|Elliott Sound Products||Semiconductor Safe Operating Area|
Safe Operating Area (SOA) for semiconductors is a little understood topic. Although the chart is generally provided in the data sheet, there is a great deal you need to know to be able to make proper use of it. Without a thorough understanding of the loudspeaker load, instantaneous voltage and currents, and what happens to transistors if the SOA is exceeded, it is easy to imagine that the supply voltage for an amplifier can be increased up to the maximum voltage allowed by the transistors used.
This is not the case at all, and this article discusses the problems faced in any amplifier design to create a reliable circuit that (ideally) can never place the power devices at risk. This is much easier said than done, unfortunately.
I suggest that the reader also has a look at Short Circuit Protection - Testing amplifiers to the limits, because these two articles cover the same topic, but from very different perspectives. Here we look at how output devices are protected, but the VI Limiting article explains what can happen if the designer gets it wrong (amongst other things).
Along similar lines, the article on The Design Of Heatsinks ties in with this topic, because safe operating area (SOA) and temperature are very much interdependent. It doesn't matter how good the transistors might be, if they aren't cooled properly then they will fail. Some aspects of this are covered below, but the Heatsinks article has a lot more detail.
Most basic analysis of a power amplifier design is done (at least initially) using circuit simulation and basic theory. None of this is at all difficult, and is essentially a matter of current analysis through the amplifier into the load. For the sake of simplicity, a resistive load is generally used for all but the most rigorous analysis, and for low powered amplifiers, this is quite sufficient.
When you make (for example) a 20 Watt amplifier using discrete components, most of the power transistors available have so much reserve current and power available that few problems will be encountered. Even a 100W amp is not a problem if the impedance is known in advance, and reasonable care will give a reliable circuit.
The problem is that real life loads are neither predictable nor reasonable, with nominal* 8 ohm loads perhaps plunging to 3 ohms or less at some frequencies, and soaring to 50 ohms or more at loudspeaker driver resonance. Four ohm loads are no better, and 2 ohm loads are a nightmare for most amps. However, as load impedance falls, speaker cable resistance can make the difference between amplifiers surviving or failing. The extra resistance may mitigate (to some extent) the inductive part of the load, but at the expense of wasted power.
One of the problems is that music is also unpredictable. Some music has a very low 'crest factor' (the ratio, in dB, between the average and peak power), so relatively high power levels are present on a more or less constant basis. Other music has a high crest factor, with a peak to average ratio of up to 20dB (a power ratio of 100:1). Classical recordings are commonly thought to have a high crest factor, but this is not always the case, with some having as little as 6dB. Yes, this is uncommon, but it can (and does) happen).
'Modern' music (a term that has a different meaning to everyone) is not immune from high crest factors, but they are less common than in unprocessed orchestral recordings (for example). A great deal of the material that you may listen to has a very low crest factor, so there is little difference between the peak and average power.
It is the combination of unpredictable loads, very different musical styles and power demands, different listening preferences and (we must never forget this one!) ... heat, that can spell doom for even the best designed amplifier if it is used outside of the original design parameters.
* Nominal - existing in name only. As used in electronics, 'nominal' refers to the expected voltage, power, impedance, etc., under 'normal' (or sometimes idealised) conditions.
So, what are the factors that determine if an amp will be reliable or an owner's nightmare? There are quite a few, and as an example I will use the P3A 60/100W amp design from The Audio Pages projects section.
The primary objective is to produce a design that has sufficient power to suit the needs of the owner - this is a very difficult, because of the vast differences in loudspeaker efficiency, preferred listening level, and type of music. Nonetheless, 70W is not an unreasonable figure, and is well suited for many speaker systems. Smaller 2-way units are very popular because of their relatively high spousal acceptance factor, and they are convenient, reasonably priced and can give very good performance.
Such systems simply will not take the full continuous power of a 250W amplifier. The sales blurb may claim they are suited to amps of '20-200W', but this often assumes a 'typical' crest factor of around 10dB (10:1), where the peak power will be 10 times the average, or in some cases even more.
