Thermal interface materials (TIMs) play a vital role in the performance of electronic packages by enabling improved heat dissipation. These materials typically have high thermal conductivity and are designed to offer a lower thermal resistance path for efficient heat transfer. For some semiconductor components, thermal solutions are attached directly to the bare silicon die using TIM materials, while other components use an integrated heat spreader (IHS) attached on top of the die(s) and the thermal solution attached on top of the IHS. For cases with an IHS, two TIM materials are used—TIM1 is applied between the silicon die and IHS and TIM2 is used between IHS and thermal solution. TIM materials are usually comprised of a polymer matrix with thermally conductive fillers such as silica, aluminum, alumina, boron nitride, zinc oxide, etc. The polymer matrix wets the contact surface to lower the contact resistance, while the fillers help reduce the bulk resistance by increasing the bulk thermal conductivity. TIM thickness varies by application but is typically between 25 μm and around 250 μm. Selection of appropriate TIM1 and TIM2 materials is necessary for the reliable thermal performance of a product over its life and end-use conditions. It has been observed that during reliability testing, TIM materials are prone to degradation which in turn leads to a reduction in the thermal performance of the product. Typical material degradation is in the form of hardening, compression set, interfacial delamination, voiding, or excessive bleed-out. Therefore, in order to identify viable TIM materials, characterization of the thermomechanical behavior of these materials becomes important. However, developing effective metrologies for TIM characterization is difficult for two reasons: TIM materials are very soft, and the sample thickness is very small. Therefore, a well-designed test setup and a repeatable sample preparation and test procedure are needed to overcome these challenges and to obtain reliable data. In this paper, we will share some of the TIM characterization techniques developed for TIM material down-selection. The focus will be on mechanical characterization of TIM materials—including modulus, compression set, coefficient of thermal expansion (CTE), adhesion strength, and pump-out/bleed-out measurement techniques. Also, results from several TIM formulations, such as polymer TIMs and thermal gap pads, will be shared.
Thermal dissipation in flip chip electronic packages is critical to the overall performance and reliability of the product. The thermal design power of a microprocessor refers to its maximum dissipated heat that needs to be removed when running standard benchmark workloads [1,2]. Heat generated in the silicon chip has to be extracted by the cooling solution through multiple interfaces as shown in Figs. 1(a) and 1(b). Packages using an integrated heat spreader (IHS) create a heat flow path to the cooling solution (see Fig. 1(a)). Thermal interface materials (TIMs) are used to create a lower thermal resistance path between the silicon chip and the copper or aluminum IHS and generally referred to as TIM1. The cooling solution is then attached to the IHS using another TIM interface, referred to as TIM2. For some packages, the cooling solution does not use an IHS and is directly coupled to the silicon die using TIM2 interface (see Fig. 1(b)).
Thermal interface materials play a crucial role in determining the overall thermal design power capability of the component. A variety of TIM systems are widely used in semiconductor packaging. They include thermal greases, polymer gels, phase change materials, gap pads, and elastomeric resin–filler composites. For the TIM1 interface, two commonly used materials are polymer-based TIM (PTIM) and solder-based TIM. PTIM is a composite made of elastomeric resin with thermally conductive filler particles. The resin is usually a silicone-like material, while the fillers are usually made of alumina, silver, or zinc oxide particles. The thermal conductivity of PTIMs is tuned by controlling the filler loading. PTIMs are generally soft and flexible with high thermal conductivity along with the ability to be dispensed and cured in packages during lid attach process. The thickness of TIM1 interface is controlled during the IHS assembly process. In this study, we focus on various methods to characterize the mechanical behavior of PTIM. The other TIM1 material used commonly is solder-based TIM, which typically consists of an indium foil that acts as a soldering medium between the silicon die and the IHS lid. The melting point of solder TIM can range from 60 °C to 210 °C depending on the alloying composition .
