Calculation of phase diagrams for lead-free power IC die attach
BY J.N. LALENA, M.W. WEISER AND N.F. DEAN
In the quest for lead-free solders, there have been few reports regarding possible replacements for the Pb-5Sn and Pb-10Sn alloys. These alloys are commonly used for die attach in power devices, because of their high melting point, thermal conductivity and thermomechanical fatigue resistance. A project has been undertaken to identify lead-free alloys for this application using the CALPHAD (CALculation of PHAse Diagrams) method.
The task of developing a new lead-free alloy necessitates a complete assessment of all available data, as well as the acquisition of new data. Designing “drop-in” replacements for lead-containing alloys and materials can be technically challenging; this article primarily focuses on thermochemical aspects of designing an alloy for level-one die attach solder used in power devices.
Power Die Attach Requirements
Within a power device package, the die, with backside metallization, is soldered to a copper leadframe that acts as a heat spreader (Figure 1). The solder must meet several stringent requirements, with devices typically generating 100 W of heat or more and normal operating temperatures exceeding 150°C.
Figure 1. Typical construction of a board mounted power device. |
The solder also serves as an electrical and thermal interface material, needing a thermal conductivity on the order of 20 to 30 W/mK. Because of a large difference in the coefficients of thermal expansion between silicon (3 x 10-6/°C) and copper (16.6 x 10-6/°C), the solder joint will also experience shear loading. Therefore, the alloy must have a low shear modulus and, therefore, good thermomechanical fatigue resistance. Finally, when the component is soldered to a PWB, the die attach material should remain below its solidus (the temperature at which melting begins) to prevent detachment of the die from the leadframe.
Presently, the use of the Pb-63Sn eutectic alloy (mp = 183°C) at the board level allows the die attach solder to possess a solidus as low as 230°C. Although they melt somewhat higher than this, Pb-5Sn (mp range = 310-320°C) and Pb-10Sn (mp range = 280-305°C) are commonly used because of their superior mechanical properties. However, once the Pb-free board-level solders are implemented, their reflow temperature could reach as high as 270°C, making this the new minimum acceptable melting point for the die attach solder. The liquidus (the temperature at which all the solid has melted) should be kept below 350°C to prevent thermal damage to the die during the soldering operation.
Figure 2. Candidate materials for Pb-free die attach, based on fundamental material properties. |
Finally, to avoid large capital investments on the part of semiconductor manufacturers, a new alloy should be compatible with existing processes and equipment. Thus, the new solder must also be ductile and malleable enough to be fabricated into wires, ribbons and preforms.
Lead-free Candidate Materials
Based on fundamental metallurgical and chemical principles, it is believed that the major components of lead-free solders will come primarily from groups 11-15 on the periodic table (Figure 2). Specifically, the elements Ag, Al, Au, Bi, Cu, Ga, Ge, In, Mg, Sb, Si, Sn and Zn hold the most promise with regard to potential systems. This is in accordance with their lower melting points and electrochemical activity, and the Hume-Rothery rules that predict stable solid solution formation over a wide range of compositions with minimal differences in the electronegativities, atomic radii, crystal structures and valence electron counts of the pure components. The only binaries from this group of 13 elements that meet the melting criteria for the die attach application are Au-Si and Au-Sn, and these are in fact sometimes used. However, in addition to their high cost, these alloys exhibit less than ideal mechanical properties.
Moving on to higher-order alloys, the number of possible systems is given by combinatorics as n!/[m!(n-m)!], where, in this case, n = 13, and m = 3 for ternary systems or 4 for quaternary systems. Of the 286 ternary systems, only 86 have published phase diagrams, but each can be ruled out based on improper melting behavior, low ductility, or both.
There has been surprisingly little published research on lead-free alternatives for power die attach applications. Zn-4.0%Al-3.0%Mg-3.2%Ga has been proposed as a possibility.1 This alloy does exhibit acceptable melting behavior, having a melting point range extending from 309 to 347°C, but it is not ductile enough to fabricate it into wire or ribbon using conventional techniques. It has also been found to exhibit poor wetting in real die attaching experiments. Ruling this alloy out still leaves a staggering 200 ternary and 714 quaternary systems for which there is little or no available research literature.
