Content addressable memory cell

[0001] 1. Field of the Invention

[0002] The present invention relates in general to a semiconductor memory and, in particular, to a non-volatile Content Addressable Memory (CAM).

[0003] 2. Description of the Related Art

[0004] Unlike conventional memories, CAMs include a plurality of CAM cells that are addressed in response to their content, rather than by a physical address. Words stored in a CAM are accessed, i.e., univocally identified, by applying corresponding data values to input terminals of the CAM. In response to applied data values, a match line of the CAM is driven so as to assert or de-assert an associated match signal, indicating whether or not a stored word matches the applied data values.

[0005] CAMs are useful in many applications, such as search engines, in which a list of data values in predetermined order is searched to identify a specific word in the CAM. By identifying which location of the CAM is coupled to a match line driven to assert the match signal, the specific word is identified.

[0006] Two types of CAM cells are typically used in CAMs: binary CAM cells and ternary CAM cells. A binary CAM cell can store a high logic value or a low logic value. A ternary CAM cell can store one of three values: a high logic value, a low logic value or a “don't care” value. When the logic values stored in the binary or ternary CAM cells of a CAM location match applied data values, then the match line coupled to that location is driven so as to assert the match signal. Otherwise, when the logic values stored in the binary or ternary CAM cells of a CAM location do not match the applied data values, then the coupled match line is driven so as to de-assert the match signal. In addition, a ternary CAM cell storing a “don't care” value provides a match (or, alternatively, a mismatch) condition irrespective of the data value applied thereto.

[0007] Both binary and ternary CAM cells can be either volatile or non-volatile. In particular, volatile CAM cells were initially implemented by exploiting the architecture of static RAM (SRAM) cells and adding transistors for realizing an output of the CAM cell connected to the match line coupled thereto. Recently, volatile CAM cells have been developed based on the dynamic RAM (DRAM) architecture, for reducing the semiconductor area occupied by the relatively high number of transistors employed in a CAM cell based on an SRAM architecture.

[0008] The volatile CAM cells, based on SRAM or DRAM cells, have a low access time (lower in SRAM than in DRAM-based CAM cells), but their content is lost during power-down. Consequently, the volatile CAM cells need be re-loaded at every power-on of the CAM, and a separate memory device of non-volatile type, such as a hard disk or an EPROM memory has to be used as a back-up storage unit for the volatile CAM.

[0009] The use of two distinct semiconductor memories (distinct chips) is disadvantageous, because very complex and expensive. A wide area on a printed circuit board is to be reserved, additional sockets and interconnections between the two memories are to be provided.

[0010] Non-volatile CAMs have been proposed. For example, U.S. Pat. No. 6,317,349 B1 discloses an architecture of ternary non-volatile CAM cell, including two floating gate transistors of the type normally used in flash memories. A floating gate transistor is non-volatily programmed by injecting charges into the floating gate, so as to modify its threshold voltage: for example, a high threshold voltage can be associated with a high logic value, while a low threshold voltage can be associated with a low logic value, or vice versa. This non-volatile CAM cell preserves its content also during power-down and moreover permits the use of a smaller number of transistors in respect to a CAM cell based on SRAM architecture.

[0011] In a CAM, the operation of searching a given data word (search operation) needs a preset phase, in which each match line is precharged to a prescribed preset voltage. When a searched word matches the word stored in a CAM location, the match line coupled thereto remains at the preset voltage, while a variation of the voltage at a match line is read as a mismatch. A drawback of the non-volatile CAM cell architecture described in U.S. Pat. No. 6,317,349 B1 is a consequence of the flow of leakage currents through the floating gate transistors with low threshold voltage during the search operation. In a mismatch condition during the search operation, if a number of CAM cells of the same CAM location conduct leakage currents, it may happen that the voltage at the match line does not vary sufficiently to be read as a mismatch. This problem is significant when the threshold voltage of the floating gate transistors is low (a lower threshold voltage implies a greater leakage current), but in any case the leakage currents imply an excessive power consumption. Furthermore, during the search operation it may happen that only one CAM cell of a CAM location shows a mismatch, while all the other CAM cells of the same location show a match. The discharge of the match line coupled to this location from the preset voltage can be very slow and in many applications auxiliary circuits might be necessary, to be associated with such a non-volatile CAM for speeding up the discharge of the match lines and, consequently, the search operation.

[0012] The Applicant also observes that a programming operation of floating gate transistors requires a relatively long time, and a preliminary block erasing operation is necessary. The erasing operation can be critical because of the problem of regulating the threshold voltage of the floating gate transistors for limiting the leakage threshold voltage.

[0013] One embodiment of the present invention overcomes the above-mentioned drawbacks.

[0014] One embodiment of the present invention provides a content addressable memory cell for a non-volatile content addressable memory, including a non-volatile storage element for storing a content digit, a selection input for selecting the memory cell, a search input for receiving a search digit, and a comparison circuit arrangement for comparing the search digit to the content digit and for driving a match output of the memory cell so as to signal a match between the content digit and the search digit. The non-volatile storage element includes at least one phase-change memory element, i.e., a memory element based on a phase-change material, for storing in a non-volatile way the respective content digit.

[0015] Moreover, one embodiment of the present invention provides a non-volatile content addressable memory, including an arrangement of a plurality of content addressable memory cells that use phase-change memory elements.

[0027] With reference in particular to FIG. 1, a functional block scheme of a CAM 100, in which the present invention is applicable, is shown. The CAM 100 includes a plurality of CAM cells 105, arranged in a two-dimensional disposition, or a matrix, of a number N of rows and a number M of columns. Each row of CAM cells 105 is controlled by a respective wordline WL_i (i=1, . . . , N) and each column is controlled by two respective bitlines, in particular a bitline BLL_j and a bitline BLR_j (j=1, . . . , M). Each CAM cell 105 is univocally identified by the respective wordline WL_i and the bitlines BLL_j and BLR_j; each CAM cell 105 is further connected to a match line ML_i, associated with the respective row.

[0028] A bit of information is written in each CAM cell 105 and the number M of columns of CAM cells 105 is equal to the length of a word stored in the CAM 100, plus possible auxiliary columns. Reading circuits (schematically shown as a block 125), including in particular sense amplifiers, are connected to the bitlines BLR_j, BLL_j and are capable of sensing the content of the CAM cells 105, while match detecting circuits (schematically shown as a block 130) are connected to the match lines ML_i and read the voltage at the match lines ML_i.

[0029] The CAM 100 also includes a number of auxiliary lines, useful for the operation thereof, such as supply voltage lines and control lines. In particular, each CAM cell 105 can be connected, for example, to a supply voltage line PL_j and/or a reference voltage line RL_i and/or a search control line SCL_i (all these auxiliary lines are represented in dash-and-dot in FIG. 1). The supply voltage lines PL_j supply a supply voltage, that, depending on the specific CAM cell embodiment and/or the operations to be carried out, may be the supply voltage V_dd of the CAM 100 (typically, from 1 to 3 V) or a higher voltage, provided by devices (not shown in the drawing) integrated on the same chip, or externally supplied to the CAM 100; the reference voltage lines RL_i supply a reference voltage V_ref (i.e., the ground) to the CAM cells 105, if required.