In electronic packages, TIM materials undergo a variety of tests to emulate or accelerate loading conditions including but not limited to elevated temperature and humidity tests, high-temperature bake tests, temperature cycling tests, and power cycling tests. During these tests, the TIM is exposed to operating conditions which can adversely affect their mechanical and thermal properties. When a package undergoes temperature change, the mismatch in coefficient of thermal expansion (CTE) of the different interfaces in the package induces stresses along the TIM interface. The package also undergoes a change in shape (convex to concave or vice versa) which impacts the thermomechanical stress distribution in the TIM interface. This can potentially lead to issues like pump-out, delamination, cracking, and void formation, which generally results in lower thermal performance of the TIM . Figure 2 depicts the behavior of TIM as it goes through reliability testing. The TIM is in pristine shape at the beginning of life. However, under test conditions, it could delaminate from the IHS or pump-out (causing air gaps). These air gaps lead to high thermal resistance, thereby causing thermal throttling of the product.
Accurate mechanical property characterization is essential for predicting the thermomechanical behavior of TIM in packages. Figure 3 shows an example contour of peel stress in TIM sandwiched between the IHS and die under operating conditions as simulated by finite element method. It can be seen that the TIM is subject to compressive stresses in the center of the die and tensile stresses at the periphery of the die. The variation of stress in TIM interface, based on its location, is due to the dynamic warpage of the package. The tensile stress of the TIM could result in delamination, pump-out, and eventually the creation of air gaps.
While thermal behavior of TIM is of primary concern in package design, the prevalence of various mechanically induced fail modes (highlighted in the earlier paragraph) emphasizes the need for mechanical characterization of TIM as well. Thermal characterization of TIM in packages like its thermal conductivity, thermal resistance, etc., is detailed in this issue . A survey of existing literature also shows the lack of a comprehensive understanding of mechanical behavior of TIMs. In this paper, we discuss different methods for characterizing the mechanical behavior of TIMs. The mechanical and material properties discussed here are, namely, the elastic modulus, compression set, interfacial adhesion, and coefficient of thermal expansion. We discuss the test methods, challenges posed by the materials, and the modifications/deviations from standard test methods necessary for successful mechanical characterization.
Mechanical Characterization Methods
In this section, the different techniques for modulus measurement in tensile, shear, and compression modes are discussed. In Sec. 2.4, a method for determining the compression set behavior is discussed. Methods for determining the adhesive strength are discussed in Sec. 2.5. Finally, Sec. 2.6 describes the methods for determining CTE of TIM.
Uniaxial Tensile Test.
A dynamic mechanical analyzer (DMA) was used for measuring the uniaxial tensile properties of TIM at different temperatures. In DMA, a small sinusoidal force is applied to the test specimen placed in a temperature controlled furnace. In this study, cured PTIM samples were investigated with TA Instruments Q800 DMA operating under tensile geometry, using multifrequency strain-controlled mode. More details on the DMA test methodology can be found elsewhere . Cured PTIM samples were cut to the 22 mm × 3 mm × 1 mm dimensions and clamped between DMA tensile grips. The sample was heated from −5 °C to 200 °C at 3 °C/min, with a typical dynamic strain of 0.1–1% and a dynamic frequency of 1 Hz. The static loading force was maintained at 125% of the dynamic force to prevent the sample from buckling. Testing soft materials such as TIM is challenging due to (a) tensile grips deforming the soft TIM while gripping, and (b) the TIM samples also deform under their own weight. Hence, alternate methods like dynamic shear strain sweep tests and uniaxial compression tests were investigated.
Materials such as PTIM, thermal greases, and gap pads exhibit viscous response upon loading. A rheometer-based strain sweep technique was found to be helpful to characterize the mechanical behavior of such TIM materials. PTIM is a polymer-based composite which exhibits viscoelastic behavior. Its complex modulus G* consists of storage modulus G′ and loss modulus G″ (or G* = G′ + iG″) which can be determined by a dynamic strain sweep (DSS) test . For a pure elastic (or “solid”) material G* = G′ in the dynamic test, while for a pure viscous (or “liquid”) material G* = G″. The ratio of G′/G″ in a viscoelastic polymer composite depends on conditions such as strain, shear rate, and temperature. This ratio G′/G″ also characterizes whether the material behaves like a “solid” or like a “liquid” under a given condition.