This has everything to do with the amplifier design, as it sets a reasonable expectation of the power needed, and sets us towards an understanding of the load the amp is expected to drive. A quick analysis of any 2-way or 3-way speaker will show that the impedance is far from flat, it has peaks and dips at various frequencies, and will only show the nominal impedance at a few frequencies.
Again, for the purpose of explanation, I must choose a speaker system, and the one used for the remainder of this article is completely imaginary. It exists in simulation only, but has a reasonably close resemblance to many small/medium sized 2-way loudspeakers. The reason for a simulated speaker is simple ... the effects of the impedance variations are easily seen, and are actually very similar to a 'real' speaker.
The primary requirement for obtaining power is voltage swing. This in turn is determined by the supply voltage, and the supply voltage and lowest impedance determines the maximum current.
Using the (nominal) ±35V supply for P3A as the example, we must accept that even for an 8 ohm loudspeaker, the minimum impedance will be lower than claimed. Six ohms is a realistic figure (assuming a well engineered speaker), but it could be less. A 4 ohm speaker can be expected to have a minimum impedance of around 3 ohms. The worst case peak dissipation of an amplifier running into a nominal 4 ohm load (3 ohms in series with 470uH for the reactive case) is a little under 180W. See the example below to see how much difference a slightly higher supply voltage can make!
Power transistors are a single pair (NPN and PNP), rated at 200W dissipation. They have adequate voltage and current capabilities and are the recommended devices for the P3A amplifier. These will survive with a ±35V supply and 'normal' hi-fi duty (the amp's design goal), but are at serious risk at higher voltages.
For the 6 ohm case, peak current will be 5.8A, or 11.6A into 3 ohms. This is with a +/-35V supply, but the transformer voltage is always quoted at full (resistive) load, so typically with normal mains voltage, the supplies can be expected to be ±38V or so with no load. Large filter caps will hold this voltage for several milliseconds, so the maximum peak currents are probably closer to 6A and 12A for 8 and 4 ohm nominal loads (respectively).
When consistent high current is drawn from a normal unregulated supply, that will cause the voltage to fall. Often, this makes the difference between survival and failure when the amp is used at high power into a difficult load.
Under ideal conditions, a transistor's power dissipation rating refers to the maximum peak power that the device can handle, with the case temperature at 25° C. At any case temperature above 25°, the power is derated (reduced) linearly, until it reaches zero at around 150° C. In some datasheets, you will see that they refer to junction temperature, rather than case temperature. Regardless to the terminology used, it is the maximum permissible temperature of the silicon die (the semiconductor junctions) that is the limiting factor. Very few transistors are designed to operate with a junction temperature above 150°C.
Using the following graph as an example, you can see that maximum dissipation (230W for the transistor shown) is 230W at a case temperature of 25°C. It falls to zero at 150°C, so must be derated by 1.84W/°C above 25°C ...
Derating Factor = Max Dissipation at 25°C / 125° - where 125° is simply the difference between the maximum temperature (150°C) and 25°C
Derating = 230 / 125 = 1.84 W/ °C
The derating factor can be determined for any semiconductor using the same method. A very few devices may be rated for higher maximum case temperatures, so simply adjust the formula to suit. For example, some MOSFETs may allow a maximum case or junction temperature of 175°C, but in general I wouldn't exceed 150°C regardless of claims. Note that although the datasheets refer to case temperature, it is all referred to the junction/ die. At maximum rated dissipation (230W for the device shown) and a case temperature of 25°C, the die will be at 150°C !
In fact, you can pick any point on the graph, and the die temperature will always be 150°C. For example, at a case temperature of 75°C and with 140W dissipation, the die will again be at 150°C. The only way you can reduce the die temperature is to keep dissipation and/ or case temperatures below the red line in the graph.
