matching system based on current mode measurement, comprising a TCAM array and a sensing amplifier, wherein the TCAM array serves as a front-stage circuit and the sensing amplifier serves as a back-stage circuit for receiving analog current signals output by each row of TCAM units in the TCAM array and converting them into digital pulses equal to the mismatch degree.

[0008] Furthermore, the current measurement method of the TCAM unit includes: canceling the original VDD pre-charging mechanism for ML, connecting the original ML to VDD to maintain a high voltage, using the TCAM output to ground as a new ML and using the output current as a measurement indicator.

[0009] Furthermore, the sensing amplifier includes an input transmission gate, a capacitor, an output transmission gate and a Schmitt trigger that are electrically connected in sequence. When the sensing amplifier is working, the input current signal passes through the input transmission gate controlled by the enable control and the output signal in sequence to charge the capacitor, the capacitor output voltage is output through the Schmitt trigger, and the output signal is fed back to the discharge transmission gate.

[0010] Furthermore, there are two input transmission gates, namely, a first transmission gate TG1 and a second transmission gate TG2, there are two discharge transmission gates, namely, a third transmission gate TG3 and a fourth transmission gate TG4, and there are two iso-inverters, namely, a first iso-inverter and a second iso-inverter;

[0011] By the enable signal V EN With the output feedback signal V SPIKE The first transmission gate TG1 and the second transmission gate TG2 are respectively controlled and connected in series to serve as isolation between the front-stage TCAM array and the measurement capacitor;

[0012] The output current of the TCAM array passes through the first transmission gate TG1 and the second transmission gate TG2 and then to the capacitor C SA Charge, so that the capacitor C SA The output voltage increases linearly, and the capacitor C SA The other end is grounded;

[0013] By the enable signal V EN With the output feedback signal V SPIKE The third transmission gate TG3 and the fourth transmission gate TG4, which are respectively controlled in reverse phase, are connected in parallel to the capacitor C SA Output terminal and reset signal V RESET A fast discharge circuit is formed between them;

[0014] Capacitor C SA The output signal passes through the first phase amplifier for amplitude expansion, the periodically flipped Schmitt trigger, and the second phase amplifier in turn to obtain the output pulse signal V SPIKE ;

[0015] Capacitor C SAThe voltage increase causes the Schmitt trigger output signal to flip, causing the capacitor C SA The discharge circuit is turned on, and the capacitor C SA The voltage decreases, forming a charge-discharge cycle. The cycle period is inversely proportional to the input current. Within the same measurement time, the number of charge-discharge cycles, that is, the number of output pulse peaks, is proportional to the input current.

[0016] The operation method of the sense amplifier includes:

[0017] Before measurement, the enable signal is turned off, the sense amplifier is disconnected from the previous storage array, and the capacitor discharge loop is turned on and completely discharged;

[0018] When the measurement starts, the enable signal controls the on-off flipping of the transmission gate, and the system automatically works according to the input circuit and outputs a digital signal in the form of a spike pulse.

[0019] Furthermore, the TCAM unit is a FeFET-TCAM unit, which is composed of two FeFETs, whose drain terminals are commonly connected to VVD, and whose gates are respectively connected to the search line SL and The sources are connected in common to the matching line ML, and the matching lines of each row of TCAM units in the TCAM array are connected.

[0020] Furthermore, the source-drain voltage of FeFET is constant, and the output current is constant during search; each column of TCAM cells in the TCAM array shares the same vertical SL and Each row of ML is connected to a post-stage sense amplifier for further signal processing.

[0021] Furthermore, the operation method of the TCAM array includes:

[0022] Before the TCAM array starts working, high voltage is used to transmit binary coded data and its opposite signal through SL and Write to array;

[0023] No pre-charging is required before the search begins. The search signal and its opposite signal are directly transmitted through SL and Input, if there is no match, ML outputs a constant current, and the output currents of each TCAM unit in each row of the TCAM array are superimposed as the input of the subsequent sensing amplifier.

[0024] Furthermore, the TCAM unit is a MOSFET-TCAM unit, which is composed of two symmetrical storage bits. Each storage bit stores data through a ring inverter. The gate is connected to the word line WL to select the connection between the MOS tube control bit line BL and the ring inverter. The output of the ring inverter and the search line respectively control the discharge circuit of VDD to ML through the gates of two series-connected MOS tubes.

