Avoiding the pitfalls of DC power supply designs

Paul Horner, MD of Advance Product Services Ltd talks about how his experiences leading a power supply repair house has enabled their engineering team to offer ultra reliable bespoke DC power supply designs and how you can avoid some of the pitfalls.

It's interesting. Being involved with switch mode power supply design for over 40 years, you learn what works and what doesn't. What once seemed like quite complex theory becomes second nature and you instinctively have a feel for what is required to make a design work.

But working in the lab as a prototype is one thing. Working faultlessly for the next 20 years is quite another. It's feedback that the design engineer seldom has the opportunity to benefit from. You rarely see the end application, let alone the condition of the components after years of operation in an industrial environment.

This is probably of little concern to the glut of far eastern manufacturers with products at throw-away prices. But there are still plenty of applications where long term reliability and build quality are paramount, and this remains a stronghold of UK design and manufacturing. Let's face it, if you are manufacturing in the UK and you are not focusing on quality - you are dead in the water.

Of course, there are a host of general parameters that effect long term reliability of a power supply. Fundamental circuit design, component selection, mechanical construction, assembly process, storage and handling all play a big role. However these tend to be well appreciated at the design and manufacturing stages.

Looking at things from the service return side gives the engineer an entirely new perspective. It allows a unique appreciation of what, in practice causes power supplies to fail in the field.

And it's not always obvious.
 
Electrolytic Capacitors

The drying out of wet electrolytic capacitors is perhaps one the most widely recognised causes of age related failure, and it is certainly prevalent. Modern demands for ever decreasing can sizes result in thinner dielectric materials and less volume of electrolyte. Although the loss of electrolyte is by some means the natural wear out mechanism, it can be slowed considerably by reducing the core operating temperature of the capacitor.

Locating caps away from other high dissipation components is one obvious example, but the core temperature is also very much influenced by the ripple current flowing through the ESR (equivalent series resistance), namely the electrolyte. A typical 105°C rated capacitor has a ripple current rating in a 105°C ambient, giving a core temperature of approx. 115°C. The specified load life under these conditions can be as low as 1,000 hours (42days), although in practise most caps will continue to operate for longer than this, albeit with reduced capacitance and or higher ESR.

Most practical applications do not subject passive components to more than 50°C, so it can be tempting to increase the ripple current above the rated maximum. This is not recommended because the temperature rise is proportional to the square of the ripple current multiplied by the ESR. Because ESR increases with time, end of life failure will occur sooner than for a cap operating at 105°C and maximum rated ripple current.

Output capacitors on small 'flyback' power supplies, operating in the discontinuous current mode are especially vulnerable to early failure due to the large ripple currents inherent in this topology, so they need specifying carefully. By comparison, continuous current flyback and 'forward' converters have typically a 20% peak to peak current ripple compared to the 100% of the discontinuous mode flyback converter.

Conversely, small (approximately 6 x 12mm) electrolytic caps commonly used in power supply control circuitry can cause problems in high local ambient temperatures, even when run at very small ripple currents. These capacitors are often used in conjunction with a high resistance connected to a HT rail to provide a start-up supply to the control circuit. Due to the very small amount of electrolyte they contain, they can dry out before any other component fails and prevent the power supply starting at turn on due to high impedance or current leakage. Often this can go completely unnoticed until the first mains blackout and subsequent restart attempt.

Careful electrolytic capacitor selection is becoming increasingly important as more and more far-eastern manufactured components enter the market and it is important to pay a good deal of attention to the detailed specification of such components. Cutting costs by using inferior capacitors is rarely money well saved when it results in a dramatic reduction in service life, potentially high warranty costs and a blemished reputation. Better cooling, larger capacitors or solid electrolyte capacitors are alternative solutions. Niobium solid electrolyte caps are a cheaper alternative to tantalum caps, which are becoming more expensive as tantalum reserves diminish. If you must use wet electrolytics, as most of us do, ignore ripple current at your peril!
 
Film Capacitors

It is not often appreciated that the ac rms voltage rating of film capacitors must be greatly de-rated for frequencies above approximately 1kHz. A popular cap, rated at 400Vdc & 250Vac is specified at just 1Vac maximum at 100kHz, so it can be easy to exceed the high frequency ac voltage rating in a power circuit. The image below is the result of exceeding the HF ac voltage rating of a 470nF 250Vac cap.

Film caps are also vulnerable to failure as a result of exceeding the repetitive rate of change of voltage (dV/dt). Metallised polyester snubber caps across switching semiconductors have been found to fail due to excessive dV/dt, where the use of polypropylene, ceramic or foil film would have been preferable.
 
Surface Mount Multilayer Ceramic Capacitors (SMD MLC)

The larger sizes (1812, 2220) of SMD multilayer ceramic caps are prone to failure when mounted on fibre glass or composite PCBs due to the different coefficients of thermal expansion of the cap and the substrate. These components can fail short circuit with devastating consequences if they are connected across a power rail.

