Air Cooling of Power Electronics Through Vertically Enhanced Manifold Microchannel Systems (VEMMS)

Semiconductor switches are essential parts of power conversion systems. They provide the means of creating alternating current flow into the transformer through switching at ever increasing frequencies. During the switching process, electrical losses occur in the form of heat dissipation. In the past decades, as power levels have increased and the switch size has decreased, a tremendous rise in heat fluxes has been observed. Thermal management of the switches has become a primary design goal in the packaging of power converters. Many aggressive cooling methods, such as jet cooling, spray cooling, and two-phase cooling, have been and are being developed which are capable of handling heat fluxes above 100 W/cm², with some capable of dissipating near or above 1 kW/cm² [1–5]. However, such cooling methods usually involve additional complexity and lack sufficient understanding of the fundamental physics of liquid–vapor flow interaction, making accurate and reliable analysis and design optimization more difficult [6]. Such complex systems may also be challenging to integrate with a typical power converter system. Air cooling, on the other hand, is considered to be reliable, since air is dielectric and benign, is widely utilized across diverse thermal management applications, and requires fewer cooling loop elements; thus it is an often simpler and more cost-effective solution when it can deliver the cooling requirements for the selected application [7]. The primary limitation of air cooling is its inherently less attractive thermophysical properties compared to some liquid-based working fluids. Because of this shortcoming, bulky heat sinks with large fans are often required. This increases the volume and mass of the cooling system, typically requiring comparable or sometimes a bigger space than the electronics themselves, leading to reduced power densities.

Despite its inherently poor thermophysical properties, the many advantages of air cooling have continued to draw the attention of researchers in the past decades. Research has been conducted to better understand and enhance the thermo-fluidic performance of air cooling [7–9]. Jet cooling has been extensively studied using air, as it has shown promising thermal performance enhancement over traditional forced convection of air with bulky heat sinks [10–12]. Recently additive manufacturing has also been utilized to enhance the air cooling performance through more optimum fluid flow distribution and enhanced heat transfer processes [13]. Specifically, on thermal management of power electronics, National Renewable Energy Laboratory has performed extensive numerical and experimental work on air cooling of power modules using high aspect ratio minichannel ducts [14]. Similarly, Wrzecionko et al. used high aspect ratio metal heat sinks for forced air cooling of actual power devices on a printed circuit board (PCB) that was directly attached on top of the heat sink, which showed good thermal performance even at elevated ambient temperatures up to 120 °C [15]. Chen et al. investigated air cooling of SiC switches using similar duct-like heat sinks with internal fin structures and used phase-change thermal grease to improve performance [16]. Wu et al. performed comprehensive design optimization studies using a genetic algorithm to design a nonconventional air cooled heat sink [17].

To improve performance over what has been reported in the literature, three major objectives are considered in this study: (1) reducing the number of layers and interfaces between the heat source and air; (2) improving the heat sink design to enable maximum heat transfer area while maintaining reasonable pressure drops; (3) creating a design with minimal footprint on board and reduced volume to enhance power density. Some of these targets, especially 2 and 3, are competing objectives. For example, larger heat sinks can provide more surface area but reduce power density due to increased volume. Smaller heat sinks with dense features, such as fins, create significant pressure drops that will choke or stall most fans. The novel vertically enhanced manifold microchannel system (VEMMS) design proposed in this study aims to find the sweet spot between these competing objectives. The detailed design and exhaustive computational fluid dynamics (CFD) analysis of the cooler was given by Yuruker et al. [18], who included two distinct designs, tailored for high and medium heat fluxes, liquid and air-cooled versions, respectively. The primary purpose of one VEMMS cooler is to provide adequate thermal management for a pair of switches on a PCB while also functioning as electrical connections to and between them.

