3588

IEEE TRANSACTIONS ON MAGNETICS, VOL. 27, NO. 4, JULY 1991

The Exchange-Spring Magnet: A New Material Principle for Permanent Magnets

Eckart F. Kneller. Member, ZEEE, and Reinhard Hawig

Abstract-The best available hard magnetic materials have a lower saturation polarization J , than many soft magnetic materials. They are chemically very reactive and also expensive on account of a substantial content of a rare earth. Therefore, it is proposed to make permanent magnets of composite materials consisting of two suitably dispersed ferromagnetic and mutually exchange-coupled phases, one of which is hard magnetic in order to provide a high coercive field, while the other may be soft magnetic just providing a high saturation J, and should envelop the hard phase regions in order to prevent their corrosion. A general theoretical treatment of such systems shows that one may expect besides a high energy product (BH),,, two uniquely characteristic features, a reversible demagnetization curve (exchange-spring) and, in certain cases, an unusually high isotropic remanence ratio B,/J,, while the required volume fraction of the hard phase may be very low, of the order 10%. The technological realization of such materials is shown to be based on the principle that all phases involved must emerge from a common metastable matrix phase in order to be crystallographically coherent and consequently magnetically exchange coupled. Finally it is shown that a corresponding composite material thus produced exhibits the characteristic magnetic behavior as predicted by the theory.

INTRODUCTION HE theoretical limit for the maximum energy product of a given magnetic material'

(BH),,, 5 J f / 4 p o (1) depends explicitly only on its saturation polarization J , = p a , , but is besides subject to the following conditions: saturation remanence B, = J, (solid ferromagnetic material, i.e., packing fraction p = 1 and alignment of easy axes parallel to the field axis), and critical fields for irreversible magnetization reversal (nucleation fields) H,, 2 J s / 2 p 0 = M , / 2 . On account of these conditions, the limit (1) may in principle be reached only with materials that have a sufficiently high nonplanar (preferentially uniaxial) magnetocrystalline anisotropy K >> Jf/4p0, where K is the magnetocrystalline anisotropy constant as defined by ( 2 ) . More generally speaking, the magnitude of the ratio K = K / ( J : / 4 p 0 ) = 4K/p&,2 characterizes the basic

magnetic behavior of the material. If K >> 1, the behavior is dominated by magnetocrystalline anisotropy; such materials are called hard magnetic (k-material). If K << 1, the behavior is governed by maghetostatic energy, and such materials are called soft niagnetic (m-material). In these terms, the limit (1) is attainable only with a k-material. A survey on the magnetic properties of some wellknown materials of both categories in Table I indicates, however, that the decisive quantity in (l), J,, is for most k-materials considerably lower than for many common m-materials, whereas the coercivity Hch of k-materials may exceed by far the value M s / 2 necessary to reach the limit (1). Moreover, the best of the k-materials contain about 25 wt% or more of a rare eakh metal, which adds to their price and raises serious problems with respect to chemical stability, while most m-materials are much less reactive and rather inexpensive. Therefore it is tempting to consider composite materials consisting of two suitably dispersed and mutually exchange-coupled phases, one of which is of the k-type thus providing a high enough nucleation field for irreversible magnetization reversal, and the other is an m-type material with M, as high as possible, in order to attain a high average saturation. In addition the m-material may envelop the k-phase regions in order to prevent their corrosion. In this work we shall first estimate the critical dimensions of the phases and discuss corresponding models of microstructures. On that basis, the typical magnetic properties of such composites are derived. Finally it is shown how a material of this kind may be realized technologically and that its magnetic (behavior corresponds entirely to the predictions of the theory. MICROSTRUCTURE Critical Dimensions In the following, all quantities referiing to a phase k or m , respectively, will be marked by a corresponding index. The microstructure must be such as to permit no easy (low energy) mechanism of irreversible magnetization reversal in either of the phases. A rough estimate of the corresponding critical dimension of the phases may be derived from the one-dimensional model in Fig. 1 consisting of a sequence along the x axis of alternating k- and m-phase regions of widths 2bk and 2b,, respectively, being crys0 1991 IEEE

Manuscript received May 27, 1991. The authors are with the Institut fur Werkstoffe der Elektrotechnik, RuhrUniversitat, Postfach 10 21 48, D-4630, Bochum, Germany. IEEE Log Number 9101828. 'The magnetic induction is written as B = po(H M ) = poH + J , with po = 4 a . lo-' Vs/Am, thus giving B and the polarization J in T, and the magnetic field H a n d the magnetization M in A / m .

0018-9464/91/0700-3588$01.OO

KNELLER AND HAWIG: EXCI-IANGE-SPRING MAGNET

3589

TABLE 1

ROOMTEMPERATURE VALUES OF

MAGNETIC PROPERTIESSOME MAGNETICALLY (k-type) AND SOME MAGNETICALLY (m-type) MATERIALS OF HARD SOFT
K =

Category k-type hard

(K

Material BaO 6FeZ03 MnBi Nd2Fe,,B Co,Sm

Crystal Symmetry hex tetr tetr hex

Tc ("C/K)

M, (lo6 A/m)
0.38 0.58 1.25 0.84

K
(lo6J/m3) 0.32 1.16 9.4 11.9

4K AM? 7.2 9.2 19 54

J3 H A
P d A
J3

4140 0.45 1.3 4.9 2.2

(lo6A/m) 1.35 2.94 12 23

(T)
1.7 3.7 15.1 28.6

(T) (lo5J / m 3 )
0.47 0.72 1.57
1.05

> 1) >
2K

HA =
(K

450/723 360/633 312/585 730/1003

M.

