Study of sintered Nd-FeB magnet with high performance of Hcj ( kOe ) + ( BH ) max ( MGOe ) > 75

We have developed a Nd-Fe-B sintered magnet of extremely high performance. The intrinsic coercivity Hcj is as high as 35.2 kOe (2803kA/m) together with the maximum energy product (BH)max of 40.4 MGOe (321.6kJ/m3). These values result in Hcj (kOe) + (BH)max (MGOe) > 75. Between 293 K (20oC) and 473 K (200oC), the temperature coefficients of remanence and intrinsic coercivity are αBr = −0.122 %/°C and αHcj = −0.403%/°C, respectively. A maximum operating temperature of 503 K (230oC) is obtained when permeance coefficient Pc = −B/H = 2. Grain boundary diffusion (GBD) technique on magnet surface has been developed to increase Hcj by 3.6 kOe without significantly decrease of Br and (BH)max. The intrinsic coercivity of the GBD treated magnet Hcj(C) has a linear relationship with that of the untreated magnet Hcj(B) between 200 K and 473 K (in unit of kOe): Hcj(C) = 1.03Hcj(B) + 2.38. The enhancement of Hcj by GBD treatment has contributions not only from the improvement of microstructure but also from the increas...


I. INTRODUCTION
Invented in 1983, 1,2 Nd-Fe-B sintered magnets possess outstanding magnetic properties.They are widely applied in various fields such as information technology, home appliance, magnetic resonance image and industry.Sustainable development and green energy create new applications in hybrid or electrical vehicle (HEV or EV), energy-saving appliance, wind mill power generator and so on. 3,4 ost of these applications require Nd-Fe-B magnets with high maximum energy product (BH) max and high thermal stability which can operate at elevated temperature beyond 200 • C. 5 Excellent thermal stability depends on high Curie temperature (T C ), low temperature coefficient of remanence (α Br ), high intrinsic coercivity (H cj ) with low temperature coefficient (α Hcj ). 6Practically high H cj is an essential factor.Unfortunately the increase in H cj always accompanies the decrease in (BH) max .Therefore the performance of the magnet should be judged by both H cj and (BH) max , rather than by individual H cj or (BH) max .For sintered Nd-Fe-B magnets the value of Q = H cj (kOe) + (BH) max (MGOe) may be used as an comprehensive index for judging their performance.The higher the value of Q is, the better performance of the magnet will be.The value Q reflects the technological level of the magnet production.Focusing on the issue of how to produce high a Electronic mail: niue@aphy.iphy.ac.cn performance (big value Q) magnets by improving the thermal stability with little sacrifices in B r and (BH) max , many efforts have been devoted and much progress has been made in recent years.Element addition or substitution is a conventional solution.However the merit and demerit are often co-existed for single element.For example, partial substitution of Fe by Co increases T C and decreases α Br but reduces H cj .0][11] However this substitution reduces the saturation magnetization M s and thus B r and (BH) max due to the antiparallel coupling of Fe and Dy/Tb magnetic moments.In addition, element abundance and supply of Dy and Tb is much more limited than Nd. 12 The third solution is to optimize the microstructure of sintered magnets by small amount addition of fusible metals (Al, Cu, Ga) or refractory metals (V, Nb, Mo, etc.) to enhance the H cj .][15][16][17] Combined addition of different elements, including proper oxygen doping, 3,6,18,19 is applied to compromise the demerits and achieve high performance.
Different from the element substitution of Dy or Tb, grain boundary diffusion (GBD) treatments have been widely studied recently.In GBD process oxides, fluorides, alloys or even pure metals of Dy or Tb are supplied on the surface of sintered magnets.1][22] H cj can be raised up to 4 ∼ 5 kOe without obvious reduction of B r .Moreover, the composition of Dy or Tb is much lower than those in conventional alloying process, which may reduce magnet cost dramatically.4][25][26] However, according to Nakamura et al., 26 the H cj enhancement effect of Dy/Tb GBD-treated magnets depends significantly on the diffusion depth in magnets.The H cj of the surface part is higher than the H cj of the inner part due to the content decrease of Dy/Tb from the surface to the center.
Appropriately reducing the grain size of sintered magnet is another way to effectively increase H cj but make little change of B r and (BH) max .Sagawa et al. 4,5,27 applied jet mill with Helium gas as medium to bring the average particle size of Nd-Fe-B powder down to ∼1 μm.To prevent such fine particles from oxidation, they invented a so-called 'press-less pressing (PLP)' technique to compact fine powder into block in air-tight and oxygen free container.The green parts were sintered in high vacuum.They succeeded in making the Dy or Tb-free Nd-Fe-B magnet of intrinsic coercivity of 20 kOe, which is 9 kOe higher than conventional powder metallurgic process.The (BH) max is 48 MGOe.Obviously this procedure has great advantage over conventional sintering and GBD process.But it also confronts a lot of difficulties in realizing mass production, such as oxygen contamination and grain size control.
These aforementioned methods make it possible to realize high performance magnet with H cj (kOe) + (BH) max (MGOe) ≈ 70. 19,28 or high temperature applications, it is necessary to keep H cj higher than 30 kOe and make (BH) max as large as possible.Manufacture process has to be optimized in every detail to eliminate possible factors of (BH) max decrease.In this work the production process has been optimized in order to increase the degree of orientation, to decrease the grain size, and to enhance the uniformity of grain size distribution.GBD procedure of fine Dy alloy powder has also been developed.Sintered Nd-Fe-B magnets with H cj > 30 kOe and H cj (kOe) + (BH) max (MGOe) > 75 are achieved and suitable for high-end applications.The intrinsic magnetic properties, extrinsic magnetic properties and microstructure were investigated to unveil their relationship with technical procedure.

