Posted on 30 January 2019

Water Cooling of Power Modules

Test cooling plate with various inserts for direct base plate cooling








Water cooling in power modules can be used for very high power inverters (MW range) as well as for low-power devices which already have a water cycle for operating reasons (e.g. car drives, galvanic installations, inductive heating). In most cases, the admission temperature of the coolant values is as much as 50...70°C when the heat of the coolant is directly dissipated to the atmosphere; in industrial plants with active heat exchangers, the temperature is about 15...25°C. The temperature difference between heatsink surface and coolant, which is lower than for air cooling, may be utilized in two ways:

  • increased power density, but with high dynamic ΔTj of chip temperature per load cycle, or
  • low chip temperature, long module life.

Figure 1 shows an example of water cooling of a 6-fold SKiiP on a water-cooled heatsink.

layout with water-cooled heatsink

Figure 1. Layout with water-cooled heatsink

The following factors influence the thermal resistance in a liquid cooler:

  • the contact area to the coolant (e.g. number of cooling channels)
  • the volumetric flow rate as a function of the pressure drop
  • the heat storage capability of the coolant
  • turbulence in the water flow
  • heat conduction and spreading in the heatsink (heatsink material)
  • the coolant temperature (depending on viscosity and density)

Enlarging the contact area of heatsink/coolant will result in improved heat transfer. The traditional cooler is subject to limitations with regard to the number of cooling channels. The pin-fin cooler features small columns protruding into the coolant, which enlarges the contact area and also ensures sufficient turbulence.

Pin-fi n-liquid cooler to enlarge the heat transfer area; a) Schematic drawing; b) Photograph of the cooling side

Figure 2. Pin-fin liquid cooler to enlarge the heat transfer area; a) Schematic drawing; b) Photograph of the cooling side

The particular shape of the liquid cooler and a sufficiently high flow velocity creates a turbulent flow which substantially reduces the heat transfer resistance between heatsink and liquid (also see the spiral shaped inserts in Figure 6). Without turbulence, a liquid film is created on the cooler surface which impairs heat transfer.

Even more so than with air coolers, an even distribution of heat sources across the heatsink surface is important for low thermal resistance. Due to the high heat transfer coefficient of some 1000 W/(m2K), the heat flow is dissipated to the cooling liquid with only minor cross-conduction. This means that essentially only those areas on which power semiconductor modules are mounted are used for cooling. Copper rather than aluminium as heatsink material will reduce the volume resistance, increase cross-conduction, thus also increasing the effective cooling area. A cooler made of copper allows for a reduction in Rth(j-a) by approximately 20% for a standard IGBT module.

Especially in water-glycol mixtures, Rth(s-a) depends on the coolant temperature. This is due the glycol viscosity, as well as the changing density of the coolant, albeit to a lesser extent. For a mixture of 50% glycol and 50% water in the temperature range of 10°C to 70°C, it was found that Rth(r-a) was reduced by about 25% between the temperature sensor and coolant.

Dependency of the heatsink resistance on the coolant inlet temperature

Figure 3. Dependency of the heatsink resistance on the coolant inlet temperature

Pressure drop and water volume, test pressure

In a closed cycle, liquid flow from the heat source and back can be generated by gravity (the heated liquid has a lower density, thus rising upwards to the heat exchanger; the cooled water sinks down to the heat source again - thermosyphon cooling). In most cases, however, a pump is used to circulate the liquid. This means that the available pumping power can be used to set the required water volume. Increasing water volume causes the thermal resistance to drop, but the pressure drop across the cooling unit also increases.

