Assume your task is to design a dc-dc converter with a defined reliability — at minimum cost. To succeed, you would have to prove that you met or exceeded the reliability target. You can do this by predicting the overall system MTBF, based on known references. It's also important to collect field data to calculate the real-life MTBF.
To understand the challenges of a specific application, you should be aware of everything that can damage dc-dc converters. This includes temperature extremes or rates of change from low to high temperature. If the rates of change are too high, thermal coefficient of expansion problems can occur.
As dc-dc converter power densities increase, their available surface area for dissipating heat decreases. Therefore, the converter's efficiency and thermal management must be optimized to maintain high reliability. Good thermal management requires all converter surfaces to be at a uniform temperature without any hot spots.
Of course, high efficiency is an important goal; however, that doesn't mean the converter's performance is necessarily adequate for a practical application. The published efficiency is usually a typical value measured at 25°C ambient, at the nominal input voltage. As the temperature rises, losses for semiconductor (usually MOSFETs) and copper traces can rise dramatically. At low line, there's increased input current loss. At high-line switching, losses increase; thus, efficiency goes down. With this in mind, worst-case efficiency is more important than the quoted typical efficiency. Whereas efficiency values can assist in comparing similar-sized converters with equal voltage and current ratings, they aren't worth much if the size, output voltage, or the output currents are different.
One technique for dissipating heat while minimizing the impact on efficiency is to use an insulated metal substrate (IMS). As shown in Fig. 1, the IMS consists of a base layer, dielectric layer, and metal layer of either aluminum or copper. To achieve reliable heat transfer, all heat-generating components must be mounted on the IMS' metal layer. An aluminum sheet has high thermal conductivity: It dissipates four times more power than the equivalent FR-4 board. The sheet spreads heat evenly and eliminates hot spots. You can easily attach a heatsink to the IMS, which provides cooling for all the components in the power train. IMS characteristics include:
- Excellent flatness — 0.0015-in., guaranteeing good heat transfer to a heatsink.
- Ruggedness that improves mechanical integrity and durability.
- Low thermal impedance, providing effective cooling due to a greater surface area.
- High reliability — because normally cool components are not heated by conduction through the material.
- No mechanical stress, due to thermal coefficient of expansion mismatch, leading to higher reliability.
- No water entrapment; it's truly water washable.
- Easy to manufacture at relatively low cost.
Copper Core Boards
Compared with IMS construction, the single-board FR-4 material is a relatively poor thermal conductor. It offers no medium for heat distribution. So, unless something is done, there are high peak temperatures on the surface and inside power components.
Several steps can be taken to optimize the thermal performance of single-board dc-dc converters. One step is to use multiple power semiconductors (SO-8, and the like) to distribute low power heat sources across the board. A second step is to use a p. c. board copper core, shown in Fig. 2. Some quarter-brick and half-brick, single-board dc-dc converters use a multilayer board with a thick copper core. The effect is that heat dissipation across the area of the board is almost uniform. With this, every part of the board area can dissipate heat. This minimizes thermal resistance and temperature rise, and leads to lower component temperatures and higher reliability.
The Cast Carrier (Figs. 3a and 3b) consists of an aluminum casting enclosing the circuit board. It marries the metal plate heat spreader heatsink interface of the IMS with the copper core board. The copper core single board is used for low-profile, low-cost converters, and IMS construction is used for high-power density or more arduous applications.
Cast Carrier is the ultimate in thermal control. It requires a casting to mirror the skyline of the added components, and to provide pits for encapsulants. The combination of multiple heat sources, copper cored board, and cast aluminum carrier is state-of-the-art thermal designing for reliability within the dc-dc converter world.
Fig. 4, on page 38, shows a thermal photograph of a dc-dc converter that has uniform heat distribution.
Managing the thermal performance of a dc-dc converter enhances its reliability. Keep in mind that the following performance-oriented characteristics can also have an impact on reliability:
- Excessive voltage across the input, across the output, between input and output, between outputs, and between any pins — including control and monitoring pins, surges, spikes, and ESD.
- Excessive current at the input, output, and ESD.
- Excessive mechanical disturbances, including shock, vibration, and bump.
- Solid and liquid contaminants, including dust and dirt, water, diesel exhaust, and conductive particles.
- Chemicals used in soldering and washing the application.
- Magnetic fields, electric fields, mobile telephones, radio transmitters, high-power equipment.
In recent years, the tremendous improvement of semiconductors and implementation of better switching topologies have boosted efficiencies. This allowed simplified single board designs up to 60A. With the increasing power demands of telecom boards and the continuous reduction of output voltage levels, the required output currents of “bricks” has increased dramatically. Until recently, digital circuit boards needed 30A to 60A. The latest very large scale integration logic designs with 1.8V, 1.5V, and 1.2V supply voltages require 75A to 100A. Along with the continuous reduction of brick size, this will lead to very high current densities for dc-dc converters.
The high currents in the transformer windings and in the components made it difficult to use a single board approach, because it's difficult to avoid hot spots and resulting degradation of brick reliability.