[0030] A row decoder 110 is connected to the wordlines WL_i for selecting the desired rows of non-volatile CAM cells 105 and for applying suitable voltages during operation. Search control circuits 115 are connected to the bitlines BLL_j, BLR_j for applying suitable voltages during operation in response to control signals provided by a column decoder 120. In particular, during reading and writing operations on the CAM cells 105 the row decoder 110 and the column decoder 120 are responsive to address codes corresponding to respective CAM cells 105. The row decoder 110 is further connected to and drives the reference voltage lines RL_i and the search control lines SCL_i, while the search control circuits 115 are also connected to the supply voltage lines PL_j.

[0031] Referring to FIG. 2A, a ternary non-volatile CAM cell 105 according to an embodiment of the present invention is shown. The ternary non-volatile CAM cell 105 is implemented by exploiting two phase-change memory (PCM) elements S1, S2.

[0032] The PCM elements S1, S2 are based on a phase-change material, typically consisting of a calcogenide (such as a Ge₂Sb₂Te₅ alloy) with resistivity changing at phase variations. The phase-change material can be reversibly switched between an amorphous, disordered phase and a crystalline, highly ordered phase. The two phases of the material exhibit different electrical characteristics; particularly, the material in the amorphous phase exhibits high resistivity and this phase can be associated with a first logic value, such as 0 (conventionally, reset state); the material in the crystalline phase exhibits low resistivity (about one hundred times lower than the resistivity of the material in the amorphous phase) and this phase can be associated with a second logic value, such as 1 (set state).

[0033] Without descending into particulars well known in the art, the material phase is stable below a given temperature T_s (such as 150° C.) and, consequently, the phase-change material can be used for implementing a non-volatile storage element, such as the PCM elements S1, S2. The material phase can be changed by heating the material above the temperature T_s. From the electrical standpoint, the phase-change material in the PCM elements S1 and S2 can be heated by causing a current to flow through a resistive element (or heater) embedded in the PCM elements S1, S2: the heat generated by Joule effect heats the phase-change material accordingly.

[0034] Considering FIG. 3, current-voltage (I-V) characteristic curves of a generic PCM element (S1 or S2) in the set and reset states are shown. A voltage between a negative terminal (“−” in the drawing) and a positive terminal (“+”) of the PCM element S1 (S2) is referred to as V_p and a current flowing from the positive terminal to the negative terminal is referred to as I_p.

[0035] If the current I_p reaches a given “set” value I_set (for example, 300 μA), the temperature of the phase-change material raises over a nucleation temperature (such as 200° C.) and, when cooled slowly (e.g., in around 100 ns), the phase-change material becomes crystalline. If the current I_p is raised up to a given “reset” value I_reset, greater than the set value I_set (roughly, twice the value I_set and, for example, equal to 500 μA), the temperature of the phase-change material raises over a melting temperature (such as 600° C.) and, when cooled rapidly (e.g., in around 10 ns or less), the phase-change material becomes amorphous. The current values I_set, I_reset, used for programming the PCM element S1 (S2), are indicated on the I-V characteristic curve of the crystalline phase.

[0036] When the voltage V_p is significantly lower than a switch value V_switch^p (typically, approximately from 1 to 1.5 V), both the set and the reset states of the PCM element S1 (S2) are stable and the resulting value of the current I_p is a signature of the resistivity of the phase-change material; the current I_p thus corresponds to the stored logic value. In FIG. 3 the resulting I-V characteristic curves of the PCM element S1 (S2) outline the difference between the high resistance (typically, from 50 to 150 kΩ) exhibited when the logic value 0 is stored and the low resistance (typically, from 2 to 4 kΩ) when the logic value 1 is stored.

[0037] In operation, if the phase-change material is in the amorphous phase (i.e., if the logic value 0 is stored), when the voltage V_p exceeds the switch value V_switch^p, the resistivity of the phase-change material becomes very similar to the resistivity in the crystalline phase and the resulting current I_p flowing through the PCM element S1 (S2) can modify the state of the phase-change material (that may become crystalline). Consequently, the stored logic value 0 might be turned into a 1; as a consequence, the voltage V_p should not exceed the switch value V_switch^p, in order to avoid the risk of spurious programming.

[0038] Referring back to FIG. 2A, the CAM cell 105 further includes four n-MOS transistors M1n, M2n, M3n and M4n. The PCM element S1 has the positive terminal connected to the bitline BLR_j and the negative terminal connected to the drain terminal of the transistor M1n. The PCM element S2 has the positive terminal connected to the bitline BLL_j and the negative terminal connected to the drain terminal of the transistor M2n.

[0039] The transistors M1n and M2n have the gate terminals both connected to the wordline WL_i and the source terminals connected together to a match node MG. The drain terminal of the transistor M3n and the gate terminal of the transistor M4n are also connected to the match node MG. The source terminal of the transistor M3n and a first source/drain terminal of the transistor M4n are connected together to the reference voltage line RL_i; a second source/drain terminal of the transistor M4n is connected to the match line ML_i and the gate terminal of the transistor M3n is connected to the search control line SCL_i. In this way, the transistors M1n and M21n act as selecting elements for the PCM elements S1, S2, respectively, while the transistors M3n and M4n act as switching elements, controlled by the voltages at the search control line SCL_i and the match node MG, respectively; in particular, the transistor M3n is used for selectively connecting the match node MG to the reference voltage line RL_i (thus for discharging the match node MG to the reference voltage).

[0040] The reading operation is performed on the selected CAM cell 105 by sensing the current flowing in each one of the two PCM elements S1, S2 by means of the reading circuits 125. For this purpose, a wordline read voltage V_read^WL (such as 3 V) is applied to the selected wordline WL_i for selecting the row of the CAM cell 105 in read mode, and a bitline read voltage V_read^BL (typically a voltage between 0.2 and 0.8 V), suitably lower than the switch voltage V_switch^p of the PCM elements S1 and S2,1 is applied to the selected bitlines BLL_j and BLR_j. At the same time, the reference voltage line RL_i is set to the ground voltage and the search control line SCL_i is set to a relatively high voltage (such as 3 V) so that the transistor M3n is turned on and strongly couples the match node MG to ground. The wordline WL_i and the bitlines BLL_j and BLR_j are for example selected supplying suitable address codes to the row decoder and the column decoder.

[0041] Differently from a reading operation, a writing operation is performed on the two PCM elements S1, S2 separately; a suitable current pulse is applied to one of the two PCM elements S1 and S2 at a time. When a low logic value has to be written into the selected PCM element S1, S2, the current pulse has to be of relatively short time duration and with an amplitude equal to the reset current I_reset. When a high logic value has to be written into the selected PCM element S1, S2, the current pulse has to be of long time duration and with an amplitude equal to the set current I_set. The selected wordline WL_i is set to a relatively high voltage (e.g., from 3 to 5 V), the reference voltage line RL_i is set to the ground voltage and the search control line SCL_i is set to a relatively high voltage (for example, between 3 and 5 V), for strongly coupling the match node MG to ground, as in the above described reading operation. One of the two bitlines BLL_j and BLR_j is raised to a suitable high voltage, for example from 3 to 5 V, while the other is set to a low voltage, typically between 0.4 and 0.7 V. Particularly, depending on the logic value to be programmed into the selected element S1, S2, for permitting the flow of the reset current I_reset or the set current I_set through the PCM element S1, S2, a suitable voltage (typically, in the range from 3.0 to 5.0 V) is applied to the respective bitline BLL_j, BLR_j for a predetermined time. Bitline driver circuits (not shown in the drawing) can act as current limiting devices for regulating the voltage at the bitlines BLL_j, BLR_j in such a way to permit the flow of a suitable current pulse, irrespective of the state of the PCM element S1, S2.

