Search Documents

By John W. Burd

A system is described for the simultaneous deposition of epitaxial layers on as many as eight substrates. A high degree of uniformity of both physical and electrical characteristics is achieved in the films. Variation of film thicknesses is consistently less than ±10pct within a wafer and from wafer to wafer within a run with the variation typically on the order of 55 pct. Composition variation of GaAs1-x PX layers within a wafer and from wafer to wafer within a run is consistently less than 51 pct. Electrical evaluation of the films by several techniques indicates excellent doping uniformity within a wafer and from wafer to wafer within a run. Mobilities for lightly doped GaAs films at 300°K are consistently >6000 cm2 v-1 sec-1 and mobilities > 7000 cm2 v- 1 sec-1 are regularly attainable. Techniques for the preparation of material with carrier concentrations from 1 x 1015cm-3 to 1 x 1019 cm-3 n-type and 5 x 1016 to 5 x 1018 cm-3 p-type are discussed. METHODS for the preparation of 111-V compounds by vapor phase reactions have been extensively reported in the literature.1-6 Almost all of the apparatus described for these various methods are suitable for processing one or at the most a very limited number of wafers simultaneously. With the recent rapid advances in the use of vapor grown GaAs for microwave oscillators and GaAs1-xPx as visible light emitters the requirements for these materials are steadily increasing. In order to satisfy these requirements it is necessary to move from a laboratory scale apparatus to one which is capable of processing a large number of wafers simultaneously. Desirable features would be a high degree of uniformity among the wafers and good reproducibility from run to run. The apparatus to be described fulfills these requirements very well. DISCUSSION The various methods reported in the literature can be classified under three headings: 1) closed tube, 2) open tube, and 3) the close-spaced method. Of these three the open-tube method is the most amenable for scale-up to a manufacturing process. It is the most versatile and the various operating conditions can be more precisely controlled than with the other two methods. A number of chemical reactions may be used to achieve vapor-phase growth of 111-V compounds. Sev-era1 of the more generally used reactions are shown in Fig. 1. All of these reactions have the following points in common: 1) generation of a volatile group III(Ga) species by the reaction of the transport agent (halide or HC1) with either Ga or GaAs, 2) introduction of the Group V(As and/or PI component, 3) a method of adding dopant, if desired, and 4) a region in which deposition from the vapor will occur and form as a single crystal epitaxial film on the substrates. The laboratory scale reactors permit the hot re-actant gases to flow into the relatively cooler deposition zone and pass successively over the several substrates which are arrayed along the long axis of the tube parallel to the gas flow. With this arrangement the composition of the reactant stream is continually changing as solid material is deposited on each successive substrate. As a result of this changing gas composition the reaction driving force also changes from substrate to substrate and the degree of uniformity of layer thickness, doping level, and so forth, is poor. This effect can be partially overcome by imposing a controlled temperature gradient along the deposition region to compensate for change in gas composition. However, even when this is done variations in layer thickness on the order of 30 to 40 pct are common and as high as 50 pct are frequently experienced between adjacent wafers in the tube. To expand this arrangement to a large number of wafers would only increase the nonuniformity from the first to last wafer in the line. From the above discussion the two undesirable features of changing gas composition and temperature gradient become evident. A reactor system which eliminates or minimizes these undesirable features is one in which the apparatus is mounted vertically as shown schematically in Fig. 2. The vertical mounting permits the disposition of a number of substrates on a suitable support so that all wafers are at the same vertical height in the furnace and hence at essentially the same temperature. By using only a single row of wafers the reactant gas mixture passes over only one substrate in its path through the reactor. Thus the two undesirable features of changing gas composition and temperature gradient are minimized. An additional design feature which further minimizes temperature variations is rotation of the substrate holder. Rotation serves to integrate any radial temperature gradient existing around the resistance heated furnace. A photograph of a reactor assembly at the completion of a run is shown in Fig. 3. MATERIAL PREPARATION Apparatus. Although any of the several chemical systems shown in Fig. 1 are adaptable for use in this apparatus the one generally used is System 2, the hydride synthesis system. This system has been de-

Jan 1, 1970

By Akinobu Kasami, Keiji Maeda, Makoto Naito, Masaharu Toyama

Gap electroluminescent diodes have been prepared by the vapor phase diffusion of zinc into n-Gap crystals which were grown from a gallium solution (10 wt pct Gap) doped with tellurium and Ga203. A marked improvement in the efficiency of the red electrolumines -cence has been achieved by heat treatment after diffusion. External quantum efficiencies of diodes annealed under optimum conditions are 0.2 to 0.6 pct at room temperature, or about 200 times higher than the efficiencies of diodes quenched after diffusion. The optimum dopant concentrations in the gallium melt from which the crystals were grown are 3to6 x at. pct Te and 4 to 8 x 10-2 mol pct Ga203. The efficient diodes are characterized by linearly graded junctions with an i-layer 0.1 to 0.2u thick. Annealing increases the emission intensity by a factor of 20 to 50 and decreases the current density to 1/3 to 1/8 that of quenched diodes at a given bias. The decrease in current is attributed to an annihilation of deep recombination centers in the depletion layer. The increase in emission intensity is interpreted in terms of an increase in lifetime of minority carriers and an increase in the relative intensity of red-to-infrared emission. The dependence of these quantities on the tellurium and oxygen doping levels is also discussed. A number of studies have been made of the red light emission from for ward-biased Gap diodes.' At room temperature this emission band is centered at 7OOO? with a spectral width of nearly 1000?. Low-tempera-ture photoluminescence indicates that this emission is due to either the radiative annihilation of an exciton bound to a pair of zinc and oxygen atoms substituting on nearest neighbor lattice sites2,3 or the radiative recombination of an electron bound to this Zn-O pair with a hole bound to an isolated zinc shallow acceptor.3 An emission band is also observed with a spectral peak at 9800?. This infrared emission has been shown to be due to the recombination of an electron trapped at an isolated oxygen deep donor with a hole trapped at an isolated zinc acceptor.4 The red emission from Gap diodes is fairly efficient at room temperature because the nearest neighbor Zn-0 pair forms a deep electron trap at 0.3 to 0.4 ev below the edge of the conduction band.2'4 In diodes grown by liquid epitaxy an external quantum efficiency of 2.1 x 10-2 (photon/electron) has been attained by heat treatment at relatively low temperatures.5 This heat treatment was found to increase the efficiency by a factor of 3 to 6. However, no detailed studies have-been reported on the effects of heat treatment. We can only cite Onton and Lorenz's work6 on the change ; in the relative intensity of red-to-infrared emission. Heat treatment has also been tried on junctions built in during growth, but contrary to expectations the efficiency decreased. In-diffusion is a simple and controllable method of fabricating p-n junctions. For Gap, zinc is generally used to form a p-type layer on n-type crystals. The emission efficiencies of in-diffused diodes are, however, extremely low in comparison with liquid epitaxial diodes.' Although efficiencies as high as 2 x 1O-3 have been reported, values from 10-6 to 10-4 are generally obtained by typical diffusion techniques. Out-diffused diodes are known to be a little more efficient than in-diffused diodes. Nevertheless, the quantum efficiency is at most 7 x 10- 3 and ordinarily of the order of 10-4.8 NO results have been reported on heat treatment of either in-diffused or out-diffused diodes. This paper reports a marked improvement in the efficiency of the red emission observed for in-dif-fused diodes as a result of heat treatment after diffusion. The method described reproducibly yields diodes with external quantum efficiencies of 2 to 6 x 10-3. The observed dependence of efficiency on annealing time and on doping level will be discussed in terms of the lifetime of minority carriers and the formation of Zn-O complex pairs. EXPERIMENTAL A) Diode Fabrication. The n-Gap crystals used in this study were grown from a saturated gallium solution by a slow cooling method.8 The Gap content in the gallium melt was fixed to 10 wt pct corresponding to a growth temperature of about 1100 Tellurium was chosen as the n-type dopant and added to the melt in concentrations ranging from 0.001 to 0.06 at. pct. Oxygen was added in the form of Ga2O3, whose concentration was varied from 0.004 to 0.2 mol pct. The resulting crystals were platelets with well-developed (111) surfaces. Typical electrical properties were Hall mobilities of 130 to 30 sq cm per v-sec and carrier concentrations of 1016 to 10" cm-3 at room temperature. Diodes prepared from crystals with relatively low doping levels, in which u = 130 to 100 sq cm per v-sec and n = 0.6 to 6 x 1017 Cm-3, were examined in detail. The p-n junctions were produced in these n-Gap crystals by the diffusion of zinc from the vapor phase by the following procedure. The platelets were carefully lapped on both sides to a thickness of 150 to 200 u while maintaining the (111) orientation. After being etched in hot aqua regia, the crystals together with the zinc were sealed in an evacuated 12 mm ID quartz ampoule 20 cm long. The crystals and the zinc were then separated from each other at opposite ends

