HFE7500 Coolant Dielectric Strength Augmentation Under Convective Conditions

High-power electronic systems will be needed to drive the increasing electrification of mobility as well as the shift from hydraulic and pneumatic actuation to electric power. One way to enable more power is through the development of modules that operate at higher voltages, which has been made possible by the emergence of wide band gap (WBG) semiconductor devices with voltage ratings exceeding 1.2 kV. Additionally, these WBG devices exhibit high-temperature tolerance and low drain-source on resistance (RDSon) capabilities, enabling high amperage power modules that further increase power density; however, packaging and cooling techniques are struggling to keep up. To meet the demands of increased power density, new advancements in packaging materials [1], packaging architectures [2], and thermal management have emerged.

Novel cooling solutions that have been explored include those aimed at increasing the heat transfer area through microchannel flow passages [3], single phase immersion cooling in dielectric fluids to unlock higher densities, particularly in computer servers [4,5], novel fluid delivery methods via sprays [6] or jets [7] for enhanced heat transfer rates and cooling uniformity [8] as well as two-phase flow [9] to exploit the latent heat of boiling. A common theme for many of these efforts is the minimization of the convective thermal resistance (R_conv), which still leaves a large component of the overall thermal resistance, specifically junction-case resistance (R_jc), largely untouched. Minimizing R_jc, which is made up of the thermal resistances of all layers between the power device and heat sink, has been the subject of numerous research activities including those seeking to replace traditional ceramic-copper substrates with more conductive alternatives [10]. While these approaches deliver some improvement in thermal conductivity, the most effective way to minimize R_jc is to bring the coolant as close to the power devices as possible.

Integrated or direct cooling of power dense electronics has attracted significant research activity including efforts by Kwon et al. [11], Birbarah et al. [12] and Jung et al. [13]. While these direct cooling efforts have demonstrated the ability to improve the thermal management of power dense electronics, the research efforts have largely focused on low voltage applications. In high voltage packages, there are implementation challenges pertaining to the risk of phenomena such as partial discharge and breakdown, which complicate implementation.

High voltage systems face a tradeoff between electric field (E-field) mitigation and thermal management. On the one hand, thick or stacked ceramic substrates [14,15] with high dielectric properties are required between the power devices and heat sink to reduce the risk of partial discharge and breakdown due to the high E-field across the system. On the other hand, this isolates the power devices from the sink, thereby limiting the ability to extract heat. Existing research has considered solutions such as double-sided cooling or high-performance cooling schemes including jet impingement [14] onto the copper metallization of the thick DBC substrates. However, such solutions are only effective up to a point, after which further increases in the heat transfer coefficient have little to no effect, due to increased (R_jc). Direct cooling with dielectric fluids represents an opportunity to meet the cooling demands in such high-V power electronics. In a recent effort, researchers at the Army Research Laboratory demonstrated direct cooled packages, including a stacked diode module [16] that was demonstrated to operate at >20 kV while being directly cooled by forced convection of a dielectric fluid (DF) of the hydrofluoroether type. However, dielectric breakdown (BD) tests on the package showed a lower breakdown voltage (V_BD) for the cooled stack compared to the uncooled stacked module, which the authors attributed to failure through the fluid or the package. This result highlights the fact that while direct cooling with DF's is a viable approach for thermal management of high voltage packages, concerns over long-term performance of the coolants and their voltage isolation characteristics remain a barrier to adoption.

