The electronics industry is continuously searching for ways to reduce the package size of circuits. As a result, parts density is increasing with every design. Hybrid devices are part of this ongoing effort to fit greater amounts of circuitry into smaller spaces.
Reducing the size of hybrid designs requires the development of smaller high-power components and the ability to place these power components alongside temperature-sensitive parts in densely packed designs. As the parts density rises, however, heat radiated by power components becomes unacceptable because of the negative impact it has on circuit performance and reliability (see "Temperature Versus Reliability,").
As a result, one of the major goals in new designs is to minimize heat inside the package through techniques such as heatsinks, heat pipes, forced-air cooling, and water cooling. Because power resistors are among the main components that contribute to heat generation, they are the focal point of much of these efforts.
Traditionally, power resistors have been available in a variety of styles, including wirewound, carbon composition, carbon film, and metal film. In their usual cylindrical construction, such resistors typically operate at a power density of 5 W/in.2 Another style of power resistor, the conventional thick-film resistor, can operate at 10 W/in.2 At the same time, thick-film chip resistors can attain a power density of about 30 W/in.2
Unfortunately, these devices transfer most of their heat through radiation. (Thick-film chip resistors are a special case, since they conduct heat through their end terminations directly to the circuit board.) And because these resistor types are difficult to heatsink, their use in high-density designs is limited.
But planar thick-film resistors offer a higher-power alternative. These components can be made with integral heat spreaders that attach to heatsink assemblies. Properly constructed, planar thick-film resistors can operate at significantly higher power densities than traditional power resistors.
Using refined processes and materials, it's possible to manufacture planar thick-film resistors capable of operation at 250 W/in.2 The performance of such resistors has been tested for samples operating at 40, 60, and 1000 W, constructed in various package styles using either air- or water-cooled heatsinks. The results obtained with these prototypes can be applied in the design of hybrid devices, as well as multichip modules.
Operation at very high power densities requires manufacturing techniques of the highest caliber. Thick-film resistors can tolerate minor defects in the resistor layer when operating at 10- to 50-W/in.2 But at 250 W/in.2, the resistors will fail as a result of minor defects, such as pinholes or airborne contaminants resting on the wet-screened surfaces during manufacture. Management of current densities, spatial thermal heat patterns, and voltage stress are all critical to success at the higher power densities.
Operation at 250 W/in.2 stresses components to much higher levels than customary. The resistor is operated at higher voltage, current, and power levels than are usually accepted in hybrid design. These higher levels also are considerably above the manufacturer's published specifications for the thick-film inks used to make the resistors. Tests performed on the experimental thick-film power resistors show the effects of this higher stress level.
Although all of the resistor samples tested were designed to function at 250 W/in.2, packaging and assembly varied according to each resistor's overall power rating. Each resistor element consisted of thick-film resistor material deposited on an alumina substrate.
In the case of the 40-W resistor, the substrate was housed in a TO-220 package, while the 60-W version was encased in a somewhat larger Z package. In both cases, solder was used to attach the package to the heat spreader.
For the 1000-W resistors, packaging consisted of an aluminum heatsink mounted to the resistor substrate. Detailed assembly instructions are given below.
There are several different material systems used to make thick-film inks. Ruthenium dioxide (RuO2) is the classic resistor material of choice in the thick-film industry. Its resistivity ranges from 1 Ω per square to 1 MΩ per square. (It is conventional practice in the thick-film industry to use a unit called "ohms per square." The basic equation for resistance is R = *l/A, where * is resistivity, l is length, and A is area, all in consistent units of length. A = thickness (t) * width (w) so that:
R = */t * l/w
The term */t is defined as the resistivity in "ohms per square.") These resistivities are 12.5 to 12.5 million times higher than those of wirewound alloys.
It's then possible to use these inks to make resistors with very high values in a very small area. Inks made with RuO2 have been optimized for resistance stability at fairly low power densities.
Another option, pure silver inks, were originally formulated for use as conductors. Their resistivity is comparable to that of wirewound alloys with a temperature coefficient of resistance that's very high, often in the range of 500 to 1000 ppm/°C.
Additionally, alloys of palladium and silver or of platinum and silver have recently become available. The temperature coefficient of resistance of these alloys is low, often in the 50-ppm/°C region, which is satisfactory for most resistor applications. Besides these alloys, many others have been made into thick-film inks to satisfy the need for a particular characteristic.