Jan 1, 1970

By James C. Word, R. M. McLouski

This paper will describe the development and application of large-scale integration techniques employed in the fabrication of a germanium multielement array. The array consists of 100 by 228 PNP bipolar transistors fabricated on 5 mi1 centers. Back-biased p-n junction techniques are used for electrical isolation of the individual elements. The end use of the array is a high resolution, large area IR sensor. The monolithic array is fabricated in 1 ohm-cm p-type germanium epitaxially deposited on 6 ohm-cm n-type substrate. Epitaxy was accomplished through the hydrogen reduction of germanium te trachloride. Di-borane was used as the dopant. Base regions are achieved by the diffusion of arsenic from doped oxide or arsine sources. Oxide-masking of the arsenic im-pzlvity was achieved by the chemical deposition of a boron doped glass. The emitter is formed by an aluminum alloy diffusion technique. Vacuum deposited aluminum is used for the emitter, interconnections, and for the contact and bonding pads. ALTHOUGH a great volume of literature pertaining to the development of large scale integration techniques (LSI) has been published for silicon and in particular silicon imaging applications,' to date only a small number of similar devices have been constructed using germanium technology.' Since the physical and chemical properties of germanium are vastly different from those of silicon, the fabrication technology for integrated structures in germanium is also different from that of silicon. In particular germanium does not possess a stable oxide as can be grown on silicon by heating in an oxidizing ambient for masking of dopants and passivation. This paper describes the application of germanium LSI techniques employed in the fabrication of a multielement infrared sensor array. The array is used in a high resolution, large area infrared sensor for operation in the 0.8- to 1.5-u spectral range. Back biased p-n junction techniques are used for electrical isolation of individual elements. Discrete germanium devices have been fabricated routinely for some time. However, mainly due to the lack of a suitable mask for selective doping and the high current leakages inherent in germanium p-n isolation, few monolithic germanium structures have been constructed. THE INFRARED MOSAIC A cross-sectional view of the array is shown in Fig. 1. The monolithic structure consists of 12,800 PNP transistor elements in a 100 by 128 matrix fab- ricated on 5 mil centers. The emitters of each line of transistors are connected together using aluminum interconnects while the strip collectors are connected together in series at right angles to the emitter lines. The selection of this structure is dictated by the readout technique involved. Access to each element transistor is obtained by applying a bias voltage to a particular collector strip and separately interrogating each emitter row. A charge storage, i.e., an integration mode is used for reading out this particular array Construction techniques available for use with germanium do not include a selective p-type diffusion capability for surface concentrations greater than 10" per cu cm and junction depths greater than about 10 u. This fact limits the type of structure that may be used. Therefore, an array of PNP transistors that did not employ p-type diffusions was chosen. The structure was fabricated by growing a 1 ohm-cm p-type epitaxial layer on a carefully prepared 6 ohm-cm n-type substrate. N-type dopants were used for the isolation and base diffusions and alloyed aluminum was used to form the emitter junctions. The array was then completed by evaporation of aluminum interconnections and contact pads. SUBSTRATE AND SUBSTRATE PREPARATION Germanium substrates of (111) orientation grown by both Czochralski and zone leveling techniques were utilized for mosaic fabrication. Czochralski substrates were preferred because of the lower dislocation densities available in this type of material. Dislocation densities for the Czochralski material were typically less than 3000 per sq cm, while those for the zone leveled material were typically less than 5000 per sq cm. All substrates were uncompensated to minimize thermal conversion problems in subsequent epitaxial and diffusion processing. Both in-house and vendor polished wafers were used. The in-house polishing technique employed consisted of an initial gross chemical etch in CP4 to remove saw damage from both surfaces. This was followed by a chemical-mechanical polishing operation of one side of the wafer. The chemical-mechanical polishing solution used was Lustrox 1000 (Tizon Chemical Co.), and consists of zirconium dioxide, sodium hypochlorite, water and a surfactant. The wafer thickness before and after polishing was typically 0.020 and 0.010 in, respectively. THERMAL CONVERSION The problem of thermal conversion of both the substrate and epitaxial layer was particularly acute because of the relatively low carrier concentrations employed in both regions. This problem has been encountered by other workers in the past.3 Without special treatment before epitaxial growth substrate conversion (n-type to p-type) and changes in the re-

