Thermal Resistance and Capacitance are Critical Parameters of Power Devices

Modern power transistor chips shrink continuously, GaN and SiC chips are by nature even smaller than silicon. Thermal resistance and capacitance are impaired.

 

Definition of the problems

From its beginnings the semiconductor industry is continuously busy to shrink their chips. Chip size not only is a cost factor, but in the case of power transistors the thermal resistance and capacitance are adversely affected. The smaller contact area between chip and housing increases the thermal resistance junction-to-case Tjc, and the thermal capacitance is reduced due to the lower mass. Newer chips thus run hotter than their predecessors for the same power dissipation, and those chips can not take as much short.term overload either.

 

SMPS start-up and output overload

In SMPS start-up and output overload can impose a sudden increase in power dissipation, the lower the thermal capacitance is the higher the temperature will rise and may reach destructive levels. While older, larger chips took such stresses easily more recent ones may fail. Users are seldomly notified or warned if a chip was shrunk, sometimes a suffix like “A” is added to the type designation, which should ring the alarm bells with engineers. Manufacturers praise the lower electrical capacitances, the faster switching but it is a rarity if they also mention the disadvantages incurred. Often the data sheet remains unchanged although the chips were shrunk. So it may happen that suddenly transistors from a new shipment fail while such failures of the same type previously never occurred.

 

Chip Shrinking

In such cases prime customers can force the semiconductor manufacturers to continue production of the larger chips, and, although the former type designation was removed from the firm’s catalog it is still available - if one knows this. Sometimes chip shrinking goes along with a change to a less expensive technology which is more delicate and sensitive to overload. In case of sudden, unexplainable failures it is advisable to check first whether these parts came from a new shipment. If yes the next step is to compare new chips with ones from earlier shipments. If these tests confirm that the new ones fail and the old ones not confront the manufacturer with the results and ask whether the chip was altered, and what the date codes of the new ones are.

Distributors can then be asked whether they still have old chips in stock. If not, one will be forced to switch to the new type; in order to get a chip which is comparable to the “old” one, it will be necessary to choose a type which is one or two sizes larger than the data sheet implies, the thermal resistance spec should be the guide. This also applies when looking for the right size transistor for a new design. Not only electrically, but also thermally a transistor must be adequate.

 

Thermal Circuits Ohm’s law

In thermal circuits Ohm’s law applies: temperature difference equals voltage, heat flow equals current, and the resistances add up. The vital parameter is junction temperature Tj. The user has no access to the junction and is unable to check its temperature. The manufacturers place test structures like diodes on the chips which accurately measure the temperature.

The user has to trust the manufacturers’ data sheet specifications of thermal resistance junction-to-case Tjc, he can only measure the case temperature, but in order to arrive at meaningful results two problems must be solved: how can the temperature of the transistor case be measured if there are high fast pulses like 360 Vpp in 10 ns? And how can the power dissipation of switching transistors be accurately determined?

 

Method of determining the transistor power dissipation

High fast pulses will disturb most temperature measuring instruments so much that they give false results, also the high capacitance of the probe may disturb the switching. It is therefore necessary to measure immediately after turn-OFF with a contact probe. The best method of determining the transistor power dissipation is this: first the case temperature of the power transistor is measured, then the transistor is exchanged against a power resistor in the same type case, e.g. TO-220. This resistor is heated up to the same case temperature from a precision supply; and this power is identical to the transistor’s. Attempts at calculating a switching transistor’s power dissipation lead at best to approximate results because switching is very complex.

Standard silicon power mosfets are in fact no simple mosfets as the symbol implies but consist of a cascode connection of a power jfet and a mosfet and associated capacitances - which is very seldom mentioned. Especially turn-OFF is complicated, and the vital capacitances are nonlinear. Figures 1 and 2 show the true internal structure of power mosfets, small signal mosfets conform to the symbol.

 

 

Figure 1: Cross sectional view of a standard silicon power mosfet shows a cascode connection of a power jfet and a mosfet. Yes, the power device is a jfet

 

This complex structure explains why it is easy to switch the mosfet ON, but the user has no control over the switching OFF, he can only switch the inside mosfet OFF, but he has no access to the power jfet, and current will continue to flow alone via the capacitances. In practice the products of the various manufacturers with the identical type designation vary widely in their turn-OFF behavior. Turn-OFF happens in stages: the current will fall rather fast to about half, but stay there for times which reach from a few ten ns (good mosfets) to over hundred before it falls to zero.