All sizes, but more especially the larger ones, are prone to failure due to mechanical stress. An example encountered recently used several SMD MLC caps under a dc power output screw terminal block which was subject to flex whenever the terminal was pressed down by tightening or loosening the screws. The subsequent fracture of the cap burnt a hole right through the pcb due to the fact that the ceramic cap body remains mostly intact even when red hot.

Avoid large SMD MLC caps on circuit boards involved with intense thermal cycling unless the substrates are matched, and never place in areas of mechanical flex or stress.
 
Power MOSFETs

Generally speaking, power semiconductors are among the group of components least prone to ageing effects. Assuming they are used within their maximum ratings and are well thermally managed they are very reliable. However they account for more than half of all service return failures.

Typically this is because their maximum ratings have been exceeded through knock-on effects of other component failures, poor circuit design, environmental influences such as spike or surge, over-temperature or mechanical stress.

In terms of the circuit design however, there are subtleties that can contribute to a surprisingly large proportion of failures which tend to be are far less well appreciated:

Problems can occur in MOSFETS where a high rate of rise of drain to source voltage (dVds/dt) causes capacitive charging of the FET gate. This can switch the FET back on while it is turning off - usually a destructive event! This is especially problematic where the "off" drive connects the gate to a voltage slightly above zero, rather than to a negative potential. A negative drive holds the gate well below the threshold voltage as the drain-source cap charges and generally provides a much more robust solution. It should be noted that the gate threshold voltage typically reduces to less than 70% of its 25°C value at maximum junction temperature.

A high dVds/dt can also cause the parasitic transistor (present in the construction of all FET devices) to turn on, especially at high temperature where more thermally generated minority carriers exist within it. If the body diode of the FET is used to clamp the drain to source voltage (as in a zero voltage switching 'ZVS' resonant converter), its reverse recovery time can be very long. This is due to the FET body diode only being moderately fast and the fact that the reverse voltage is only the "on" voltage the FET, typically around 1V. As the body diode is in fact the collector-base junction of the parasitic transistor, the unrecovered charge carriers cause the parasitic transistor to turn on when Vds rises rapidly, allowing large currents to flow in the device. To make matters worse, the diode recovery time is even longer at higher temperatures.

There is a final scenario which sounds like it has come straight from science fiction! It is known as Single Event Burnout (SEB). SEB research carried out as long ago as 1996 showed that a high voltage MOSFET, biased off, supporting a voltage near to its maximum rating can suffer an avalanche failure caused by a single sub atomic particle colliding with a silicon nucleus. Subsequent research has shown that even at ground level, neutrons from cosmic ray collisions in the upper atmosphere can cause random failures in high voltage MOSFETs over and above the rate predicted by MTBF data from manufacturer's life tests. Reducing the maximum Vds by even 6% has been shown to decrease SEB failures by an order of magnitude.

All the above scenarios are exacerbated by high junction temperature, so there is much to be gained from running FETS well below their maximum temperature and voltage ratings, and careful consideration is needed if the body diode is utilised in the application.
 
Optocoupler Ageing

Most designers have a good appreciation of electrolytic capacitor ageing, but we also see many age related failures due to optocouplers. Generally this manifests itself as a reduction in the effective current transfer ratio (CTR) over time. This doesn't sound too serious until you recognise that optos are commonly used to enable the converter stage of a power supply across a primary to secondary isolation barrier. A degraded opto can and often does render the entire power supply inoperable and as such can be considered a high failure risk.

The primary piece parts of an optocoupler are a photo-detector IC and an infrared emitting LED (typically Gallium Arsenide). Experimental analysis has shown that the LED is the only portion of the optocoupler that has a significant impact on life, with light output degradation leading to a decrease in CTR. Furthermore, it is the actual current through the LED which is by far the most dominant factor.

For longest possible service life therefore, it is desirable to allow at least 50% margin for a reduction in CTR over time and to drive the LED at as low a current as possible for the required CTR.
 
Spike & Surge

The majority of engineers are aware of the catastrophic effects of high transient energies on the input and output lines of a power supply. Indeed, voltage fluctuations on the local grid are commonplace and the variance in the quality of the AC mains from location to location can be surprisingly large. However, a typical power supply which meets EN 61000-4-5 (basic immunity test for surge) does not guarantee low susceptibility in the field.

The financial rewards of producing reliable products over and above the basic EMC standard are usually very worthwhile. A certain UK manufacturer saw their warranty costs fall by £2.7million per year after spending less than £100K on improved immunity.

In the UK at least, a relatively small proportion of energy fluctuations on the grid originate from lightning strikes. Contrary to popular belief it is not a direct strike which causes the most problems, but the voltage induced on overhead lines from the magnetic field of indirect strikes. Some of the largest discharges have been confirmed at >200,000A and there are often several discharges per strike. You could probably even measure some transient voltage induced in a paper clip lying on your desk within 500m of a storm if you were quick enough with the scope!