[0042] The CAM cell 105 can be programmed in four possible storage states, depending on the logic values written in the two PCM elements S1, S2, as defined in Table 1.

[0043]

TABLE 1
S1	S2	CAM cell
logic value	logic value	storage state
1	1	not used
1	0	1
0	1	0
0	0	Always Match (AM)

[0044] The two storage states in which one of the two PCM elements S1, S2 is set to the logic value 1 and the other to the logic value 0 are used to store one bit of a data word, hereinafter referred to as a content bit. In particular, when the PCM element S1 is set to 1 and the PCM element S2 is set to 0, the content bit is equal to 1; in the opposite case, the content bit is equal to 0 (the opposite convention can of course be adopted). The storage state in which both the two PCM elements S1 and S2 are set to the logic value 1 is referred to as Always Match (AM) state. The storage state in which both the two PCM elements S1 and S2 are set to the logic value 0 is not intended for use.

[0045] During a search operation of a given word, the search control circuits 115 simultaneously apply to the bitlines BLR_j and BLL_j voltages corresponding to the bits to be searched, hereinafter referred to as input or search bits. The row decoder 110 selects all the rows of the CAM 100, by applying a wordline search voltage V_search^WL (typically, from 3 to 5 V) to every wordline WL_i. In addition, the row decoder 110 precharges all the match lines ML_i to a relatively high preset voltage V_ML (typically, from 1.0 to 3.0 V) and sets all the reference voltage lines RL_i to the ground voltage; all the search control lines SCL_i are temporarily brought to a voltage sufficiently high for turning the transistor M3n on, thereby presetting the voltage of the match node MG to the ground voltage.

[0046] Supposing that the content bit in a given CAM cell 105 is 0, the respective PCM element S1 shows a high resistance, while the PCM element S2 shows a low resistance. If an input bit equal to 1 is being searched (thus, the input bit does not match the content bit in the CAM cell 105), the bitline BLL_j is set to a bitline search voltage V_search^BL (from 0.8 to 1.0 V) and the bitline BLR_j is kept at a low search voltage V_low (from 0.4 to 0.6 V). The voltage at the match node MG depends on a partition of the voltage (V_search^BL−V_low) across the PCM element S1 and the transistor M1n, and the PCM element S2 and the transistor M2n. The PCM element S1 shows a high resistance and, consequently, the impedance of a circuital branch between the bitline BLR_j and the match node MG, consisting of the PCM element S1 and the transistor M1n, is high and does not significantly influence the voltage at the match node MG. On the contrary, the PCM element S2 shows a low resistance and the node MG is pulled towards the bitline search voltage V_search^BL, because the transistors M2n is on. Therefore, the transistor M4n turns on and the match line ML_i is pulled towards the ground voltage, i.e., a voltage lower than the preset voltage V_ML. The match detecting circuits 130 sense the difference between the preset voltage V_ML and the voltage dynamically reached at the match line ML_i, thus detecting the bit-mismatch situation.

[0047] Only one bit-mismatch in a row is sufficient to cause the associated match line ML_i to be pulled towards the ground voltage, thus a bit-mismatch in a row causes a word-mismatch to be declared. If no CAM locations contain the searched data word, all the match lines ML_i are pulled to ground, and the match detecting circuits 130 thus declare a miss.

[0048] If an input bit equal to 0 is being searched (thus, the input bit matches the content bit in the CAM cell 105), the bitline BLR_j is set to the bitline search voltage V_search^BL and the bitline BLL_j is kept at the low search voltage V_low. The PCM element S2 shows a low resistance and, being the transistor M2n turned on, the match node MG is pulled towards V_low. Therefore, the transistor M4n is kept turned off, revealing a bit-match. The match line ML_i is kept at the preset voltage V_ML only if a bit-match occurs in all the CAM cells 105 of the associated row; the fact that the voltage at the match line ML_i does not change from the preset voltage V_ML is read by the match detecting circuits 130 as a word-match.

[0049] During the search operation, the response of a CAM cell 105 storing a content bit equal to 1 is similar to the above described one, thanks to the architectural symmetry of the CAM cells 105.

[0050] If the CAM cell 105 is in the AM storage state, then both the PCM elements S1 and S2 show a high resistance and the node MG is always kept at the low search voltage V_low; the transistor M4n is turned off irrespective of the voltage applied to the bitlines BLL_j, BLR_j. In this way, whichever the input bit fed to the CAM cell 105, a situation similar to a bit-match is produced. The AM state is useful, for example, for adapting a CAM with a certain number M of columns to the use in applications in which the word to be searched has a number of bits lower than M.

[0051] It is also possible to mask one or more of the input bits to be applied to the CAM. In fact, by setting both the bitlines BLL_i and BLR_i to the low search voltage V_low, a bit-match is produced whichever the content bit.

[0052] Alternatively, before starting a search operation, a reference voltage V_ref (for example, 1.8 V), higher than the ground voltage, can be applied to the reference voltage line RL_i, while the preset voltage V_ML at the match line ML_i can be equal to the ground voltage. In this alternative preset condition, when a bit-mismatch occurs, the match line ML_i voltage does not undergo the complete swing from the preset voltage V_ML to the reference voltage V_ref, as in the previous case, but the voltage swing is now equal to V_search^BL−V_ML−V_th4, where V_th4 is the threshold voltage of the transistor M4n. Consequently, the power consumption is reduced at each search operation.

[0053] It is observed that the bitline search voltage V_search^BL to be used depends on the threshold voltage of the transistor M4n; a higher threshold voltage imposes the use of higher search voltages at the bitlines BLL_j and BLR_j for turning the transistor M4n on in case of a bit-mismatch.

[0054] Furthermore, a combination of the above described operation conditions is possible, i.e., the node MG can be preset to a higher voltage than the ground voltage, for reducing the voltage swing at the node MG when a higher bitline search voltage is needed for turning the transistor M4n on.

[0055] According to still another alternative, both the reference voltage line RL_i and the match line ML_i can be preset to a relatively low voltage, e.g., 1.8 V. In this case, the bitline search voltage V_search^BL shall be high enough to make the transistor M4n conductive even in presence of a non-negligible body effect. The search operation can include an active probing, in which the match line ML_i is initially pulled down to the ground voltage for a relatively short time. If the voltage dynamically reached at the match line ML_i is the ground voltage, the transistors M4n in all the CAM cells 105 of the row are off and a word-match is thus detected (the match line ML_i sees a high impedance); differently if the voltage dynamically reached at the match line ML_i raises towards the voltage at the reference voltage line RL_i, the transistor M4n in at least one CAM cell 105 is on, therefore a word-mismatch is determined to have occurred (a low impedance exists between the reference voltage line RL_i and the match line ML_i).

[0056] This arrangement of CAM cells 105 is defined as of NOR-type, because the transistor M4n in the generic cell, when turned on, acts as a parallel connection element between the match line ML_i and the reference voltage line RL_i. It is observed that an AND-type arrangement can also be implemented, as schematically shown in FIG. 2B, e.g., by connecting a further transistor M0n of the generic CAM cell, driven by the match node MG, in series to the match line ML_i^IN−ML_i^OUT, while the transistor M4n has the gate terminal connected to a match control line MCL_i. In this way the transistors M0n of the CAM cells of a row are connected in series along the respective match line.

