Improved Thermal Conductivity in Microelectronic Encapsulants

The paper summarizes a recent research effort carried out at both Dexter Electronic Materials Division and The Dow Chemical Company. The interdependence of thermal conductivity with crucial properties of polymer-based encapsulants has been investi­gated. The selection of the filler and the resultant filler content in semiconductor molding com­pounds are discussed in view of the practical limitations of composite thermal conductivity. Theoretical predictions based on the Nielsen model are considered. The temperature dependence of composite thermal conductivity has been measured and the effects of different types of polymer matrices on the temperature dependence are shown.

Practical semiconductor-grade molding compounds resulting from this work are extensively investigated.

Introduction

For most of the semiconductor industry’s history, engineering efforts have been focused on higher levels of integration, improved performance, speed, and yield. This has resulted in ever-increasing chip sizes and circuit density. The inability to adequately conduct heat away from the chip has become yet another engineering constraint in many new product designs.

In plastic encapsulated devices, their thermal problems have traditionally been moderated by the use of high-cost embedded heat sinks. However, these packages tend to be susceptible to thermal cracking and are of limited utility in the thinner package configurations. Many engineers continue to press tor ultra-high thermal conductivity plastic encapsulants as the ideal solution to problems of thermal management.

The major ingredient in semi­conductor molding compounds is filler. These are blended into the polymer matrix to modify and improve select composite proper­ties. Improvements in one property, however, often require concessions in another. Such is the case with fused silica, the filler used most commonly in semiconductor encapsulants.

Currently, most large chips and small package devices are transfer molded using a polymer composite consisting of 20–30 wt.% (30–45 vol. %) epoxy resin and 70–80 wt. %. (55–70 vol. %) fused silica (a non-crystalline form of quartz) filler. This cost-effective encapsulant provides excellent performance In most key areas including dimensional stability (a low coefficient of thermal expansion), high moisture resistance, and superior stress and electrical characteristics. It is, however, a rather poor thermal conductor (0.5–1.0 W/mK) due to the low thermal conductivity of the fused silica (1.5 W/mK), approximately seven times higher than that of the neat epoxy resin.

Formulators are seeking improvements by replacing fused silica with higher thermal conduc­tivity fillers while maintaining all other encapsulant’s crucial properties at the current level. A balance is required that maximizes the thermal dissipation effect of the electrically insulating molding compound while achieving superior device performance.

Metals, for example, though excellent thermal conductors are unsuitable for these applications, due to their high electrical conductivity. Some promising materials include thermally conductive ceramics such as crystalline silica, silicone carbide (SiC), aluminum nitride (AIN), boron nitride (BN), beryllium oxide (BeO), and diamond. However, concerns about availability and certain other properties, especially toxicity, make some of these fillers unattractive for commercial use.

Within the scope of the presented study, several thermally conductive ceramics in combina­tion with different polymer matrices have been examined. The interdependence of thermal conductivity with other key properties of the polymer composites is discussed.

Thermal conductivity measurements

Three techniques, Laser Method (1), C-matic, Steady State High Temperature (2), and Transient Hot Strip Method (3) were used to determine the thermal conductivity of neat resins and polymer ceramic composites.

Thermo-mechanical analysis

Thermo-mechanical (TMA) measurements were carried out using the DuPont instruments Model 943.

Sample preperation

Four different thermosets, silicone elastomer (commercially avail­ able), epoxy (commercially available cresol epoxy novolac), polycyanate (4,5) and polybenzo­ cyclobutene (6) resins and three thermoplastic polymers, polypro­ pylene (commercially available), Nylon 11 (commercially available) and crystalline polyamide (7) were employed as a matrix material.

Ceramic fillers tested as thermally conductive additives included fused silica, crystalline silica, aluminum oxide, diamond, aluminum nitride, boron nitride, silicon carbide, and graphite fibers.

The blending of composites and molding of specimens was done according to standard industry methodology and as required by the various measurement techniques.

Results and Discussion

Neat resins

As indicated in Table 1, the thermal conductivity of the evaluated neat resins varied between 0.24–0.4 W/mK. The measured coefficients of thermal expansion of the epoxy, polycyanate, and BCB resins were found to be in the 59.2–86.6 ppm/°C range. Even higher values were measured for thermoplastic polymers, Nylon 11 and polypropylene, CLTE = 184 ppm/°C and 143 ppm/°C, respectively. Crystalline polya­mide, although thermoplastic material, had a CLTE value of 73 ppm/°C. Although the polymer portion of composites provides excellent electrical insulation and environ­mental protection, the polymers do not offer adequate thermal conductivity or dimensional stability. (8-10)

Fillers

The fillers, evaluated in this study, are listed in Table 2. With the exception of fused and crystalline silica intrinsic thermal conductivities of the materials employed were above 40 W/mK. Ceramics of the adamantine structure (11), such as diamond, SiC, BeO, AIN, and BN possess high intrinsic thermal conductivities. As expected, the coefficients of thermal expansion of the tested ceramics are low, within the 0.5–7 ppm/°C range.

