Bridge-type magnetic random access memory (MRAM) latch

[0003] 1. Field of the Invention

[0004] The invention generally relates to memory technology. In particular, the invention relates to non-volatile magnetic memory.

[0005] 2. Description of the Related Art

[0006] Computers and other digital systems use memory to store programs and data. A common form of memory is random access memory (RAM). Many memory devices, such as dynamic random access memory (DRAM) devices and static random access memory (SRAM) devices, are volatile memories. A volatile memory loses its data when power is removed. For example, after a conventional personal computer is powered off, the volatile memory is reloaded upon a boot up. In addition, certain volatile memories, such as DRAM devices, require periodic refresh cycles to retain their data even when power is continuously supplied.

[0007] In contrast to the potential loss of data encountered in volatile memory devices, nonvolatile memory devices retain data for long periods of time when power is removed. Examples of nonvolatile memory devices include read only memory (ROM), programmable read only memory (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), flash memory, and the like. Disadvantageously, conventional nonvolatile memories are relatively large, slow, and expensive. Further, many types of conventional semiconductor nonvolatile memories arm relatively limited in write cycle capability and typically can only be programmed to store data about 10,000 times in a particular memory location. This prevents a conventional non-volatile memory device, such as a flash memory device, from being used as general purpose memory.

[0008] An alternative memory device is known as magnetoresistive random access memory (MRAM). An MRAM device uses magnetic states to retain data in its memory cells. Advantageously, MRAM devices are relatively fast, are nonvolatile, consume relatively little power, and do not suffer from a write cycle limitation. There are at least three different types of MRAM devices, including giant magneto-resistance (GMR) MRAM devices, magnetic tunnel junction (MTJ) or tunneling magneto-resistance (TMR) MRAM devices, and pseudo spin valve (PSV) MRAM devices. GMR MRAM devices separate at least two ferromagnetic layers with a metallic layer. In a MTJ MRAM device, at least two ferromagnetic layers are separated by a thin insulating tunnel barrier. A PSV MRAM device uses an asymmetric sandwich of the ferromagnetic layers and metallic layer as a memory cell, and the ferromagnetic layers are driven so that they do not switch at the same time.

[0009] An interface circuit detects and latches the memory state of a magnetic memory cell so that the contents of the cell can be provided to another circuit, such as a central processing unit (CPU). Disadvantageously, conventional circuits that latch the state of the magnetic memory cell are relatively sensitive to variations in transistor parameters such as gate length. These sensitivities can disadvantageously increase production costs and can also cause data to be incorrectly read from the magnetic memory cell. Embodiments of the invention solve these and other problems by providing a latching circuit that is relatively less sensitive to variations in transistor parameters.

[0010] The invention relates to techniques to read a stored state in a magnetoresistive random access memory (MRAM) memory device. Advantageously, an interface circuit or latching circuit according to an embodiment of the invention is relatively insensitive to transistor parameter variations, such as gate length. This advantageously allows the interface circuit to be manufactured less expensively and also allows the interface circuit to reliably detect logic states with relatively small differences in resistance.

[0011] A bit line in an MRAM device is segmented into a first portion and a second portion. An interface circuit compares the resistance associated with a first portion and a second portion of a first bit line to the resistance associated with a first portion and a second portion of a second bit line to determine the logical state of a cell in the first bit line. The interface circuit includes a reset circuit to allow the circuit to achieve a balanced state by coupling outputs of the interface circuit to a relatively low resistance. Decoupling of the outputs allows cross-coupling within the interface circuit to latch the outputs to a logical state corresponding to the stored magnetic state, thereby allowing the stored state of a cell to be read.