A rheometer is used to collect the rheological data of PTIM materials. Freshly dispensed PTIM is cured inside the rheometer at 125 °C. A DSS shear test is run with parallel plate configuration and constant frequency or shear rate (1 rad/s) at 125 °C. Typically, G′ starts orders of magnitude higher than G″ at low strain. But as strain increases, G″ crosses over G′ and becomes higher than G′. The strain where G′ and G″ cross over each other is known as crossover strain. In theory, when G′ > G″ at low strain, the material is solid-like and more likely to recover its deformation; when G′ < G″ at high strain, the material is liquid-like and less likely to recover its deformation upon load removal. Basically, the crossover strain is a gauge of the ability of a material to recover its deformation once the load is removed.
Pump-out is considered to be a failure mechanism of PTIM during reliability test, especially temperature cycling test . In TIM1 application, because the package warps dynamically with temperature while the IHS shape remains more or less unchanged during temperature cycling tests, the PTIM between the die and IHS experiences cycles of compression and tension. Depending on the level of strain in compression and in tension, a PTIM material may or may not be able to recover its deformation caused by these compression–tension cycles. If PTIM does not recover, pump-out failure is observed during temperature cycling tests. Hence, accurate characterization of TIM combined with warpage optimized package design can minimize or eliminate pump-out risk in packages.
Uniaxial Compression Test.
This section discusses the method of testing-cured TIM candidates using a uniaxial load fixture. Upon curing, a typical TIM hardens into a tough sheet that is prone to cracking. As highlighted earlier in Sec. 2.1, gripping a TIM can be challenging making tensile test methods unfeasible. But these materials can be tested in compression by sandwiching the TIM sample between two compression platens. Standard compression tests at room temperature can be applied directly to test these samples due to the large strains sustained by these samples as well as the large changes in the area of the sample during compression. An in-house built miniature load frame with precision DC motor driven actuator was used for testing TIM at room temperature, as shown in Fig. 4. A compression subpress was adapted to maintain parallelism between the top and bottom platens throughout the test. This was critical for testing thin samples in the thickness range of 500 μm and lower. A viewing window made of thick distortion free quartz glass was used as top platen. The glass platen enabled real-time recording of the area of the sample during compression using a camera framing at 9 frames/s. For soft TIM, the change in the area during compression was typically in the range of 5–30% of the original area depending on the filler loading and sample thickness. This enabled computation of true axial stress in the TIM sample.
Closed-loop DC motor driven linear actuator has a maximum axial load limit of 300 N. The platen compresses the sample at a velocity of 5 μm/s. This enabled quasi-static testing of TIM samples. Samples were typically compressed to a maximum linear strain limit of 50%. During initial testing phases, variations in sample thickness and detection of initial zero strain point lead to test repeatability issues. Due to the thin and soft nature of the TIM samples, detecting the point of contact between the loading platen and the sample was prone to uncertainty. In order to address this issue, the sample was tested first to about 50% strain and the corresponding stress in the sample at peak strain was determined. Subsequent tests were conducted by compressing the samples directly to the maximum stress limit. This simplified the test procedure by removing the need to detect loading initiation point. The test started with the loading platen being few tens of micrometers above the sample. Once the test started, the platen made contact with the sample and registered a force response corresponding to the displacement. The strain values were determined during postprocessing of the recorded force–displacement data.
Two main datasets were collected during the test: (a) force–displacement data from the load cell and encoder at 100 samples/s, and (b) real-time snapshots of the sample during the test at 9 frames/s. Postprocessing algorithms were used for analyzing the test data. Determination of the cross section area of the sample was performed using image processing. True stresses and strains were then computed and plotted. At least three samples were tested for each TIM material for repeatability.
Compression Set Test.
