Search Documents

By R. J. Schaefer, M. E. Glicksman

Direct, in situ observation of metallic alloys during solidification reveals that in some systems crystal multiplication occurs by means of a dendrite remelt-ing process, without the necessity of independent nu-cleation events. The solidification is observed on optically clean surfaces of supercooled molten alloys. In one system (Sn-3 wt pct Bi) , a fine-grained equi-axed structure is produced by the complete disintegration, via remelting, of transient initial dendrites. In contrast, several other metallic systems show little or no evidence of remelting activity. The behavior of these metallic systems is consistent with that expected on the basis of published observations of dendrite remelting in transparent organic alloy systems. It is concluded that dendrite remelting is an important crystal multiplication process in some metallic alloys, but is velatively unimportant in some other alloys. The detachment of dendrite branches from the primary stem by a remelting process has been observed by Jackson, Hunt, Uhlmann, and seward1 during the solidification of some transparent organic alloy systems. The branches released by this remelting process can drift free of the primary stem and serve as seeds for the formation of numerous independent crystals. This proliferation of new crystals does not require repeated independent nucleation events, but only modification of the initial solidification structure. The mechanism of dendrite remelting, as described by Jackson et a1.1 and by Tiller and O''Hara,2 is dependent on the transport of both heat and solute and is, therefore, expected to operate only in alloys, and not in pure substances. It was found in the transparent organic alloy systems that remelting occurred primarily in the constricted regions near the base of dendrite branches, where the branches had grown through the solute-rich liquid layer surrounding the primary dendrite stem. It was suggested that impure solid had formed in these constricted regions at a relatively low temperature and had later become unstable as further solidification of purer material raised the temperature. Jackson et al. proposed that the equiaxed zone of metal alloy castings originates from a similar remelting process, and they demonstrated the operation of this mechanism in "ingots" poured from saturated salt solutions. Although these substances were especially selected to show solidification behavior similar to that of metals,3 the remelting mechanism has, until now, been subject to the objection that its operation has not been directly observed in a metallic system. Considering the fact that thermal properties such as diffusivity may differ by as much as two or three orders of magnitude between metals and organic compounds, this objection was considered to be possibly serious. A method has now been developed by which it is possible to make direct optical observations of metals during solidification. This method has revealed transient solidification processes which would not be detected by metallographic examination of totally solidified specimens. In particular, a grain-refinement process in a metallic alloy has been observed to be a direct consequence of the total disintegration of an initial dendritic structure. The influence of an initial, transient solidification structure on the final solidification structure is demonstrated here for the first time in a metallic system. I) EXPERIMENTAL The technique of using a transparent flux to maintain an optically clean metal surface during solidification has been described by us elsewhere.* The

Jan 1, 1968

By G. B. Gibbs

Various semiempirical equations which relate defect-formation energies and entropies with elastic and thermal properties of a metal crystal are shown to be related by a simple model for lattice energy. An illustrative comparison of theory with experiment is made using self-diffilsion data for fee metals. SEVERAL workers have related the free energy of defect formation with elastic and thermal properties of a crystal lattice, particularly elastic bulk modulus and Debye temperature. The present note shows how the various approaches are related by a simple model for lattice energy. A comparison of theory with experiment is given using the self-diffusion parameters of fee metals. 1) A MODEL OF THE CRYSTAL LATTICE The internal energy per atom in a metallic lattice may be written U=Ui-Ue [1] where Ui and Ue refer to ion-ion interaction and valence electron energies, respectively. The volume per atom V has an equilibrium value 52 for which

Jan 1, 1968

By A. Grant Elliot, H. W. Lavendel

qualitative picture has been developed to describe the oxidation behavior of TaSi2-coated tantalum and CbSi2-coated columbium. These systems have a significantly lower inherent oxidation resistance than MoSi2-coated molybdenum does. This stems primarily from the fact that Ta2O5 and Cb2O5 are nearly as stable thermodynamically as SiO2, whereas MoO2 or Moos are not. Further, diffusion of silicon in the Ta- and Cb-Si system is considerably slower than in the Mo-Si system. These ,factors prohibit the mechanism of selective oxidation of- silicon which accounts for the oxidation resistance 01- MoSi2-coated molybdenum. The silicide can be stabilized by adding suitable Modifiers which increase the thermodynamic stability of the silicate formed during oxidation. Modifiers, such as aluminum, can be inroduced into solid solution in the coating. in controlled amounts through proper selection of the source in the pack cementation process of coating fov~rzatiorz. Addition of aluminum to TaSi2, coatings on tantalum was effective in moderately increasing the oxidation resistance. EXTENSIVE experimental work and analysis have established the nature of the oxidation behavior exhibited by MoSi2- and MoSi2 -coated molybdenum-base alloys, and defined the conditions for maximum protection against oxidation of the substrate.'-* The oxidation resistance of MoSi2 in the temperature-pressure range of 1100°C-PO2 > 10-5 atm to 1900°C— PO2 > 10-1 atm is due to the formation at the surface of a continuous film of SiO2 which results from selective oxidation of silicon. Under the prevailing kinetic conditions, this film is stable toward the molybdenum silicide with which it comes in contact. Initially molybdenum oxidizes also, but it forms volatile species. SiO2, however, nucleates and grows as a condensed phase. Once a continuous film of SiO2 has formed, the oxidation rate falls to that observed for the oxidation of pure silicon indicative of diffusion through the oxide film as the rate-controlling mechanism. This oxidation behavior is of course highly dependent upon temperature and oxygen pressure. Bartlett and Gage13 and Bartlett, McCamont, and Gagelb define precisely this dependence in terms of the oxygen partial pressures and silicon diffusivities required to support a stable SiO2 film. At low temperatures (near 500°C—the "pest" region) silicon diffuses too slowly to be selectively oxidized. Hence, molybdenum and silicon oxidize readily in proportion to their stoi- chiometry. At high temperatures and low pressure, SiOz dissociates to form volatile SiO(g), and a protective film cannot be maintained. Application of the MoSiz/Mo system is limited to temperatures below 1900oC, the eutectic between MoSi, and MO5Si3.5 The oxidation behavior of MoSi2-coated molybdenum is essentially the same as that outlined above with the exception that the MoSi2 is not in equilibrium with the molybdenum substrate. At the temperatures under consideration silicon will diffuse rapidly into the molybdenum eventually converting the coating to MosSi3.4 The rate constant for subsequent decomposition of Mo5Si3 into Mo3Si plus silicon, and/or the diffusivity of silicon through Mo3Si then becomes low enough to allow active oxidation of both molybdenum and silicon with subsequent degradation of the specimen. A stable silica film can be formed but at temperatures and/or oxygen partial pressures higher than those required with MoSi2 present as a source of si1icon.l, 4 Because of the similarity between the silicides of molybdenum and those of columbium and tantalum one would expect similar oxidation behavior for coatings in the respective systems. This is not entirely the case, however, as shown by the experimental results reported herein. Regarding tantalum and columbium disilicide coatings on tantalum and columbium substrates, respectively, the oxygen arriving at the surface of the coating partitions itself nearly equally between the metal and the silicon, and a two-phase oxide layer (Me2O5 plus SiO2) is always formed. The diffusion of silicon in the tantalum and columbium silicides is relatively slow, compared to that in the molybdenum silicides, which further enhances this equipartitioning of oxygen. Thickening of the coating during service by inward diffusion of silicon into the substrate is correspondingly slow, and the effective thickness of the coating at the roots of cracks and defects is only slightly changed providing high probability for premature coating failure. Furthermore, the SiO2 glass that is generated is not thermodynamically stable with respect to the coating. The metal silicide tends to reduce the SiO2 liberating either free silicon or SiO. The situation can be improved by suitably modifying the coating such that the stability of the protective glass which is generated during service is increased. Thus, selective oxidation of silicon and the modifying agent will occur, and the silicide coating will not tend to reduce the oxide layer. Modifying agents can be introduced into the coating by the pack cementation process. Using sources containing the modifier at controlled chemical potentials allows control of the coating composition. Partially substituting aluminum for silicon in TaSi2 coatings by forming a Ta(Si,Al)2 solid solution was effective in moderately increasing the oxidation protection.