Direct forced convective cooling, in particular, exposes the fluid to changing conditions in terms of flow and temperature, which have been shown to significantly increase or decrease the voltage shielding capacity of DF's. Breakdown data from several reviewed studies [17–20] are plotted in Fig. 1 as a function of flow velocity, indicating a clear impact of enforced flow. To crystallize the effects of flow and facilitate comparison, the data from each study is plotted as V_BD/V_BD,0 where V_BD is the average breakdown voltage at a given flow velocity and V_BD,0 is the respective average breakdown voltage under static flow conditions. In these works, the augmentation of breakdown strength with flow is attributed to the sweeping action of flow on charge carriers and impurities, whereas the reverse impact of flow is attributed to, among other factors, the onset of turbulence that provides low pressure regions that act as nucleation sites for breakdown and triboelectric charging of the fluid, which results in local E-field enhancement. The observed differences in the data from the reviewed studies is due to several factors including varying flow conditions, flow geometries, electrode configurations and other factors all of which can influence breakdown strength. However, while this information is available for hydrocarbons of paraffinic and isoparaffinic type, of which mineral oils for transformer cooling applications are notable examples, there remains a lack of information on fluorinated compounds, including perfluorocarbons and hydrofluoroethers such as HFE7500. These fluids, which possess low viscosity over a broad temperature range, are well suited to electronics cooling applications when it comes to pumping power requirements.

In direct cooling applications where the fluid is subject to forced convection, a better understanding of the effects of flow conditions needs to be built to ensure safe implementation of these types of DF's in high voltage, high power electronics cooling. This effort seeks to advance the feasibility of direct cooling by assessing the voltage blocking characteristics of fluorinated dielectric fluid through relevant temperatures, flow rates, and electric fields.

The breakdown characterization efforts focused on HFE7500, a hydrofluoroether type DF developed by 3 M. HFE7500 is nontoxic and nonflammable, colorless and exhibits good thermal and chemical stability while being compatible with most common metals and hard plastics [21]. Additionally, HFE7500 is environmentally friendly with no-ozone depleting potential as well as little global warming potential. In Table 1, the thermophysical properties of HFE7500, as obtained from the manufacturer's data sheet, are presented [22]. Efforts were made to verify some of these properties using Differential Scanning Calorimetry for thermal properties and Rheometry for viscosity. As shown in Table 1 and corroborated by rheometry, HFE7500 exhibits low viscosity even at low temperatures, reducing pumping requirements and making it well suited to power electronics cooling.

A setup that enables DF testing through relevant temperatures, flow, and E-field was assembled. As shown in Figs. 2 and 3, the setup consists of a flow loop that features a pump ① (INTG7-060) for flowrate control and an inline heater ② for temperature regulation. The AC heater is rated for 120 V 400 W and is used in tandem with a PID controller to maintain fluid temperature at a desired value. Feedback to the PID controller is provided by a thermocouple ③ placed upstream of the test section. To apply a field to the fluid inside the test section ④, a Matsasuda Rb60 power supply   Ⓐ rated for 30 kV, 2 mA is used. In the current permutation of the apparatus, the voltage from the power supply is applied to a pair of point-point electrodes  Ⓑ , whose spacing is precisely controlled to vary E-field intensity. The point-point electrodes were chosen to minimize flow perturbation and capture the effects of E-field concentrations analogous to those at sharp corners, which represents worst-case E-field concentrations within a high voltage package [23]. Electrode spacing is varied relative to a reference point, where the electrodes are brought into electrical contact as verified with a multimeter's continuity test. The desired electrode spacing (d) can then be achieved by offsetting each electrode by a distance equal to d/2 relative to the reference point. For precise positioning, each electrode is attached to a micrometer actuator  Ⓔ , which is used to specify the offset to within a resolution of 10 μm.

Voltage application to the electrodes   Ⓑ   is controlled via a combination of a labview program Ⓒ and an NI USB-6001 DAQ system   Ⓓ  . The two are used in tandem to generate an analog signal (Vcon-in) to the RB60 supply  Ⓐ   . This external control voltage is varied from 0 to +5 Vdc, with the upper bound corresponding to a 30 kV output. During testing, Vcon-in is ramped such that the voltage applied to the electrodes increases at a rate of 100 Vdc/s. Throughout operation, the RB60 supply   Ⓐ outputs analog signals Vmoni and Imoni, which make it possible to monitor the supplied voltage as well as the leakage current flowing between the electrodes, respectively. The Imoni signal ranges from 0 to 5 Vdc, with the upper limit corresponding to a leakage current of 2 mA. To stop each test, a 1 mA leakage current limit is prescribed in the labview Ⓒ code.

