Thermal Management of Planar Magnetics

Planar transformers and inductors for high-frequency converters differ radically from conventional magnetics in that they do not use magnet wire.

Instead, windings are copper-foil lead frames or flat copper spirals laminated onto thin dielectric substrates. These windings are stacked on flat low-profile ferrite “E” cores, which are glued together with fine-grain epoxy. Thin Mylar, Kapton or high-temperature Nomex films provide the necessary interwinding insulation.

Key benefits of planar technology include:

Suitability for power levels from a few watts to 30 kW
Low-package profile — only 60-mm high for 20 kW
Efficient high-frequency operation — 98% to 99% up to 3 MHz and above
Excellent repeatability due to pretooled components
Low-leakage inductance
Easily terminated multiple windings
Minimum skin effect
Standard outlines compatible with application-specific custom designs
Usability in both square- and sine-wave topologies
Compatibility with sophisticated thermal management techniques.

Core losses in ferrite-cored transformers suitable for operation above 100 kHz are lowest when the core “hot spot” is about 100°C. However, it isn't economic to operate at this low a temperature. A maximum of about 130°C is usually dictated by winding-insulator limitations, although higher-temperature materials are available at increased cost. These same hot-spot criteria apply to all ferrite-cored designs.

Copper losses are less troublesome in planar than in round-wire designs. Due to skin effects, losses are concentrated on the surface of wafer-thin wide flat conductors, so there's little copper wastage. Because adjacent layers are stacked tightly together like pages in a book, cooling isn't difficult. By contrast, the same skin effects in round-wire designs waste copper, while minimal line contact between adjacent wires hampers heat extraction.

For these reasons, raw materials are better used in planar designs, yielding smaller size and lower cost. Most efficient heat extraction is achieved by conduction — mounting the transformer onto a heatsink or equipment chassis — rather than by convection (radiation) cooling. The low-profile flat nature of a planar design provides an ideal large-area mounting surface for conduction cooling the core. The flat spiral-wound copper-winding layers also lend themselves to conduction cooling via metal spacers clamped between the outer-winding layers and the heatsink.

Test Principle

The electrical efficiency of a properly dimensioned planar transformer at full-load is approximately 98% to 99%. This performance is achieved when copper losses and core losses are about equal. When core and copper losses differ markedly, efficiency suffers badly, falling to 95% to 96% at a loss ratio of four in either direction.

DC copper losses vary with temperature:

R_T = R₂₅ [1 + 0.004 (T - 25)]

Where:

R_T = wire resistance at T, R₂₅ = resistance at 25°C, and T = operating temperature (°C)

Skin effect and proximity losses then boost effective ac resistance. As a result, real copper losses exceed those estimated from I²R_T.

Core losses vary with flux density and frequency as well as temperature. When they exceed about 100 mW/cm³, heatsinking is usually desirable. Cores driven with square waves have lower core losses than those driven with sine waves. Copper losses are much greater with square waves than with sine waves. These complex interrelationships suggest that testing a transformer to establish its “hot spot” temperature isn't a simple matter.

One method in common usage is to heat the transformer by energizing its series connected windings with dc. The amplitude of the current is chosen such that I²R_T = rated power loss. This method is somewhat misleading for two reasons. First, the effective value of R is not R_T, so an accurate estimation of power dissipation is difficult. Second, is the high amplitude of dc necessary for rated dissipation without core losses? This means copper losses will be twice their real value, so the temperature profile in the transformer will be distorted.

A more logical technique is to test the transformer in its real application at rated power. However, with high-power throughput, it's difficult to measure small power losses. At 99% efficiency, detecting 20 W in 2 kW requires better than 1% accuracy, which is not easy to achieve.

A better approach is to energize the transformer from a square-wave inverter at rated frequency with its secondary short-circuited. The RMS voltage applied to the primary is then adjusted until the measured power at the primary equals expected losses when the transformer is operating normally. With this method, core losses are not significant due to the artificially low-primary voltage. Still, the presence of high-frequency skin effects and the right ratio of primary-to-secondary currents ensures that temperature profiles reflect real operation more closely than with dc.

Test Setup

In the circuit of Fig. 1, the transformer under test (TUT) is energized from a half-bridge inverter consisting of MOSFETs Q1 and Q2 together with capacitors C1 and C2. The inverter runs at 75 kHz. A Pearson #411 current transformer measures primary current, while primary voltage is monitored with a Tektronix high-voltage differential probe. The digital storage scope's math processor multiplies these inputs together to calculate and display power.