II. EXPERIMENTAL
Raw materials was prepared according to a nominal composition of Pr 2.8 Nd 8.7 Tb 1.9 Dy 0.3 (Cu,Al,Ga) 0.6 Co 1.5 Fe bal B 5.7 .The addition of Co, Tb and Dy could improve the Curie temperature and magnetocrystalline anisotropy field of the Nd 2 Fe 14 B-structure phase in magnet.The dopants of fusible elements of Al, Cu and Ga could positively influence the wetting behavior of the liquid phase during sintering, then enhance the magnetic decoupling of the grains 29 and the corrosion resistance. 30Alloy flakes were prepared by strip-casting technique.The molten alloy was kept at 1400 • C ∼ 1450 • C, and the cooling water at 35 ∼ 40 • C. The rotating rate of casting wheel was set to 1.1 m/s and the thickness of the alloy flakes was in an range of 0.3 ∼ 0.5 mm.The alloy flakes were crushed by hydrogen decrepitation (HD) process into coarse powder of hundreds microns.With nitrogen gas jet mill, the coarse powder was further milled into fine powder with D 50 of 6.63 μm.The particle size distribution of fine powder was characterized by a laser particle size analyzer.Under the protection of nitrogen gas, fine powder was aligned in a constant magnetic field of 18.2 kOe (1.45 MA/m) together with alterative field pulses of peak field 40.2 kOe (3.20 MA/m) parallel or antiparallel to the constant field.The powder was pressed to density of 4.2 g/cm 3 at a pressure of 120 MPa in the direction perpendicular to magnetic field.The green compacts were then isostatically compacted in oil cylinder under a pressure of 200 MPa.Sintering was undertaken in vacuum at 1080 • C for 3 hours to reach full density and quenching to room temperature by Ar gas blower.Post-sintered magnets were heat-treated in two stages at 900 • C for 2 hours and 500 • C for 3 hours, respectively.This kind of magnet is called "reference magnet" or "Specimen A".
To improve the performance of the "reference magnet", the sample preparation routine was optimized in strip casting, jet milling and sintering.Started from the same nominal composition, kept the same temperature range of molten alloy and cooling water as the 'reference magnet', alloy strip casting was conducted at rotating rate of 1.5 m/s using a specially designed copper wheel with the diameter 610 mm.The thickness of the alloy flakes was in a range of 0.2 ∼ 0.4 mm.The jet milling was also optimized to narrow the particle size distribution with D 50 of 5.28 μm.To prevent grains from abnormal growing the sintering was at lower temperature of 1020 • C for longer time of 6 hours, compared with the sintering condition of Specimen A. The same two-stage heat-treatment as the Specimen A was taken after sintering.The magnets prepared by this routine are referred as "optimized magnets" or "Specimen B".
The third way of sample preparation was in the same routine as that for "optimized magnet" except an additional grain boundary diffusion (GBD) treatment.The post-sintered magnets were cut into cylinders of dimensions 10.0 × 10.0 mm 3 or blocks of 50.0 × 48.0 × 3.8 mm 3 .The cylinders and blocks were coated with Dy alloy powders of several microns in thickness and then were heated at 750 • C ∼ 1000 • C for 20 min to 6 h, which was regarded as GBD treatment.Afterwards, the magnets were annealed in the same two-stage condition as Specimen A and Specimen B. The magnets prepared by this routine are referred as "GBD treated magnets" or "Specimen C".
The chemical composition was determined by inductively coupled plasma (ICP) spectrophotometry.The contents of oxygen, nitrogen and hydrogen were determined by an ONH analyzer (ELTRA ONH2000).
The phase identification and the microstructure were analyzed by X-ray diffraction (XRD) and scanning electron microscope (SEM).
The magnets samples were sliced into cubic of 1.5 × 1.5 × 1.5 mm 3 for measuring magnetization.A Lakeshore 7410 vibrating sample magnetometer (VSM) was used to measure the temperature dependence of magnetization (M) in an applied field of 300 Gs (23.9 kA/m) to determine the Curie temperature T C .The magnetization curves were measured by a superconducting quantum interference device (SQUID) VSM with the maximum field of 70 kOe (5.57MA/m).
Because of the pole-piece saturation effect in close-loop measurement, the room temperature demagnetization curves were measured by a pulse field magnetometer PFM12.cn with the peak field up to 100 kOe (7.96 MA/m).The demagnetization curves of specimens from 120 • C to 200 • C were measured by a Magnet-Physik permagraph C-300.Samples for close-loop measurements were 10.0 × 10.0 mm 3 cylinders.The temperature-dependent irreversible flux losses were deduced from the magnetic flux differences of the magnets at 20 • C and a fixed temperature (up to 250 • C) for 2 hours.The magnetic flux was measured by Lakeshore Helmholtz coil and fluxmeter.The permeance coefficients (P c = −B/H) of samples were chosen as 0.5, 1.0 and 2.0, respectively, by cutting samples from blocks with the same thickness of 3.8 mm and different diameters.The maximum operating temperature for a magnet with a certain P c value is defined by the temperature point at which the absolute value of irreversible flux loss is 5%.The corrosion resistance is characterized by the weight loss measured after the accelerated aging test.A pressure cook test (PCT) chamber was employed under the test condition of 120 • C and 100%RH (relative humidity) with condensation of water vapor (according to JEDEC standard: JESD22-A102D, Accelerated Moisture Resistance-Unbiased Autoclave).  up to 200 • C. Contrastively, both Specimen A and B have knee points on their B-H curves at this temperature (see Fig. 1(b)).