[0012] One embodiment according to the invention includes an MRAM memory latching circuit that includes a reset circuit, a first n-type transistor, a second n-type transistor, a first p-type transistor, and a second p-type transistor. The memory latching circuit can be used to latch the stored state of a corresponding memory cell and can also be used as an interface between the memory cell and another device, such as a central processing unit (CPU). The reset circuit includes a first terminal, a second terminal, and a third terminal. The reset circuit couples the second terminal to the third terminal in response to a first state of an input signal applied to the first terminal, and the reset circuit decouples the second terminal from the third terminal in response to a second state of the input signal applied to the first terminal. Coupling of the second terminal to the third terminal effectively disables cross-coupled feedback across the memory latching circuit. The first n-type transistor includes a source coupled to a first portion of a first bit line. The second n-type transistor includes a source coupled to a first portion of a second bit line. The first p-type transistor includes a source coupled to a second portion of the second bit line, where a drain of the first p-type transistor is coupled to a drain of the first n-type transistor and to a first output. A gate of the first p-type transistor is coupled to a gate of the first n-type transistor, to a drain of a second n-type transistor, to a drain of a second p-type transistor, and to the second terminal of the reset circuit. The second p-type transistor includes a source coupled to a second portion of the first bit line, where the drain of the second p-type transistor is coupled to the drain of the second n-type transistor and to a second output. A gate of the second p-type transistor is coupled to a gate of the second n-type transistor, to the drain of the first n-type transistor, to the drain of the first p-type transistor, and to the third terminal of the reset circuit. In one embodiment, the transistors are metal-oxide semiconductor field-effect transistors (MOSFETs).

[0013] Another embodiment according to the invention includes an MRAM latching circuit. The latch includes a first terminal adapted to couple to a first portion of a first bit line, a second terminal adapted to couple to a second portion of the first bit line, a third terminal adapted to couple to a first portion of a second bit line, and a fourth terminal adapted to couple to a second portion of the second bit line. The memory latching circuit further includes an input terminal adapted to receive a reset signal, where a first state of the reset signal couples a first output to a second output, and where a second state of the reset signal decouples the first output from the second output so that the first output and the second output latch in opposite directions in response to an imbalance in resistance between the first bit line and the second bit line.

[0014] Another embodiment according to the invention includes a bit line in an MRAM. The bit line includes a first portion and a second portion coupled to magnetically sensitive resistors. The first portion is adapted to couple to a voltage source and to an interface circuit The second portion is adapted to couple to a voltage reference (e.g., ground) and to the interface circuit. The first portion is electrically connected to the second portion through the interface circuit. In one embodiment, the first portion and the second portion contain about the same number of magnetically sensitive resistors.

[0015] Another embodiment according to the invention includes an MRAM device. The MRAM device includes a plurality of word lines and a plurality of bit lines. A bit line in the plurality of bit lines is coupled to a first portion of magnetically sensitive resistors and is also coupled to a second portion of magnetically sensitive resistors. A memory state of a cell in the MRAM device can be determined by passing a first current through a corresponding word line and a second current through at least one corresponding bit line. Where the MRAM device is a GMR, a corresponding bit line is lower in resistance when a magnetic field in a hard layer of the cell is aligned with a field induced by the first current and the second current than when the magnetic field is not aligned. The number of resistors coupled to first portions and second portions of bit lines can be the same. In one embodiment, at least one pair of bit lines is coupled to a different number of resistors in the corresponding first portion and second portion than another pair of bit lines.

[0016] One embodiment according to the invention includes a method of determining a magnetic state in an MRAM cell. The method includes applying a reset signal to an interface circuit such that at least one output of the interface circuit is reset to a balanced state, removing the reset signal, comparing an amount of resistance associated with a first portion of a first bit line to an amount of resistance associated with a first portion of a second bit line, comparing an amount of resistance associated with a second portion of the first bit line to an amount of resistance associated with a second portion of the second bit line, and generating an output with a logic state corresponding to the magnetic state stored in the cell based on the comparisons.

[0017] Another embodiment according to the invention includes a method of latching a magnetic state in an MRAM cell, such as a GMR cell. The method includes applying current to a word line corresponding to the MRAM cell. In a rest state, the method includes neutralizing cross-coupled feedback of an interface circuit to allow the interface circuit to balance in response to current applied through first and second portions of a first bit line and through first and second portions of a second bit line, where the current applied to the first bit line and the current applied through the second bit line are substantially equal, and where the first bit line corresponds to the MRAM cell. In a latched state, the method includes enabling cross-coupled feedback across the interface circuit that allows an imbalance in resistance from a stored state of the cell to latch the cell in a corresponding logical state.

