This application is based upon and claims the benefit of priority from Japanese Patent Applications No. 2012-198269, filed Sep. 10, 2012; and No. 2013-177298, filed Aug. 28, 2013, the entire contents of all of which are incorporated herein by reference.
Embodiments described herein relate generally to a magnetoresistive memory device, such as a magnetoresistive random access memory (MRAM), from which data is read by using a differential current.
In any system on a chip (SoC), a circuit such as cache operates not at all times. If the leakage path is cut while the circuit is not operating, the energy consumption can be reduced because no current leaks.
The static RAM (SRAM) used in the conventional SoC consumes more energy at leakage than in active mode. If the power supply is disconnected to cut the leakage, however, the data stored will be erased. In order to reduce the leak energy, a voltage that would not erase the data may be applied (in retention mode) in some cases. However, the electrostatic capacitance of the SRAM power supply is so large that the leak current may not be reduced even if the power supply is disconnected frequently.
In order to reduce the leak current, it is proposed that an MRAM should be used as nonvolatile memory. The MRAM can, however, hardly operate at high speed if the current flowing to each memory cell is small. This is because data is written in the memory cell when a current flows in the memory cell, and is read from the memory cell by detecting the current flowing in the memory cell. It is therefore desired to provide a magnetoresistive memory in which sufficient current flows in each memory cell and which can therefore operate at high speed.
In general, according to one embodiment, a magnetoresistive memory device includes first and second bit lines, a memory cell, third and fourth transistors, and fifth and sixth transistors. The memory cell comprises a first magnetoresistive element and a first transistor serially connected between the first bit line and a first power supply line, and a second magnetoresistive element and a second transistor serially connected between the second bit line and the first power supply line, and gate electrodes of the first and second transistors are connected to a word line. The third and fourth transistors have current paths inserted in the first and second bit lines, and gate electrodes of the third and fourth transistors are connected, respectively, to the second and first bit lines provided on a side opposite to the memory cell. The fifth and sixth transistors are connected to the first and second bit lines of the third and fourth transistors on the side opposite to the memory cell, and are configured to be turned on when data is read from the memory cell.
In the MRAM, a current equal to or larger than a specific value is supplied to the memory cell 11, thereby writing data in the memory cell 11. To read the data, a current of such a value as would not destroy the data in the memory cell 11 must flow to the memory cell 11. In order to limit this current, a clamp transistor 14 is provided between, and connected to, the selected memory cell 11 and the sense amplifier 13. A clamp transistor 15 is provided between, and connected to, the reference cell 12 and the sense amplifier 13. Since the clamp transistor 14 is provided between the selected memory cell 11 and the sense amplifier 13, the data can be read from the memory cell 11, while prevented from being destroyed.
However, the use of the clamp transistor 14 reduces the current read from the memory cell 11.
As shown in
Nonetheless, a current loss occurs because the clamp transistor 14 is used, which has such current-voltage characteristic as shown in
Therefore, this embodiment uses magnetic tunnel junction (MTJ) elements as memory cells to perform differential reading, thereby to achieve a high-speed operation.
Moreover, in this embodiment, a part charged at the time of activation and a part discharged are separated. Therefore, a faster transition from a state in which the leakage current has been suppressed to a state in which the memory cells can work, can be obtained and power consumption can be reduced.
Various embodiments will be described with reference to the accompanying drawings. In the drawings, the components of each embodiment, which are identical to those of any other embodiment, are designated by the same reference numbers and symbols.
The memory cell array 21 is composed of a plurality of sub-arrays 21a. In each sub-array 21a, a plurality of local bit lines BL and a plurality of word lines WL are arranged. The local bit lines BL extend at right angles to the word lines WL. As will be described later, one memory cell is arranged between a pair of local bit lines and is connected to one word line WL. The word line WL may be selected by a row decoder 22.
In parallel to the local bit line BL, a global bit line GBL is arranged. The global bit line GBL extends over a plurality of sub-arrays 21a. The global bit line GBL and the local bit line BL are selectively connected by a transistor, as will be described later. The global bit line GBL is connected to a global read/write circuit 23. The global read/write circuit 23 includes at least a write circuit (not shown) and a sense amplifier SA. The write circuit supplies write data to the global bit line GBL to write data. The sense amplifier detects the current read to the global bit line GBL. As will be described later, each local bit line BL is composed of a pair of bit lines, and each global bit line GBL is composed of a pair of read bit lines and one write bit line.