Jan 1, 1968

By J. L. Swedlow, M. L. Williams

Two standard techniques for assessing fracture initiatlon in sheet specimens involve measurements of 1) the applied load of fracture which, via linear fuacture mechanics, is used to find the fracture tough/less SC (or SIC), and 2) the velative displacement of fracture of the opposite free faces of the crack (COD). In the past, the quantitative association between these two parameters has not been determined in detail, owing to the mathematical difficulties of computing. the stress and strain fields in the combined elasto-plastic flow near the crack. Using a numerical technique Irecently developed and applied to a 7075- TO' aluminum plane stress specimern, the two parameters noted have been determined and associated quantitatirely. CONSIDERABLE effort has been expended in selecting a rigorous yet practical manner for predicting fracture. The usual point of departure is the modified Griffith criterion in which the mechanism of energy dissipation is reflected in Sc (or SIC), the parameter of "fracture toughness". Sc. differs from the surface energy used by Griffith in perfectly brittle fracture in that it includes the effect of plastic flow in the vicinity of the crack tip. Because plastic flow alters the stresses and displacements predicted from the usual linear theory of fracture mechanics, and because analysis of the nonlinear stress-strain field has presented great obstacles: only estimates of several pertinent quantities have heretofore been available. Nevertheless, two parameters have often been measured experimentally althougll usually just at the point of fracture. These are the energy release rate, or fracture toughness 9, (or I,), and the crack-opening displacement (COD). Such observations have been used in connection with frarture criteria and have been most helpful in design, although their application requires care with respect to changes in geometrical configuration from that upon which the criteria were based 1,2 Recently, however, a reasonably accurate and comprehensive method for analyzing the elasto-plastic flow has been devised and applied to plane stress specimens having arbitrary octahedral stress-strain curves.3 It thus becomes feasible to compute, albeit numerically, both the fracture toughness and crack opening as functions of the applied tensile loading on a typical specimen geometry. As a result, it is now possible to associate these two parameters quantitatively, and thereby to associate data obtained by one method with data pertaining to the other. ELASTIC ANALYSIS For the case of linear elasticity, it is relatively easy to deduce the connection between various measures of crack deformation. Using, for example, the singularity analysis, it is found that the leading term of the Airy stress function x f°r symmetrically applied loadings on an isotropic medium is

Jan 1, 1968

By S. W. Zehr, G. S. Murty, W. A. Backofen

AN observation by Squires, Weiner, and phillipsl has stimulated interest in a mechanism of deformation at high temperature (above -0.5 of the absolute melting point) that is not usually thought to be involved in the creep of metals. They found precipitate-free or "denuded" zones along grain boundaries normal to the stress axis in tensile specimens of a hydrided Mg-1/2 wt pct Zr alloy after slow straining at 450° and 500°C. Their suggestion that the zones were a result of directional diffusion of magnesium atoms was supported with calculations from Harris and Jones2 which also tended to identify the process as Nabarro-Herring (N-H) creep3,4—a process that is characterized by Newtonian viscosity, with stress (o) directly proportional to strain rate (C), and one that proceeds by lattice diffusion. Material transport by diffusion along grain boundaries is an alternative basis for Newtonian viscosity, as demonstrated by Coble,5 and is favored over N-H creep for fine grain size and lower temperature.6 In either case, the strain rate sensitivity of flow stress, defined as m = d log o/d log E (which commonly appears in the empirical a = KEW), is unity. The experiments to be described here gave metallo-graphic indications of diffusional creep which were no less clear than those just cited, but the associated m (as measured by the method of Ref. 7) was much more like that of a Bingham material. For true Bingham behavior, (a — ao) E, where oo is an initial yield stress. Therefore, even if flow after yielding should occur by diffusional transport, m must increase from 0 as < is raised to reach unity only in the limit as One result is that the resistance to tensile necking conferred by Newtonian viscosity is lost, which would become increasingly apparent through the formation of more sharply defined necks as 6 is lowered. That trend appeared in these experiments. Most of the work was done on an extremely finegrained alloy of magnesium with -6 wt pct Zn and ½ wt pct Zr (Mg ZK60); the initial mean-intercept grain diameter, L, was 0.55 µ. Shown in Fig. l(a) is a germanium-shadowed carbon replica of the surface of a specimen that was coated with silicone oil,* heated to 270°C under a helium atmosphere, and then stretched locally -200 pct at an initial E = 3.3 x 10-5 sec-1. Under

Jan 1, 1969

By P. Bolsaitis, L. Skolnick

A model of metallic solutions is postulated which explains the energy of formation of alloys on the basis of changes in electron density around solute and solvent atoms and changes in pairwise interactions between atoms. The model is based on the Wigner-Seitz cell model of metals, with the electronegativities of the metals as the parameter for the electron-attracting power of the constituent atoms. In contrast to models based on the description of cohesive energy in terms of pairwise interactions only, which require an empirical, adjustable correction parameter to the Berthe-lot relations in order to predict negative energies of formation, the electron cell model accounts for both negative and positive energies on the basis of parameters of the pure components. Calculated heats of mixing based on published parameters of the pure metals are found to compare favorably to experimental values, despite the absence of any adjustable parameters in the model. THE formulation of an "exact" quantum mechanical model for the energy of cohesion of metals reduces ultimately to the many-body problem and for a complete solution would require the knowledge of wave functions and energy levels of all the particles in the system. The case of alloys is further complicated by the fact that the potential fields around the constituent atoms are different and the matching of boundary conditions in terms of the resulting wave functions is still an unresolved mathematical problem. It appears that in spite of the numerous investigations that have led to an elucidation of many properties of the metallic state, the formulation of the Hume-Rothery rules in 1931 marks the lone accomplishment toward simple, qualitative prediction of the energy of formation of alloys. More recently Friedel&apos;&apos; and Arafa have utilized the modern concepts of the electronic structure in metals in combination with the Hume-Rothery rules to arrive at quantitative estimates of the energy of formation of alloys with low solute contents. The relative success of the Hume-Rothery rules and Friedel&apos;s and Arafa&apos;s studies encouraged broaching the subject of formulation of a model based on realistic physical parameters and of conceptual and mathematical simplicity that could be applied to alloys regardless of concentration. On the basis of the free-electron theory, the cohesive energy of pure metals can be expressed as a function of "valence" (number of free electrons per atom), lattice parameter (or density), and the ioniza-tion energy. In addition to the electronic contributions to cohesion, one has to consider the pairwise exchange interaction energy between neighboring ion cores and the contribution from van der Waals forces. The energy changes upon alloying are viewed as resulting primarily from a redistribution of electron densities around the ions (a simplification of the more basic concept of a change of the wave functions of the electrons in the presence of a perturbing potential field). In addition there is a change in the pairwise interactions between neighboring ions which is usually the only factor taken into account in commonly used adaptations of the quasichemical theory. The electronegativities of the metals forming the alloy are used as the measure of the electron-attracting power of the ions in the alloy and the difference in electronegativities as the driving force for changes in electron energy on alloying. The interplay of pairwise interactions and electronic forces results in positive or negative energies of formation of alloys. Results obtained for systems to which the free electron model is best applicable are in good agreement with experimental values. 1) COHESIVE ENERGY OF METALS IN TERMS OF THE ELECTRON CELL MODEL The simplest physical model of a metal is that of a lattice of positive charges immersed in a sea of uniform, negative charge distribution represented by the

