Substrate Attach Design Guidelines for Bonding Alumina to Aluminum

Substrate Attach Design Guidelines for Bonding Alumina to Aluminum

In an effort to achieve higher density microelectronic packages, various ceramic substrate packaging techniques are being introduced. For instance, one of these techniques eliminates the need for a carrier so that a substrate is bonded directly to a housing. The advantages of this technique include the elimination of an intermediate piece of hardware (the carrier) and the associated costs for drawings, manufacturing, assembly and documentation with a slight improvement in electrical performance. On the other hand, the benefit of a modular design, which has lower replacement costs, is sacrificed as is remote testing capability. The carrierless substrate attach approach is wholly dependent on the integrity of the adhesive attachment system. Electrical performance desiring minimal bond line thickness opposes the structural requirements of increased bond line thickness. This article discusses design guidelines developed for two different adhesives to determine the minimum bond line thickness required for a given substrate size to withstand the effects of the thermal cycling. This minimum thickness is determined through material property testing, finite element analysis (FEA) and validation, by test, of the analytical predictions.

Theodore Trembinski

AIL Systems Inc.

Deer Park, NY

©1995 International Microwave Packaging Society . Reprinted with permission from International Microwave Packaging Society Conference Proceedings, September 1995; and AIL System Inc.'s Topics in Engineering, Vol. VII, 1996, pp. 4-1-4-18.

The mechanical substrate attach effort assessed the structural integrity of an adhesive joint that bonds dissimilar materials together, namely alumina and aluminum. In the past, the standard design practice of mounting such dissimilar materials having a large thermal coefficient of expansion (TCE) mismatch was to solder the alumina substrate to a Kovar® carrier (alumina and Kovar having a matched TCE) and attach the substrate/carrier assembly to an aluminum housing (alumina/Kovar having a mismatched TCE to aluminum) using screws. Thus, the stresses generated due to thermal loading were absorbed by the carrier/hardware combination and isolated from the fragile substrate. Current design practices have eliminated the Kovar carrier and associated hardware and involve bonding the substrates to the aluminum housings directly.

While adhesives exist that are meant for applications involving dissimilar material with large differences in TCE, care still must be taken with regard to the bond line thickness. Since the stress that produces failure in the adhesive is a function of temperature, substrate size, the difference in TCE of the two adherends, the elastic modulus, the thicknesses of all rigid materials and the thickness of the adhesive joint (and with most variables predetermined in this application), the only variable remaining to reduce the stress for a given adhesive is the thickness of the adhesive joint. The proper selection of the adhesive joint thickness can ensure the mechanical success of the carrierless design.

The Purpose

The purpose of the mechanical substrate attach effort was to create a guideline to aid in the design of assemblies in which substrates are attached to housings by an adhesive. These assemblies are subjected to thermal cycling through either test (for example, MIL-STD-883, Method 5011.21) or the thermal cycling experienced during actual usage. The assembly configuration addressed in this study was an alumina substrate attached to an aluminum housing by adhesive. Two adhesives were evaluated and designated adhesive A and adhesive B. The resulting design guideline is presented in the form of a curve plotting substrate size vs. minimum adhesive thickness for each type of adhesive. The curve can be used to determine quickly the minimum bond line thickness required for a given substrate size.

Design Curve Validation

A validation experiment was conducted to measure the accuracy of the analytically predicted design curves. The experiment addressed the three regions of each curve through bonded samples of alumina to aluminum such that their size and bond thickness yielded points in the three regions of the curves. The regions include the area above the line where an adhesive failure would be expected, the area on or close to the line where failure might be expected and the area below the line where no failures would be expected.

The experiment involved thermally cycling three different substrates (1" ¥ 1", 2" ¥ 2" and 2" ¥ 4" by 0.025"-thick alumina) in quantities of three samples each, bonded to a typical aluminum housing (0.188" thick) with each of the two adhesives. Note that thinner aluminum housings may not yield similar results. Each sample was subjected to three sets of 10 thermal cycles (-65° to +150°C) and five drop tests (1500G shock). The first set of testing fulfilled MIL-STD-883 requirements and the remaining sets represented additional cycles/drops that could be recognized by a unit that has undergone rework or repair.

To achieve substrate size/adhesive thickness combinations in each of the three regions of the curves for the different adhesives, the samples were constructed such that the 2" ¥ 4" samples should be well above the line and fail; the 2" ¥ 2" samples should be on or near the line and go either way, depending on resulting bond line thickness; and the 1" ¥ 1" samples should survive the test failure free. Since the adhesives are available in standard preform thicknesses, the substrate size was chosen to yield the required data points. The adhesive preform thickness was 0.003" and 0.002" for adhesives A and B, respectively. Table 1 lists the test results. All substrates were inspected visually after bonding and after completion of each phase of testing. The failure criterion for the adhesives was a visible sign of adhesive cracking, which usually was accompanied by lifting and/or complete separation of the substrate from the housing. All adhesive A failures exhibited complete separation of the substrate from the housing, while all adhesive B failures ranged from lifting of a corner to lifting of 50 percent of the substrate from the housing but never complete separation. The design curves are considered valid since correlation between analytical prediction and the validation experiment was good.

