Rate-Controlling Process of Compound Growth in Cu-Clad Al Wire during Isothermal Annealing at 483–543 K

Abstract

The rate-controlling process of compound growth in the Cu-clad Al (CA) wire was metallographically examined in the temperature range of 483–543 K (210–270°C). CA wires were prepared by a wire drawing technique, and then isothermally annealed for various times up to 3.456 Ms (960 h) in this temperature range. During isothermal annealing, the α₂, γ₁, δ, η₂ and θ phases form as layers at the original Cu/Al interface in the CA wire. However, at 483–513 K (210–240°C), the α₂, γ₁, δ and η₂ phases could not be differentiated from one another in a metallographical manner. The layer of the θ phase is designated layer 1, and that of the α₂, γ₁, δ and η₂ phases is called layer 2. The mean thickness of each layer increases in proportion to a power function of the annealing time. Such a relationship is called a power relationship. The exponent of the power relationship takes values between 0.22 and 0.36 for layer 1 and those between 0.39 and 0.50 for layer 2. The exponent smaller than 0.5 indicates that boundary diffusion as well as volume diffusion contributes to the layer growth. However, the contribution of boundary diffusion is more remarkable for layer 1 than for layer 2. The activation enthalpy for the proportionality coefficient of the power relationship was estimated to be 43 and 73 kJ/mol for layers 1 and 2, respectively. The former value is smaller than the latter value. This is attributed to the contribution of boundary diffusion being greater for layer 1 than for layer 2.

Fig. 3 The total thickness l of the intermetallic layer versus the annealing time t shown as open rhombuses, squares and circles for T = 483, 513 and 543 K (210, 240 and 270°C), respectively. The values of k and n in eq. (2) were estimated by the least-squares method.

1. Introduction

Fuel efficiency is one of the most important issues in the automobile industry for each country. Since a lot of electric wires are used in the automobile, reduction of total weight of the electric wires improves the fuel efficiency. As a result, replacement of Cu wires by Al wires gradually proceeds.^1–3) The total-weight reduction of the electric wires can be realized also by combination of Cu and Al. A typical example of the combination is Cu-clad Al (CA) wire, where the CA wire is an Al wire coated with Cu. The CA wire is widely used for electronic parts as a conductive material in the automobile industry.

The CA wire is ordinarily produced by wire drawing process. Owing to this process, the hardy Cu/Al interface is actualized by solid-state bonding without formation of intermetallic compounds (IMC’s). On the other hand, the CA wire utilized near the motor in the automobile is heated at temperatures of about 373–473 K (100–200°C). During heating at such temperatures, a layer of IMC’s may form at the Cu/Al interface and grows gradually. The IMC layer is brittle and electrically resistant. As a consequence, the growth of the IMC layer deteriorates mechanical and electrical properties of the CA wire. There, for assurance of the reliability of the product, information on the growth behavior of the IMC layer during heating is essentially important.

Gueydan et al.⁴⁾ experimentally observed the growth of the IMC layer. In their experiment, CA wires were annealed at temperatures of 573–673 K (300–400°C). Their observation shows that the IMC layer is composed of the γ₁ (Cu₉Al₄), η₂ (CuAl) and θ (CuAl₂) phases and grows according to a parabolic relationship. Here, the parabolic relationship means that the total thickness of the IMC layer is proportional to the square root of the annealing time. Such dependence of the layer thickness on the annealing time indicates that the IMC growth is controlled by volume diffusion.⁵⁾

In contrast, the growth behavior of the IMC layer was experimentally examined in the temperature range of 423–543 K (150–270°C) in a previous study.⁶⁾ In this temperature range, CA wires were isothermally annealed for various times of 43.2 ks to 3.456 Ms (12–960 h). The experimental result indicates that the IMC layer consists of the α₂ (Cu₃Al), γ₁ (Cu₉Al₄), δ (Cu₃Al₂), η₂ (CuAl) and θ (CuAl₂) phases. Furthermore, the total thickness of the IMC layer is proportional to a power function of the annealing time, and the exponent of the power function takes values of 0.23–0.44. Hence, unlike the observation by Gueydan et al.,⁴⁾ the parabolic relationship does not hold under the experimental conditions in a previous study.⁶⁾ Nevertheless, in that study,⁶⁾ attention was focused on the influence of pre-heating at 523 K (250°C) for 10.8 ks (3 h) on the growth behavior of the IMC layer during the subsequent isothermal annealing in the temperature range mentioned above. For this purpose, the parabolic relationship was conveniently assumed to evaluate the temperature dependence of the layer growth. The evaluation was compared with the corresponding one reported by Gueydan et al.⁴⁾