= 1) CO hex

1120/1393

1.40

0.53

0.84

0.70/0.60

0.88/0.75

1.76

6.2

2 /kW
m-type soft
(K

< 1) <
Ms

HA

U-Fe FeZ3B6 Fe3B

cub cub tetr

760/1043 425/698 510/783

1.70 1.35 1.28

0.047 0.01 m0.2

0.05

0.03 0.39

0.85 0.67
0.64

1.06 0.86 0.80

2.13 1.70 1.60

9.0 5.7 5.1

m
0

H-v
b, =, , 6

= const.

I I
I

b ,

decreases

Fig. 1. Schematic one-dimensional model of the microstructure and the micromagnetic structure of the exchange-coupled composite material as a basis for the calculation of the critical dimensions of the phase regions. (a) Saturation remanence. (b)-(c) > (d) Demagnetization in an increasing reverse field H a t a constant overcritical width of the m-phase, b,, > b,,,8. Demagnetization at decreasing width b,, + bcm.

3590

IEEE TRANSACTIONS ON MAGNETICS, VOL. 27, NO. 4, JULY 1991

tallographically coherent and exchange coupled through the phase boundaries [1 1 . The magnetocrystalline anisotropy is assumed uniaxial (for simplicity) in both phases, with both easy axes being parallel to the z axis perpendicular to x . The agisotropy energy density depends on the angle 4 between M and the easy axis as

Ek

K sin2 4,

(2)

with K > 0. It should be noted at this point that for the k-phase there is, per definition, K k = ~ K , / P , , M ? ~ >> 1 , and for the m-phase, K , = 4Km/p,,M:,,, << 1 . The ratio K k / K , is of the order lo2 to l o 3 , according to Table I, and is due mainly to the magnitude of the ratio Kk/K,,, of the order lo2, whereas Msk and M,, are mostly of the same order, with the ratio M:m/M:k not exceeding 10 in practical cases. The exchange energy density may be written in the form

EA = (3) where A is a constant of the order lo-" J / m at room temperature, but depends on the Curie temperature T, and : , I on temperature Tas A 0 T , [ M , ( T ) / M , ( O ) ] * and I, is f y the present model the angle in the yz plane between M, and the z axis. With these quantities, the energy per unit area of a 180" Bloch wall in a homogeneous material may be written approximately as y = 6K 6A(.rr/6)*, (4) where 6 is the wall thickness. In equilibrium, y(6) has a minimum (dy/d6 = 0), from where the equilibrium quantities 6,, yo are obtained:

where the wall will invade into the k-phase, thus leading to irreversible magnetization reversal of both the m- and the k-phase regions. The corresponding critical field H,, is definitely lower than the anisotropy field of the k-phase H,,, < HAL = 2 K k / M s k ,yet it will have about the same order of magnitude. The coercive field H,M is defined by M(H,M) = 0 and is much smaller, HcM << H,,,, as Msm > Msh and also because it has been assumed that b, >> bk, and thus the demagnetization curve between M , ( H = 0) and M(H,M) = 0 is completely reversible. If 6, is now reduced to val, ues b, <,,6 H,, remains unchanged, but H,M increases because, for H < H,,,, the thickness of the 180" walls in 6,,. the m-phase is essentially confined to 6, = b, Hence the critical width of the m-phase b,,, giving the maximum coercivity HcM, is determined by the condition (7) with 6, = b,,. From (4) finds for small 6, (6, << So,> that y m (6,) one = 6J,(~/6,)~, from where the energy density EYm = y,/6, = A , ( T / ~ , ) ~Inserting this result into (7) and . setting 6, = b, yields the critical dimension of the mphase: b,, = T(A,/~&)'/~.

(8)

6, = a ( ~ / ~ ) ' / 2
yo = 2 a ( A
*

(5)

(6) Critical dimensions for a high-energy magnetization reversal in the one-dimensional system in Fig. 1 are obtained from a consideration of the reversal process. If one assumes that the hard k-phase has a reasonable thickness, 6,k = T ( A k / K k ) ' / * corresponding to about its e.g., bk critical thickness (see below), then, starting from the saturation remanence along the easxdirection +z (Fig. l(a)) with an increasing reverse field H , the magnetization will invariably begin to change (reversibly) in the soft m-phase. For large b,, b, 2 ,a = T ( A , / K , ) 1 / 2 >> 6ok 2: b , k (because K , << &), two equilibrium 180" walls _will form reversibly in the m-phase (Fig. l(b)). When H is further increased (Fig. l(c)), these walls will be reversibly compressed towards the k-phase boundary, and the energy density in these walls will increase above its equilibrium value Eym = y m / 6 , > Ex,, = yorn/6,,, while the magnetization in the k-phase Msk remains essentially unchanged, as Kk >> K,. This process will continue until E?, approaches the equilibrium energy density Eyok of a wall in the k-phase
=
Ym/6m

K)1'2.

With representative values A , = lo-" J / m , Kk = 2 * lo6 J/m3, one gets forb,, the order b,, = 5 nm. For the k-phase a critical thickness cannot be derived theoretically (Brown's paradox [ 2 ] , [3]). For practical purposes it seems reasonable to take bckabout equal to the equilibrium wall thickness in the k-phase bck = 6ok = n ( A k / K k ) ' / 2as has been assumed initially. Since mostly , Ak < A , on account of the generally low Curie temperatures of the k-materials (see Table I), this gives for bck about the same magnitude as b,,:
bck

bcm*

(84

Eyok

Yok/Aok

2Kk,

(7)

