Available online at www.sciencedirect.com

ScienceDirectScripta Materialia 95 (2015) 35–38

www.elsevier.com/locate/scriptamat

On the direct nucleation and growth of ferrite and cementite withoutaustenite

A. Fillon,a X. Sauvage,a,⇑ B. Lawrence,b C. Sinclair,b M. Perez,c A. Weck,d E. Cantergiani,d

T. Epicierc and C.P. Scotte

aUniversity of Rouen, CNRS UMR 6634, GPM, FrancebDepartment of Materials Engineering, The University of British Columbia, Vancouver, BC V6T 1Z4, Canada

cUniversity of Lyon, INSA Lyon, MATEIS, UMR CNRS 5510, FrancedMechanical Engineering Department, University of Ottawa, Ottawa, ON K1N 6N5, Canada

eCanmetMATERIALS, Hamilton, ON L8P 0A5, Canada

Received 4 September 2014; revised 18 September 2014; accepted 19 September 2014Available online 11 October 2014

The direct nucleation and growth of ferrite and cementite has been investigated in an Fe–C metallic glass using X-ray diffraction, transmissionelectron microscopy and atom probe tomography. The nucleation and growth proceeds via primary crystallization of ferrite, without the formationof austenite. There is no carbon gradient at ferrite/amorphous interfaces, indicating that the transformation rate is controlled by the low mobility ofthese interfaces compared to the mobility of C atoms. A thermodynamic description is proposed to explain the delayed nucleation of cementite.� 2014 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.

Keywords: Ferrite; Cementite; Metallic glass; Crystallization

Optimization of the microstructures of steels, andhence of their properties, often depends strongly on thecontrol of phase transformations during thermomechanicalprocessing [1]. The decomposition of austenite is of partic-ular interest as it is the parent phase for most of the micro-structures of importance, e.g. martensite, bainite orpearlite. When the transformation rate is relatively low,the austenite (face-centered cubic) transforms at low tem-perature into ferrite (body-centered cubic, bcc) and cement-ite (orthorhombic) [2]. This transformation proceeds viadiscontinuous precipitation where austenite grain bound-aries act as nucleation sites, both for pro-eutectoid ferriteor pro-eutectoid cementite, and also for pearlite when thesetwo phases grow simultaneously [3,4]. Thus, these resultingmicrostructures are strongly controlled by the mechanismsof nucleation. In addition, it is relatively well establishedthat the rate-controlling mechanism is determined by thecarbon diffusion in austenite or along boundaries [3,4].Then, an interesting question is how ferrite and cementitewould nucleate and grow in the absence of prior defectsin austenite (namely grain boundaries), or from a parentphase other than austenite. Experimental investigationshave shown that it is impossible to directly nucleate thesephases from the melt. Indeed, a very fast cooling rate leadsto the nucleation of the metastable e phase [5,6] or to the

http://dx.doi.org/10.1016/j.scriptamat.2014.09.0251359-6462/� 2014 Acta Materialia Inc. Published by Elsevier Ltd. All rights

⇑Corresponding author; e-mail: xavier.sauvage@univ-rouen.fr

formation of a metallic glass [5–8]. The crystallization ofsuch a metallic glass upon annealing strongly depends onthe carbon content. If the carbon content is below20 at.%, then it proceeds at low temperature via primarycrystallization in two steps. First, ferrite nucleates, followedby the formation of carbides [7,8], as in the ternary Fe–C–Bsystem [5]. If the carbon concentration exceeds 25 at.%,then the amorphous phase transforms completely into car-bides [9], as in Mn–C and Cr–C systems [10]. In the case ofprimary crystallization, the kinetics of nucleation andgrowth into the amorphous structure is often controlledby the partitioning of elements [11,12]. This involves atomicdiffusion in the amorphous matrix and eventually the for-mation of concentration gradients at crystal/amorphousinterfaces during the primary crystallization process [13].Thus, the transformation rate could be influenced by largedifferences in the diffusivity of interstitial species (e.g. car-bon in iron) in an amorphous and crystalline phase [14].The aim of this work was first to understand how ferriteand cementite nucleate from a parent phase other than aus-tenite, namely a metallic glass. Second, by investigatingconcentration gradients through the amorphous/crystallineinterfaces we studied the influence of carbon diffusion in thepossible cooperative growth of these two phases.