Thermodynamic Calculations
The most efficient method of screening alloys from such a large number of potential systems for further in-depth studies will involve phase equilibria calculations. The melting behavior of an alloy can be obtained from the calculated liquidus and solidus projections, as well as appropriate isopleths (vertical section through a solid diagram resembling a binary phase diagram). This data can then be used as the criterion for investigating an alloy's mechanical properties. The CALPHAD approach has been applied extensively in recent years toward materials design, including several solder alloys: Ag-Bi-Sn,2 Ag-Cu-Sn,3,4 Ag-Sn-Zn,5 Al-Mg-Zn,6 Bi-Sb-Sn,7,8 Bi-Sn-Zn,9 Cu-Mg-Zn,10 Cu-Sn-Zn,11 In-Sb-Sn,12 and Bi-In-Sn-Zn.13 The basic premise is that thermodynamic data for systems of fewer components can be used to make phase equilibria predictions on higher-order systems.14
There are several commercially available CALPHAD computer programs,15-17 and the equilibrium calculations used in all of the programs can be described as iterative techniques for minimizing the total Gibbs energy of the system. Generally, constraints (such as composition range) are set and some initial guess of the equilibrium state is made so that the total Gibbs energy can be calculated. Then, a minimization routine, such as the Lagrangian multiplier method, is used to estimate new values for the extensive variables that will cause the Gibbs energy to be decreased. Convergence is obtained when the difference in calculated Gibbs energy between iterative steps has reached some small enough value.14 The coefficients of the Gibbs energy functions are derived from experimental data for all known phases in the system and stored in a database that the program accesses to make the phase equilibria calculations. The Gibbs energy of liquid and disordered solid solution phases are normally expressed by the regular solution model:14
|
null
where xi is the mole fraction of element i, given by the alloy's composition, and °Gi is the reference state of element i. The terms on the right-hand side of Equation 1 represent the Gibbs energies of a mechanical mixture of the constituents of the phase (terms 1-3), the entropy of mixing for an ideal solution (term 4), binary interaction terms (terms 5-7), and a ternary interaction term (term 8).
Equation 1 applies to random mixtures, or substitutional solutions, of atoms in a liquid or disordered solid phase. However, sub-lattice models more appropriately describe interstitial solutions, intermetallic compounds and intermediate phases. Interstitial phases are not very predominant in solder systems because the atomic radii of the different elements are close. Quite often, though, a stoichiometric intermetallic line compound, with a unique lattice structure, forms. These phases can be described with a two sub-lattice model.14
The formation of intermetallic line compounds with a definite stoichiometry or intermediate phases with a homogeneity range in a solder alloy can be detrimental as they essentially have non-metallic properties. Their electrical and thermal conductivity and ductility are typically much lower than the parent metals.
In the absence of a ternary assessment (also known as optimization), the last term on the right-hand side of Equation 1 would not be included in the phase equilibria calculations. In this case, the binary data are merely extrapolated into the ternary system when the third component is added, because the algorithms cannot predict the existence of an unknown ternary phase in the absence of experimental data for that phase. Experience has shown extrapolation of n-1 data into an n-th order metallic system to work well for n < 4, because of the rarity of true quaternary phases.18 Conversely, for n = 3, ternary assessments can become critical for accurate predictions, because the contribution from the last term of Equation 1 to the Gibbs energy can be substantial. Furthermore, neglecting the existence of an intermetallic line compound in a ternary system or of an intermediate phase formed by any of the numerous types of invariant reactions (such as eutectoid, peritectic or peritectoid reactions), would clearly lead to incorrectly predicted liquidus and solidus boundaries.