Thermal interface materials are subjected to multiple thermomechanical loading scenarios inside a microelectronics package. One of the vital properties of TIM is its ability to spring-back or recover its original shape upon deformation. For instance, under normal use, TIM material that is sandwiched between the silicon die and IHS lid undergoes tensile and compression loading depending on the temperature of the package and the spatial location of TIM on the die surface. Once the TIM is compressed and then unloaded, it does not recover its original state completely. This volumetric change can contribute to voiding and/or delamination.
Good adhesion of TIM to various surfaces like silicon die, IHS lid, and heat sink is essential for reliable thermal performance of the package. With change in temperature, the various materials in the package expand at a rate dictated by their CTE which result in package shape change due to CTE mismatch. When package shape changes, the TIM could undergo delamination due to tensile and/or shear deformations. Such delamination events can lead to degradation in thermal performance of the package by introducing voids or pump-out of TIM. Couple of techniques used for TIM adhesion strength characterization is described below: (i) simple pull and shear test and (ii) double cantilever beam (DCB) test.
Simple Pull and Shear Tests.
Pull and shear tests were devised to measure the bond strength between a silicon puck attached to an IHS puck using a thin and uniform layer of PTIM at ambient temperature of 25 °C. Figure 6 shows the schematic of a test sample in shear and pull (tensile) mode. The pucks and bond footprint had a diameter of around 12 mm. The TIM was bonded to silicon and IHS pucks using a pneumatic actuator, which had a goniometer and a translation stage for tilt correction so that the silicon puck could be made parallel to the IHS puck. The pneumatic actuator also had a built-in hard stop that allowed for the bond thickness to be adjusted and a built-in heater for in situ curing of the PTIM. The actuator was used to compress a sandwich of silicon puck, PTIM, and IHS puck while simultaneously heating the stack to cure the PTIM. Cured coupons were then gently removed from the tool and attached to an aluminum base plate using cyanoacrylate glue. Pull studs with clevis holes were then attached to the topside of the puck coupons. In a similar fashion, shear test coupons were fabricated as well. Shear testing was used to understand if there were any significant differences in PTIM bond strength when subjected to tensile and shear loading.
Double Cantilever Beam Test.
The DCB test uses fracture mechanics to characterize the adhesive strength. In DCB, the interface of interest is sandwiched between two elastic beams which are then pulled apart to slowly grow a pre-existing crack along the interface. Critical mode I strain energy release rate (GIC) is computed at multiple locations to characterize the adhesion strength as a function of crack length. Strain energy release rate is the energy dissipated per unit area of the newly created surface during fracture. Since the interface toughness tested at low mode-mixity is lower than at other mode-mixity factors typically experienced by interfaces in real units [10,11], the DCB test provides data which envelope, in a lower-bound sense, the loading seen in a real package. Another salient feature of DCB is the ease of producing a mixed-mode loading condition during the test by using cantilever beams of unequal thickness . This test is called asymmetric DCB. Finally, multiple values of GC can be extracted from a single DCB sample which reduces the number of samples needed .
Adhesion characterization using DCB requires significant investment of time into developing a sample geometry and sample preparation process that yields adhesive failure modes. Sample preparation and geometry optimization is also a unique process for different types of adhesives and adherents. The general configuration of adherents and PTIM is shown in Fig. 7. An asymmetric DCB sample with 1500 μm thick beam made from IHS lid and a 300 μm thick silicon beam was used for testing. The beams were 12 mm wide and 50 mm long. Sandwiched between the two pieces was a layer of TIM whose bond line thickness was controlled using spacers at either end of the sandwich. Initially the PTIM layer thickness was around 50 μm which yielded mixed adhesive and cohesive failure modes more often than adhesive failure modes. A DCB sample with asymmetric beam thickness typically yields more adhesive failures. However, after several rounds of trial and error, it was apparent that the 50 μm thick PTIM led to a high degree of resin damage due to maximum filler particle dimensions being close to PTIM thickness. This caused the interfacial precrack to progress cohesively through the material and sometimes adhesively. After multiple rounds of testing, a 300 μm thick layer of PTIM was found to yield adhesive delamination consistently.