A half-bridge topology immunizes the transformer from “flux walking” and possible saturation. Power loss in the TUT depends on the dc bus voltage set with the variac. This could also be done by pulse-width modulation of the dc bus, but this results in very narrow power pulses with dangerously high-peak to average currents at the low-output voltages needed here.

The short-circuited configuration shown is preferred for the precise measurement of transformer losses. However, the inverter is dimensioned to work reliably from the full-rectified mains voltage, with or without pulse-width modulation, for verification of the Payton transformer behavior under normal working conditions.

Control Circuit

Fig. 2 illustrates the control circuit employed to switch Q1 and Q2 at 75 kHz. A 150-kHz clock frequency is generated by IC1 connected as an astable multivibrator. By using one of its half-frequency complementary outputs, operation at 75 kHz with a true 50% duty cycle is assured. The clock signal is converted into a pair of complementary outputs incorporating “dead time” by IC2.

Dead time is the pause between the turn off of one MOSFET and the turn on of the other, to prevent simultaneous conduction or “shoot through.” It's set here to a generous 1 µs through R2 and C5. Long dead times don't limit performance in this circuit, where the output voltage needed for transformer testing in a short-circuit mode is relatively low. Were the circuit to be used for full-power testing, it would desirable to shorten the dead time.

The two complementary pulse trains from IC2 feed an IXBD4410/4411 bridge-driver chipset, IC3/IC4, which translates them into appropriate drive voltages for the two MOSFETs. While the ground-referenced low-side Q2 is driven directly by IC3, signals for the “floating-ground” high-side Q1 are transmitted across the 1200-V isolation barrier to IC4 via a miniature pulse transformer.

Supply voltage for the floating IC4 is derived from the low-side 15-V bus by bootstrap diode D1 and capacitor C7. Because the switches in a phase-leg topology may be subjected to very high dv/dts, IC3 and IC4 apply negative gate bias to both Q1 and Q2 during the off periods. The chipset shuts the system down in the event of overcurrents.

Test Procedure

This test is to compare cooling arrangements for Payton planar transformers. First, the secondary winding is short-circuited at its output terminals to minimize I²R losses in the link. The transformer is then mounted on a heatsink, or on an equipment chassis serving as a sink, using a clamp or epoxy adhesive. The various interfaces are treated with thermal “filler” as required. Finally, with power loss in the transformer adjusted to the required value, the setup is run for one hour to stabilize thermally. All temperature measurement points are taped matte black to facilitate accurate readings with a STEINEL reflected-IR digital thermometer. Temperature measurements are shown in the tables.

Test Results

Test 1: Natural convection cooling, no heatsink. Total heat output is the sum of convection losses from surfaces exposed to free air and conduction losses from those attached to a heatsink. Because convection is less effective than conduction, core surface temperature was measured with convection cooling alone. The TUT was set on a pair of triangular “V” blocks in free air.

The functional spec of a Payton #5844 transformer, chosen as the TUT, lists 15-W losses at 1500-W full power. This corresponds to a hot-spot temperature of 115°C with natural cooling. To dissipate 15 W in the TUT, the dc bus was 52.5 V. Temperatures were measured at three locations (Fig. 3) on the top surface of the core: the geometrical center, hot-spot over the winding window and core edge. The results for this and subsequent tests shown in Table 1. The ambient temperature, T_A, was 21°C.

Test 2: Hybrid conduction/convection cooling, transformer mounted on an extruded aluminium heatsink.

In Test 2.1, a 10-cm × 2-cm by 2-mm steel bar and two 5-mmΦ bolts were used to clamp the transformer to an extruded aluminium cooling fin. The bar was thermally insulated from the transformer by a triangular metal fulcrum placed between them. With this arrangement, the lower surface of the core was conduction cooled to the sink, the rest being convection cooled to air. Fin sink-to-ambient thermal resistance, R_SA was 1.7K/W and the interface was dry. Heatsink surface temperature, as checked around the periphery of the core, was reasonably uniform.