Jan 1, 1970

By R. A. Burmeister, G. P. Pighini, P. E. Greene

An open tube vapor phase epitaxial growth system has been used for large area (multiple substrate) growth of GaAs1-xPx on GaAs substrates. The GaCl-GaCl transport reaction is used with either a GaAs or Ga (nonsaturated) source. Selenium and tellurium have been used for donor impurities, and zinc as an acceptor. The useable substrate area in this system is approximately 20 sq cm. The uniformity of thick-ness of the epitaxial layers are typically better than ±5 pct across a given wafer. Electrical and optical measurerments indicute comparable uniformity in electrical and luminescent properties within a wufer. The application of this system to the large scale pro-duction of GaAs1-x Px for display devices, both discrete and arrays, is discussed. Typical electrical and luminescent properties of light emitting diodes fabricated front material produced by this technique are presented. THE most promising materials currently being utilized for visible injection electroluminescence are GaAs1-xPx, Ga1-xAlxAs, and Gap. All have reasonably efficient emissions in the red portion of the visible spectrum at room temperature; Gap also has an efficient green emission.' At present, GaAs1-xPx has a technological advantage over Ga1-xAlxAs and Gap for display applications, since relatively large (several sq cm) areas of GaAs1-xPx suitable for use in electroluminescent devices may be readily grown by vapor phase growth techniques. In contrast, the preparation of Gap and Ga1-xAlxAs for electroluminescent device applications generally employs solution growth techniques which are at present not well suited for the growth of large areas of uniform thickness and doping level. The capability of uniform growth over large substrate areas and the use of multiple substrates is necessary for the practical utilization of electroluminescent devices. This is particularly important when quantity production or monolithic devices are required. Furthermore, in many display applications arrays of light emitting devices are used, the individual elements of which are of a size resolvable by the unaided eye. Thus the overall dimensions of display are substantially larger than those of most semiconductor devices. It is also necessary to achieve a high degree of control over the growth parameters to attain the required degree of reproducibility of materials properties for electroluminescent devices. In the case of GaAs1-xPx it is necessary to accurately and precisely control the phosphorus content of the alloy, both on a macroscopic and microscopic scale, in addition to the parameters generally associated with epitaxial growth such as thickness and doping level. This value is critical, as it has a major effect on the performance of electroluminescent devices. This paper describes the epitaxial growth of GaAsl-xPx suitable for electroluminescent display devices using a system developed specifically for this purpose, and which contains several novel features. The results of studies of selected physical properties of the epitaxial layers are also discussed. Finally, a brief summary is given of the characteristics of display devices fabricated from GaAsl-xPx grown in this system. EXPERIMENTAL A) Reactants. A number of techniques suitable for the vapor phase epitaxial growth of GaAs1-xPx have been reported in the literature.'-' The method selected for this investigation is that in which the Ga is transported by the GaC1-GaCI3 reaction in an open tube process. The results reported here were obtained using either the combination of GaAs, AsC13, and pH3, or Ga, AsH3, pH3, and HC1 as the initial re-actants.* The ASH3 and pH3 were obtained as dilute *Several different sources of supply were used for these reactants, y~elding comparable results._____________________________________________________ mixtures in HZ; the HC1 was obtained from the reduction of AsC13 by Hz at elevated temperatures. Both selenium and tellurium were employed as donor impurities, and zinc as an acceptor impurity. Selenium was introduced in the form of H2Se, tellurium in the form of tellurium-doped GaAs, and zinc in the form of diethy1 zinc. B) Apparatus. The prinicipal difference between the apparatus used in the present study and that of Tietjen and Amick,8 in addition to size and other related design features, is that RE induction heating is utilized in place of resistance heated furnaces. Induction heating was selected for this application because it appears to have several advantages, including: 1) It is possible to keep all fused silica portions of the apparatus at temperatures well below those of the reaction zone, thus minimizing a possible source of contamination. 2) The thermal mass of an induction heated system can be made small, thus reducing the total time required for the growth process. 3) Sharp temperature profiles (desirable for high deposition efficiency) are easily achieved. 4) The volume of the system for a given substrate area can generally be made smaller than a comparable resistance heated unit. This results in shorter time

Jan 1, 1970

By Donald E. Meyer

High current-low voltage EB-gun evaporation in an oil-free ultra-high vacuum system was found to be necessary, though not sufficient, for stability (300°C, 106 v per on) of aluminium gate MOSFET's and MOS capacitors not stabilized by a phosphorous glaze. five characteristics of the equipment used: 1) Vacuum purification of the aluminum charge, 2) Ionization of the evaporant by the electron beam, 3) X-ray formation, 4) Residual gases during evaporation, and 5) Metal film structure were studied as Possibly significant in MOS fabrication. EVAPORATION of contact metals common to the semiconductor industry historically has been accomplished with oil diffusion pump systems and various resistance heated evaporant sources as dictated by the type of metal evaporated. To meet a need for greater reliability of semiconductor devices, other metallization methods were developed. A good example would be application of the moly-gold contact system to integrated circuits with deposition by RF or triode sputtering.' More recently, fabrication of stable metal-oxide-silicon devices and circuits has put new demands on metallization. The purity of the thin metal films composing MOS structures is critical, particularly at the metal-oxide interface, and ultra-high vacuum metallization using sputter-ion pumping and electron beam gun (EB-gun) evaporation are well suited for the task. At this laboratory aluminum has been the most common contact-gate metal for both MOS capacitors and MOSFET's. In the earliest work with MOS capacitors, aluminum was evaporated from wetted tungsten filaments using both diffusion pump and ion pump vacuum systems. In spite of clean oxide techniques these capacitors were unstable under bias-tempera-ture stressing. Only after a switch to EB evaporation of aluminum were stable capacitors produced. Using the same techniques it was possible to make MOSFET's with equivalent stability. Stability data for a discrete MOSFET is shown in Fig. 1. This is a "clean" oxide gate (no phosphorus stabilization or no etch back of a thicker gate) having a thickness of lOOO? thermally grown on the (111) plane. Gate length after diffusion was 0.24 mils, and the devices were hermetically sealed. Stressing conditions were 300°C and 106 v per cm applied alternately as a positive and negative field for 10 min, 50 min, and 4 hr for a total stress time of 10 hr. An initial shift in turn-on voltage of 0.1 v was detected for 10 min of positive bias. All evidence at this laboratory indicated that while EB-gun evaporation of ultra-high purity aluminum was not sufficient for 300°C stability, it did seem to be necessary. There may well then be something inherent in the EB-gun deposition used which enhanced stability, and probably no single factor existed but rather a series of factors. It is the purpose of this paper to report on some of the investigations carried out to learn more about EB-gun evaporation in ultra-high vacuum systems. EXPERIMENTAL DESCRIPTION The EB-gun was self accelerated, had a maximum power rating of 10 kw, and used a water-cooled copper crucible able to hold a 20-g aluminum charge. The electron beam was bent 180 deg and focused by an electromagnet which also provided movement of the beam across the crucible. Normal power conditions in this work were 9 kv and 300 to 600 mamp. The gun can be described as high-cur rent/low-voltage and was quite different in its mechanism of operation from EB-guns with much higher acceleration potentials. An oil-free vacuum system capable of 5 x 10- l0torr, a quartz crystal rate and thickness monitor and a quadruple mass spectrometer completed the evaporation system, Fig. 2. A typical evaporation cycle consisted of a 3 to 4 hr pumpdown to the upper l0-9 range and evaporation at l0? per sec with the pressure in the bell jar not rising above 1 x 10"7 torr. Thickness control was 5 pct or less and could be automatically monitored and controlled. Five phenomena associated with the EB evaporation and considered as possible contributors to Ma performance included a purification effect, ionization of evaporating aluminum, X-rays, constitution of vacuum ambient during evaporation, and film structure dependence upon evaporation rate. These phenomena are now discussed. Vacuum Purification. The design of the EB-gun permitted purification of the aluminum charge by vacuum outgassing. Particular features included an efficiently water-cooled copper hearth with a capacity of over 20 g of aluminum and the capability for sweeping the beam across the charge. Such capacity meant that aluminum had to be added only after about every fifth evaporation. A new charge was not required each evaporation as is necessary with filament evaporation. An oxide "scum" which appeared on the charge could be completely cleared from the top hemisphere of the charge by sweeping with the beam prior to opening the shutter. An indication of the purifying effect was obtained by a series of analytical measurements on incoming aluminum, after melting but with little vacuum out-gassing, after 30 min outgassing, and the evaporated film itself. Either a solids (spark source) mass spectrometer or an emission spectrometer were used for analyzing the aluminum charge. Analysis of the evapo-