Figure 2.1 - Typical Power Derating Curve
Based on this, it is obvious that keeping the temperature down is critical, since elevated temperatures reduce the available dissipation, and reduce any safety margin that has been incorporated into the design. High temperatures also affect device life, and the hotter a transistor runs, the shorter its expected life.
To place the thermal issues into perspective, there is a calculation in the next section that shows average transistor dissipation to be about 40W (42.5W actually), when driving a 4 ohm reactive load at full power using ±35V supplies. For the sake of simplicity, we could assume about half that to be the continuous average with music at the highest level before clipping. 20W continuous does not sound like very much, but the thermal resistance from junction to air (per transistor) might be around 5.5°C/ W ...
Rth(j-c) = 0.54°C/ W (junction to case, assume 0.5°C/ W for simplicity)
Rth(c-h) = 1°C/ W (case to heatsink - a very good figure, and difficult to achieve in practice)
Rth(h-a) = 4°C/ W (heatsink to ambient, based on a 1°C/ W heatsink and 4 transistors [two amps])
... so the die temperature rise is 110°C. Now, add the ambient temperature - say 25°C - but it could be a lot higher!. The die temperature is therefore 135°C, so the case temperature will be 10°C cooler (based on 0.5°C/ W j-c), or about 125°C. At that temperature, the continuous allowable power dissipation is reduced to about 45W, but when the instantaneous safe operating area is considered (see below) we are too close to the thermal limit of the transistors - even a minor obstruction over the heatsink could be enough to tip the balance. Remember that the goal is to keep the transistor die as cool as possible, not to try to get the most power from a device that we can without killing it outright.
There is a lot more than just simple resistive load dissipation though. Figure 2.2 shows the power in a transistor driving a resistive load at the onset of clipping. As you can see, the power increases until the voltage reaches the halfway point between zero volts and full supply. After that, it goes down again - in a perfect amplifier, dissipation will fall to almost zero at the clipping point. For Figure 2.2, the applied voltage was ±35V, with a 3 ohm resistive load.
Figure 2.2 - Transistor Power Dissipation
The peak power is (using 6 ohms and 35V supplies) (35/2)² / 6 = 51W, or 102W for a 3 ohm load. Average dissipation is difficult to calculate because of the waveform, but my simulator tells me that it is 15 and 30 watts respectively. With music signals in real life, it is extremely hard to calculate the average, and it's simpler to measure the heatsink temperature rise and work backwards. Note that for the simulations, zero quiescent current has been assumed - in real circuits, this just adds to the average dissipation.
Where things rapidly get out of hand is with the loudspeaker load - it is not resistive (or even close to resistive) for 99% of all loudspeakers. The impedance and phase angle of a loudspeaker varies, and as phase angle changes from zero degrees (voltage vs. current), dissipation increases further.
For the 3 ohm case, a reactive load (at 45° phase angle) can be simulated by using a 470uH inductor in series with the 3 ohm load (for a frequency of 1kHz). With this combination, the peak transistor dissipation is almost 200W, with an average of about 43W - note that the peak transistor power has doubled, and the average has increased by 1.414. Of particular interest is that the maximum power occurs at the voltage zero crossing point, when the maximum voltage is across the device. This is what causes transistors to fail, and the higher the voltage, the greater the risk.
Figure 2.3 - Voltage, Current and Power Dissipation
While average power is well within the maximum ratings, the peak has reached the maximum device power, and we are now constrained by the SOA of the devices. Remember, this is with a supply voltage of +/-35V - higher voltages will create higher peak powers with real loudspeaker loads! The diagram above is based on a rather simple (but still extremely useful) series connection of a 3 ohm resistor and 470uH inductor. A real speaker system is far more complex (and harder to analyse), but it still follows the same rules. Of course, all variations we see are highly frequency-dependent. The amplifier 'sees' a very complex load and signal, but its job is to provide the required voltage and current at any instant in time - no more and no less.