As the voltage will rise quite fast, the product voltage times current can reach destructive levels. Experienced engineers know this and will specify only the products of the manufacturers they tested. In other words: by no means is it acceptable to buy from any manufacturer who produces parts with the same designation. Failures caused by slow turn-off have nothing to do with thermal resistance and capacitance, but it is evident that a large chip will take overloads easier.

 

Figure 2: Equivalent circuit of the above structure. Both drawings are from Harris Semiconductor application notes.

Figure 3: Dependence of semiconductor life on the junction temperature: 10-6 failures per hour vs. time.It should be noted that there are more influences on the service life, e.g. moisture will influence plastic encapsulated devices. SMD pack-ages are more affected than through-hole devices because there is only a thin layer of plastic

 

Maximum Operating Junction Temperature

A frequent misunderstanding concerns the meaning of the “maximum Tj “ in the data sheets. By no means is it implied that a component can be operated at or even near this temperature! The manufacturer guarantees only that the part will function up to this temperature, but no more. Quite independent of the maximum junction temperature the law of halving life for every 9 K applies! This is why important users of electronics as a rule prescribe a maximum operating junction temperature like 110 or even 90 C in order to ensure a meaningful service life resp. low failure rates.

Applications where such components are stressed close to their maximum ratings, such as in the vicinity of combustion engines in autos are possible because it is often forgotten that an auto is only operated for a few thousand hours in its life, and that the worst operating conditions occur only seldomly, too, e.g. during traffic jams in summer.

The exponential dependence of life on the junction temperature meaans that every degree counts: the cooler a component runs the longer it will function. There are large scale products like smartphones which are designed for two years, but the bulk of electronic products has to last for many years, even up to 15 or even 30 years. The designer can only meet such requirements by operating the components at low junction temperatures.

 

Thermal resistance and capacitance

The newest Coolmos chips became so small that the thermal resistance in TO-220 of some goes up to 2 K/W while the older ones were mostly below 1 K/W. This means e.g., at a power dissipation of 10 W, an increase of 10 degrees C in junction temperature with such new devices over their predecessors.

This problem is no better with the new materials GaN and SiC: these chips are much smaller than silicon chips of the same Rdsons, a fact which is often overlooked. The thermal conductivity of GaN is lower than that of silicon while that of SiC is higher. Both materials take higher temperatures than silicon so that even though the masses are so small the temperature increase in case of short-term overload may rise even above 250 C without causing a failure. Due to plastic package limitations most parts are specified for a continuous Tj of max. 150 C.

Thermal capacitance equals electrical capacitance; it is charged up to a certain temperature by a thermal current like a capacitor which is charged up to a certain voltage by a current. Thermal capacitance is defined by:

 

Cth = V x ρ x cp V = volume (m3), ρ = density (kg/m3),

cp = specific heat (J/kgK)

 

In practice only the product ρxcp counts which does not vary much between materials. Data sheets rarely specify thermal mass resp. capacitance, they show a graph of the thermal impedance vs. time with the single pulse energy as the parameter. The higher the thermal mass, the more short-term overload a chip can take without failure.

In SMPS, if soft start is not provided, start-up will overload for a moment because all electrolytics are empty. Simple output current limiting circuits are fairly slow and also cause a momentary overload which is more dangerous because the power devices are already hot.

 

Mounting considerations

Transistor thermal resistance Tjc is a component property, the designer has only influence on the other thermal resistances: case-to-cooling surface and mostly also an insulator. Here quite a few problems are hidden. For the benefit especially of our young engineers the following hints: Neither the surface of the transistor case nor that of the mounting hardware are flat, also the surface finish is often poor.

The actual contact area available for the heat transfer is hence much smaller than the area of the transistor case, causing an increased thermal resistance. Also the method of mounting the transistor influences it substantially. Rivets are completely out. Screws are a solution which has become obsolete for several reasons.

Apart from the fact that mounting with screws is labor intensive and expensive, it is only acceptable if the mounting torque is limited and a spring washer is used. Limiting the torque is essential because if the screw is tight-ened too much the effect shown in Figure 4 will ensue: the case will be lifted so that the contact area is severely reduced.

 

Figure 4: What happens if the screw is tightened too much: the rather soft case metal will flow, bend and lift the case off, causing a severely reduced contact area and thus increased thermal resistance.