Jan 1, 1970

By S. Sumski, M. B. Panish, R. T. Lynch

The liquidus isotherms in the gallium-rich corner of the Ga-Al-P phase diagram have been determined from 1000" to 1200°C and at I100°C the corresponding solidus isotherm was obtained. A simple thermody-namic treatment which permits calculation of the solidus and liquidus isotherms is discussed. A technique which was previously used for the growth of GaxAl1-xAs was used for the preparation of solution epitaxial layers of GaxAl1-xP and ALP. An approximate value of 2.49 i 0.05 ev for the band gap of Alp at 300°K was obtained and the ternary phase data were used to estimate a value of 36 kcal per mole for the heat of formation 0f Alp at that temperature. The Gap-A1P crystalline solid solution is one in which there exists the possibility of obtaining crystals with selected energy gaps, within the limits imposed by the energy gaps of Gap and Alp. Such crystals are of considerable interest because of their potential value for optoelectronic and other solid-state devices. Furthermore, it has been amply demonstrated for GaAs and GaP,'-7 that device, or bulk materials grown from gallium solution generally have more efficient radiative recombination than materials prepared in other ways. This presumably due to the lower gallium vacancy concentration in such material.= Small crystals of GaXAl1-xP and A1P have been grown from solution,8-10 and A1P has been grown from the vapor," but neither have previously been grown by liquid epitaxy. In this paper we present the ternary liquidus-solidus phase diagram of the Ga-A1-P system in the region of primary interest for solution epitaxy, and discuss the thermodynamic implications of that phase diagram with particular reference to the liquidus and solidus isotherms in the gallium-rich corner of the GaxAl1-xP primary phase field and to the A1-P system. Several measurements of the absorption edge of GaxAl1-xP crystals have been made and the width of the forbidden gap of A1P has been estimated from these measurements. EXPERIMENTAL The differential thermal analysis technique used to determine the liquidus isotherms and the optical measurements used in this work are similar to those described previously12 for the Ga-Al-As system, ex- thermocouples in the thermopile for added sensitivity. The materials used were semiconductor grade Ga, Gap, and Al+ The composition and temperature range at which DTA studies could be done was quite restricted. The upper temperature was limited by the chrome l-alumel thermopile to about 1200°C, and the highest aluminum concentration to about 5 at. pct by low sensitivity caused by the reduced solubility of Gap with increasing aluminum concentration in the liquid. DTA studies were not possible at 1000°C and below because of the low sensitivity caused by low solubility of Gap in the Ga-A1-P system. The cooling rate for these studies was about 1°C per min. No heating studies were done because of limited sensitivity. Supercooling probably does occur, but our experience with other 111-V systems indicates that it is no greater than about 10 to 15.c. Solid solubilities were determined by analyzing epitaxial layers of GaxAl1-xP grown from the liquid, with an electron beam microprobe. The layers were grown on Gap seeds by a tipping technique in which the layer is grown over a short-temperature range (20" to 50°C) on the seed from a solution of known composition. The tipping technique reported by Nelsson1 for GaAs could not be used, particularly for solutions containing appreciable amounts of aluminum, because of the formation of an A1203 scum on the liquid surface. A system was therefore designed, which would effectively remove the oxides mechanically, so that uniform wetting and crystal growth could occur. This tipping technique has already been described in detail." The best control over the composition of the re-grown layer was obtained when the tipping was done at a temperature which corresponded to the temperature of first formation of solid for the solution being used. Generally, therefore, a solution was prepared by adding the amounts of Ga, Gap, and A1 required to yield a solution which would be completely liquid above the tipping temperature with solid precipitating below that temperature. For most of the work reported here, the 1100°C isotherm of the ternary was used. It was generally necessary to heat the solution to 50" to l00. C above the tipping temperature to dissolve all of the Gap in a reasonable length of time. The epitaxially grown layers were used both for optical transmission measurements to aid in the estimation of the way in which the absorption edge changed with aluminum concentration, and for the electron beam microprobe analyses to provide data for the determination of the solid solubility isotherm. RESULTS AND DISCUSSION Liquidus Isotherms in the Ga-A1-P Ternary Phase Diagram: Thermodynamic properties of the system. The only thermal effect studied in this work was that