Pressure drop across a water cooler

Figure 4. Pressure drop across a water cooler for 2 different lengths (incl. 90 mm end pieces) dependent on the volumetric flow rate (water-glycol mixture 50%:50%, diagonally opposite inlet /outlet, Ta = 55°C)

The length data in Figure 4 applies to a water cooled heatsink profile including 90 mm end pieces, e.g. …/390 means 300 mm heatsink profile + 90 mm end piece. This image also shows that an elongation of the profile by 66% from 180 mm to 300 mm will only increase the pressure drop by around 15%. It thus follows that the major part of the pressure drop is caused by the end pieces. This is not surprising, since there is narrowing in the cross section at the connection pieces, a surface for water distribution, and 4 changes in direction which are responsible for the pressure drop. If a greater volume of water is to be pumped through the cooling circuit with reasonable pumping power, large pipe diameters are needed. In addition, care should be taken to ensure the following for the cooling cycle:

  • no narrowing in the cross-section,
  • no slam-shut valves,
  • as few directional changes (elbows) as possible

Thermal resistance as a function of the volumetric flow rate of the water cooler for various lengths and a 50:50 water/glycol mixture

Figure 5. Thermal resistance as a function of the volumetric flow rate of the water cooler for various lengths and a 50:50 water/glycol mixture

The SKiiP systems with water cooler supplied by SEMIKRON are subjected to a leak test at a test pressure of 6 bar. The recommended operating pressure is 2 bar. Nearby a known operating point, Rth(s-a) can be determined as a function of the volumetric flow rate according to the following equation:

 R_{th(s-a)2} = R_{th(s-a)1} \cdot \Big (\frac {\dot{V}_1}{\dot{V}_2} \Big )^k

where K = 0.3…0.5

Coolant, cooling cycle, and chemical requirements

The typical heat transfer medium to be used for liquid cooling is often water or a glycol/water solution (anti-freeze). More rarely, deionized water or insulation oil (fluorocarbons and PAO = synthetical hydrocarbons) is used. Due to its high heat retention capability (specific heat capacity cp = 4.187 J/g·K), water is generally preferred to oil or glycol for heat dissipation. Water can either form a closed circuit and be air cooled by means of a heat exchanger, or fresh water is used which runs off after flowing through the cooling unit. Deionized water, which is characterized by low electrical conductivity, can be used in closed circuits. From the outset, fresh water is noticeably conductive, but this is of minor importance for semiconductor components with internal insulation since the cooling water remains de-energized in contrast to non-insulated components.

It is important to choose a liquid that is compatible with the cooling circuit and provides either corrosion protection or a minimum risk of corrosion. To ensure corrosion protection in SEMIKRON water cooled aluminium heatsinks, the glycol content must amount to at least 10%. Manufacturers of antifreeze mixtures even call for a higher minimum glycol content to avoid undercutting the necessary concentration of corrosion inhibitors for non-ferrous metals. The hardness degree of the cooling water must not exceed 6. At least for coolant temperatures above 60°C, it is recommended to use a closed cooling circuit. Some of the explanations and tables below dealing with coolants originate from an application manual by Lytron Inc. Table 3 provides recommendations regarding which metals and liquids are compatible in the cooling circuit.

  Water Glycol mixtures Deionized water

Non-conductive liquids (Fluoro-inert, PAO)

Copper X X   X
Aluminium   X   X
Stainless steel X X X X

Table 3. Materials and compatibility of liquids

Fresh water

Water is the most effective cooling liquid due to its high thermal capacity. It is recommended to use a closed circuit. Depending on its chemical composition, fresh water or tap water may cause rust formation in metals. Chloride, for example, which can usually be found in tap water, is corrosive. Fresh water should not be used for a liquid cooling circuit if it contains more than 25 ppm of chloride or sulphates. The proportion of calcium and magnesium in water must also be observed, since both minerals cause limescale on metal surfaces, thus reducing the thermal performance of the heatsinks (Figure 6).