A. Extrinsic magnetic properties and thermal stability of magnets
From Table I and Fig. 2 it can be recognized that all the extrinsic magnetic parameters decrease with the increase of temperature.The difference of B r between Specimen A and Specimen B keeps at 0.1 ∼ 0.2 kGs (see Fig. 2(a)).The (BH) max of Specimen B are 0.4 ∼ 0.6 MGOe higher than Specimen A (Fig. 2(c)).The B r of Specimen C is nearly the same as Specimen A in the whole temperature range.Because of the improvement of squareness in demagnetization curve, the (BH) max of Specimen C is slightly higher than Specimen A. Obvious change occurs on the intrinsic coercivity H cj .With optimized routine H cj of Specimen B enhance by 1.1 ∼ 1.8 kOe compare to Specimen A. The GBD treatment adds another 2.2 ∼ 3.6 kOe to H cj (see Fig.  C),where the magnetic flux (T) was measured at room temperature after the magnet exposed at high temperature T for 2 hours.It can be seen that all the absolute values of irreversible loss are less than 3% below 200 • C, no matter what the treatment and P c are.Therefore even the samples of optimized magnet with P c no less than 0.5 can be stably used at 200 • C. At higher temperature, the absolute values of irreversible losses increase drastically with decrease of P c .The samples of optimized magnet show larger loss comparing to the samples of GBD treated magnet of the same P c at the same temperature.The maximum working temperature for a magnet with a certain P c value is usually defined as the temperature at which the absolute value of irreversible loss is 5%.It can be recognized from Fig. 3 that the maximum working temperatures of Specimen C are 5 ∼ 10 • C higher than Specimen B. In other words GBD treated magnets have better thermal stability than optimized magnets.The maximum working temperature of GBD treated magnet with P c = 2.0 is higher than 230 • C, which means that GBD magnets may meet the demands of many applications such as hybrid or electrical vehicle (HEV or EV), energy-saving appliance, wind mill power generator and so on.

