1. Field of Invention
The present invention is related to magnetic random access memory and in particular to reference averaging to enhance the read out margin.
2. Description of Related Art
Information is stored in a magnetic random access memory (MRAM) in the form of parallel and anti-parallel magnetic states of a magnetic tunnel junction (MTJ). The resistance of the parallel state is lower than that of the anti-parallel state. Data is stored in an MTJ by whether the state of the magnetic material of the MTJ is parallel or anti-parallel. Read out of the data in a sense amplifier is current flow from the MTJ compared to a reference that is positioned midway between current flow from an MTJ that has a magnetic parallel state and one that is anti-parallel.
To reduce readout variation and to produce a better fit with the threshold for deciding the state of an MTJ, a constant reference voltage or a constant reference current have been tried, but constant voltage and constant current references usually can not track temperature and process variations of the MTJ cells. These constant references cannot track the parasitic resistance and capacitance in the signal paths. Therefore, constant voltage and current references are not suitable for high speed operation. Self reference scheme have been considered, but it usually requires a cell to be read twice at different time intervals, which causes a longer time interval, charge loss, and possible influence of disturbance.
U.S. Pat. No. 7,239,537 B2 (DeBrosse et al.) is directed to a calibrated magnetic random access memory current sense amplifier, wherein a plurality of trim transistors that are individually activated to compensate for device mismatch relative to the data and sense sides of a sense amplifier. In U.S. Pat. No. 7,167,389 B2 (Iwata.) a magnetic random access memory is directed to a use of reference cells to help facilitate read and write operations. U.S. Pat. No. 7,038,959 B2 (Garni) is directed to a sense amplifier and a method for sensing an MRAM cell, comprising data cells, reference cells and dummy cells. U.S. Pat. No. 6,822,895 B2 (Yamada) is directed to a reference bit line connected in common for a plurality of bit lines, wherein a sense amplifier connected to a bit line and the reference bit line. U.S. Pat. No. 6,700,814 B1 (Nahas et al.) is directed to the use of a mock MRAM array and a mock sense amplifier, wherein a control circuit maintains current through cells of the mock MRAM at a value proportional to a reference current through variations in average bit cell resistance.
An article by Thomas W. Andre, Joseph J Nahas, Chita K. Subramanian, Bradley J Garni, Halbert S. Lin, Asim Omar and William L. Martino, Jr. titled “A 4 Mb 0.18 um 1T1MTJ Toggle MRAM With Balanced Three Input Sensing Scheme And Locally Mirrored Unidirectional Write Drivers”, IEEEE Journal of Solid State Circuits, vol. 40, pp. 301-309, January 2005, is directed a five level CMOS technology with a single toggling magneto tunnel junction to achieve a small chip size using unidirectional programming currents controlled by mirrored write drivers and a balanced three input current mirror sense amplifier for a read operation. An article by Gitae Jeong, Wootoung Cho, Sujin Ahn, Hongsik Jeong, Gwanhyeob Koh, Youngnam Hwang and Kinam Kim, titled “A 0.24 um 2.0 V 1t1MTJ 16 Kb Nonvolatile Magnetoresistance RAM With Self Reference Sensing Scheme”, IEEE Journal of Solid State Circuits, vol. 38, pp. 1906-1910, November 2003, is directed to a one transistor and one magnetic-tunnel-junction magneto resistance RAM using a self reference sensing scheme for reliable sensing margin.
It is an objective of the present invention to generate a reference current for an MRAM sense amplifier by averaging a plurality of at least two sense amplifier reference cells.
It is further an objective of the present invention to provide a differential sense amplifier for use with a single MTJ MRAM.
It is still further an objective of the present invention that the differential sense amplifier comprises an improved reference averaging scheme obtained by averaging the reference currents.
It is also still further an objective of the present invention to produce reference averaging over one word of n bits.
In the present invention a sense amplifier for a magnetic random access memory (MRAM) comprising a plurality of magnetic tunnel junction (MTJ) cells is described. The MTJ cells have two states, a parallel magnetic state and an anti-parallel magnetic state. Information is stored in the form of the two magnetic states. The resistivity of the MTJ cell when in the parallel state is lower than the resistivity in the anti-parallel state. Thus by reading current through the MTJ cell, a sense amplifier can determine which state and therefore what information is stored in the MTJ cell.
To determine of which magnetic state the MTJ cell is in, a reference current must be applied to the sense amplifier. If the reference current is a fixed bias applied to the sense amplifier, the effects of process variation and temperature on the circuitry will not be taken into consideration as well as circuit parasitic effects. These variation effects can be partially overcome by using two reference cells, one reference cell having a parallel magnetic alignment and a relatively low resistance Rp, and the other reference cell having an anti-parallel magnetic alignment and a relatively higher resistance Rap. Ideally, a combination of the currents from the two references can be used to produce a reference current that is midway between the current for the parallel aligned reference cell and the current for the anti-parallel reference cell. However, localized process and temperature variations can skew the combination of the two currents in favor of one state or another. To overcome the skewing of the reference current caused by effects of localized process and temperature variation, the current for the parallel aligned and anti-parallel aligned reference cells is done over two sense amplifier data words for 2n MTJ cells. This broader averaging has a tendency to smooth out local variations and produce a reference current that is closer to the midpoint between the parallel aligned reference cell and the anti-parallel aligned reference cell.