First bit line BLt and second bit line BLc correspond to the local bit line BL shown in
The memory cell MC is composed of an MTJ 1, an MTJ 2, an N-channel MOS transistor T1, and an N-channel MOS transistor T2. MTJ 1 and MTJ 2 are each composed of a storage layer, a reference layer and an insulating film interposed between the storage layer and the reference layer (not shown). In
One end (i.e., reference layer) of MTJ 1 is connected to the first bit line BLt, and the other end (i.e., storage layer) of MTJ 1 is connected to one end of the transistor T1. The other end of the transistor T1 is connected to a power supply line GND. One end (i.e., reference layer) of MTJ 2 is connected to the second bit line BLc, and the other end (i.e., storage layer) of the MTJ 2 is connected to one end of the transistor T2. The other end of the transistor T2 is connected, together with the other end of the transistor T1, to the power supply line GND. The gate electrodes of the transistors T1 and T2 are connected to the word line WL. The power supply line is connected to, for exampled, the ground an N-channel MOS transistor (not shown). Nonetheless, the power supply line need not be set to the ground potential. It may be set to any other potential instead.
The first bit line BLt, which is used as a local bit line, is connected to a read global bit line RGBLt via an N-channel MOS transistors T13 and T11 that are connected in series to each other. The second bit line BLc, which is used as a local bit line, is connected to the read global bit line RGBLc via N-channel MOS transistors T14 and T12 that are connected in series to each other.
The gates of the transistors T11 and T12 receive a signal read<i>. The signal read<i> is activated by a column address to read data. In response to the signal read<i>, the transistors T11 and T12 connect the first bit line BLt and second bit line BLc, which are use as local bit lines, to the read global bit lines RGBLt and RGBLc, respectively.
The gate electrode of the transistor T13 is connected to the connection node of the transistors T14 and T12. The gate electrode of the transistor T14 is connected to the connection node of the transistors T13 and T11. That is, the transistors T13 and the transistor T14 are cross-coupled to each other.
An N-channel MOS transistor T15 is connected between the ground and the first bit line BLt which extends between the transistor T13 and the memory cell MC. Further, an N-channel MOS transistor T16 is connected between the ground and the second bit line BLc which extends between the transistor T14 and the memory cell MC. The gate electrode of the transistor T15 is connected to a write global bit line WGBLc. The gate of the transistor T16 is connected to a write global bit line WGBLt.
The first bit line BLt is connected to the write global bit line WGBLt via a P-channel MOS transistor T17. The second bit line BLc is connected to the write global bit line WGBLc via a P-channel MOS transistor T18. The gates of these transistors T17 and T18 receive a signal/Write<i>. This signal/Write<i> is activated by a column address to write data.
First, the signal/write<i> and the word line W1 are set to low level and high level, respectively, and the signal read<i> is set to low level, in order to write data in the configuration described above.
When the signal/write<i> is set to low level, the transistors T17 and T18 are turned on. Then, the first bit line BLt and the second bit line BLc are set to high level and low level, respectively, if the data on the write global bit line WGBLt is “1” (high level) and the data on the write global bit line WGBLc is “0” (low level). As the word line WL is set to high level, the transistors T1 and T2 are turned on. As a result, current flows from the first bit line BLt to the second bit line BLc via the memory cell MC.
The data on the write global bit line WGBLt and the data on the write global bit line WGBLc turn the transistor T16 on, and the transistor T15 off. The transistor T16, now turned on, sets the second bit line BLc to the ground potential, though the second bit line BLc is not set to the ground potential because of its threshold voltage since both the write global bit line WGBLc and the second bit line BLc are connected by a P-channel MOS transistor. Thus, data “1” is set to MTJ 1, and data “0” to MTJ 2.
Assume that the data on the write global bit line WGBLt is “0” (low level), and that the data on the write global bit line WGBLc is “1” (high level). Then, the first bit line BLt is set to low level, and the second bit line BLc is set to high level.
The data on the global bit line WGBLt and the data on the global bit line WGBLc turn the transistor T15 on, and the transistor T16 off. The transistor T15, now on, sets the first bit line BLt to the ground potential. As a result, data “1” is set to MTJ 2, and data “0” is set to MTJ 1.
