Filler-depletion layer adjacent to interface impacts performance of thermal interface material

When installing thermal interface material (TIM) between heat source and sink to reduce contact thermal resistance, the interfacial thermal resistance (ITR) between the TIM and heat source/sink may become important, especially when the TIM thickness becomes smaller in the next-generation device integration. To this end, we have investigated ITR between TIM and aluminum surface by using the time-domain thermoreflectance method. The measurements reveal large ITR attributed to the depletion of filler particles in TIM adjacent to the aluminum surface. The thickness of the depletion layer is estimated to be about 100 nm. As a consequence, the fraction of ITR to the total contact thermal resistance becomes about 20% when the TIM thickness is about 50 μm (current thickness), and it exceeds 50% when the thickness is smaller than 10 μm (next-generation thickness).


Filler-depletion layer adjacent to interface impacts performance of thermal interface material
Susumu Yada, Takafumi Oyake, Masanori Sakata, and Junichiro Shiomi a When installing thermal interface material (TIM) between heat source and sink to reduce contact thermal resistance, the interfacial thermal resistance (ITR) between the TIM and heat source/sink may become important, especially when the TIM thickness becomes smaller in the next-generation device integration.To this end, we have investigated ITR between TIM and aluminum surface by using the timedomain thermoreflectance method.The measurements reveal large ITR attributed to the depletion of filler particles in TIM adjacent to the aluminum surface.The thickness of the depletion layer is estimated to be about 100 nm.As a consequence, the fraction of ITR to the total contact thermal resistance becomes about 20% when the TIM thickness is about 50 µm (current thickness), and it exceeds 50% when the thickness is smaller than 10 µm (next-generation thickness).C 2016 Author(s).All article content, except where otherwise noted, is licensed under a Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
[http://dx.doi.org/10.1063/1.4941354]Recent innovation in nanotechnology has made it possible to manufacture highly integrated circuits with nanometer-size components.2][3] In standard thermal management, the heat generated in the chip is transferred to heat sink (e.g.heat spreaders or thermal fins).While the thermal contact resistances due to imperfect contact between the chip and the heat sink can bottleneck the heat dissipation, thermal interface materials (TIMs) are used to improve the thermal contact. 4IMs are typically composed of mechanically-compliant resin to fill the gap at the contact, and thermally-conductive fillers to lower the thermal resistance.
For a given TIM, in theory, the thinner the TIM is the smaller the thermal resistance is.However, the TIM often needs to comply with the dynamic deformation of the contact.Figure 1 shows a typical contact between a chip and a heat spreader.In this case, the heat spreader contacting the chip is installed by being bonded to the circuit board.Since the chip (e.g.silicon) and the circuit board (e.g.organic material) have largely different coefficients of thermal expansion, 5 the temperature-rise during the operation deforms the circuit board and thus the heat spreader with respect to the chip (Fig. 1).This gives rise to a dynamic gap between the chip and the heat spreader, 6 therefore, TIM needs to be thicker than the maximum gap, which is about 50 µm in the current technology. 7This means even by using a TIM with thermal conductivity of several Wm −1 K −1 , TIM can occupy a significant fraction of the total thermal resistance of integrated devices.
][10][11] Another line of effort has been made to reduce the dynamic gap by matching the thermal expansion coefficients or through integration design. 12,13With a realistic scope that thickness of the gap (thus TIM) will become smaller in near future, interfacial thermal resistance (ITR) between TIM and heat sink should become more important as its impact increases with decreasing TIM thickness.The importance of ITR has been pointed out by Campbell et al. 14 for a curing TIM.They measured thermal resistance of cured epoxy-based TIM in contact with silicon or aluminum (Al) by using laser-flash method.By the differential analysis (i.e.subtracting the separately measured thermal resistance of suspended TIM), they have found that ITR can be significant and the values range from 2 mm 2 KW −1 to 20 mm 2 KW −1 .Furthermore, Gowda et al. 15 observed the interface between the cured TIM and Al by computed tomography, acoustic microscopy, and scanning electron microscopy, and showed the existence of delamination and filler-depleted/resin-rich region at the interface.
Although previous works have shown that the defects can give rise to significant ITR, it is not clear if they are generated during the curing process or it is an intrinsic property of the interface.To this end, in this work, we have studied ITR between non-curing TIMs and Al.Three different convectional TIMs with Silicone-oil resin (G776 , G765, X23-7921-5) from Shin-Etsu Chemical 16 are used.As summarized in Table I, these TIMs have different thermal conductivities and viscosities but similar density.Since the above-mentioned differential analysis cannot be applied to non-curing TIMs, direct measurement of the ITR is required.][19] The measurement results reveal the presence of intrinsic ITR even without the curing process.We then further evaluated the importance of the ITR by comparing with the internal TIM resistance for various TIM thicknesses.