Jan 1, 1969

By John P. Hager, Igor A. Wilkomirsky

The thermodynamic properties of the Pb-Ag system have been determined between 775° and 950°C by the cell: Electrotransport measurements on silica-saturated PbO-SiO2 melts established that the conduction was entirely ionic and that lead is present only in the divalent state. The activities of lead and silver exhibit Positive deviations from ideal behavior. In dilute solutions the activities of lead and silver follow Henry&apos;s law. The values obtained for yhb and yig at 1200°K are 1.49 and 1.69, respectively. Determination of a portion of the liquidus curve agrees well with the established phase diagram. provides an accurate means for the thermodynamic study of liquid alloy systems. However, Wagner and wernerl have shown that cells of type I are limited to alloy systems for which the reaction at the alloy-electrolyte interface will be extremely small. The traditional electrolyte for cells of type I has been a low melting point mixture of group IA chlorides, to which a small amount of the compound ACIZa has been added. However, the choice of a chloride electrolyte may not always be practical, in that the Gibbs free energies of formation of the compounds ACIZ, and BCIZb may not be sufficiently different to minimize Reaction [I]. On the other hand, the Gibbs free energies of formation of the oxides of metals A and B may be sufficiently different to minimize Reaction [I]. In this case, the use of molten oxide electrolyte would be clearly advantageous. In the present study, the thermodynamic properties of the Pb-Ag system have been investigated using the cell: Previous galvanic cell studies on the Pb-Ag system2&apos;= are not reliable due to the displacement reaction between the alloys and the chloride electrolyte. Wagner and ~erner,&apos; using Laitinen and Liu&apos;s&apos; value at 450°C for the standard electrode potential of Reaction 111, have estimated the percent error in the activity of lead bPb) when using a chloride electrolyte. At 1000°C the error for xpb = 0.10 is 12 pct. Taking the Gibbs free energies of formation for PbClz and Ag cI,&apos; and the equations of Wagner and Werner,&apos; the error in apb at 1000°C and Xpb =0.10 was calculated to be 140 pct. However, using the Gibbs free energies of formation for Pb O&apos; and an estimated value for Ag 20, the error in apb when using a molten oxide electrolyte at 1000°C was estimated to be less than 1 pct for The virtual reaction of cell I1 for the passage of n Faradays of positive electricity from left to right in the cell is: where 5 is the Faraday equivalent (23,063 cal per v equivalent), E is the electromotive force of the cell in volts, and n is the number of Faradays of electricity required for the transfer of 1 g-atom of lead through the cell. The value of n should be equal to 2 since lead is generally divalent in oxide melts. However, the presence of either tetravalent lead ions or electronic conductivity through the molten oxide electrolyte will result in a value of n greater than 2. Bockris, Kitchener, and Davies&apos; have determined the amount of ionic conductance in a liquid PbO-SiO2 mixture (xPbO= 0.50) at 920°C by measurement of the volumes of CO and C& evolved at a carbon anode during electrolysis. From their data a value of 2.13 * 0.16 was calculated for n. Although their measurements indicate the possible presence of multivalent lead ions and/or electronic conductivity, their measurements were not at silica saturation and no definite conclusions can be drawn. However, as a result of the electrotransport measurements described in the next section the ;due of n can be taken as equal to 2.

Jan 1, 1969

By W. Jolley

Decarburized iron and Fe-3.28 pct Ni alloys were impact and tension tested in the temperature range of ambient to 4°K. It was found that alloying with nickel improves the fracture properties of the ferrite matrix. The improvement in toughness of iron alloyed with nickel is attributed to the effect of the nickel addition on the temperature and strain rate dependence of the yield stress as well as on slip behavior. The Fe-Ni alloy exhibits wavy slip when strained at temperatures down to at least 50°K. This behavior contrasts with that of the unalloyed iron where a change It is a generally accepted fact that of the common alloying elements of steel only nickel and manganese improve both the room-temperature yield strength and the notch impact properties. It has been argued for a number of years that manganese refines the ferrite grain size, the pearlite spacing, and the grain boundary carbides in medium- and low-carbon steels and thereby increases resistance to brittle fracture.&apos;&apos;&apos; This hypothesis recently was confirmed for a 0.04 pct C iron containing manganese, but at the same time from wavy to planar slip occurs at about 120°K. The observed lower temperature and strain rate dependence of the lower yield stress in the Fe-Ni alloy as compared to the unalloyed iron may be attributed to this difference in slip behavior. In contrast to suggestions published previously, transmission electron microscopy showed that the uniform dislocation distributions and the lack of dislocation tangles which are produced by low-temperature straining of bcc materials cannot be attributed to the absence of cross slip at these temperatures. no increase in notch toughness was observed in a decarburized Fe-Mn alloy.3 It thus appears that manganese only increases the toughness of Fe-Mn-C alloys by refining grain boundary carbides and has no toughening effect in carbon-free iron. On the other hand, the mechanism whereby nickel improves toughness has not been established. Hodge et a1.4 have clearly shown that a 3.28 pct Ni addition lowers the impact transition temperature of 0.02 pct C steel approximately 35°C when the grain size is maintained constant. Recently, Wullaert and Tetelman6 stated that nickel refined the grain boundary carbides of an Fe-C alloy and consequently increased the toughness. They also observed the previously reporteda-&apos;

