Journal of Magnetism and Magnetic Materials 192 (1999) 334-342



                                                   GMR applications

                                                         J.M. Daughton*
                       Nonvolatile Electronics, Inc., 11409 Valley View Road, Eden Prairie, Minnesota, MN 55344, USA

                                     Received 19 August 1998; received in revised form 17 September 1998



Abstract

   Giant magneto-resistance (GMR) materials have magneto-transport properties which determine their suitability for
applications in magnetic field sensors, read heads, random access memories, and galvanic isolators. Each of these
applications for GMR materials is discussed, and desirable materials development are described.   1999 Elsevier
Science B.V. All rights reserved.

Keywords: GMR; Sensors; Read heads; Random access memory; Isolators



1. Introduction                                                          2. GMR material properties

   Ten years after the discovery of giant mag-                              Magnetoresistance (the percent change in resis-
netoresistance (GMR), commercialization of the                           tivity with respect to its lowest value under applica-
technology is evidenced by product introductions                         tion of a range of magnetic fields) and the magnetic
in magnetic field sensors and read heads for hard                        field required to achieve the full range of the mater-
drives. This comparatively short introduction time                       ial's resistance (saturation field) are two of the most
was facilitated by the prior existence of similar                        important material properties for GMR material
products using anisotropic magnetoresistance                             applications. The rate of change of resistance with
(AMR) materials. While the primary driving force                         magnetic field (sensitivity) determines the signal
behind the industrial introduction of GMR mater-                         generated by a magnetic field when the field is less
ials is their high magnetoresistance, other material                     than the saturation field.
properties of GMR materials are extremely impor-                            Table 1 compares some typical magnetoresistance,
tant to applications. This paper first examines some                     saturation field, and sensitivity values for a range of
of these material properties, and then discusses                         GMR materials. A magnetic sandwich [1] is a spin
some current and potential applications of GMR                           valve [2] without a pinning layer. Granular films [3]
materials. Some desirable future advances in GMR                         are thicker films made of immiscible magnetic and
materials will be discussed.                                             nonmagnetic conductors. Spin dependent tunnel-
                                                                         ing (SDT) structures [4] use a different conduction
                                                                         mechanism from those materials commonly called
  * Tel.: #1-612-996-1607; fax: #1-612-996-1600; e-mail:                 GMR materials, but for purposes of discussion they
daughton@nve.com.                                                        are treated as GMR materials in this paper.

0304-8853/99/$ - see front matter   1999 Elsevier Science B.V. All rights reserved.
PII: S 0 3 0 4 - 8 8 5 3 ( 9 8 ) 0 0 3 7 6 - X



                          J.M. Daughton / Journal of Magnetism and Magnetic Materials 192 (1999) 334-342                  335

Table 1
Comparison of typical properties of GMR materials

                   % Magneto- resistance             Saturation field (Oe)         Sensitivity (%/Oe)       Comments

AMR                2                                 5-20                          0.4
Multilayer         10-80                             100-2000                      0.1                      Hysteresis
Granular           8-40                              800-8000                      0.01                     Hysteresis
Spin valve         5-10                              5-50                          1.0                      Thermal?
Sandwich           5-8                               10-40                         0.5
CMR                100                               1000                          0.1                      High TCR
Tunneling          10-25                             5-25                          2.0                      High R