[0057] Compared to a volatile CAM cell based on the SRAM architecture, the proposed non-volatile CAM cell 105 allows reducing the number of transistors, and thus the area of semiconductor occupied. At the same time, the desired non-volatility is achieved by means of the PCM elements, which can be easily integrated slightly modifying pre-existing fabrication processes of integrated circuits, permitting a simple architecture of the non-volatile CAM.

[0058] Furthermore, the exploitation of storage elements made by a phase-change material permits the implementation of a non-volatile CAM intrinsically hard to radiations, easily and quickly re-programmable (a preliminary erasing process is not required) and even operatively faster than a non-volatile CAM exploiting floating gate transistors.

[0059] In the above-described non-volatile CAM cell, a current ranging from 2 μA to 8 μA is required by each cell during a search operation. This means that the size of the CAM might be limited by the maximum available power.

[0060] FIG. 4A shows a non-volatile CAM cell 105, according to a further embodiment of the present invention (the elements corresponding to those depicted in FIG. 2A are denoted with the same reference numerals and their description is omitted for the sake of simplicity). The two PCM elements S1, S2, acting as non-volatile storage elements, have the negative terminals connected to the bitlines BLR_j and BLL_j, respectively.

[0061] The n-MOS transistors M1n and M2n, with the gate terminal connected to the wordline WL_i, have the source terminals connected to the negative terminals of the PCM element S1 and S2, respectively, and the drain terminals connected together to the match node MG.

[0062] Two p-MOS transistors M3p and M4p replace the n-MOS transistors M3n and M4n of the embodiment of FIG. 2A. The p-MOS transistor M3p has the drain terminal connected to the node MG (consequently to the drain terminals of the n-MOS transistors M1n, M2n) and the source terminal connected to the supply voltage line PL_j, which provides the supply voltage V_dd. The gate terminal of the p-MOS transistor M3p is connected to the search control line SCL_i, which is used for driving the transistor M3p (as in the case of the transistor M3n shown in FIG. 2A). The transistor M3p acts as a switch for providing the supply voltage V_dd to the node MG. The p-MOS transistor M4p has the source terminal connected to the match line ML_i and the drain terminal connected to the reference voltage line RL_i, which provides the ground voltage. The gate terminal of the p-MOS transistor M4p is connected to the node MG and the voltage at the node MG drives the transistor M4p.

[0063] The reading operation is performed by selecting a row of CAM cells 105, i.e., by setting the wordline WL_i and the bitlines BLL_j and BLR_j to voltages between the ground voltage and the supply voltage V_dd (e.g., the wordline WL_j is set to 2.0 V). Then, the search control line SCL_i is set to the ground voltage, the p-MOS transistor M3p turns on and the node MG raises to the supply voltage V_dd, turning the p-MOS transistor M4p off. For reading the logic value stored in either one of the PCM elements S1, S2, the corresponding bitline BLL_j or BLR_j is pulled to a lower voltage (e.g., 1.0 V) than the voltage at the match node MG.

[0064] During the writing operation, the CAM cell 105 is selected (and the n-MOS transistors M1n, M2n are turned on) by applying the supply voltage V_dd to the corresponding wordline WL_i; the corresponding bitlines BLL_j and BLR_j are initially set to the supply voltage V_dd. The search control line SCL_i is set to the ground voltage, turning the p-MOS transistor M3p on, and, in order to deliver a suitable current pulse to each one of the PCM elements S1, S2, the bitlines BLL_j and BLR_j are alternatively pulled towards ground. The bitline driver circuits can regulate the voltage at the bitlines BLL_j, BLR_j, as described with reference to the embodiment of FIG. 2A. Also in this embodiment, the CAM cell 105 can be set in either one of the four storage states defined in Table 1 (the storage state in which both the PCM elements S1 and S2 show low resistance is still not used).

[0065] Before starting a search operation, the wordline WL_i reaches a voltage such that each one of the n-MOS transistors M1n and M2n conducts a prescribed current I_SN (for example, 800 nA) in the case in which the respective source terminal is at ground. The bitlines BLL_j and BLR_j are preset to a positive low voltage, for example between 0.5 and 1.0 V, suitable to keep the transistors M1n, M2n turned off. A voltage at the search control line SCL_i is such that the p-MOS transistor M3p conducts a current I_SP, lower than I_SN (for example of 400 nA); the node MG is thus preset to the supply voltage V_dd, the n-MOS transistors M1n, M2n being initially turned off. The match line ML_i is preset to the preset voltage V_ML (for example, between 1 and 3 V).

[0066] For starting the search operation, the voltage at the bitlines BLL_j, BLR_j is varied according to the input bit to be searched. Let it be supposed that the input bit applied to the CAM cell 105 is 1 and that the content bit of the CAM cell 105 is 0 (the PCM element S1 exhibits high resistance and the PCM element S2 exhibits low resistance). The input bit 1 is applied to the CAM cell 105 by pulling the voltage at the bitline BLL_j down to the ground voltage, while the bitline BLR_j is left at the preset voltage. In this condition, the transistor M1n is turned off, while the transistor M2n conducts a current almost equal to I_SN, because the source terminal thereof is approximately at the ground voltage. Consequently, a current equal to the difference I_SN−I_SP (800 nA−400 nA=400 nA in the example) pulls the match node MG down towards the ground voltage, the transistor M4p is turned on, and thus the match line ML_i is also pulled down towards ground (i.e., the ground voltage plus the threshold voltage of the transistor M4p), indicating a bit-mismatch (and consequently, a word-mismatch).

[0067] When the input bit 0 is applied to the CAM cell 105, the voltage at the bitline BLR_j is pulled down to the ground voltage, while the bitline BLL_j is left at the preset voltage. In this case the transistor M2n is turned off, while the transistor M1n is turned on. However, the high resistance exhibited by the PCM element S1 limits the current that can flow through the transistor M1n to a value smaller than I_SN (the source terminal of the transistor M1n is not in this case pulled down to ground). Consequently, the match node MG is not pulled down to ground, remaining substantially at the supply voltage V_dd; the transistor M4p is thus kept off and the voltage at the match line ML_i does not move from the preset voltage V_ML, indicating a bit-match.

[0068] Compared to the CAM cell of FIG. 2A, this second embodiment of CAM cell allows limiting the current required at every search operation.

[0069] It is observed that the series connection of the p-MOS transistor M3p and the n-MOS transistor M1n (M2n) may make the CAM cell difficult to be written in the case the PCM elements S1, S2 need a large current to be set/reset; in fact, the current that is allowed to flow through the PCM element S1 (S2) is limited by the series connection of the transistors M3p and M1n (M2n) for writing the CAM cell 105.

[0070] A further embodiment of ternary non-volatile CAM cell 105 according to the present invention, adapted to overcoming this possible problem, is shown in FIG. 4B (the elements corresponding to those in FIG. 2A and FIG. 4A are denoted with the same reference numerals and their description is omitted for the sake of simplicity). With respect to the embodiment illustrated in FIG. 4A, two diodes D1, D2 are added, each one having the cathode connected to the source terminal of a respective one of the transistors M1n, M2n, and the anode connected to an additional writing control line WCL_i associated with all the CAM cells of a same row.