[0025] Furthermore, the voltage and circuit structure of the VDD to ML discharge loop remain stable during the search process, and the loop outputs a constant current; each row of TCAM cells in the TCAM array shares a pair of horizontal bit lines BL and Write the opposite signal, each column of TCAM cells shares a pair of vertical write lines WL and Control write unit, share a pair of longitudinal search lines SL and Implement search functionality.

[0026] Furthermore, the operation method of the TCAM array includes:

[0027] Before the TCAM array starts working, the write line WL and Control, the opposite signal from the pair of bit lines BL and Write to the ring inverter and store;

[0028] No pre-charging is required before the search begins. The search signal and its opposite signal are directly transmitted through SL and If the input does not match, ML outputs a constant current, and the output currents of each unit in each row are superimposed as the input of the subsequent sensing amplifier.

[0029] The beneficial effects of the present invention are as follows:

[0030] 1) The present invention makes full use of the matching discharge characteristics of TCAM to realize the design of a new TCAM approximate search array and measurement circuit, realizes the measurement of the number of search mismatch bits, and obtains the scalability of the storage array;

[0031] 2) The current mode sensing amplifier in the present invention can realize the conversion of analog current signal to digital spike pulse, so as to measure the current magnitude;

[0032] 3) For CMOS-TCAM design, the ML voltage of the existing TCAM decreases rapidly inversely with the degree of mismatch after pre-charging, and the difference between different degrees of mismatch cannot be quantitatively obtained. The output current of the CMOS-TCAM based on the current mode increases proportionally with the degree of mismatch, and the current size corresponding to the unit mismatch is constant, thereby achieving quantitative measurement without being limited by the array scale.

[0033] 4) For FeFET-TCAM design, based on the beneficial effects of CMOS-TCAM, it has non-volatile storage characteristics, lower energy consumption and smaller area, which can increase the storage array scale of the TCAM system. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Schematic diagrams of an existing TCAM unit (a) and an improved TCAM unit (b) proposed by the present invention;

[0035] Figure 2 The structure diagram of the FeFET-TCAM unit (a) and the FeFET-TCAM array (b) it consists of;

[0036] Figure 3 is a diagram of the current mode sense amplifier structure;

[0037] Figure 4 is a structural diagram of a CMOS-TCAM cell (a) and its constituent CMOS-TCAM array (b);

[0038] Figure 5 The simulation waveforms (a) and statistical results (b) of the FeFET-TCAM design are shown in Figure 1.

[0039] Figure 6 The simulation waveforms (a) and statistical results (b) of the CMOS-TCAM design are shown in Figure 1.

[0040] Figure 7 It is the extended sawtooth simulation waveform of the current mode sense amplifier design;

[0041] Figure 8 is the actual pulse spike output simulation waveform of the current mode sense amplifier design. DETAILED DESCRIPTION

[0042] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] Example 1

[0044] See also Figure 1-2 , an approximate search TCAM matching system based on current mode measurement, including a TCAM array and a sense amplifier, wherein the TCAM array is used as a front-stage circuit, and the sense amplifier is used as a back-stage circuit, for receiving the analog current signal output by each row of TCAM cells in the TCAM array and converting it into a digital pulse equal to the mismatch degree. The current measurement method of the TCAM cell includes: canceling the original VDD to ML pre-charging mechanism, connecting the original ML to VDD to maintain a high voltage, using the TCAM to ground output as a new ML and using the output current as a measurement indicator.

[0045] The TCAM unit of this embodiment is preferably a FeFET-TCAM unit. The overall structure and operation flow of the FeFET-TCAM unit and its array are as follows:

[0046] like Figure 2 As shown, each FeFET-TCAM unit consists of two symmetrically connected FeFETs, the drain of the FeFET is connected to VDD, and the gate is connected to the search line SL and The source is connected to the match line ML, and each row of TCAM cells in the TCAM array is connected to the match line. The FeFET source-drain voltage remains constant, and the output current is constant during search. In the TCAM array, multiple FeFET-TCAM cells are connected in parallel to ML to form a row, and each column of FeFET-TCAM cells shares the same vertical SL and ML is connected to the sense amplifier as output.

[0047] The operation process of the entire FeFET-TCAM array is as follows:

[0048] (1) Before the FeFET-TCAM array starts working, the binary coded data is written into the corresponding FeFET-TCAM. The operation scheme is shown in the following table. When writing, a reverse high voltage signal (-4V) is added to reset the storage unit (state 0), and a forward high voltage signal (4V) can write the storage content to state 1. When searching, a read voltage (1V) is used to read the current to avoid affecting the storage content.