As mentioned earlier, the power relationship holds between the total thickness of the IMC layer and the annealing time, and the exponent of the power function is equal to 0.23–0.44 at 423–543 K (150–270°C). The exponent smaller than 0.5 shows that boundary diffusion as well as volume diffusion contributes to the rate-controlling process of the IMC growth. As a consequence, unlike the evaluation in a previous study,⁶⁾ the value of the exponent should be appropriately considered to estimate the rate-controlling process. In the present study, the kinetics of the IMC growth in the CA wire was experimentally examined by a metallographical technique at temperatures of 483–543 K (210–270°C). The rate-controlling process was discussed on the basis of the experimental result.

2. Experimental

Experiment was conducted according to the procedures reported in a previous study.⁶⁾ A wire drawing technique was used to prepare CA wires with Cu and Al volume fractions of roughly 0.15 and 0.85, respectively. Thus, the area fractions of Cu and Al on the cross-section of the CA wire are about 0.15 and 0.85, respectively. Here, the diameter of the initial CA rod was 10 mm, and that of the final CA wire was 1.5 mm. The Cu/Al interface in the CA wire was strongly bonded by the wire drawing technique. The CA wires were isothermally annealed at temperatures of 483–543 K (210–270°C). The annealing time was 0.6912–3.456 Ms (192–960 h) at 483 K (210°C) and 43.2–691.2 ks (12–192 h) at 513 and 543 K (240 and 270°C). Cross-sections of each annealed CA wire were mechanically polished using # 180–1000 emery papers and diamond paste with size of 1 µm. The mechanically polished cross-section was chemically etched with a solution containing ammonia and hydrogen peroxide as the main components. The microstructure of the chemically etched cross-section was observed mainly by scanning electron microscopy (SEM). Transmission electron microscopy (TEM) was partially used for the microstructure observation. A focus ion beam (FIB) technique was utilized to prepare TEM specimens. The chemical composition of each phase in the TEM specimen was determined by energy dispersive spectrometry (EDS).

3. Results and Discussion

3.1 Microstructure

A bright field (BF) image of a TEM specimen for the CA wire annealed at T = 513 K (240°C) for t = 10.8 ks (3 h) is shown in Fig. 1, where the Cu/Al interface is located edge on. Here, T is the annealing temperature measured in K, and t is the annealing time measured in s. In Fig. 1, the area with black contrast on the left-hand side is the Cu specimen, and that with white contrast on the right-hand side is the Al specimen. As can be seen, various layers with different gray contrasts are observed between the Cu and Al specimens. For identification of the gray layers, EDS measurements were carried out along the direction normal to the original Cu/Al interface.

Fig. 1

Bright field (BF) TEM image of the CA wire isothermally annealed at T = 513 K (240°C) for t = 10.8 ks (3 h).