[0071] The writing control line WCL_i is normally kept at the ground voltage, except during a writing operation, when the writing control line WCL_i is brought to the supply voltage V_dd. In order to write the CAM cell 105, one among the bitlines BLL_j and BLR_j, depending on the content bit to be stored, is pulled down to the ground voltage, permitting the flow of a current through the corresponding diode D1 or D2. The currents flowing through the diode D1, D2 is injected into the respective PCM element S1, S2, thus contributing to the programming current thereof.

[0072] Alternatively, the diodes D1, D2 can be replaced by selectively activatable conductive elements, such as p-MOS transistors, for contributing to the current flowing through the PCM elements S1 and S2 during a writing operation.

[0073] In a further alternative embodiment of the present invention, the p-MOS transistor M4p can be replaced by a n-MOS transistor, for a better drive of the match line ML_i towards a low-level voltage. Before starting the search operation, the wordline WL_i is set to the voltage such that each one of the n-MOS transistors M1n, M2n conducts the prescribed current I_SN, as in the above described embodiment. However, in this case the search control line SCL_i has to be preset to the supply voltage V_dd, for turning the transistor M3p off, and the input bit is imposed to the bitlines BLL_j, BLR_j, i.e., at least one of the bitlines BLL_j, BLR_j is at the ground voltage, for presetting the node MG to the ground voltage by the flow of the small current I_SN. The search operation starts when the voltage at the search control line SCL_i is pulled down to the bias voltage needed to cause the current I_SP flowing through the transistor M3p.

[0074] Referring to FIG. 4C, an alternative embodiment is shown, in which the ternary non-volatile CAM cell 105 has a complementary architecture compared to that depicted in FIG. 4B. In detail, the two n-MOS transistors M1n, M2n are replaced by two p-MOS transistors M1p, M2p, and the two p-MOS transistors M3p, M4p are replaced by two n-MOS transistors M3n, M4n, respectively.

[0075] In this embodiment, the PCM elements S1 and S2 have the positive terminals connected to the bitlines BLL_j and BLR_j, and the negative terminals connected to the source terminals of the p-MOS transistors M1p and M2p, respectively. The n-MOS transistor M3n has the source terminal connected to the reference voltage line RL_i; the n-MOS transistor M4n has the source terminal connected to the match line ML_i and the drain terminal connected to the supply voltage line PL_j. The two diodes D1 and D2, if provided, have the anodes connected to the source terminals of the transistors M1p and M2p, respectively, and the cathodes connected to the writing control line WCL_i. It is observed that the diodes D1, D2 might be represented by the pn junction formed by the p-type source region of the transistors M1p and M2p and the respective n-well, wherein the transistors are formed; to this purpose, the n-well should be directly connected to the writing control line WCL_i for being biased to the desired voltage (the connection is represented in dash-and-dot line in the drawing).

[0076] The operation voltages have to be changed accordingly and, in particular, during the writing operation the writing control line WCL_i, normally kept at the supply voltage V_dd (for example, from 1 to 3 V), is brought to the ground voltage. In order to write the CAM cell 105, the wordline WL_i is pulled down to the ground voltage and one of the bitlines BLL_j and BLR_j at the time is pulled up to a voltage suitable for setting or resetting the corresponding PCM element S1, S2, depending on the content bit to be stored, while the other bitline is kept at the ground voltage.

[0077] For presetting the match node MG to the ground voltage, the search control line SCL_i is biased to a voltage such that the n-MOS transistor M3n can conduct the current I_SP, when the source terminal thereof is at the ground voltage. The wordline WL_i is biased to a voltage such that each of the p-MOS transistors M1p and M2p can conduct the current I_SN, when the source terminal thereof is at the supply voltage V_dd. Being the bitlines BLL_j, BLR_j preset to a low voltage (or the ground voltage), the p-MOS transistors M1p, M2p do not conduct current, and the match node MG reaches the ground voltage. The match line ML_i is preset to the ground voltage.

[0078] For starting the search operation, at least one of the bitlines BLL_j, BLR_j is pulled up to the supply voltage V_dd, depending on the input bit to be searched. In case of a bit-mismatch, the match node MG moves towards the supply voltage V_dd, thus turning the transistor M4n on; the match line ML_i is thus pulled up towards the supply voltage V_dd, provided by the supply voltage line PL_j, minus the threshold voltage of the transistor M4n. When the input bit matches the content bit, as well as when the CAM cell 105 is programmed in the AM storage state, the match node MG is kept at the ground voltage and, consequently, the transistor M4n of the CAM cell 105 is kept off.

[0079] In general, the search operation can also be dynamic, i.e., the voltage at the bitlines BLL_j and BLR_j can be pulsed accordingly to the input bit, with pulses having duration of the order of nanoseconds. The bitlines BLL_j and BLR_j are preset to a relatively low voltage and the voltage pulses are from this preset voltage to the ground voltage in the case of the embodiments of FIGS. 4A, 4B, and to the supply voltage V_dd in the case of the embodiment of FIG. 4C. The transistor M3n or M3p, connected between the match node MG and the reference voltage line RL_i or the supply voltage line PL_j, can be used as a simple switch for presetting the node MG. During the dynamic search operation, the transistor M3n or M3p is turned off and the voltage at the match node MG can move, turning the transistor M4n or M4p on, only if a bit-mismatch occurs, i.e., only if the voltage pulse is applied to the bitline BLL_j, BLR_j connected to the PCM element S1, S2 exhibiting a low resistance.

[0080] Referring to FIG. 5A, a ternary non-volatile CAM cell according to a further embodiment of the present invention is illustrated (the elements corresponding to those in FIGS. 2A–4C are denoted with the same reference numerals and their description is omitted for the sake of simplicity). The CAM cell 105 is coupled to two supply voltage lines PL1_j, PL2_j of the CAM 100; in detail, the two PCM elements S1, S2 have the positive terminals connected to the supply voltage lines PL1_j, PL2_j, respectively.

[0081] This further embodiment of the CAM cell 105 exploits five transistors M1p, M2p, M4p, M5n and M6n rather than four as in the above described embodiments. The negative terminals of the PCM elements are connected to the source terminals of the p-MOS transistors M1p and M2p, respectively. The p-MOS transistors M1p and M2p are controlled by the voltage at the wordline WL_i (as in the above described embodiments) and have the drain terminals connected to the bitlines BLL_j and BLR_j, respectively; in this way, the p-MOS transistors M1p and M2p act again as selecting elements for the PCM elements S1 and S2. Furthermore, the n-well of the p-MOS transistors M1p and M2p (i.e., the n-type semiconductor region(s) wherein the p-MOS transistors M1p and M2p are formed) is biased to the voltage at the writing control line WCL_i. The p-MOS transistor M4p has the source terminal connected to the match line ML_i and the drain terminal connected to the reference voltage line RL_i, which is set to the ground voltage.