Jan 1, 1970

By R. P. Dolan, W. M. Shedd, B. L. Buchanan

Analytical expressions are derived from empirical data relating the basic physical device parameters to the radiation dose. To put in perspective and justify the approach taken, the overall problem and methods of attempted solution of radiation damage to transistors in a missile or satellite environment is briefly examined. The main emphasis is on a new criteria for neutron radiation hardness and the device design implications resulting from using this criteria. Contrary to other derived criteria, the derived bipolar hardness relation depends mainly on the initial lifetime rather than the base transit time. It is shown, that contrary to previously accepted beliefs, the inherent attainable radiation tolerance of JFET's is at least a factor of ten greater than the radiation tolerance ofbipolars. Experimental results on JFET's designed for maximum neutron radiation tolerance are presented. The accuracy, utility, and range of possible validity of various published lifetime and carrier removal damage constants are discussed. It is well known that transistors are subject to damage due to radiation at much lower radiation levels than those which would cause structural damage to a missile or satellite. The three principal categories of radiation damage to transistors are displacement, ionization, and thermomechanical. The main cause of displacement damage is fast neutrons, and the main effects which permanently degrade transistor operation are minority carrier lifetime reduction and carrier removal. A second-order effect is mobility reduction. The main cause of ionization damage is gamma radiation which results in permanent effects due to charge build-up in the oxide and transient effects due to photo-current generation. The transient effects are usually handled by using circumvention techniques in the electronics system. The extent of the transistor degradation due to displacement and ionization damage depends strongly on the transistor type, whereas the extent of thermomechanical damage does not depend on the transistor type, but depends mainly on packaging and metallization which are common to all transistor types. The standard transistor types are categorized in Fig. 1 and typical cross-sections are shown in Fig. 2. The original reason for the designation of the two classes of transistors was that the unipolar operation depends on only one carrier (majority) and the bipolar operation depends on both minority and majority car- riers. The bipolar is then seriously degraded by minority carrier lifetime reduction, whereas the unipolar or field effect transistor is not. One of the two principle classes of field effect transistors, the junction field effect transistor (JFET), depends strongly on carrier removal. However, the insulated gate (MIS) is little affected by either carrier removal or lifetime changes, but is virtually eliminated as a radiation resistant device because of permanent charge build-up in the insulator caused by ionizing radiation. The bipolar and JFET suffer only minor effects due to charge build-up in the oxide passivation when compared1 to the MOS. This is because the active region of the MOS is adjacent to or part of the oxide layer. It is then the bipolar and JFET which are potentially resistant to permanent radiation damage. DERIVATION OF RADIATION TOLERANCE RELATIONS Criteria. The bipolar is by far the most widely used transistor type. The usual criteria for hardening bipo-

Jan 1, 1970

By Y. C. Liu, G. A. Alers

Using X-ray pole-figure information and the single-crystal elastic constants, the angular variation of an elastic modulus in a polycrystal can be calculated and compared to measured clalues. This procedure has been used in the past to provide supporting ecidence for a particular choice of ideal orientations in the description o the X-ray pole figure. Since it is well-known that the absolute value oj an elastic modlus oj a poly crystal cannot be acczrralely calculated from the single-crystal elastic constants, there exists a serious question about the reliability oj this modulus method in the study of textures. This paper deriz-es sez3eral simple fornzulae for the ntzgular lariatiion of Young's modulus in the plane o a rolled sheet sing the same ssuinptions which lead to the lack oj detinztion of the absolute c'alre OJ the elastic ,nodulus. The vesults show that, although the absolute isallre oj Young's modulus depends strongly on the assumptions made, the shape of the modulus vs angle curve is not very sensitive to the assumptions and thus can be used reliably in texture studies. Exatzples oj' using this criteriotz are illustrated in the text. NeARLY all poly crystalline materials have some preferred orientation in the arrangement of their grains and hence exhibit anisotropy in their properties. This preferred orientation is usually measured and studied by X-rays which yield detailed information on the statistical distribution of grain orientations in the form of pole figures. The usefulness of these pole figures in theories of texture formation and in predictions of properties has come to depend almost entirely on the characterization of the pole figure by one or more ideal orientations. Unfortunately, this characterization is not unique and different authors often present different ideal orientations to characterize a given pole figure. In order to distinguish between these different choices of ideal orientations, it is necessary to go beyond X-rays and consider other physical properties which reflect the texture' simply because any choice of ideal orientations must predict correctly the observed anisotropy in all physical properties. The most common example of this procedure has made use of the angular variation of Young's modulus with respect to the rolling direction in the plane of a rolled sheet.3 4 Here the anisotropy can be measured easily and it can also be calculated from the single-crystal elastic constants and the chosen ideal orientation. Unfortunately the methods used to calculate this anisotropy were chosen for their simplicity rather than for their mathematical rigor so that there exists a serious question as to the validity of conclusions reached by this modulus method. It is well-known5 that the absolute value of the Young's modulus of a polycrystalline aggregate cannot be calculated very accurately because different mathematical averages over the grains yield different results. It is one purpose of this paper to examine the effect of different averaging procedures on the calculated anisotropy of Young's modulus in the rolling plane of rolled sheets in order to see if this property can be used reliably in texture studies. This particular modulus and type of material were chosen because the modulus is easily measured and rolled sheet is technologically important. The results show that, although different averages yield different absolute values for Young's modulus, the shape of the modulus vs angle curve is not very sensitive to the averaging procedure and thus can be used as a criterion to check the various ideal orientations used to characterize the X-ray pole figure. Examples of this checking procedure are given in the final section. Another purpose of this paper is to examine the possibility of using the anisotropy of Young's modulus as an indicator of texture without reference to pole figures or ideal orientations. Such a use would require that different textures yield quite different shapes for the modulus vs angle curve so that the distinction between textures would be quite clear. The experimental results presented in the final section indicate that there is easily enough variation in shapes to allow modulus measurements to be useful texture indicators, par-

Jan 1, 1967

By C. M. Wayman, S. R. Rauze

It was recently reported' that the initial transformation in (100) iron whiskers grown in the bcc condition can be nucleated on heating in regions of the whisker which are not necessarily the hottest. Although this behavior was observed1 several times, its sporadic nature precluded any documentation other than occasional visual observations during hot-stage optical-microscope heating experiments. The purpose of this communication is to report how, in one instance, the a-y transformation in such a (100) iron whisker was initiated in a region which was clearly at a lower temperature than an adjacent region which did not transform at all.