[0018] These and other features of the invention will now be described with reference to the drawings summarized below. These drawings and the associated description are provided to illustrate preferred embodiments of the invention and are not intended to limit the scope of the invention.

[0019] FIG. 1 is a perspective view illustrating a giant magneto-resistance (GMR) cell in a spin valve mode.

[0020] FIG. 2 is a schematic top-down view illustrating an array of GMR cells.

[0021] FIG. 3 illustrates a GMR cell in a pseudo spin valve (PSV) mode.

[0022] FIG. 4 illustrates a conventional latching circuit.

[0023] FIG. 5 illustrates a latching circuit according to an embodiment of the invention.

[0024] FIG. 6 is a schematic of a bit line of an array of magnetoresistive random access memory (MRAM) cells according to an embodiment of the invention.

[0025] Although this invention will be described in terms of certain preferred embodiments, other embodiments that are apparent to those of ordinary skill in the art, including embodiments that do not provide all of the benefits and features set forth herein, are also within the scope of this invention. Accordingly, the scope of the invention is defined only by reference to the appended claims.

[0026] A magnetoresistive random access memory (MRAM) stores data in magnetic states of its memory cells. The electrical resistance of the cell varies depending on the stored magnetic state of the cell. The stored state of the cell is detected by sensing the difference in resistance.

[0027] While illustrated in the context of giant magneto-resistance (GMR) memories, the skilled artisan will appreciate that the principles and advantages described herein are applicable to other types of magnetic memories, such as tunneling magneto-resistance (TMR) MRAMs.

[0028] FIG. 1 is a perspective view illustrating a GMR cell 100 in a spin valve mode. The GMR cell 100 includes a word line 102 and a bit line 104. In a GMR cell, the bit line 104 is also known as a sense line. The bit line 104 contains magnetic layers. Data is stored in a cell body portion of the bit line 104 by simultaneously applying current through the word line 102 and the bit line 104. The direction of the current in the bit line 104 determines the polarization of the magnetic field that stores the logical state of the data. The polarization can be clockwise around the bit line 104 for a first state, and counterclockwise around the bit line 104 for a second state. Where the magnetic memory cell corresponds to a tunneling magneto-resistance (TMR) configuration, a TMR cell is disposed between the bit line 104 and the word line 102.

[0029] To read data from tee GMR cell 100, currents are again applied to the word line 102 and the bit line 104 corresponding to the GMR cell 100. The resistance encountered by the current applied to the bit line 104 varies depending on the logical state stored in the magnetic layers. A cell with a larger resistance exhibits a larger voltage drop in response to the current than a cell with a smaller resistance.

[0030] FIG. 2 is a schematic top-down view illustrating an array 200 of GMR cells. A plurality of cells are arranged into the array 200 in a memory device. The array 200 of cells includes a plurality of word lines 202 and a plurality of bit lines 204. An individual cell within the array 200 is selected by applying current through the corresponding word line and the corresponding bit line. Data is not stored or read in a cell where current flows through only one of the cell's word line or bit line. Thus, data is only stored in a cell where the applied word line current and the applied bit line current intersect (i.e., both currents are present at the same cell).

[0031] FIG. 3 illustrates a GMR cell 300 in a pseudo spin valve (PSV) mode. The GMR cell 300 includes a word line 302 and a bit line 304. The bit line 304 of the GMR cell 300, which is also known as a sense line, further includes a first magnetic layer 306, a conductive layer 308, and a second magnetic layer 310. The first magnetic layer 306 and the second magnetic layer 310 are mismatched so that the first magnetic layer 306 is magnetically “softer” than the second magnetic layer 310. The mismatch in magnetic properties can be obtained by making the first magnetic layer 306 relatively thin as compared to the second magnetic layer 310, by selecting a relatively soft magnetic material for the first magnetic layer 306 and a relatively hard magnetic material for the second magnetic layer 310, or by making the first magnetic layer 306 thinner and magnetically softer than the second magnetic layer 310.