Jan 1, 1969

By R. R. Judd, H. W. Paxton

The rate of dissolution of cementite was studied in three low-carbon materials: a zone-refined Fe-C alloy, an Fe-0.5pct Mn-C alloy, and a commercial low-carbon steel. The materials were spheroidized, ad then held isothermally at temperatures above the Al. The isothermal anneal was interrupted periodically by a water quench and the specimens were analyzed by quantitative metallography for the amount of aus-tenite formed during the anneal. The results of this study were compared with an analytical model for the process, which assumes that carbon diffusion in aus-tenite is the rate-controlling step for the cementite dissolution process. The correlation between the model and the experimental data is excellent for the zone-refined Fe-C alloys; however, the Fe-0.5 pct Mn-C alloys and the commercial steel deviate from the calculated model. This deviation is thought to be a result of manganese segregation between the carbide and the matrix. The rate of nucleation of austenite at carbide interfaces was reduced by the manganese addition and enhanced by the presence of ferrite-ferrite grain boundaries. PREVIOUS investigations of the nucleation and growth of austenite from ferrite-carbide aggregates are not entirely satisfying for at least one of several reasons. The most prevalent of these is a lack of quantitative data. Engineering studies have been run on many steels with little control over important parameters such as composition and initial aggregate structure. The data obtained are valid only for material with identical chemistry and thermal history. A more informative approach to the problem of aus-tenitization would be to determine the mechanism that controls the rate of solution of carbide in austenite and how it is modified by alloying elements. This information could then be used to calculate an austeniti-zation rate for any material, provided its composition and structure are known. The object of the present work is to establish the rate-controlling step for cementite dissolution in Fe-C austenite and to investigate the modification of this rate by small manganese additions. The composition and structure of the material used were carefully controlled and all measurements were designed to allow a quantitative analysis of the kinetic process that controls the austenitization rate. A MODEL FOR DISSOLUTION OF CEMENTITE Cementite dissolution has been analyzed mathematically by a model that approximates the material used in the experiments. This model postulates a regular ar-array of identical cementite spheroids with 4 C( diam, embedded in a grain boundary- free ferrite matrix. The analysis provides a detailed description of the dissolution of one carbide spheroid and a generalization of the solution by summation over all the carbides in the material. The carbides may be isolated by defining identical, space-filling cells of ferrite around them. If the cell dimensions are greater than the diameter of the austenite sphere resulting from complete dissolution of the carbide, and no interaction (through diffusion in ferrite) takes place between cells during the dissolution process, the model need concern only one cell, since the solution in each cell is identical. In the experimental material, the dimensions of the cell, the carbide, and the final austenite sphere are approximately 24, 4, and 8 p, respectively; use of the single cell is therefore justified. The experimental observations are made on the austenite nodules that form around each carbide during the dissolution process. The model concerns the growth of these austenite nodules. The attendant shrinking of the carbide can be obtained from the same analysis by an extension of the calculations. Several a priori assumptions are necessary to make the analysis of the growth problem tractable. They are: 1) carbon diffusion through the austenite nodule is the rate-controlling process; 2) local equilibrium exists at all interfaces, 3) the austenite nucleus that forms on each carbide instantaneously envelops the carbide; 4) during the austenite growth process, the diffusion flux of carbon in ferrite is insignificant; 5) a quasi-steady state exists in the austenite concentration field; that is, at any instant during the dissolution process, the austenite carbon concentration gradient closely approximates that for a steady-state solution; and 6) the effects of capillarity on the dissolution rate of the carbides can be neglected. Referring to Fig. 1, a mass balance at the y-a interface for an infinitesimal boundary movement gives: Where rb is the outer radius of the austenite shell, C1 and C are carbon concentrations at the interface in austenite and ferrite, respectively, see Fig. 2, is the diffusion coefficient of carbon in austenite for the concentration of carbon at the interface, and t is time. The fifth assumption permits the austenite carbon concentration to be approximated by the Laplace solution for the spherical case. Therefore, where C(Y) is the carbon concentration at r, and A and B are constants. Local interfacial equilibrium fixes the boundary conditions for the diffusion problem. They are:

Jan 1, 1969

By Arnt Solbu, Jomar Thonstad

The reaction between CO, and aluminum in cryolite-alumina melts in contact with aluminum has been studied by passing CO2 over the melt. In unstirred melts a homogeneous reaction between dissolved metal and dissolved CO2 was observed. In stirred melts in which convection was induced by bubbling argon through the melt, the dissolved metal apparently reacted mainly with gaseous CO2. The rate of formation of CO increased slightly with increasing depth of the melt, and it did not depend on whether CO2 was passed over or bubbled through the melt. The rate of formation of CO increased with increasing area of the metal/melt interface and with the application of anodic current to the metal. It is concluded that the dissolution of metal into the melt is the rate-determining reaction. THE current efficiency in aluminum electrolysis is determined by the rate of the recombination reaction between the anode gas and the metal: 2A1 + 3CO2—A12O3 + 3CO [1] as originally stated by Pearson and waddington.1 The occurrence of this reaction in cryolite-alumina melts in contact with aluminum was first verified experimentally by Schadinger.2 Thonstad3 has shown that the reaction may proceed further to give free carbon: 2A1 + 3CO— A12O3 + 3C [2] Normally only a few percent of the CO formed undergoes such reduction. The mechanism of these reactions has not yet been clarified. Aluminum, as well as CO,, is soluble in the melt. The solubility of aluminum in cryolite-alumina melts at around 1000°C corresponds to 75 x 10- 6 mole A1 per cu cm,4 while that of CO2 is only 3 x 10-6 mole CO, per cu cm.5 Taking into account the stoichiometry of Reaction [I], the ratio between dissolved aluminum and dissolved CO2 available for the reaction in a saturated melt is about 40. Therefore, as will be shown in the following, the reaction probably mainly occurs between gaseous COa and dissolved aluminum. The dissolved aluminum presumably consists of subvalent ions of aluminum and sodium.4&apos;6 Since the interpretation of the present results is not dependent upon the nature of this solution, the dissolved metal will be designated solely as Al+ in the following. The reaction can then be divided into four steps: A) dissolution of metal, e.g., 2A1 + Al3 — 3A1+ [3] B) diffusion of dissolved metal through a boundary layer; C) transport of dissolved metal through the bulk of the melt; D) Reaction [1]. If dissolved CO, takes part in the reaction, three additional steps embodying the dissolution and transport of CO2 must be added. schadinger2 observed, when bubbling CO2 through the melt, that the rate of formation of CO (in the following designated rfco) did not depend on the distance from the metal surface. The results also indicate that the rate of bubbling did not affect the rfco. When passing CO, over the melt, Revazyan7 found that the loss of metal did not depend on the depth of the melt above the metal or on the flow rate of CO2, and concluded that Step A is rate-determining. In an unstirred melt, however, Gjerstad and welch8 found that the rfCo decreased with increasing depth of the melt, indicating that step C was rate-determining. It thus appears that the rate control of the process depends on the experimental conditions, particularly on the convection. In the present measurements the reaction has been studied in unstirred as well as in stirred melts. EXPERIMENTAL AND RESULTS The experiments were carried out at 1000°C in a Kanthal furnace with a 10-cm uniform temperature zone (±0.l°C). The melts were made up of "super purity" aluminum (99.998 pct), hand-picked natural cryolite, and reagent-grade alumina. In experiments where alumina crucibles were used, the alumina content in the melt was close to saturation (13.5 wt pct9); otherwise it was 4 wt pct. Pure Co2 (99.85 pct) was passed over the melt, and the exit gas was analyzed for CO2 and CO by the conventional absorption method.3 From the weighed amount of CO (as CO2) the rfco was calculated as the number of moles of CO formed per min per sq cm of the surface area of the melt. The amount of carbon formed by Reaction [2] was not determined. As already indicated the rfco is much higher than the rfC, by Reaction [2]. Since the rfC probably is proportional to the rfco, the measured rfco should then the proportional to, but slightly lower than, the total rate of Reactions [I] and 121. In general the scatter of results obtained in duplicate measurements was ±5 to 10 pct, while within a given run a precision of ±3 to 5 pct was obtained. The various crucible assemblies that were used will be described below. Measurements in Unstirred Melts. When carrying out aluminum electrolysis in small alumina crucibles. Tuset10 observed that after solidification the lower part of the electrolyte was gray and contained free metal, while the upper part near the anode was white and contained no metal. One may test for the presence of free metal by treating with dilute hydrochlorid acid.