The measured system described in Fig. 2(a) is composed of three layers: SiO 2 substrate, Al thin layer, and TIM.The target of measurement is the ITR between the Al layer and TIM, and SiO 2 layer is used to support the Al layer.Al layer is deposited by vacuum evaporation on the SiO 2 surface.The roughness of Al surface was measured to be less than 4 nm by the atomic force microscopy.Al layer also plays the role of transducer for the TDTR measurement.In TDTR method, Al surface is instantaneously heated up by the pump beam, and the temperature response of the Al surface is measured through the change in reflectivity that is proportional to temperature (thermoreflectance) by detecting the power of the probe beam.Ti-sapphire laser with wave length of 800 nm and pulse repetition rate of 80 MHz is used for both pump and probe beams.Note that the thermoreflectance of Al is large around 800 nm. 20The wave length of the pump beam was shifted to 400 nm by a  BIBO crystal to reduce the noise in the detection of the probe beam.To further realize high signal to noise ratio, the pump beam is modulated at a frequency (varied from 1 MHz to 11 MHz) and the corresponding signal from the probe beam was picked up by the lock-in-amplifier. 18,19,21The modulation determines the thermal penetration depth as d =  α/π f , where α is thermal diffusivity and f is modulation frequency. 22For example, the above frequency range gives the penetration depth in the range between 100 nm and 300 nm when α is 0.1×10 −6 m 2 s −1 .
The target thermal properties of the sample can be obtained by fitting the signal time history with the heat conduction model. 18,19We assume that the pump beam completely passes through the SiO 2 layer and is only absorbed at the surface of Al, and the generated heat is conducted in both directions to SiO 2 and TIM. 19For the sake of simplicity, we assumed one-dimensional heat conduction because the radius of probe beam and pump beam (13 µm and 18 µm, respectively) is sufficiently larger than the penetration depth.The model incorporates ITR as a thermal resistance with infinitesimally small thickness (Kapitza thermal resistance model 23,24 ).Thermal conductivities and volumetric heat capacities of Al and SiO 2 were taken from the bulk values 238 Wm −1 K −1 and 1.48 Wm −1 K −1 , and 2.42×10 6 J/m3 K and 1.65×10 6 J/m 3 K, respectively.The thickness of Al layer (34 nm) and ITR of Al/SiO 2 interface (0.005 mm 2 KW −1 ) were obtained by performing TDTR measurement for SiO 2 /Al sample before depositing TIM with the two variables as the fitting parameters.Note that Al thickness is made as thin as possible in the range that fully absorbs the laser to maximize the sensitivity of Al/TIM interface.The fitting was done for the ratio of the real part (V in ) to the imaginary part (V out ) of the TDTR signal, -V in /V out .
The properties of the three different TIMs used in this study are shown in Table I.These TIMs are composted of Silicone-oil resin and oxide/metal particle fillers. 16The TIMs have different viscosities but they all have enough mechanical compliance to the Al surface.First, we performed TDTR measurements with the modulation frequency of 11 MHz i.e. penetration depth of about 100 nm.Then, if we fit the TDTR signal with the heat conduction model using the configuration shown in Fig. 2(a) with ITR of Al /TIM as a parameter as shown in Fig. 3(a)-3(c), the value of ITR lie in the range between 0.4 and 0.6 mm 2 KW −1 .One observation is that these values are quite large for being the Kapitza resistance 23 i.e. intrinsic thermal resistance between two different materials that are in perfect contact.Another observation is that, as shown in Fig. 3, the signal obtained using TIM nearly collapses with that obtained when replacing TIM with pure Silicone oil despite that thermal conductivity of bulk TIM (Table I) is considerably larger than that of Silicone oil (0.15 Wm −1 K −1 ).It should be noted that the sensitivity analysis 25 (Fig. S1 (a-c) 26 ) suggests that the TDTR signal is sensitive to the value of thermal conductivity.This means that the apparent thermal conductivity of TIM in the region adjacent to the Al/TIM interface (i.e.within the thermal penetration length from the interface) is as low as that of Silicone oil.Therefore, it is likely that the thermally conductive fillers are not functioning in the region due to the presence of the filler depletion layer.Such filler-depletion layer (FDL) has been observed for a curing TIM 14,15 but the current result suggests that it can also occur for non-curing TIM even without phase transition.
