&cN H.PM mar lor 0'- ZONE REFINING i* : ' 4 '96 7 -l"WIINng(on -WA 88185 A(r (*&LA 217: 13 ' This simple technique, in which a solid is refined by passing a liquid zone through it, has been of profound value in those technologies that call for materials of extremely high purity by William G. Pfann t a third of the elements and In zone refining, which is probably the tories, told me I could spend half my undreds of organic and inorganic most important zone melting technique, time on fundamental research of my own compounds have been poduced a number of molten zones are passed choosing. I chose to study "slip bands" their purest form by a simple tech- through the charge in one direction. producid by deforming single ~;.~stalsof que called zone refining. If two of Each moving zone carries a fraction of lead containing a fraction of a percent of e elements, germanium and silicon, the impurities to the end or, in some antimony. I recognized that if I used a not been available in the desired cases, to the beginning of the charge, standard freezing technique to grow the y, the whole development of solid- thereby the- remainder aTld crystals, the antimony would tend to seg- electronics would have been crip- concentrating impurities at one or both regate, thus producing a crystal wit11 n . Other elements, when highly puri- ends. In the 15 years since my original nonuniform distribution of the solute. r- paper on the subject I have been asked To remedy this I thought to grow the over and over: "How could such a sim- crystal by passing a short melted zone of research. All of this has hap- ple idea have been missed all these dec- along the ingot in the hope of leveling since the basic paper on zone ades? How did you think of it?" out the distribution of antimonv. This I really cannot answer eitl~erquestion. procedure, now called zone leveling, one of a Several years after my papel. had ap- did not strike me as particularly remark- nown as peared I learned that Peter Kapitza, able and nothing came of it at the time. s can be the Russian physicist, had actually per- I assumed that such a sim~leidea must itions or fo~meda one-pass zone melting opera- be common knowledge. Also, about that le melt- tion and had described it in a papel. pub- time Bell Laboratories became heavily e trav- lished in 1928. He was concerned \\lit11 engaged in technical problems of World -ge (or crystal growth, however, and apparently \Val- 11. It was not until years later, ortion did not recognize the pote~ltial of a when the need arose for ge~manium molten zone as a distributor of solutes, of uniform purity for transistors, that I es, or and it would seem that neither did ally again thought of zone leveling. Sudden- ponents, depends on the size, num- of the hundreds of physicists who must ly the lightning of zone refining struck: a and direction of travel of the zones, have read his paper. molten zone moving through an ingot n the My own conception of the process could do more than level out im~urities- prop- dates back to 1939, when the late Earle it could remove them. istri- E. Schumacher, head of the metallurgy With the help of colleagues at Bell laboratory of the Bell Telephone Labora- Laboratories zone- refinink was quickly developed into a successf;~ manufactur- ing procedure. The purity achieved was spectacular. The harmful impurities- those affecting transistor properties- were reduced to less than one itom of ' impurity in 10 billion atoms of germax um. This ratio is eq-to a grain oT3ZFKa freight-car load ot sugar. Sfnce then zone refinineu has been a~-' plied successfully to many other semi- otes diffusion in the liquid region. The cylinder travels to the left at the rate of about entimeters an hour, thus sweeping the molten zones to the right. In addition the cyl- conductors and metals, as well as to has a ratchet motion resembling that of a typewriter carriage: after traveling slowly organic and inorganic compounds. left one heater spacing it jumps quickly to the right the same distance. Thus the In order to understand zone refining, and zone melting in general, one must understand what is meant by the distri- Jrn . ' buson coefficient. A solution of salt in solution tends to rise steadily until all tribution coefficient, designated k ; water provides a familiar example. When the liquid is frozen. When the solute is an effective value, designated k. b 1 the temperature of such a solution is not table salt but a more typical sub- whole success of zone refining d liti lowered to the freezing point, the first stance, the crystals become progressive- on the fact that under ordinary fre 511 t ipcrystals to form do not contain the ly richer in the solute