Jan 1, 1969

By P. Bolsaitis, L. Skolnick

Vapor pressures of silver over ternary alloys of six different copper to gold ratios have been measured by means of the Knudsen effusion technique. The data obtained in the temperature range of 1330° to 1375° have been used to estimate the integral excess free energies of the ternary alloys and the activities of copper and gold. The results compare favorably to the predictions based on the electron cell model of alloys recently formulated by the authors. THE orifice effusion method for measuring vapor pressures of materials in condensed phases was first introduced by knudsen, and has since been used extensively for the determination of thermodynamic properties of condensed phases. The basis of the measurement is the effusion, through a small orifice, of vapor in equilibrium with a condensed phase. Assuming free molecular flow into high vacuum, the vapor pressure, P (in mm Hg), of the effusing species is calculated from the following equation:&apos; where w (g) is the weight of effusing species lost during a run, t (sec) the time length of the run, A, (sq cm) the area of the effusion orifice, T ("K) the temperature at which the run is conducted, M the molecular weight of the effusing species, and k the Clausing factor, a constant that corrects for deviations from free molecular flow caused by the finite thickness of the orifice walls.17 The determination of partial and integral thermodynamic variables from vapor pressure data of ternary alloys is possible only if the vapor pressures of all but one component are negligible or can be esti-

Jan 1, 1969

By R. L. Skaggs, R. A. Molsberger

DARKEN1 has pointed out that in most binary liquid metal systems the thermodynamic behavior is relatively simple in the terminal regions. Between the terminal regions. the thermodynamic behavior is more complex. To describe this behavior. he has defined the stability and the excess stability as shown: Qualitatively, an inflection in the free energy of mixing curve results in a change in sign, from positive to negative, of the stability function and is associated with a phase separation. Conversely, a large, positive stability over a narrow composition range is indicative of compound formation. Darken has shown a number of binary systems where the excess stability is constant in the terminal regions and varies sharply in between. The work described in this communication was undertaken to study the variation of stability and excess stability in a binary system as a function of temperature. In addition, this work provides means for the direct comparison of previously existing equilibrium (activity) and calorimetric (molar heat of mixing) data on the Hg-Sn system. Hildebrand2 measured the vapor pressure of mercury over Hg-Sn alloys at 595°K. Calorimetric studies were made by Kawakami3 at 523°K and by Kleppa4 at 423°K. In this work, the equilibrium vapor pressure of mercury was measured over a series of Hg-Sn alloys at 453" and 503°K by use of fused salt manometer with one arm exposed to mercury vapor. Deflections of the fused salt manometer were balanced by argon pressure on the other arm. Argon pressure was subsequently measured with a dibutyl phthalate manometer. Tin activities were determined by graphical integration of the Gibbs-Duhem equation making use of the familiar Darken5 a function. The activities of mercury and tin showed positive deviation from Raoult&apos;s law at 453 and 503°K as Hildebrand observed at 595 K. In Fig. 1 are plotted log activity of mercury and tin vs reciprocal temperature. The plot contains activities obtained from calorimetric studies. Hult-gren6 has calculated activities from Kleppa&apos;s4 data, and Kubaschewski7 has similarly treated Kawakami&apos;s7 data. Presumably. in each case a regular solution was assumed. For regions where a two-phase field

Jan 1, 1970

By H. J. Levinstein, H. J. Guggenheim

RbFeF3 is a transparent ferromagnet with a large faraday rotation which permits the direct observation of magnetic domain structures in bulk crystals. If the position of dislocations within the crystal were known a study of domain wall dislocation interactions would be possible. In this paper an etch pitting solution is described which gives a one-one correspondent between the etch pits and the dislocations intersecting the (100) surfaces of RbFeF3 crystals. In addition, RbFeF3 is shown to have sufficient plasticity so that known arrays of dislocaticms can be put into the crystal. It is concluded that RbFeF, is an ideal material for the investigation of domain wall dislocation interactions. THE influence of lattice defects on the motion of domain walls in ferromagnetic materials has been studied extensively both theoretically and experimentally.4 The experimental data obtained has been qualitative in nature due to the lack either of a quantitative knowledge of the magnetic domain array in the material or of the nature of the dislocation configuration in the material. At room temperature RbFeFJ has the cubic perovskite structure. It is ferromagnetic below 86°K.5,6 It is also transparent and has a large faraday rotation in the magnetic state.7 The domain structure of this material is observable in three dimensions in crystals which are millimeters in thickness.&apos; Since RbFeF3 satisfies the requirements of having an easily visible three-dimensional magnetic domain structure, it was of interest to determine if the dislocation structure of the material could be determined and if the material was sufficiently plastic so that dislocations could be introduced into the as grown crystals. This is a report of the results of that investigation. EXPERIMENTAL PROCEDURE The crystals of RbFeFs used in this experiment were grown by a technique described elsewhere.5 In order to obtain flat strain-free surfaces, the samples were first cut to the crystal orientation desired and the surfaces were polished by employing standard metallographic techniques. The samples were then immersed in a solution of 1 to 3 g of KOH in 100 ml of water. This solution removes the damaged surface layers uniformly leaving a film on the surface of the sample. Etch pits are obtained by immersing the sample in an approximately 50-50 solution of 30 pct peroxide and concentrated acetic acid. The film formed during the KOH treatment turns deep red in color in the etching solution and flakes off the sample. The etching time required depends upon the pit size desired and

Jan 1, 1970

By J. Szekely

A rrlethod is presented for the estitnation of oxygen pickup by teettzed molten steel streams. Of the mechanisrt~s considered for oxygen absorption, physical entrainment appeared to be the most significant. The estent of this physical entrainment may be calculated by the application of boundary layer theory. It was found that the values of oxygen pickup thus predicted were in wry good agreement with data reported in the literature, in the case of short, well-defined metal streams. This entrainment model does not provide an adequate description of oxygen pickup by long streams, exoosed to external winds and where breakup of the streatn is likely. WheN molten steel is teemed into a mold or tundish it will absorb oxygen contained in its environment. Recently interest has been expressed in the measurement and possible prediction of the extent of this oxygen absorption: this interest owes its existence to two principal factors: a) the desirability of providing a quantitative assessment of the role played by the absorbed oxygen in impairing the quality of (vacuum degassed. stainless, or continuously cast) steel; b) the need for providing guidelines for stream protection methods. The purpose of this communication is to suggest a simple method for the prediction of oxygen absorption by continuous metal streams and to compare the predicted values with the data available in the literature. It is hoped that this method may provide at least an initial framework for the interpretation of additional data that might become available. FORMULATION Consider a continuous metal stream falling into a molten metal pool as depicted in Fig. 1. This system may absorb oxygen due to the following two mechanisms: 1) physical entrainment of gases (including oxygen) into the molten pool where reactions. thus absorption, will occur; 2) mass transfer (convec:tive diffusion) of oxygen to the metal stream. 1) Physical Entrainment. The physical entrainment may be calculated by using the concept of the diaplace-ttzent tlzickness of the laminar boundary layer, which will be stated briefly in the following.&apos; When a fluid flows past a solid surface, a portion of the fluid close to the surface is retarded due to fric-tional forces. This region in which the velocity of the fluid differs from that in the main stream is called the boundary layer. An illustration of this is given in Fig. 2 where it is seen that the thickness of the boundary layer increases as we move along in the x direction. In the laminar region, when Re, < 10&apos; the thickness of the boundary layer is given by the following relationship: f =4.64 Re;1" [1] where 61 is the boundary layer thickness, x is the length along the leading edge, Us is the velocity of the free stream, (i is the viscosity of the fluid (gas in this instance).

