PII:

Acta mater. Vol. 47, No. 17, pp. 4427±4434, 1999 # 1999 Acta Metallurgica Inc. Published by Elsevier Science Ltd. All rights reserved. Printed in Great Britain S1359-6454(99)00317-1 1359-6454/99 $20.00 + 0.00

TEM STUDY ON THE ORIGIN OF CABBAGE-SHAPED MICA CRYSTAL AGGREGATES IN MACHINABLE GLASSCERAMICS A. GEBHARDT{ 1, T. HOÈCHE 1{, G. CARL 1 and I. I. KHODOS 2 Otto-Schott-Institut fuÈr Glaschemie, Friedrich-Schiller-UniversitaÈt, Fraunhoferstraûe 6, D-07743 Jena, Germany and 2Institute of Microelectronics Technology and High Purity Materials, Russian Academy of Sciences, 142432 Chernogolovka, Russia

1

(Received 22 June 1999; accepted 26 July 1999) AbstractÐIn micaceous glasses cabbage-shaped mica crystal aggregates are formed at certain compositions and distinct heat-treatment schedules. It is experimentally proved that a liquid±liquid phase separation is responsible for this peculiar behaviour. Based on transmission electron microscopy observations of glasses and glass-ceramics, a novel growth mechanism is proposed for cabbage-shaped mica crystal aggregates in machinable mica glass-ceramics. In this model, crystallization takes place along isocompositional lines at or not far from the surface of the liquid±liquid phase separation droplets. Due to high growth rates, di€usion has only little impact on the crystal growth. Finally, experimental reasons for earlier misinterpretations are discussed. # 1999 Acta Metallurgica Inc. Published by Elsevier Science Ltd. All rights reserved. ZusammenfassungÐBei ausgewaÈhlten GlaÈsern aus dem Glimmersystem werden unter bestimmten Temperbedingungen kohlkopfartige Glimmeraggregate beobachtet. Wir erbringen den experimentellen Nachweis, daû eine FluÈssig±¯uÈssig-Entmischung fuÈr dieses PhaÈnomen verantwortlich ist. Auf der Grundlage von transmissionselektronenmikroskopischen Untersuchungen an GlaÈsern und Glaskeramiken, schlagen wir ein neues Wachstumsmodell fuÈr die kohlkopfartigen Glimmeraggregate in maschinenbearbeitbaren Glimmerglaskeramiken vor. Demzufolge erfolgt die Kristallisation entlang von Linien gleicher Zusammensetzung direkt an oder aber unweit der Ober¯aÈche der EntmischungstroÈpfchen. Aufgrund hoher Wachstumsraten haben Di€usionsvorgaÈnge einen zu vernachlaÈssigenden Ein¯uû auf das Kristallwachstum. Schlieûlich werden die experimentellen Ursachen fuÈr fruÈhere Fehlinterpretationen beleuchtet. # 1999 Acta Metallurgica Inc. Published by Elsevier Science Ltd. All rights reserved. Keywords: Electron di€raction; Transmission electron microscopy; Biomaterials, bone; Microstructure; Phase transformations, nucleation, growth

1. INTRODUCTION

In the early 1980s, cabbage-shaped mica crystal aggregates in machinable glass-ceramics were reported [1, 2]. Later, the particular arrangement of ¯uorophlogopite crystals was found to result in a considerable improvement of the machinability of the glass-ceramic material as compared with glassceramics containing the commonly found mica platelets [3]. Although an extended survey of the aggregates phenomenology will be given elsewhere [4], essential points shall be recalled brie¯y. Cabbage-shaped mica crystal aggregates are formed from certain glass compositions only and are closely connected with distinct heat-treatment schedules. The structure of the aggregates generally consists of a central amorphous sphere surrounded by a number of sub{Present address: Vitron Spezialwerksto€e GmbH, Otto-Schott-Straûe 13, D-07745 Jena, Germany. {To whom all correspondence should be addressed.