[0082] The n-MOS transistor M5n has a first drain/source terminal connected to the bitline BLL_j (i.e., connected to the drain terminal of the p-MOS transistor M1p), a second drain/source terminal connected to the gate terminal of the p-MOS transistor M4p and the gate terminal connected to the negative terminal of the PCM element S2. Symmetrically, the n-MOS transistor M6n has a first drain/source terminal connected to the bitline BLR_j (i.e., connected to the drain terminal of the p-MOS transistor M2p), a second drain/source terminal connected to the gate terminal of the p-MOS transistor M4p and the gate terminal connected to the negative terminal of the PCM element S1. Consequently, the second drain/source terminals of the n-MOS transistors M5n, M6n and the gate terminal of the p-MOS transistor M4p are connected together to the match node MG. The n-MOS transistors M5n, M6n act as switch elements controlled by the voltage reached at the negative terminals of the PCM elements S2, S1.

[0083] The four possible storage states for this CAM cell 105 are defined in Table 2.

[0084]

TABLE 2
S1	S2	CAM cell
logic value	logic value	storage state
0	0	not allowed
0	1	1
1	0	0
1	1	Always Match (AM)

[0085] When the PCM element S1 is set to the logic value 1 and the PCM element S2 is set to the logic value 0, the content bit is equal to 0; in the opposite case, the content bit is equal to 1 (again, the opposite convention can be adopted). The storage state in which both the two PCM elements S1 and S2 are set to the logic value 1 is referred to as an Always Match (AM) state, while the storage state in which both the two PCM elements S1 and S2 are set to the logic value 0 is not used.

[0086] During the writing operation, the wordline WL_i and the two bitlines BLL_j and BLR_j are set to the ground voltage. The write control line WCL_i is also set to the ground voltage, in order to correctly bias the diodes formed by the p-type source/drain regions of the transistors M1p and M2p and the respective n-type well wherein they are formed, thereby increasing the current flowing through the PCM elements S1 and S2. The supply voltage lines PL1_j and PL2_j are set to voltages adapted to permitting a suitable current pulse to be applied to the PCM elements S1, S2. The two PCM elements S1 and S2 are typically programmed sequentially and one of the two supply voltage line PL1_j, PL2_j is set to a sufficiently high voltage (e.g., from 3 to 5 V), while the other one is set to the ground voltage. Particularly, when the logic value 0 has to be written in the PCM element S1, S2, the voltage applied to the respective supply voltage line PL1_j, PL2_j is regulated by driver circuits (not shown in the drawing) for permitting the flow of the reset current I_reset through the diode of the transistor M1p, M2p, and thus the PCM element S1, S2. When the logic value 1 has to be written in the PCM element S1, S2, the voltage applied to the respective supply voltage line PL1_j, PL2_j is regulated by the driver circuits for permitting the flow of the set current I_set. The two PCM elements S1 and S2 could also be programmed in parallel, provided that the write control line WCL_i, (typically, a metal line), is sufficiently conductive.

[0087] Both the reading operation and the search operation are carried out by setting the reference voltage line RL_i to the ground voltage, and the supply voltage lines PL1_j and PL2_j and the writing control line WCL_i to the supply voltage V_dd. In particular, in order to sense the current flowing through the CAM cell 105, the wordline WL_i is set to the wordline read voltage V_read^WL (e.g., from 0.8 to 1.0 V) and both the bitlines BLL_j and BLR_j are set to a bitline read voltage V_read^BL (e.g., from 1 to 2 V).

[0088] Before starting a search operation, the bitlines BLL_j and BLR_j are both set to the supply voltage V_dd (e.g., from 1 to 3 V), and a suitable positive voltage V_search^WL is applied to the wordline, so as to turn the transistors M1p and M2p on. In this way, the source terminal of the p-MOS transistors M1p, M2p reaches the supply voltage V_dd, being the drain terminal thereof at the supply voltage V_dd, and the n-MOS transistors M5n and M6n are turned on, presetting the match node MG at the supply voltage V_dd, provided by the bitlines BLL_j, BLR_j, minus the threshold voltage of the transistors M5n, M6n (i.e., the p-MOS transistor M4p is kept turned off at this stage).

[0089] In order to carry out a search operation, the wordline WL_i is kept to the search voltage V_search^WL. The search voltage V_search^WL must be low enough to turn on the p-MOS transistors M1p and M2p, at the same time avoiding the risk of having the switch voltage V_switch^P across the PCM elements S1 and S2. Thus, the search voltage V_search^WL must satisfy the following conditions:
V_search^WL<V_dd−|V_th1,2|
and
V_search^WL>V_dd−V_switch^p−|V_th1,2|+ΔV,
where V_th1,2 is the threshold voltage of the p-MOS transistors M1p, M2p and ΔV is a security margin.

[0090] Let it now be supposed now that the content bit is 0 and that the input (search) bit is 1. For accomplishing the search operation, the bitline BLR_j is set to a relatively low positive voltage VL_L, while the bitline BLL_j is set to the supply voltage V_dd. Consequently, the voltage at the gate terminal of the n-MOS transistor M5n moves towards the low positive voltage V_L, set onto the bitline BLR_j, since the PCM element S2 exhibits a high resistance and the transistor M2p is on. On the contrary, the gate terminal of the n-MOS transistor M6n approaches the supply voltage V_dd, because the PCM element S1 exhibits a low resistance and the supply voltage line PL1_j is set to the supply voltage V_dd. The n-MOS transistor M6n turns on, pulling the match node MG towards the low positive voltage V_L. Then, the p-MOS transistor M4p is turned on and the match line ML_i is pulled down towards the ground voltage, provided by the reference voltage line RL_i, revealing the bit-mismatch (and a word-mismatch). It is underlined that the bitline BLR_j needs not necessarily be set to the ground voltage, being sufficient to set the low positive voltage V_L, assuming that such a voltage has a value suitable to turn the p-MOS transistor M4p on (V_L<V_ML−V_th4, where V_th4 is the threshold voltage of the p-MOS transistor M4p).

[0091] When the input bit is 0, the bitline BLR_j is set to the supply voltage V_dd, while the bitline BLL_j is set to the low positive voltage V_L. Consequently, the voltage at the gate terminal of the n-MOS transistor M6n moves towards the supply voltage V_dd, provided by the supply voltage line PL1_j, because the PCM element S1 exhibits a low resistance. The gate terminal of the n-MOS transistor M5n is also at the supply voltage V_dd, provided by the bitline BLR_j, because the transistor M2p is on, and the match node MG thus moves to an intermediate voltage between the supply voltage V_dd at the bitline BLR_j and the low positive voltage V_L at the bitline BLL_j. For this reason, it might be preferable not to set the voltage of the bitline BLL_j too close to the ground voltage, so as to ensure that the p-MOS transistor M4p is kept turned off, revealing the bit-match condition (the match line ML_i is kept at the preset voltage V_ML if all the CAM cells 105 of a row reveal a bit-match).

[0092] In the AM storage state, the two n-MOS transistors M5n, M6n are turned on, because the two gate terminals thereof are always at the supply voltage V_dd, as in the above described matching condition (both the PCM elements S1, S2 exhibit a low resistance). In order to ensure that the transistor M4p is kept off, it might be preferable not to drive one of the two bitlines BLR_j, BLL_j to the ground voltage, but to stop at a relative low positive voltage, e.g., the above mentioned voltage V_L. Also in this case the importance of the choice of the low search voltage V_L is to be outlined.