Figure 1 illustrates the evolution of the VEMMS cooling concept from a small flat traditional heat sink to its final form. A flat heat sink with minimum footprint, shown in Fig. 1(a), enables other vital electrical components (such as gate drivers and passives) to be placed very close to the switches, however, the heat transfer surface area is not sufficient to provide adequate cooling. Figure 1(b) shows a laterally expanded flat heat sink, which can provide the necessary heat transfer performance but reduces the maximum achievable packaging density, as it pushes away the electrical components farther away. By bending this heat sink vertically, as shown in Fig. 1(c)), the necessary heat transfer surface area can be preserved while the footprint utilization on the board is minimized. In order for the design to perform adequately with air, the pressure drop of the heat sink must be low enough so that low power and small fans can provide sufficient air flow. This can be done by converting the straight microchannel concept shown in Fig. 1(c) into a quasi-manifold microchannel concept as shown in Fig. 1(d). This enables air to be distributed among the outlet holes on the sides of the cooler, which increases the heat transfer surface area while reducing the flow length of air traveling between the system inlet and outlet (ambient within the converter enclosure, see Fig. 2(a)), thus effectively reducing the pressure drop.

The finalized design of the air cooled VEMMS cooler is shown in Fig. 2(b)). Air enters the cooler from the top opening, flows through narrow vertical channels that are 0.5 mm in width, separated by 0.5-mm-thick, 2-mm-tall fins formed on the side walls, and finally exits through small, 0.5-mm-diameter with 1-mm-pitch outlet holes on the sides. These outlet holes also provide a large heat transfer surface. The side walls act as a heat spreader, conducting heat away from the heat generating switches. The cooler is made out of an electrically and thermally conductive material, such as silver or copper, and is directly bonded on the switches. This enables the cooler to serve two functions, as it also provides the means for the electrical connection between the two switches, as shown in Fig. 2(b). Although this work investigates performance in a single-sided-cooling configuration, the VEMMS approach can be utilized for double-sided cooling where the switches are sandwiched between two coolers and thermal resistance is nearly halved. For more details on the inception of the vertically enhanced manifold microchannel concept and extensive CFD analysis, readers are referred to another study by Yuruker et al. [18].

Section 2 of this study presents the successful manufacturing of the first prototypes, assembly of an experimental test loop, and thermofluidic performance characterization of the cooler.

The concept demonstrator of the air-cooled VEMMS is manufactured out of silver, using a process called additive manufacturing-assisted lost-wax casting. In this manufacturing method, a design is initially additively manufactured from sacrificial wax. This part is then submerged inside a ceramic bed, where molten metal is poured through a small inlet tube, melting the wax, replacing it, and taking the shape of the 3D printed part. Thus, it is considered to be a semi-additive manufacturing process. The part is then removed from the ceramic bed, the ceramic residue is broken and cleaned, and the surfaces are postprocessed to reduce surface roughness. A picture of the demonstration unit with its outer dimensions next to a one-cent coin is shown in Fig. 3. The prototype, including the air domain inside it, is 1.98 cm³ in volume, weighs ∼9.8 grams, and its dimensions are within ±0.1 mm tolerance. This cooling approach is designed to be ultimately used in a DC–DC converter with 10-kW power output. Thus, the reduced size of the VEMMS coolers of this study, along with the advanced transformer assembly introduced by Yuruker et al. [19], enables power densities up to 20 kW/L. A single VEMMS cooler will be used in thermal management of a pair of bare die SiC switches (see Fig. 2), operating at 1 MHz frequency at elevated voltages. Previous work was used to determine the heat loads for this case study, where a 10-kW dual active bridge DC–DC converter was studied and electrical analysis resulted in ∼25 W heat load per switch, adding up to a total of ∼200 W of losses for two full bridges containing eight switches [19,20]. Considering a bare die SiC switch from CREE company (Durham, NC) [21] with dimensions of 4 mm × 6 mm, the heat flux was computed to be within 80–105 W/cm². For most applications, this magnitude of heat flux is high enough to require significant heat spreading, usually via heat pipes and large heat sinks, or more advanced methods such as liquid cooling [22–28]. However, the air cooled VEMMS is, by design, capable of handling such heat fluxes, ensuring that the switches operate below their maximum recommended temperature of ∼200 °C [21].