Coefficient of Thermal Expansion.
As highlighted earlier in Sec. 2.5, characterizing the CTE of substrate, die, IHS, and TIM is important to explain and also model the mechanical behavior of packages under thermomechanical loading conditions. To determine the CTE of TIM materials, two techniques, namely, thermomechanical analysis (TMA) and three-dimensional (3D) digital image correlation (DIC) are discussed in the next paragraphs.
Thermomechanical Analysis Method.
A TMA tool is designed to accurately track displacement of test specimen under a range of temperatures. Due to their high sensitivity, they are capable of measuring the Tg, glass transition temperatures, as well as the CTE of the sample . In this study, CTE measurements were conducted using a TMA, in the expansion mode. A cured TIM specimen, typically with dimensions of approximately 4 mm × 4 mm × 2 mm, was placed on top of the quartz stage inside the TMA. The TMA expansion probe, also made of quartz, was lowered to be in contact of the top surface of the specimen. A typical loading force of 0.01–0.02 mN was applied to the probe. During each test, the sample was heated first from 25 °C to 260 °C at 10 °C/min, then it was cooled to −50 °C and equilibrated at −50 °C. A second heating scan from −50 °C to 260 °C was followed with a heating rate of 7 °C/min. The first heating scan was designed to eliminate any possible stresses or moisture in the sample, while the second heating scan was used to calculate CTE. CTE1 is calculated between 0 °C and 40 °C, while CTE2 was typically calculated between 200 °C and 240 °C. It is to be noted that CTE1 and CTE2 in this case are not related to the glass transition temperature (Tg) of PTIM, which is less below −80 °C.
For very soft TIM samples, TMA does not work well, since the TIM can flow during the test and the TMA probe can sink into the sample even under a small loading force. In such cases, alternate methods like the 3D digital image correlation need to be used.
Three-Dimensional Digital Image Correlation Method.
Three-dimensional DIC based strain measurement technique is a noncontact full-field method of measuring CTE in materials. In this technique, surface strains are computed from speckled images of the test specimen as it is being heated. Under the absence of any mechanical interactions, pure thermal strains can be determined . A stereo vision system is used to capture the image of the sample to extract 2D strains on the warped sample surface. DIC software was used to extract in-plane strains from the speckled images of the sample. Materials were tested from 25 °C to 260 °C inside a convection oven. Due to the tackiness observed on such materials, dry lubrication had to be used between the sample and the oven surface. Figure 8 illustrates the measurement setup where a thin (1 mm) packed layer of graphite powder was used as a dry lubricant or antistick agent to allow for free expansion of the sample once exposed to high temperature. For this experiment, 1 mm TIM samples were created following current heat spreader attach procedures, and cured samples were then cut to about 5 mm × 5 mm and placed in the oven plate with graphite powder as illustrated in Fig. 8.
In Sec. 2, various methods for characterizing TIM mechanical and material properties were discussed. Three different PTIM materials, namely, material A, material B, and material C, were characterized using these methods. All the three materials are silicone-based PTIM with different filler loadings and filler sizes. In this section, the results from the different characterization techniques are discussed later. Tensile, shear, and compression modulus measurements are discussed in Secs. 3.1, 3.2, and 3.3, respectively. In Sec. 3.4, the results of compression set experiments are presented. Section 3.5 discusses the results from adhesion test methods, including pull test, shear test, and DCB tests on PTIM. Results from CTE measurements are discussed in Sec. 3.6.
Modulus Determined From Dynamic Mechanical Analyzer Tests.