To achieve these goals, Fe–C amorphous films �500 nmthick were obtained by reactive magnetron sputtering usingCH4 as reactive gas and a Fe target (purity 99.5%) (seedetails in Ref. [15]). Films containing about 15 at.% C were

reserved.

36 A. Fillon et al. / Scripta Materialia 95 (2015) 35–38

deposited on Si substrates for X-ray diffraction (XRD)analysis, on standard Al foil for transmission electronmicroscopy (TEM) characterization (the substrate was dis-solved in NaOH solution) and finally on interstitial-free(IF) steel to investigate the influence of a pre-existing inter-face (amorphous/ferrite) on the nucleation and growthmechanisms. Films were annealed under vacuum(10�6 mbar) up to 500 �C. Crystalline phases were identifiedby XRD using a Rigaku diffractometer (Cu-target X-raysource and scintillation detector). Cross-section TEM foilswere prepared using a focused ion beam (FIB; NVISION-40 Zeiss). Samples were observed in a JEOL ARM200Fmicroscope operated at 200 kV. Energy-filtered (EF)TEM images were recorded using a Gatan Imaging Filter(GIF Quantum). In situ annealing experiments were carriedout to follow growth kinetics using a Gatan Tantalum heat-ing holder in an aberration-corrected Titan environmentaltransmission electron microscope operated at 300 kV.Composition gradients across crystalline/amorphous inter-faces were measured by atom probe tomography (APT).Samples were prepared using FIB and then analyzed inan energy-compensated atom probe (EcoTAP CAMECA)using electric pulses (repetition rate 30 kHz and 20% pulsefraction) at 80 K. Carbon quantification was done follow-ing the procedure proposed by Sha et al. [16].

Both XRD measurements (Fig. 1b) and selected-areaelectron diffraction (SAED) performed in the transmissionelectron microscope (not shown here) confirmed that theas-prepared material was fully amorphous. APT analyseswere carried out to check both the composition of the

Fig. 1. (a) 3-D reconstruction of a volume analyzed in the as-prepared Fe–Cafter isothermal annealing (3600 s) at various temperatures; (c) in situ TEM bshowing the successive crystallization of ferrite and cementite (circles and sq

metallic glass (Table 1) and the distribution of carbonatoms within the structure. As shown on the 3-D recon-struction (Fig. 1a), no significant composition variationswere observed and the computed frequency distributionof carbon atoms corresponded relatively well to a randomdistribution. During isothermal annealing for 1 h at lowtemperature (250 �C) the ferrite phase nucleated first(Fig. 1b), indicating a primary crystallization process con-sistent with data reported on the crystallization of similarmetallic glasses prepared by splat cooling [7,8]. At highertemperatures (300 �C and above), peaks corresponding toboth ferrite and cementite are detected. In situ TEM alsoconfirmed that the ferrite nucleates before cementite (seeFig. 1c), which is qualitatively very consistent with theXRD data shown in Figure 1b. At completion of the trans-formation the material exhibits a nanoscaled structure(Fig. 2a). Equiaxed and slightly faceted ferrite grains thathave nucleated by primary crystallization are homoge-neously distributed and exhibit an average diameter of50–100 nm. They are embedded in much larger grains(200–400 nm) identified as cementite (Fe3C). EFTEM datashowing the carbon distribution (Fig. 2c) clearly indicates,as expected, a strong partitioning of carbon between thetwo phases. In the carbides, the large density of stackingfaults, clearly visible on the TEM bright-field image(Fig. 2a), gives rise to characteristic streaks on the diffrac-tion pattern (Fig. 2b). Such faulted carbides have been pre-viously reported in high-carbon crystallized Fe–C metallicglasses (e.g. [10]) as well as in tempered martensite [17]. Itwas concluded that this transformation occurred without

film showing the distribution of C atoms; (b) XRD spectra recordedright-field images taken after holding for 3600 s at 250, 350 and 450 �C,uares are used as a guide to follow the growth).