At low power levels, dc-dc converters without a baseplate are common. Instead of a two-board construction, they use a single multilayer board — holding all the pins, transformer windings, and components. This concept has recently been applied to high current designs with headline ratings up to 60A. In most applications, these converters can't deliver 60A at the maximum ambient temperature and the least favorable input voltage. They will require heavy derating and/or airflow to keep these components below their operating limits.
Component failure rates generally decrease as applied stress levels decrease. Thus, derating or operating components at levels below their ratings (for current, voltage, power dissipation, temperature, etc.) will increase reliability. You can achieve this derating by circuit design (minimizing applied stress), component selection (using components with ratings well above the applied stress), and thermal design (reducing the component's operating temperature).
Most estimates of failure rate are based on an operating temperature, input voltage, and output power. Using the converter at an output power or current different from that used for the reliability calculation can yield a reliability different from that predicted. You can use this to your advantage by operating the converter at an output power less than the maximum permissible value, such as derating. The power dissipated in many internal components — including most of the high-power devices in the power conversion chain — is approximately proportional to the output power or current. This means that derating the output current by x% will reduce the component temperature rise by at least x% and by more for components such as MOSFETs and magnetics, where the power dissipation is proportional to the current squared. The lower power dissipation leads to a corresponding lower failure rate. Consequently, the power system designer can decrease the converter's failure rate by reducing its output power.
Many new converter designs use high-efficiency synchronous rectifiers, reducing the overall loss. Smaller losses allow a reduction in the derating, but in the race for “amps/in.2” this gain is immediately employed to provide increased output current or to reduce the converter size. This dictates that the new generation converters are not so different from traditional designs: They, too, need high-speed forced air-cooling and/or heatsinks and/or derating. Our newest converters, which employ a copper core board, allow less derating and a lower junction temperature for the MOSFETs.
A necessary procedure for manufacturing that enhances reliability is burn-in. Burn-in keeps infant mortality in the factory, rather than allowing it in the field. This can be done at the part, board, or system level. All dc-dc converters should go through a burn-in process. Failure rates and times need to be recorded and analyzed to ensure the burn-in period is long enough to bring out all cases of infant mortality.
If you can't achieve the required level of predicted reliability solely by derating and thermal management of one dc-dc converter, then you can add more converters in parallel. Although more components are present to fail, the consequence of one failure may be completely overcome, requiring the system to have two failures before compromising functionality.
In today's competitive market, a commitment to product quality and reliability is a necessity. Customers have high expectations for the reliability of the products they buy, and the companies that don't meet those expectations fail. You already know the advantages of building reliable products: reputation grows, costs shrink. The most successful companies meet these market demands for quality by using design for reliability principles. You should be able to integrate reliability considerations into the entire product design process, right from the start. That way, you design reliability into the product, not patched on later, when problems arise. Companies practicing design for reliability find that it results in fewer design changes and iterations, lower manufacturing costs, lower warranty and service costs, more profit, and happy customers. An important element of the design for reliability process is reliability prediction, which allows you to predict product failure rates.
Reliability predictions provide a quantitative basis for evaluating product reliability. You can use the information from a reliability prediction to guide design decisions throughout the development cycle. When an initial design concept is proposed, a reliability prediction can indicate the design feasibility from a reliability standpoint. Even though these early stage predictions are based on limited design information and are approximate at best, they can give direction to your design decisions, which may be critical to product success. For example, you might have a requirement of a 200,000 hr MTBF for a product. If your predicted value is 35,000 hr, the current design concept may not be feasible. At this point, you can modify the design concept or revise the requirement. A predicted value of 500,000 hr can give you confidence in your design concept, at least as far as reliability is concerned.
You can use various reliability prediction techniques, depending on your knowledge of your design. You can make an early estimate by comparing your product with products of similar function or complexity, of known reliability. Generally, this will be a crude estimate at best, as the many differences in design details between the products are not accounted for. As more details of your design are known, more accurate methods become available. These methods use part failure rate models, which predict the failure rates of parts based on various part parameters, such as technology, complexity, package type, quality level, and stress levels. The best-known failure rate prediction method in communications is Bellcore (now called Telcordia Technologies). This handbook offers documented procedures for predicting electronic product reliability, providing a standard basis for comparing reliability numbers.
Predictive methods attempt to predict the reliability of a part based on some model typically developed through empirical studies and/or testing. An attempt is made to identify critical variables such as materials, application environmental and mechanical stresses, application performance requirements, duty cycle and manufacturing techniques. Typically, a base failure rate for the component is assigned, and this is multiplied by factors for each critical variable identified. An example of this method is Bellcore TR-332. Using this standard, reliability can be predicted early (in fact, before parts are made) and effort is minimal. The predictive models assume a constant failure rate over the lifetime of a product. This ignores higher failure rates typically seen at the beginning and end of component life, infant mortality, and wear-out, respectively. Predictive methods can provide a relatively accurate reliability estimate in cases where good studies have been done to analyze field failures.
Predictive methods are also useful in providing a relative ranking of reliability between alternative designs, but the absolute reliability numbers (or failure rates) obtained with these methods will rarely be indicative of real-life performance. This is the best way to demonstrate the real reliability level of a product. It requires a statistically significant population in the field, and a reasonably long time in the field.
For more information on this article, CIRCLE 333 on Reader Service Card