  Comparison between multilayers and granular                           There are several important thermal character-
structures shows an advantage in GMR and sensi-                      istics of magnetoresistive materials. TCR has al-
tivity for multilayers. It could be postulated that                  ready been mentioned. (GMR and SDT devices
the granular structure is easier to fabricate, but no                have satisfactory TCRs on the order of 1500 and
serious obstacles have been encountered in making                    !700 ppm/°C, respectively). Short term (a few
multilayers, especially at the so-called `second peak'               hours) thermal stability is important in fabrication
thickness for the nonmagnetic conductor in the                       of devices, where modern processes frequently ex-
multi-layer [5]. With computer-controlled sput-                      ceed 200°C. Where integrated circuits are com-
tering equipment designed for good thickness                         bined with the GMR materials, the temperatures
uniformity, multilayers are relatively easy to manu-                 can be even higher. Long term (thousands of hours)
facture, and hence there is no obvious need for the                  stability is essential to product reliability. Except
granular structure at this point in time.                            for automotive and certain industrial applications,
  Colossal magnetoresistance (CMR) can give very                     operating temperatures rarely exceed 125°C. The
large changes in resistance with applied magnetic                    magnetic sandwich is stable to operating temper-
fields [6], but would probably be difficult to apply                 atures above 200°C. Thermal properties of spin
because of operating temperature (generally well                     valves have not been as desirable, but have recently
below room temperature), and even more so be-                        improved greatly with the use of iridium manga-
cause of an extremely high temperature coefficient                   nese [7] and nickel manganese [8] as antiferro-
of resistivity (TCR). This latter property would                     magnetic pinning layers, and hence operating
make it difficult to compensate for temperature                      temperature is less of a concern for spin valves than
shifts so as to distinguish between temperature and                  it was a few years ago.
magnetic field in a practical application. With                         Magnetostriction has not been reported as
CMR properties as they now stand, the material                       a problem in GMR materials, but there is potential
would not find widespread use.                                       for problems, particularly in devices with small
  The other materials listed in Table 1 are finding                  features. Most reported alloys are nominally non-
use in practical devices. AMR materials (permalloy                   magnetostrictive in bulk, but magnetostriction in
thin films) have good sensitivity at fields below                    thinner layers may present problems with shifting
about 10 Oe, but will yield a smaller signal than                    magnetic properties. This area is worthy of more
spin valve, sandwich, or tunneling materials at                      research than has been reported to date.
fields higher than about 5 Oe. The sensitivity of                       The resistance of a device is also an important
multilayers is not as high as for the other three, but               parameter. In GMR devices the resistance is a func-
there are applications which use several hundred                     tion of the length and width of the device, with the
Oe field, and in these the other GMR structures                      sheet resistivity usually in the 2-20  /square range,
and the AMR structure would be saturated, mak-                       where its number of squares is its ratio of the length
ing the multilayer a better choice based on output                   to width. For very small devices which are con-
in higher field applications.                                        strained to be only a few squares (as in memory and



336                  J.M. Daughton / Journal of Magnetism and Magnetic Materials 192 (1999) 334-342

read heads), the resistance will be on the order of
10-20  . SDT devices, on the other hand, have
a resistance inversely proportional to device area.
Typical resistances vary considerably depending on
processing parameters, and values ranging from
10 to 10    m are common. Even resistances at
the lower limits of this range will result in a square
micron device with 10   resistance, and several
orders of magnitude higher are easily attainable.                    Fig. 1. Top view of two GMR bridge sensor chips.
  SDT devices are relatively new, but their intrinsi-
cally high sensitivity and high resistance are very
promising for some applications. Two potential
problems with spin tunneling are: (1) magneto-
resistance declines for voltages across the device
exceeding approximately 0.1 V, with an irreversible
breakdown occurs at about 2 V and (2) uncertainty
in the ruggedness of very thin barriers with respect
to elevated temperatures. In addition to these prob-
lems, their high resistance can result in long RC
time constants, particularly where high density re-
quirements dictate small areas.


3. Current and potential applications of GMR
materials

  There are several current or potential applica-
tions for devices based on GMR or SDT materials.
Announced products include magnetic field sensors
and read heads for hard drives. SDT materials
show promise for improved signal/noise ratios,
which could extend the use of GMR sensors to
much lower magnetic fields. Several companies (in-
cluding IBM) have announced GMR read head                      Fig. 2. GMR multilayer resistance (left) and bridge output
products for hard drives. There are several large              (right) as a function of applied field.
development programs in the US and Europe
aimed at using GMR or SDT materials for non-
volatile random access memory. A derivative of                 of 10-100 Oe using GMR materials with saturation
GMR sensors can make it possible to isolate digital            fields of 200-300 Oe. Fig. 1 is a picture of two
and analog signals in networks to preserve data                bridge sensor chips prior to chip packaging. Fig. 2
integrity and to prevent transient damage to com-              shows a GMR multilayer's response to magnetic
ponents elsewhere in the system. In this section, the          field along with the resulting output of a Wheat-
status of each application is discussed in turn along          stone bridge using a current source or voltage
with some future projections.                                  source to drive the bridge, with operating temper-
  Sensors - The first commercial GMR sensors,                  ature as a parameter. With a current source, the
which were introduced in 1995, use multilayer                  bridge output decreases with temperature at the
GMR materials [9]. The use of shields and flux                 rate of about 4% per 100°C, whereas the output
concentrators allow the sensors to operate at fields           declines at the rate of about 30% per 100°C with