Jan 1, 1967

By V. Kondic, W. M. Rumball

HUNT and Jackson' have recently suggested that binary eutectics might be classified according to the entropy of fusion of the component phases. Such a classification was in fact proposed2 some time ago by the authors and the purpose of this communication is to present some supporting evidence. As pointed out by Hunt and Jackson' typical metals of low entropy of fusion, ?S, tend to form normal eutectics (using Scheil's terminology3). Where one phase is metallic (low ?S) and the other nonmetallic (high AS) then anomalous structures are formed. These anomalous structures may either be regular (e.g., Bi-Sn)or irregular (e.g., A1-Si). Finally Hunt and Jackson suggest that where both phases are of high ?S regular faceted structures are formed. In the proposed classification2 the ratio of the entropies of fusion is taken as the quantitative parameter, i.e., ?Sa/?SB = E, where e > 1.0. An extensive literature survey has been carried out2,4 for reliable photomicrographs of eutectic structures. The structures, numbering 151, have been classified into two classes, viz., normal (i.e., lamellar and rodlike) and anomalous. The latter group has been arbitrarily subdivided into regular and irregular anomalous eutectics. Details and sources of the structures are given elsewhere.2,4 In the absence of better thermodynamic data the value of E has been determined for each of the systems making the following assumptions and approximations: l) The entropy of fusion of solid solutions may be calculated additively;5 ii) Inter metallic compounds are treated similarly but with the addition of the following term for fully ordered structures.5 -4.573(Ni lOg10 Nl + N2 log10 N2) where Nl and N2 are the atomic fractions. It might be argued that the entropies of fusion are meaningless as the eutectic phases form at some temperature (T,°K) below their true melting point (Ta or Tb). As AS, = Lo/To if La the latent heat of fusion of a, is assumed independent of temperature, the actual entropy change AS& is given by La/Te. The ratio of entropies of fusion then becomes Fig. 1 shows E' plotted against cumulative totals for normal and anomalous systems from which it is seen that normal eutectics predominate at low values of E'; e.g., some 90 pct of normal systems lie below E' = 1.5 with 65 pct of anomalous systems above this value. Fig. 1 also shows that the regular anomalous structures occur mostly at low values of c' possibly suggesting growth more akin to normal than anomalous growth. The choice of c as a classification parameter was based on nucleation concepts, it being observed that metals widely separated in Sundquist and Mondolfo's8 nucleation series were anomalous and those close together normal. The basis of these classification views is really outside the intended scope of this communication but it is hoped to explain them fully in the near future.7 Having arrived at this entropy of fusion classification by nucleation considerations it is interesting to note that Hunt and Jackson's ideas are based on growth considerations. Elsewhere one of the authors4 has shown that an even better classification of the metallographic structures reviewed is obtained from purely conventional heterogeneous nucleation concepts. It has been found that systems having a value of the parameter +b less than 0.28 are almost certainly normal and above this anomalous: V, is the gram atomic volume. This parameter is nevertheless closely related to c and c', these being an approximation of p. The parameter is also related to the other classification3,8,9 which have met with any success in the past. All these classifications are to be reviewed shortly by the authorsU7 A significant thing about the classifications based on

Jan 1, 1968

By B. D. Lichter

In a previous paper, Oelsen, Schuermann, and Hey-nertl pointed out the possibility of obtaining complete thermodynamic functions for alloy systems from calorimetric measurements alone. Specifically, if the heat content of a heterogeneous, multicomponent system is known as a function of temperature and composition, then it is possible to derive the heats, entropies, and free energies of formation of liquid phases and of solid intermediate phases of fixed composition. The authors present equations and graphical techniques applicable to eutectic systems and systems with intermediate phases without solid solubility at lower temperatures. eijering has considered possible extension of the treatment to systems with complete solid-state miscibility at lower temperatures; however, experimental difficulties practically restrict the method to systems with very limited miscibility in the solid state. The purpose of this note is to present an alternative derivation of the heats and entropies of formation of liquid phases and of solid intermediate phases from heat-content measurements alone. Fig. 1 shows three types of phase diagram to which the analysis may be applied. If the heat content is determined as a function of composition x, at temperature T, we have quite generally and the heat of mixing of the liquid is obtained directly from heat-content measurements. If the temperature T lies in a two-phase region (solid-liquid) then

Jan 1, 1967

By Charles L. Mantell, B. G. Shah

SINCE the beginning of the industry, commercial electrolytic manganese has followed the Shelton patent' among others, which called for the addition of sulfur dioxide in controlled amounts, to a purified manganese sulfate-ammonium sulfate at a controlled pH. Every attempt to eliminate sulfur dioxide additions has resulted in operating difficulties, chiefly solution of the deposited metal and lowered current efficiencies. Many replacements for sulfur dioxide have been proposed and studied, chiefly sulfur compounds, urea, thiourea, and proteins similar to those employed successfully in copper refining. It has been suggested that sulfur dioxide is a re-ductant for higher-valence manganese compounds, or that it is reduced by manganese metal in the deposit to hyposulfites or its compounds, and that these might affect the cathode film or the deposition potential, or the overvoltage of manganese. The additives were studied following the method of Mantel1 and Ferment2 who determined the feasibility of current-potential curves as an analytical tool for monitoring manganese elec trowinning solutions for metallic impurities: nickel, cobalt, silver, copper, zinc, molybdenum, cadmium, magnesium, and sodium. The individual effect of each on the standard polarization was determined for a range of concentrations. Not all affected the polarization curve; cadmium, magnesium, and sodium had no effect. Nickel, cobalt, silver, copper, and zinc yielded a correlation between transition current and concentration of impurity.

Jan 1, 1968

By J. A. Lund, A. M. Lawson

THE structures of quenched eutectoid and hypereu-tectoid Fe-Cu alloys have been examined by X-ray diffraction and by optical and electron microscopy. The relevant portion of the Fe-Cu phase diagram is given in Fig. 1. Several values have been reported for the solubility of copper in a Fe at the eutectoid temperature, including 1.4 wt pct by orton' and 2.13 wt pct by Wreidt and Darken.2 Alloys containing 4.8 and 7.5 wt pct Cu have been prepared from two batches of Ferrovac E iron (0.005C and 0.013C) and electrolytic tough pitch copper (99.90 Cu). Specimens of dimensions 0.8 by 0.8 by 0.1 in. were cut from hot-rolled material, austenitized, and quenched into brine at 25°C. Similar specimens were also cooled in the furnace, after a short soaking treatment at 1100°C, at a rate estimated to be 10°C per min in the temperature range of 900" to 500°C. X-ray diffractometry was performed on quenched and on furnace-cooled bulk specimens using iron radiation. Only a bcc phase could be detected in all samples. Lattice parameters were calculated from measurements on four lines, made up of two resolved high-angle doublets. Comparisons were made against a silicon standard. Results are shown in Table I. Based on X-ray diffraction studies of a series of alloys of lower copper concentrations, orton' estimated that the tattice constant of a Fe at 25°C increases 0.0014A per wt pct Cu in solution. Extrapolating Norton's data, it may be concluded that all copper was retained in solution in the brine-quenched 4.8 pct Cu alloy, and that approximately 5 pct Cu was also held in solution on quenching the 7.5 pct Cu alloy. Typical microstructures of brine-quenched specimens, as observed by optical microscopy, are shown in Figs. 2 and 3. Precipitation of the E phase (copper-rich) occurred along the y grain boundaries on quenching the 7.5 pct Cu alloy, rendering them readily detectable in the microstructure, Fig. 3. The original boundaries in the 4.8 pct Cu specimens were gener- ally not discernible, however. This indicates that the z phase which was seen in the higher-copper alloy was probably of proeutectoid origin. Thin foils for transmission electron microscopy were obtained from bulk specimens by spark machining to 0.006 in. thickness, followed by electrolytic polishing. Typical transmission electron micrographs