Additional equipment such as a thermistor ⑤ (SEN-AP008B0), pressure sensor ⑥, flow sensor ⑦ (FTB2002-C) as well as an inline coolant filter ⑧ (INS-FLTR03) for keeping impurities from the test cell are also incorporated into the flow loop. The data from the sensors ⑤⑥⑦ is recorded separately using an Arduino Mega 2560 board. The test cell where the fluids are subjected to high E-field is additively manufactured using Peopoly Moai Hi-Temp Nex Resin that handles up to 180 °C. This makes it possible to drive the fluid to high temperatures to account for the growing push for elevated coolant inlet temperatures that reduce power and size demands on the heat exchanger/condenser side.

During testing, the breakdown voltage of the dielectric fluid was determined by monitoring the increase in the leakage current between the electrodes with increasing voltage. As indicated in Fig. 4, the breakdown voltage (V_BD) of the fluid is then identified as the voltage at which a sharp rise is observed in the measured leakage current due the development of a conductive path through the fluid. Based on this principle, the breakdown voltage of HFE7500 was tested under varying flow velocities and fluid temperature.

As the fluid passes through the test section, the flow regimes are not only controlled by the internal channel geometry (cross-field flow) but also by flow over the cylindrical electrodes. Since, as illustrated in Fig. 5, the flow field has both geometries potentially participating in the local fluidic phenomena, the Reynolds numbers associated with each flow type are computed to crystallize any impact on breakdown mechanisms. In the computation of these Reynolds numbers, ρ is the density of HFE7500, V is the flow velocity and μ is the dynamic viscosity. For flow over the electrodes, the characteristic length is the needle diameter (D) whereas for channel flow, this corresponds to the hydraulic diameter D_h, which is determined using the cross-sectional area A and the perimeter P of the channel.

To illustrate any dependency of breakdown strength on the flow regimes, the dielectric strength of HFE7500 is plotted as a function of Re_channel and Re_cyl in Fig. 6. For these tests, temperature and electrode gap were fixed at room temperature and 1 mm, respectively. Furthermore, each data point in the figure represents 10 breakdown measurements of HFE7500. Results indicated that cross-field flow can deteriorate the breakdown voltage, while higher flowrates in the range of Re_channel > 15,000 can result in an enhancement of the breakdown voltage. Figure 6 also shows that turbulent conditions overlap for cross-field flow and for flow over the electrodes. For channel flow, the onset of turbulence is typically taken to occur at 2300 < Re_channel< 4000. At the same time, for flow over the electrodes, vortex shedding, which is brought about by the detachment of the boundary layer as the fluid passes over the electrodes, starts at Re_cyl≈ 90 and transitions to turbulence at 150 < Re_cyl< 300 [24]. In this Re_cyl range, flow in the wake of the electrodes is characterized by random vortex formation and, similarly to channel flow, low pressure regions that can act as nucleation sites for breakdown. The fluctuation of breakdown strength observed within these transitional ranges for Re_cyl and Re_channel can be attributed to turbulent conditions, which influence local pressure and in turn bubble nucleation and breakdown. Temperature-dependent measurements (see Fig. 7) indicated a slight decrease in breakdown strength with increasing temperature for Re_channel< 17,000, while no discernible trend can be observed with Re_channel> 17,000.

where d denotes the gap length in mm and A and n are constants. Table 2 summarizes the values of A and n associated with the experimental data. Predictions made using these parameters are plotted in Fig. 8 along with 95% confidence level (CL) intervals for the fitted line. The model, which describes the experimental data well particularly at high Re_channel values, can be used to make decisions on the layout of direct cooled packages to minimize the risk of a discharge event.