A typical result of the EDS measurement for Fig. 1 is shown in Fig. 2. In this figure, the vertical axis indicates the mol fraction y_i of component i (i = Cu, Al), and the horizontal axis represents the distance x. Furthermore, open squares and circles show the values of y_Cu and y_Al, respectively. According to the result in Fig. 2, the brightest gray-layer on the Al-side is the θ (CuAl₂) phase. In contrast, the darkest gray-layer on the Cu-side is the Cu solid-solution phase produced by diffusion induced recrystallization (DIR).⁷⁾ DIR in the Cu(Al) system was observed also by Gueydan et al.,⁴⁾ Vandenberg et al.⁸⁾ and den Broeder et al.⁹⁾ The notation A(B) indicates that a solute B diffuses into either a pure metal A or a binary A–B alloy of the A-rich single-phase according to convention. The region produced by DIR is hereafter called the DIR region. Although a couple of gray layers with intermediate contrasts are recognized between the DIR region and the θ phase, they cannot be distinguished from one another in Fig. 2. In the binary Cu–Al system,¹⁰⁾ six stable compounds exit at T ≤ 636 K (363°C). They are the α₂ (Cu₃Al), γ₁ (Cu₉Al₄), δ (Cu₃Al₂), ζ₂ (Cu₅Al₄), η₂ (CuAl) and θ (CuAl₂) phases. At the highest annealing temperature of T = 543 K (270°C), the α₂, γ₁, δ, η₂ and θ phases were observed, but the ζ₂ phase was not recognized. This was also reported in a previous study.⁶⁾ Although the α₂, γ₁, δ and η₂ phases cannot be differentiated from one another in Figs. 1 and 2, we may consider that these four compounds are produced also at T = 483–543 K (210–240°C). Hereafter, the layer composed of various compounds is merely called the intermetallic layer.

Fig. 2

Mol fraction y_i of component i (i = Cu, Al) versus the distance x measured by EDS along the direction shown as a horizontal solid line in Fig. 1.

3.2 Growth behavior of compounds

As mentioned in Section 2, SEM was mainly used for the microstructure observation. From SEM images of the annealed CA wires, the total thickness l of the intermetallic layer was estimated by the following equation.⁶⁾   

\begin{equation} l = \frac{A}{w} \end{equation} (1)

Here, A and w are the total area and the total length of the intermetallic layer, respectively, on the cross-section. The result is shown in Fig. 3. In this figure, the vertical axis indicates the thickness l, and the horizontal axis represents the annealing time t. Both axes are shown in logarithmic scales, and open rhombuses, squares and circles indicate the results of T = 483, 513 and 543 K (210, 240 and 270°C), respectively. As can be seen in Fig. 3, the thickness l monotonically increases with increasing annealing time t at all the annealing temperatures. The higher the annealing temperature is, the faster the intermetallic layer grows. Furthermore, the open symbols for each annealing temperature lie well on a straight line. This means that l is expressed as a power function of t by the equation   

\begin{equation} l = k\left(\frac{t}{t_{0}}\right)^{n}, \end{equation} (2)

where t₀ is unit time, 1 s, which is adopted to make the argument t/t₀ of the power function dimensionless. The proportionality coefficient k has the same dimension as that of the thickness l, and the exponent n is dimensionless. Using the experimental points plotted in Fig. 3, k and n in eq. (2) were determined by the least-squares method, as indicated by dotted, dashed and solid straight-lines for T = 483, 513 and 543 K (210, 240 and 270°C), respectively. The determined values of k and n are represented in Fig. 3. According to these straight-lines, the exponent takes values of n = 0.33, 0.37 and 0.38 for T = 483, 513 and 543 K (210, 240 and 270°C), respectively.

Fig. 3

The total thickness l of the intermetallic layer versus the annealing time t shown as open rhombuses, squares and circles for T = 483, 513 and 543 K (210, 240 and 270°C), respectively. The values of k and n in eq. (2) were estimated by the least-squares method.

As mentioned earlier, the intermetallic layer is composed of the α₂, γ₁, δ, η₂ and θ phases. However, the α₂, γ₁, δ and η₂ phases cannot be distinguished from one another at T = 483–543 K (210–240°C). Hereafter, the layer of the θ (CuAl₂) phase is called layer 1, and that of the α₂, γ₁, δ and η₂ phases is designated layer 2. From the SEM images used for the measurement in Fig. 3, the mean thickness l_i of layer i (i = 1, 2) was determined by an equation similar to eq. (1). The total thickness l of the intermetallic layer corresponds to the sum of l₁ and l₂ as follows.   

\begin{equation} l = l_{1} + l_{2} \end{equation} (3)