Conventional RuO2-based resistor inks perform poorly in surge applications. Their oxide and glass layers aren't good at removing heat from the printed resistor. To overcome this deficiency, thick-film manufacturers are developing materials specifically for surge protection, which is required in many telecom and power-supply applications. Surge materials must not only withstand large power spikes, they also must dissipate heat effectively. In repetitive surge situations, heat can build up quickly. Proper packaging is required to aid the heat flow from these materials.
As substrate materials, aluminum nitride and beryllium oxide are attractive for their thermal properties. But the high cost and limited availability of these products makes alumina the substrate of choice. Also, beryllium oxide is very toxic, and it's considered a health hazard if any particulate matter becomes airborne.
To construct a resistor in the TO-220 package, thick-film material was applied to the face of a 0.4- by 0.4- by 0.040-in. alumina substrate. The backside of the substrate was coated with a standard silver conductor and copper plated. The substrate was then soldered to the metal frame using 309°C eutectic solder, which is a good filler between the rough ceramic and the smooth metal.
Power was applied to the unit until the surface of the resistor reached 150°C. Without a heatsink, the unit operated at roughly 3.5 W. When attached to a heatsink, the unit reached 150°C at a power of 40 W for a power density of 250 W/in.2
The thermal resistance from resistor to heatsink can be determined from the plots of resistor and heatsink temperatures over a range of power levels (Fig. 1). The thermal resistance of 2.5°C/W at 150°C reflects the substrate thickness. The package mounts to the heatsink using a single bolt, which may actually warp the component away from the heatsink. Additional information on this type of problem is provided elsewhere.1
A Z-package resistor was made in the same manner as the TO-220 package. The idea behind this package is that the units fit together to conserve space on the printed-circuit (pc) board or heatsink. The resistor area was about 50% larger than the previous package, and the free-air power dissipation was proportionately higher at roughly 5 W.
The power levels obtained with a heatsink were also 50% higher. This unit reached a 150°C resistor temperature at 60 W (Fig. 2), versus 40 W for the TO-220 resistor. At 150°C, the thermal resistance measured 1.67°C/W, which reflects the sample size and substrate thickness.
To create a 1-kW planar resistor, a resistive spiral was printed on a 2- by 2- by 0.040-in. alumina substrate. The pattern was 270 squares long, and the resistance depended on the particular ink used (Fig. 3). The offset spirals were connected in the middle, making the resistive path long. This long path is good for reducing the voltage/length in surge applications, spreading heat over the substrate, and achieving larger resistance values.
Three different resistive materials were tried—a conventional RuO2-based thick-film material, a conventional silver conductor, and a resistive ink specifically designed for surge applications. The specific surge-resistor material used was the 4500 series EMCA-REMEX product with its recommended termination material. Typical resistance values were between 1 and 15 Ω. All samples were covered with a dielectric.
Before the units were tested, they were mounted to a heatsink. Since the substrate would be pressed very tightly against the heatsink, the heatsink had to be very flat. Fly-cutting the heatsink's surface achieved this flatness.
Heatsink design was important. A commercially available water-cooled heatsink (Aavid 416701) with widely spaced copper tubing and thin aluminum webbing between the tubing didn't adequately cool the package. There was a 150°C difference between the temperatures measured directly over the water pipe and at the midpoint of the aluminum webbing.
As an alternative, an experimental heatsink was created using a finned aluminum block (Aavid 66975) submerged in flowing tap water. A high thermal-conductivity (2.3 W/m-Kelvin) silicone paste was applied as an interface between the heatsink and the substrate. An aluminum top plate was placed over the assembly and bolted to the heatsink (Fig. 4). To isolate the resistor from the aluminum heatsink, a second alumina substrate was placed between the resistor substrate and the top plate. The interface was then filled with heatsink compound. Designers should note that electrical isolation requirements depend on the package's operating voltage. Creepage distances for various voltages are well known, and insulation must be designed accordingly.
Heat Transfer Optimized
The top plate supplied the required contact pressure and had provisions for leads to exit the package. High contact pressure and thermal heatsink compound were used to optimize the transfer of heat from the thick-film substrate to the heat spreader. Standard semiconductor practice provided a reference point in determining the proper amount of contact pressure.
Motorola offers an application note that stresses the importance of obtaining about 500 lb/in.2 of contact pressure to reduce the thermal resistance of the interface between heat source and heatsink to 0.1°C/W/in.2 The 1000-W example met this specification, achieving 0.025°C/W on a 2- by 2-in. substrate.
