Abstract
Increased demands on large scale server system packaging density have driven the need for new, more challenging electronic component cooling solutions. One such application required the development of a large form-factor printed circuit board assembly with multiple power transformer devices to be cooled via a common heat spreader. Thermally coupling the multiplicity of devices to the heat spreader was completed using a compliant thermal interface material. Given the mechanical tolerance range, the strain rate dependency of the interface material and the mechanical load limitations of the electronic devices, finite element analysis and empirical evaluation techniques were applied to ensure the anticipated interface gaps were established and that the initial and residual mechanical loading effects were understood. A characterization of the thermal interface material’s mechanical properties was completed for analysis input. Coupling this input with the geometric and stiffness properties of the assembly’s structural elements provided predictions of both the initial as well as the residual mechanical assembly loads. Once completed, experiments using pressure sensitive film and piezo-resistive film load cells were completed to correlate with the acquired analytical predictions.
Key Words: electronic assembly, thermal interface material, pressure sensitive film, piezo-resistive film
Introduction & General Application Description
As high-end computer electrical designs have evolved, so must their cooling solutions. To address this, advanced techniques must be developed to analyze and verify the performance of these cooling solutions, early in the computer’s development. As an example, a new high-end server was recently developed utilizing a large array of voltage transformer modules (VTMs) mounted to a common printed circuit board (PCB). Traditionally, voltage transformation had been previously performed within the power supplies remote to the source of load, but in this case, is performed closer to semiconductor chips, to increase end-to-end power delivery efficiency by reducing the physical distance of the power distribution network (i.e., path lengths of high current flow).
In the specific application, an array of 37 VTMs, each measuring approximately 22 mm x 33 mm in size was attached to a PCB that is approximately 380 mm x 760 mm in size. These VTMs dissipate power in use, and need to be sufficiently cooled to maintain proper functionality and reliability. The selected cooling solution employed a thin layer of thermal interface material (TIM) that individually thermally coupled the back of the VTMs to a common, global, heat spreader. The heat spreader was slightly larger than the PCB. The nominal TIM material design thickness was 1.5 mm, but because of assembly tolerances, can vary by approximately +/-0.5 mm. The TIM was carefully selected for its thermal and mechanical properties, which were required to be sufficiently stable to ensure cooling of the VTMs for the life of the product. By our definition, thermal reliability includes the material not breaking down, nor migrating out of the gap between the VTMs and the heat spreader during the life of the product. The compression and separation characteristics of the TIM were also carefully determined and used as the basis for the process placing the TIM between the VTMs and the heat spreader, and then compressing the TIM (in a controlled manner), until the heat spreader was assembled in its final position. In this manner, the TIM was compressed in a means that (1) assured proper VTM-TIM coverage and (2) prevented excessive (damaging) compressive loads imparted to the VTMs. Given the assembly’s complexity, it was anticipated that some assemblies would fail in-line manufacturing testing. In these cases, a cost effective rework solution was required. Once again, understanding the TIM characteristics was critical to ensure the selected TIM would support heat spreader removal without imparting high (damaging) tensile loads on the VTMs or the PCB.
Early in the design process, finite element (FE) analysis was performed to confirm the viability of the mechanical aspects of the cooling solution. This analysis evaluated the stresses during initial assembly, during the life of the assembly and during rework operations. The primary advantage of this FE analysis was its ability to identify any design concerns prior to initial hardware build. This FE analysis (as well as the TIM selection) relied heavily on experimental mechanical characterization of the TIM. Special fixtures were used to carefully measure the TIM response to compression and separation. These measured characteristics were then used to develop the equations used as inputs to the FE analysis. and reworked, without damaging any of the components. Besides post test destructive analysis, two in situ measurement during the assembly process. Inherently, PSF had the advantage of being readily applied at each module location, but had the disadvantage that it only showed the maximum pressure during a process. In addition, it also did not show when (time at occurrence) that maximum pressure was or if the pressure decayed with time. As such, after learning which module locations load ramped up (during the assembly process), and later decayed. This technique provided an enhanced understanding of the had to rely on the FE analysis.
