Other designs using single, large area transistors or power modules often need to contend with local hot spots and rely on extrinsic heat spreaders to help mitigate this issues. However, FCML implementations leveraging smaller devices – though having a higher device count – benefit from an intrinsic heat spreading effect[80]. However, removing the heat from each transistor is still required for operation at high power. Previously, FCML inverters employing chip-scale GaN transistors used a compressible thermal interface material (TIM) to mate the devices to a solid heat sink or cold plate [66–68, 143], or otherwise required the heat sink be machined to precise tolerance [118]. Thus, a modular design approach, where heat sinks are attached via spring compression to each chip-scale GaN device, presented an appealing opportunity.
A modular heat sink design for FCML or any other converters requiring the individual heat sinking of multiple, discrete power transistors was originally proposed in [80, 144]. As
Surface-Mount Nut Inlet Baffle
Compression Plastic Screw Spring
Flow Flow
(a) Mounting strategy of [80].
Enclosure Fastener Compression Metal Screw Spring
Flow Flow
Inlet Baffle
(b) Proposed mounting strategy.
Figure 3.1: Comparison of previous and proposed modular heat sink mounting strategies. Note, the threaded insert in the right figure is only conceptual; the interference fit described later in this section made this specific consideration obsolete.
such, the approach will only be summarized here with the remainder of the section reserved for advances in the design, manufacturing and performance. Essentially, this approach fo-cuses on cooling each power device individually rather than attempting to provide cooling to the entire converter with a single heat sink. This can provide an opportunity to minimize heat sink mass, reduce manufacturing tolerances (especially the need to exactly match vari-ations in component height), and allow for the use of a higher performance TIM that might otherwise be impossible with a single heat sink. Additionally, just as a modular approach is applied at the system level to mitigate design complexity and support scalability, modularity in heat sink design can also be leveraged for similar benefits.
3.2.1 Single Switching Cell
The original conception of the modular heat sink intended the heat sinks to mount to the converter PCB through screws threaded into surface-mount nuts [80]. As this scheme con-tinued to evolve, incremental advances in pad design and registration features were included to properly align the surface mount nuts during reflow in [1]. However, thermal performance continued to fall short of the expected improvements, and heat sink installation proved to be tedious and yielded inconsistent quality. Major issues arose from nut misalignment (in spite of improvements to the pad design), the use of plastic (electrically-insulating) screws that were prone to strip and cross-thread, and the narrow spring selection available for the form factor of the design. This last issue was driven by three competing constraints: a short enough free length to allow the screw to seat in the corresponding hole in the heat sink; or few enough turns so as not to go “solid” (completely compressed, i.e., rigid and no longer applying compressive force) when the screw was fully threaded; and a suitable spring constant to apply the correct contact pressure. Assembly was further complicated by the
Compression
Spring Alignment Peg Snap-Fit
(a) Cross-section render.
Spring Alignment Peg Snap-Fit
(b) Installation photo.
Figure 3.2: Examples of the snap-fit concept with spring alignment and interference features.
ongoing need to apply careful alignment and threading to two screws per heat sink for each device on the PCB – up to 72 manual insertions per converter for a dual-interleaved, 10-level module.
As such, this chapter addresses both the manufacturability and overall performance is-sues of the previous work. The first major update changed the plane by which the spring was compressed from the PCB surface (by way of threading the screw into the surface-mount nut) to the surface of heat sink ducting that was already present in the design. This update is illustrated in Fig. 3.1, where Fig. 3.1a illustrates the previous approach and Fig. 3.1b illus-trates the approach developed in this work. By doing so, the range of acceptable spring free lengths was expanded, increasing the quantity of applicable, commercially available springs, listed in Table 3.1. Here, beyond mechanical fit, the primary characteristic was whether the spring, when compressed, would exhibit a contact pressure corresponding to the test conditions of the TC-5022 TIM that would be used [145]. TC-5022 was identified to provide superior thermal performance to other, compressible TIMs used in prior work. Additional considerations included spring cost, the extent to which the difference between free length and compressed length could absorb manufacturing tolerances, and whether the ends of the spring were milled flat to ensure a flush mate with the compressing surfaces. The highlighted row in Table 3.1 is the part number that best fit this criteria.
The second major change was to retain heat sinks (via captive attachment) in the con-verter enclosure housing by means of an interference, or snap-fit. This design is illustrated in the mechanical model of Fig. 3.2a, where registration pegs align each spring to the com-pression plane of the enclosure and interference features retain the heat sink when springs are uncompressed. Fig. 3.2b shows one heat sink in mid-assembly, where springs are being aligned with registration pegs on the enclosure, with the other heat sink retained with the deformable interference features.
Table 3.1: Selection of appropriately sized miniature springs from Acxess Spring.
Length [mm] Rate Force
Part Number Free Compr. Delta [N/mm] [N]
PC356-3175- 6000-MW-6756-CG-N-MM 6.76 5.70 1.06 1.77 3.73 PC356-3048- 6130-MW-6350-CG-N-MM 6.35 5.43 0.92 1.97 3.61 PC305-2388- 9000-MW-6350-C -N-MM 6.35 5.43 0.92 1.35 2.48 PC356-2388- 7000-MW-5994-CG-N-MM 5.99 5.43 0.56 3.78 4.24 PC406-2896-10300-MW-5944-CG-N-MM 5.99 5.43 0.56 2.11 2.37
3.2.2 Complete Converter
The design modifications described above drastically improved converter manufacturability of the modular heat sink design. In this way, when the converter is leveraging dual-sided cooling, an entire half of the converter could be assembled at once by first installing heat sinks on a single enclosure side (shown in Fig. 3.3a), then mounting the enclosure to the PCB.
Fig. 3.3b shows how this process further streamlined TIM application where, when using a stencil, application could proceed quickly from heat sink to heat sink. Note, the polyimide stickers visible in Fig. 3.3b are part of a pre-assembly process to electrically insulate heat sinks from high-voltage nodes present at the flying capacitor terminals.
Finally, an illustration of the proposed air-cooling configuration for the full converter module is shown in Fig. 3.4. Here forced-air is provided at inlets on either side of the converter. This cooling air runs across heat sinks on the outside of both the front and back of the PCB, then exhausts from the middle of the converter. While a single flow path from left to right or vice versa (i.e., across both sets of heat sinks on a single converter face and in
(a) Fully installed heat sinks. (b) Stenciling of TIM.
Figure 3.3: Installation of heat sinks on an entire face of a module with dual-sided cooling.
Exhaust
Exhaust
Airflow Inlet AirflowInlet
Figure 3.4: Illustration of dual-inlet cooling of an FCML-based power module. One inlet is provided for each interleaved phase, with exhaust air designed to exit out the middle of each module.
one direction) would potentially provide a more ideal flow path, such a configuration would cause one phase to receive warmer cooling air and thus run hotter than the other. The implications of such an imbalance in cooling are discussed further in Section 4.3, while the consequences of the proposed cooling configuration are described in the following section.