B. Microstructure of magnets and related alloys
Large field images observed by optical microscope are shown in Fig. 4 for reference magnet (Fig. 4(a)), optimized magnet (Fig. 4(b)) and GBD treated magnet (Fig. 4(c)).The images were taken on the surface perpendicular to magnetic orientation direction.From Fig. 4(b) and 4(c) it can be estimated that the grain sizes of matrix phase are from 2 to 15 μm with an average of 7 μm.There is no abnormally growing large grain and even no two grains join together.The net-like black lines of etched RE-rich phase obviously indicate that fine matrix grains are well separated from each other.This metallurgic structure is essential for Nd-Fe-B magnets to achieve high coercivity.For comparison, the image of reference magnet (Fig. 4(a)) shows an average grain size of 10 μm.The notable difference of Fig. 4(a) to Fig. 4(b) and 4(c) is the existence of 20 μm large grains and sub-micron fine grains.Abnormally grown large grains in reference magnet are responsible for the relatively low H cj .In addition, the very fine grains will decrease both B r and H cj because of their strong tendency of oxidation and misalignment.
Several efforts have been made in optimized manufacturing routine to result in such differences.The technique of strip casting is helpful for the starting R-Fe-B alloy to uniformly disperse RE-rich phase and depress the growth of α-Fe phase. 32With optimized structure of tundish the temperature in melt pool keeps homogeneous during alloy cast. 33Together with optimized wheel spinning rate an ideal microstructure of strip-casting alloy have been achieved.shows that the ultra-fine zone contains submicron isometric grains of matrix phase.During jet milling some of these submicron isometric crystals may keep in multi-crystalline status which is proven to be harmful to powder orientation. 34The remanence perpendicular to the orientation direction B r ⊥ was measured on FIG. 6. Particle size distribution of conventional crushing process and optimized process.
Specimen A and B magnets.They are 0.89 kGs and 0.80 kGs, respectively.Dividing B r ⊥ by B r in Table I one can get the ratio of B r ⊥ / B r as 7.0% for reference magnet and 6.2% for optimized magnet.The degree of orientation in reference magnet is slightly distorted.Other submicron isometric crystals may be crushed into sub-micron single-crystalline particles.Then they are eliminated into dust collector.The total rare earth composition will decrease and the liquid-phase sintering process will be negatively affected. 34The ICP analysis revealed the composition results.The total rare earth compositions are 30.95wt.% for optimized magnet and 30.60 wt.% for reference magnet, even though they started from the same nominal composition.Therefore the metallurgical structure of strip-casting alloy plays a critical role in optimizing the magnetic properties.
Narrow distribution of fine grains in optimized magnet is also ensured by optimized crushing process.Fig. 6 compares the particle size distribution of optimized and conventional crushing process.The optimized crushing process leads to a very narrow particle size distribution.The measurement of particle size distribution reveals that the D 50 is 5.28 μm for optimized crushing process, while 6.63 μm for the conventional process.The very large particles may lead to abnormal grain growth during sintering.The over-grown grains will damage the coercivity and the squareness of demagnetization curve.It is worth noting that the D 50 is comparable to the width of the columnar grains in alloy strips.Together with low temperature and long time sintering, the optimized magnet with fine grains is realized to ensure high coercivity.
Normally a two-step annealing at temperature of 900 • C for 2 hours and 500 • C for 3 hours was applied to the post-sintered magnets.Fig. 7(a) and Fig. 7(b) compare the microstructures of magnets before and after annealing, respectively.In as-sintered magnet (Fig. 7(a)) patches of RE-rich phases can be seen to concentrate in triple-junctions and corners of main phase grains.It is hard to recognize any thin RE-rich phase penetrating into two-grain boundary.However, in two-stage annealed magnet (Fig. 7(b)) there are continuous thin RE-rich phases along the grain boundaries with scattered RErich concentrated junctions.After GBD treatment, more distinct, continuous and thinner RE-rich phase between the grain boundaries can be seen in the magnet (Fig. 7(c)).Apparently, two-stage annealing and the GBD treatment uniformly disperse the RE-rich phases into grain boundaries to completely isolate the matrix phase grains.
During manufacturing process O, H, or N contamination is inevitable and sensitive to magnetic properties.Oxygen content in strip-casting alloy flake was determined by ONH analyzer as 100 ppm.In optimized magnet and GBD treated magnet the oxygen contents increased to be 1000 ppm.The hydrogen content is about 1.3 ppm and the nitrogen content is less than 110 ppm.The X-ray diffraction pattern of grinded magnet shows a Nd 2 Fe 14 B-structure phase together with a tiny hint of R 2 O 3 phase.Supposing all oxygen atoms exist in form of R 2 O 3 , it can be deduced that there is about 0.71 wt.% of R 2 O 3 in magnets.Over dosage of oxygen contamination is considered to be harmful for forming smooth thin RE-rich phase, thus decreases the coercivity dramatically. 35ig. 8(a) shows the metallographic image of optimized magnet on the surface perpendicular to magnetic orientation direction.The image is binarized to whiten the Nd 2 Fe 14 B-structure main phase and leave the grain boundary as black (Fig. 8(b)).Based on the area method of the Quantitative Metallographic Analysis (QMA) system, together with the software Image-Pro Plus (IPP), the area ratio of the main phase are estimated.Three different zones of 0.6 × 0.5mm 2 have been selected and analyzed under the magnification of 500×.The area ratios of main phase are 94.6%, 94.9% and 94.6%, respectively.The average value is 94.7%.With this area ratio the main phase grains are well separated by black net lines.The RE-rich phase is sufficient to form nonmagnetic intergranular phase to completely isolate matrix phase.If there were too much oxygen, the R-rich phase would be insufficient.Neighbor grains of main phase have tendency to join each other directly.This phenomenon is equivalent to abnormal grown large grains in magnet which decreases intrinsic coercivity.