Jan 1, 1970

By B. Chalmers, L. Bäckerud

The growth temperature of dendrite tips has been measured in a binary alloy system, Al-Cu, as a function of rate of formation of solid phase. Metallographic examination has rerealed the dendritic structure dereloped during different stages of the freezing process and factors influencing the undercooling, such as solute buildup, surface tension, and kinetic effects, have been estimated. Practical implications of the results are discussed. THE development of a dendritic growth pattern during the solidification of metals and alloys has an influence at least equal to that of grain size on the properties of the casting or ingot being produced. and consequently much interest has been devoted to the study of this phenomenon. A review of the knowledge gathered up to 1964 has been given by Chalmers.&apos; Among recent publications in the area the works of Jackson et al 2 and Flemings and his coworkers3,4 are of special interest. This work is an experimental study of the factors which influence the development of a dendritic pattern during the solidification of a typical binary alloy system, A1-Cu, when there is no heat source in the liquid. This means that the temperature and the temperature gradient are controlled by the heat extraction from and latent heat evolved by the growing crystals. EXPERIMENTAL PROCEDURE Samples were prepared from a master alloy made up of aluminum, 99.995 pct, and copper. 99.99 pct, of the concentrations given below and of sufficient size to fill cylindrical graphite molds of 35 mm ID and 35 mm height. Bottom. wall. and lid of the graphite molds were all of 6 mm thickness, Fig. 1. Thin ceramic thermocouple shields could be introduced at different positions in the sample through holes in the lid. The assembly was heated in a small vertical resistance furnace to a suitable temperature. 700° to 720°C. which is well above the melting point of the sample. It was thereafter taken out of the furnace and chilled at selected rates, and the temperature changes in the sample were recorded. For high sensitivity in the temperature recordings, a reference thermocouple was held in a solidifying pure aluminum melt. which was very well insulated and therefore was at a constant temperature. 660°C, for nearly half an hour. The thermocouples were connected so that the difference voltage was fed into the recorder of which the most sensitive range was 1 mV. Temperature variations could be read to within 0.l°C. The thermocouples used were of the Chromel-Alumel type (28- or 38-gage wires) and the recorder a two-pen Moseley strip chart recorder. The different rates of chilling were obtained in the following way. a) The graphite crucible was wrapped in one or two layers of an insulating. feltlike material (Fiberfrax paper from the Carborundum Co.). The drop in temperature per unit time after complete solidification could thereby be varied from 8" to 18°C per min. This corresponds to a rate of heat extraction in the given samples from 0.08 to 0.17 cal cm-&apos; sec-I. 6) The bare graphite mold was placed free on a piece of firebrick. The measured temperature drop was around 50°C per min (-0.5 cal cm-" &apos; sec- I). c) By blowing compressed air through a tube with many holes surrounding the mold in a ring shape. the rate of heat transfer could be raised to 150&apos; to 200°C per min (-1.7 cal cm-3 sec- I). d) Cooling rates around 1000 C per mill (-9.4 cal cm-3 sec-l) were obtained by substituting the air in the device used in c by water. The range of cooling rates used in these experiments correspond to technical casting processes from slow solidification as obtained in sand molds to very rapid freezing which occurs in some continuous casting and welding processes. EXPERIMENTAL RESULTS The A1-Cu system was chosen for the experiments: the relevant part of the phase diagram is reproduced in Fig. 2. Observation of the Center Temperature. When a thermocouple is inserted in the center of the melt. the type of cooling curves obtained at slow solidification rates differs little from numerous published curves from similar systems. and the solidification pattern

Jan 1, 1970

By R. Taggart, R. H. Ericksen, D. H. Polonis

Transmission electron microscopy techniques hare been used to study the spontaneous marfensite phase that forms during the thinning of Ti-Cr alloys. The structure of this phase has been found to differ. from the hcp martensite and has been identified as a clisforted fcc lattice. The spordaneous marlensile in this system is interpreted as a metastable isomorph of the equilibriunz TiCr, phase. WHEN thin-foil specimens for transmission electron microscopy were prepared from metastable alloys of the Ti-Cr system, a spontaneous transformation occurred&apos; when the foils contained the untrans-formed 0 phase or two-phase mixtures of the 0 and w phases. The purpose of this paper is to examine the relationship that exists between the crystal structure of the spontaneous product and that of the matrix. The differences between the martensite transformation that occurs in the bulk material and the spontaneous reaction that occurs during the preparation of thin foils are compared with the observations that have been reported for similar phenomena in a commercial titanium alloy.&apos; A spontaneous reaction that exhibits characteristics similar to a martensite transformation has been reported to occur in thin foils prepared from the retained metastable structures of several alloy systems.1&apos;3"7 ~ull~ observed such a spontaneous reaction in 0 brass and attributed the apparent increase in the martensite transformation temperature to a decrease in the non-chemical free energy associated with bulk strain energy. He also showed that the spontaneous product did not have a specific habit plane, but that its orientation relationships to the parent matrix were dependen on the thickness of the transforming region. It has also been pointed out by Breedis6 that in thin films the absence of the constraints imposed by a bulk matrix can cause different requirements for the martensite plate interface and lattice distortion which is reflected bv a dependence of the transformation mechanism on the specimen geometry. In the case of the spontaneous reaction reported for Au-Zn alloys7 a definite habit plane was determined and shown to correspond with the habit plane calculated on the basis of the Bowles and Mackenzie crystallographic theory of the martensite transformation.&apos;- lo EXPERIMENTAL PROCEDURE A series of titanium alloys containing 5.5, 5.9, 6.4: 7.6, 8.0, 11.3, and 18.7 at. pct Cr were prepared from high-purity (iochrome) chromium and iodide-grade titanium by levitation melting. The experimental procedure used for sectioning the alloys and for thin-foil preparation has been reported in an earlier paper.&apos; X-ray diffraction patterns were obtained using a 114-mm Debye-Scherrer camera and Cu-Ka radiation. In order to minimize fluorescent radiation from the specimen, a strip of aluminum foil 0.002 in. thick was placed in contact with the X-ray film. After a specimen was removed from the camera: an additional X-ray experiment was performed, to insure that no spurious lines were obtained due to the presence of the aluminum foil. The interstitial impurity analysis for the alloys used in this study was as follows: oxygen, 100 ppm: hydrogen, 30 ppm: nitrogen, 28 ppm. EXPERIMENTAL RESULTS Electron diffraction patterns were obtained from several individual crystals of the spontaneous marten-