There are other potential sources for computational error, of importance to solders, that can arise from an incomplete understanding of the system being studied. For example, if miscibility gaps are known to exist and are well characterized, the computer algorithms can account for them. However, unknown miscibility gaps can result in local minima in the Gibbs energy space, which adversely affect the program's ability to locate the global minimum.14 Miscibility gaps are fairly common in, for example, some Al-containing systems and Bi-containing systems.
Calculated Phase Diagrams
Phase diagram calculations were made with the program MTDATA, ver. 4.71, and the NPL Solders Database, ver. 2.2.19 The accuracy of the CALPHAD predictions was ascertained by comparing published, experimentally determined phase diagrams with the calculated ones for several systems. It has been a useful simplification to distinguish three cases.
First, there are ternary systems (A-B-C) with non-existent or negligible ternary interactions. These consist of binary subsets (A-B, A-C, B-C) containing no intermetallic or intermediate phases. Additionally, each binary subset exhibits complete liquid miscibility between components, and there may be complete or limited solid miscibility between components for no more than two subsets. This minimizes the likelihood for formation of a ternary solid solution phase. As expected, for this case there is excellent agreement between the calculated and published phase diagrams. Figure 3 illustrates this for the liquidus projection for Al-Si-Zn.
Figure 3. The (a) calculated and (b) published20 liquidus projections for Al-Si-Zn. |
The second category involves ternary systems that have not been fully thermodynamically assessed and in which an A-B-C ternary solid solution phase may form. This will occur when there is solid miscibility among all of the binary subsets. However, in this category there is still no intermetallic or intermediate phases in any of the subsets. Shown in Figure 4 are the calculated and experimental liquidus projections for Bi-Pb-Sn, for which this case applies. There is close agreement between the two methods for this situation.
Figure 4. The (a) calculated and (b) published21 liquidus projections for Bi-Pb-Sn. |
Finally, there is the case of ternary systems composed of binary subsets that contain a range of intermetallic compounds and/or intermediate phases, as well as possible ternary solubility interactions, and which lack a ternary assessment. In this category, there can be varying levels of complexity. For example, in addition to an A-B-C ternary solid solution, there may exist the possibility for an (A,C)xBy compound of fixed stoichiometry, in which there is preferential substitution for a binary element on one of the sublattices by the third element. There can also be more complex intermetallic compounds with significant variation in stoichiometry. Finally, there may be interstitial ternary phases formed, in which the third element resides in what may be ordinarily empty octahedral or tetrahedral holes between the close packed atoms of the other two elements. Generally, the more complex the binaries, the potentially less accurate the phase equilibria calculations in the absence of a ternary assessment. Figure 5 depicts this on the Cu-Sb-Sn system, for which the agreement is poor.
Discussion
The CALPHAD method is a useful tool for obtaining reliable phase equilibria predictions on simple to moderately complex solder alloys. However, it is important to realize its limitations when applied to more complicated systems, in which there may be unknown interactions. Fortunately, experimental validation of phase boundaries is easily achieved with differential scanning calorimetry (DSC) or differential thermal analysis (DTA). In any case, a computational/experimental procedure is the only reasonable approach for studying the enormous number of unexplored lead-free solder systems for possible use in the power die attach application, which has been relatively ignored in the literature. The advantage of CALPHAD is that these systems can be screened for further study based on their melting behavior in a much more timely manner than by experiment alone.
Our computational work thus far indicates that the addition of significant quantities of a third metal to an existing binary system tends to result in wide pasty ranges with the liquidus surfaces sloping downward and the solidus level below the melting point of the lowest melting pure component. This would be predicted from simple freezing point depression considerations. Hence, the closer one is to the desired melting point to begin with in the binary systems under study, the better. This presents a major problem, however, in that the metals with melting points nearest 300°C are cadmium and thallium, which are considered more toxic than lead.