The results for l, l₁ and l₂ are shown as open circles, squares and rhombuses, respectively, in Fig. 4. Figure 4(a), 4(b) and 4(c) indicates the results of T = 483, 513 and 543 K (210, 240 and 270°C), respectively. In Fig. 4, the vertical and horizontal axes represent the logarithms of l_i and t, respectively, as in Fig. 3. The values of l₁ and l₂ are smaller than that of l, and hence the open squares and rhombuses are more markedly scattered than the open circles. Nevertheless, we can consider that l_i is proportional to a power function of t according to the following equation of similar form to eq. (2).   

\begin{equation} l_{i} = k_{i}\left(\frac{t}{t_{0}}\right)^{n} \end{equation} (4)

The values of k_i and n for l_i in eq. (4) were calculated from the experimental points in Fig. 4 by the least-squares method as shown with straight lines. The calculated values of k_i and n are shown in Fig. 4. As can be seen, n = 0.22, 0.24 and 0.36 at T = 483, 513 and 543 K (210, 240 and 270°C), respectively, for l₁, and n = 0.46, 0.50 and 0.39 at T = 483, 513 and 543 K (210, 240 and 270°C), respectively, for l₂. Thus, n is smaller for l₁ than for l₂ at each annealing temperature. On the basis of the values of n, we will discuss the rate-controlling process of the compound growth.

Fig. 4

The thicknesses l, l₁ and l₂ versus the annealing time t shown as open circles, squares and rhombuses, respectively: (a) 483 K (210°C), (b) 513 K (240°C) and (c) 543 K (270°C). The values of k_i and n in eq. (4) were estimated by the least-squares method.

3.3 Rate-controlling process of compound growth

The values of n are plotted against the annealing temperature T as open circles with error bars in Fig. 5, where the error bar shows standard error. In Fig. 5, the vertical and horizontal axes show n and T, respectively. Figure 5(a) and 5(b) indicates the results of l₁ and l₂, respectively. The value of n is 0.5 when volume diffusion governs the layer growth of the intermetallic layer.^5,11–19) On the other hand, at low temperatures where volume diffusion is frozen out, boundary diffusion can control the layer growth. In such a case, n becomes less than 0.5 on condition that grain growth takes place in the intermetallic layer.²⁰⁾

Fig. 5

The exponent n versus the annealing temperature T shown as open circles with error bars: (a) layer 1 and (b) layer 2. The common value of n simultaneously estimated from various open symbols in Fig. 8 is also indicated as a solid circle with error bars.

According to the result in Fig. 5(a), n is much smaller than 0.5 for l₁. On the other hand, n is merely slightly smaller than 0.5 for l₂ in Fig. 5(b). Thus, for layers 1 and 2, boundary diffusion as well as volume diffusion contributes to the layer growth. However, the contribution of boundary diffusion is greater for layer 1 than for layer 2. For layer 1 in Fig. 5(a), n monotonically decreases from 0.36 to 0.22 with decreasing annealing temperature from T = 543 K (270°C) to T = 483 K (210°C). This means that the contribution of boundary diffusion becomes more remarkable at lower temperatures than at higher temperatures. In contrast, for l₂ in Fig. 5(b), there is no systematic dependence of n on T, and thus n is considered rather insensitive to T. This means that the growth of layer 2 is controlled mainly by volume diffusion and partially by boundary diffusion independently of T at T = 483–543 K (210–270°C).

In a manner similar to Fig. 5, the values of n for l in Fig. 3 are plotted against the annealing temperature T as open circles with error bars in Fig. 6. Also in Fig. 6, the vertical and horizontal axes indicate n and T, respectively. Since the growth of the intermetallic layer is controlled by boundary and volume diffusion, n is smaller than 0.5 and takes intermediate values between l₁ and l₂ at each annealing temperature in Fig. 6.

Fig. 6

The exponent n for l versus the annealing temperature T shown as open circles with error bars. The corresponding results reported by Gueydan et al.⁴⁾ and Meguro et al.²¹⁾ are also indicated as open squares and rhombuses, respectively, with error bars.