Jan 1, 1970

By G. L. F. Powell, L. M. Hogan

Samples of silver and Ag-O alloy, 0.12 wl pet, have been undercooled to a maximum of 250°C by melting in a slag of commercial soda-lime glass. Grain refinement occurred in undercooled silver samples whet the degree of undercooling exceeded a critical value which varied from 153° to 175°C, depending on the melting conditions. In under cooled Ag-0 samples, the grain structure was fine and equiaxed at degrees of undercooling larger than 50oC. The grain refinement in both silver and Ag-0 alloy samples was the result of recrystallization during or immediately following freezing. When the oxygen content was increased, the degree of undercooling at which recrys-tallization occurred decreased. INVESTIGATIONS of the grain structure of strongly undercooled melts of nickel and cobalt1-4 have revealed a change in room-temperature grain size as a function of undercooling. At large degrees of undercooling the grain size was fine, but samples undercooled by smaller amounts exhibited a much coarser grain structure. The reported undercooling at which this grain size transition occurred in nickel varied between 140° and 175&apos;C undercooling. Walker1 attributed this grain size change to an enhanced nuclea-tion rate. Cavitation at a melt-solid interface was con sidered to generate a high-pressure wave. This, in turn, produced nucleation in the liquid as a result of an elevated freezing point increasing the undercooling in the melt. A similar grain size transition was observed in undercooled silver between 133° and 153°C undercooling.5 The grain size transition in undercooled silver was found to be dependent on oxygen con tent. More recently, the authors5 showed that a simila transition occurred in an undercooled Cu-O alloy, =0.08 pet, but not in oxygen-free copper undercooled more than 200°C. The transition was attributed to recrystallization of the solidified grains after freezing; the greater the degree of undercooling, the more complete the recrystallization. In view of this effect of oxygen content on grain size transition, additional undercooling experiments were carried out with silver. This paper reports the results of these studies on silver and Ag-O alloys undercooled more than 50°C EXPERIMENTAL 1) Materials. The silver was supplied by Matthey Garrett Pty., Ltd., as fine granulate, silver mini&apos; mum 99.9 pet, and contained the trace impurities shown in Table I. The silver originated from two different refineries resulting in the difference in impurity content shown in Table I. However, the variation in composition had no discernible effect on the undercooling behavior. Samples of the silver were melted in an open-ended, vertical, cylindrical resistance furnace, wound with Kantha] wire, the temperature of which was adjusted by means of a variable transformer, 2) Melting and Undercooling. Bulk samples of silver were undercooled by preoxidation and removal of oxidized impurities in a glass slag, as previously described,5 but special precautions were necessary to allow comparison between oxygen-free silver and a Ag-O alloy. Samples free of oxygen were prepared by first melting weights of granulate varying from 200 to 400 g in clean vitreous silica crucibles in air for approximately 15 min. A sample was frozen and a complete glass cover formed over the top surface by adding granulated glass rod. The sample was then melted and frozen several times until no oxygen bubbles appeared in the glass slag. Subsequent thermal analysis indicated that the oxygen content was negligible. The glass acted as an efficient barrier to absorption of oxygen from the air. Ag-O alloys were prepared by melting silver in silica crucibles with only a partial cover of glass, so that the top surface of the melt was in contact with air. The silver was held molten in air for several hours at 1000° to 1050°C, in the expectation that the oxygen in solution would reach equilibrium with atmospheric oxygen. At T - 1000°C the equilibrium oxygen content was calculated as 0.12 wt pet O, using Sieverts law and Eq, [l], derived by Mizikar et a!.7 from the data of Sieverts and Hagenacker8 for 1 atm of oxygen:

Jan 1, 1970

By J. Lewis, J. Harper, M. F. Ashby

A grain boundary in a metal interacts with second-phase particles, which exert a pinning force (first estimated by Zener) on the boundary opposing its motion. We have computed the shape of boundaries which interact with more or less spherical second-phase particles and have constructed a soap-film model to reproduce the shape of the boundary surface. An important result is that measurement of this shape allows the pressure, or driving force, on the boundary to be measured. We hare applied this technique to grain boundaries in two alloys and hate measured the pinning force exerted by single second-phase jwrticles on the boundaries. It is in good agreement with Zener&apos;s estimate. J\. boundary between two grains, or two bulk phases, interacts with small inclusions or particles of a second phase, whether they are gas or solid. This interaction means that the boundary, forced to migrate by a difference in free energy between the material of the two grains or phases which it separates, exerts a force on a particle which it touches, tending to drag it forward. (The movement of inclusions through metals under the influence of this force, has, in fact, been observed. 1-3) Equally, the particle can be thought of as exerting a pinning force on the boundary, tending to hold it back. Zener (in a celebrated private communication4) first realized that this interaction, and the resulting pinning force, existed. His calculation of its magnitude was crude but adequate: a spherical inclusion of radius r blanks off an area nr2 of the boundary on which it sits; since the boundary has an energy of rMM x per unit area, the blanlung-olf decreases the energy of the system by MM: this energy is returned to the system if the boundary is pulled free from the inclusion— a forward movement of the boundary by a distance r will do this—so that the maximum pinning force is Trrym.M- A similar argument can be made for inter-phase boundaries. The nature of the particle itself did not enter this, or two subsequent treatments.5,6 When it is considered, tic leifthe energyoftheb a) The boundary may enter and pass through the particle if the energy of the boundary is lower within the particle than in the matrix. Fig. l(r/). Certain coherent precipitate particles behave like this. h) More usually, the boundary will bend round the particle, enveloping and bypassing it. Fig. l(b). In doing so, it changes the structure and energy of the interface between the particle and its matrix. This means that the boundary does not touch the particle surface at right angles, as early treatments assumed,5&apos;9 but at some angle a which depends on this change in surface energy of the particle, and can be calculated from the equilibrium of surface tensions. Most precipitate particles and inclusions behave like this. Gas bubbles or liquid drops can be regarded as belonging to either group. The progress of bypassing is conveniently measured by the angle shown in Fig. 1. When the nature of the particle is ignored, its maximum pinning force is exerted when - 45 deg. When it is considered, this critical value of is found to depend on a and thus on the nature of the particle. The maximum pinning force lies between nryMM and 2jtjMM (Appendix 1). not very different from Zener&apos;s result. In reality, a boundary between crystals has a specific energy and tension which varies with the orientation of the boundary. Furthermore, recent experiments7 indicate that such a boundary is not atom-ically smooth, but has steps on it: migration of the boundary corresponds to the sweeping of these steps across the boundary surface, like the Frank model of crystal growth from the vapor. This means that the interaction of a boundary with particles should really be considered in terms of the way in which particles hinder the movement of these steps. To suppose a grain boundary or interphase boundary to be smooth, and to ignore the variation of its energy with orientation, is to liken it to a soap film. The advantage of this soap film approximation, which we have used throughout this paper, is that interaction energies and boundary shapes can be calculated easily. We have done this by numerical computation and by using a soap film model, and have compared the results with grain boundaries in an aluminum-based and a copper-based alloy. It turns out that the shape of the boundary which bulges between particles allows the pressure an it to be calculated; that is, the local driving force an the boundary can be measured. This has allowed us to check the Zener relationship experimentally.