sequent cabbage leaves. The latter are separated by glassy interspaces (cf. Fig. 1). Since its discovery, the phenomenon has attracted considerable attention resulting in attempts to interpret the ®ndings in terms of a growth mechanism. The most prominent idea [5, 6] assumes a bent lattice due to deviations from the ideal stoichiometry. This is since cabbage-like aggregated mica crystals 3+ proved to be always enriched in Al as compared with stoichiometric phlogopite. It is known [7] that surplus aluminium can substitute for both silicon (located inside the tetrahedral layer of the phlogopite crystal structure) and magnesium (in the octahedral layer). The mis®t between the tetrahedral and the octahedral layer (see, e.g. Ref. [8]) which can usually be accommodated by a certain rotation of tetrahedra [9] is considered to be no longer sucient. In the picture of Elsen et al. [5, 6], therefore, mica crystals develop bent lattice planes upon reducing stresses evolving from the layer mis®t. On the one hand, this unconventional hypothesis is subject to a number of critical points. First, it is

4427

4428

GEBHART et al.: CRYSTAL AGGREGATE

generally considered unusual that crystals respond to evolving stresses during growth by becoming bent. This argument particularly applies to nonpolar crystal structures and phlogopite belongs to this group. Second, radii of bending (> 1 mm for cabbage-like mica aggregates) are at least two orders of magnitude larger than observed for instance with chrysotile (maximum bending radius < 10 nm) [10]. Third, it is dicult to imagine how long-range order information on sense and radius of bending shall be transferred. On the other hand, this model is capable of describing the fact that at elevated heat-treatment temperatures plane crystals are forming from glasses of identical composition. It is assumed that 2+ with rising temperature the mobility of Mg ions is increasing, resulting in crystals of a composition closer to stoichiometric phlogopite than at lower temperature. However, bending radii of mica aggre-

gates are also sensitively dependent on heating rates or more precisely on the length of stay within certain temperature ranges and Elsen et al.'s model can hardly explain this ®nding. In the present contribution we present our most recent results on the phase separation behaviour of glasses from which cabbage-like aggregated mica crystals are formed. Our ®ndings prove that a liquid±liquid phase separation takes place in these glasses. Only, the droplet phase has not been recognized until now since the traditional manner to study phase separations by replica ®lms in the TEM is not applicable to this particular materials system. 2. EXPERIMENTAL

2.1. Preparation of glasses and glass-ceramics Glass batches of the composition given in Table 1 (glasses 1, 2, and 3) were obtained by melting the

Fig. 1. SEM microscopy of a cabbage-shaped mica crystal aggregate. In addition to the amorphous central part, the separation of the mica ``leaves'' is clearly visible.

GEBHART et al.: CRYSTAL AGGREGATE

4429

Table 1. Nominal composition of the glasses studied

Glass Glass Glass Glass

1 2 3 4

Na2O (mol%)

K2O (mol%)

MgO (mol%)

Al2O3 (mol%)

SiO2 (mol%)

TiO2 (mol%)

Fÿ (mol%)

3.2 6.4 ± 4.1

3.2 ± 6.4 3.1

17.4 17.4 17.4 16.8

16.0 16.0 16.0 16.9

48.7 48.7 48.7 42.9

± ± ± 3.5

11.5 11.5 11.5 12.7

raw materials at 14508C in a Pt crucible for about 1.5 h. After an additional hour of homogenization at 15008C, the melts were fritted and remelted at 15508C for another 2 h. Finally, the melts were cast into heat-resistant steel moulds followed by slow cooling from 6808C to room temperature at about 10 K/min. During the above-mentioned heat treatments, a loss of ¯uorine (about 30%) occurs. The glass-ceramic material studied was obtained by crystallizing glass 1 for 2 h at 9008C using heating and cooling rates of about 10 K/min.