 

The professional method uses spring clips which press upon the plastic body; this exerts the pressure where the chip sits and prevents any bending of the case. Also the spring keeps the force up over time even if an insulator softens and yields. A cost saving and effective cooling method is to use a U-shaped aluminum chassis and to arrange all power semiconductors along the circumference of the e.c. board; the spring clips press the components against the side walls of the chassis.

Care must be taken regarding the transistor leads, the material is quite brittle and does not take much bending. If bending is necessary it should be limited to once, and the leads must be supported near the body. Also no mechanical stress must be left on the leads after mounting, otherwise failures are programmed! Connection with wires solves this problem, but is rare because it is expensive and increases lead resistance and inductance.

SMD power devices are a special case because the standard FR-4 material is a very poor heat conductor. The large copper areas shown on the data sheets are never realized in practice. Power dissipations are thus limited to a few watts at best. Higher values require the use of heat sinks.

For heat sinks manufacturers specify thermal resistances, but these are obtained under special testing conditions and thus variables. In any application where the heat sink is mounted otherwise the figure does not apply.

 

Interface materials, Insulation

Most switching transistors have the drain connected to the tab resp. case. In offline-SMPS the drains are on line potential, high insulation against the cooling surface, mostly the chassis which is connected to safety earth, is required.

 

The insulation between drain and chassis has to fulfill three requirements:

1. Insulation good for at least 1 KVrms 50 Hz up to 4 KVrms; also creepage requirements have to be observed, e.g. 8 mm; in case spring clips are used which press upon the body, with TO-220 8 mm can not be realized against the drain, in this case the spring clip must be also insulated.
2. Good thermal conductance,
3. Low capacitance and low dielectric losses.

 

The latter is often overlooked; the high and steep pulses at the drain, e.g. of a PFC switching transistor 360 Vpp in 10 ... 20 ns cause enormous dielectric currents through that insulator which produce dielectric losses in the first place and, secondly, these currents flowing into the chassis generate strong emi. The dielectric losses heat the insulator which goes by undetected because this heat can not be discerned from the heat contributed by the transistor.

The current and the emi disturbances are proportional to the dv/dt, hence faster switching will improve the efficiency somewhat, but cause higher stress on the insulators and increased cost and space for emi filtering components.

These days there is much talk about the faster switching of GaN and SiC transistors, the truth is that those are mostly cascodes which switch in such circuits in about 5 ns, the very same results are obtained with Si Coolmos cascodes, because - what is not conveyed - the switching speed in a cascode is solely determined by the lower transistor which is a standard Si mosfet; it is immaterial which kind of transistor is upstairs, also a high fT npn will do. These GaN and SiC cacodes have true advantages in such circuits where also reverse currents flow and they can operate at much higher junction temperatures.

 

Figure 5: Dependence of insulation properties on frequency of typical materials, here of polyester foils, shows the drastic decrease. At typical SMPS frequencies around 100 KHz a 75 um foil drops from 200 V/um to 25 V/um, at 1 MHz almost nothing is left! Nevertheless, PE is the most used material for SMPS inductive components. Such graphs are rarely available for thermally conductive insulation materials.50 Hz material specifications have no meaning at these frequencies and will lead to grossly false designs

 

The insulating properties of all insulators deteriorate drastically with increasing temperature, also their life is shortened, ceramics excepted. Ceramics like alumina are ideal: from the electrical standpoint, they take high voltages and have extremely low losses and capacitances, they take high temperatures without ageing, but in practice they are difficult to handle and expensive. The best material is beryllium oxide but it is poisonous. Even if the surfaces of ceramics are lapped they can hardly be mounted without thermally conductive grease on both sides. Only spring clips may be used although most ready-made plates have holes.

Plastic insulation materials age, the more, the higher the operating temperature which is, as a rule, quite high for thermally conductive insulation.

 

Figure 6: The 5 IEEE classes for insulation materials. The graph shows the dependence of life on the operating temperature. This is especially important for electronic gear which is continuously on. Take class 130 as an example: at 100 C life is only 2 x 104 which is a little more than 2 years

 

In practice 3 kinds of materials are used:

1. silicone rubber, filled with ceramic powder,
2. so-called phase-change materials.
3. Gels.

 

Earlier mica and silicone grease, filled with silica, alumina or zinc oxide, were customary, but the latter suffers from severe disadvantages: Grease tends to migrate, leaving a film which fouls up connectors, prevents adhesion and soldering.