First, the signal read<i> is set to high level, and the word line WL is set to high level. Further, the signal/write<i> is set to high level.
Both the read global bit line RGBLt and the global bit line RGBLc are pre-charged to high level.
When the signal read<i> is set to high level, the transistors T11 and T12 are turned on, and the transistors T13 and T14, which are cross-coupled to each other, are also turned on. An electrical charge is therefore transferred from the read global bit lines RGBLt and RGBLc to the first and second bit lines BLt and BLc, respectively.
Since the word line WL is at high level, the transistors T1 and T2 of the memory cell MC are turned on. Assume that MTJ 1 and MTJ 2 data “0” and data “1” are held in MTJ 1 and MTJ 2, respectively. Then, the resistance of MJT 2 is almost twice the resistance of MTJ 1. Hence, the current flowing between MTJ 1 having low resistance and the power supply line GND becomes larger than the current flowing between MTJ 2 and the ground. That is, the current flowing via MTJ 1 to the first bit line BLt is twice as much as the current flowing via MTJ 2 to the second bit line BLc.
A current, which is a difference between the currents flowing in the first and second bit lines BLt and BLc, flows to the read global bit lines RGBLt and RGBLc connected to the second bit lines BLt and BLc via the transistors T11 and T12, respectively. The sense amplifier SA, which is connected to the read global bit lines RGBLt and RGBLc, detects the difference between the currents flowing, respectively, in the read global bit lines RGBLt and RGBLc.
To read data, the global bit lines WGBLt and WGBLc are equalized to voltage V. If the voltage of the signal read<i> is set to V in this state, the transistors T11 to T14 operate in saturated region. Assume that the connection node N1 of the transistors T11 and T13 is at voltage V1, and that the connection node N2 of the transistors T12 and T14 is at voltage V2. Then, the gate-source voltage Vgs of the transistor T11 is V−V1 (Vgs=V−V1), and the gate-source voltage Vgs of the transistor T12 is V−V2 (Vgs=V−V2). Further, the source voltage Vs of the transistor T13 is V−V2−V1, and the source voltage Vs of the transistor T14 is also V−V2−V1. The potential difference between the first and second bit lines BLt and BLc can therefore be zero at the time of reading data. Hence, the difference between the currents flowing, respectively, in the first and second bit lines BLt and BLc can be reliably transferred to the global bit lines WGBLt and WGBLc.
Moreover, MTJ 1 and MTJ 2 are prevented from assuming a data-written state because the potential difference between the first and second bit lines BLt and BLc is zero. This can prevent the data in MTJ 1 and MTJ 2 from being destroyed.
Once the transistor T14 has been cut off, the potential at the connection node N2 remains at high level. The transistor T13 is therefore maintained in on-state, and current flows in the first bit line BLt. The ratio of the current flowing in the first bit line BLt to the current flowing in the second bit line BLc (or difference between these currents) can therefore be increased, and the sense amplifier can reliably detect the data stored in the memory cell.
Further, since the transistors T13 and T14 are controlled by, respectively, the potential at the connection node N2 and the potential at the connection node N1, the ratio of the current flowing in the first bit line BLt to the current flowing in the second bit line BLc can be determined at high speed. That is, the sense amplifier can perform its function at high speed.
According to the first embodiment, the transistors T13 and T14 are cross-coupled to each other, the transistor T13 is controlled by the potential at the connection node N2, and the transistor T14 is controlled by the potential at the connection node N1. Therefore, the time for cutting off the transistor connected to the small-current side can be shortened. As a result, the ratio of the currents between the first bit line BLt and the second bit line BLc can be determined at high speed. This enables the sense amplifier to operate a higher speed.
In addition, the transistors T13 and T14, which are cross-coupled to each other, can reduce the potential difference between the first and second bit lines BLt and BLc to zero. The current difference between the first and second bit lines BLt and BLc, which achieve differential reading, can therefore be correctly transferred to the read global bit lines RGBLt and RGBLc.
The first and second bit lines BLt and BLc and the read global bit lines RGBLt and RGBLc are selectively connected by the transistors T11 and T12. The transistors T11 and T12 can therefore isolate the first and second bit lines BLt and BLc from the read global bit lines RGBLt and RGBLc, respectively, ultimately raising the state transition speed.