                      J.M. Daughton / Journal of Magnetism and Magnetic Materials 192 (1999) 334-342              337





















                                  Fig. 3. Digital GMR sensor and output characteristics.


a voltage source. The difference is due to the back-            the 10-100 Oe range. Two areas for improving
ground resistance, which increases at the rate of               GMR materials for these applications are:
about 20% per 100°C, whereas the GMR values                     (1) Elimination of virtually all hysteresis over the
decline at only 4% per 100°C. Fig. 3 shows a digital            field operating range and (2) Self biasing GMR
magnetic field sensor which includes a bridge and               materials which give a bipolar (positive for positive
integrated electronics, also shows the output of that           fields and negative for negative fields) change in
chip with applied field. The hysteresis of the output           resistance under applied fields.
is mostly due to designed-in hysteresis in the on-                 For sensors which detect perturbations of the
chip electronics. A new GMR magnetic field                      earth's magnetic field, or the earth's field itself
sensor for angular position sensing was introduced              (compasses), sensitivity (signal/field) and noise are
in 1997 [10].                                                   the most critical parameters. Spin dependent tun-
  The two largest categories of applications for                neling materials have the potential to be very im-
magnetoresistive sensors are: (1) sensing the posi-             portant to low magnetic field sensor applications
tion or speed of a ferrous body by using an auxili-             (below 10\ Oe). Tunneling devices can have
ary permanent magnet, which magnetizes the body                 20-25% equivalent magnetoresistance [11,12], and
to be sensed and (2) sensing the position or speed              if properly biased, can give a linear output with
of a ferrous body using the earth's magnetic field to           respect to applied field with little hysteresis [13].
magnetize the body. A third large potential applica-            Although thorough noise measurements in SDT
tions area is current sensing, which will be dis-               materials has not been published, there are indica-
cussed later in the paper.                                      tions that noise in SDT materials is lower than for
  In the first category it is necessary that the field          equivalent GMR materials. The combination of
from the body be higher than the earth's magnetic               lower noise and higher sensitivity indicate a poten-
field so that the earth's magnetic field doesn't create         tial for SDT sensors to reach low field sensitivi-
a large error. A magnetic field on the order of 10 Oe           ties not yet achieved by existing magnetoresistive
should be sufficient. In fields higher than about               technologies.
100 Oe, a Hall sensor is generally adequate and                    Read Heads - New GMR read head products
inexpensive. Thus, practical applications for GMR               have been introduced by IBM and others. Al-
sensors of the type discussed here are for fields in            though little product description is available, it is