The Approach

Since an evaluation by test, that is, thermally cycling approximately 100 assemblies to failure, would be too time consuming and costly, an alternative had to be chosen. The best solution for this task was simulation by analysis. FEA was chosen as the tool to simulate the many different assembly configurations subjected to thermal cycling without having to build any hardware.

Before any analysis could be performed, the material properties of the adhesives at -65°C had to be measured. Some properties were available from the vendors (TCE and Poisson's ratio). However, the shear modulus and shear strength had to be derived by experimentation. The need for material properties prompted a series of coupon tests to obtain the necessary properties.

Seven FEA models of an alumina substrate bonded to an aluminum housing were constructed. The sizes ranged from 0.25" ¥ 0.25" to 4" ¥ 4", and all were 0.025" thick. The adhesive thickness used for each substrate size was varied at seven discrete values ranging from 0.0005" to 0.0200" for a total of 49 FEA models for each adhesive type. Subsequently, 98 analyses were performed to yield the necessary data points to create the design guideline curves for both adhesive types.

Coupon Testing

The purpose of the coupon testing was to measure select material properties at both the hot and cold temperature extremes of MIL-STD-883 (+150° and -65°C, respectively) and to correlate FEA predictions to actual test results to ensure the accuracy of the modeling technique and adhesive properties being used. Once the FEA results showed close correlation to test results, the same FEA methodology was used to simulate the alumina-adhesive-aluminum configuration with a high level of confidence.

Curing and assembly of the test samples were important steps in the coupon testing phase of the study. Prior to the assembly of the coupons, each aluminum strip was inspected for flatness and held to a flatness criterion of 0.0015". The pieces were etched chemically prior to assembly. Both of the adhesives were cured using the manufacturer's recommended cure schedule.

Coupon testing followed the American Society for Testing and Materials' (ASTM) standard test method ASTM D1002 for the lap shear testing.² Also, testing of the strength properties of an adhesive in tension by tensile loading was conducted. Table 2 lists the coupon test results. The test lots consisted of at least four samples per lot and the results were averaged for each lot. Testing was conducted at three different temperatures: -65°, +20° and +150°C. Three types of coupon testing were conducted with each adhesive: a single lap shear, a double lap shear and a tension test. The single lap shear test was conducted per ASTM D1002, which also can be used as an inspection criterion for material acceptance. The double lap shear test follows the same procedure as the single lap, but was conducted to determine if the pure shear properties of the adhesive differed from the single lap shear test results. The tension test was conducted to determine how the tensile properties compared to the shear properties. The single and double lap shear results for both adhesives were approximately the same for the -65° and +20°C temperature tests. The +150°C test yielded unanticipated results, which may be due to testing the materials at their cure temperature.

An FEA Model Description

The FEA models were constructed and analyzed using the ANSYS computer-aided engineering system. The alumina substrate, adhesive and aluminum housing were constructed using three-dimensional solid elements. Only one-quarter of each assembly needed to be modeled due to geometric symmetry. Since this work was a parametric study of substrate size and adhesive thickness, the finite element aspect ratio was kept constant so as not to bias the results.

The finite element aspect ratio was unity and the finite element size was 0.05" ¥ 0.05". This self-imposed constraint yielded over 12,000 finite elements for the largest model. The only finite element dimension that was permitted to change was the thickness. The adhesive layer was modeled with four elements through the thickness since test problem results indicated that further mesh refinement was unnecessary. Variations in the element aspect ratio had a greater effect on the results and were removed from the analysis by keeping the ratio constant for all analyses.  

Assumptions

All material properties were assumed to be isotropic and linearly elastic. Material properties not supplied by the vendor were calculated based on the coupon testing results. For both adhesives, the shear modulus was calculated based on the lap shear test results. Table 3 lists the -65°C cold temperature material properties used in the FEA for both adhesives.

Loading Conditions

The analysis was based on the thermal cycle from +150° to -65°C. This temperature range was based on the adhesive cure temperature of +150°C and the MIL-STD-883 cold temperature extreme of -65°C. It was assumed that when heated to +150°C during the cure cycle all materials undergo free thermal expansion. Once the adhesive cures and the assembly cools down and attempts to return to its original room temperature dimensions (each material proportionally to its respective TCE), internal stresses arise from the TCE mismatch.