[0049] Table 1 FeFET-TCAM operation scheme (voltage unit: V)

[0050]

[0051] (2) In the search phase, the search signal is directly sent from SL and Input. If it matches the storage content, both FeFETs are turned off and no current flows through ML; if there is a mismatch, only one of the two FeFETs is turned on and ML outputs a constant current. Each row of ML is connected in parallel with multiple storage units, and the total output is the sum of the output currents of each unit.

[0052] The overall structure and operation flow of the current mode sense amplifier are as follows:

[0053] like Figure 3 As shown, the sensing amplifier includes a first transmission gate TG1, a second transmission gate TG2, a capacitor C SA , the third transmission gate TG3, the fourth transmission gate TG4, the first isotropic converter 4, the Schmitt trigger 1 and the second isotropic converter 2. By the enable signal V EN With the output feedback signal V SPIKE The first transmission gate TG1 and the second transmission gate TG2 are connected in series to isolate the front-stage TCAM array from the measurement capacitor. The output current of the TCAM array passes through the first transmission gate TG1 and the second transmission gate TG2 and then to the capacitor C SA Charge, so that the capacitor C SA The output voltage increases linearly, and the capacitor C SA The other end is grounded. EN With the output feedback signal V SPIKEThe third transmission gate TG3 and the fourth transmission gate TG4, which are respectively controlled in reverse phase, are connected in parallel to the capacitor C SA Output terminal and reset signal V RESET A fast discharge circuit is formed between capacitor C SA The output signal passes through the first phase amplifier for amplitude expansion, the periodically flipped Schmitt trigger, and the second phase amplifier in turn to obtain the output pulse signal V SPIKE Capacitor C SA The voltage increase causes the Schmitt trigger output signal to flip, causing the capacitor C SA The discharge circuit is turned on, and the capacitor C SA The voltage decreases, forming a charge and discharge cycle. The cycle period is inversely proportional to the input current. Within the same measurement time, the number of charge and discharge cycles, that is, the number of output pulse peaks, is proportional to the input current.

[0054] The entire current mode sense amplifier operates as follows:

[0055] (1) Preparation stage, enable signal V EN The first transmission gate TG1 is turned off and disconnects the connection between SA and the TCAM array, while TG4 is turned on to ensure that the capacitor voltage is maintained at V RESET .

[0056] (2) Search phase, enable signal V EN Set to low level, TG1 is turned on and TG4 is turned off. At this time, the TCAM array is connected to the sense amplifier. As the current is input, the capacitor C SA The voltage rises, the Schmitt trigger 1 flips, and the output pulse signal V SPIKE The second transmission gate TG2 is turned off to block the current input, and the third transmission gate TG3 is turned on to make the capacitor C SA Quick discharge to V RESET , and then repeat the process again.

[0057] Example 2

[0058] The difference between this embodiment and embodiment 1 is that the TCAM unit of this embodiment is preferably a CMOS-TCAM unit. The overall structure and operation flow of the CMOS-TCAM unit and its constituent array are as follows:

[0059] like Figure 4 As shown in the figure, each CMOS-TCAM unit contains 2 symmetrical storage bits, a total of 16 MOS tubes. Taking the storage bit on the left as an example, for the input part, the drains of MOS tube M1 and MOS tube M3 are connected to BL and The gates are connected to WL, the sources are respectively connected to the two ends of the ring inverter, and the ring inverter is composed of two inverters; for the storage part, the two inverters are connected in a ring to realize the storage function; for the output part, the drain of MOS tube M5 is connected to VDD, the gate is connected to the output of the ring inverter, the source is connected to the drain of MOS tube M7, the gate of MOS tube M7 is connected to SL, and the source is connected to ML. MOS tube M5 and MOS tube M7 form a discharge branch.

[0060] The operation process of the entire CMOS-TCAM array is as follows:

[0061] (1) Before the CMOS-TCAM array starts working, the binary coded data is written into the corresponding TCAM cell. When writing, as shown in the following table, all search lines are kept at low voltage. In the first stage Set high voltage, write content from BL input, and Keep low voltage, in the second stage WL is set to high voltage, reverse write content from Input, BL and WL keep low voltage.