Figure 2.4 - Simulated Loudspeaker System
The combination of resistors, capacitors and inductors simulates the 2 drivers in the system, along with their crossover networks. For simplicity, no impedance correction networks have been included, and the loudspeaker is vented (note the double low frequency peaks shown in Fig. 2.5). This is a typical response, but remember that this is only the electrical response of the system - acoustically, it might be good, bad or indifferent (the electrical response gives clues, but cannot be used to predict the acoustical performance - I would expect it to be very ordinary however, based on the lack of impedance correction that is clearly visible looking at the phase angle and impedance curves).
Figure 2.5 - Impedance and Phase Response Of Simulated Speaker
While the impedance is more or less as expected, the phase is another matter. At a phase angle of other than zero, the voltage and current are not simultaneous - the current may occur before the voltage (leading phase, capacitive load) or after the voltage (lagging phase, inductive load).
This is a major problem for amplifier designs, since at any phase other than zero, the power delivered to the load decreases, while the transistor dissipation increases. At 45°, peak transistor dissipation doubles, and power into the load is halved. This is worst-case operation, and is the point where transistor dissipation is at its highest.
As the impedance rises with increasing frequency, the load appears as an inductor, and when it falls with increasing frequency, it is capacitive. Note how little time is spent at zero degrees phase shift! This means that at nearly all frequencies in the spectrum, the amplifier sees not a resistive load, but a highly reactive (and variable) load. A significant part of the frequency range has over 30° phase shift, and the amplifier will be working nearly twice as hard as you thought it would.
A bipolar junction transistor (BJT) has a negative temperature coefficient. As temperature rises, the junction voltage falls, and gain increases. Transistors are not perfect - there are always minute flaws in the fabrication, causing tiny variations in the characteristics of different parts of the transistor die.
Modern fabrication techniques have minimised these to a huge extent, but they still exist. Even the resistance of the conductive layers within the device becomes very significant at high currents, so perfect current distribution cannot happen.
Now, there is a sequence of events than can (and does) occur within the transistor. If the instantaneous power dissipation is too high, parts of the transistor die will get hotter than others. This means that the junction voltage falls, and the gain increases - but only at the most sensitive part(s) of the die. If Vbe (base to emitter voltage) falls and gain increases - at one spot in the transistor - it will naturally take more of the current, which means it gets hotter, so it takes even more of the current (and so on). This can happen in a few milliseconds! That part of the transistor will quickly exceed the maximum permissible temperature, and the transistor will short-circuit internally.
All of this has happened in perhaps 10 milliseconds, and the case is not even warm. This phenomenon is called 'second breakdown' (or secondary breakdown), and is the single greatest reason for transistor failure in a working circuit.
Data sheets usually have a full set of graphs and charts, showing the various device parameters as a function of voltage, current and frequency. In the design phase, all are important, but the most important of all are the two that are most often overlooked by hobbyists and experimenters - thermal derating and safe operating area.
From the data sheet for the MJL4281A, Fig. 3.1 shows the SOA curve for these devices. Non-repetitive peak currents of up to 30A are permissible for 10ms, but only for collector voltages up to 30V, and only with the junction temperature at 25 degrees. This is a peak power of 300W (the device rating is 230W), but it must be stressed that these conditions must not be allowed to continue beyond the time specified - 10ms is not very long!
Figure 3.1 - SOA Curves for MJL4281A/4302A
If the time is extended, then the peak current is reduced for a given voltage, and for 1 second, the maximum rated current (15A) may only be drawn at collector-emitter voltages below 15V. This region is limited by the maximum rated current of the transistor, and will never allow continuous operation at maximum power. Remember thermal derating? This is where it must be applied.
So far, all this looks pretty good if you look at it in conjunction with the demands outlined above, and it even looks as if it would be safe with 4 ohm loads at greater than rated ±35V. Appearances can be deceptive though! Remember that all peak currents and power dissipations referred to were for a junction temperature of 25 degrees - no transistor can maintain that temperature in real life, since there is thermal resistance between the die and case, and further thermal resistance between case and heatsink. See Heatsink Design for more information on thermal resistance and heatsinking of transistors.