Jan 1, 1967

By E. T. Turkdogan, Klaus Schwerdtfeger

OelSEN and Maetz1 detected some 20 years ago the existence of a miscibility gap in iron oxide-CaO-P2O5 slags melted in iron crucibles at about 1400°C. Because of the importance of this system for the dephos-phorization of steel in the basic Bessemer process, equilibria between liquid iron and selected iron oxide-CaO-P2Q slags have been measured since by numerous investigators.2-5 When in equilibrium with metallic iron, the iron oxide of the slag is present mainly as FeO. In connection with oxygen-blowing steelmaking processes, it is useful to know the phase relations in the slag system at higher oxygen pressure, when major parts of the iron oxide are present as Fe2O3. This problem was investigated by Turkdogan and Bills7 by equilibrating the oxide mixtures contained in platinum crucibles with CO2-CO mixtures at 1550°C. It was found that increasing the Fe2O3 content decreases the composition range of the miscibility gap strongly so that the miscibility gap has almost disappeared at pco2/pco = 75. This result was refuted by the careful work of Olette et a1.,''' who equilibrated their slags with controlled Ha-H2-Ar gas mixtures. Their equilibrium measurements, at 1600°C and at oxygen pressures of 5 x 10"* and 10"5 atm, showed that the oxidation state of the iron has almost no influence on the formation of the miscibility gap. The present experiments were undertaken to check the previous results of Turkdogan and Bills. The experiments were performed at 1625°C in the strongly oxidizing atmosphere of air (PO2 = 0.20 atm) for which no experimental data are available. About 10 g of slag were melted in platinum crucibles and held at constant temperature for 1 hr. After equilibration, the crucible was rapidly pulled out of the furnace and cooled in air. The platinum crucible was removed from the sample. The two slag layers were carefully separated with a small diamond disc, and the surface of the top layer, which may have changed its oxidation state during cooling, was removed. The slags were crushed and analyzed chemically for CaO, P2O5, Fe2+, and Fetotal. The starting mixtures were prepared by sintering the desired amounts of reagent-grade 2CaO . P2O5 - H2O, CaCO3, and Fe2O3. Sintering and subsequent crushing were done three times to ensure homogenization. Molybdenum wire resistance heating was used. The furnace was provided with a recrystallized alumina reaction tube which was left open to air at the top. The temperature was controlled electronically. The reported temperature was measured with a Pt/Pt-10 pct Rh thermocouple and is estimated to be accurate within +5°C. The composition of the equilibrated melts is given in Table I. For the graphical illustration of these quaternary slags the type of projection suggested by Trömel and Fritze10 was used. In this representation, Fig. 1, the composition point of a mixture within the tetrahedron Fe2O3-CaO-P2O5-FeO is projected into the Fe2O3-CaO-P2O5, triangle (triangle I) so that the direction of projection is parallel to the side FeO-Fe2O3, and into the triangle Fe2O3-P2O,-Fe0 (triangle 11) so that the direction of projection is parallel to the side CaO-P2O5, of the tetrahedron. The projected point has the coordinates wt pct CaO, wt pct P205, and wt pct (FeO + Fe2O3) in triangle I and wt pct FeO, wt pct Fe2O3, and wt pct (CaO + PzO5) in triangle 11. Both triangles are turned into the same plane around the Fe203-P20, side of the tetrahedron. An illustration of the projection of a quaternary point in the present system is shown in Fig. 1. The advantage of this type of projection is that all four components for an equilibrium curve can be read directly from the diagram. The present results are shown graphically in Fig. 2. The curves depicting the miscibility gap are dashed in parts where no experimental points were obtained. The composition range covered by the miscibility gap

Jan 1, 1968

By N. J. Olson, G. W. Toop

A regular solution is characterized by a finite molar heat of mixing and entropy due only to ideal or random mixing. The thermodynamic properties of binary regular solutions which exhibit constant atomic coordination number and constant atom-atom intefac-

Jan 1, 1967

By J. E. Hilliard

AS is well-known, the length per unit volume, Lv, of any lineal feature (such as grain edges in a polycrys-talline specimen) can be estimated from a count of the number of point intersections with a random plane of polish. Specifically:' Lv = 2NA in which NA is the expected number of intersections per unit area of the plane. As with all other measurements in quantitative metallography, the estimation of Lv is subject to a sampling error, the magnitude of which will depend on the number of observations. The analysis is usually performed by counting the number of intersections falling within a closed test figure of known area applied to a succession of different areas on the plane of polish. The test figure can be drawn directly on the screen of a projection microscope or else superimposed on the image of the structure by means of a reticule in a focusing eyepiece of an ordinary microscope. The use of micrographs is very inefficient unless, as in the present study, these are also needed for some other purpose. Consider an analysis in which n applications are made of a test figure within which the expected number of intersections is N. The expected total number of intersections counted is then given by NT = nN If (N) is the standard deviation in N for one application of the figure and if the latter is applied to non-overlapping regions so that the N's are independent in the statistical sense, then the standard deviation in Nt is given by: a(NT) = a(N)vn Eliminating n from Eqs. [2] and [3] we obtain: Thus, k will be a constant for a given specimen and test figure. From Eqs. [ I] and [4] we find As is to be expected, therefore, the fractional error in the estimation of Lv is inversely proportional to the square root of the total number of intersections counted. It is very desirable to have at least a rough estimate of k before starting an analysis, since it will then be possible to decide in advance by use of Eq. [5J how many counts should be made in order to achieve the desired accuracy in Lv. Without such information the investigator, unless lucky, will either do needless (and tedious) work or else obtain a result with too large an error. It is very easy to calculate k for the limiting case of N — 0, i.e., when the test figure is vanishingly small. In this case there is a negligible probability of two or more intersections falling within the figure. There are therefore only two possible outcomes: a count of one with a probability of, say, p and a zero count with a probability of (I-p). Thus and Therefore: We might expect k to depend on N and therefore on the size of the test figure. Such a dependence was observed2 in the analogous case involving the estimation of grain-boundary area per unit volume by a count of the number of intersections of a test circle with traces of the boundaries on the plane of polish. In this case the theoretical value of k at the limit N — 0 was 1.41. It was also predicted that k would display an initial sharp decrease with increasing N. Experimentally, k was observed to decrease to a value of 0.62 at N - 6 and thereafter remain approximately constant. Furthermore, the value of k for N > 6 was found to be insensitive to the form of the structure provided no pronounced anisotropy or grain-size variation was present. In order to determinewhether k would exhibit a similar dependence on N in the estimation of grain-