Figure 5. The (a) calculated and (b) published22 liquidus projections for Cu-Sb-Sn. |
The proper melting point is not the only criterion to be met though. Once an alloy with the proper melting behavior is identified from the phase equilibrium calculations, it must be evaluated with regards to ductility, compliance, thermal and electrical conductivity, and die/substrate wetting. With the exception of wetting, thermodynamics tell us little or nothing about these other properties. Thus, it is important to judiciously select the order the systems are studied. For example, it is expected that a ductile alloy would be unobtainable from a system composed of intrinsically brittle end members (e.g., Ga-Ge-Sb), even in the absence of intermetallic phases. Likewise, the presence of many intermetallics will tend to result in an alloy that is brittle even if the end members are not (e.g., Ag-Au-Sn), in addition to possibly decreasing the accuracy of the CALPHAD predictions.
The mechanical properties of an alloy are also microstructure-dependent. For example, small grain sizes, which can sometimes be obtained by proper solidification processing conditions, are known to provide better deformation relaxation mechanisms. However, this is not always easily achieved in the manufacturing process. It is also crucial that solder also possess adequate mechanical properties. This is important for both its ease of use and the reliability of the solder joint.
Conclusion
The multitude of requirements for power IC die attach has precluded an easy solution. From a performance standpoint, lead is ideally suited for this application. A very large number of systems have been modeled and have not yet identified a so-called “drop-in” lead-free replacement æ one that has all of the thermal and mechanical properties of the high-lead alloys. Most systems that meet the melting criteria are either too hard and brittle (zinc alloys), too costly (gold alloys), or too toxic (thallium alloys). Interestingly, it has been reported recently that high-lead solders were formally exempt from pending European legislation – because of a lack of any replacement.
AP
J.N. Lalena, senior research scientist, M.W. Weiser, technical product manager, and N.F. Dean, research manager, can be contacted at Honeywell, 15128 East Euclid Avenue, Spokane, WA 99216; 509-252-2180; Fax: 509-252-8743; E-mail: [email protected], [email protected] and [email protected].
References
- T. Shimizu et al., Journal of Electronic Materials, 28, 1172, 1999.
- U.R. Kattner and W.J. Boettinger, Journal of Electronic Materials 23, 603, 1994.
- K.W. Moon et al., Journal of Electronic Materials 29, 1122, 2000.
- I. Ohnuma et al., Journal of Electronic Materials 29, 1137, 2000.
- H. Ohtani, M. Miyashita and K. Ishida, Journal of Japanese Institute of Metals 63, 685, 1999.
- P. Liang et al., Thermochimica Acta 314, 87, 1998.
- G. Ghosh, M. Loomans and M.E. Fine, Journal of Electronic Materials 23, 619, 1994.
- H. Ohtani and K. Ishida, Journal of Electronic Materials 23, 747, 1994.
- D.V. Malakhov et al., J. Phase Equilibria 21, 514, 2000.
- P. Liang, et al., Calphad 22, 527, 1998.
- T. Jantzen and P.J. Spencer, Calphad 22, 417, 1998.
- S. Ishihara et al., Journal of Japanese Institute of Metals 63, 695, 1999.
- S.W. Yoon, et al., Acta. Mater. 45, 951, 1997.
- N. Saunders, A.P. Miodownik, CALPHAD (Calculation of Phase diagrams): A Comprehensive Guide, Pergamon Press, 1998.
- B. Sundman, B. Jansson and J.O. Andersson, Calphad 9, 153, 1985.
- G. Eriksson and K. Hack, Metallurgical and Materials Transactions B 21B, 1013, 1990.
- R.H. Davies et al., Applications of Thermodynamics in the Synthesis and Processing of Materials, eds. P. Nash and B. Sundman, TMS, Warrendale, PA, 1995.
- U.R. Kattner, JOM, 49, 14, 1997.
- Computed by means of MTDATA, the NPL Databank for Metallurgical Thermochemistry.
- S.A. Mey, K. Hack, Z. Metallkd. 77, 454, 1986.
- T.-H. Ho, W. Hofmann and H. Hannemann, Z. Metallkd. 44, 127, 1953.
- J.M. Blalock, Jr., J.V. Harding and W.T. Pell-Walpole, Metallography, Structures and Phase Diagrams, Vol. 8, Metals Handbook, 8th ed., ASM International, Materials Park, OH, 1973.