Since an MRAM is used as memory cell, not using any clamp circuit, thereby decreasing the current loss. The power consumption can, therefore, be reduced.
The local bit lines are connected to read global bit lit lines, in one-to-one correspondence. The connection is not limited to this. A plurality of local bit lines may be connected to one read global bit line. In this case, the number of sense amplifiers required can be decreased.
As described above, the currents flowing in the first and second bit lines BLt and BLc must be limited during the differential reading, in order to prevent the destruction of the data stored in the memory cell MC.
The transistors may have greater current-driven ability than designed because of the manufacture process conditions or the temperature ambient to the transistors may be relatively high.
In either case, the bias voltages applied to the transistors T11 and T12 can be controlled, thereby to control the currents flowing in the first and second bit lines BLt and BLc.
In view of this, the gate biases of the transistors T11 and T12 can be controlled in the second embodiment.
As shown in
The memory cell MCL is a replica of the memory cell MC. The N-channel MOS transistors T22, T23, T25 and T26 are replicas of the transistors T11 to T14.
The gate electrodes of the P-channel transistors T21 and T24 are connected to each other, and is connected to the gate electrode of the transistor 22 and the connection node of the transistors T21 and T22. The gate electrode of the transistor T23 is connected to the connection node of the transistor T22 and T23.
The gate electrode of the transistor T25 (i.e., the output terminal of the current mirror circuit) is connected to the connection node of the transistors T24 and T25, and is connected via the switch SW to the gate electrodes of the transistors T11 and T12. The gate electrode of the transistor T26 is connected to the connection node of the transistors T25 and T26.
In the configuration described above, the switch SW is turned on when data is read and the bias circuit 21 is connected to the gate electrodes of the transistors T11 and T12. In this state, the current flowing in the constant current source 22 is mirrored to the memory cell MCL. The voltage based on the current supplied from the constant current source 22 is therefore applied to the gate electrodes of the transistors T11 and T12. As a result, the gate voltages of the transistors T11 and T12 are thereby held at a constant value.
According to the second embodiment, the gate voltages of the transistors T11 and T12 are controlled by the bias circuit 21 comprises a replica circuit composed of transistors T11 to T14, a replica circuit of the memory cell MC, and a current mirror circuit including the constant current source 22. Hence, even if the transistors have greater current-driven ability than designed because of the manufacture process conditions or the temperature ambient to the transistors is relatively high, an excessively large current is prevented from flowing to the first and second bit lines BLt and BLc, merely by controlling the gate voltages of the transistors T11 and T12. This prevents the data stored in the memory cell MC from being destroyed as it is read from the memory cell MC.
It has been explained how the second embodiment operates if the transistors have greater current-driven ability than designed. The second embodiment is designed to operate well if the transistors have smaller current-driven ability than designed.
The clock signal CLK is also supplied via an inverter circuit 35 to the gate electrodes of P-channel MOS transistor T31 and N-channel MOS transistor T32, which constitute an inverter circuit 36. The transistors T31 and T32 are connected in series between the power supply and the ground. The connection node of these transistors T31 and T32 is connected to the gate electrodes of the transistors T11 and T12 mentioned above.
Further, a P-channel MOS transistor T33 is provided between, and connected to, the power supply and the connection node of the transistors T31 and T32. This transistor T33 receives, at gate electrode, the output signal of the NOR circuit 34.
How the bias circuit 31 of
As
In the modification described above, the bias circuit 31 boosts the gate biases of the transistors T11 and T12. The current-driven ability of the transistors T11 and T12, which may be lower than designed, can therefore be enhanced. As a result, the currents flowing in the first and second bit lines BLt and BLc can be reliably transferred to the read global bit lines RGBLt and RGBLc.
The tolerance voltage the cell has at data writing will be explained with referenced to
As specified above, two transistors T11 and T2 and two MTJ 1 and MTJ2 are provided in the write path to write data in the memory cell MC provided between, and connected to, the first bit line BLt and the second bit line BLc. To provide a sufficient write current, the low-level potentials of the first and second bit lines BLt and BLc set to negative voltages. The transistors T1 and T2 can be reliable if MTJ 1 and MTJ 2 undergo a voltage drop (IR drop). The bias the transistors T1 and T2 need to acquire reliability is given as follows:
Vwl−IwR+ΔV<Vtol (1)
where Vwl is the level to which the word line WL is set when the word line WL is selected, Iw is the current flowing in the memory cell MC, R is the resistance of MTJ 2, ΔV is the absolute value of the negative voltage, and Vtol is the tolerance voltage of the transistors.