[0093] As shown in the alternative embodiment of FIG. 5B, the CAM cell 105 of FIG. 5A can be modified by substituting the p-MOS transistors M1p, M2p with the n-MOS transistors M1n, M2n. The N-MOS transistor M1n has the drain terminal connected to the bitline BLR_j, i.e., to the first drain/source terminal of the n-MOS transistor M6n. The n-MOS transistor M2n has the drain terminal connected to the bitline BLL_j, i.e., to the first drain/source terminal of the n-MOS transistor M5n. The positive terminal of the PCM element S1 is connected to the source terminal of the n-MOS transistors M1n and to the gate terminal of the n-MOS transistors M5n, while the negative terminal is connected to the reference voltage line RL_i, used to deliver the ground voltage. The positive terminal of the PCM element S2 is connected to the source terminal of the n-MOS transistors M2n and to the gate terminal of the n-MOS transistors M6n, while the negative terminal is connected to the reference voltage line RL_i.

[0094] The writing operation of the CAM cell 105 is accomplished in two steps, being the two PCM elements S1 and S2 programmed sequentially. A high voltage (e.g., from 3 to 5 V) is placed onto the wordline WL_i and one of the bitlines BLR_j, BLL_j, while the other bitline is kept at the ground voltage. In particular, similarly to the above described embodiment of FIG. 5A, the bitline BLL_j, BLR_j connected to the PCM element S1, S2 to be programmed is set (by driver circuits not shown in the drawing) to the voltage adapted to permit a suitable current pulse to flow through the respective PCM element S1, S2.

[0095] The four possible storage states for this further embodiment of CAM cell 105 are defined in Table 3.

[0096]

TABLE 3
S1	S2	CAM cell
logic value	logic value	storage state
1	1	not allowed
0	1	1
1	0	0
0	0	Never Match (NM)

[0097] The storage state in which both the two PCM elements S1, S2 are set to the logic value 0 is referred to as Never Match (NM) state, while the storage state in which both the two PCM elements S1, S2 are set to the logic value 1 is not allowed, for the reasons explained later on.

[0098] Before starting a search operation, both the bitlines BLL_j and BLR_j are preset to the supply voltage V_dd, and a sufficiently high positive voltage V_search^WL is applied to the wordline WL_i, so as to turn the transistors M1n and M2n on. In this way, at least one of the n-MOS transistors M5n and M6n is turned on, presetting the match node MG to a voltage close to the supply voltage V_dd minus the threshold voltage of the transistors M5n and M6n (i.e., the p-MOS transistor M4p is kept turned off).

[0099] The wordline search voltage V_search^WL must be high enough to turn the n-MOS transistors M1n and M2n on, avoiding at the same time the risk of having the switch voltage V_switch^P across the PCM elements S1 and S2. In this way, the wordline search voltage V_search^WL must satisfy the following conditions:
V_search^WL>V_th1,2
and
V_search^WL<V_switch^p+V_th1,2−ΔV,
where V_th1,2 is the threshold voltage of the n-MOS transistors M1n, M2n.

[0100] Supposing that the content bit of the CAM cell is 1, when the input (search) bit is 1, the bitline BLL_j is left at the supply voltage V_dd and the bitline BLR_j is set to the ground voltage. Both the n-MOS transistors M5n and M6n turn off, keeping also the p-MOS transistor M4p turned off (bit-match). On the contrary, when the input bit is 0, the bitline BLL_j is set to the ground voltage and the bitline BLR_j is set to the supply voltage V_dd. In this way the n-MOS transistor M5n turns on, moving the voltage at the match node MG towards the ground voltage and turning the p-MOS transistor M4p on. The match line ML_i is thus pulled towards the ground voltage, revealing the bit-mismatch.

[0101] The NM storage state (both the PCM elements S1 and S2 exhibit a high resistance) permits to turn the p-MOS transistor M4p on both in the case in which the input bit is 1 and in the case in which the input bit is 0, achieving always a bit-mismatch.

[0102] It is observed that the storage state in which both the PCM elements S1, S2 exhibit a low resistance is not allowed because the match node MG could remain floating, and this condition has to be avoided.

[0103] In the CAM cell embodiment illustrated in FIG. 5B the n-MOS transistors preferably are low-threshold voltage transistors, in order to increase the speed and the reliability of the CAM cell 105. In fact, the transistors M5n and M6n are driven by the voltage drop across the PCM elements S1, S2, limited to a value lower than the switch value V_switch^p. If the transistors M5n, M6n have a low threshold voltage, the voltage swing at the gate terminal thereof can be more efficiently exploited for driving the match node MG, i.e., for transferring the bitline voltage to the match node MG.

[0104] Thanks to the fact that PCM elements are exploited as storage elements, the non-volatile CAM cell according to the present invention is faster than CAM cells exploiting floating gate transistors, especially in the write and erase operations. It is possible to accomplish a search operation in a time of the order of ten nanoseconds, while the time of a writing operation, although depending on the logic value to be programmed in the PCM elements, does not exceed the hundred of nanoseconds.

[0105] Moreover, the area occupied by the CAM cell on the semiconductor chip is smaller than that of the SRAM-based CAM cells (which moreover is volatile), due to the lower number of the transistors exploited and the compactness of the PCM elements.

[0106] It is observed that during a search operation, currents flow through the CAM cells, because the PCM elements are resistive components. However, this is not a major concern, because these currents, and consequently the power consumption, can be greatly reduced by carefully regulating the wordline search voltage V_search^WL.

[0107] FIG. 6 shows an alternative embodiment of the present invention adapted to reducing the power consumption during the search operations. The non-volatile CAM cell 105 according to this further embodiment of the invention has a split architecture, with a non-volatile part and a fast but volatile part.

[0108] The non-volatile part includes the n-MOS transistor M1n with the gate terminal connected to the wordline WL_i. The n-MOS transistor M1n acts as the selecting element for a PCM element S, having the positive terminal connected to the source terminal of the n-MOS transistor M1n and the negative terminal connected to the reference voltage line RL_i (delivering the ground voltage). The non-volatile part of CAM cell 105 further includes a p-MOS transistor M7p with the source terminal connected to the supply voltage line PL_j, delivering the supply voltage V_dd. The p-MOS transistor M7p has the gate terminal connected to a bitline BLP_j, used for writing/reading the PCM element S, and the drain terminal connected to the drain terminal of the transistor M1n. The p-MOS transistor M7p is used to provide the suitable currents to the PCM element S during operation.

[0109] The volatile part of the CAM cell 105 includes two inverters 610 and 615, particularly CMOS inverters, and an n-MOS transistor M8n. The input terminal of the inverter 610 (node DN in the drawing) is connected to the drain terminal of the transistors M1n, M7p and to a first source/drain terminal of the transistor M8n, while the output terminal of the inverter 610 is connected to the input terminal of the inverter 615; the output terminal of the inverter 615 is connected to a second source/drain terminal of the transistor M8n. The transistor M8n receives an enabling signal EN at the gate terminal thereof; the enabling signal EN is generated inside the CAM 100 and can assume the value of the supply voltage V_dd, for turning the n-MOS transistor M8n on, or the value of the ground voltage, for turning the n-MOS transistor M8n off. In this pass transistor configuration the n-MOS transistor M8n acts as a controlled switch, which can open or close a loop formed by the two inverters 610, 615. When the loop 610, 615 is closed, the two inverters 610, 615 implement a volatile storage element (a latch) similar to a memory cell of a Static RAM memory. Furthermore, the volatile part of the CAM cell 105 includes the n-MOS transistors M5n, M6n and the p-MOS M4p. The n-MOS transistors M5n has the gate terminal connected to the output terminal of the inverter 615, while the n-MOS transistors M6n has the gate terminal connected to the input terminal of the inverter 615 (node D in the drawing). Similarly to the previous embodiments, the transistors M5n, M6n are connected to the two bitlines BLL_j, BLR_j, used during a search operation.