To assess the performance of the cooler experimentally, a test loop is shown in Fig. 4(a) was built. Air supply to the system was provided by a valve-controlled tap that contains pressurized air supplied by the facility. The valve is connected to a rotameter for flow rate measurements. After the rotameter, air reaches the VEMMS cooler through a custom designed and 3D printed plastic adaptor as shown in Fig. 5. The 3D printed fixture has monolithic pressure ports on its sides for pressure drop measurements near the inlet region of the cooler. An electronic pressure transducer, in parallel with mechanical pressure gages, was installed to record the pressure drop from the inlet of the cooler to the ambient.

Initially, adiabatic tests were performed to characterize the VEMMS cooler's flow rate versus pressure drop behavior, as shown in Fig. 6. The tests were repeated twice to prove the repeatability of the results. Furthermore, pressure drop was simultaneously measured during diabatic tests and plotted on the same graph in Fig. 6. A good agreement was obtained between previously conducted CFD predictions by Yuruker et al. [18] and the experimental findings of this work, where maximum deviation was less than 13% and mean deviation was 5%, which fall within the experimental uncertainty, validating the CFD predictions of our previous work [18].

To mimic heat loads dissipated at a pair of high-frequency SiC switches, two similarly sized AlN-based thin-film resistor heaters (see Fig. 5, top left corner) were utilized [29]. The thickness of the thermal grease used as thermal interface material (TIM) was expected to be between 100–200 μm, and its thermal conductivity is 6 W/m-K [30]. The power input to the heaters was recorded using two multimeters, one for current and one for the voltage drop across the two heaters connected in parallel. The temperature of the chips was recorded using T-type thermocouples placed on top of the AlN heaters and held in place with insulating foam, as shown in Figs. 4(b) and 5. This ensures an adiabatic boundary condition on the top side, and thus temperature differential between the thermocouples and the heaters was negligible.

where V is the voltage and I is the electrical current applied to the heater and precisely measured via two multimeters using the four-wire method.

Throughout the experiments, the power of the two heaters was kept constant at 12.36±0.1 W; therefore, the temperature of the heat source was the highest at the lowest flow rate, and it gradually decreased as the flow rate increased. To verify the assumption that heat losses to the environment were negligible, a simple natural convection and radiation calculation was performed. With a highly conservative assumption of emissivity of 1 for the polished silver surface, an isothermal cooler outer surface temperature equal to heater temperature, and a natural convection heat transfer coefficient of 5 W/m²K, the total heat losses were calculated to be less than 0.51 W even for the lowest flow rate condition (0.5 g/s), where the temperature difference between the cooler and ambient was the highest (∼60 K). The 0.51-W loss corresponds to ∼4% of the total heat input to the AlN resistor heaters, which was precisely measured using the four-wire method. From 1 g/s on, the conservatively calculated loss value became <2.5%; for the highest flow rate, the loss was only 1.3% of the total power input to the heaters. Since the actual emissivity of polished silver is expected to be lower than 1, the actual loss percentages are even smaller. This analysis result validates the assumption that losses through radiation and natural convection are negligible.

where $δf$ is the error in the metric function of interest (thermal resistance in this case) and $φ1,φ2,.$ are the measured independent variables of this function, with $δφ1$ being the individual uncertainties of each measured variable, given in detail in Table 1.