Electrical power was applied to the device, and the temperature was measured in the center of the package on the second or insulating alumina substrate. Power in each unit was increased until the part failed or reached 1300 W.
The above tests were applied to a conventional thick-film resistor sample, a silver-conductor sample, and a surge-resistor sample. The samples that survived the continuous power testing were put on surge testing.
The samples were connected to the heatsink as previously discussed and subjected to 104-J pulses. Pulses were supplied by charging a 52-µF capacitor to 2000 V and discharging through the samples at roughly 2-min. intervals. Temperature versus time was monitored to see how the package heated and dissipated heat under surge conditions.
Initially, samples failed at the interface of the conductor and the resistor. This problem was corrected by redesigning this connection. Since the sample resistances ranged between approximately 1 and 15 Ω, the current densities became large at maximum power. Any bottleneck in the current path resulted in a premature failure. Also, the initial spiral design was improved by increasing the width of the resistor trace.2 A wider resistor trace has more thermal contact with the substrate, and ultimately, the heatsink.
Samples made with conventional RuO2-based resistor material failed very quickly, and at relatively low power. A few survived to 800 W, but only for minutes. The oxides and glass phase in these resistors are poor heat conductors.
The next samples had spirals made with silver-conductor paste, and consequently had low resistance (0.7 Ù). From a materials standpoint, these samples should perform better. The low thermal-conductivity oxide phases in the conductor are balanced by the silver's high thermal conductivity. As expected, these samples outperformed the RuO2-based resistor material.
The silver-conductor samples operated at 1000 W for two hours without any change in properties. Because of the low resistance of these samples, the spirals were drawing roughly 40 A. The temperature of the alumina insulating substrate measured over time was 100°C (Fig. 5). Evaluation of these samples is ongoing with life testing, and the effects of long-term operation are under investigation.
Very good performance also was achieved by the samples made with surge-resistor material. The particular sample was powered to 1000 W, and the measured temperature on the insulating substrate reached about 120°C. When the sample power was increased to 1300 W, the temperature rose to approximately 140°C, as measured over the course of a four-hour test (Fig. 6). These tests have been extended to eight hours with identical results and additional life testing is under way.
The resistance of the surge-resistor samples before testing was 14.21 Ù, and it didn't change appreciably after testing. Surge-resistor material is designed with heat removal in mind, and these tests demonstrate the material's ability to conduct heat effectively. A low glass content and high palladium and silver content can account for this behavior.
The surge handling of these samples also was good. The samples dissipated the 104-J pulses effectively with no change in resistance. The temperature measured at the insulating alumina substrate rose just a few degrees above the heatsink temperature, but determining the exact value of the temperature excursions isn't easy because of the electronics' sampling rate. Nevertheless, temperature measurements for a sampling of the 100 pulses applied indicated that temperature excursions weren't too large (Fig. 7).
The results presented show that a power density of 250 W/in.2 is feasible if the package can remove the heat to a heatsink, and the heatsink can quickly dissipate the heat. For small devices with lower total power, a standard finned heatsink cooled by air may be adequate. Meanwhile, the heatsink may need to be water cooled at high power. The heatsink temperature is a function of heatsink design, fluid heat capacity, temperature, and flow rate.
Conventional RuO2-based resistor materials don't hold up well at power levels up to 1000 W. The low-resistance silver-conductor inks can withstand this power level, but whether they can support high current levels for extended periods of time needs to be addressed. Furthermore, the ultimate resistance that these materials can supply is typically very low.
The surge inks work well in continuous operation and surge applications. Surge inks also allow a wider range of resistance values than low-resistance inks, but the choices remain limited. Higher-resistance values require more insulating phase in the formulation, which hampers heat removal. *
The authors would like to thank Peter Bokalo of EMCA-REMEX for design suggestions and for providing the surge-resistor samples. The electrical testing by Steve Walzac also was much needed and appreciated.
- Motorola application note AN1040, "Mounting Considerations for Power Semiconductors." Prepared by Bill Roehr, staff consultant, Motorola Semiconductor Sector.
- D. Raesner, CTS Corporation; P. Bokalo, J. Robertson, EMCA-REMEX Products. "Design and Process Considerations for Thick Film Surge Resistors To Increase Reliability," Passive Components Symposium, CARTS, Nice, France, Oct 7-11, 1996, pp 3-10.