as they freeze conditions equilibrium is far from Jgc same concentration of salt as the original from a more and more concentrated solu- achieved. If freezing of, say, a solution 1g; solution does. In fact, the ice crystals tion. Concomitantly the freezing point or an alloy took place so slowly that the .er: contain a significantly lower concentra- of the solution drops steadily as the sol- entire solid at all times had the composi. rhc tion because the salt redistributes itself ute concentration of the remaining solu- tion of the solidus curve in the phase di.' PO' between the solid phase and the liquid tion rises. agram, the last clystals to form A phase while the crystal is being formed. All of this can be summed up in a have the same composition as the 5n: - The distribution coefficient is defined as phase diagram that shows how the com- solid. This implies, in turn, that ng the concentration of the solute in the position of the liquid phase (bounded by frozen solid would have preci -111 '- solid divided by its concentration in the the "liquidus" curve) is related under same solute concentration as t Jo. liquid; thus the coefficient is less than equilibrium conditions to the composi- nal solution. Paradoxically eq oli one. Indeed, for ordinary table salt in tion of the solid (bounded by the "soli- freezing consists in two simultaneous io~ \ . freezing water, the distribution coeffi- dus" curve) for all concentrations of the and opposed processes: the re ril cient is close to zero, meaning that salt solute in the solvent, together with the the solute by the advancing s es: is almost totally excluded from ice crys- freezing point for each concentration interface and the diffusion of the solute 01 :11; - . tals as they form. As a consequence the [see illustration below]. back into the solid that has frozen. .. .. concentration of solute in the remaining There is an equilibrium, or ideal, dis- In practice equilibrium fre n I most impossible to achieve if ai ing with substantial volumes of material. 81 Diffusion rates in solids are almost al- , )e ways too low to permit back n; the solute. For practical pur fore, the solute-rejection p n only one that counts. In 5 normally contain a much h tration of the solute than original solution. In normal freezing, ba effectively prevented if t interface advances at a to 20 centimeters per is also usually slow e concentration of solu equalized by diffusi Under these conditi tribution along an in piedicted by a norm some constant valu solute concentratio constant k is reasonable. mixing is rapid eno tribution coe5cien a binary system of two conipoaents, A and B, clarifies the relations is based. The diagram shows &at 'adding solute B to solvent A low- olution. Conversely,-adding A to B raises the freezing ning works because the composition of the solid that differs from the composition of the solution. Thus if a to its freezing point, 2, the solid that forms, 3, contains ina1liquid:The exclusioqof some B from the solid raises the con- aining liquid, 4, which.now ha&,a~owerfreezing point, 5. Again excludes B and raises the concentration of B in the liquid. 7, and so n isrthe equilihrium distriht&Gc>c&tficien< k,,, defined as the ratio .... -,,. tion to "liquidus" ton&trati.on \at a given .temperature. Here that effective distribution coefficient, k, is usually higher than k,,. form concentration of solute B in solvent NORMAL FREEZING e illustrations at right]. Let the concentration of B in A be repre- by Co. It is not necessary that the n ordinary rapidly frozen cast- be a mixture of sintered pow- 1 m .,-a- ZONE MELTING en two long rods side by side . .-*..;......................... ....................... .".'.'.::.. *............... se cross section at any point corre- ..... joL+:;. :::.::.:1:.::.1: ..:..: ....:.:.uNMELT.ED.~~L~ 6.:. :.......... to concentration C,. ............: lten zone, say a tenth of the length of the ingot, is formed by a heat- RING HEATER- A ing coil placed at one end of the charge. .-'The heating coil is then moved slowly ORDINARY FREEZING AND ZONE MELTING both alter the distribution of a solute in charge. As it advances, tlie first a solvent. Oddly enough, one-pass zone melting (bottom) is actually less effective as a re- eeze behind it has a concentra- fining method than normal freezing (top).Thia can be inferred from the illustration below. .: tion of solute B equal to C,, times the dis- tribution coefficient k, or kc,,.Since k is less than one, the newly formed solid contains less of solute B tha~ithe original- es. This means that some of 8 ncentratioil C,, is ted into tlie zone at its leading molten zone and also in the solid that at the trailing interface.