Dynamic mechanical analyzer test results on cured samples of materials A and C are shown in Fig. 9. The Y-axis shows the DMA storage modulus of both PTIM A and C normalized against the modulus value of material A at 25 °C. It is observed that as the temperature decreases from 25 °C to below 0 °C, the storage modulus of material A increases rapidly, similar to what has been observed earlier . Further analysis by differential scanning calorimetry revealed that such an increase in modulus with decreasing temperature is due to the crystallization of silicone oil in the material. When temperature increases from 25 °C to 200 °C, the storage modulus of material A decreases to less than 50% of its value at 25 °C, as shown in Fig. 9. On the other hand, material C is much softer than material A due to two possible factors: (a) relatively lower filler content in material C compared to material A and (b) difference in resin formulations of elastomers in materials C and A, resulting in overall reduction of PTIM hardness in material C. This also leads to 10% lower room temperature modulus compared to that of material A. Material C also showed an increase in modulus with decreasing temperature, but the slope is much smaller because of the softness of this material. For even softer materials, DMA becomes a less reliable method to characterize the mechanical behavior because of various issues as described in Sec. 2.1.
Modulus Determined From Direct Shear Strain Sweep.
Dynamic strain sweep tests on materials A and B were performed at constant frequency or shear rate of 1 rad/s and at 125 °C. Both PTIM materials demonstrate viscoelastic behavior in DSS test. According to DSS test, as shown in Figs. 10(a) and 10(b), material A has crossover strain of 0.8%, which is almost 2 orders of magnitude lower than crossover strain of material B which is 71%. Based on the above rheological analysis, material B is expected to have higher pump-out risk than material A. In-house temperature cycling tests on actual lidded packages show that the thermal properties of packages with material B degrade more than material A. Further inspection of these temperature cycled packages revealed the primary mechanism of the degradation to be the pump-out mechanism. TIM performance predictions based on DSS crossover strains appeared to be in good agreement with reliability performance of actual packages.
Polymer-based TIM degradation also occurs during extended bake tests. Unlike temperature cycle tests, temperature in bake test is maintained constant during the entire duration of the test. Since the package is held at a constant temperature, there is no dynamic warpage-based deformation of the TIM. Unlike pump-out fail mode seen in temperature cycling tests, bake tests lead to a different fail mode in the TIM. In the case of PTIM, storage modulus G′ increases over time during bake, due to side reactions other than the main cure reaction catalyzed by platinum catalyst. Comparing the storage modulus at time-zero versus post 72 h bake for both materials, material A shows a threefold increase in hardening, while material B exhibits 22 times increase in hardness. Both PTIM materials experience different levels of hardening caused by 72 h bake at 125 °C, with material A hardening at slower rate than material B.
Generally, the end-of-line PTIM is the softest and it has the lowest stress induced by package dynamic warpage. As bake testing continues, PTIM becomes “harder” which in turn creates more stress induced at similar strains. When the increasing stress surpasses the adhesion strength of the IHS–PTIM–die joint, a delamination could occur leading to thermal performance degradation. Based on PTIM hardening mechanism, material B is expected to have higher bake degradation than material A. Materials A and B have different filler sizes and resin formulation. PTIM needs to have minimal hardening during bake test in order to achieve good bake reliability performance. However, PTIM hardening rate is not the only property that impacts bake degradation. Other factors such as thickness of TIM could significantly impact the PTIM degradation as well.
Uniaxial Compression Test.
Uniaxial compression tests were conducted on thin PTIM samples made from materials A, B, and C that are discussed here. Samples were compressed up to maximum axial strains of 40%. Figure 11 shows the typical stress–strain plot of TIM under compressive loading. The PTIM material was first compressed to 10% strain limit and unloaded. Then, it was compressed to 20% strain and unloaded, followed by maximum strain of 40%. This cyclic loading of PTIM shows the extent of permanent deformation in PTIM for low, medium, and high strain limits. At low strains of under 10%, PTIM recovers most of the compressed thickness. At 20% strains, the PTIM sample retained up to half of the deformed thickness. At 40% strains, the sample recovered up to one-third of deformed shape. Another interesting observation is the continuous nature of stress–strain curve with multiple load–unload cycles. Such a phenomenon is also observed in rubber like elastomers. A comparison of stress–strain curves for the three PTIM materials, namely, A, B, and C, is shown in Fig. 12. Materials B and C exhibit similar loading behavior up to peak strains of 40%. During unloading, material B exhibited a larger permanent deformation of up to 15% compared to material A which is ∼10%. Material C was the softest of the three PTIMs tested. At peak strains of ∼22%, the material permanently deformed to 45% of its initial thickness. During temperature cycling, higher permanent deformation in compression leads to larger extent of delamination in tension. Although all three materials show varying extents of deformation during compression, the most ideal TIM will be one that remains elastic throughout the compression cycle. Any deformation during compression is recovered completely during unload phase. However, in practice such an elastomer with rigid fillers cannot be formulated easily. Based on the compression tests, it is possible to identify a suitable TIM with minimal permanent deformation.