As a reference substance in which straight mica platelets are observed only after crystallization, glass 4 (which is very similar to glasses 1±3 except for the addition of some mol% titanium) was included (see Table 1). 2.2. Microstructural characterization Transmission electron microscopy (TEM) replica ®lms were prepared as detailed in Ref. [11] both from freshly broken and chemically etched glasses. Besides the commonly used 5% HF, ethylenediami-

Fig. 2. TEM bright-®eld micrograph showing an enlarged section of a cabbage-shaped mica crystal aggregate as shown in Fig. 1. Plane mica segments of some 200 nm width are interconnected by smallangle tilt boundaries.

4430

GEBHART et al.: CRYSTAL AGGREGATE

netetraacetate (EDTA) solution (sensitive to di€erences in the Mg content) and Al[NO3]3 solution ÿ (sensitive to di€erences in F content [12]) were used to generate a detectable surface topology. Furthermore, 3 mm disks of glasses 1±4 were prepared in close analogy to ordinary TEM samples as detailed below with the di€erence that ion-beam etching until perforation was performed only on one side at an incidence angle of 158 avoiding the oscillating motion of the sample (2p turn). From bulk glass-ceramic blocks, about 0.5 mm thick slabs were cut using a diamond wire saw, subsequently plane-parallel ground to 0100 mm thickness, polished on one side and dimpled to a residual thickness of 10±15 mm from the unpolished side. + Finally, double-sided Ar beam etching at an acceleration voltage of 2.5 kV and an ion-beam current of 1.5 mA using low incidence angles (68) was

applied until perforation of the foil was just achieved. A Hitachi H-8100 II transmission electron microscope operating at 200 kV acceleration voltage was used for TEM characterization of the microstructure. EDX analyses were performed using an attached Link ISIS analyser (Oxford Instruments) equipped with an atmospheric thin window. 3. RESULTS

Our detailed TEM study of cabbage-shaped mica crystal aggregates in machinable glass-ceramics revealed a previously unreported ®nding: as depicted in the TEM bright-®eld micrograph (Fig. 2), individual mica crystals are not bent. Instead, they consist of a series of equally spaced smallangle tilt boundaries separating mica segments of

Fig. 3. TEM bright-®eld micrograph proving the change of lattice orientation with position in the mica ``leaf'' (insets: selected-area di€raction patterns obtained from areas indicated, systematic 00l rows pointing in the radial direction are indicated by arrows).

GEBHART et al.: CRYSTAL AGGREGATE

4431

Fig. 4. TEM replica ®lm of glass 3 etched with a 1:1 mixture of 5% HF and 20% HNO3 for 10 s. Note that the glass is only slightly rougher than the MoO3 reference face in the upper right.

slightly di€erent orientation. This is also evident from the discretely rotated series of 00l re¯ections to be seen in Fig. 3. As determined by selected-area electron di€raction, adjoining segments di€er by some 2±2.58. Taking into account average width of the segments (200±300 nm) and their respective radial position in the cabbage-like aggregates, the diameter of enclosed spheres can be estimated to be about 5 mm. The earlier impression of bent crystals arose from spatially badly resolved images of a large number of such segments summing up to a virtually homogeneously bent ``cabbage leaf'' (cf. Fig. 1). The direct consequence of this new experimental ®nding is that mica crystals are forced by external conditions to not grow straight, since the generation of such a high density of grain boundaries drastically increases the lattice energy as compared with straight-grown crystals. Hence, crystals would never adopt such a growth form if not forced to do so. On the other hand, assuming the growth direction to follow a curve, the generation of dislocations (®rstly randomly distributed) allows the reduction of elastic stresses that would exist without dislocation generation. These ideas indicate that the reason for the peculiar morphology observed has to be sought in the glass structure. From the appearance of the cabbage-shaped mica crystal aggregates (particularly their spherical, amorphous central parts) it is obvious that phase separation droplets are primary candidates. As detailed elsewhere, this idea is