The devices must be derated by 1.84 W / °C above 25° (see Fig 2.1), which gives zero dissipation at 150° C. The thermal resistance from junction to ambient air (via the case, insulating washer and heatsink) can be expected to be around 1.5-2° C/W (for a big heatsink), so all dissipation limits quoted can be expected to be as little as 1/2 of those shown in the specifications.
That means that the 230W transistor is really only capable of a dissipation of around 120W at typical (relatively high) operating temperatures. As a result, at ±35V with a 3 ohm resistive + 3 ohm reactive load (representing a typical 4 ohm speaker either side of resonance), the maximum limits will be exceeded with a continuous (steady state) load!
Although this is completely true, in reality there are two things that will ensure that the amp remains functional (for many years) - the nature of music itself, and the collapse of the power supply under sustained load. However, continuous operation at full power into a reactance that gives a 45° phase angle will cause the amp to fail, even with ±35V supply rails.
The variable nature of music, where the frequency and instantaneous amplitude are continually changing, means that potentially destructive signals do not last long enough to cause a problem, however increasing the supply voltage or reducing the load impedance further will almost certainly cause device failure. As you can see from the chart, brief excursions into the 'unsafe' area are permissible - look at the 100ms and 10ms limits.
Likewise, the bigger the heatsink, the better. The thermal resistances that cause the semiconductor die to operate at a much higher temperature than you may expect are the limiting parts of the equation - and they cannot be eliminated - at least not sensibly. It is generally considered uneconomical to provide a refrigeration system to keep the transistor temperature at low enough temperatures to avoid problems.
Ultimately, all bipolar transistors will experience second breakdown if pushed too hard. The SOA curves for the transistors you plan to use must be examined carefully, and the design must avoid the second breakdown area (shown in Figure 3.1) at all times. Even a brief excursion into this prohibited area can cause instantaneous failure. You can see that the second breakdown region is non-linear for the MJL4281/4302 devices, and like all bipolar transistors they become more susceptible to second breakdown as the voltage across the transistor increases.
This is common, but may not be visible on the SOA graph for some devices. There are protection circuits that include multiple 'break points' to accommodate the discontinuities in the SOA curve, and this is appropriate if the output devices (and in some cases the driver transistors as well) are being pushed to their limits. In most cases this should not be necessary and can only be done successfully if thermal compensation is used as well - mostly, it is not!
While there are some who claim that there are 'rules of thumb' that can be applied, I disagree because the characteristics of transistors are different depending on type and manufacturers' optimisation techniques. Each case should be looked at on its own merits, and the datasheet examined carefully to be certain. From the speaker impedance and phase charts shown above, you can get a good idea of the peak power that the transistors will be subjected to. 45° phase shift is the worst case, because with greater phase displacement the power delivered to the load is reduced, and so is transistor dissipation. With a speaker load, there is always a resistance in series with the reactive components, and that limits the maximum transistor dissipation.
Having discounted the idea of any 'rules-of-thumb', I'm going to give you one anyway . Let's assume that you want to deliver 100W into 8 ohms, so you need a power supply with ±42V rails (I'm going to ignore losses here). The amp must also be able to drive nominal 4 ohm loads, so expect the minimum impedance to be 3 ohms. Worst case (resistive load) dissipation is therefore ...
I = V / 2 / R = 21 / 3 = 7 Amps
P = V / 2 * I = 21 * 8 = 168 Watts (peak)
This accounts for the resistive part of the load, and as we saw above, the reactive part of the load causes dissipation to double. Just like second breakdown, we aren't interested in the average dissipation - this influences the size of heatsink needed, but not the transistor's safe area. Therefore, Ppeak will be ...