[0032] The GMR cell 300 stores the logical state in the second magnetic layer 310. A relatively high magnetic field is required to switch the magnetization of the second magnetic layer 310. The magnetization of the second magnetic layer 310 is read by switching the magnetization of the first magnetic layer 306, which can be switched with a relatively low magnetic field generated by applying currents to the corresponding word line 302 and the corresponding bit line 304. The resulting magnetization of the first magnetic layer 306 is either parallel or anti-parallel to the magnetization of the second magnetic layer 310, which stores the state. When the magnetization in the first magnetic layer 306 is relatively aligned with or in parallel with the magnetization of the second magnetic layer 310, the electrical resistance of the GMR cell 300 is lower than when the magnetization of the first magnetic layer 306 is relatively not aligned with or is anti-parallel to the magnetization of the second magnetic layer 310. In one embodiment, current is applied in the word line 302 in both directions to peruse the magnetization of the second magnetic layer 310. The difference in electrical resistance of the bit line 304 is then sensed, thereby allowing the stored logical state of the GMR cell 300 to be retrieved. In a TMR memory cell, a cell body is formed between the bit line and the word line, and the conductive layer 308 is replaced by a relatively thin insulating layer, which is thin enough to allow the tunneling of electrons from one magnetic layer to the other magnetic layer to detect memory states.

[0033] FIG. 4 illustrates a conventional latching circuit 400. The conventional latching circuit 400 is disadvantageously relatively sensitive to transistor parameter variations, such as gate length. The conventional latching circuit 400 detects a logic state by comparing the resistance of a first magnetically sensitive resistor R_A402, with the resistance of a second magnetically sensitive resistor R_B404. Although the first magnetically sensitive resistor R_A402 and the second magnetically sensitive resistor R_B404 are schematically illustrated with a single resistor symbol, the first magnetically sensitive resistor R_A402 and the second magnetically sensitive resistor R_B404 correspond to a combination of multiple resistances from a bit line or a sense line that corresponds to multiple memory cells in an array.

[0034] The resistance of the first magnetically sensitive resistor R_A402, the resistance of the second magnetically sensitive resistor R_B404, or the resistances of both resistors can vary depending on the stored state of the selected memory cell, which is the cell corresponding to the cell at the intersection of the applied word line current and the applied bit line current.

[0035] Initially, a reset transistor M₅406 is switched “on” by providing a relatively high signal to the gate of the reset transistor M₅406 from an input terminal 408. The “on” state of the reset transistor M₅406 balances the conventional latching circuit 400 by electrically tying the gate of a first n-type transistor M₁410 to the gate of a second n-type transistor M₂412. The first n-type transistor M₁410 and the second n-type transistor M₂412 are matched. The “on” state of the reset transistor M₅406 also electrically ties the gate of a first p-type transistor M₃414 with the gate of a second p-type transistor M₄416. The first p-type transistor M₃414 and the second p-type transistor M₄416 are also matched.

[0036] To illustrate the latching operation of the conventional latching circuit 400, the operation of the conventional latching circuit 400 will now be described with the first magnetically sensitive resistor R_A402 at a lower resistance than the second magnetically sensitive resistor R_B404. When the resistance of the first magnetically sensitive resistor R_A402 is lower than the resistance of the second magnetically sensitive resistor R_B404, the source of the first n-type transistor M₁410 should be at a lower voltage relative to voltage at the source of the second n-type transistor M₂412. Since the gate of the first n-type transistor M₁410 and the gate of the second n-type transistor M₂412 are tied to the same potential, the first n-type transistor M₁410 is relatively more conductive from drain to source than the second n-type transistor M₂412 because of the larger gate to source voltage of the first n-type transistor M₁410.

[0037] To latch the conventional latching circuit 400, the reset transistor M₅406 is turned “off” by deactivating the relatively high input signal previously applied to the gate of the reset transistor M₅406 via the input terminal 408. When operating as intended, the conventional latching circuit 400 latches into an electrical state determined by the mismatch in resistance between the first magnetically sensitive resistor R_A402 and the second magnetically sensitive resistor R_B404. When the first magnetically sensitive resistor R_A402 has a lower resistance than the second magnetically sensitive resistor R_B404, the conventional latching circuit 400 should switch to a state where a first output OUT₁418 is at a “low” state and a second output OUT₂420 is at a “high” state.