C. Intrinsic magnetic properties and intrinsic coercivity before and after GBD treatment
The intrinsic magnetic properties of the rare-earth transition-metal intermetallic compounds (R 1 ,R 2 ,. . .R i ,. . .,R n ) 2 (T 1 ,T 2 ,. . .T j ,. . .T m ) 14  of Fe is 1.37 at.%.Based on the T C -x curve of Curie temperature T C against Co content x in Nd 2 (Fe 1-x Co x ) 14 B, 8,16,36 the T C of the strip-casting alloy should be 324 • C. For comparison of prediction and measurement, the density of magnet has to be obtained first.From X-ray diffraction the lattice parameters of Nd 2 Fe 14 B-structure main phase for all specimens have the same values of a = 0.879 nm and c = 1.218 nm.It indicates that GBD treatment does not affect main phase.If taking the lattice parameters of individual R 2 (Fe,Co) 14 B intermetallic compound (R = Pr, Nd, Tb, Dy) in accordance with the composition of Pr 0.41 Nd 1.27 Tb 0.28 Dy 0.04 Co 0.26 Fe 13.74 B, the lattice parameters of a = 0.879 nm and c = 1.216 nm may be obtained, which are well agreed with the experimental results.From the room temperature measurement, we obtained the density of optimized magnets and GBD treated magnets.They have the same value of ρ m exp = 7.621g/cm 3 .The magnetization curves of an optimized magnet (Specimen B) and a GBD treated magnet (Specimen C) were measured in the applied fields up to 70 kOe along the easy magnetization direction (parallel to the magnetic orientation direction) and the hard direction (perpendicular to the magnetic orientation direction).The magnets were cubic with side length of 1.5 mm.And one edge was chosen as parallel to the orientation direction.Fig. 9 gives the typical magnetization curves at temperature of 20 • C. The open circuit measurements were corrected by a demagnetization factor N = 0.26. 37The saturation polarization 4π M s is deduced from the magnetization curve along easy direction with density of ρ m exp = 7.621 g/cm 3 .The 4π M s is 13.3 kGs for the optimized magnet and 13.2 kGs for the GBD treated magnet.These values are slightly smaller than that of the strip-casting alloy flake and the predicted value.
The magnetocrystalline anisotropy field H a is estimated from the cross point of extension lines of the 4π M-H curves along easy and hard directions (Fig. 9).In both directions, the polarization 4π M of the GBD treated magnet is slightly smaller than that of the optimized magnet.The magnetocrystalline anisotropy field H a has almost the same value of 104 kOe for both magnets.This value is higher than the predicted value of 99.7 kOe from composition.The misalignment of main phase grains of the magnet may affect the extrapolation of magnetization curve along hard magnetization direction.
The measured value of Curie temperature T C is nearly the same as 334 that the optimized magnet and the GBD treated magnet have almost the same values of intrinsic magnetic properties.In other words, the GBD treatment has little effect on the intrinsic magnetic properties of the main phase of the magnets.Therefore the intrinsic coercivity is enhanced by GBD treatment while the remanence and the maximum energy product almost keep unchanged, which is one of the outstanding advantages of our GBD technique.If the same amount of H cj enhancement is expected, the conventional process may need more Dy to replace Nd.The reduction in B r is estimated to be 0.8 ∼ 1.0 kGs 40,41 rather than 0.1 to 0.2 kGs.