Ceramic fillers can provide improvements in thermal conductivity and dimensional stability of the composite. These properties must, however, be carefully balanced with other characteristics that influence the final encapsulant.

Polymer/Ceramic Composites

Theoretical predictions of thermal conductivity of polymer/ceramic composites have been well documented in the literature (e.g. 12-16). Since this paper is mainly concerned with the realistic and practically achievable improve­ments in balanced properties of a molding compound, the basic Nielsen model predictions (13, 14) of composite thermal conductivity are adequate for the discussion.

Effects of the Filler on λ_c

As discussed by Bigg (14), when the intrinsic thermal conductivity of the filter (λ_f) is greater than 100 times that of the polymer matrix (λ_p), there is no significant improvement in the composite thermal conductivity.

Effects of the volume fraction on λ_c

The most important experimental parameter for the fabrication of thermally conductive polymer composites is the volume fraction of the filter. The relative effect of the volume fraction and the filler’s intrinsic conductivity is illustrated in Figure 1. The curves represent the theoretical relationships predicted by the Nielsen model tor the spherical filters of the λ_f/λ_p= 10, 100, and 1000 ratios. An appreciable increase in the filler’s intrinsic conductivity from the λ_f/λ_p = 10 to the λ_f/λ_p = 100 ratio.

Figure 1 further illustrates the importance of the filler’s volume fraction. A substantial improve­ment in relative thermal conductivity above 0.5–0.6 volume fraction filler is clearly predicted. The implication of the model suggests that the increase is limited to the factor of 20 (λ_c/λ_p ≥20). The experimentally determined increase of composite thermal conductivity as a function of filler content is shown in Figures 2 and 3. Figure 2 shows the behavior of crystalline silica-containing epoxy-based composite. lt shows an appreciable increase in λ_c only above 0.4 volume fraction. The maximum loading achieved in this case was 70 vol.%. In Figure 3, the solid lines represent the theoretical relationships predicted by the model tor the λ_f/λ_p = 10, 100, and 1000. As can be seen, the experimental data determined for various polymer matrix materials and several different al fillers, follow relatively well the predictions of the model.

In accordance with the theoretical predictions, the data collected tor the fused silica (λ_f/λ_p = 6.25) containing composites at approx. 0.48-0.60 volume fraction only shows the improvement of composite thermal conductivity from λ_c/λ_p = 1.8 to λ_c/λ_p = 2.6. It can be shown that even at a maximum theoretical loading of 0.63–0.68 volume fraction, the λ_c/λ_p of a fused silica-filled encapsulant would not be substantially increased. The improvement in the relative thermal conductivity, λ_f/λ_p estimated for the fillers satisfying the condition of the critical λ_f/λ_p = 100 ratio was in the 3–12 range. In an agreement with the theory, the λ_c/λ_p of the diamond dust (λ_f/λ_p > 1000) containing composites, was not significantly higher compared to that of the λ_f/λ_p > 100 fillers.

It should be realized that it is, indeed, very difficult to maintain a continuous polymer phase at loadings approaching and exceeding the 0.6 volume fraction. Improvements in the packing density of the spherical fillers can be accomplished by various approaches. For example, employing bimodal or multimodal particle size distribution filler (17) has been known to increase the maximum volume fraction. The specific gravity of the filler is yet another important characteristic greatly affecting the achievable packing density. For example, using a relatively low-density fused silica (d = 2.2) filler, higher volume fractions can be achieved when compared to the higher density fillers, such as AIN (d = 3.26) or Al₂O₃ (d = 3.99).

The effect of a particular polymer matrix on the resulting composite thermal conductivity is rather indirect and reflects the ease of fabrication and acceptance of the filler as well as variations in interfacial resistance between the discrete filter particles and their non-random dispersion in the matrix.

Bond percolation aspect, the formation of thermally conductive clusters at high loadings, and the shape factor A effect have been well discussed (15, 16) and partially experimentally verified by Bujard (12, 18). Although we have observed very similar manife­station of the cluster formation, the detailed discussion of the effects is beyond the scope of this paper.

High Aspect Ratio/Oriented Filler

A limited number of experiments were carried out using non­ spherical fillers. The effects of orientation, parallel versus transverse, of the high aspect ratio filler, graphite fibers dispersed in an epoxy matrix, is summarized in Table 3. Only by orienting the fibers parallel to the heat flow was a large increase in thermal conductivity noted from 0.63 to 16.7 W/mK. Unfortunately, ordering fibers directionally to maximize heat flow is not feasible beyond laboratory scale, nor are carbon fibers suitable for this application due to their high electrical conductivity.