Volume Fractions o Phases f The optimum geometry of the microstructure mini? , mizes the volume fraction of the k-phase + = vk/v (vk = volume of k-phase; V = total volume of material) under the conditions of about equal lateral dimensions of the two phases, b,, = bck (equation 8(a)), and the chemical protection of the k-phase by the m-phase. A mathematical solution of this problem would be of no practical value, because in technologically realizing such a material one must put up with the microstructure that nature produces with a suitable pair of phases. We can tell immediately, however, that the type of microstructure to look for is a homogeneous precipitation of a k-phase in an m-phase, and not vice versa. With the reasonable assumptions that precipitates with diameters of the order of some nm are spherical (minimum surface to volume ratio) and are spatially distributed approximately according to an fcc lattice (equal diffusion paths), as shown in Fig. 2 . One obtains uk = a / 2 4 d 2 = 0.09. Virtually the same result is obtained with a bcc lattice: z.'k = n ! 3 / 6 4 2 0.09. :

KNELLER A N D HAWIG: EXCHANGE-SPRING MAGNET

\
/

3591

I
Fig. 2. Schematic model of a favorable microstructure. The hard magnetic k-phase regions (encircled) are spherical precipitates on an fcc space lattice in soft magnetic rn-phase. At bck = b,,, the volume fraction of the k-phase is uk = ?r/24d2 = 0.09.

Fa-+

Once uk is known, the average saturation magnetization of the material is given by M, = ukMsk (1 - Vk)M,,. (9) Under the initial assumption Msk < M,, and with z.k = 0.09, (9) yields M, = M,,, whereas for platelets of alternating k- and m-phases as in the model Fig. 1, uk = 0.5 and M, = (Msk M,,)/2.

It is for this specific and quite typical magnetic behavior, in a sense resembling a mechanical spring, that such magnets have been termed exchange-spring magnets. Their striking reversibility in conjunction with a high remanence and a high coercivity distinguishes them uniquely from the conventional single ferromagnetic phase permanent magnets, where the demagnetization curves reflect essentially the distribution of the critical switching fields and are, therefore, mainly irreversible (Fig. 3(c)). In order to further illustrate these features, some minor loops are drawn in the Fig. 3(a)-3(c), as they will be obtained upon reducing to zero and reincreasing the field at various points along the demagnetization curve. The recoil permeability p, of an exchange spring magnet is expected to be about 5 times as large as that of a conventional magnet with equal coercive field and saturation magnetization. As for the general shape of the demagnetization curve M ( H ) , it is immediately obvious from the exchangespring mechanism that an optimum microstructure (b, = bcm) will yield a normal convex M ( H ) curve (Fig. 3(a)) between M , and M = 0, similar to a conventional permanent magnet (Fig. 3(c)). Whereas an overaged microstructure (b, >> b,) must lead to a quite characteristic shape of the demagnetization curve being concave throughout from short below the remanence on until saturation in the reverse direction. In any case, the exchange coupling between the phases produces a shape of the saturation loop like that of a uniform material showing no indication of the presence of two phases with extremely different magnetic hardness. If there was no exchange coupling, one would get a constricted loop as in Fig. 3(d). Saturation Remanence Ratio m, = M r / M F The value of m, depends on the phases employed. A quantitative calculation of m, for a given pair of phases is in general difficult, because it requires the micromagnetic treatment of complex many body magnetic systems. Therefore, we shall describe here only the character of the problem and on that basis will derive approximate solutions for two simple cases in order to illustrate how the parameters of the system determine the result. It will be generally assumed that the microstructure originates by precipitation of the k-phase in an m-matrix and thus corresponds in essence to the model Fig. 2, and that the number of k-precipitates in a single m-grain is large enough to apply statistics where appropriate. Moreover, it is assumed that the k-phase has a uniaxial crystal structure, e.g., tetragonal or hexagonal, with the ck-axis being the easy magnetic axis, whereas the m-phase may have any symmetry, in particular cubic symmetry. Since there must be magnetic exchange coupling between the k- and the m-phase regions, the phases must be crystallographically coherent. This implies that the orientations of the ck-axes (easy axis of the k-precipitates) in the m-matrix are not arbitrary, but must be parallel to speof cific crystallographic axes [hokoZo] the m-crystal lattice. Since the precipitate will generally form at a temperature

MAGNETIC BEHAVIOR Hysteresis Loop and Demagnetization Curve According to the foregoing analysis, the demagnetization curve after previous saturation of such a material will be reversible in reverse fields H < H,,, i.e., before the magnetization of the k-phase begins to switch, as is illustrated schematically in the Fig. 3(a) and 3(b). At H < H,,, the material has unidirectional anisotropy [ 11 on account of the exchange coupling between the two phases. This and the following statements with respect to reversibility hold independently whether the material is aligned or isotropic (see also the section on remanence). For a given pair of phases, the reversible range in M , AM,,,, depends on the volume fraction of the hard phase v k (or the soft phase U = 1 - uk, respectively), on the , ratio M,,/Msk, and on the lateral dimension of the mphase b,. At fixed uk and M,,/M,k, AM,,, is smallest for b, = b,, (the optimum microstructure, Fig. 3(a)) and increases as b, > b,, (the overaged state, Fig. 3(b)), because H,, remains constant. At large U,, U , = 0.8, for example, AM,,, may well exceed the saturation remanence AM,,, > M , (Fig. 3(b)).

3592

IEEE TRANSACTIONS ON MAGNETICS,

VOL. 27, NO.