[0038] When the reset transistor M₅406 is turned “off,” the gates of the first n-type transistor M₁410 and the first p-type transistor M₃414 are decoupled from the gates of the second n-type transistor M₂412 and the second p-type transistor M₄416. Because the drain of the first n-type transistor M₁410 is relatively more conductive than the second n-type transistor M₂412, the voltage at the first output OUT₁418 is relatively low as compared with the voltage at the second output OUT₂420. The relatively low voltage at the first output OUT₁418 is cross-coupled to the gate of the second n-type transistor M₂412 and to the gate of the second p-type transistor M₄416. A relatively low voltage on the gate of the second n-type transistor M₂412 tends to turn the second n-type transistor M₂412 “off” (i.e., less conductive). A relatively low voltage on the gate of the second p-type transistor M₄416 tends to turn the second p-type transistor M₄416 “on” (i.e., more conductive). This causes the voltage of the second output OUT₂420 to rise.

[0039] A relatively high voltage on the second output OUT₂420 is cross-coupled to the first n-type transistor M₁410 and to the first p-type transistor M₃414. A relatively high voltage on the gate of the first n-type transistor M₁410 tuns the first n-type transistor M₁410 more “on” (i.e., more conductive). A relatively high voltage on the gate of the first p-type transistor M₃414 causes the first p-type transistor M₃414 to turn “off” (i.e., less conductive). This causes the voltage of the first output OUT₁418 to go lower. The process continues until the first output OUT₁418 is at a “low” state and the second output OUT₂420 is at a “high” state.

[0040] The skilled artisan will appreciate that relatively closely matched transistors are difficult and expensive to manufacture. In addition, the magnitude of the difference in resistance between one logic state and another logic state is relatively small, e.g., a few percent. Where, for example, the drain currents of the first p-type transistor M₃414 and the second p-type transistor M₄416 are mismatched by more than the difference in resistance between logic states, the final state of the latch will be undesirably determined by the transistor mismatch rather than the stored logic state, and the logic state may not be correctly reported. The drain current in the first p-type transistor M₃414 or in the second p-type transistor M₄416 is approximately inversely proportional to the gate length, L, of the transistor. Thus, for the conventional latching circuit 400 to properly report data, the gate lengths of the first p-type transistor M₃414 and the second p-type transistor M₄416 should be matched to at least the difference in resistance between logic states.

[0041] FIG. 5 illustrates an improved latching circuit 500 according to an embodiment of the invention. The latching circuit 500 more reliably detects logical states of magnetically sensitive resistors, such as those used in GMR and in TMR memory devices, than the conventional latching circuit 400 described above in connection with FIG. 4. The latching circuit 500 is advantageously relatively insensitive to transistor parameter variations, such as gate length. This advantageously allows the latching circuit 500 to be manufactured less expensively and also allows the latching circuit 500 to reliably detect logic states with relatively small differences in resistance.

[0042] Although the latching circuit 500 is illustrated with n-channel and p-channel metal-oxide semiconductor field-effect transistors (MOSFETs) for the n-type and p-type transistors, respectively, it will be understood by one of ordinary skill in the art that the circuit can also be implemented with other technologies. For example, silicon-germanium (SiGe) bipolar transistors, silicon bipolar transistors, gallium arsenide metal semiconductor field-effect transistors (GaAs FETs or MESFETs), heterojunction bipolar transistors (HBTs), SiGe bipolar complementary metal-oxide semiconductor (BiCMOS) and the like can also be used.

[0043] The latching circuit 500 detects a logic state by comparing the resistance of a first bit line with the resistance of a second bit line. When the magnetic cells are GMR cells, the first bit line and the second bit line include a plurality of magnetically sensitive resistors for their respective cells that are connected in series. When the magnetic cells are TMR cells, the first bit line and the second bit line carry current coupled from word lines through the TMR cells in addition to current applied to the bit line. In the illustrated latching circuit 500, the first bit line is segmented into two portions. The bit line can be segmented into more than two portions. In one embodiment, the bit line is segmented into two approximately equal portions.