Compression Tests on Gap Pads.
Using the previously described uniaxial compression test setup, two different elastomeric thermal pad materials D and E were tested to evaluate their compressive behavior. One sample per material was used, in a circular shape with a diameter of 12.7 mm and a nominal thickness of 2 mm. The strain behavior was tested to three different values, in increasing order to understand recovery behavior as well as capture the strain-dependent behavior.
As seen in Fig. 11, for the elastomeric thermal pads tested there was a material clearly stiffer in compression, thermal pad E. This material required, for all strain levels, roughly a fivefold greater stress applied than its softer counterpart. A point worth mentioning is that the area change from start to finish was roughly 20% for both these materials, meaning that the stress would have been overpredicted if the true stress was not accounted for. The current metrology is suitable to provide a higher confidence stress–strain curve for this type of TIM.
Furthermore, upon close inspection the recovery behavior can be seen to be superior for the stiffer thermal pad E after the load is released after the first two strain levels. That is, thermal pad D stores more plastic deformation than thermal pad E and therefore recovers less after the load is released. After the first strain level, thermal pad D recovers a normalized strain of around 0.09% versus 0.13% of pad E, and at the second level the trend is repeated with values of ∼0.1% versus 0.16%. For this set of materials, if stiffness was the driving factor, thermal pad D would be deemed the superior choice since it is the more yielding of the two. However, if the main property of interest was the recovery behavior, then thermal pad E would presumably be the better option if the strain levels expected are in the range of the levels tested. As previously stated before, the ideal case would be the combination of lower stiffness coupled with higher recovery behavior in compression to minimize contact resistance during operating conditions.
Compression Set Tests.
One of the vital mechanical properties of TIM is its ability to spring-back or recover its original shape upon deformation. Section 3.3 discusses the characterization of TIM samples compressed cyclically up to three different strains and measures the extent of recovery after unloading. An alternate method of testing the TIM samples during compression is the compression set test. ASTM D395 describes the method of measuring compression set tests for elastomers such as rubber . Typical sample thickness requirements for these tests are either 12.50 mm or 6.25 mm. In this study, we utilize a modified compression set technique for TIM samples which are much thinner (less than 150 μm) compared to the typical requirements of the ASTM deviating from the sample size requirements of the ASTM D395 standard. For these experiments, sample dimensions of 120 μm thickness and 4 mm diameter were created and cured by sandwiching between two thin glass slides. Figure 13 shows the normalized compression set values for PTIM B and C. Material B exhibits 40% higher value of compression set compared to material C. Under the specific conditions tested, material C would be better at filling a void after being compressed and allowed to recover. This study does not factor the presence of air/voids or the different filler densities for these materials. This metrology illustrates an important mechanical behavior of TIM which needs to be considered along with thermomechanical characterizations during material down-selection for a semiconductor package.
Pull tests and shear tests were done on TIM samples to measure the strength of the TIM–IHS lid interface. Although the targeted failure mode is adhesive failure of the TIM to the surface it is bonded to, such a fail mode was not observed consistently in repeated measurements. However, it is useful to distinguish gross differences between adhesive types or thermal treatments. As shown in Fig. 14, failure loads could vary by as much as 70% of the average failure load. However, some differences in mean strength were noted for the different PTIM materials. Little-to-no change was noted after baking the coupons.