4, JULY 1991

EXCHANGE SPRING

CONVENTIONAL

r..................
I
I

TWO INDEPENDENT PHASES

*
I

I I I
/

--a----

(d)

reversible i r reversible

Fig. 3 . Typical demagnetization curves M ( H ) (schematic). (a) Exchange-spring magnet with optimum microstructure, 6," = 6cm. As (a), but with overaged microstructure, 6,,8> 6<,". Conventional single ferromagnetic phase magnet. (d) Mixture (b), > (c) of two independent ferromagnetic phases with largely different hardness (constricted loop).

above the Curie temperatures of both phases, it may be presumed that the ck-axes of the k-precipitates are distributed equally among these distinguished orientations [hokolol. Now we consider a single spherical (in order to exclude shape anisotropy) m-grain and disregard demagnetization effects.* Since Kk >> K,,,, the resulting saturation rema'This is justified because in the polycrystalline material each grain is imbedded in magnetic material having a comparable mean magnetization.

nence of the k-precipitates in the grain i i r k will then point in the direction [h, k, Islo, which is symmetric with respect to the ck-axes, [h&olo] and forms the sm_allestangl5 8 with the previously applied 2aturation field H. Hence Mrk is in general not parallel to H. For simplicity, we shall assume for the following that all of these symmetry axes [h, k, Z] , with respect to [h,$olo]are crystallographically equivalent and form the same angle (Yk with all possible ck-axes [hdcolo3. In such cases the relative magnitude of i$i,.k beCOmeS Mrk.Msk = mrk = COS CYk.

II

KNELLER AND HAWIG: EXCHANGE-SPRING MAGNET

3593

The magnetization of the soft m-matrix is exchange coupled with the magnetization of the k-precipitates along their mutual phase boundari5s. Therefore, the resuQng remanence of the mLmatrix M,, will be parallel to Mrk. The relative size of Mrm,M,,,,/M,, = m, will, however, be greater than mrk = cos ak, because the exchange coupling within-the m-matrix will smooth out the local magnetization M,,(3) (? = position vector) between the nearly fixed (because Kk >> K,) magnetization directions in the surrounding k-precipitates in such a way that the t_otal free energy is minimized. Thus the magnetization M,,(3)of the m-matrix in the remanent state will be, in general, inhomogeneous, and the resultant mrm must be calculated micromagnetically from the condition of minimum total energy, which may be complicated. Consequently there exist in each grain of the material j, depending on the crystal symmetry of the m-phase and the crystallographic coherency conditions for the k-precipitates, one or several equivalent axes [h,k,l,] for the total resultant saturation remanence vector $whose relI, ,, ative magnitude
"i-1

54.5", giving mrmZ 0.79. With these results, (10) yields with M, from (9), mi Z 0.78. Since Mrj will always lie in the [l 1 11 direction next to the direction of the previous saturation field H, (cos 8) = 0.87, and we obtain from (1 1)

Mr~/~s = (l/Ms)[ukmrkMsk

+ (l

uk)mrmMsml

(10) is the same for all grains. For a polycrystalline sample of magnetically independent grains with their crystallographic axes oriented at random, the relative remanence rn, is then obtained by averaging over the angles-8 between the direction of the previous saturatio? field H and the corresponding directions [h,k,l,]Oof M,, in the grains:
m, = M,/M,
=

i.e., an isotropic remanence ratio rn, considerably above 0.5, although the angular distribution of the easy axes of the uniaxial magnetically hard phase constituents is by all means isotropic in a polycrystalline sample with random grain orientations. This result is a consequence of the exchange coupling between the k- and the m-phases as well as within the magnetically soft m-matrix, and also to some extent of the cubic symmetry of the soft phase. Thus it may be concluded that a large isotropic remanence ratio rn, > 0.5 is a possible but, with regard to Example 1, not a universal feature of exchange-spring magnets. In connecting these two results with the foregoing characteristics of the hysteresis loop, it appears that a largely reversible demagnetization curve in conjunction with an isotropic saturation remanence ratio m, 2 0.5 may be considered a criterion for the presence of the exchangespring mechanism.

Nucleation Field H,,, and Coercive Field HcM

Of the characteristic properties of an exchange-spring magnet, the nucleation field H,,, for irreversible magnetization reversal and likewise the coercive field HcM are the most complex and least reliably predictable quantities. Only rough estimates will be attempted. With regard to (7) and the way it was derived one would get for a completely aligned microstructure as in Fig. 1 H,,, = 2&/pdM,,, and for an isotropic polycrystal about one half of this value
Hno

mi (cos

e).

(1 1)

Any numerical evaluation of (1 1) requires the knowledge of the crystal symmetry of the m-phase, the crystallographic orientations [hokolo]of the ck-axes, and the volume fractions and saturation magnetizations of the phases. This is illustrated by two simple examples. Example 1 : The m-phase has a uniaxial symmetry (hexagonal or tetragonal), and the ck-axesof the k-precipitates form parallel to the cm-axis of the m-matrix, i.e., [hokolol = [OOlI. In this case, mrk = mrm = 1, and from (10) with M, from (9), m , = 1. Since 8 may have any value 0 I85 90" in space, (cos 8) = j:l2 cos 8 sin 8 d8 = 0.5. With these values, (11) yields for a random polycrystalline sample
(1la) independently of the volume fractions and magnetizations of the phases. However, this material may be aligned. Example 2: The m-phase is cubic, the ck-axes form parallel to the cube edges of the m-matrix, [hoke$] = [loo], u = 0.1, and M,, = 2 Msk. In that case, M,, is k along [h,k,l,] = [ 1 1 1 ] , i.e., (Yk = 54.5" and mrk = cos a k = 0.58. mrm,on the other hand, must be calculated micromagnetically. However, a lower bound of mrmmay be obtained by averaging over all space angles 5 (Yk =

= Kk/pdMsm.

(12)

m, = m, (cos 8) = 0.5

With representative quantities Kk = 2 lo6 J/m3, p&l,, = 1.8 T, this would give for H,,, the order lo6 A/m, which seems rather high and may be considered an upper limit to H,,,. For an optimum microstructure, b, = b,, [see (8)], one would expect HcM 2: H,,,. For an overaged microstructure, i.e., b, > b,,, H,M will depend on b, as

Forb, = b,, = ?r(A,/2Kk)'I2 [see (8)], (13) transforms into (12), HcM = H,,,. With representative values A, 2: lo-" J / m , p a s m = 1.8 T, (13) becomes HcM = 3 10-"/bi A/m, giving with, e.g., b,,, = 10 nm, HcM = 3 lo5 A/m.