Jan 1, 1967

By S. A. LoPilato, J. T. Smith

SEVERAL techniques have been evaluated for cooling the throat area of rocket-nozzle inserts to prevent erosion or fracture of the inserts during exposure to high operating temperatures and pressures. Self-cooled composites, such as infiltrated tungsten, have shown considerable promise for these applications. These composites can be fabricated by infiltrating the refractory matrix with a lower-melting metal or non-metal which is insoluable in tungsten. During firing, the lower-melting substance undergoes a series of continuous endothermic phase changes which absorb sufficient heat to maintain the nozzle matrix material at or below a safe operating temperature. At present, the state-of-the-art has reached a point where the refractory matrix density and microstruc-ture have been established as the critical parameters to be controlled for optimum self-cooled performance.'" Nondestructive testing has been utilized to ensure that porous tungsten billets were of uniform density and free from cracks or other defects prior to fabrication into rocket nozzle inserts. Ultrasonic-velocity techniques have been employed because of their reliability and reproducibility. The purpose of this note is to present and interpret the test data developed for porous tungsten matrices. For evaluation purposes, tungsten powders with Fisher subsieve sizes of approximately 4 and 18 p have been obtained for study from four commercial sources: the General Electric Co., Wah Chang Corp., Fansteel Metallurgical Corp., and Kennametal, Inc. Porous tungsten matrices were fabricated from the powders by hot pressing in induction-heated graphite die assemblies. Typical billets ranged from 3 in. diam, 2 in, high to 5 in. diam, 5 in. high. The ultrasonic-velocity measurement technique employed was through transmission, pulse echo at a frequency of 1.0 mc using matched barium titanate crystal transducers. The testing equipment included a pulsed ultrasonic generator, an amplifier, a specimen holder, and an oscilloscope, as shown in Fig. 1. The porous tungsten thickness was measured to i 0.0005 in, with precision micrometers while the oscilloscope achieved 0.02 psec time measurement accuracy. The time for a sound wave to pass through the billet, the bulk wave velocity, was calculated to 0.5 pct using Eq. [l]: V-} where V = bulk wave velocity, in. per psec, T = specimen thickness, in., and / = time, psec. To develop inspection standards, the bulk density of a porous tungsten billet was determined by the ASTM water-immersion technique.3 The ultrasonic-velocity measurements were completed at seventeen positions for each billet and averaged to develop a characteristic standard velocity for a particular tungsten powder at several density levels. A plot of ultrasonic velocity vs density is presented in Fig. 2 for six different tungsten powders. This graph has been established as

Jan 1, 1967

By H. C. Fiedler

THE development of the cube-on-edge secondary re-crystallization texture in Si-Fe strip depends upon the ability of inclusions to restrain normal grain growth. This ability is determined by the number and size of inclusions, which, for a given composition, is controlled by the thermal history of the material during the prior processing.' It has been shown2 that increasing the silicon content decreases the solubility of sulfur, and it is the purpose of this note to show the consequences of this decreased solubility for secondary recrystallization. Heats were made as previously described and cast into graphite molds.' The compositions are shown in Table I. Slabs 1 in. thick were cut from the ingots and either hot-rolled from 1000°C in the as-cast condition or after one of the heat treatments described later. The processing to the final gage of 12 mils followed the previous procedure.' To avoid cracking, the 5.2 pct Si alloy was cold-rolled at 150" to 200°C rather than at room temperature. The percent cube-on-edge texture developed by isothermal heat treatments was determined by dividing the maximum torque measured on disk sam-

Jan 1, 1968

By F. E. Rizzo, L. R. Bidwell, D. F. Frank

THE use of galvanic cells for the determination of oxygen activities has spread rapidly since the suitability of calcia-stabilized zirconia as a solid electrolyte material was first demonstrated by Kiukkola and Wagner.' In many of these investigations:-7 including several at ARL, the free-energy changes for various oxidation-reduction reactions involving the oxides FexO, NiO, and Cu2O have been measured as a means of assuring that a cell apparatus is functioning properly. As a result, a large quantity of data pertinent to the calculation of the high-temperature standard free energy of formation (?GO) of these oxides has been accumulated. AGFexo has been carefully measured in many independent investigations, using a variety of techniques, and consequently is accurately known. chipmans has used this informa-

Jan 1, 1968

By Alexander Coucoulas

FABRICATION of thin-film circuits deposited on glass substrates requires the formation of permanent electrical contacts. Since thin films are in the order of a few thousand angstroms thick and are deposited on nonconductive surfaces such as glass, new methods of lead attachment are being pursued in the microelectronics industry. Ultrasonic welding is being considered in thin-film technology since joining can be accomplished between a wide variety of dissimilar materials.l,' Ultrasonic welding is accomplished by a transducer-coupling-tip system which converts high-frequency electrical energy into mechanical vibrations and delivers it into the weld region. During the process the tip oscillates in a direction parallel to the surface of the weldments, inducing shear stresses in the work pieces which are clamped between the tip and anvil. Though the basic bonding mechanism is not clearly understood, it has been described as closely akin to friction welding. In the course of an investigation concerned with ultrasonic we!ding of aluminum wires to tantalum thin films (1400A) deposited on glass, low weld strengths were related to glass failure. Upon investigating this condition, an improved method of bonding was developed where the aluminum was held at a temperature where a minimum amount of strain hardening occurred during the introduction of ultrasonic energy. Experimental Procedure. Aluminum wires (0.015 in,) were bonded to tantalum thin films with an optimized set of welding parameters (power, time, clamping force) employing a commercial 100-w 60-kc ultrasonic welder. Forty-five degree shear-peel tests, where the lead is pulled at a 45-deg angle with respect to the substrate at 0.2 in. per min, produced a large variation from zero to the strength of the wire, Table I. Inspection of the weld region generally revealed: glass cracking before and after testing; the removal of glass nuggets from the substrate by the lead during testing, Table I; thin-film penetration. In an effort to improve the shear-peel strength, the aluminum between the tip and substrate was heated above ambient temperature through an argon gas media prior to and during welding with the same parameters employed above. With the addition of heat: wire deformation was increased; shear-peel tests all failed at the weld-wire junction, leaving the bonded portion undisturbed; no evidence of glass cracking or film penetration prior to or after testing was observed. In order to prevent failure at the weld-wire junction due to excessive wire deformation, 0.035-in. aluminum wires were welded to tantalum thin films with an optimized set of welding and heating parameters. As

Jan 1, 1967