338                        J.M. Daughton / Journal of Magnetism and Magnetic Materials 192 (1999) 334-342

                                                                        One way to reduce demagnetizing effects in the
                                                                     horizontal stripe would be to make the magnetic
                                                                     films (and the nonmagnetic interlayer) thinner
                                                                     while retaining high GMR. This would have the
                                                                     benefit of increasing resistance and improving
                                                                     signal level (provided current densities can be
                                                                     higher).
                                                                        Another way would be to reduce the magneti-
                                                                     zation of the magnetic layers. This raises funda-
                                                                     mental issues about the relationship between
                                                                     magnetization and GMR. Intuitively, one would
                                                                     suspect that magnetic moment and GMR are lin-
       Fig. 4. Spin valve read head sensor (from Ref. [14]).         ked by some necessary relationship through spin-
                                                                     polarized electrons (conduction and bound), but
believed that spin valve material is etched into                     this has not been demonstrated.
a narrow stripe, and the magnetization of the pin-                      Still another useful development would be spin
ned layer is oriented across the stripe [14].                        valves with high field sensitivity, but with lower
Fig. 4 shows how the spin valve stripe is then                       exchange coupling. Once again, this raises a ques-
oriented with respect to the media in order to detect                tion about the fundamental relationship between
stored data.                                                         GMR and exchange constant. Can high GMR co-
  In the last several years, substantially improved                  exist with low ferromagnetic exchange? One
materials antiferromagnetic material (for example,                   might think that lower exchange would imply a
iridium manganese [7] and nickel manganese [8])                      low ordering temperature. This problem has been
has improved stand-off field and operating and                       overcome in recording media by doping grain
processing temperature tolerance of the spin valves                  boundaries so as to reduce exchange coupling be-
used. Further improvement through the addition of                    tween grains without affecting the material proper-
a thin ruthenium film and another ferromagnetic                      ties within the grains. Could something similar be
layer has demonstrated still better temperature and                  done with spin valve structures?
field characteristics [15].                                             Because the magnetic field from the media de-
  With shrinking geometries forced on the read                       creases very rapidly with vertical distance perpen-
head designer by industry demands for higher den-                    dicular to the stripe, it is important that the spin
sity, demagnetizing effects in very narrow horizon-                  valve be sensitive to fields very near the outside
tal sensor stripes will become a major challenge.                    surface of the head. The rapid decrease is due both
For example, a 100 As thick permalloy stripe                         to the nature of the field from the media, and to
0.25  m wide has a self demagnetizing field of ap-                   magnetic shields, lying on either side of the stripe,
proximately 200 Oe in the center of the stripe. At                   which prevent `reading' data from nearby bits.
the edges of the stripe, demagnetizing effects tend to               Shield materials are thick and have high permeabil-
`pin' the magnetization along the edge, thus making                  ity, thus creating a low reluctance path to shunt off
magnetization near the edges of the stripe very                      the signal flux.
insensitive to magnetic fields. Exchange coupling of                    One potential method of lowering the effect of
magnetization near the edge region tends to reduce                   demagnetizing fields is the use of vertical stripes as
sensitivity of materials short distances away from                   suggested by Pohm [16]. Demagnetizing problems
the edge, an effect which gets worse with shrinking                  are also reduced, but not eliminated. An optimum
geometries. These demagnetizing effects limit the                    GMR material for this configuration would be to
sensitivity to the fields supplied by the media, and                 maximize magnetic moment so as to provide as low
although clever techniques can ameliorate the de-                    a reluctance path as possible for the media signal
magnetizing effects, they become more difficult to                   flux to follow the sensor. If this solution were to be
implement as linewidths keep shrinking.                              implemented, finding techniques to stabilize the



                     J.M. Daughton / Journal of Magnetism and Magnetic Materials 192 (1999) 334-342                 339















                                           Fig. 5. Pseudo-spin valve concept.


edges of the vertical stripe through process or
design techniques would also be a challenge.
  MRAM - Spin valve [17], tunneling [18,19], and
pseudo-spin valve [20,21] structures have all been
proposed for high density nonvolatile random ac-
cess memory. The use of GMR materials to replace
anisotropic magnetoresistive (AMR) materials
shows promise to ameliorate one of the most diffi-
cult problems which has faced MRAM technology
- that of a small signal size, leading to relatively
long read access times for memory applications.
  Honeywell was the first to demonstrate an oper-
ating memory chip using GMR materials [22]
using sense linewidths of approximately 2  m. Sub-
micron GMR memory cells required an improved                              Fig. 6. Pseudo-spin valve read method.
mode of operation, and one of the more promising
proposals is the pseudo-spin valve. Fig. 5 shows the
cell concept where two magnetic films are separ-               decreases when a magnetic field is swept from
ated by a thin conducting layer. These layers are              a negative to a positive value as shown in Fig. 6
etched into stripes sufficiently narrow to constrain           (from 1 to 2). Most of the available range of resist-
the magnetizations in the stripes to lay along the             ance is observed, and a stored `1' and a stored `0'
long axis of the stripe. A conductor layer etched              have opposite signs.
into a strip line is place over the stripe to apply               Stable MRAM cells with sense linewidths of less
a magnetic field when a current is passed through              than 0.5  m [20,21] have been demonstrated, but
it. One of the magnetic films switches at a lower              not without careful end-shaping to avoid magnetic
magnetic field than the other. This is accomplished            anomalies. It is interesting to recall the commonly
either by the two films having different thicknesses           held view of only a few years ago, that if magnetic
or composition. Data is stored in the magnetization            thin films were etched into geometries less than
layer requiring the larger magnetic field for                  a wall width in diameter, then the magnetization
reversal. The softer film can switch back and forth            would be `single domain'. However, both experi-
without the storage film switching. The mag-                   mental data from several sources and calculations
netoresistive property is used to read out data                [17] by Zhu et al. indicate that not to be the case.
by observing whether the resistance increases or               Magnetic anomalies occur at a very small