further supported by recent small-angle X-ray diffraction data [4]. However, replica ®lms prepared in the usual manner (etching of freshly broken surfaces with a 1:1 mixture of 5% HF and 20% HNO3) do not show any phase separation (see Fig. 4). Although the replica ®lm from the glass is rougher than that of the crystalline MoO3 reference face, phase separation droplets larger than some 10 nm can be excluded using this route of preparation. However, etching with HF mainly reveals di€erences in the silica content since by the reaction: SiO2 ‡ 6HF 4 H2 ‰SiF6 Š ‡ 2H2 O water soluble H2[SiF6] is formed. Therefore, pronounced etching of silica-rich parts of the microstructure will occur. If, however, the droplet phase in question has the same or only little di€erent silica content as compared to the matrix glass, nothing should be found. Moreover, attempts to 2+ ÿ preferably attack Mg (EDTA solution) and F [Al(NO3)3 solution] failed. Since chemical etching proved to be unable to + show the phase separation sought, Ar ion-beam etching as commonly used for the preparation of TEM foils was applied. Ion-beam etching is more sensitive to chemical bonding than to speci®c chemical elements and, as shown, in Figs 4(a)±(c), a surface morphology can in fact be generated indicating spherical droplets of a few micrometres in diameter. In the middle of these droplets, the TEM foil is much thicker than between the spheres as

4432

GEBHART et al.: CRYSTAL AGGREGATE

Fig. 5. TEM micrograph of an Ar+ ion-beam etched sample (2p turn) of: (a) glass 1; (b) glass 2; (c) glass 3; (d) reference glass 4.

shown by atomic-force microscopy, AFM (di€erence about 50 nm) and in the TEM as well (up to 100 nm){. From these thickness determinations it is evident that di€erent thickness as a consequence of locally varying etching rates is responsible for the contrast observed [Figs 5(a)±(c)]. For comparison, in Fig. 5(d), a TEM micrograph of a glass of similar composition (from glass 4, plate-like mica crystals are formed) is shown after applying identical preparation steps. The di€erence between glasses 1± {The thickness measurement in the TEM was performed by adjusting a sharply focused electron-beam spot on the sample. After a few minutes, contamination hillocks are formed on the entrance and the exit face of the foil. These hillocks can be separated by tilting the sample as far as the tilting range permits. The projected separation of the contamination spots along with the tilting angle adjusted can be combined to calculate the thickness of the foil.

3 and glass 4 is obvious since in the latter no indication of a topographic contrast is evident. Due to experimental diculties, it was impossible to determine the di€erence in chemical composition of droplet and matrix phase. On the one hand, phase separated droplets cannot be recognized in very thin sections commonly used for EDXS analyses. On the other hand, in thicker parts of the TEM foil (several hundred nanometres thick) where droplets can be distinguished from the matrix, analyses are spoilt by overlapping droplets and matrix, thickness approaching the limits of the absorption correction made in TEM-EDXS analyses codes and beam broadening. Furthermore, radiation damage is a severe problem resulting in large error bars. For these reasons, no statement on the chemical di€erence between droplet and matrix phase could be made until now.