The thermal cycling favours the migration of the grease which then leaves voids under the device, causing increased thermal resistance and higher Tj, even failure. It is also deleterious to eyes and hands and unpopular in production environments.

The first material class fulfills all requirements, but care must be taken not to use too thin materials, the transistor cases will cut into the material, for offline-SMPS purposes 0.4 mm material is a minimum.

 

Phase-change materials

Phase-change materials are solid and can be applied like any other solid material; as soon as the transistor heats up to its operating temperature this material starts to soften and flow, filling the gaps and holes like a grease, but it does not migrate. The thermal conductance is the same as that of grease. The gel materials deliver the best performance, they transfer the heat at the lowest necessary pressure and don’t migrate either.

Of course each insulator contributes additional thermal resistance. Typically a TO-220 transistor silicone rubber, ceramic filled, insulator of 0.4 mm shows 4 K/W which is much compared to the transistor’s Tjc of e.g.1.5 degrees/W, not to speak of large dies in professional TO-3 cases which sport 0.1 degree/W. Taking the above example 10 W of dissipation would create a ΔT = + 55 K between junction and cooling surface. At a typical 85 C ambient temperature in a SMPS and at its housing the Tj would reach 140 C which is not acceptable.

Either the dissipation must be reduced or an overtemperature sensor on the cooling surface (chassis) switches off when 60 C are exceeded. This example illustrates, by the way, that the currents and power dissipations on data sheets have mostly lttle bearing in practical applications. Fictitious 25 C specs are theoretical, no power device operates at 25 C case.

 

Figure 7: Interface thermal resistance of TO-220 without grease for several older interface materials.Curve 1 ThermalfilmCurves 2 and 3 micaCurve 4 Hard anodizedCurve 5 and 6ThermalsilCurve7 Bare jointCurve 8 Grafoil

 

Thermal runaway

Thermal runaway is a condition which leads to eventual destruction of a power device. If a transistor heats up, its power dissipation increases, this holds for all types. In mosfets the Rdson increases. The higher dissipation causes a temperature increase. This is a thermal closed loop system, similar to an electrical closed loop. As long as the loop gain stays below unity the system is stable. Each ΔP creates a ΔTj. Two situations must be considered. In the first scenario the cooling surface remains at a fixed temperature. When the power dissipation rises, the temperature difference (equals voltage) between the junction and the cooling surface increases, causing increased heat flow (equals current).In practice, this is rare, consider a popular execution of a SMPS, the U-shaped chassis, also called open frame, which serves as the cooling surface for the power devices; this will heat up, so that the temperature difference between junction and cooling surface will remain constant or shrink.

An increased power dissipation will hence not cause an increased heat flow. The consequence is that the junction temperature must rise according to the sum of Rth,jc and Rth, c-to-chassis . Now comes the dependence of the transistor losses on its temperature into play: Rdson rises with temperature and mostly in a nonlinear fashion. Each ΔP causes a higher ΔTj the higher the temperature already is. In other words: the ΔP necessary to generate a given ΔTj becomes ever smaller the hotter the junction, eventually it becomes zero, and this is the point of runaway. The vital fact is that in such a situation the increase in power dissipation comes about automatically, without an increase in the product voltage times current. The loop gain is now unity, the temperature will increase fast to destruction.

 

Figure 8: TO-220 with thermal grease and the same materials as above.

 

It is hence necessary, in a professional design, to check whether there is the danger of thermal runaway; with the aid of the curve Rd-son vs. junction temperature in the data sheet and the known thermal resistances this can be done. Each SMPS requires an overtemperature sensor which switches off before destruction can set in. The right place is mostly the chassis in the vicinity of the power devices Quite independently an electrical overload protection must exist, if this is not provided an output overload will cause a sudden increase in junction temperature; before the chassis is heated up so much that the sensor responds, the transistor will be destroyed. There are other temperature sensitive devices, e.g. low forward voltage Schottky diodes; they suffer from the disadvantages of low reverse voltage and very high leakage currents which can cause thermal runaway.

 

About the Author

Dr.-Ing. Artur Seibt is a professional electronics design lab consultant with specialization in SMPS with 40 yrs. experience incl. SiC, GaN, D amplifiers. Inventor of current-mode control (US Patent) and He is also an expert in EMI design.

 

This article originally appeared in the Bodo’s Power Systems magazine.