[0062] Table 2 CMOS-TCAM write scheme (voltage unit: V)

[0063]

[0064] (2) In the search phase, the signals are shown in the following table: WL, BL and All keep low voltage, search signal from SL and Input. If it matches the storage content, both discharge branches are turned off and no current flows through ML; if there is a mismatch, only one of the two discharge branches is turned on and ML outputs a constant current. Each row of ML is connected in parallel with multiple storage units, and the total output is the sum of the output currents of each unit.

[0065] Table 3 CMOS-TCAM search scheme

[0066]

[0067] The functions and effects of the above-mentioned embodiments 1 and 2 are further illustrated and demonstrated through the following simulation experiments:

[0068] 1. Simulation conditions

[0069] For the CMOS-TCAM, the 45nm model was used to simulate it on the Hspice software, and for the FeFET-TCAM, the simulation was performed on the Spectre software. Both built a 32-bit single-row TCAM storage array.

[0070] To verify the scalability, the applicant conducted multiple searches, with the search mismatch increasing bit by bit starting from bit 0. For example, for a 4-bit storage array with a storage content of 4'b0000, the input search data were 4'b 0000, 4'b 0001, 4'b 0011, 4'b 0111, 4'b 1111. The single search time and search interval were both 5ns.

[0071] For the current mode sense amplifier, the applicant also performed simulations on Hspice software and directly input the CMOS-TCAM output current signal into the sense amplifier circuit for experiments.

[0072] 2. Simulation results

[0073] 1) Verification of scalability

[0074] 1.1) Figure 5 (a) The simulation results of FeFET-TCAM are given. The output current can maintain good stability within one search time. Read the current under different mismatches and perform linear regression analysis. Figure 5 As shown in (b), the two have a good linear relationship, and the current corresponding to the unit bit mismatch is 766.8nA, which confirms the scalability of the array.

[0075] 1.2) Figure 6 (a) The simulation results of CMOS-TCAM are given, which once again well verify the effective improvement of scalability of the present invention. The output current corresponding to the unit bit mismatch is 43.5uA, which is 57 times smaller than the output current of CMOS-TCAM.

[0076] 2) Energy consumption and delay evaluation

[0077] The following table shows the energy consumption and latency measurement results of the two TCAM designs.

[0078] Table 4 TCAM (64*64) array performance comparison

[0079]

[0080] As can be seen from the table, the introduction of FeFET has significantly reduced power consumption, and the search energy of FeFET-TCAM is 57 times less than that of current-mode CMOS-TCAM. This fully demonstrates the advantages of FeFET in the storage field and also solves the high power consumption problem of the current-mode solution.

[0081] 3) Sense amplifier function verification

[0082] Figure 7 and Figure 8The simulation results of the current mode sense amplifier are given. It can be observed that the number of pulse spikes corresponds to the number of mismatch bits. We confirm that the proposed scheme achieves mismatch measurement of up to 4 levels of current. When the mismatch is low, we can see that an additional smaller sawtooth charge and discharge waveform appears at the end of the search time. Its purpose is to reserve a certain time margin to deal with the negative effects of nonlinear discharge time.

[0083] The above embodiments are used to illustrate the present invention rather than to limit the present invention. Any modification and change made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. An approximate search TCAM matching system based on current mode measurement, characterized in that: It includes a TCAM array and a sense amplifier, wherein the TCAM array is used as a front-stage circuit, and the sense amplifier is used as a back-stage circuit, and the sense amplifier is used to receive the analog current signal output by each row of TCAM cells in the TCAM array and convert it into a digital pulse equal to the mismatch degree; The sensing amplifier includes an electrically connected input transmission gate, a capacitor, an output transmission gate, a directional converter and a Schmitt trigger. The input current signal passes through the input transmission gate controlled by the enable control and the output signal in sequence to charge the capacitor. The capacitor output voltage is output through the Schmitt trigger, and the output signal is fed back to the discharge transmission gate.

2. The approximate search TCAM matching system based on current mode measurement according to claim 1, characterized in that: The current measurement method of the TCAM unit includes: canceling the original VDD pre-charging mechanism for ML, connecting the original ML to VDD to maintain a high voltage, using the TCAM output to ground as a new ML and using the output current as a measurement indicator.