The voltage blocking performance of dielectric liquids has drawn significant research attention, but their breakdown mechanisms remain to be fully understood, nonetheless. Broadly speaking, two predominant theories of pure liquid breakdown have emerged in literature: electronic breakdown theory and bubble/cavitation initiated breakdown [25]. The former involves the generation of free charge carriers at the liquid-electrode interface via field ionization or field emission. The continuous prebreakdown conduction currents observed in Fig. 4 are partly attributed to these mechanisms [26]. In the field ionization process, neutral molecules adjacent to a positive electrode (anode) are ionized by valence electrons tunneling from the liquid into the metal surface, and thus injecting positive ions in the fluid. This process is governed by the difference between the ionization energy of the liquid (I) and the work function of the electrode surface (φ), i.e., the energy required to free an electron from the surface of the electrode. Field emission on the other hand takes the reverse path with electrons being emitted from the tip or surface irregularities of a negative electrode (cathode), where the work functions (φ) are much lower compared to the rest of the electrode and local field enhancement by a factor of up to 300 is possible [27]. Consequently, at moderate applied voltages (2–5 kV), electrons at these preferred injection sites on the cathode surface gain enough energy to tunnel through the potential energy barrier and into the liquid in a phenomenon known as field emission [25,28]. Once injected, these electrons acquire energy from the field (E) and are accelerated toward the anode. The electronic breakdown theory postulates that some of the electrons gain sufficient energy to ionize liquid molecules on collision, creating positive ions and more electrons that initiate avalanche and eventually breakdown. The onset of avalanche is met when the energy acquired by an electron (eEλ) along its mean free path (λ) is equivalent to the ionization energy (chv) of the liquid molecule (where hv is the quantum of energy lost in the ionization process and c is a constant) [25].

Given the dense nature of the liquid phase, it is a point of contest that injected electrons can gain enough energy to trigger Townsend-type impact ionization and electron multiplication in the liquid phase [29,30]. This has led to the rise of an alternative breakdown theory centered around bubble-initiated discharge. The bubble or cavitation theory proposes that a gaseous or vapor phase of lower dielectric strength triggers total breakdown of the liquid medium. Bubble formation in pure liquid dielectrics may be a consequence of several processes [1] including gas pockets on electrode surfaces, cavitation due to turbulence, and gaseous products released through the dissociation of liquid molecules upon impact with electrons. Electronic processes can also initiate bubble formation. For example, the injection of electrons by field emission at a point cathode or surface asperities can induce local heating resulting in the formation of gaseous cavities [27,31]. Once formed, these bubbles represent low density regions where the electron mean free path is significantly greater than in the surrounding liquid medium. This allows electrons to acquire enough energy from the local field to trigger impact ionization and avalanche discharge within the bubble and release more energy. The vapor phase discharges result in conduction current fluctuations, similar to those observed before breakdown in Fig. 4, while also facilitating the propagation of the low-density front across the gap [27]. For cases where the point electrode is negative (cathode), these low-density regions develop into streamers [29] that propagate toward the anode, eventually causing total breakdown of the liquid.

Streamer initiation for a point anode is less understood and less reproducible; however, it propagates at higher velocities and leads to lower breakdown [32,33]. Given the positive DC polarity applied to the point electrodes in this study (see Fig. 3), point anode-initiated streamers are expected to play a role in breakdown. According to Ingebrigsten et al. [34], there exists a voltage-dependent critical volume around a point electrode where the field is strong enough to initiate avalanche within a fluid. When a seed electron is emitted from the cathode, it is injected at the location most favorable to initiate avalanche required to vaporize the fluid. For a point positive electrode, on the other hand, the likelihood that field ionization supplies a favorably placed seed electron is a stochastic event that leads to a distribution of avalanches in the vicinity of the point anode with varying capacity to vaporize the fluid. Consequently, unlike at the cathode, streamer inception voltages at a point anode are less reproducible and more affected by external factors such as hydrostatic pressure [35] and mechanical cavitation due to fluid motion [36]. This is observable in the flow-dependent data in Fig. 6 where there is a large spread in the breakdown of HFE7500 as function of enforced motion, especially for Re_channel ranging from 2000 to 6000 due to the onset of internal flow turbulence. Even at internal flow Re_channel below this range, vortices due to flow over the cylindrical electrodes can induce low pressure regions for streamer nucleation and reduced breakdown strength. In the absence of enforced fluid motion, mechanical cavitation can still ensue due to electrohydrodynamic (EHD) motion [31,37–39]. With EHD motion, the free charge generated by field emission or field ionization at the electrode interface is driven toward the oppositely charged electrode by electrophoretic forces (also called Coulomb forces). As the ions move, they collide and transfer some of their momentum to neutral liquid molecules, inducing bulk flow in the direction of the ion. The ratio of the induced liquid motion to ion drift velocity is always greater than one for liquids, highlighting the prominent role played by the induced motion in the transport of charge carriers (electroconvection) [25]. Furthermore, at strong enough fields, the induced flow becomes turbulent [40], which influences local pressure and in turn streamer inception and breakdown of the liquid medium. This phenomenon is a key driver of BD for cases in this study where the fluid is not subjected to forced convection. Imposing external forced motion on the fluid can however introduce a sweeping action on ionic charge carriers, which can disrupt the turbulence induced by EHD motion from fully developing in the interelectrode gap. Increasing temperature should also increase the propensity for the formation of the low-density regions for BD nucleation due to changes in fluid properties such as viscosity, density, and surface tension; however, within the tested range, increasing temperature only resulted in marginal reduction of BD strength for the laminar and transitional flow regimes.