Table 1. Temperature readings (°C) of planar transformer with various forms of heatsinking.Test Condition Location 1 Location 2 Location 3 Fin 1 Natural convection cooling 100 106 95 Not applicable 2.1 Hybrid conduction/convection cooling with aluminium extrusion 64 69 58 42 2.2 Same as test 2.1, but with “Thermalcote II” thermal grease 62 67 56 42 2.3 Same as test 2.1 but with “Copperease” grease 62 67 56 42 2.4 Same as test 2.2 but with upper cooling fin 56 65 Not measured 42 2.5 Same as test 2.4 but with U-shaped aluminium bracket 53 58 Not measured 43 2.6 Same as 2.5 but with Kunze #KU-ALC “phase-change” aluminium foil 47 50 Not measured 44 2.7 Same as 2.5 but with 6-mm Al spacers 52 56 Not measured 46

The same setup was used in Tests 2.2 and 2.3. However, in Test 2.2 Thermalloy “Thermalcote II” silicone-free thermal grease was applied between core and sink. In Test 2.3, a different grease, the Comma “Copperease” copper-particle-filled high-temperature grease was used.

In Test 2.4, the steel bar/fulcrum arrangement was replaced with a 11-cm × 5-cm × 6-mm finned-aluminium extrusion, in direct contact with the transformer core and bolted to the heatsink (Fig. 4). This extrusion conducted heat away from the upper surface of the transformer, where it was dispersed to the air by natural convection without transfer to the heatsink below. Both upper and lower interfaces were treated with Thermalcote II.

Next, in Test 2.5, the miniature extrusion of Test 2.4 was removed, and the transformer was clamped to the heatsink with a fabricated “omega”-shaped aluminium bracket (Fig. 5). Here the core-heatsink interface was coated with Thermalcote II, with all other interfaces being dry. In this case, heat was extracted from the core by a complex mix of conduction and convection effects. As long as the sink temperature beneath the bracket feet remains lower than the bracket temperature, heat originating in the core transfers to the sink by conduction. At the same time, additional heat is evacuated to the surrounding air by convection from all surfaces of the clamp.

In Test 2.6, the same setup as test 2.5 is maintained, but with Kunze #KU-ALC “phase-change”-coated aluminium foil lining all interfaces between bracket, transformer and sink. Before testing, the assembly was oven-cured at 75°C to ensure complete transformation of the phase-change coating from solid to liquid. When this occurs, at about 60°C, all interface-voids are automatically filled as the thermally conductive compound flows to expel entrapped air.

And finally in Test 2.7, the setup is as above in 2.6. However, in this case 6-mm-thick aluminium spacers have been trapped between the heatsink and each end of the winding bundle where these latter exit the core, as well as between the winding bundle and upper bracket. The two lower pads conduct heat generated by copper losses directly to the fin, while the two upper pads do this via the bracket, which is long enough for this purpose.

Table 2. Temperature readings (°C) of planar transformer mounted to sheet metal enclosure.TestConditionCore CenterWinding WindowCore Edge Core Panel Core Panel Panel 3.1 Kunze thermal compound #KU-CR 58 45 61 44 45 3.2 Bergquist Silpad 800 60 44 63 43 43 3.3 Bergquist “Flexible Hi-flow” 57 45 60 46 46 3.4 Kunze graphite foil 58 46 60 45 45 3.5 Metal-particle loaded epoxy adhesive 74 41 87 41 39 3.6 With metal spacer 70 42 80 43 40

From these test results (Table 1), several observations can be made:

Conduction cooling of one surface to a heatsink produces a dramatic reduction of core temperature versus natural convection cooling.
Coating the transformer-heatsink interface with thermal grease yields further improvement. The relatively low heat density makes the choice of grease noncritical.
“Fin” cooling the core's upper surface of improves heat removal from the transformer.
Heatsinking four faces of the core has beneficial effects, because heat generation in the core is volume related. Four-sided cooling is improved by lining the inside of the bracket with phase-change coated aluminium foil.
In a well-designed transformer, copper losses should equal core losses. Using metal spacers to extract heat from the windings pays off. The presence of upper and lower spacers raises the core temperature, since heat transferred from winding to bracket reduces the latter's ability to cool the core. However, winding temperatures are much lower, confirmed by higher sink temperatures reflecting additional heat flow.
Lower spacers used alone reduce winding temperatures less without heating the core.

Test 3: Hybrid conduction/convection cooling, transformer mounted to a sheet-metal equipment enclosure.

Additional measurements were made to quantify the cooling capability of a typical equipment enclosure, doubling as an inexpensive heatsink. Unlike purposely made extruded fins with smooth machined surfaces, the surface of a punched metal panel is neither smooth nor flat. Being thin, it also lacks rigidity, so any attempt to apply tensile force from mounting screws provokes bowing, with predictable effects on thermals.