[0110] During a writing operation, the bitline BLP_j is set to the ground voltage, and the p-MOS transistor M7p is turned on. In addition, the enabling signal EN is set to the ground voltage, in order to turn the n-MOS transistor M8n off and thus keep the loop 610, 615 open; in addition, the bitlines BLL_j and BLR_j are kept floating. In this way, the two inverters 610, 615 are not allowed to latch any logic value during the writing operation. Voltages V_set^WL, V_reset^WL are placed on the wordline WL_i, allowing the flow of currents at least equal to the currents I_set, I_reset, respectively, through the PCM element S.

[0111] If the logic value 0 has to be written into the PCM element S, the reset voltage V_reset^WL (typically, 3 V for this embodiment) is applied to the wordline WL_i. The reset voltage V_reset^WL at the gate terminal of the n-MOS transistor M1n has to assure that a current at least equal to the reset current I_reset flows through the PCM element S, irrespective of the possible state of the phase-change material; in particular, the reset voltage V_reset^WL has to be higher than the voltage V_switch^p for assuring a phase change in case the material is in the amorphous phase. The p-MOS transistor M7p, driven by a suitable bitline voltage, can be used as a current limiting device to limit the current to the reset value I_reset. The phase-change material, when cooled rapidly (e.g., by rapidly decreasing the reset voltage V_reset^WL at the gate terminal of the n-MOS transistor M1) amorphizes exhibiting high resistance.

[0112] If the logic value 1 has to be written into the PCM element S, the set voltage V_set^WL (typically, 1.5 V) is applied to the wordline WL_i. The set voltage V_set^WL at the gate terminal of the n-MOS transistor M1 has to assure that a current at least equal to the set current I_set flows through the PCM element S, irrespective of the possible state of the phase-change material; in particular, the set voltage V_set^WL has to be at least equal to the switch value V_switch^p for assuring a phase change in case the material is in the amorphous phase. Also in this case the p-MOS transistor M7p can be used as a current limiting device to limit the current to the set value I_set. The phase-change material, when cooled slowly (by slowly decreasing the voltage at the gate terminal of the n-MOS transistor M1n) crystallizes exhibiting low resistance.

[0113] The supply voltage line PL_j can also be used as a verify line, enabling, for example, a Direct Memory Access (DMA) to the PCM element S of a selected CAM cell 105 in the matrix for establishing the status thereof by sensing the current.

[0114] It is observed that the non-volatile CAM cell 105 of the embodiment of FIG. 6 is binary, only two storage states being possible. The two storage states correspond to the logic complement of the logic value stored in the PCM element S. When the PCM element S stores the logic value 0, the CAM cell storage state is 1 and, when the PCM element S stores the logic value 1, the storage state is 0.

[0115] After the writing operation, the logic value, stored in the PCM element S, needs to be transferred at the node DN (transferring operation). The bitline BLP_j is biased to a voltage such that the p-MOS transistor M7p is turned on and the node DN can be driven up/down by the current flowing through the PCM element S, for example the ground voltage. The wordline WL_i, controlling the transistor M1, has to be brought to a voltage lower than that used in the writing operation. In detail, a transferring value V_T (typically, about 0.8 V) lower than the switch value V_switch^p, is placed on the wordline WL_i, in order to prevent any spurious programming of the PCM element S during the transferring operation, but suitable to keep the transistor M1n turned on; thus, the transferring value V_T needs to be higher than the threshold voltage of the transistor M1n. During this transferring operation, the enabling signal EN is kept at the ground voltage and then the loop 610, 615 is kept open.

[0116] The PCM element S exhibits high resistance when the logic value 0 is stored therein, and the p-MOS transistor M7p pulls in this case the voltage at the node DN up towards the value of the supply voltage V_dd. On the contrary, the PCM element S exhibits low resistance when the logic value 1 is stored therein, and the n-MOS transistor M1 pulls the voltage at the node DN down towards the ground voltage.

[0117] As for the previous embodiments, before starting a search operation the match line ML_i is preset to the preset voltage V_ML and then, after the transferring operation, the enabling signal EN is set to the supply voltage V_dd and the n-MOS transistor M8n turns on, closing the loop 610, 615. In a way similar to a volatile SRAM cell, the loop 610, 615 quickly latches the logic value corresponding to the value of the voltage at the node DN and provides the logic complement thereof at the node D. Consequently, when the node DN is at the ground voltage, the node D is at the supply voltage V_dd and vice versa. It has to be observed that the transistors M7p, M1n can be turned off after the transferring operation.

[0118] Consequently, when the PCM element S stores the logic value 1, the node DN reaches a voltage value close to the ground voltage, turning the n-MOS transistor M5n off, while the node D reaches a voltage value close to the supply voltage Vdd, turning the n-MOS transistor M6n on. If the input bit 1 is being searched, the bitline BLR_j is set to the supply voltage V_dd and the bitline BLL_j is set to the ground voltage. In this way, the n-MOS transistor M6n, turned on, pulls the match node MG towards the supply voltage V_dd, turning the p-MOS M4p off; the match line ML_i is kept at the preset voltage V_ML, revealing the matching condition.

[0119] Otherwise, if the input bit 0 is being searched, the bitline BLL_j is set to the supply voltage V_dd and the bitline BLR_j is set to the ground voltage. In this way, the n-MOS transistor M6n, turned on, pulls the match node MG towards the ground voltage turning the p-MOS M4p on; the match line ML_i is discharged towards the ground voltage revealing the mismatching condition.

[0120] It is observed that the transferring operation is required not only after the programming operation of the CAM cell, but also at every power-on of the CAM 100. However, the simple structure of the CAM cell 105, that combines the properties of both non-volatile and volatile memory elements, permits a fast transferring operation, lasting only few tens of nanoseconds. During a search operation the PCM element needs not be sensed, i.e., the resistive element need not be run through by any current, consequently achieving a reduction of power consumption. In fact, the splitting of the CAM cell 105 into the non-volatile part and the fast part permits a separate use thereof.

[0121] Alternatively, the n-MOS transistor M8n, which is in pass transistor configuration, can be replaced by a CMOS pass transistor, if needed, for a better drive of the signal closing the loop 610, 615. Furthermore, the p-MOS transistor M4p, used for discharging the match line ML_i in mismatching condition, can be replaced by a n-MOS transistor and the connections of the two gate terminals thereof has to be inverted. In this last case the preset operation must be executed by setting the bitlines BLL_j and BLR_j to the ground voltage.

[0122] However, the concepts of the present invention are also applicable when the CAM cells are connected to a different number of bitlines and supply voltage lines; similar considerations apply if the CAM cells are connected to different auxiliary lines, such as further control lines.

[0123] All of the above U.S. patents, U.S. patent application publications, U.S. patent applications, foreign patents, foreign patent applications and non-patent publications referred to in this specification and/or listed in the Application Data Sheet are incorporated herein by reference, in their entirety.

[0124] Naturally, in order to satisfy specific requirements, a person skilled in the art may apply to the solution described above many modifications and alterations all of which, however, are included within the scope of protection of the invention as defined by the following claims.