GEBHART et al.: CRYSTAL AGGREGATE 4. DISCUSSION

It was shown that there are liquid±liquid phase separation droplets in glasses forming cabbageshaped mica crystal aggregates with diameters of about 5 mm. The latter ®gure is in remarkable agreement with the diameter of the inner amorphous sphere found in cabbage-shaped mica crystal aggregates. The chemical deviation of the droplets from the matrix glass, however, could not be determined. Nevertheless, etching experiments with a variety of acids proved that there should be no di€erence in the silicon content between both constituents of the phase separation. According to our etching experiments described earlier, magnesium and ¯uorine are most likely to also not vary by a signi®cant amount comparing droplet and matrix phase. Taking into account these previously unreported ®ndings, a new hypothesis on the mechanism of formation of cabbage-shaped mica crystal aggregates can be made: in glass systems where spherical mica aggregates are formed, crystallization of mica occurs only in a very narrow compositional range. This composition is found either directly at the interface between phase-separated droplets and matrix or some distance away at some appropriate point in the radial composition gradient that may be supposed. Because the compositional range necessary for the crystallization of mica is so narrow, crystal growth strictly follows isocompositional lines. This mechanism, however, will work only if the di€usional mobility of at least one of the key components is strongly restricted. Such conditions cannot only arise from overall low di€usivities but are highly facilitated by high crystal growth rates of the mica crystals. In a recent study, HoÈche et al. were able to determine linear growth rates for mica crystals along the sheet plane to be in the order of 1 mm/s at 11008C and 150 nm/s at 9008C [13]. However, the glass used in the earlier study did contain more SiO2 but less Na2O, K2O, MgO, and Al2O3. As detailed in the following, due to this compositional departure of the glasses mica crystals are growing from, it can be concluded that the viscosity is signi®cantly reduced and hence di€usion and consequently mica growth rates for the glass used in this study are substantially enhanced as compared to the earlier system. In general, the crystal growth rate is given by the expression (see for instance Ref. [14]):    w Dm G ˆ G0 1 ÿ exp ÿ …1† Z kT where w accounts for the mechanism of crystal growth, Z is the viscosity and Dm is the supersaturation. At high undercoolings Dm=kTG  t1, so that the term in square brackets can be neglected. The parameter w is constant in the case of normal growth, in the case of spiral growth it is pro-

4433

portional to Dm. The latter can be expressed through the melting enthalpy DHm in the form:   Dm DHm Tm ÿ T ˆ : …2† T Tm Tm It is seen that if we compare two compounds from which the same crystals are formed then the ratio of the crystal growth in the two cases will be   G1 Z2 DHm1 …Tm1 ÿ T † Tm2 1 : …3† G2 Z1 DHm2 …Tm2 ÿ T † Tm1 As the viscosity ratio is changing much faster than the term in square brackets the latter can be neglected in equation (3). The viscosity of the investigated composition at 9008C is about 6  105 Pa s. Therefore, it will crystallize about an order of magnitude faster than phlogopite plates observed in an earlier study [13, 15] where the glass of a di€erent composition (57.3 mol% SiO2, 11.5 mol% Al2O3, 14.8 mol% MgO, 2.6 mol% Na2O, 2.5 mol% K2O, 10.2 mol% ÿ ÿ F , 1.1 mol% Cl ) has a viscosity of 5  106 Pa s at the same temperature. Hence, it can be concluded thatÐunless not directly measured in this studyÐthe linear growth rate of mica crystals perpendicular to the c-axis is already at 9008C assumed to be about 1.5 mm/s. Such a high growth rate will drastically reduce the impact of di€usion on the growth mechanism. One might imagine that (upon following the isocompositional line) the mica lattice is capable of allowing for some bending by appropriate compensation mechanisms including tetrahedral rotation. But bending is strongly limited due to elastic stresses evolving. Therefore, in the next step (above some energy threshold determined by the degree of elastic bending), dislocations (mainly of edge character) are introduced to allow the overall crystal to grow ``bent''. This process decreases the elastic energy of the mica crystal. Later on, the lattice energy can be further reduced by an appropriate rearrangement of the dislocations in dislocation boundaries resulting in the observed small-angle tilt boundaries separating the individual crystal segments. Since ``cabbage leaves'' are hollow spheres, the above mechanism applies to all directions tangential to the surface of the sphere. Therefore, it is most likely, that some two-dimensional arrangement of small-angle grain boundaries on the sphere is formed along distinct crystallographic directions. The experimental proof of this hypothesis, however, is extremely dicult. If the ®rst sphere of the cabbage-shaped mica aggregate (i.e. the ®rst leaf) is formed that way, there will be radial composition gradient in the surrounding glass centred in the middle of the enclosed droplet. At some distance from the surface of the newly formed crystalline ``leaf'', the chemical composition just corresponds to that required for mica