Fig. 3. Fe–C metallic glass deposited on the IF steel and annealed for1 h at 250 �C. (a) Bright-field TEM image (corresponding SAED inset)showing nanoscaled ferrite grains that have homogeneously nucleatedbut also a large density of ferrite grains nucleated at the interface withthe IF substrate. (b) 3-D reconstruction of a volume analyzed by APTand showing the distribution of C (red) and Fe (black) atoms. Thecomposition profile was computed through a ferrite/amorphousinterface (sampling volume thickness 1 nm). (For interpretation ofthe references to colour in this figure legend, the reader is referred tothe web version of this article.)

Fig. 2. (a) TEM bright-field image of the fully crystallized Fe–C alloyaged at 300 �C for 3600 s, showing nanoscaled a-Fe grains embeddedin Fe3C. (b) SAED pattern taken in a Fe3C grain and involving severala-Fe grains. (c) EFTEM map combining the C-K (green) and Fe-K(blue) edge signals. (For interpretation of the references to colour inthis figure legend, the reader is referred to the web version of thisarticle.)

Table 1. Phase composition (at.%) obtained from APT analyses.

As-prepared 1 h 250 �Camorphous phase

1 h 250 �Cferrite

Fe 85.7 ± 0.1 75.2 ± 0.13 99.78 ± 0.03C 13.6 ± 0.1 24.7 ± 0.13 0.04 ± 0.01Others <0.03 <0.03 <0.03

A. Fillon et al. / Scripta Materialia 95 (2015) 35–38 37

the formation of austenite or the epsilon carbide phase.This latter phase has been reported for similar metallicglasses obtained by splat cooling [6], though it seems thatits stability is strongly linked to the silicon content of thealloy [18,19]. It is interesting to note that carbon partition-ing occurred exclusively during the primary crystallizationstage via carbon diffusion in the amorphous phase. Toinvestigate the influence of a pre-existing interface on thenucleation process, an amorphous film with the same car-bon content was deposited on IF steel (single-phase bcc fer-rite). As shown on the bright-field TEM image in Figure 3a,such an interface acts as a preferential site for the nucle-ation of ferrite. Numerous ferrite grains are seen alongthe substrate interface, these exhibiting a higher nucleationdensity compared to the bulk. Due to the high density ofnuclei at the interface, the ferrite growth seems to be con-strained in all directions except perpendicular to the origi-nal boundary. In analogy with the decomposition ofaustenite, one might expect the formation of cementite inthe regions between these ferrite grains due to the partition-ing of carbon into the amorphous matrix leading to a pearl-itic-like structure. However, in the present case, even if alarge amount of carbon is rejected into the parent phase,cementite does not nucleate and there is no cooperativegrowth of ferrite and cementite. The carbon distributionin this partly crystallized state was investigated by APTanalysis (Fig. 3b). The carbon concentration measured inthe ferrite (Table 1) is in relatively good agreement withthe solubility of carbon in ferrite predicted from the a-Fe/Fe3C phase diagram [2]. This is not an obvious resultas we could have expected deviations since here the ferriteis in contact with the amorphous phase rather than withFe3C. It is also interesting to note that, even if the compo-sition of this amorphous phase is extremely close to thecomposition of cementite (Fe3C, 25 at.%), the latter phasehas not nucleated at this stage. From these compositionmeasurements, and assuming that the molar volume of all