In commercial dielectric liquids, the breakdown is also heavily influenced by the presence of impurities. These impurities can take the form of gaseous inclusions of lower breakdown strength and liquid contaminants such as water globules [41]. Commercial liquids also contain solid impurities which can be in the form of suspended solid particles or fibers. The suspended particle theory [25] of breakdown suggests that when a field is applied to a dielectric liquid of permittivity ε_DF containing a particle of permittivity ε_p, the said particle gets polarized and is subjected to a force exerted by the applied field. For ε_{p >} ε_DF, which is generally the case for solid particles and water globules, this force directs the particle toward the region of highest stress, i.e., the uniform field region in plate-plate or sphere-sphere electrode configurations subject to DC voltage. Once inside this uniform field region (dE/dx = 0), no force acts on the particle allowing it to reach an equilibrium state. An electric flux concentration then forms at the particle surface due its higher permittivity relative to the surrounding medium. The higher flux concentration attracts more particles, creating a particle chain that may eventually bridge the gap and initiate breakdown. In cases with too few particles to bridge the gap, local field enhancement due to their presence may induce microdischarges that can lead to early breakdown [25].

where R is the initial radius of the droplet of water, ε_DF is the permittivity of HFE7500 (i.e., ε_DF= ε₀ε_r where ε_r is the relative permittivity of HFE7500 and ε₀ is the permittivity of a vacuum), and σ is the surface tension of the water globule (0.043 N/m). Assuming a 1 μm initial water globule in HFE7500, this critical field inside the globule would be about 13 kV/mm, which is well within the range of values reached in this study, highlighting the significant effect that even microscopic quantities of water can have on breakdown.

Efforts were made to minimize the influence of particulates including the incorporation of a mesh-based filter upstream of the test chamber as well as the tracking of the sequence in which tests were performed, which provided a way of capturing the effects of impurity generation. These impurities could include dissolved water, gaseous products of liquid molecule dissociation and even particulates due to electrode erosion and other components of the flow loop such as the pump. The significance of particulate generation on the breakdown data from Fig. 6 was examined through a screen test in jmp software. The standard least squares was used to construct a model with breakdown as the response and model effects being Re_channel, same day test sequence and the interaction between the two. The Prob > F statics, which assess the significance level of the variables, are reported in Table 3. The Prob>F or p-value is the probability that the null hypothesis is true (i.e., the predictor does not contribute to the response). P-values closer to zero reject the hypothesis and thereby confirming the significance of the predictor. P-values < 0.01 are considered statically significant in this study. As shown in Table 3, Re_channel is significant in explaining the variance in breakdown strength and while test sequence might have an effect, it is not significant. Based on the Prob > F value for the interaction between the two variables, it can also be concluded that no interaction is detected.