[0044] A first portion of the first bit line is schematically illustrated as a first magnetically sensitive resistor R₁502, and a second portion of the first bit line is schematically illustrated as a second magnetically sensitive resistor R₂504. A first portion of the second bit line is schematically illustrated as a third magnetically sensitive resistor R₃506, and a second portion of the second bit line is schematically illustrated as a fourth magnetically sensitive resistor R₄508. It will be understood by one of ordinary skill in the art that the portions of the first bit line and the second bit line schematically illustrated by the first magnetically sensitive resistor R₁502, the second magnetically sensitive resistor R₂504, the third magnetically sensitive resistor R₃506, and the fourth magnetically sensitive resistor R₄508 include combinations of multiple magnetically sensitive resistors.

[0045] The combined resistance (i.e., the series resistance) of the first magnetically sensitive resistor R₁502 and the second magnetically sensitive resistor R₂504, the combined resistance of the third magnetically sensitive resistor R₃506 and the fourth magnetically sensitive resistor R₄508, or both combined resistances can vary depending on the stored state of the selected memory cell, which is the cell corresponding to the cell with the applied word line current and the applied bit line current.

[0046] Initially, a reset transistor M₅510 is switched “on” by providing a relatively high signal to the gate of the reset transistor M₅510 from an input terminal 512. The illustrated reset transistor M₅510 is an n-type transistor. However, the skilled practitioner will appreciate that a p-type transistor can also be used. The “on” state of the reset transistor M₅510 balances the latching circuit 500 by electrically tying the gate of a first n-type transistor M₁514, the gate of a second n-type transistor M₂516, the gate of a first p-type transistor M₃518, and the gate of a second p-type transistor M₄520 together. The first n-type transistor M₁514 and the second n-type transistor M₂516 are preferably matched to each other. The first p-type transistor M₃518 and the second p-type transistor M₄520 are also preferably matched to each other.

[0047] To illustrate the latching operation of the latching circuit 500, the operation of the latching circuit 500 will now be described with the reset transistor M₅510 switched “on” and with the combined resistance of the first magnetically sensitive resistor R₁502 and the second magnetically sensitive resistor R₂504 (i.e., the resistance of the first bit line) at a lower resistance than the combined resistance of the third magnetically sensitive resistor R₃506 and the fourth magnetically sensitive resistor R₄508 (i.e., the resistance of the second bit line).

[0048] When the combined resistance of the first magnetically sensitive resistor R₁502 and the second magnetically sensitive resistor R₂504 is lower than the combined resistance of the third magnetically sensitive resistor R₃506 and the fourth magnetically sensitive resistor R₄508, then one of the first magnetically sensitive resistor R₁502 or the second magnetically sensitive resistor R₂504 is lower in resistance than the corresponding one of the third magnetically sensitive resistor R₃506 or the fourth magnetically sensitive resistor R₄508. The other resistor in each series resistance should have about the same resistance (e.g., if the resistance of the first resistor R₁502 is lower than the resistance of the third resistor R₃506, then the resistance of the second resistor R₂504 is approximately equal to the resistance of the fourth resistor R₄508).

[0049] When the resistance of the first magnetically sensitive resistor R₁502 is lower than the resistance of the third magnetically sensitive resistor R₃506, the first n-type transistor M₁514 is relatively more conductive from drain to source than the second n-type transistor M₂516 is from drain to source because the lower resistance of the first magnetically sensitive resistor R₁502 results in a larger voltage from gate to source in the first n-type transistor M₁514 than in the second n-type transistor M₂516. When the resistance of the second magnetically sensitive resistor R₂504 is lower than the resistance of the fourth magnetically sensitive resistor R₄508, then the source of the second p-type transistor M₄520 is at a higher electrical potential than the source of the gate of the first p-type transistor M₃518. Since the gates of the second p-type transistor M₄520 and the gate of the first p-type transistor M₃518 are tied to the same potential when the reset transistor M₅510 is switched “on,” the second p-type transistor M₄520 has a relatively larger source to gate voltage than the gate of the first p-type transistor M₃518. The relatively larger source to gate voltage of the second p-type transistor M₄520 as compared to the source to gate voltage of the first p-type transistor M₃518 causes the second p-type transistor M₄520 to be relatively more conductive from source to drain than the first p-type transistor M₃518.