Temperature Dependence of λ_c and λ_c/λ_p

It is generally known (10) that the thermal conductivity of amorphous polymers slightly increases with increasing temperature up to the glass transition temperature at which point conductivity declines with further increases in temperature.

Figure 4 shows the experimentally determined temperature depen­dence (10–90 °C) of thermal, conductivity of the epoxy and silicone elastomer based AIN filled composites. As expected, the composite thermal conductivity of the epoxy (T_g = 145 °C) based formulations slightly increases and that of the silicone elastomer (T_g = –80°C) decreases with the increasing temperature within the same 10–90 °C range. Within the same 10–90 °C range, the intrinsic thermal conductivity of AIN decreases with increasing temperature (11).

Figure 5 shows a similar relationship expressed In terms of the relative thermal conductivity, λ_c/λ_p, as a function of temperature. As illustrated, the relative thermal conductivity for both epoxy and silicone elastomer-based encapsulants decreases with the increasing temperature. It is our observation that temperature dependence of intrinsic composite thermal conductivity (λ_c) is predominantly affected by polymer matrix and that the relative thermal conductivity (λ_c/λ_p) reflects the filler’s characteristics and is expected to be independent of filler loading (18).

Practical Aspects of Molding Compounds

The study has focused primarily on thermal conductivity of composites. This property is dominated by the filler loading. At the volume fractions needed to significantly influence thermal conductivity, other properties of the encapsulant may be significantly compromised. Examples are shown in Table 4.

Since the use of highly conductive fillers at maximum loading is clearly required to achieve high thermal dissipation, concerns regarding dimensional stability and thermal stress become very important. The basic ingredients of the encapsulant, resins, and inorganic fillers, differ significantly in terms of coefficient of linear thermal expansion (C_LTE), Tables 1 & 2. Typically a coefficient of thermal expansion below 17 ppm/°C is desired when designing low-stress encapsulant, because the C_LTE matches that of copper lead frames.

The coefficient of thermal expansion of the encapsulant linearly decreases with the increasing volume fraction of the filler (20). On an equal volume basis, 52.4 vol.%, the coefficients of linear thermal expansion of the AIN-filled thermoset resins were found to be in the same range as those determined for the common, fused silica-containing materials, CTE = 15–25 ppm/°C. On the other hand, the CTE values of the Nylon 11 and polypropylene (commercially available thermoplastic resin) composites filled to approximately 53 vol.% were In the higher region 76–104 ppm/°C and 63–83 ppm/°C, respectively. Of the thermoplastic resins tested, only the crystalline polyamide (CPA) composites, CTE = 12–26 ppm/°C, compared well with the materials based on the thermoset resins.

Reliance on substitution of the more thermally conductive crystalline form SiO₂ or Al₂O₃ or the fused silica at maximum volume fraction is a simple and commercially available modifi­cation. Examination of the ladder study using quartz summarized in Table 5, however, provides several reasons why such a conversion may not be desirable. Even small quantities of quartz significantly degrade stress performance, as evidenced by the increasing CTE and rapidly decreasing stress performance of the resultant molding compound.

As it was stated earlier, many engineers are predicting the need for much higher thermally conductive encapsulants in order to reduce chip face temperatures. lt would, therefore, be instructive to study the effect of composite thermal conductivity (as isolated from other molding compound variables) on chip operating temperature and high-temperature service life.

Figure 6 shows the dependence of θJ_c as a function of the thermal conductivity of the molding compound in a standard package of configuration. The initial doubling of the thermal conductivity over the base fused silica-filled composite shows a significant advantage: however, further equally large increases meet with increasingly limited improvements. Also, several recent papers (21, 22) have concluded that factors other than temperature may play a significant role in extending the high-temperature operational life of ICs.

Conclusion

There are practical limits to a composite’s thermal conductivity as predicted theoretically and verified experimentally.

Molding compounds formulated to maximum conductivity, using the highest conductivity fillers at the highest feasible volume fraction, did not provide either the expected nor needed thermal performance gain. And they still required very significant concessions in other areas of performance. This suggests that a balance needs to be sought that maximizes the thermal dissipation effect of the molding compound while still achieving superior device protection as evidenced by low-stress dimensional stability and moisture resistance.

Filler blends consisting of fused silica and aluminum nitride may eventually offer a three-fold increase in thermal dissipation with minimal performance concessions. Currently, available products still require that a choice be made between high-stress crystalline silica encapsulants where λc = 6-9 W/m°K and truly low-stress compounds based on filler blends of fused silica, SiC, and Al₂O₃ where λc = 4-6 W/m°K.
Increasing thermal conductivity beyond these ranges was shown to only marginally decrease chip surface temperature.

Further advances may be achievable if suitable techno­logies, such as randomly oriented, non-conductive whiskers can be developed and commercialized (23).