The intrinsic coercivity may be expressed by the following model: [42][43][44] The first term on the right-hand side reflects the contribution of the magnetocrystalline anisotropy field H a associated with the microstructure effect represented by parameter α.The second term 'N eff M s ' is an effective magnetostatic interaction, where N eff is effective demagnetization parameter which mainly depends on the shape and size of the grains of main phase.Here the parameters α and N eff are purely phenomenological and are usually determined by plotting H cj /M s versus H a /M s at different temperatures.
H cj /M s varies with H a /M s at different temperatures is plotted in Fig. 10, which is similar to the H cj /M s -H a /M s plotting on Pr-Fe-B magnet by Sagawa et al. 43 It can be seen that H cj /M s has a linear relationship with H a /M s for both the optimized magnet and the GBD treated magnet below 340 K.By linear fitting the data between 200 K and 340 K, we obtain (in unit of kOe): for the optimized magnet, and for the GBD treated magnet.Our α values are about 2.3 times larger than 0.36 on the Pr-Fe-B magnet achieved by Sagawa et al., 43 which indicate that the microstructures in our magnets are very much improved to transfer H a to H cj .The H cj (C) of the GBD treated magnet as a function of the H cj (B) of the optimized magnet at different temperatures is shown in Fig. 11.It can be seen that the H cj (C) has a very good linear relationship with the H cj (B) within the whole temperature range of measurement.By linear fitting the data, we obtain (in unit of kOe): By bringing equations (2) into equation ( 4) and taking M s (B) ≈ M s (C) into consideration, we have (in unit of kOe): Equation ( 5) is slightly different from equation (3).It has been observed that the modification of coercivity on Nd-Fe-B magnet by the GBD process is sensitive to rare-earth oxides with different rare earths. 22The Tb oxide is the best to increase the H cj due to Tb 2 Fe 14 B with the highest uniaxial anisotropy, while the Sm oxide drastically decreases the coercivity due to Sm 2 Fe 14 B with strong planar anisotropy.This indicates that although the magnetocrystalline anisotropy field H a keeps unchanged on the entire Nd 2 Fe 14 B-structure main phase during the GBD process, but the H a of the surface layers of the main phase grains is influenced.By the GBD treatment, the enhancement of intrinsic coercivity H cj is caused not only by the improvement of microstructure but also by the increase of H a in the grain surface layer.Suppose each main phase grain in sintered magnet is composed of kernel and surface layer.The kernel and surface layer have different intrinsic magnetic properties, especially the magnetocrystalline anisotropic fields.And the GDB treatment only affects the properties of surface layer.Then we may write the following relationship for H cj (C) (assuming M s (B) ≈ M s (C)): Re-writing equation ( 5) in the form of equation ( 6) and taking the coefficients of α(B) and N eff (B) from equation (2), one can get α = 0.025, H a = 2.80 kOe, and N eff = 0.13.The H cj enhancement seems to be from two parts respectively: the microstructure improvement by 3% ( α/α(B)) and the H a increase of the surface layers of the main phase grains by 2.80 kOe.And the H cj is weakened by the magnetostatic interaction increase by 3% ( N eff / N eff (B)).Improving H cj is an important way to increase thermal stability of magnets. 6Contradicting to the large magnetocrystalline anisotropy field H a of Nd 2 Fe 14 B-type main phase, the intrinsic coercivity H cj of sintered magnet is far lower than H a .For example, at room temperature the H a is 76 kOe for Nd 2 Fe 14 B compound.But for sintered Nd-Fe-B magnet without heavy rare earth, the corresponding intrinsic coercivity H cj is only 11 kOe for conventional manufacture routine.The ratio of H cj to H a