3. The approximate search TCAM matching system based on current mode measurement according to claim 1, characterized in that: There are two input transmission gates, namely a first transmission gate TG1 and a second transmission gate TG2, there are two discharge transmission gates, namely a third transmission gate TG3 and a fourth transmission gate TG4, and there are two iso-inverters, namely a first iso-inverter and a second iso-inverter; By the enable signal V EN With the output feedback signal V SPIKE The first transmission gate TG1 and the second transmission gate TG2 are connected in series as the front-stage TCAM array and the measurement capacitor C SA Isolation between The output current of the TCAM array passes through the first transmission gate TG1 and the second transmission gate TG2 and then to the capacitor C SA Charge, so that the capacitor C SA The output voltage increases linearly, and the capacitor C SA The other end is grounded; By the enable signal V EN With the output feedback signal V SPIKE The third transmission gate TG3 and the fourth transmission gate TG4, which are respectively controlled in reverse phase, are connected in parallel to the capacitor C SA Output terminal and reset signal V RESET A fast discharge circuit is formed between them; Capacitor C SA The output signal passes through the first phase amplifier for amplitude expansion, the periodically flipped Schmitt trigger, and the second phase amplifier in turn to obtain the output pulse signal V SPIKE ; Capacitor C SA The voltage increase causes the Schmitt trigger output signal to flip, causing the capacitor C SA The discharge circuit is turned on, and the capacitor C SA The voltage decreases, forming a charge-discharge cycle. The cycle period is inversely proportional to the input current. Within the same measurement time, the number of charge-discharge cycles, that is, the number of output pulse peaks, is proportional to the input current. The operation method of the sense amplifier includes: Before measurement, the enable signal is turned off, the sense amplifier is disconnected from the previous storage array, and the capacitor discharge loop is turned on and completely discharged; When the measurement starts, the enable signal controls the on-off flipping of the transmission gate, and the system automatically works according to the input circuit and outputs a digital signal in the form of a spike pulse.

4. The approximate search TCAM matching system based on current mode measurement according to any one of claims 1 to 3, characterized in that: The TCAM unit is a FeFET-TCAM unit, which consists of two FeFETs, whose drain terminals are connected to VVD and whose gate terminals are connected to the search line SL and The sources are connected in common to the matching line ML, and the matching lines of each row of TCAM units in the TCAM array are connected.

5. The approximate search TCAM matching system based on current mode measurement according to claim 4, characterized in that: The source-drain voltage of FeFET is constant, and the output current is constant during search; each column of TCAM cells in the TCAM array shares the same vertical SL and , each row of ML is connected to a post-stage sense amplifier for further signal processing.

6. The approximate search TCAM matching system based on current mode measurement according to claim 5, characterized in that: The operation method of the TCAM array includes: Before the TCAM array starts working, high voltage is used to transmit binary coded data and its opposite signal through SL and Write to array; No pre-charging is required before the search begins. The search signal and its opposite signal are directly transmitted through SL and Input, if there is no match, ML outputs a constant current, and the output currents of each TCAM unit in each row of the TCAM array are superimposed as the input of the subsequent sensing amplifier.

7. The approximate search TCAM matching system based on current mode measurement according to any one of claims 1 to 3, characterized in that: The TCAM unit is a MOSFET-TCAM unit, which consists of two symmetrical storage bits. Each storage bit stores data through a ring inverter. The gate is connected to the word line WL to select the connection between the MOS tube control bit line BL and the ring inverter. The output of the ring inverter and the search line control the discharge circuit of VDD to ML through the gates of two series-connected MOS tubes.

8. The approximate search TCAM matching system based on current mode measurement according to claim 7, characterized in that: The voltage and circuit structure of the VDD to ML discharge loop remain stable during the search process, and the loop outputs a constant current; each row of TCAM cells in the TCAM array shares a pair of horizontal bit lines BL and Write the opposite signal, each column of TCAM cells shares a pair of vertical write lines WL and Control write unit, share a pair of longitudinal search lines SL and Implement search functionality.

9. The approximate search TCAM matching system based on current mode measurement according to claim 8, characterized in that: The operation method of the TCAM array includes: Before the TCAM array starts working, the write line WL and Control, the opposite signal from the pair of bit lines BL and Write to the ring inverter and store; No pre-charging is required before the search begins. The search signal and its opposite signal are directly input through SL and ML respectively. If they do not match, ML outputs a constant current. The output currents of each unit in each row are superimposed as the input of the subsequent sensing amplifier.

Images