In the cell selected, which is shown in
IwR−ΔV (>0) (2)
This positive value satisfies the relation of the formula (1), ensuring the reliability of the cell selected.
IwR−ΔV+Vds (>0) (3)
This satisfies the relation of the formula (1), ensuring the reliability of the cell selected.
Thus, in the memory cell array used in the first and second embodiments, each cell has a sufficient tolerance voltage at the data writing.
In the pattern of
On one part of the source-drain region common to the transistors T1 and T2, MTJ 1 and MTJ 2 are provided. MTJ 1 and MTJ 2 are connected to bit lines BLt and BLc, respectively. MTJ 1 and MTJ 2 are arranged on one side of a word line WL. The source-drain region common to the transistors T1 and T2 is connected, at the other side, to a power supply line GND. This layout pattern has two power supply lines GND, each having a contact CP. The contacts PC are arranged on the other side of the word line WL.
In the pattern of
On one part of the source-drain region common to the transistors T1 and T2, MTJ 1 and MTJ 2 are provided. MTJ 1 and MTJ 2 are connected to bit lines BLt and BLc, respectively, and are arranged on the sides of the word line WL, respectively. The other part of the source-drain region of the transistor T1 and T2 is arranged between the first and second parts WL1 and WL2 of the word line WL, and is shared by the transistors T1 and T2. On the other part of the source-drain region, the contact CP of the power supply line GND is arranged. This pattern has one power supply line GND.
In the pattern of
On one part of the source-drain region common to the transistors T1 and T2, MTJ 1 and MTJ 2 are provided. MTJ 1 and MTJ 2 are connected to bit lines BLt and BLc, respectively, and are arranged between the first and second parts WL1 and WL2 of the word line WL. On the other part of the source-drain region of the transistors T1 and T2, the contact CP of the power supply line GND is arranged. This pattern has two power supply lines GND.
The pattern layout of the memory cell array is not limited to those shown in
In a high-speed memory capable of performing differential reading, a memory cell MC is provided between, and connection to, first and second bit lines BLt and BLc, as described above. Since two MTJs are used in it, the memory device may consume more power than otherwise. If serial writing is performed on two MTJs, a sufficiently high voltage can hardly be applied across each MTJ in some cases, inevitably because of the IR drop in the MTJ. Consequently, it may become difficult to acquire a write current of Iw or more. The transistors T1 and T2 constituting the memory cell should better be thin-film transistors, for the purpose of reducing the sell size. Hence, the voltage VDD on the word line WL cannot be raised.
In view of this, a write transistor is connected to the memory cell MC in the, third embodiment, thereby mitigating the influence of the IR drop at the time of writing data.
More specifically, a write transistor T3 is connected in parallel to the transistors T1 and T2 as shown in
The use of the write transistor T3 can reduce the IR drop at the time of writing data. Reliability can therefore be ensured in writing data in any selected cell.
In a cell of a selected column, i.e., one of the cells constituting a non-selected row, a potential higher than the potential at the transistors T1 and T2 is applied, as shown in
Also in a cell of a non-selected column shown in
According to the third embodiment, the write transistor T3 is connected in parallel to the transistors T1 and T2 that constitute the memory cell MC. To write data, not only the transistors T1 and T2, but also the transistor T3 is turned on. Therefore, the IR drop is reduced at the time of writing data, enhancing the reliability of data writing.
Since the write transistor T3 is used, the memory cell array has more transistors than in the first and second embodiments. However, the third embodiment can has a far larger write margin than the first and second embodiments.
To write data, only the transistor T3 may be selected, not selecting the transistor T1 or the transistor T2. In this case, the energy for selecting the word line WL can be reduced, and the line delay can be more decreased than in the first and second embodiments. As a result, the memory device of the third embodiment can perform high-speed data writing.
