Acta mater. 48 (2000) 4035–4043 www.elsevier.com/locate/actamat THE EFFECT OF PHASE SEPARATION ON SUBSEQUENT CRYSTALLIZATION IN Al 88 Gd 6 La 2 Ni 4 A. K. GANGOPADHYAY*, T. K. CROAT and K. F. KELTON Department of Physics, Washington University, One Brookings Drive, Campus Box 1105, St. Louis, MO 63130-4899, USA ( Received 3 January 2000; received in revised form 19 May 2000; accepted 26 June 2000 ) Abstract—To understand the crystallization (devitrification) kinetics of Al-rich Al–RE–TM (RE = rare earth, TM = transition metal) amorphous alloys, the devitrification of Al 88 Gd 6 La 2 Ni 4 was studied by in situ electrical resistivity and transmission electron microscopy (TEM) investigations. This glass compo- sition was chosen because it shows a well-defined glass transition temperature (T g ) and transforms to α-Al on partial devitrification. Surprisingly, we show that crystallization appears to be preceded by a phase separ- ation into Al-rich and solute-rich amorphous regions, having a typical dimension of 40 nm. TEM studies reveal a preferential rapid nucleation of α-Al at the interface of the phase-separated regions. A Johnson– Mehl–Avrami analysis of crystallization-induced changes in the electrical resistivity shows limited agreement with the theory, with an Avrami exponent n close to unity. This is explained by a simple numerical model that is consistent with the microstructure; i.e., rapid nucleation at the phase boundary followed by diffusion- limited growth.  2000 Acta Metallurgica Inc. Published by Elsevier Science Ltd. All rights reserved. Keywords: Rapid solidification; Phase transformations; Resistivity; Transmission electron microscopy; Het- erogeneous nucleation 1. INTRODUCTION The recent discovery of Al-based metallic glasses by Inoue et al. [1, 2] and others [3], and Zr-based bulk metallic glasses [4], has led to renewed fundamental interest in amorphous metallic alloys. These amorph- ous alloys crystallize (devitrify) to a nanostructured composite material that contains nanometer-sized crystallites embedded in an amorphous matrix, often further enhancing key physical properties above those of the metallic glasses [5–7]. The mechanisms under- lying this process, however, are poorly understood and not easily controlled. Clearly, the high density (>10 23 /m 3 ) of extremely small grains indicates an unusually high nucleation rate and a slow growth velocity. In the Zr–Ti–Cu–Ni–Be bulk metallic alloys, this may be explained by phase separation in the undercooled liquid or glass prior to crystallization [8, 9]. Theoretical [10, 11] and experimental investi- gations [12–16], however, suggest that phase separ- ation is unlikely in the Al-rich (>80 at.%) Al–RE– TM (RE = rare earth, TM = transition metal) metallic glasses. Various competing models, based on hetero- geneous nucleation [16], the presence of a large den- sity of quenched-in nuclei [14, 15] or sub-nanometer- sized crystallites [12, 13] in the amorphous phase and * To whom all correspondence should be addressed. 1359-6454/00/$20.00  2000 Acta Metallurgica Inc. Published by Elsevier Science Ltd. All rights reserved. PII:S1359-6454(00)00196-8 a new kind of homogeneous nucleation mechanism (“linked flux”) [17, 18], have been proposed. How- ever, the correct mechanism has yet to be identified. Calin and Ko ¨ster [16] studied the microstructure development of Al 85 Ni 5 Y 10 and Al 90 Ni 6 Nd 4 alloys as a function of annealing time and temperature. They observed an initial rapid rise in the nuclei density fol- lowed by near saturation at longer times, suggesting time-dependent heterogeneous nucleation followed by site saturation. Assuming a reasonable size for the sites to be effective, however, the large density of such heterogeneous sites (  10 23 /m 3 ) would corre- spond to an impurity level of 1% or more. This is much higher than the contamination level typically found (  10 13 /m 3 ) in conventional metallic glasses [19]. In contrast, two other groups [12–15] suggested that the transformation is basically a manifestation of the growth on pre-existing, quenched-in nuclei/sub- nanocrystallites that form during melt quenching. It is, however, unclear from one group’s work [12, 13] whether the suggested “sub-nanocrystallites” are fun- damentally different from quenched-in nuclei. Assuming a typical cluster size of 100 atoms and 10 23 /m 3 quenched-in nuclei, both reasonable numbers, at least 10 25 atoms/m 3 would be in crystal grains. Assuming a molar weight of 35 g and a density of 3.5 g/cm 3 , the estimated number of atoms is  6×10 28 /m 3 for these Al–RE–TM alloys. This