When plotted against mass flow rate, the effective thermal resistance of the cooler is seen to decrease with increased flow rate, as expected. The curve follows an asymptotic decay, where the caloric and convective terms in the effective thermal resistance given in Eq. (1) reduce as a function of the flow rate. The caloric resistance R_cal is defined as the difference between the mean temperature of the fluid in the channel and the inlet temperature of the fluid. It is therefore inversely proportional to the product of the mass flow rate and specific heat of the fluid [31–33], since the amount of temperature rise between the inlet and outlet are largely dictated by mass flow rate and the specific heat of the fluid. Therefore, in the limit where mass flow rate approaches infinity, the caloric resistance also approaches zero, as the temperatures of the fluid at the inlet and outlet become equal in the limit. This is the main reason for the decay in R_eff with the mass flow rate.

Similarly, the convective resistance R_conv, which is inversely related to the heat transfer coefficient and surface area, also reduces with the increased flow rate. Increased fluid velocity increases Reynolds number, and thus the Nusselt number, which enhances the heat transfer coefficient, leading to a reduction in R_conv. Yet, it is important to note that increased mass flow reduces only the caloric and convective terms in the thermal resistance network, but the R_eff curve shown in Fig. 8 approaches to a nonzero asymptotic value. This is because conductive resistance at the base and side walls of the cooler is not a function of flow rate, but rather a function of geometry and the properties of the solid material.

Consequently from Fig. 8, it is determined through extrapolation that the asymptotic value when flow rate approaches infinity using 100 microns of solder (k = 60W/m-K), is ∼0.8 K/W. Further reduction of this value is possible yet would be quite difficult, as the selected solid material, i.e., silver already has a very high thermal conductivity (∼405 W/m-K).

After both the pressure drop and thermal resistance curves of the cooler were characterized, the cooler was subjected to the air flow provided by a commercially available low-power miniature fan [34]. In order to achieve an adequate heat removal rate, sufficient flow must be supplied by the fan, which is dictated by the pressure drop across the inlet and outlet of the cooler. The existing test setup was revised as shown in Fig. 9. An adaptor was additively manufactured to direct flow from the fan to the cooler. There are two pressure ports on the adaptor, one close to the exit of the fan, and one at the inlet of the cooler. The pressure difference between the atmosphere and pressure port closer to the fan create the pressure drop across the fan.

As was discussed in Yuruker et al. [18], the primary advantage of the air-cooled version of VEMMS is its reduced pressure drop versus flow rate characteristics, as a means to increase the flow rate from the fan. However, even though the selected fan is capable of providing 800 Pa static pressure, the pressure drop through the rotameter at the required flow rates would be too high, preventing us from mimicking realistic, real-life operational conditions and reducing the achievable flow rates. Thus, we could not directly measure air flow rate. However, the airflow rate can be inferred by measuring the pressure drop across the VEMMS cooler and finding the corresponding flow rate using the previous experimental pressure drop characterization curve shown in Fig. 6.

Thermal resistance results for fan tests are shown in Fig. 10. Through plugging the measured temperatures of the heater and inlet fluid, as well as the electrical input to the heaters into Eqs. (1)–(3), thermal resistances corresponding to the fan flow rates are obtained. Once these thermal resistance values are graphed at their predicted flow rates, they land very close to the CFD results with 150-micron TIM. This shows the internal consistency of the fan tests and previous characterization experiments shown in Fig. 8 and validates the prediction of mass flow rates during fan tests.

For R_th = 2.6K/W and Q_sw = 25W, T_rise is calculated to be 130 °C. This means that through the utilization of air cooled VEMMS heat sinks, switches can operate safely at up to 70 °C ambient air temperature. The ultimate goal is to utilize double-sided cooling with solder instead of TIM, where the other side of the board is also populated with VEMMS coolers, and to have the heat dissipated by the switches flow in both directions. In such a scenario, by ensuring good conductance through the board, one can assume that equivalent thermal resistance will be half that of what single-sided cooling can achieve. This leads to a thermal resistance of ∼1 K/W, which enables inlet temperatures up to 150 °C for 50 W total heat load (25 W loss per switch). Alternatively, if inlet temperature is 25 °C, switches are allowed to dissipate up to 170 W total (87.5 W each), which corresponds to ∼ 360 W/cm² maximum allowable heat flux.