phases is similar, the volume fraction of ferrite was esti-mated to be �40%. Importantly, the concentration profilesmeasured across the ferrite/amorphous interfaces (Fig. 3c)clearly show that there is no significant carbon gradientin the amorphous phase (the 3–5 nm gradient can be attrib-uted to the convolution of the sampling volume with thecurved interface). There is no measurable carbon enrich-ment near the boundary, indicating that the mobility of car-bon atoms is much higher than the mobility of the growinginterface. From in situ TEM observations it proved possi-ble to estimate the mean velocity of the ferrite/amorphousinterface during the growth regime. It turns out that thefastest growth rate of ferrite crystals at 250 �C was typicallyfrom 10 to 60 nm in diameter over a period of 5 min. Thisleads to a mean interface velocity of about 5 nm min�1.Previous studies on the diffusion of interstitial elements inamorphous Fe suggest a slightly higher diffusivity than inthe crystalline phase [14], but also that this diffusivity mightbe affected by the structural relaxation of the glass duringannealing. Thus, one can conservatively make a lowerbound estimate of the diffusivity of carbon in the amor-phous structure by using the diffusivity in bcc ferrite, wherethe diffusion coefficient is taken to be: D = D0 exp [�Q/(RT)] (with D0 = 6.2 � 10�7 m2 s�1, Q = 80 kJ mol�1,R = 8.31 J K�1 mol�1 [1]). Using this estimate the effectivediffusion distance (k � (6 D t)1/2 with t the diffusion time) ofcarbon atoms is �103 nm in 1 min at 250 �C. This suggeststhat the transformation rate is too slow to cause the build-

38 A. Fillon et al. / Scripta Materialia 95 (2015) 35–38

up of a carbon gradient at the interface. Therefore, bothour experimental observations (APT data, Fig. 3) and theseestimates indicate that, unlike the pearlitic transformation[3,4], in the present case the rate of growth of the ferriteinterface is not controlled by carbon diffusion but ratherby the interface mobility. Furthermore, the rapid redistri-bution of C atoms inside the amorphous phase makes it dif-ficult to obtain locally high carbon concentrations duringthe crystallization. This is in contrast to the decompositionof austenite, during which the relatively high mobility ofthe ferrite/austenite interface leads to a buildup of carbonat the transformation front and therefore a preferentiallocation for cementite nucleation. Clearly, compared to fer-rite, the nucleation of cementite seems more difficult in thepresent material. This could be due to a higher nucleationbarrier or a lower nucleation driving force. The nucleationof cementite has also been reported to be difficult in mar-tensite tempered at low temperatures (below 300 �C), eventhough the mobility of carbon is expected to be high underthese conditions [17,18]. For long aging times in martensite(>103 s) cementite should appear at 250 �C but it is typi-cally preceded by transition carbides or is observed to formon defects where the nucleation barrier is reduced (e.g. twinboundaries) [18]. As in martensite, the cementite nucleationin the amorphous phase might be limited by the need forsignificant rearrangement of Fe atoms. At low tempera-tures, the mobility of Fe atoms is rather limited. It has beenproposed that at such low temperatures carbides couldform in tempered martensite via a displacive transforma-tion [20,21]. It is hard to imagine a similar process beingavailable in a disordered amorphous structure.

To better understand the two-step crystallization pro-cess of the Fe–C metallic glass, one should consider thedriving force for nucleation. A representation of the molarGibbs energy of the liquid phase (which is, thermodynam-ically speaking, not far from the amorphous state), ferriteand cementite is shown as a function of the carbon contentin Figure 4 (calculations made with the TCFE7 database).Note that the curve for the amorphous/liquid phase indi-cates a minimum of enthalpy close to the eutectic composi-tion. In the Fe–B system and other alloys based on theseelements, a good glass-forming ability is typically achievedfor about 20 at.% B as well [22]. Thus, Figure 4 clearlyshows that at the beginning of crystallization, the driving

Fig. 4. Representation of the molar Gibbs energy curves of thedifferent phases involved in the crystallization process (estimate doneat 298 K with Thermocalc software using the TCFE7 database).Primary crystallization (DG1), followed by secondary polymorphousreaction (DG2). DGn

i represent the driving force for nucleation of the iphase (i = a-Fe or Fe3C).

force for the nucleation of ferrite (DGna-Fe) is much higher

than that of cementite (DGnFe3C), triggering the primary

crystallization (DG1). In consequence, the carbon contentin the amorphous phase increases. This leads to a smalldecrease of the enthalpy but a progressive increase of thedriving force for the nucleation of cementite. Then, in a sec-ond step, if the temperature is high enough to overcome thenucleation barrier, the amorphous phase with a carboncontent close to the cementite composition crystallizes(DG2), forming large cementite grains surrounding thenanoscale ferrite clusters (Fig. 2a).