340                  J.M. Daughton / Journal of Magnetism and Magnetic Materials 192 (1999) 334-342

(&0.2  m) length scale. Methods of preventing
these anomalies, or perhaps living with them, or
even using them for memory applications will be
interesting research topics for the next several
years.
  Some of the GMR materials improvements sug-
gested for read heads would not apply to MRAM,
because unlike read heads, data storage stability is
essential to MRAM. The energy well created by the
magnetic moment and the coherence of the mag-
netization (exchange) demand maintaining both
magnetization and exchange at relatively high                            Fig. 7. Isoloop concept for GMR isolator.
values.
  Galvanic Isolators - In the discussion of GMR
magnetic field sensors, an important application
category of magnetic sensors was deferred. Current
sensing is a primary application area for magnetic
sensors. Magnetoresistive sensors have an intrinsic
advantage over Hall sensors in that they are sensi-
tive to in-plane fields. This makes it possible to
integrate a coil on chip to generate a field from
a current. Hall effect sensors generally use a fer-
romagnetic or ferrite toriod with a flux gap to
generate a field to which they are sensitive. An                  Fig. 8. Comparison of opto-isolator and GMR isolator.
on-chip coil on Hall sensors could also be used, but
the field efficiency (field per unit current) would be
relatively low compared to the in-plane field pro-
duced by under planar windings, and the field
sensitivity of Hall sensors is much lower. As a con-
sequence, the GMR sensor with an integrated coil
requires at least an order of magnitude lower cur-
rent for the same output as a Hall sensor with an
integrated coil.
  An interesting extension of the GMR sensor
technology is the GMR isolator [23]. Fig. 7 illus-
trates the concept, where an on-chip coil driven by
an external current creates a magnetic field sensed
by an on-chip GMR sensor. The coil circuit and the
sensor circuit are separated by a few microns of                        Fig. 9. Photomicrograph of GMR isolator.
high dielectric strength. Thus, data can be passed
through this device by driving the coil and sensing
magnetic field, and the coil circuit and sensing               barrier, and the light activates a semiconductor
circuits don't have a direct connection path to each           device. In a GMR isolator, a current in a coil
other, i.e. they are isolated. Isolation voltages are          produces a magnetic field which extends across an
typically 1000-3000 Volts. Fig. 8 shows how this               insulating barrier, and the magnetic field is sensed
concept compares to the presently utilized opto-               by a GMR sensor.
isolators. In opto-isolators, a current input to an               The GMR isolator has been reduced to practice
LED causes light to be emitted across an insulating            in several forms. Fig. 9 shows a photomicrograph



                      J.M. Daughton / Journal of Magnetism and Magnetic Materials 192 (1999) 334-342                         341

Table 2
Some potential improvements to GMR materials

Application                       GMR materials                                  Desired developments

General                           All                                            Temperature stability
                                                                                 High Sensitivity
                                  Tunneling                                      Higher conductance
                                                                                 Higher roll-off voltage
Field Sensors,                    Multilayer                                     Low hysteresis
Isolators                         Sandwich, SV Tunneling                         Bipolar response
Read heads
Horizontal                        Spin valve                                     Thinner films
                                                                                 Lower moment
                                                                                 Lower exchange
Vertical                          Sandwich                                       Higher moment
                                                                                 Edge property control
MRAM                              Sandwich, SV Tunneling                         No magnetic anomalies



of an integrated isolator. A linear bipolar inte-               ately 400°C for an hour would enhance the integra-
grated circuit acts as the substrate for the GMR                bility of GMR materials with integrated circuits. In
sensors and integrated coil. This GMR isolator                  the case of SDT devices, lower resistances are re-
measures about 1 mm on a side and operates at 50                quired for low field sensors and memory. A specific
Mbaud. Much faster GMR isolators are in evalu-                  resistivity of 1000    m would be very desirable.
ation. The inherent advantages of this technology               Finally, if a very high on/off resistance ratio ('10)
are small size, integration with silicon circuits, and          can be attained under room temperature condi-
high speed potential ('1 GBaud). First prototype                tions through the application of a magnetic field,
products are planned for this year.                             GMR devices could have potential for functions
  The most desired materials for the isolator are               which now use transistors.
similar to those for magnetic field sensing. High                  Ten years after the discovery of GMR materials,
GMR with low saturation fields are important.                   products using GMR materials are being produced.
Because direction of current in an on-chip coil must            Wider applications of these materials would benefit
be sensed, it is also important that the material               from continued advances in understanding and de-
have a bipolar response.                                        velopment of GMR materials.