Minerals Recommended limit
Calcium < 50 ppm
Magnesium < 50 ppm
Chloride < 25 ppm
Sulphates < 25 ppm

Table 4. Recommended upper limits for ions in cooling water

Cooling channels and turbulators

Figure 6. Cooling channels and turbulators (spirals); a) New; b) After extended use with unsuitable cooling liquid and heavy limescale

Deionized water

Deionized water is free from ions such as natrium, calcium, iron, copper, chloride, and bromide. The deionization process removes harmful minerals, salts, and other impurities which may cause corrosion or limescale. Compared to tap water and most other liquids, deionized water has a high electric resistance and is an excellent insulator. But it easily turns acidic when it comes into contact with air. Carbon dioxide in air dissolves in water, thus resulting in an acidic pH-value of approximately 5.0. Pressure compensation vessels must therefore be separated from air by means of a membrane. This limits the maximum fluctuation of the temperature range for the coolant. It may be necessary to use anticorrosives in applications with deionized water. Connection pieces should be nickel coated. Copper leads are incompatible with the use of deionised water for cooling plates or heat exchangers. Leads made of stainless steel are recommended.

Inhibited glycol and aqueous solutions

Due to the corrosive effect of water and the often necessary frost resistance, open or closed circuits with pure water are hardly ever used. Ethylene Glycol Water (EWG) and Propylene Glycol Water (PWG) are the two most frequently used solutions in liquid cooling applications. Ethylene glycol has positive thermal properties such as a high boiling point, low freezing point, stability across a wide temperature range, and a relatively high specific heat capacity and thermal conductivity. It also features low viscosity, meaning the piping requirements are less strict. PGW is used for applications where toxicity might be a problem. The glycol used in cars should not, however, be used in a cooling system or heat exchanger, since it contains a silicate based rust inhibitor. These preventives may turn solid and cause deposits to form on the surfaces of heat exchangers and in doing so impair their efficiency. Glycol solutions should contain an anticorrosion agent.

By adding glycol, for example, the heat retention capability of the coolant will diminish (e.g. 3.4 J/kg·K for an addition of 50% glycol and a coolant temperature of 40°C). Since the viscosity and specific weight of the coolant will increase, the thermal resistance from heatsink to coolant Rth(s-a) will increase substantially together in line with the percentage of glycol. Compared to pure water, 10% glycol will cause an increase in Rth of around 15%, while adding 50% glycol will cause an increase of 50...60%. If the glycol content is increased to as much as 90%, Rth will double. Please note that these statements also depend on the flow conditions in the heatsink and the coolant temperature.

Influence of water-glycol mixture ratio on R th(s-a) for different flow rates

Figure 7. Influence of water-glycol mixture ratio on Rth(s-a) for different flow rates

Mounting direction and venting

When setting up the cooling circuit, care must be taken that cooling is not blocked by air bubble build-up. The best mounting setup therefore consists of vertical channels, while the worst is horizontal channels on top of each another, since the top channel accumulates the air bubbles.

The preferred flow direction is upwards with the inlet at the bottom and the outlet at the top in the control cabinet. Loops in the water flow, i.e. an "up and down" arrangement in the cabinet is disadvantageous. In that case, vent valves would be required in the cooling circuit above the power semiconductors. After the cooling unit has been filled, a test run with the highest flow rate should be performed for an extended period of time (> 0.5 h) without exposing the unit to normal electrical operation. A high flow rate (l/min) may remove existing air bubbles that might have been generated when the unit was first set up.

When designing parallel cooling circuits, steps must be taken to ensure that the pressure drop is the same in all parallel channels (same number and length of heatsinks, tubes, number of directional changes). The inertia of water in the direction of flow must be observed, since this is responsible for straight water flow, even if equally long parallel paths turn off sideways. Baffles to feed water into the branching channels must be used.

Tips on how to arrange water coolers

Figure 8. Tips on how to arrange water coolers

Setting up the water circuit

Figure 9. Setting up the water circuit

Other liquid cooling possibilities


Microchannel coolers are a special type of liquid cooler (Figure 10). In the DBC process, the microchannel cooler is furnished with several copper foils that are inserted between two DBC layers. These foils are punctured such that the holes are offset against each other. In the ceramic substrate of the bottom DBC layer there is a liquid inlet and outlet. Due to the offset bores in the sheet copper stack, a turbulent flow is created even at a low coolant flow rate which ensures good heat transfer from the component being cooled to the cooling liquid. In this way, a very good cooling effect can be achieved with a relatively low pressure drop and little coolant. A disadvantage is the high risk that channels might get clogged up by dirt or scaling or the steep temperature gradient within a module due to the low water quantity.