In conclusion, the crystallization of Fe–C metallic glassinvolves a primary crystallization process of ferrite. Con-trary to the decomposition of austenite, the transformationrate is controlled by the interface mobility, and due to thehigh mobility of carbon in the amorphous phase, there isno carbon buildup at the interface, favouring the nucle-ation of cementite. One explanation, illustrated qualita-tively here, is that the thermodynamic driving force forferrite relative to cementite can explain the relative easeof nucleation of the former relative to the latter.

Prof. Philippe Maugis is gratefully acknowledged for dis-cussions on the thermodynamic description of the system. Thiswork was performed with the funding support of ANR andNSERC through the GRACOS project (ANR-09-BLAN-0412).The CLYM (Centre Lyonnais de Microscopie, www.clym.fr) isgratefully acknowledged for access to the environmental transmis-sion electron microscope.

[1] R.W.K. Honeycombe, H.K.D.H. Bhadeshia, Steels Micro-structure and Properties, Edward Arnold, London, 1995.

[2] J. Chipman, Metall. Trans. 3 (1972) 55.[3] S.E. Offerman, L.J.G.W. Van Wilderen, N.H. Van Dijk, J.

Sietsma, M.Th. Rekveldt, S. Van Der Zwaag, Acta Mater. 51(2003) 3927.

[4] A.S. Pandit, H.K.D.H. Bhadeshia, Proc. R. Soc. A 467 (2011)508.

[5] E. Hornbogen, I. Schmidt, in: B. Cantor (Ed.), RapidlyQuenched Metals III, vol. 1, Metals Society, 1978, pp. 261–264.

[6] I. Schmidt, Z. Metallkd. 74 (1983) 561.[7] P.G. Boswell, G.A. Chadwick, J. Mater. Sci. 11 (1976) 2287.[8] P.H. Shingu, K. Shimomura, K. Kobayashi, R. Ozaki, Mater.

Sci. Eng. 23 (1976) 183.[9] E. Bauer-Grosse, Thin Solid Films 447–448 (2004) 311.

[10] E. Bauer-Grosse, G. Le Caer, J. Phys. F: Met. Phys. 16 (1986) 399.[11] K. Kajiwaraa, M. Ohnuma, D.H. Ping, O. Haruyama, K.

Hono, Intermetallics 10 (2002) 1053.[12] C.F. Conde, M. Millan, A. Conde, J. Non-Cryst. Solids 232–

234 (1998) 346.[13] Maria Teresa Clavaguera-Mora, Narcis Clavaguera, Daniel

Crespo, Trinitat Pradell, Prog. Mater. Sci. 47 (2002) 559.[14] W. Chambron, F. Lancon, A. Chamberod, J. Non-Cryst.

Solids 61–62 (1984) 895.[15] C.P. Scott, C. Sinclair, A. Weck, Scr. Mater. 65 (2011) 763.[16] W. Sha, L. Chang, G. Smith, L. Cheng, E. Mittemeijer, Surf.

Sci. 266 (1992) 416.[17] G. Miyamoto, J.C. Oh, K. Hono, T. Furuhara, T. Maki,

Acta Mater. 55 (2007) 5027.[18] Jae Hoon Jang, In Gee Kim, H.K.D.H. Bhadeshia, Scr.

Mater. 63 (2010) 121.[19] Sigemaro Nagakura, Toshiaki Suzuki, Michiko Kusunoki,

Trans. Jpn. Inst. Met. 22 (1981) 699.[20] K.W. Andrews, Acta Metall. 11 (1963) 939.[21] M. Audier, J.P. Simon, P. Guyot, Acta Metall. 34 (1986) 1983.[22] A.L. Greer, Acta Metall. 30 (1982) 171.