Figure 3(a)-3(c) also shows that the TDTR signal obtained for pure Silicone oil agrees well with the heat conduction model with the bulk thermal conductivity of Silicone (0.15 Wm −1 K −1 ) without taking the Kapitza resistance between the Silicone oil and Al into account.On the other hand, the sensitivity analysis in Fig. S2 26 shows that, with the current setup, the sensitivity of the TDTR signal to the Kapitza resistance is too small to quantify the value unless it is larger than about 0.03 mm 2 KW −1 .These results suggest that the Kapitza resistance between FDL and Al is smaller than 0.03 mm 2 KW −1 , and thus, much smaller than the thermal resistance of FDL.Therefore, in the following we ignore the Kapitza resistance between FDL and Al.To gain insight into the thickness of FDL, we performed additional TDTR measurements with larger thermal penetration depths by decreasing the modulation frequency.Here, we adopted a heat conduction model shown in Fig. 2(b), where the TIM layer is divided into FDL with the same thermal conductivity as Silicone oil and the filler layer with the same thermal conductivity as bulk TIM.Thermal conductivity and volumetric heat capacity of FDL are set to 0.15 Wm −1 K −1 and 1.5×10 6 J/m 3 K respectively, 16 and the properties of TIM layer is set to the values shown in Table I.There is no thermal resistance between FDL and the TIM layer because the interface is merely an imaginary one.
The model allows us to evaluate the thickness of FDL by making the thickness a fitting parameter.The parameters other than the thickness of FDL is fixed to the value mentioned above.Decreasing the modulation frequency to 1 MHz increases the thermal penetration depth to 300 nm, and enhances the sensitivity of the TDTR signal to the thickness of FDL.For this thermal penetration depth, the sensitivity of the TDTR signal to the variable parameters are shown in Fig. 4 taking the case of G776 TIM.Figures 4(a), 4(b), and 4(c) correspond to when the FDL thickness is 100 nm, 200 nm, and 300 nm, respectively.These plots show that the FDL thickness has enough sensitivity when the thickness is 100 nm or 200 nm.
As shown in Fig. 4(d), the fitting curve based on the model well reproduces the TDTR signal.The obtained FDL thicknesses are listed in Table I.Each value in Table I is the average taken from 4 independent measurements.The FDL thickness for all the TIMs is around 100 nm with moderate variation depending on the kind of TIM.As seen in the Table, the variation does not correlate with viscosity nor thermal conductivity of the TIM.We speculate it might correlate with the filler density and size of the filler particles, however, because the density difference is small and the size distribution is large in the conventional TIMs, we are not able to investigate the aspect in this work.Such investigation with more controlled TIM samples remains to be our future task.
We now evaluate the impact of FDL layer to the total thermal resistance of TIM.Based on the above analysis, the thickness and thermal conductivity of FDL are 100 nm and 0.15 W m −1 K −1 , which result in thermal resistance of 1.33 mm 2 KW −1 .On the other hand, thermal conductivity of the TIM is set to 6 Wm −1 K −1 , which is the largest value among those of the conventional TIMs (Table I). Figure 5 shows the resulting total and partial thermal resistances as a function of the TIM thickness.When the TIM thickness is 50 µm (current technology), FDL occupies about 20% FIG. 5. Fractions of FDL and internal thermal resistances to the total thermal resistance as a function of the TIM thickness.
Reuse of AIP Publishing content is subject to the terms at: https://publishing.aip.org/authors/rights-and-permissions.IP: 157.82.14.101On: Wed, 02 Mar 2016 00:47:00 of the total TIM resistance.The fraction of FDL thermal resistance increases as the TIM thickness decreases, and becomes larger than the internal thermal resistance of TIM for thickness smaller than 10 µm.This means that FDL already holds significant impact to the TIM thermal resistance in current technology and will become dominant when the thickness decreases to 10 µm in near future.
In summary, we conducted direct measurement of thermal resistance of TIM/Al interface using three different conventional non-curing TIMs.The result reveals that the presence of filler-depletion layer (FDL) adjacent to the TIM/Al interface even in non-curing TIMs.FDL reduces the thermal conductivity of the layer to that of Silicone-oil resin.By adopting a heat conduction model accounting for FDL and expanding the thermal penetration depth in the TDTR measurements, we identified the FDL thickness to be around 100 nm.Finally by comparing the thermal resistance of FDL with the internal resistance of TIM, we showed that FDL occupies a significant fraction (∼20%) of the total thermal resistance for the nominal thickness (50 µm) in the state-of-art integration technology.Furthermore, when the thickness is reduced to 10 µm, which is likely be achieved in near future, the thermal resistance of FDL will dominate over the internal resistance of TIM.This implies the importance to prevent formation of FDL in the future thermal management of integrated devices.

015117- 2 Yada
FIG. 1.(a) A schematic diagram of the cross sections of the circuit board, chip, TIM, and heat sink.R interface denotes the interfacial thermal resistance (ITR).(b) Sketch of the dynamic thermal stress and deformation (not to scale) during the operation due to difference in coefficients of thermal expansion (CTE).

FIG. 2 .
FIG. 2. Diagrams of the sample structures (a) for measuring interfacial thermal resistance between TIM and Al and (b) for measuring thickness of the filler depletion layer (FDL) with thermal conductivity of Silicone-oil resin.

FIG. 4 .
FIG. 4. (a)-(c) Sensitivity of the TDTR signal to the variable parameters of model in Fig. 2(b) for G776 TIM.Figures (a), (b), and (c) correspond to when the FDL thickness is 100 nm, 200 nm, and 300 nm, respectively.The modulation frequency of TDTR is 1.11 MHz.(d) Normalized TDTR signal and best-fitted model profile using the system in Fig. 2(b).

TABLE I .
Properties of TIM and measured thickness of FDL.