D. Corrosion resistance before and after GBD treatment
The corrosion resistance of magnets has been examined by Pressure Cooker Test (PCT) for 96 hours and 192 hours under the condition of 120 • C, 100%RH relative humidity, with condensation of water vapor (according to JEDEC standard: JESD22-A102D, Accelerated Moisture Resistance-Unbiased Autoclave).The weight loss of a sintered magnet is defined by WL = (W 1 −W 0 )/S 0 in which W 0 and W 1 are the weight of the sample before and after test, and S 0 is the surface area of the sample before test.It was found that both magnets with and without GBD treatment have little difference on weight loss (see Table III).It was also found that the weight loss on the two pole surfaces (c-axis as the normal direction) is much bigger than that of the side surfaces (in which the c-axis lies).This anisotropic corrosion phenomena was also observed by others. 45,46 o, the weight losses of unit surface area from PCT were calculated in two ways: weight loss divided either by total surface area or only by area of two pole surfaces.The absolute value of weight losses for both samples are ≤3.2 mg/cm 2 based on the total surface, and ≤6.2 mg/cm 2 based on the two pole surface.The latter is almost doubled.The PCT result indicates that our GBD treatment has no negative effect on the corrosion resistance of the magnets.

IV. CONCLUSION
In this work, we have developed a new procedure to significantly improve H cj and keep a high maximum energy product of sintered Nd-Fe-B magnet.This new procedure is an optimized routine compare to the conventional one.It consists of an optimized composition design, optimized strip casting, jet milling and low temperature sintering, especially an excellent grain boundary diffusion (GBD) technique.
A magnet with remarkably high performance has been achieved with intrinsic coercivity H cj as 35.Taking the measurement uncertainty into consideration, the intrinsic properties of 4π M s , H a and T C values of the GBD treated magnet are nearly the same as the optimized magnet without GBD treatment.It indicates that our new GBD process has no obvious effect on the Nd 2 Fe 14 B-structure main phase in the magnets.The Dy diffusion mainly affects the microstructure of the optimized magnet and increase the magnetocrystalline anisotropic field in surface layer of main phase grain to enhance the intrinsic coercivity H cj without significant reduction of the remanence B r and the maximum energy product (BH) max .Our new GBD method also brings no deterioration in corrosion resistance of magnet.

FIG. 1 .
FIG. 1. Demagnetization curves of (a) 4π M-H at 20 • C and 200 • C, (b) 4π M-H and B-H at 200 • C of reference magnet (Specimen A, the dot line), optimized magnet (Specimen B, the dash line) and GBD treated magnets (Specimen C, the continuous line) at 20 • C and 200 • C. (Close-loop measurement and sample dimension: 10.0 × 10.0 mm 3 )

Fig. 1 (
Fig. 1(a) shows the 4π M-H demagnetization curves of reference magnet (Specimen A), optimized magnet (Specimen B) and GBD treated magnet (Specimen C) at temperatures of 20 • C and 200 • C, respectively.Fig. 1(b) shows B-H curves together with 4π M-H curves at 200 • C. The extrinsic magnetic parameters between 20 • C and 200 • C are summarized in Table I and Fig. 2. Comparing to Specimen A, Specimen B with optimized routine has its (BH) max increased by 0.6 MGOe and H cj by 1.2 kOe at 20 • C. According to a high magnetic performance description, by adding data of H cj and (BH) max in Gaussian unit we can get H cj (kOe) + (BH) max (MGOe) = 72.5.The Specimen A only has this value of 70.7.Obviously the optimized routine significantly improves the performance of the magnet.In addition, Specimen C (GBD treated magnet) has B r of 12.8 kGs (1.28 T), H cj of 35.2 kOe (2803 kA/m) and (BH) max of 40.4 MGOe (321.6 kJ/m 3 ).The B r decreases slightly compare to Specimen B. The H cj increases drastically with 3.6 kOe by GBD treatment which brings the value of H cj (kOe) + (BH) max (MGOe) up to 75.6.At 200 • C the B-H curve of GBD treated magnet has no obvious knee point.This implies that the GBD treated magnets can be applied at temperature 2(b)).Subsequently the optimized routine has positive effect to increase the sum of H cj (kOe) and (BH) max (MGOe) by 1.6 ∼ 2.4.Meanwhile the GBD treatment improves the sum of Specimen C by 2.1∼3.1 within the whole temperature range compare to Specimen B. The temperature coefficient of B r is defined as α Br = (B r (T) − B r (T 0 ))/(B r (T 0 )(T − T 0 )).The corresponding temperature coefficient of H cj is α Hcj = (H cj (T) − H cj (T 0 ))/(H cj (T 0 )(T − T 0 )).Between 20 • C and 200 • C the temperature coefficients of B r and H cj are α Br (B) = −0.124%/ • C and α Hcj (B) = −0.425%/ • C for optimized magnets.These values become α Br (C) = −0.122%/ • C and α Hcj (C) = −0.403%/ • C for GBD treated magnets.While α Br is nearly the same value, α Hcj is reduced by about 5%.This indicates that the GBD treatment enhances the intrinsic coercivity efficiently at all measured temperatures so that temperature coefficients of H cj have improved.The GBD treatment has little effect on the temperature coefficient of remanence.Fig. 3 and Table II illustrate the irreversible flux loss of the magnets of Specimen B and C for different permeance coefficients P c = −B/H = 0.5, 1.0, 2.0 after exposure at the temperatures of 26 • C, 120 • C, 150 • C, 180 • C, 200 • C, 220 • C, 230 • C, 240 • C and 250 • C for 2 hours, respectively.The irreversible flux loss 31 is defined by ( (T) − (20 • C))/ (20