The averaged reference current can be further fine tuned by adjusting the numbers of parallel aligned and anti-parallel aligned reference cells. This is useful because the sweet spot for reference current depends on the distribution of parallel and anti-parallel current. The best reference level may not be in the midpoint of the parallel aligned cell and the anti-parallel aligned cell. It is also useful for product trimming and MTJ screening to identify weak bit.
In a fully symmetrical differential sense amplifier design, a common mode bias is obtained by averaging the signal and the reference current. As with the non-differential sense amplifier the averaging of reference currents over one or more words of n bits produces a reference current that is close to the ideal midpoint between the current for a parallel aligned cell and an anti-parallel aligned cell.
This invention will be described with reference to the accompanying drawings, wherein:
An MTJ memory cell has a resistance in the parallel state (Rp) that is lower than the resistance in the anti-parallel state (Rap). A common gate current amplifier is usually used as shown in
The sense amplifier margin can be defined as a figure of merit (FM), the signal to variation ratio, defined as:
FM=(Is−Ir1)/(s(Is+s(Ir1)2)1/2 EQ. 1:
where s(Is) and s(Ir1) are the standard deviations of the signal current Is and reference current Ir1, respectively, assuming Gaussian distribution.
Previously, the reference current in a single MTJ of an MRAM is obtained by averaging current from Rp and Rap so that it will be a mid value. However, the resulting average current is usually not exactly at the midpoint spot of the sensing threshold. Also the reference current still has variation of about 1/(2)1/2 of the signal variation.
A similar sense amplifier design is shown in
In
A first method shown in
A second method shown in
A third method of averaging variations in reference currents is to connect the output node “on” in all sense amplifiers together making node “on” the same for all n sense amplifiers and including variations in the PMOS mirror circuits and the NMOS clamp transistors in the averaging.
A combination of the three methods shown in
FM=(Is−Ir1)/s(Is) EQ 2:
assuming that s(Iref)=(½)1/2s(Is), the improvement in signal to variation margin becomes, FM=(1.5)1/2.
Shown in
In
The first method is to connect B nodes of each sense amplifier together, which will average together variations in the reference MTJ cells, parasitic resistance in the reference paths, switches, clamp transistor M5 identified in
The second method is to connect only node A together in each sense amplifier which does not average variations of clamp transistor M5 and mirror transistor M7 identified in
The number of reference cells to be averaged can be ranged from two to any large number, for example the length of a word, which can be 38 bits, but other word length such as 16 or 64 bits, plus parity bits, are also practical. The average over a word length is particularly attractive for block read, where all bits in a word are closely placed in one physical location. Sometimes, the bits in a word are scattered among the entire chip. In this case, the two or four reference averaging scheme for the differential sense amplifier shown in
While the invention has been particularly shown and described with reference to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made without departing from the spirit and scope of the invention.
Number | Name | Date | Kind |
---|---|---|---|
6700814 | Nahas et al. | Mar 2004 | B1 |
6822895 | Yamada | Nov 2004 | B2 |
7038959 | Gami | May 2006 | B2 |
7113437 | Schweickert et al. | Sep 2006 | B2 |
7167389 | Iwata | Jan 2007 | B2 |
7239537 | DeBrosse et al. | Jul 2007 | B2 |
7596045 | DeBrosse et al. | Sep 2009 | B2 |
20030103395 | Ooishi | Jun 2003 | A1 |
20040090841 | Pemer et al. | May 2004 | A1 |
20040120200 | Gogl et al. | Jun 2004 | A1 |
20050083747 | Tang et al. | Apr 2005 | A1 |
20070242549 | Klostermann et al. | Oct 2007 | A1 |
20090010045 | Ueda | Jan 2009 | A1 |
20090086534 | DeBrosse et al. | Apr 2009 | A1 |
20100061144 | Davierwalla et al. | Mar 2010 | A1 |
20100302838 | Wang et al. | Dec 2010 | A1 |
20110176350 | Jung et al. | Jul 2011 | A1 |
20110188305 | Yang | Aug 2011 | A1 |
20110194361 | Kawahara et al. | Aug 2011 | A1 |
Entry |
---|
“A 4-Mb 0.18-um 1T1MTJ Toggle MRAM With Balanced Three Input Sensing Scheme and Locally Mirrored Unidirectional Write Drivers,” by Thomas W. Andre et al., IEEE Journal of Solid-State Circuits, vol. 40, No. 1, Jan. 2005, pp. 301-309. |
“A 0.24-um 2.0-V 1T1MTJ 16-kb Nonvolatile Magnetoresistance RAM With Self-Reference Sensing Scheme,” by Gitae Jeong et al., IEEE Journal of Solid-State Circuits, vol. 38, No. 11, Nov. 2003, pp. 1906-1910. |
PCT Search Report PCT/US2012/070319 Mail date—Mar. 1, 2013. |
Number | Date | Country | |
---|---|---|---|
20130176773 A1 | Jul 2013 | US |