Acknowledgments

The authors gratefully appreciate the outstanding effort of Mark Cox of The Dow Chemical Company in coordinating and bringing together the research and resources of both companies that made this joint paper possible. The authors would also like to thank Valerie McLaren of Dexter Electronic Materials Division for the use of her experimental results with diamond dust fillers.

References

1. W.J. Parker, et. af., •Flash Method of Determining Thermal Diffusivity, Heat Capacity and
   Thermal Conductivity,” J Appl. Vol. 32, pp. 1679-1684,
   1961.
2. K.G. Coumou and A.P. Tye; “A Laboratory Instrument tor Rapid Determination of Thermal Conductivities in the Range 0.4-5 W/mK,” presented at the _7th European Thermophys,cal Properties Conference Antwer­ pen, Belgium, June 30-July 4,
   1980.
3. J.J. DeGroot, et. al., “Transient Hot Strip Technique,” Physica. vol. 75, pp. 454-482,
   1974.
4. E.P. Woo, et. al., “Novel Polyaromatic Cyanates,” U.S. 4, 751,323, 1988.
5. E.P. Woo, et_ al., “Production of Polytriazines from Aromatic Polycyanates with Cobalt Salt of Carboxylic Acid,” U.S. 4, 528,
   366, 1985.
6. R.A. Kirchhoff, et. al., “Benzocyclobutenes in Polymer Synthesis,” presented at the 18th International SAMPE Technical Conference, October 7-9, 1986.
7. A. Onder, et. al., “New Thermoplastic  Transparent Polyamides•, presented at the 39th Annual SEP ANTEC Conference, pp. 883-886, 1981.
8. N.L. Hancox, “The Thermal Conductivity of Composite Materials,” Proceedings of the International Conference (A89- 19976-06-39),  Southampton, England; Apr. 13-15, 1988.
9. J.D. Bolt, et. al., “Polymer/Ceramic New Properties,” Advanced Materials & Processes Ine. Metal Progress, Vol. 7 pp. 32-35, 1988.
10. D.W. Van Krevelen, Properties of Polymers; Elsevier Scientific Publishing company, CH. 17, pp. 395-402, 1976.
11. GA. slack, et. al., “The Intrinsic Thermal Conductivity of Aluminum Nitride” J Phys. chem. solids Vol. 48, No. 7, pp. 641-647, 1987.
12. P. Bujard and J.P. Ansemet, ”Thermally Conductive Aluminum Nitride-Filled Epoxy Resin” presented at the 5th IEEE SEM THERM Symposium, 1989.
13. L.E. Nielsen, “Thermal Conductivity of Particulate-Filled Polymers,” J. Appl. Polym. Sci., 17, 3819, 1973.
14. D.M. Bigg, “Thermally Conductive Polymer Composi­ tions,” Polymer Composites; Vol. 7, No. 3, June 1986.
15. R.L. McCullough, “General­ ized Combining Rules tor Predicting Transport Properties of Composite Materials,” Composite Science and Technology. Vol. 22, pp. 3-21, 1985.
16. G.W. Milton and K. Golden, Thermal Conductivity 18, 1985; Plenum Press N-Y, pp. 571-582.
17. D.I. Lee, “Packing of Spheres and its Effect on The Viscosity of Suspensions, Journal of Paint Technology. Vol. 42, No. 550, pp. 579-587, November 1970.
18. P. Bujard, “Thermal Conduc­tivity of Boron Nitride Filled Epoxy Resins: Temperature Depend­ ence and influence of Sample Preparation,” presented at the lntersociety Conference on Thermal Phenomena in the Fabrication and Operation of Electronic Components, pp. 41, Los Angeles, 1988.
19. H.E. Bair, et. al., ”Thermomechanical Properties of IC Melding Compounds,” Polymer Engineering and Science, Vol. 30, No. 10, pp. 609-617, May 1990.
20. A.S. Raghava, “Thermal Expansion of Organic and Inorganic Matrix Composites: A Review of Theoretical and Experimental Studies,” Polymer Composites, Vol. 8, No. 1, p. 1-11, February 1988.
21. A.A. Gallo, “Effect of Mold Compound Components on Moisture-lnduced Degradation of Gold·Aluminum Bands in Epoxy Encapsulated Devices,” IEEE IAPS Symposium, 1990.
22. M. Nakao, et. al., “Degradation of Brominated Epoxy Resin and Effects on integrated-Circuit-Device Wire­ bonds,” Polymeric Materials for Electronic Packaging and Interconnection, J. Lupinsky and R. Moore, eds., Chapter 35, 1989.
23. D. Hansen, et. at., “Heat Conduction in Metal-Filled Polymers: The Role of Particle Size, Shape, and Orientation,” Polymer Engineering and Science, Vol. 15, No. 5, May 1975.