[0050] Following the foregoing process, a relatively low signal is provided from the input terminal 512 to the gate of the reset transistor M₅510 to switch the reset transistor M₅510 to an “off” state to thereby enable the latching action of the latching circuit 500. When the reset transistor M₅510 is switched “off,” the gates of the first n-type transistor M₁514 and the first p-type transistor M₃518 are decoupled from the gates of the second n-type transistor M₂516 and the second p-type transistor M₄520.

[0051] When the first n-type transistor M₁514 is relatively more conductive than the second n-type transistor M₂516, the voltage at a first output OUT₁522 is relatively low as compared with the voltage at a second output OUT₂524. The relatively low voltage at the first output OUT₁522 is cross-coupled to the gate of the second n-type transistor M₂516 and to the gate of the second p-type transistor M₄520. A relatively low voltage on the gate of the second n-type transistor M₂516 tends to turn the second n-type transistor M₂516 “off” (i.e., less conductive). A relatively low voltage on the gate of the second p-type transistor M₄520 tends to turn the second p-type transistor M₄520 “on” (i.e., more conductive). This causes the voltage at the second output OUT₂524 to rise.

[0052] A relatively high voltage on the second output OUT₂524 is cross-coupled to the first n-type transistor M₁514 and the first p-type transistor M₃518. A relatively high voltage on the gate of the first n-type transistor M₁514 turns the first n-type transistor M₁514 more “on” (i.e., more conductive). A relatively high voltage on the gate of the first p-type transistor M₃518 causes the first p-type transistor M₃518 to turn “off” (i.e., less conductive). This causes the voltage of the first output OUT₁522 to go lower. The process continues until the first output OUT₁522 is at a “low” state and the second output OUT₂524 is at a “high” state.

[0053] When the second p-type transistor M₄520 is relatively more conductive than the first p-type transistor M₃518, the voltage at the second output OUT₂524 is relatively high as compared with the voltage at the first output OUT₁522. The relatively high voltage at the second output OUT₂524 is cross-coupled to the gate of the first n-type transistor M₁514 and to the gate of the first p-type transistor M₃518. A relatively high voltage on the gate of the first n-type transistor M₁514 tends to turn the first n-type transistor M₁514 “on” (i.e., more conductive). A relatively high voltage on the gate of the first p-type transistor M₃518 tends to turn the first p-type transistor M₃518 “off” (i.e., less conductive). This causes the voltage at the first output OUT₁522 to decrease.

[0054] A relatively low voltage on the first output OUT₁522 is cross-coupled to the second n-type transistor M₂516 and to the second p-type transistor M₄520. A relatively low voltage on the gate of the second n-type transistor M₂516 turns the second n-type transistor M₂516 further “off” (i.e., less conductive). A relatively low voltage on the gate of the second p-type transistor M₄520 causes the second p-type resistor M₄520 to turn “on” (i.e., more conductive). This causes the voltage of the second output OUT₂524 to increase. The process continues until the first output OUT₁522 is at a “low” state, and the second output OUT₂524 is at a “high” state.

[0055] As illustrated above, when the combined resistance of the first magnetically sensitive resistor R₁502 and the second magnetically sensitive resistor R₂504 is less than the combined resistance of the third magnetically sensitive resistor R₃506 and the fourth magnetically sensitive resistor R₄508, the latching circuit 500 switches to a state where the first output OUT₁522 is at a “low” state, and the second output OUT₂524 is at a “high” state.