Finite Element Model of the PCB Assembly:
The finite element analysis was setup to proceed in four steps as follows:
1. The heat spreader was brought down to the PCB, compressing the TIM at a specified strain rate
2. The screws securing the heat spreader to the PCB were preloaded to their specified torque level
3. Initial VTM compression forces are recorded (VTM TO compression loads)
4. TIM allowed to relax for 120 minutes, then compressive forces recorded (VTM residual loads)
Figure 1. (a) CAD Geometry of PCB Assembly (Heat Spreader not shown) (b) Assembly Finite Element Model
Figure 2. (a) Simplified TIM, VTM and PCB assembly (b) Detailed View of FEM components
TIM Material – FEM Inputs:
A thermally conductive, pre-cured, silicone gel was characterized and selected for use in this application. The thermal conductivity was measured to be 2.4 +/-0.1 W/mK (Figure 3), however, because of its high viscosity (> 40,000 Poise), wettability at interfaces was poor, adding approximately 300C-mm2/W to unit thermal resistance. For a nominal bond-line of 1.5 mm, this represented an approximate 5 percent increase in the interface’s resistance. Besides acceptable thermal performance, this material also required an inherently low compression mating force as well as administered low tensile separation strength. These were key requirements to protect the functional integrity of the VTMs during heat spreader mating and during rework (i.e., when it is necessary to remove the common heat spreader). TIM characterization was completed in two phases. The first phase focused on acquisition of time zero material properties and processing, while the second phase assessed the long term stability of the material after exposure to a variety environmental stress conditions.
Figure 3. TIM Thermal Conductivity vs. Applied Bond-line Thickness
Parallel plate rheometry was used to measure the storage modulus versus temperature. (Figure 4). The room temperature modulus of 70-80 Pa was used in finite element modeling to estimate stresses at critical locations. The storage modulus was also measured during a temperature ramp of 30C/min up to 1500C and it was determined that there was no significant decrease in flow to warrant using heat during the attachment of the heat spreader.
The mating force applied to the heat spreader was limited to 250 kg; determined by assuming a uniform loading of all 37 modules to 90% of the compressive limit. Figure 5 is a plot of the TIM bond line vs. time and mechanical loading. As shown, the ultimate bond lines were not achieved with 250 kg mating force and time alone, but only after engaging the screws and securing to the standoffs were stable bond lines achieved (i.e., 1.2 to 1.7 mm). Given the typical strain rate dependency of the TIM with respect to stiffness during compression, instantaneous loads can be very high when the screws were engaged. Therefore, in order to ensure that the individual compressive loads on individual VTMs remained within the force limits, in situ force measurements were made. This work defined the required screw fastening sequence and rate to keep the instantaneous loading on any single VTM within the safe region.
Figure 5. Bond Line Measurement During Assembly vs. Time
Next, tensile adhesion testing was conducted on bonded samples of single components and aluminum plates. The tensile stress to separate the aluminum plate was consistently less than 0.04 MPa, which is below the upper limit rating of 0.05 MPa for the component. In addition, two cells of adhesion samples were exposed to 1000 hours of 1250C and 675 hours of 500C/80% RH, respectively. Again, tensile stress to separate remained stable at less than 0.04 MPa, thus assuring that the heat spreader could be removed without damage to components even after long periods of time under operational conditions.
Given the above TIM stiffness properties, a piece-wise, multi-linear material model was used to approximate the modulus of the TIM. Force versus deflection data, for a variety of strain rates, was obtained from INSTRON® material tests on a sample that was of similar dimension to the actual application. A different material model was used as input to the FEM for each actuation strain rate case. To model the TIM’s force decay rate, seen in the INSTRON® tests, a pseudo-thermal contraction was enforced. This was accomplished by creating an artificial coefficient-of-thermal expansion for the TIM, which when combined with a small decrease in temperature, enforced only on the TIM, matched the Force vs. Time INSTRON® data (Figure 6).
Figure 6. Force Decay Rate of TIM vs. Time at Different Applied Rates of Strain
Finite Element Analysis Results:
The finite element model was subjected to a three separate total assembly times: 5 sec., 30 sec. and 120 seconds. The VTM loads at each site were recorded at T0 and at 120 minutes after assembly, to examine the initial and residual compressive loads on the VTM’s. Figure 7 shows the VTM site numbering scheme. Figure 8 and Figure 9 show a comparison of the VTM loads per site for the T0 and 120 minute assembly times, respectively. The results in Figure 8 show that slower actuation time reduces peak compression loading. The final actuation time was selected to provide the necessary VTM mechanical loading safety margin, while optimizing manufacturing throughput. In addition, examining the residual VTM loads showed that none of the sites resulted in a tensile force on the TIM material (a key indicator for interface reliability).