Buildings in Europe whose AC power is carried by overhead wires can reckon on having 80-120 surges every year due to lightning. These are typically limited to around 6kV because the standard domestic style mains socket flashes over at the rear connectors at around this voltage and acts like a spark gap suppressor! Industrial premises with only 3ph supplies can see much more. A modest strike of 15,000A would induce around 10kV on a transmission line 150m away (even when buried in the ground). Heavy industrial switchgear, large photocopiers & laser printers, HVAC systems, electric motors and thyristor devices are all notorious for imposing spike and surge on transmission lines, and not always at lower energies than lightning strikes.

It is not widely appreciated that even if such transient energies do not cause instant catastrophic failure, repeated exposure has a proven degenerative effect, particularly with highly integrated silicon devices. Call it transient ageing if you will. It has a significant impact on long term reliability.

In all cases, a well-considered surge protection stage is essential but is often overlooked or poorly optimised. Indeed, there are a great many variables to consider and not every engineer appreciates the subtleties of the various protection devices available.

Looking specifically at the input of an AC-DC power supply, it is desirable to place surge protection devices in both the line-to-line and line to earth positions, giving both common and differential mode protection. Metal oxide varistors (MOV's) or VDR's, are the most commonly used device in low-cost applications. However, a MOV may not be able to successfully limit a very large surge from an event such as a lightning strike where the energy involved is many orders of magnitude greater than it can handle. We have seen many designs where the power supply has a scattering of varistors on the input with no sacrificial protection (e.g. a dedicated thermal fuse). The result is that the first high energy surge to arrive either causes the varistors to explode, often accompanied by a large plasma discharge which destroys everything else in the vicinity, or the main input fuse to blow. Either way the power supply fails and has to be returned for service the same way as if there were no protection fitted at all!

An important characteristic to consider with MOV's is that they degrade when exposed to a few large transients or many smaller ones. As they degrade, their trigger voltage falls lower and lower, ultimately leading to thermal runaway of that particular device. Therefore to ensure good long term reliability, correct voltage rating is essential. It is also worth noting that selecting a device with a higher energy (joule) rating typically increases the life expectancy exponentially. 

It is common to see multiple MOV's in parallel to increase the overall joule rating of the network, however unless specifically matched sets are used, each MOV will have a slightly different non-linear response when exposed to the same overvoltage. This invariably leads to current hogging and premature failure of the individual device.

Thus the 'effective' surge energy of the network is dependent on the MOV with the lowest clamping voltage, and the additional parallel MOV's do not provide any benefit. Furthermore, because each MOV has a relatively high leakage current (typically around 0.5mA at working voltage for a 20mm device), using many devices in parallel can lead to unacceptably high earth leakage currents.

The other two devices commonly used in protection networks are transient voltage suppressor diodes (commonly referred to as Transorbs and also sold under the name Transil) and gas filled discharge tubes (GDT's). Whereas the practical response times of MOV's are in the 40-60ns range, suppressor diodes respond to spikes within 1 - 10 pico-seconds, mostly limited by the inductance of the connecting circuitry. This makes diodes ideal for suppressing sub-nanosecond spikes generated by the many thyristor controlled devices sat on the mains supply. Sub-nanosecond spikes show up, do their damage, and are gone before MOV's even notice.

Diodes also have the added benefit that they do not degrade with repeated surges which means they can be selected with clamping voltages much nearer to the AC working voltage than with MOV's. The disadvantage of suppressor diodes is that they offer a lower 'cost/energy handling' ratio in comparison to other devices and they tend to be physically larger for the same energy rating.

However, if space and cost are not critical, they are one of the most effective devices available for suppressing fast energy transients.

Gas discharge tubes consist of two electrodes surrounded by a special gas mixture in a sealed glass or ceramic enclosure. The gas is ionized by a high voltage spike which causes an arc to form between the electrodes and current to flow. GDT's can conduct more current for their size compared to diodes and MOV's but are crucially different in that they continue to conduct until the source voltage has dropped close to zero. This has huge implications for DC and indeed has to be considered carefully for AC whereby it is quite possible to have a full half cycle of mains energy to absorb in addition to the initial spike or surge energy. It is critical therefore that this follow-on current is controlled.

Like MOV's, gas discharge tubes have a finite life and can only handle a few very large transients. The typical failure mode is a modified trigger voltage or, if subject to very high energies, a dead short. GDT's take a relatively long time to trigger; 100nS pulses 500v above rated voltage will often be completely unsuppressed. However, gas discharge tubes offer the highest energy handling capabilities of all protection devices and have exceptionally low capacitance.

By far the most effective suppression networks utilise a combination of components to give high energy, high current capability with a very fast response time. Parallel devices are to be avoided unless using specifically matched sets and thermally vulnerable devices must be protected by dedicated components.

Any design which neglects a well optimised surge and spike suppression network can expect substantially increased failures rates in the field.

Paul Horner is Managing Director at Advance Product Services Ltd.