As mentioned in Section 1, the growth of the intermetallic layer in the CA wire was experimentally observed at T = 573–673 K (300–400°C) by Gueydan et al.⁴⁾ According to their observation, the intermetallic layer consists of the γ₁ (Cu₉Al₄), η₂ (CuAl) and θ (CuAl₂) phases and the total thickness l of the intermetallic layer is proportional to a power function of the annealing time t. For the power relationship expressed by eq. (2), they reported that n = 0.5 ± 0.1 at all the annealing temperatures.⁴⁾ Their values of n are plotted against the annealing temperature T as open squares with error bars in Fig. 6.

In a study by Meguro et al.,²¹⁾ sandwich Al/Cu/Al diffusion couples were isothermally annealed at T = 693–753 K (420–480°C) for various times up to t = 5.598 Ms (1555 h). In their experiment, the γ₁, δ, ζ₂, η₂ and θ phases were clearly observed in the diffusion couple annealed at T = 753 K (480°C) for t = 5.598 Ms (1555 h). At this annealing temperature, the α₁ phase is not a stable compound.¹⁰⁾ Hence, all the five stable compounds¹⁰⁾ form under their annealing conditions. Also for their result, eq. (2) holds between the total thickness l of the intermetallic layer and the annealing time t. They reported that n = 0.57, 0.54 and 0.57 at T = 693, 723, and 753 K (420, 450 and 480°C), respectively.²¹⁾ Their values of n are also plotted against the annealing temperature T as open rhombuses with error bars in Fig. 6. If the interface reaction at the moving interface controls the layer growth, n is equivalent to unity.^22–30) The exponent n greater than 0.5 indicates that the interface reaction contributes to the rate-controlling process of the layer growth. According to their observation,²¹⁾ the intermetallic layer grows into both the Cu and Al specimens. The chemical compositions of the ζ₂, η₂ and θ phases are close to the equilibrium ones, but those of the γ₁ and δ phases are shifted from the equilibrium ones to the Al-rich ones. This means that the local equilibrium is not realized at the moving Cu/γ₁, γ₁/δ and δ/ζ₂ interfaces. Among these three moving interfaces, the migration rate is the greatest for the Cu/γ₁ interface. Thus, it is concluded that the values n = 0.54–0.57 is attributed to the interface reaction at the moving Cu/γ₁ interface.

The rate-controlling process of the layer growth is schematically depicted in Fig. 7. The interdiffusion across the intermetallic layer and the interface reaction at the moving interface are serial processes. Hence, the slower process between the interdiffusion and the interface reaction controls the layer growth. On the other hand, the volume diffusion and the boundary diffusion across the intermetallic layer are parallel processes. Thus, the faster process between the volume diffusion and the boundary diffusion governs the interdiffusion. As can be seen in Fig. 6, n < 0.5 at T = 483–543 K (210–270°C), n = 0.5 at T = 573–673 K (300–400°C), and n > 0.5 at T = 693–753 K (420–480°C). Consequently, the interdiffusion controls the layer growth at T ≤ 673 K (400°C), and the interface reaction contributes to the rate-controlling process at T ≥ 693 K (420°C). Furthermore, the volume diffusion governs the interdiffusion at T = 573–673 K (300–400°C), and the boundary diffusion contributes to the interdiffusion at T ≤ 543 K (270°C).

Fig. 7

Schematic for rate-controlling process of compound growth.

The values of l₁ and l₂ in Fig. 4 are again plotted against the annealing time t in Fig. 8(a) and 8(b), respectively. In Fig. 8, the vertical and horizontal axes represent l_i (i = 1, 2) and t, respectively, and open rhombuses, squares and circles show the results of T = 483, 513 and 543 K (210, 240 and 270°C), respectively. From these open symbols in Fig. 8, the independent values of k_i and the common value of n for layer i in eq. (4) were estimated by the least-squares method. The estimation provides k₁ = 1.7 × 10⁻⁸, 2.6 × 10⁻⁸ and 5.6 × 10⁻⁸ m with n = 0.29 for layer 1 at T = 483, 513 and 543 K (210, 240 and 270°C), respectively, and k₂ = 1.3 × 10⁻⁹, 4.2 × 10⁻⁹ and 9.8 × 10⁻⁹ m with n = 0.45 for layer 2 at T = 483, 513 and 543 K (210, 240 and 270°C), respectively. Using these values of k_i and n, the dependence of l_i on t was calculated from eq. (4). The results of T = 483, 513 and 543 K (210, 240 and 270°C) are shown as dotted, dashed and solid lines, respectively, in Fig. 8. The values of n for layers 1 and 2 in Fig. 8 are represented as solid circles with error bars in Fig. 5(a) and 5(b), respectively. On the other hand, the values of k₁ and k₂ are plotted against the annealing temperature T as open circles with error bars in Fig. 9(a) and 9(b), respectively. In Fig. 9, the vertical axis shows the logarithm of k_i, and the horizontal axis indicates the reciprocal of T. As can be seen, the open circles for each layer are located well on the corresponding straight line. Thus, the temperature dependence of k_i is expressed by the following equation.   