In such straits, the best interface materials are those capable of filling deep voids to minimize interface resistance. For test purposes, the transformer was attached to the inside face of a vertically oriented switched-mode power supply (SMPS) enclosure panel. Temperature measurement points were kept to the minimum necessary for performance assessment. One was at the geometric center of the core surface, the second on the core surface over the window where the windings lie.

Readings were taken at these two locations and directly opposite them on the back side of the panel. A backside reading also was taken under the edge of the core. The transformer was mounted to the panel with the same omega-shaped bracket used earlier. In all cases, tests were run with 15 W dissipated in the transformer. Ambient temperature was 21°C. Table 2 shows the test results.

A variety of interface materials were tried in these experiments. For example, Kunze thermal compound #KU-CR was used in Test 3.1 This high-conductance silicone-free compound is a nonliquid waxlike substance supplied in block form and housed in a convenient dispensing cartridge much like lip balm. It's easy to apply and isn't at all messy. Its thermal conductivity is twice that of conventional thermal grease. It was used to coat all interfaces between transformer, bracket and mounting panel

Bergquist Silpad 800, employed in Test 3.2, is a recent addition to the Bergquist range of thermally conductive insulation materials. Silpad 800 is recommended for low-cost applications where the highest possible thermal conductivity is required, despite low available clamping force. Although electrical isolation isn't generally needed when mounting transformers, the supple nature of Silpad 800 lends itself to filling the large voids found between poorly mating surfaces. All interfaces between the TUT, bracket and sink were lined with the material.

Bergquist Flexible Hi-flow applied in Test 3.3 is a phase-change coated aluminium foil similar to Kunze #KU-ALC. Like all phase-change materials, Flexible Hi-Flow must be heated to at least 61°C before it liquefies. All interfaces were lined with the material, and the assembly was baked at 75°C to properly condition it prior to test.

In Test 3.4, the interface material was Kunze graphite foil. This naturally occurring material has excellent thermal conductivity and anisotropic properties. An anisotropic material has higher thermal conductivity horizontally than vertically, making it useful for heat spreading.

A metal-particle-loaded epoxy adhesive such as that used in Test 3.5 provides a low-cost mountdown. With this approach, no investment is needed for tooling, production or attachment of a metal bracket. The downside is that conduction cooling is limited to the bottom-side alone. The TUT was glued to the SMPS panel with a very thin film of Soudal steel-loaded epoxy resin.

From the test conducted above, it's clear there is little difference in thermal performance among the materials tested (Table 2). This is probably due to the low heat density across the large surface areas involved, diminishing the importance of interface ΔT. However, this judgment doesn't apply to epoxy resin, with its limited conduction cooling.

When adhesive is used to mount a transformer, there's no risk of the panel bowing that bedevils attachment with brackets and screws. The windings can then be cooled with metal spacers in Test 3.6, as before. The beneficial effect of spacers is apparent.

Conclusion

The benefits from cooling Payton Planar transformer are largely economic. Strapping it to an existing equipment panel costs far less than the savings in core-size reduction or elimination of high-tech insulating materials.

Conduction cooling is the most effective. Because core losses are volume-related, four-side cooling is the best, easily arranged with an omega mounting bracket made from aluminium stock. Thick stock should be used to encourage heat conduction to the sink below. Interface thermals are minimized by use of an appropriate filler. Since heat densities are low in transformer applications, costly high-performance fillers are not normally needed. And because the surface associated with wraparound straps is large, convection should be encouraged by anodizing exposed aluminium surfaces black. Although more expensive, incorporating vestigial finning on the clamp is effective.

The stacked flat PCB or lead-frame windings of a planar transformer can be conduction cooled too, by trapping metal spacers between the exposed ends of the windings and the heatsink. Winding temperature can be further reduced with additional spacers to the omega bracket above, but at the expense of a small rise in core temperature. Any technique that cools the windings pays off, in that power throughput is often limited by insulation temperature. Reducing temperature here reaps cost dividends.

Bolting a transformer to an extruded heatsink isn't difficult. This isn't necessarily the case with a crudely stamped equipment chassis, which almost certainly will not be flat. The application of a low-thermal-resistance interface filler to expel air from the voids is highly recommended. Any modern interface material will do this job well.

For more information, contact Jim Marinos at [email protected] or www.paytongroup.com.