3594

IEEE TRANSACTIONS ON MAGNETICS, VOL. 27, NO. 4, JULY 1991

With regard to (12), H,, and similarly HcM must vary with temperature about as Kk/Ms,. In particular, if the Curie temperature of the k-phase is substantially lower than that of the soft m-phase Tck c T,, (see Table I), HcM will decrease rapidly with rising temperature and at Tck will reach the low value of the soft phase. TECHNOLOGICAL REALIZATION Microstructure Essential conditions for the microstructure of such materials are a fine and regular dispersion of phases on a scale of the order 10 nm (bck = b,, = 5 nm) and exchange coupling between the k- and m-phase regions, which implies the crystallographically coherent formation of two phases with generally different structures. Such conditions are known to arise during the continuous decomposition of a metastably supersaturated phase that has crystallized from a glassy state (cf. e.g., [4]). This means that materials with the desired properties may be produced by liquid quenching of a suitable alloy to the glassy state and subsequent heating of the glass. In fact the usefulness of the glassy state as a precursor for producing permanent magnet materials has been widely recognized for over ten years (see, e.g., [5]-[l l]), mainly because very finely grained microstructures may be obtained by this method. It seems likely that some of the materials produced in this way were of the type discussed here, but that fact was apparently not recognized. The first indication of typical features of such a mechanism was gathered from a recent publication of Coehoorn et al. [ l l ] (cf. also [9], [lo]) who reported that liquid quenching and subsequent heating of alloys in the vicinity of Nd4Fe77B leads to remarkable permanent magnet properties (up to HcM = 280 kA/m, (BH),,, = 95 kJ/m3) in a state in which the material was found to consist of 85 % definitely soft magnetic phases (73 % Fe3B and 12% a-Fe) and only 15% hard phase (Nd2FeI4B). Moreover, it was particularly noted in that publication that the material exhibits an unusually high isotropic remanence ratio m, > 0.5. Therefore these3 and similar alloys, especially Nd3.8Fe77.2B19 [alloy (A)], were prepared and studied in our own laboratories with particular attention to the structural development during heating of the glass, which was observed by a continuous X-ray structural analysis technique [ 121 (DTXD = Ilynamic Temperature X-ray Diffraction). As a result, the work of Coehoorn et al. [ 113 was largely confirmed. In particular, in the optimum magnetic state of the alloy (A), as obtained by annealing the glass for 10 min at 675"C, the material was found to consist of exactly the phases given in [ 111 and reached the remarkable coercivity HcM = 160 kA/m. The history of this optimum state seemed different,
3Some samples were kindly made available to us by Dr. R. Coehoorn of the Philips Research Laboratories, Eindhoven, The Netherlands.

however, from that described in [I 11. Namely, the DTXD photograph of the crystallization process of alloy (A) in Fig. 4(a) suggests (cf also [13]) in contrast to [l 13 that none of the phases present in the optimum magnetic state (a-Fe, Fe3B, Nd2Fe14B)emerges directly from the glassy state. Instead the glass appears to crystallize completely in the metastable Cr2,C6-type phase Fe23B6, whose very broad and therefore barely detectable reflexions seem to result from its supersaturation with the comparatively large Nd atoms. Almost immediately after crystallization, the Fe23B6phase decomposes, first and to the most part into Fe3B and some a-Fe and finally, at a somewhat higher temperature, also Nd2Fe14B.One gets the impression that the formation of Fe3B and a-Fe (none of which dissolves an appreciable amount of Nd) further increases the supersaturation in Nd of the remaining Fe23B6phase regions up to a concentration where these transform into Nd2Fe14B. Moreover, all of these three phases may well be crystallographically coherent when all of them actually develop from a common crystalline Fe23B6phase matrix, thus providing an essential condition for the exchange-spring mechanism. The DTXD photograph in Fig. 4(a) is, however, not clear enough to serve as an ultimate proof of this view. Following up on these ideas, the constitutional question could finally be settled as follows. The primary crystallization from the glass of the phase Fe23B6is essential, because this phase turned out to be the only one of all phases involved that is capable of dissolving metastably an appreciable amount of the large Nd atoms such that the hard phase Nd2Fe14B later precipitate homogeneously can in that matrix. In addition, Fe23B6is magnetically soft with a comparatively high magnetization (see Table I), and has cubic crystal symmetry, thus favoring a high isotropic remanence ratio m,, as has been ekplained. Consequently, starting from the alloy (A), the Fe23B6 matrix phase and the magnetically essential hard phase Nd2Fe14Bwere stabilized by small additions of the elements Si and V, which at the same time prevent the early formation of Fe3B and a-Fe without having appreciable effects on the magnetic properties. These structural effects are clearly seen in comparing the DTXD recording of a corresponding alloy Nd3,8Fe73.3B18.0Sil [alloy (B)] in .0V3.9 Fig. 4(b) with that of Nd3,8Fe77,2B19 [alloy (A)] in Fig. 4(a). The large widths of the Fe23B6reflexions are in fact a consequence of the dissolved large Nd atoms, as mentioned already. Upon annealing the glass of alloy (B) for 10 min at 675"C, the hard phase Nd2Fe14Bis precipitated in an almost pure Fe23B6phase matrix, as is seen in the Guinier photograph Fig. 5(a). There is only a trace of Fe3B detectable, whose phase must appear upon the precipitation of Nd2Fe14B from the Nd supersaturated Fe2& matrix in order to accommodate the excess boron. Most likely, Fe3B forms shells around the precipitates. The estimated portions of phases are about 90% Fe23B6 + Fe3B and 10% Nd2Fe14B.Some of the Nd is still dissolved in the Fe23B6 matrix, as is indicated by the reflexion widths of this