FIG. 2 . 1 FIG. 3 .
FIG. 2. The temperature dependence of (a) remanence B r , (b) intrinsic coercivity H cj , (c) the maximum energy product (BH) max and (d) H cj (kOe) + (BH) max (MGOe) of reference magnet (Specimen A, the dot line with triangle symbol), optimized magnet (Specimen B, the dash line with square symbol) and GBD treated magnet (Specimen C, the continuous line with circular symbol).

FIG. 4 .
FIG.4.Optical microscopic images with the normal direction parallel to magnetic orientation direction of (a) reference magnet, (b) optimized magnet, (c) GBD treated magnet.

Fig. 5 (
a) and Fig.5(b)show the SEM back scattered electron images (BSE) of strip-casting alloy prepared by optimized process and conventional method, respectively.It can be seen that the thicknesses of the strip-casting alloy are 0.2 ∼ 0.4 mm.The columnar grains (the dark areas) are about 3 ∼ 5 μm in width and well separated by the uniformly distributed RE-rich phase (the white lines).Across the thickness direction from the wheel-contact surface to the free surface, the grains of optimized process are uniformly distributed (Fig.5(c) and Fig.5(e)).Different from the optimized strip-casting process, in conventional stripcasting alloy there are ultra-fine grain zones on the wheel-contact surface (areas with white elliptic mark in Fig.5(b) and Fig.5(d)).Further enlarged image (Fig.5(f))

FIG. 7 .
FIG. 7. BSE images of magnets with the normal direction parallel to magnetic orientation direction, (a) optimized magnet before annealing, (b) optimized magnet after annealing, (c) GBD treated magnet.

FIG. 8 .
FIG.8.The metallography images of the cross-section with the normal direction parallel to magnetic orientation direction before binarization (a) and after binarization (b).

FIG. 9 .
FIG.9.Initial magnetization curves of directions along the easy magnetization direction and the hard magnetization direction for the optimized magnet and the GBD treated magnet.
FIG.10.H cj /M s varies with H a /M s for the optimized magnet (Specimen B) and the GBD treated magnet (Specimen C) at different temperatures.

FIG. 11 .
FIG. 11.The intrinsic coercivity of the GBD treated magnet H cj (C) versus the intrinsic coercivity of the optimized magnet H cj (B) at different temperatures.
2 kOe (2803 kA/m) and maximum energy product (BH) max as 40.4MGOe (321.6kJ/m 3 ) at 20 • C.They give H cj (kOe) + (BH) max (MGOe) > 75.The temperature coefficients of remanence and intrinsic coercivity between temperature of 20 • C and 200 • C are α Br (C) = −0.122%/ • C and α Hcj (C) = −0.403%/ • C, respectively.A maximum operating temperature of 230 • C is obtained with permeance coefficient P c = −B/H = 2.It is found that the intrinsic coercivity of the GBD treated magnet H cj (C) has a linear relationship with that of non-treated magnet H cj (B) between 200 K and 393 K: H cj (C) = 1.03H cj (B) + 2.38.The enhancement of H cj by GBD treatment has contributions not only from the improvement of microstructure but also from the increase of H a in the grain surface layer.
The H cj value of GBD treated magnet at the temperature of 100 • C is beyond the measuring range of the Magnet-Physik permagraph C-300. a

TABLE III .
4eight loss of PCT (120 • C, 100%RH relative humidity, with condensation of water vapor).isHcj /H a = 14.5%.Many activities have been focusing on narrowing this gap and increase the ratio.By reducing the jet milling particle size to 1μm, the H cj can reach 20 kOe and H cj /H a = 26%.4Inthis work, at temperature of 20 • C, the H cj /H a = 31.6kOe/104kOe= 30.4% for the optimized magnet and H cj /H a = 35.2kOe/104kOe= 33.8%for GBD treated magnet.The advantage of our GBD process is to enhance the intrinsic coercivity about 11.4%, but only to reduce the remanence about 1.5%.