where R_th is the thermal resistance achieved with a pressure drop constraint in place, which must have either been achieved experimentally using a real fan/pump, or demonstrated otherwise that the assumed constraint and the corresponding thermal resistance are realistically achievable. Otherwise, a thermal resistance value selected at a random flow rate in the absence of a realistic pressure drop constraint would render the calculated conductance density impractical and arbitrary. V is the total volume occupied by the heat sink in liters. The conductance density of the air-cooled VEMMS heat sink using a miniature low-power fan at 12 V [34] is calculated to be 194 W/K-L or 266 W/K-L, depending on whether the bonding is done using TIM or solder, respectively.

Table 2 compares the performance of the air cooled VEMMS approach over some other state-of-the-art cooling systems for power converters. Thermal resistances and the corresponding heat sink volumes (calculated for a pair of switches, if more than two are present) are given in the table. It can be seen that the conductance densities of air cooled VEMMS are orders of magnitude better than other systems, comparable to some of the liquid cooling solutions.

where P_converter is the total electrical power converted by the converter, and V_fb is the total volume of the full bridge assembly, which includes four SiC switches. For a full bridge shown in Fig. 11, which utilizes double-sided cooling with four VEMMS coolers (two on each side of the board, directly bonded on switches, and are 10 mm apart from each other) and two fans, the total volume of the bounding box of the full bridge assembly is 188.4 cm³, which leads to an electrical power density of 84.5 kW_e/L.

A multifunctioning, novel air cooled heat sink for thermal management of power electronics were introduced. The proposed vertically enhanced manifold microchannel system (VEMMS) cooler was successfully manufactured, using a combination of conventional and additive manufacturing techniques. The system's thermofluidic performance was experimentally investigated. Good agreement was obtained between the experimental findings reported here and the numerical analysis results in Yuruker et al. [18]. A thermal resistance of as low as 2.6 K/W, including the contribution from TIM layer, was demonstrated using a <1.5 cm² heat sink footprint and <2 cm³ volume, which corresponds to a 194 W/K-L conductance density. The thermal resistance can be further reduced if double-sided cooling with solder instead of TIM is utilized, to a projected value as low as ∼1 K/W. Once integrated into a 10-kW_e DC–DC power converter, a 20 kW_e/L overall converter power density and 84.5 kW_e/L full bridge power density can be achieved, demonstrating the reduced size advantage of VEMMS cooling.

Future work should include the integration of the proposed thermal system into a power converter in which VEMMS heat sinks are used in the thermal management of SiC power switches operating at high voltage and frequency.

The authors thank the U.S. Army Research Laboratory (ARL) for funding this work as part of the Silicon Carbide Advanced Packaging of Power Semiconductors-II (SCAPOPS-II) program, under contract number W911NF-18-2-0118. The authors extend their appreciation to Dr. Lauren Boteler, Dr. Miguel Hinojosa, and Dr. Michael Fish from ARL for their technical contributions throughout the course of this project.

• U.S. Army Research Laboratory (ARL) (Contract No. W911NF-18-2-0118; Funder ID: 10.13039/100006754).

COD =

conductance density (K/W-L)

Cp =

specific heat (kJ/kg-K)

DC =

direct current

k =

thermal conductivity (W/m-K)

l =

length (m)

$m˙$ =

mass flow rate (g/s)

PCB =

printed circuit board

PD =

electrical power density (kW_e/L)

Q =

heat (W)

R =

thermal resistance (K/W)

T =

temperature (°C)

TIM =

thermal interface material

VEMMS =

vertically enhanced manifold microchannel system

Subscripts

Ag =

silver

amb =

ambient

cal =

caloric

Cond =

conductive

conv =

convection

e =

electrical

eff =

effective

fb =

full bridge

sw =

switch

TIM =

thermal interface material