KNEl-LER AND

3595

(b)

Fig. 4. DTXD photograph (temperature Tversus diffraction angle 40) taken with CO-K,, radiation at the heating rate of 50 K/h. The initial state is liquid-quenched glass. (a) Nd3 xFe,, .B,, [alloy (A)]. The phases are (see also 1131): A = FezlB,; a = bet Fe$: b = Nd2Fel,B: c = a-Fe (bcc): d = N d , +eFe,B,; e = bct Fe2B; f = unknown tetragonal phase: g = ?-Fe (fcc). (b) Nd, &,, oV, ,Sil [alloy (B)]. The phases are (see also [ 131): a = Fe,,B,,; b = bct Fe,B; c = Nd?Fe,,B; d = a-Fe (bcc): e = Nd, +,Fe4B4:f = bet Fe,B: g = unknown tetragonal phase.

phase. The coercivity in this state is about f f c M = 200 kA/m. In the optimum magnetic state of alloy (B) obtained by

annealing the glass for 10 min at 715C (Fig. 5(b)), the amounts of Fe3B and Nd2Fe14B have increased at the expense of Fe23B6,which now contains much less Nd as is

3596

IEEE TRANSACTIONS ON MAGNETICS. VOL. 27. NO. 4. J U L Y 1991

50

100

40

150

Fig. 5 . Guinier photograph taken with Co-Koc radiation at room tcmperature. Ditfraction angle 40. ( a ) N d , ,Fe,, 3B,, ,,V, ,Si, [alloy (B)] glass, annealed for I O min at 675C. (b) As (a). annealed for I O rnin at 715C. Optimum magnetic state. The reflexions of the various phases are identified in Fig. 4.

indicated by the sharper reflexions. No a-Fe is detectable. In that state, one can estimate about 80% soft phases, The coerFe23B6+ Fe,B, and 20 % hard phase Nd2Fe14B. civity has increased to nearly HcM= 300 kA/m. Thus the constitution of the alloy (B) in the optimum state corresponds quite clearly to the material concept and corresponding microstructure (Fig. 2) as described before. After annealing the glass of alloy (B) for 10 min at 775"C, Fe,,B6 has disappeared completely and a-Fe has formed, such that the alloy (B) consists now of the same phases in about the same proportions as the alloy (A) in its optimum magnetic state, and the coercivity has dropped to about the same lower value HcM= 150 kA/m. On the grounds of this final result, the context between the two cases (A) and (B) appears to be this: Both cases have in common that the glass crystallizes completely in the Fe23B6phase. In case (B) the further development begins with the prein cipitation of the hard phase Nd2Fe14B the magnetically soft matrix Fe2,B,, thereby producing directly a microstructure corresponding to the model in Fig. 2. Thereafter, the Nd depleted Fe2,B6 matrix decomposes gradually into Fe3B and a-Fe, which are also soft magnetic, such that the properties of the microstructure are in principle preserved. In case (A), the further development begins with the decomposition into Fe3B and a-Fe, thereby increasing the Nd concentration in the remaining Fe23B6phase regions until these transform into the hard phase Nd2Fe14B,i.e., the hard phase forms here not as a primary precipitate in the Fe23B6phase, but from the Nd enriched remainder of the decomposition of FeZ3B6into Fe,B and a-Fe. Thus the final states are again virtually equal in both cases: A magnetically soft matrix containing very fine and rather homogeneously distributed inclusions of a hard phase, according to the model concept in Fig. 2. It is remarkable that the crystallographic coherency and consequently the exchange coupling between the phases is retained, as is evident from the magnetic properties. Magnetic Properties of the Alloy (B) in the Optimum State In Fig. 6 are reproduced the dc demagnetization curves after previous saturation in the opposite direction and

some minor loops along the demagnetization branch as measured on the alloy (B) in the optimum state (Fig. 6(a)) and in an overaged state (Fig. 6(b)). These curves are to be compared to the schematic of Fig. 3(a) and 3(b), respectively. Both Fig. 6(a) and 6(b) show the typical characteristics of the exchange-spring mechanism, namely a high (isotropic) remanence ratio m, > 0.5 and a high degree of reversibility in fields below HcM, recoil pennethe ability p, = 1 M / H being of the order of 5 . In the optimum magnetic state (10 min annealing of the glass at T,, = 715"C, see also Fig. 5(b)) after applying a maximum field of 1200 kA/m for saturation, the following quantities were obtained: J , = poM, 2 1.7 T; m, = M,/M, = 0.76; H c M = 290 kA/m; HcB= 250 kA/m; (BH),,, = 105 kJ/m3. 10 min annealing at lower temperatures, 675C I Tu < 715"C, raises m, to 0.80 and = reduces H ( Mto lower values, H c M 200 kA/m (see Fig. 5(a)), as decreases. Overaging, on the other hand, either by prolonging the annealing time at 7 15"C or by increasing T,, above 7 15"C, reduces all these quantities. It is interesting to analyze the dc demagnetization curve M ( H ) between the saturation remanence M , and the reverse saturation -M, in terms of the reversible and the irreversible portions of the total magnetization change as a function of the reverse field. The irreversible portion is described by the dc field demagnetization remanence Md (H), being the remanence acquired after saturation in one direction and subsequent application of a dc field H in the opposite direction. Fig. 7(a) shows a plot of the reduced quantity D ( H ) = [M, - Md(H)]/2Mr = -AM,,,,,(H)/2Mr versus H a s measured on the alloy (B) in the optimum state. The derivative dD(H)/dH = f(H,,) is the distribution function of critical fields H,, for irreversible magnetization reversals. The experimental curve D ( H ) may be interpreted according to the model in Fig. 1, i.e., by purely inhomogeneous magnetization rotation (Kondorskij mechanism : [14]), assuming equal probabilities 0 angles 0 5 I!? I 90" in the xz plane between the field H a n d the z axis. In that case, H,,(I9) = H,(O)/cos I9 = H,,,/cos 0. When the reverse field H increases from H = 0, no irreversible magnetization reversal occurs as long as H < H,,,. When H increases further to some value HI = H,,,/cos 01, for example, all regions with 0 II!? 5 8, have reversed, and