\begin{equation} k_{i} = k_{i0}\exp \left(-\frac{Q_{ki}}{RT}\right) \end{equation} (5)

Here, k_i0 is the pre-exponential factor, and Q_ki is the activation enthalpy. From the open circles in Fig. 9, k_i0 and Q_ki were estimated by the least-squares method as shown with solid lines. The estimation provides k₁₀ = 7.3 × 10⁻⁴ m and Q_k1 = 43 kJ/mol for k₁ and k₂₀ = 6.5 × 10⁻³ m and Q_k2 = 73 kJ/mol for k₂. As previously mentioned, the growth of layer 1 is controlled predominantly by boundary diffusion, but that of layer 2 is governed mainly by volume diffusion. The activation enthalpy of diffusion coefficient is much smaller for boundary diffusion than for volume diffusion. This tendency influences the value of Q_ki. This is the reason why Q_k1 is smaller than Q_k2. The rate-controlling process of compound growth may be usually evaluated only from the activation enthalpy. As mentioned earlier, however, information on the rate-controlling process is reflected by the exponent n in eqs. (2) and (4). Such information is essentially important for reliable evaluation of the rate-controlling process. There is a rather complicated relationship between the activation enthalpy in eq. (5) and those of the interdiffusion coefficients for the relevant phases. Such a relationship for reactive diffusion in a hypothetical binary system was quantitatively analyzed using a mathematical model in previous studies.^11–13) According to this quantitative analysis, the rate-controlling process cannot be conclusively evaluated from the activation enthalpy in a straightforward manner. The activation enthalpy may be merely complementarily used for confirmation of the evaluation.

Fig. 8

The mean thickness l_i of layer i versus the annealing time t shown as open rhombuses, squares and circles for T = 483, 513 and 543 K (210, 240 and 270°C), respectively: (a) layer 1 and (b) layer 2. The independent values of k_i and the common value of n in eq. (4) were estimated by the least-squares method.

Fig. 9

The proportionality coefficient k_i of layer i versus the annealing temperature T shown as open circles with error bars: (a) layer 1 and (b) layer 2.

4. Conclusions

The kinetics of compound growth due to reactive diffusion in the Cu-clad Al (CA) wire was experimentally examined in a metallographical manner. The CA wires were isothermally annealed at temperatures of T = 483–543 K (210–270°C) for various periods up to 3.456 Ms (960 h). Owing to isothermal annealing, the intermetallic layer composed of the α₂, γ₁, δ, η₂ and θ phases is produced at the original Cu/Al interface in the CA wire. At T = 483–513 K (210–240°C), however, the α₂, γ₁, δ and η₂ phases could not be distinguished one another by SEM and EDS. The layer of the θ phase is called layer 1, and that of the α₂, γ₁, δ and η₂ phases is designated layer 2. According to the observation, the mean thickness l_i of layer i (i = 1, 2) is proportional to a power function of the annealing time, and the exponent of the power function is 0.22–0.36 for layer 1 and 0.39–0.50 for layer 2. Thus, for both layers 1 and 2, the layer growth is controlled by boundary and volume diffusion at T = 483–543 K (210–270°C). However, the contribution of boundary diffusion to the layer growth is more remarkable for layer 1 than for layer 2. From the annealing temperature dependence of the proportionality coefficient of the power relationship, the activation enthalpy was estimated to be 43 and 73 kJ/mol for layers 1 and 2, respectively. Since boundary diffusion contributes to the layer growth more markedly for layer 1 than for layer 2, the former activation enthalpy is much smaller than the latter activation enthalpy.