                                                                References
4. Desirable developments
                                                                 [1] J. Daughton, P. Bade, M. Jenson, M. Rahmati, IEEE
  Besides the never-ending quest for high mag-                       Trans. Magn. 28 (1992) 2488.
neto-resistance ratios and lower saturation fields,              [2] B. Dieny, V. Speriosu, S. Metin, S. Parkin, B. Guerney,
there are several developments which would aid in                    P. Baumgart, D. Wilhoit, J. Appl. Phys. 69 (1991) 4774.
the commercialization of GMR materials which                     [3] A. Berkowitz, J. Mitchell, M. Carey, A. Young, S. Zhang,
                                                                     F. Spada, F. Parker, A. Hutten, G. Thomas, J. Appl. Phys.
may not be obvious to researchers concentrating                      73 (1993) 5320.
on GMR material development. These were dis-                     [4] J.S. Moodera, L.R. Kinder, T.M. Wong, R. Meservey,
cussed along with GMR materials and their ap-                        Phys. Rev. Lett. 74 (1995) 3273.
plications, and are summarized in Table 2.                       [5] S. Parkin, A. Li, D. Smith, App. Phys. Lett. 58 (1991) 2710.
  Materials which have little or no hysteresis                   [6] L. Chen, T. Tiefel, S. Jin, T. Palstra, R. Ramesh, C. Kwon,
                                                                     IEEE Trans. Magn. 32 (1996) 4692.
would be very valuable for magnetic field sensors.               [7] H. Yoda, H. Iwasaki, T. Kobayashi, A. Tsutai, M. Sahashi,
Tolerance for very high temperatures of approxim-                    IEEE Trans. Magn. 32 (1996) 3363.



342                        J.M. Daughton / Journal of Magnetism and Magnetic Materials 192 (1999) 334-342

 [8] S. Mao, N. Amin, E. Murdock, J. Appl. Phys. 83 (1998)           [15] J. Leal, M. Kryder, J. Appl. Phys. 83 (1998) 3720.
       6807.                                                         [16] A. Pohm, J. Anderson, R. Beech, J. Daughton, IEEE
 [9] J. Daughton, J. Brown, E. Chen, R. Beech, A. Pohm,                   Trans. Magn. 34 (1998) 1486.
       W. Kude, IEEE Trans. Magn. 30 (1994) 4608.                    [17] Y. Zheng, J. Zhu, IEEE Trans. Magn. 34 (1998) 1063.
[10] Siemens Product Information 03.98, Ordering Number              [18] J. Daughton, J. Appl. Phys. 81 (1997) 3758.
       B143-H7098-G1-X-7600.                                         [19] W. Gallagher, S. Parkin, Y. Lu, X. Bain, A. Marley,
[11] R. Jansen, J.S. Moodera, J. Appl. Phys. 83 (1998)                    K. Roche, R. Altman, S. Rishton, C. Jahnes, T. Shaw, G.
       6682.                                                              Xiao, J. Appl. Phys. 81 (1997) 3741.
[12] Y. Lu, X. Li, G. Xiao, R. Altman, W. Gallagher, A. Marley,      [20] B. Everitt, A. Pohm, R. Beech, J. Daughton, IEEE Trans.
       K. Roche, S. Parkin, J. Appl. Phys. 83 (1998) 6515.                Magn. 34 (1998) 1060.
[13] M. Tondra, J. Daughton, D. Wang, R. Beech, A. Fink,             [21] J. Shi, T. Zhu, M. Durlam, E. Chen, S. Tehrani, Y. Zheng,
       J. Taylor, J. Appl. Phys. 83 (1998) 6688.                          J. Zhu, IEEE Trans. Magn. 34 (1988) 997.
[14] C. Tsang, R. Fontana, T. Lin, D. Heim, V. Speriosu, B.          [22] G. Granley, Honeywell, Private communications.
       Gurney, M. Williams, IEEE Trans. Magn. 30 (1994)              [23] T. Hermann, W. Black, S. Hui, IEEE Trans. Magn. 33
       3801.                                                              (1997) 4029.