A Discussion of FEA Stress Results

The first analysis was of the lap shear coupon test itself. The FEA model, shown in Figure 1 , was built to simulate the lap shear test for both adhesives. The load at failure was applied to the model and the resulting deflected shape, shown in Figure 2 , and shear stress were predicted. The adhesives' shear stress was compared to the lap shear strength measured during coupon testing. The shear strength of adhesives A and B was 1082 and 2553 psi, respectively, at -65°C. The FEA results predicted the shear stress to be 1172 and 2743 psi for adhesives A and B, respectively, as shown in Figure 3 . Both sets of results correlated to within eight percent of their respective shear strengths. Since these initial results correlated favorably, the same modeling technique was applied to the alumina substrate models.

Seven cases were analyzed, varying the substrate size from 0.25" sq. to 4" sq. The exact substrate dimensions for each case are listed in Table 4 . Figure 4 shows the FEA model for case 7. Seven analyses for each case were conducted for the adhesive bond lines of 0.0005", 0.001", 0.002", 0.003", 0.005", 0.010" and 0.020".

Symmetry boundary conditions were applied to the quarter symmetry model boundaries and the thermal stress analyses were conducted over a temperature range from +150° to -65°C. The maximum shear stress from each analysis is listed in Table 5 .

Typical FEA results are shown in Figure 5 , where the maximum shear stress of adhesives A and B are shown for a 1" sq. substrate with a 0.002" bond line. A family of curves then were generated for each adhesive. The adhesive thickness vs. shear stress for a thermal cycle temperature range of +150° to -65°C is shown in Figure 6 .

The points where each curve crosses the shear strength or failure line (1028 psi for adhesive A, 2553 psi for adhesive B) were recorded. These intersection points now represent the bond line at failure for each adhesive for a given substrate size and thermal cycle temperature range. Another set of points where each of the curves crosses the allowable (factor of safety of two) lines was recorded. Assuming a safety factor of two based on the maximum shear strength of the adhesives (541 and 1227 psi for adhesives A and B, respectively) yielded an allowable stress line. The intersection points now represent the desired or preferred bond line for each adhesive for a given substrate size and thermal cycle temperature range. Two additional curves of substrate size vs. adhesive thickness are shown in Figures 7 and 8 for adhesives A and B, respectively. Diagonal length was chosen as the ordinate scale to generalize the curve for any rectangular substrate even though the FEA considered only square substrate geometries.

Conclusion

Using the developed design guideline charts, a designer would enter the chart with a known substrate size, move across to the allowable curve and then down to the abscissa to obtain the adhesive thickness that is based on a safety factor of two as a function of the adhesive shear strength. This selection would be considered good design practice. Use of an adhesive thickness between the allowable and failure curves involves some risk and should be avoided. To ensure the mechanical success of the adhesive bond, it is recommended that at least the thickness determined by the allowable curve be used since other factors that were not accounted for could have adverse effects on the adhesive bond. These variables, which were beyond the scope of this work but were accounted for with the safety factor of two, include temperature ranges other than +150° to -65°C, thermal shock and fatigue (the number of cycles to failure at a given temperature range).

The analysis was conducted for one thermal cycle since no fatigue data were readily available for the adhesives and the creation of these data also would be beyond the scope of this effort. It can only be assumed that numerous thermal cycling would tend to weaken the adhesive bond.

Acknowledgment

The author wishes to thank T. Stenswold for his assistance in the development and preparation of the test coupons, J. Streich for the many useful technical discussions and for reviewing the manuscript, and M. Dituri for preparing the manuscript.

References

1. MIL-STD-883, "Evaluation and Acceptance Procedures for Polymeric Adhesives."

2. ASTM D1002, "Strength Properties of Adhesives in Shear by Tension Loading (metal-to-metal)."

Theodore Trembinski received his BSME and MSME degrees from the Polytechnic Institute of New York in 1982 and 1986, respectively. His expertise lies in the application of the finite element method to structural dynamic analysis and design. Trembinski's most recent projects include the Ball-grid Array and Substrate Attach IRAD programs and the structural dynamic FEA of the L-band antenna when subjected to the Navy MIL-S-901 high impact hammer shock. Previously, he was responsible primarily for FEA in the EF-111A System Improvement Program's AVIP effort. He was also employed by the Analysis and Design Application Co. (ADAPCO), a private consulting firm, where he provided design and analysis support to the major gas turbine engine manufacturers. Currently, Trembinski is a senior engineer in the Mechanical Engineering and Analysis Department of AIL Systems Inc. He is a member of the American Society of Mechanical Engineers and the International Electronics Packaging Society.