As described above, when data of a memory cell MC is read in
For example, if a resistance of MTJ1 is higher than a resistance of MTJ2, the current Icell2 flowing to MTJ2 is greater than the current Icell1 flowing to MTJ1 as indicated by PE1 in
However, if the current flows from the reference layer to the storage layer as shown in
In view of this, in the fourth embodiment, the current to flow to the memory cell MC when the data is read flows in a direction opposite to the direction shown in
In
In addition, a local sense amplifier LSA is connected between the first bit line BLt and the second bit line BLc, and the global bit lines RBLt and RBLc. The local sense amplifier LSA is composed of P-channel MOS transistors T42, T43, T44, T45 and T46. A current path of the transistor T42 is inserted into and connected to the first bit line BLt. A current path of the transistor T43 is inserted into and connected to the second bit line BLc. Gate electrodes of the transistors T42 and T43 are connected, respectively, to the second bit line BLc and the first bit line BLt on a side opposite to the memory cell MC. That is, the transistors T42 and T43 are cross-coupled to each other.
The first bit line BLt is connected to a gate electrode of the transistor T44. The second bit line BLc is connected to a gate electrode of the transistor T45. One ends of current paths of the transistors T44 and T45 are connected to each other. The other end of the current path of the transistor T44 is connected to the global bit line RBLt. The other end of the current path of the transistor T45 is connected to the global bit line RBLc.
The transistor T46 is connected between a connection node of the transistors T44 and T45 and the supply node of the power supply VDD. A signal S46 is supplied to a gate electrode of the transistor T46. The signal S46 is set at a low level when the data is read from the memory cell MC, and is generally set at a high level.
The operation of reading the data from the memory cell MC in the above-described configuration will be hereinafter described.
When the data is read from the memory cell MC, the global bit lines RBLt and RBLc, the first bit line BLt, and the second bit line BLc are set at a low level, unlike the first embodiment.
After this, the signal S41 of the current source CS set at the high level is set at a constant voltage of 0V or higher, and the signal S46 of the local sense amplifier LSA set at the high level is set at the low level.
Simultaneously, when the word line WL is set at the high level, the transistors T1 and T2 are turned on, the current Icell1 flows from the current source CS to the first bit line BLt via MTJ1, and the current Icell2 flows to the second bit line BLc via the transistor T2 and MTJ2.
It is considered here that the resistance of MTJ1 is set to be double the resistance of MTJ2. The current Icell1 flowing via MTJ1 becomes a half of the current Icell2 flowing via MTJ2 as indicated by PE1 in
When the resistance of MTJ2 is set to be double the resistance of MTJ1, oppositely to the above case, the current Icell2 flowing via MTJ2 becomes a half of the current Icell1 flowing via MTJ1 as indicated by PE2 in
According to the fourth embodiment, the currents flow to MTJ1 and MTJ2 via the transistors T1 and T2, respectively, from the current source CS in the memory cell MC when the data is read. That is, since the current flows from the storage layer side to the reference layer side in each of MTJ1 and MTJ2, the data is rarely destroyed due to Disturb.
Since the data is rarely destroyed due to Disturb, the read current can be increased. For this reason, a current equal to or higher than the upper limit of the current shown in
Furthermore, since the transistors T42 and T43 are cross-coupled to each other, the time for cutting off the transistor connected to the small-current side can be shortened, similarly to the first embodiment. As a result, the ratio of the currents flowing to the first bit line BLt and the second bit line BLc can be determined at high speed. The sensing operation of the local sense amplifier LSA can be therefore accelerated.
Moreover, the signals of the first bit line BLt and the second bit line BLc can be amplified and supplied to the global bit lines RBLt and RBLc, respectively, by the local sense amplifier LSA. Variation in amplitudes of the global bit lines RBLt and RBLc having great wiring capacitance can be therefore suppressed.
While certain embodiments have been described, these embodiments have been presented by way of example only, and are not intended to limit the scope of the inventions. Indeed, the novel embodiments described herein may be embodied in a variety of other forms; furthermore, various omissions, substitutions and changes in the form of the embodiments described herein may be made without departing from the spirit of the inventions. The accompanying claims and their equivalents are intended to cover such forms or modifications as would fall within the scope and spirit of the inventions.
Number | Date | Country | Kind |
---|---|---|---|
2012-198269 | Sep 2012 | JP | national |
2013-177298 | Aug 2013 | JP | national |