r,

KNELLER A N D HAWIG. EXCHANGE-SPRING MAGNET

3591

.: .... --L

....... . +
~

(a) Fig. 6. Demagnetization curve M ( H ) measured after previous saturation in the opposite direction in the field H = 1200 k A / m on the alloy (B). (a) in the optimum magnetic state. h,,, = h,,,, (cf. also Fig. 5(b)), (b) in an averaged state. h,,, >> h ,,,,

the total irreversible change in magnetization is -AM,,,,, = 2M, cos 8 d8 = 2Mr sin 8 ) , from where D ( H , ) = sin 8 ) . The best fit of a corresponding plot of sin 8 versus H,,,/cos 8 to the experimental curve D ( H ) is obtained with the value H,, = 3.0 lo5 A/m. This curve is drawn in Fig. 7(a). Hence most of the distribution f(H,,) is due to the distribution of angles 8, whereas the mechanism of the irreversible magnetization reversal and the k- and mphases involved in it are the same throughout the sample. Measurements of the reversible portion -AM,,, ( H )/M, versus H are given in Fig. 7(b) for the same sample as Fig. 7(a). The reduction with Mr serves only the correlation with Fig. 7(a) and has no theoretical background. The initial (below the maximum) and final (high H ) parts of the curve in conjunction with Fig. 7(a) reflect in an obvious manner properties of the hysteresis loop. The most remarkable feature of the curve is the sharp maximum at H = H,,,, i.e., precisely where it is to be expected on account of the proposed magnetization mechanism. Fig. 8 shows the coercivity HcMof the alloy (B) in the optimum state as a function of the temperature T. As is

seen, the extra coercivity that is gained by the exchange coupling of the soft m-phase to the hard k-phase decreases with rising T and vanishes at the Curie temperature of the hard phase Nd2Fe14B (T(+= 580 K, see Table
1).

CONCLUSION
A theoretical treatment is proposed for permanent mag-

net materials consisting of two suitably dispersed ferromagnetic and mutually exchange-coupled phases, one of which is hard magnetic (high magnetocrystalline anisotropy K ) and thus provides a high resistance against irreversible magnetization reversal, while the other may be magnetically soft having a high saturation induction J,yto provide a high average magnetization of the material. In practice there may be more than one phase of each category, the only conditions being that they are all equally distributed and exchange coupled among each other. The following results were obtained. The optimum microstructure of corresponding materials is a homogeneous distribution of the hard phase in a magnetically soft ma-

3598

IEEE TRANSACTIONS ON MAGNETICS, VOL. 27, NO. 4, JULY 1991

(b)
Fig. 6 . (Continued.)

trix, with the lateral dimensions of both phases being about equal and equal to the critical diameter of the hard phase b, 2: bk = bck(e.g., Fig. 2). In this case, the required volume fraction of the expensive hard phase(s) Vk is small, of the order l o % , and if the hard phase is chemically very reactive, e.g., in cases of rare earth compounds, it is protected against corrosion by the soft matrix. The magnetic behavior of such materials is uniquely characterized by a reversible demagnetization curve, i.e., maximum recoil permeability and, in certain cases, also by an unusually high isotropic remanence ratio m, = M , / M , > 0.5. With regard to the small portion of hard magnetic phase, the energy product (BH),,, is potentially very high, mainly due to the high average saturation induction and the high barrier against irreversible magnetization reversal that may be attained in these composites. The technological realization of such composites is found to be based on the following metallurgical principle. Both the hard and the soft phases must be crystallographically coherent as a necessary condition for mutual exchange coupling and, therefore, must emerge from a

common matrix phase. This matrix phase must basically be magnetically soft, and the magnetically hard inclusions in it (schematic model in Fig. 2) may be obtained by, e.g., homogeneous precipitation, or by transformation of a suitably composed and distributed remainder of a primary continuous decomposition of the matrix phase into other soft magnetic phases. Both ways have been shown to lead to comparable products with comparable magnetic properties. The key product in this process, a metastable common matrix phase of a suitable composition, is most conveniently obtained as the first crystallization product upon heating of a corresponding alloy glass, or directly by crystallization during liquid quenching of the melt. Both ways are equally suited to produce such highly metastable key phases. Finally, it is shown that a specific composite material produced along these lines exhibits precisely the characteristic magnetic properties as predicted by the theory. It should be emphasized that this specific material is just one representative example of a large class of two or more phase composites with exchange-spring behavior, which may be produced after the same technological prin-

II

KNELLER AND HAWIG: EXCHANGE-SPRING MAGNET

3599

1 .o

a 0.8
C .-

cn

0.6

2 >

$
0.4

B
sine
0
In

t-

L i 0.2
W

I
0

O w

0
1

v H,/cose s (theoretical)
I
I I

1.o

0.8
L

f 0.6 >

5
'

?I

0.4
0.2
/
/ / / /

d
1 1

0 0

Hno
(b)

5
H in 105~/m

1 0

Fig. 7. Analysis of the total magnetization change - A M ( H ) = - A M , , , , ( H ) - AM,,(H), along the demagnetization curve M ( H ) , after previous saturation in the opposite direction in the field H = 1200 kA/m of the alloy (B) in the optimum magnetic state. (a) Measurements of the irreversible portion D ( H ) = - A M , , , , , ( H ) / 2 M r versus H . The theoretical curve sin 8 versus H,,,/cos 0 corresponds to the Kondorskij model [ 141 of completely inhomogeneous rotation. H,,,, = 300 kA/m from the best fit of the theoretical curve to the experimental data. 0 is the angle between the field axis and the z axis (see Fig. I ) . (b) Measurements of the reversible part - A M , , , ( H ) / M , showing a sharp peak at H = H,,,, = 300 kA/m. The reduction with M, serves the correlation with (a) and has no theoretical background.

ll-1

.. .