Jan 1, 1970

By A. H. Larson, R. J. McClincy

Laboratory-scale decopperizing experiments with multiple sulfur addifions were conducted at 330°C on ternary Pb-Cu alloys containing, as the third elenlent, Sn, Ag, As, Sb, Bi, Zn, and Au, common impurities in lead blast-furnace bullion. For silver and tin, an increased rate and extent of &apos;cofifier removal was obsert3ed. The elements As, Sb, Zn, Au, and Bi had no effect or less effect as compared to sulfur additions with no i)npurily additions. THE production of primary lead in the blast furnace yields an impure lead frequently containing such impurities as copper. antimony. arsenic. tin, gold, silver iron, oxygen. and sulfur. By cooling this lead to a temperature near its melting point. most of the iron, sulfur, and oxygen and part of the other impurities are removed in the form of a dross. With incipient solidification of the lead, the copper concentration wil have been reduced to 0.02 to 0.05 pct. depending upon the concentration of the other impurities. according to Davey.&apos; Since copper interferes with the treatment of silver after the desilverizing process, it is desirable to decrease the copper content of the lead still fur-ther before the lead is desilvered. The decopperizing of the lead is accomplished by stirring a small quantity. approximately 0.1 pct. of elemental sulfur into the lead at a temperature near its melting point, 330" to 360°C. The copper is removed as a copper sulfide which constitutes a small fraction of a voluminous dross consisting mostly of lead sulfide and entrained metallic lead. The residual copper concentration following the decopperizing operation is frequently as low as 0.001 to 0.005 pct. Thi fact has aroused considerable interest because the equilibrium copper concentration of lead in contact with solid PbS and solid Cu2S is at least an order of magnitude greater, 0.05 pct Cu at 330C. 1, 2 Most investigators have suggested that various impurities in the lead bullion are responsible for the very low copper concentrations frequently encountered in practice. There is little agreement, however? as to which of the impurities are helpful and which are not.3"11 Also. few investigators have sought to explain the mechanisms responsible for the removal of copper to very low concentrations. Willis and Blanks9 have proposed that a nonstoichiometric copper-deficient cuprous sulfide forms in place of the supposed Cu2S. Being copper-deficient, this sulfide phase would possess a low copper activity, and the diffusion of copper dissolved in the liquid lead into this phase would be greatly facilitated. Pin and wagner2 have investigated the removal of copper from liquid lead by studying the effect of impurity-doped lead sulfide on the decopperizing of pure Pb-Cu alloys. Samples of the doped PbS were held in contact with copper-saturated lead for 1 week at 33&apos;7°C. They reported a beneficial effect on decopperizing with bismuth and antimony and no effect with tin or silver. which is directly opposite to the results observed in practice and those reported by Davey 3 and this studv. The purpose of this paper is to describe the effects of certain additive elements on the extent to which copper can be removed fro111 liquid lead by successive additions of sulfur. The impurity elements were added individually to prepared Pb-Cu alloys. The resulting ternary alloys as well as a binary Pb-Cu alloy were then decopperized with repeated additions of sulfur. EXPERIMENTAL Materials. Granulated test lead with a purity of 99.999 pct and the additive elements Cu. Ag. Sb. Bi. Zn. Sn. and Au with purities of 99.99 pct were American Smelting and Refining Co. research-grade materials. The major impurities in the lead were 1 ppm each of iron and copper. all others being less than 1 ppm. The arsenic used was a technical-grade arsenic of 98+ pct purity. Reagent-grade flowers of sulfur were melted under argon to provide small pieces free of fines. Apparatus. The decopperizing experiments were carried out in a 25-mm-OD by 375-mm-long Pyrex tube sealed at one end. The tube was mounted vertically in a resistance-heated. hinge-type tube furnace controlled to within ±lcC. Temperature measurement was accomplished by means of a standardized chromel-alumel thermocouple sealed into the base of a silica. paddle-type stirring rod. All decopperizing experiments were carried out under an argon atmosphere. Procedure. A Pb-Cu starting alloy containing 0.05 pet Cu was prepared under carbon and poured into cold tap water to produce shot. The ternary alloys were prepared by melting together 100 g of the starting alloy and a sufficient amount of the impurity element to yield the desired concentration. The resulting alloy was then homogenized in a Pyrex tube at 450C with continuous stirring. The furnace temperature was then lowered to the operating temperature of 330°C. When thermal equilibrium had been obtained at the operating temperature, individual additions of 0.2 pct (0.2 g) of solid sulfur were added to the melt and stirred in. Stirring was continued for a period of 3 min. discontinued for 5 min. and resumed for the remaining 2 min of a 10-min cycle. This cycle was repeated for as many sulfur additions as desired. When the decopperizing experiment had been completed the lead bullion was quenched and samples of the bullion and dross phases were taken for analysis. Results. The results obtained in the decopperizing

Jan 1, 1970

By Earl G. Alexander, W. R. Runyan

The nature of silicon growths which form on amorphous silicon dioxide during epitaxial deposition in unmasked areas was investigated. Octahedral silicon crystals nucleate and grow on the oxide at pinholes, adjacent to foreign matter, and probably at oxygen-deficient sites. Generation of these oxygen-deficient sites on the oxide surface is shown to be thermody-namically favorable under usual deposition conditions. The conditions which affect saturation (SiCl4 concentration, flow rate, temperature) determine the number of sites on which silicon will nucleate and grow. Reaction product HCl is likely responsible for suppressing silicon growths on the oxide near exposed areas of silicon. Evidence is given that the oxide-silicon interaction occurs in hydrogen only and involves water as a transport agent. SINCE shortly after the first epitaxial layers of silicon and germanium were grown over large areas, efforts have been underway to grow selectively on specified portions of the surface. Such selective growth should be particularly applicable to monolithic integrated circuits. Usually the approach has been to use a mask to delineate the desired area,&apos;-7 but the VLS methoda can also be used, and direct programmed atomic beam deposition has been suggested.9 The masking techniques may embody two forms: in one the mask is a material which can later be removed along with any growths attached to it; in the other, attempts are made to prevent growth on the mask, so that over a whole slice the only deposition that will occur is in the unmasked regions. The present work is concerned with only the latter case, and specifically with silicon. Joyce and Baldrey1 suggested the use of thermally grown silicon dioxide as a mask material and showed that for the proper choice of operating conditions (total flow, SiCl4 concentration, total deposition time, and temperature) spurious growth could be minimized and that if the deposition times were not too large, and the criteria not too stringent, a usable process occurred. Later Schnable et al 4 observed that after extended deposition times, during which most of the silicon dioxide became covered with growth, there was always a band of oxide around the exposed silicon area that was quite free of spurious silicon crystallites. This latter observation is particularly intriguing since it extends the hope that with proper precautions masking can be much more versatile than it presently is. EXPERIMENTAL PROCEDURE Silicon substrates oriented to within 1.5 deg of a (111) plane were prepared by etch polishing lapped

Jan 1, 1967