Diagram showing cross-section of a microchannel cooler

Figure 10. Diagram showing cross-section of a microchannel cooler

Phase transformation cooling

Phase transformation cooling utilizes the fact that a given heat quantity (evaporation heat) - the amount of heat depends on the liquid - is required to evaporate a liquid in order to transform the liquid heat carrier into a gaseous state. If the gas condenses, this heat quantity will then be dissipated again. If you manage to keep this cycle of evaporating and condensing going in a closed vessel, large heat quantities may be transported from the point of evaporation to the point of condensation. Gravity and capillary forces will be sufficient to keep the heat carrier moving - pumps are not necessary. Various kinds of cooling equipment use phase transformation for heat transport.

Evaporative cooling

The coolant evaporates at the hot spots, e.g. a power module, gas bubbles will rise and condense on the colder case or separate condenser (Figure 11).

CAUTION: If the heat flow density is too high, a compact layer of vapor may be created at the heat source. This interrupts the thermal contact between heat source and liquid, meaning that cooling will be abruptly stopped (Leidenfrost effect).

Diagram showing evaporative cooling

Figure 11. Diagram showing evaporative cooling

Spray cooling, jet cooling

These cooling methods use the principle of spraying the liquid coolant onto the surface either as droplets or by a jet (Figure 12). To some extent, the heat of evaporation is utilized here, too. Cooling may be applied from one side or both sides. The coolant evaporates at the spot where it hits the surface and condenses at colder spots. Typical coolants are often inert liquids such as fluorinated hydrocarbons which are available with a wide range of boiling points. Water cannot be used for direct spraying onto chips because its electric conductivity gets too high even after short use, and this would cause shorting at the chip edges.

It is advantageous if the coolant hits the chip directly, since this will result in optimum cooling directly at the point of heat generation and the exchange of coolant at the spot which is to be cooled occurs rapidly. The disadvantages of spray and jet cooling are the low amounts of evaporation heat produced by the fluorinated hydrocarbons, the complexity of the cooling arrangement, the high pressure of 3 to 15 bar prevailing in the entire cooling system, and the risk of nozzle clogging (Ø some 0.1 mm). Another problem is the densely packed bond wires lying side by side which often obstruct direct spraying of the chips.

Principle behind spray cooling (a) and jet cooling (b)

Figure 12. Principle behind spray cooling (a) and jet cooling (b)

Direct base plate cooling

This cooling method eliminates the thermal resistance of heatsink and thermal paste layer between the module and the cooling liquid by directly mounting the module and its base plate over an opening in the heatsink. The necessary sealing is ensured by an O-ring. Rth(j-a) can be reduced by about 25% with this method. Two different types can be implemented. With the first type, the module base plate has a structured surface (pin fins) which is immersed into the cooling liquid contained in a trough. For the second variant, the company Danfoss has coined the term "ShowerPower®". Here, a plastic insert with many parallel holes in the heatsink opening creates a turbulent and vertical flow which ensures good and even cooling (Figure 13). The advantage of the latter solution is the low cost manufacture of the plastic inserts as compared to structured base plates, while its disadvantage is the reduced contact area and the high pressure drop.

Test cooling plate with various inserts for direct base plate cooling

Figure 13. Test cooling plate with various inserts for direct base plate cooling


For more information, please read:

Heat Transfer in Power Semiconductor Devices

Cooling Methods for Power Semiconductor Devices

Forced Air Cooling of Power Modules

Thermal Modeling of Power Module Cooling Systems


VN:F [1.9.17_1161]
Rating: 4.0/6 (2 votes cast)
Water Cooling of Power Modules, 4.0 out of 6 based on 2 ratings

This post was written by:

- who has written 197 posts on PowerGuru - Power Electronics Information Portal.

Contact the author

Leave a Response

You must be logged in to post a comment.