3600

IEEE TRANSACTIONS ON MAGNETICS, VOL. 21, NO. 4, JULY 1991

0,4-

t 5 I
0

0,2-

03

-- -bn-

Fig. 8. Coercive force hoHcM versus temperature T as measured on the alloy (B) in the optimum magnetic state. The end of the steep decrease in H c M ( T )corresponds to the Curie temperature of the hard phase Nd2Fe,,B, TcK = 580 K. The low coercive force at T > 580 K is that of the soft magnetic phases Fe23B, (T, = 698 K) and Fe3B (T, = 783 K). The curve is completely reversible.

ciples as described also at quite different compositions in the same alloy system and likewise in other alloy systems.
ACKNOWLEDGMENT The authors are indebted to Dr. Y. Khan for the DTXD photographs and his invaluable help in the structural analysis, to Dip1.-Min. H. Wibbeke and again to Dr. Khan for many stimulating discussions at the start, and to Miss G . Driessen for her untiring technical assistance.

REFERENCES
[ l ] W. H. Meiklejohn and C. P. Bean, Phys. Rev., vol. 102, p. 1413, 1956; Phys. Rev.,vol. 105, p. 904, 1957. [2] W. F. Brown, Jr., Rev. Mod. Phys., vol. 17, p. 15, 1945. [3] S. Shtrikman and D. Treves, Magnerism, vol. 1 1 G . T. Rad0 and H. 1, Suhl, eds. New York: Academic, 1963. [4] E. Kneller and Y. Khan, Z. Metallkde, vol. 78, p. 825, 1987. [ 5 ] J. J. Croat, J. Magn. Magn. Mar., vol. 24, p. 125, 1981. [6] J. J. Croat, J. F. Herbst, R. W. Lee, and F. E. Pinkerton, J. Appl. Phys., vol. 5 5 , p. 2078, 1984. [7] J. J. Becker, J. Appl. Phys., vol. 5 5 , p. 2067, 1984. [8] K. H. J. Buschow, D. B. de Mooij, and H. M. van Noort, J. Less Common Metals, vol. 125, p. 135, 1986. [9] K. H. J. Buschow, D. B. de Mooij, and R. Coehoorn, J . Less Common Metals, vol. 145, p. 601, 1988. [lo] R. Coehoorn, D. B. de Mooij, J. P. W. B. Duchateau, and K. H. J. Buschow, J. d e Phys., vol. 49, p. 669, 1988. [ I l l R. Coehoorn, D. B. de Mooij, and C. De Waard, J . Magn. Magn. Mat., vol. 80, p. 101, 1989. [12] Y. Khan, J. Phys. E: Sci. Instrum., vol. 18, p. 1054, 1985. [13] Y. Khan, Phys. State. Sol. ( a ) , vol. 122, K1, 1990. [14] E. Kondorskij, J. Phys. USSR,vol. 2, p. 161, 1940.

Eckart F. Kneller (M89) was born in Magdeburg, Germany, in 1928. He received the Dipl. Phys. degree and the Dr. rer. nat. degree in physics from the Technische Hochschule (TH) Stuttgart, Germany, in 1951 and 1953, respectively. In 1960, he received the venia legendi for metallurgy and metal physics, and in 1966 he was appointed Apl. Professor at the TH Stuttgart. From 1953-1960 and from 1963-1967, he was Senior Scientist at the Max-Planck-Institut fur Metallforschung, Stuttgart. During the winter terms 1958/1959 and 1959/1960 he was a guest lecturer at the Technische Hochschule Wien, Austria, and the University Gottingen, Germany, respectively. From 1960-1962 he worked as a consultant at the IBM Thomas J. Watson Research Center, Yorktown Heights, NY. Since 1967 he is Full Professor of Materials Science and Director of the Institute for Electrical Engineering Materials at the Ruhr University, Bochum, Germany. During 1968/1969 and 1987 he was Dean of the Faculty of Electrical Engineering at the Ruhr University. He is the author of Ferromagnerismus, Springer (1962), and coauthor and coeditor with A. E. Berkowitz of Magnerism and Metallurgy, Academic Press (1969). He has contributed tables of magnetic data to Landolr Barnsrein, Vol. 1119, Springer (1962), and an article on fine particle theory to the Encyclopedia of Physics, Vol. XVIII12, Springer (1966). He has also published numerous articles in scientific journals on topics in magnetism, magnetic materials, magnetic recording, metallurgy, and metallic glasses. Professor Kneller is a member of the Deutsche Gesellschaft fur Metallkunde and a Fellow of The Institute of Physics, London, Great Britain. He is a member and presently Vice President of the Rheinisch-Westfalische Akademy of Sciences. Reinhard Hawig was born in Gelsenkirchen, Germany, in 1951. He received the Dip1.-lng. degree and the Dr.-lng. degree in electrical engineering from the Ruhr University, Bochum, Germany, in 1985 and 1990, respectively. Since 1985 he has been with the lnstitut fur Werkstoffe der Elektrotechnik, Ruhr University, Bochum. He is currently working on glass forming alloys and permanent magnet materials. He is a member of the Verband Deutscher Elektrotechniker (VDE).