4434

GEBHART et al.: CRYSTAL AGGREGATE

crystallization. Therefore, spaced by some glassy residue the next sphere is nucleating. Following the azimuth growth mechanism described above, sphere by sphere is formed subsequently. One point to be covered further by our new growth model is the behaviour observed upon heat treatment of the same glass at elevated temperatures, where plane phlogopite platelets are formed. To explain this, it is necessary to assume that the miscibility gap responsible for the phase separation phenomena found is closed towards higher temperature. This will stop phase separation droplets from occurring which in turn means that there is no spherical isocompositional line the growth has to stringently follow. An important but still open question is what compositional di€erences distinguish droplet and matrix phase of the liquid±liquid phase separation. Preliminary attempts to detect the chemistry of the phase separation are less encouraging since we already measured EDX line scans of the glass 1±3 TEM samples in the scanning electron microscope. The application of an iterative thin-®lm analysis algorithm (the latter allows for the inclusion of the foil thickness in the ®tting routine) to evaluate the compositional di€erences, however, revealed no signi®cant deviations. 5. CONCLUSIONS

It was shown that individual ``leaves'' of cabbage-shaped mica crystal aggregates are comprised of a repeated sequence of some 200 nm wide segments. Due to a slight misorientation of adjoining segments, low-resolution electron microscopy raises the impression of bent crystals. The direct consequence of this observation, however, is that cabbage-shaped mica crystal aggregates are forced to grow this way by some external condition. The liquid±liquid phase separation we were able to experimentally prove is most likely to be responsible for this peculiar behaviour. Based on transmission electron microscopy observations of glasses and glass-ceramics, a novel growth mechanism for cab-

bage-shaped mica crystal aggregates in machinable mica glass-ceramics is developed. In this model, crystallization follows isocompositional lines at or not far from the surface of the liquid±liquid phase separation droplets. Such behaviour is only possible if di€usion has only little in¯uence on the crystal growth. However, this seems to be the case since the estimated very high growth rates do restrict diffusional control. AcknowledgementsÐThe authors are grateful to Dr I. Avramov, Bulgarian Academy of Sciences, for helpful discussions on the theoretical background. We acknowledge the ®nancial support by Deutsche Forschungsgemeinschaft Bonn Bad Godesberg (INK 6).

REFERENCES 1. HoÈland, W., Naumann, K., Seifert, H.-G. and Vogel, W., Z. Chemie, 1981, 21, 108. 2. HoÈland, W., et al., Glass Technol., 1983, 24, 318. 3. Vogel, W. and HoÈland, W., Angew. Chemie, 1987, 26, 527. 4. Gebhardt, A., Carl, G., HoÈche, T. and Lembke, U., Glastech. Ber. Glass Sci. Technol., to be submitted. 5. Elsen, J., King, G. S. D., HoÈland, W., Vogel, W. and Carl, G., J. Chem. Res. (M), 1989, 1253. 6. Elsen, J., King, G. S. D., HoÈland, W., Vogel, W. and Carl, G., J. Chem. Res. (S), 1989, 160. 7. Hallas, E., Carl, G. and HoÈland, W., Wiss. Z. Friedrich-Schiller-Univ. Jena, Naturwiss., 1985, 34, 775. 8. Bailey, S. W., in Micas, ed. S. W. Bailey. Mineralogical Society of America, Washington, DC, 1987, pp. 1±12. 9. McCauley, J. W. and Newnham, R. E., Am. Mineralogist, 1971, 56, 1626. 10. Jagodzinski, H., Neues Jb. Miner. Mh., 1953, 7, 97. 11. Vogel, W., Horn, L. and VoÈlksch, G., J. Non-Cryst. Solids, 1982, 49, 221. 12. Seeber, W. T., Ph.D. thesis, Friedrich-SchillerUniversitaÈt, Jena, 1987. 13. HoÈche, T., Habelitz, S. and Avramov, I., Acta mater., 1999, 47, 735. 14. Gutzow, I., Kashchiev, D. and Avramov, I., J. NonCryst. Solids, 1985, 73, 777. 15. HoÈche, T., Habelitz, S. and Khodos, I. I., J. Cryst. Growth, 1998, 192, 185.