To supplement the time-dependent analysis of the breakdown data, Ultraviolet/visible (UV/Vis) spectroscopy was used to assess the extent of particulate generation following repeated discharge events. This was done by analyzing the difference in the light absorbance of various fluid conditions/treatments including a pristine HFE7500 sample and a sample that was drawn from the flow loop reservoir after 95 breakdown events. A contaminated sample that showed significantly reduced breakdown strength was also tested to crystallize the change in absorption due to contaminants. The tests were performed using air as the background and covered a broad range of wavelengths to detect any changes. The results in Fig. 9 show negligible water content as indicated by the absence of a water peak, which would appear in the 1000 nm wavelength range. A closer examination of the peaks in the 280–320 wavelength range also reveals that the peak at 287 nm wavelength shifted for the fluid that has undergone repeated discharge events, which indicates degradation of the fluid itself. Furthermore, a peak that is not present in the pristine nor the 95BD sample is present at 292 nm for the contaminated sample indicating particulate uptake that is not apparent for the other two samples.

Considering R to be 10⁻⁹ m, the ionic drift velocity of HFE7500 is estimated to be 6.85 × 10⁻⁹ m²/Vs. For static flow conditions wherein average E-fields reached values as high as 16.8 kV/mm before breakdown, this corresponds to an ionic drift velocity of 0.12 m/s. On the other hand, at the highest Re_channel, the average flow velocity through the test section was as high as 1.9 m/s, lending support to the idea of the sweeping action of flow on space charges and impurities.

In Fig. 10, it can be observed that higher flowrates exhibit a slow build up in leakage current compared to static flow conditions. This suggests that due to the sweeping effect, there is a slow buildup of space charge hindering the formation of an intense local field at the point electrodes and resulting in a breakdown that is more systematic. This contrasts with the more arbitrary breakdown at static and low flow (Re_channel = 2957) conditions where impurities and cavity-initiated breakdown dominate.

As part of the objective to enable direct cooling for improved thermal management of high-voltage, high-power modules, efforts have been undertaken to assess the voltage blocking characteristics of hydrofluoroether type dielectric fluid. The characterization tests are performed over a range of relevant temperature and flow conditions. Results indicated that cross-field flow can deteriorate the breakdown voltage at low velocities, while higher flowrates in the range of Re_channel > 15,000 can result in an enhancement of the breakdown voltage. The observed drop in dielectric strength can be attributed to various mechanisms, primarily turbulence-induced cavitation that facilitates breakdown nucleation. While impurities can cause reduced breakdown strength, Re_channel > 15,000 can have a sweeping effect on these impurities as well as on ionic charge carriers, hindering the formation of an intense local field at the point electrodes and resulting in delayed inception of breakdown. The sweeping action of charge carriers and impurities is supported by I–V plots which show a slow buildup of prebreakdown conduction currents at Re_channel = 22175. This contrasts with the I-V curves for static and Re_channel = 2957, wherein breakdown occurs arbitrarily with no sustained rise in current, pointing to the dominant role played by impurities, flow-induced cavitation as well as rapid ionization and avalanche that generate fast-moving low-density streamers that trigger breakdown. As temperature increases, the ability for cavity-initiated breakdown also increases due to the changes in fluid properties such as viscosity, density, and surface tension. This is, to a certain extent, supported by the slight drop observed in breakdown strength with increasing temperature for Re_channel in the laminar and transitional regime, while, conversely, no effect is seen for the fully turbulent Re_channel.

Fluid motion and impurity content are seen to be significant drivers of the voltage blocking characteristics of HFE7500, and any implementation of direct cooling of high voltage power electronics would need to adopt careful filtering and impurity monitoring measures. Additionally, while enforced motion is shown to lower breakdown strength of HFE7500, using flow velocities of Re_channel > 15,000 can take advantage of the sweeping action of flow to restore breakdown strength.

The authors would like to thank the Center for Power Optimization of Electro-Thermal Systems (POETS) for their ongoing support. Any opinions, finding, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. The authors would also like to thank Michael Lynn for his help on the additive manufacturing needs of this study and Yalin Wang for his input in the assembly of the high voltage test setup.

• National Science Foundation (Award No. 2014-00555-04; Funder ID: 10.13039/100000001).

The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.