After inspecting the internal morphology of the thermal interface, we turn to fully assembled TIM structures. A full TIM is usually composed of two sides thermally connected with composite materials, in this case nanotubes. As nanotubes are grown on a silicon substrate, it serves as one heat bath of the TIM. On the other side, as we saw, a polymer layer is covering the top of the nanotubes for increasing conductance and usually capped with an extra layer for a second heat bath. As we already described, the nanotubes loose their vertical alignment and they form big clusters during this fabrication process. The impact of the manufacturing on the nanothermal properties is a key interest for the TIM efficiency improvement.
In this last section, we attempt to explore these nanothermal properties using three different approaches. As 3D probing and access to buried interfaces are required, all approaches involve a sample preparation step that will be described. First, we used a similar polishing tool as for BEXP but in this case, the polishing angle was set at∼90◦ to obtain a perpendicular cross-section. The second sample
third TIM was polished on one of its edges using FIB. Each sample fabrication method offer advantages and disadvantages that we describe. SThM measurements on each sample were not fully conclusive being affected by the tip-sample contact area. However, this work provides a good test bed for future TIM characterisation. Perpendicular BEXP
The sample used for perpendicular BEXP was as described above with a copper heat sink over the polymer layer. Fig. 5.11 represents the sample and how it is mounted for SThM characterisation. Note that as there was more material to be polished than in traditional BEXP, the process took around three times longer (∼ 15 hours). However, the resulting surfaces were smooth compared to other
methods described later.
Figure 5.11: Perpendicular BEXP schematic, not to scale. The sample is
mounted on its edge and the polished side is accessible with the SThM probe. We observed a dip between the polymer layer and the copper that could have the following origins (see Fig. 5.12). First, from the TIM fabrication, it is possible that the copper and polymer layer were not fully in contact, especially on the sample where we performed the polishing. This is supported by the depth of this gap which was measured to be at least 400 nm with AFM whereas the space between the polymer and the copper was around 2µm large.
The other explanation could be that the perpendicular cross-sectioning impacts the interface between the polymer and the copper. As Ar ions impinges the sample from the silicon side, the polymer is removed first, followed by the copper. The
polymer is expected to be easily polished by these nearly 90◦ incident angle ions
rays while the copper, formed of heavy atoms will be less affected. Therefore, the ions rays, after having removed the polymer materials strike the heavy copper atoms and could then deflect, or even heat the copper, and affect the polymer below. This way, a dip could be formed at the copper-polymer interface.
Figure 5.12: SEM image of the full TIM as obtained from perpendicular nano-
cross-sectioning. The scale bar is 2µm. The dip between the polymer and copper
is observed.
We performed SThM measurements in ambient environment (see Fig. 5.13). The top of the SThM scan shows a small region of the silicon substrate and at the bottom we observe the polymer layer. The former appears more thermally conductive (darker) than the latter, as expected from the bulk thermal conduc- tivity of these materials. Aside from the disorder already observed in the CNT layer, various SThM signals are received. It is hard to distinguish the components of this signal as topography artefacts are likely to dominate due to the highly inhomogeneous surface of the CNTs.
Figure 5.13: Topography (left) and SThM (right) maps of the BEXP perpendic-
ularly polished sample.
TIM sample section via TEM preparation procedure
In this second sample, a TEM sample preparation method was used. A small piece of the material is isolated using precise manipulation systems. Then it is encased in a platinum frame and attached to a grid/holder as shown on Fig. 5.14. Ambient SThM setup was also used here for thermal characterisation. We remark that this very fragile size of the sample, surrounded by voids, was creating many challenges for performing experiments.
Figure 5.14: Optical microscope image of the TIM sample prepared for TEM
imaging.
Results are displayed on Fig. 5.15. All layers appear clearly on the SThM signal. We observe a general decrease in the signal from a high silicon response to a low copper one. We attributed this observation to the different air volume surrounding the SThM probe at each position. When the probe is in contact with the silicon, on the outer part of the TEM section, the effective air volume connecting the probe to the sample is smaller than when the probe contacts the copper where the whole probe body and sensor are close to the TEM section holder. Therefore more heat is transferred to the sample when scanning the copper and polymer whose response appear more conductive than the silicon, which is highly unlikely.
Figure 5.15: Topography (left) and SThM (right) maps and profile of the TEM-
like prepared TIM sample.
Studies of FIB sectioned TIM sample
The last sample examined for nanothermal properties of full thermal interface materials was prepared by FIB polishing in a small region of its edge. To this end, the traditional copper heat sink was replaced by a thin gold layer covering the polymer (see Fig. 5.16). To fully understand its thermal behaviour, a new
experiment was designed. Vacuum SThM measurements are performed in the FIB sectioned area while a temperature gradient is applied between both sides (Fig. 5.16).
Figure 5.16: (a) Schematic of the FIB section sample experiment. While applying
a temperature gradient on both sides, the temperature distribution is measured along the TIM. (b) SEM image of the FIB section.
To realise this temperature gradient, a new setup was built within the vacuum chamber as shown on Fig. 5.17. The sample is mounted on a L-shape copper support glued on a Peltier plate. This enables to keep the silicon heat sink side at a constant temperature (∼ 300 K). Then, to create a temperature gradient,
a small 22 Ohm resistor was used as a heat source. This resistor was thermally connected to the sample gold layer side via a gold coated copper foil. To ensure constant mechanical contact, small magnets where used on both sides of the sam- ples, therefore maintaining mechanical contact between the foil and the sample. Then, we could approach the FIB section with the SThM (Fig. 5.17). Note that, similarly to the TEM like sample, the area of interest was small and challenging to reach with the SThM system, due to poor optical positioning setup.
Figure 5.17: TIM temperature distribution setup and voltage bias applied on
the heating resistor.
The temperature gradient was then created while applying a square voltage on the resistor (Fig. 5.18). The absolute value of the temperature rise could not be estimated. A large sample heating cycle (75s) was chosen in order to clearly observe
the heating effect on the image. Results are displayed on Fig. 5.18. We note first that the FIB section leaves very rough surfaces even for the silicon material. On the top of the image, the probe is mechanical contact with the sample. We can observe the different layers in the TIM. The slow square heating is visible as creating an oscillation in the SThM signal. However, when the mechanical contact is removed, a residual variation is also observed in the SThM signal. We attributed this to heat transfer from the resistor to the probe base through the whole microscope.
Then, to distinguish the signal variation coming from the probe and the one coming from the tip, we need to subtract the non contact oscillation from the in contact signal. The resulting image is also shown on Fig. 5.18. As we observe, after subtraction, the oscillation largely disappears, meaning that most of the heating related signal variations are linked to the probe base heating.
Figure 5.18: Topography (top left) and SThM (top right) temperature maps
of the FIB sectioned sample when the probe is in- and out-of-contact with the sample. Schematic applied voltage on the resistor to create a temperature gradient (bottem left). Resulting SThM map (bottom right) obtained by substracting the non-contact signal from the in-contact one.