Shear tests conducted show the differences in shear strength and tensile strength for the same type of PTIM, as seen in Fig. 15. The PTIM material appeared to be up to two times stronger in shear than in tension.
Apart from tensile pull and shear tests, DCB testing offered greater sensitivity to distinguish PTIM types and differences in thermal treatment or use conditions. Figure 16 shows the normalized adhesion strength for the three PTIMs tested in this study. Material A showed the highest initial strength followed by material C and then material B. Material A showed significant decrease in strength after 24 h of aging, making it weaker than the other two PTIMs tested. Material C showed only a slight decrease in strength with aging, and material B showed a slight increase. It is to be noted that the average adhesion strength of TIMs is 2 orders of magnitude weaker than typical epoxy (used in packaging) to silicon. The very low nature of adhesion also explains the poor performance of TIM when in tension.
Coefficient of Thermal Expansion Measurements.
Figure 17 shows a typical example of the relative sample length as a function of temperature for material A when characterized using TMA. In the plot, the sample length was normalized against the length at 25 °C. As evident from the plot, the total thermal expansion from 50 °C to 260 °C is about 1.7%, which is considerably higher than that of copper or other typical underfill materials. CTE measurements on material C using TMA were not possible due to the soft nature of the samples. Hence, 3D DIC based CTE characterization method was utilized.
Figure 18 shows the normalized CTE curves measured from samples made from materials A, B, and C. CTE values computed from strains along X and Y directions are plotted. In all the three cases, the CTE values were driven by that of silicone resin properties. As observed in the TMA results, the values of CTE are about 1 order of magnitude higher than that of copper. Of the three materials, material B has this highest filler loading and hence has the lowest CTE value among the three PTIM candidates tested.
Thermal interface materials are critical to package performance and reliability over the life of the product. However, like any material, TIM also degrades over time when subjected to temperature, moisture, and cyclic loading, and that deteriorates the thermal performance of the product. The underlying cause for TIM degradation during use condition and reliability testing is usually related to change in its mechanical behavior. Mechanical characterization of TIM poses a lot of challenges owing to its softness and very small thickness. In this paper, various relevant metrologies developed at Intel for mechanical characterization of TIM overcoming the above challenges were described in detail. The metrologies covered here include the following: (i) tensile, shear, and compression tests for modulus measurement; (ii) compression set test; (iii) adhesion strength measurement using pull, shear, and double cantilever beam tests; and (iv) CTE measurement using thermomechanical analyzer and 3D digital image correlation based methods. Even though most of the results were obtained using PTIM samples, these metrologies are easily applicable to other TIM types, like thermal greases, thermal gap pads, and phase change materials, as well.
Knowing the mechanical behavior of TIM materials would enable selection of the right TIM for any given product. For instance, if pump-out risk is of primary concern, then crossover strain from the dynamic strain sweep shear test could be used to rank materials and pick the best candidate. From a package design perspective, it is important to have TIM materials under compression during package use and testing. But, as observed in the compression set test data collected on PTIM samples, TIM materials undergo permanent shape change under compression. Hence, choosing TIM materials with optimal performance in both tension and compression is critical to package performance. Adhesion data collected on PTIM and thermal grease samples using the DCB method clearly indicate that the adhesion strength of these TIM materials to both die and IHS lid surfaces is pretty weak compared to other epoxies used in packages like underfill or IHS sealant. This is not a surprise as these TIM materials are very soft compared to say underfill.
Moreover, the mechanical characterization data, like modulus and CTE, can be used in finite element method simulation models to risk assess TIM materials for use in a given product. This would help minimize the number of reliability tests to run on the product and thus help save collaterals, resources, time, and money.
The authors would like to acknowledge Gaurang Choksi, Boxi Liu, Sergio Chan, Mohammad Kabiri, Michael Drake, David McCoy, and Tony Cortez for their valuable contributions to this study.