[0056] Advantageously, the latching circuit 500 is relatively less sensitive to variations in transistor parameters, such as channel length. The conventional latching circuit 400 described in connection with FIG. 4 is relatively sensitive to variations in the first p-type transistor M₃414 and the second p-type transistor M₄416. Equations 1 and 2 express a relationship of the sensitivity of the conventional latching circuit 400 to variations in channel length. $\begin{array}{cc}\frac{\Delta L}{L}=\frac{\Delta R}{R}& \left(\mathrm{Eq}.1\right)\\ \frac{\Delta L}{L}=\left(\frac{1}{2}\right)\left(\frac{\Delta R}{R}\right)& \left(\mathrm{Eq}.2\right)\end{array}$

[0057] Equation 1 applies when the first n-type transistor M₁410 and the second n-type transistor M₂412 are relatively large and the inverse of the transconductances, i.e., g_m⁻¹, of the first n-type transistor M₁410 and the second n-type transistor M₂412 are relatively small as compared to the amount of the resistance of the first magnetically sensitive resistor R_A402 or the second magnetically sensitive resistor R_B404. Equation 2 applies when the first n-type transistor M₁410 and the second n-type transistor M₂412 are relatively small, and when the inverse of the transconductance (1/g_m) of the first n-type transistor M₁410 and the second n-type transistor M₂412 are about the same as the resistance of the first magnetically sensitive resistor R_A402 or the resistance of the second magnetically sensitive resistor R_B404. Equations 1 and 2 illustrate that in the conventional latching circuit 400, the variability of the channel lengths of the first p-type transistor M₃414 and the second p-type transistor M₄416 should be relatively small to ensure proper detection or translation of stored data.

[0058] In contrast to the conventional latch of FIG. 4, the latching circuit 500 of FIG. 5 is relatively insensitive to variations in channel lengths of the first p-type transistor M₃518 and the second p-type transistor M₄520 because of the inclusion of the second magnetically sensitive resistor R₂504 and the fourth magnetically sensitive resistor R₄508 between the voltage source V_DDand the respective sources of the transistors. Mismatches in transistor channel lengths are relatively small as compared to the resistances of the second magnetically sensitive resistor R₂504 and the fourth magnetically sensitive resistor R₄508, thereby allowing the latching circuit 500 to detect relatively small differences in resistance and allowing the production of the latching circuit 500 with a relatively large amount of variability in transistor parameters, such as channel lengths.

[0059] FIG. 6 is a schematic of a bit line 600 of an array of MRAM cells according to an embodiment of the invention. The illustrated bit line 600 includes a first portion 602 and a second portion 604, and is shown in a GMR configuration. It will be understood by one of ordinary skill in the art that in a TMR configuration, a bit line can also be apportioned between a first portion and a second portion, but that the TMR cells couple to the bit line in parallel from corresponding word lines.

[0060] The first portion 602 and the second portion 604 each include a plurality of magnetically sensitive resistors that store data in magnetic states. The resistance of the first portion 602 and the resistance of the second portion 604 can vary over a broad range. The amount of resistance of the first portion 602 and the amount of resistance of the second portion 604 can be equal, but do not have to be equal. For example, one portion can have 60% of the resistance, and the other portion can have 40% of the resistance. In addition, the proportion of allocated resistance in portions of bit lines can vary within an array, although preferably not within pairs of bit lines coupled to an interface circuit. In one embodiment, the resistance of the first portion 602 and the resistance of the second portion 604 are approximately equal.

[0061] In one embodiment, the first portion 602 and the second portion 604 do not electrically connect within the array of GMR cells. The resistance of the magnetically sensitive resistors vary in accordance with the stored magnetic state and can be translated to logical states with an interface circuit, such as the latching circuit 500 described in connection with FIG. 5.

[0062] A first terminal 606 of the bit line 600 is provided to connect, either directly or indirectly, the bit line 600 to the voltage source V_DD. A second terminal 608 of the bit line 600 is provided to interface with a corresponding interface circuit, such as the latching circuit 500. A third terminal 610 of the bit line 600 is also provided to interface with the interface circuit. A fourth terminal 612 of the bit line 600 is provided to connect, either directly or indirectly, the bit line 600 to a reference voltage (e.g., ground). The illustrated configuration of the bit line 600 advantageously allows an interface circuit to be produced with a relatively large tolerance for transistor variations, such as channel lengths.

[0063] Various embodiments of the invention have been described above. Although this invention has been described with reference to these specific embodiments, the descriptions are intended to be illustrative of the invention and are not intended to be limiting. Various modifications and applications may occur to those skilled in the art without departing from the true spirit and scope of the invention as defined in the appended claims.