REFERENCES

• 1)   K.  Oba: Fujikura Tech. J. 42 (2013) 77–80.
• 2)   K.  Taguchi,  T.  Shimada,  J.  Yoshimoto,  T.  Kuwabara and  Y.  Akasofu: SEI Tech. Rev. 84 (2017) 125–130.
• 3)   S.  Koch and  H.  Antrekowitsch: BHM Berg. 152 (2007) 62–67.
• 4)   A.  Gueydan,  B.  Domenges and  E.  Hug: Intermetallics 50 (2014) 34–42.
• 5)   M.  Kajihara: Acta Mater. 52 (2004) 1193–1200.
• 6)   T.  Kizaki,  M.  O and  M.  Kajihara: Mater. Trans. 60 (2019) 895–901.
• 7)   M.  Kajihara and  W.  Gust: Acta Metall. Mater. 39 (1991) 2565–2574.
• 8)   J.M.  Vandenberg,  F.J.A.  den Broeder and  R.A.  Hamm: Thin Solid Films 93 (1982) 277–280.
• 9)   F.J.A.  den Broeder,  M.  Klerk,  J.M.  Vandenberg and  R.A.  Hamm: Acta Metall. 31 (1983) 285–291.
• 10)  T.B. Massalski, H. Okamoto, P.R. Subramanian and L. Kacprzak: Binary Alloy Phase Diagrams, Vol. 1, (ASM International, Materials Park, OH, 1990) p. 142.
• 11)   M.  Kajihara: Mater. Sci. Eng. A 403 (2005) 234–240.
• 12)   M.  Kajihara: Mater. Trans. 46 (2005) 2142–2149.
• 13)   M.  Kajihara: Defect Diffusion Forum 249 (2006) 91–96.
• 14)   M.  Kajihara: Mater. Trans. 47 (2006) 1480–1484.
• 15)   M.  Kajihara and  T.  Yamashina: J. Mater. Sci. 42 (2007) 2432–2442.
• 16)   M.  Kajihara: Mater. Trans. 49 (2008) 715–722.
• 17)   M.  Kajihara: J. Mater. Sci. 44 (2009) 2109–2118.
• 18)   M.  Kajihara: Mater. Trans. 51 (2010) 1242–1248.
• 19)   M.  Kajihara: Mater. Trans. 53 (2012) 1896–1904.
• 20)   A.  Furuto and  M.  Kajihara: Mater. Trans. 49 (2008) 294–303.
• 21)   K.  Meguro,  M.  O and  M.  Kajihara: J. Mater. Sci. 47 (2012) 4955–4964.
• 22)   Y.  Muranishi and  M.  Kajihara: Mater. Sci. Eng. A 404 (2005) 33–41.
• 23)   T.  Hayase and  M.  Kajihara: Mater. Sci. Eng. A 433 (2006) 83–89.
• 24)   K.  Mikami and  M.  Kajihara: J. Mater. Sci. 42 (2007) 8178–8188.
• 25)   M.  Kajihara and  Y.  Tejima: J. Phys. Conf. Ser. 165 (2009) 012091.
• 26)   Y.  Tejima,  S.  Nakamura and  M.  Kajihara: J. Mater. Sci. 45 (2010) 919–928.
• 27)   M.  O,  Y.  Takamatsu and  M.  Kajihara: Mater. Trans. 55 (2014) 1058–1064.
• 28)   M.  Hashiba,  K.  Masui and  M.  Kajihara: Mater. Trans. 56 (2015) 30–39.
• 29)   Y.  Yamamoto and  M.  Kajihara: Mater. Trans. 42 (2001) 1763–1770.
• 30)   Y.  Yamamoto and  M.  Kajihara: J. Electron. Mater. 37 (2008) 1710–1720.