Ppeak = P * 2 = 168 * 2 = 336 Watts
Remember that this is the real peak power that the devices must be able to handle, and they must be able to do so at elevated temperatures. We want to be safe, so we have to choose a temperature that is realistic, given the type of service for which the amp is designed (home theatre, live sound, disco, etc.). For the sake of the exercise, we'll assume a fairly safe usage such as domestic hi-fi, and assume that the transistor die may get to 75°C (we'll use a really good heatsink). The transistor peak power dissipation must never exceed the maximum allowable, so we have to ensure that peak power remains below 140W (using MJL4281/4302 devices).
Based on these quick assumptions, we will need 2.4 transistor pairs to handle the power. Obviously we can't get 0.4 of a transistor, so we will need 3 pairs of output devices to handle the load. It will be possible to reduce this to 2 pairs if the amp is intended only for hi-fi and has excellent heatsinks, includes a thermal fan and an over-temperature cutout to ensure that the heatsink remains cool enough to keep the transistor die temperature within acceptable limits.
Referring to the SOA graph above, we see that it is permissible to exceed the DC or 1 second limits, provided the time is kept short. It is fortunate that most music is quite dynamic, so the occasional excursion into 'prohibited' territory will normally be of fairly short duration and cause no problems. However, the 10ms limits should never be exceeded or failure is very likely.
This is by no means an exhaustive examination of the requirements, because there may be other factors that come into play in normal usage. It is up to the amp designer to make sure that everything has been accounted for, and to provide the necessary protection if the output devices are pushed close to their limits. These undertakings are not trivial, so the information shown here is certainly not 'gospel' - these are basic guidelines only. There is no substitute for rigorous testing and simulation to make sure that you haven't overlooked anything.
Don't forget about the driver transistors! They are just as much at risk as the output devices if they are under-specified.
It's also worth noting that MOSFETs do not suffer from second breakdown. All MOSFETs have a very different SOA curve from that of bipolar transistors, but switching (vertical) MOSFETs have a number of very exciting ways that can result in destruction when used in linear circuits. No manufacturer of switching MOSFETs recommends their use in linear circuits, and this tells you immediately that they aren't suitable. They can be used with extreme caution, but their characteristics are not really suited to audio use at all. In particular, switching MOSFETs are very sensitive to temperature, and it can be quite difficult to prevent thermal runaway. Gain linearity is also poor, so distortion will be higher than expected.
Lateral MOSFETs (as made by Renesas, Semelab/MagnaTEC, etc.) are very different, and are limited only by the average power - provided that output current and voltage ratings are not exceeded. Lateral MOSFETs have the ability to effectively shut themselves down if they get too hot (because ratings are exceeded), and they will normally recover once they cool down again. However, do not rely on this as a protection scheme, as continuous or repetitive overloads will lead to device failure. Gain linearity is not as good as most audio grade bipolar transistors, and it is normal to increase the overall circuit gain to allow more feedback. Overall results are usually very good though, and because they are so robust many very high power professional power amps used lateral MOSFETs until the advent of Class-G amplifiers. These provide far greater overall efficiency than Class-AB. See Project 101 for a good example of a lateral MOSFET amplifier (Class-AB) intended for hi-fi applications.
Maximum Current: The emitter area determines the maximum current capability of the device, there are many design options for emitters, for audio transistors nearly all manufacturers use perforated emitter (also called mesh emitter) designs. The perforated emitter design also gives better gain linearity than a regular 'interdigitised' emitter finger design (double or single comb). The other benefit for perforated emitter designs is silicon area utilisation, you can put a lot more emitter in a given piece of silicon with this design type for lower cost. One trade-off of the perforated design is switching, devices won't be as fast but we really don't need good switching capabilities for linear transistors and for audio this is a non issue. The bonding wire size depends on the current rating of course. Typical is around 15 mils (0.38mm) aluminium for audio high current devices.
Maximum Power Dissipation: Die size is the main parameter, ON Semi's Thermal Characterization Lab has done extensive studies and created some formulas for each package type so they can predict the thermal resistance (J-C) for any Die size in a given package. There are other factors besides die size that can affect the power dissipation, like solder line thickness, solder alloy, die thickness, etc. For good SOA performance good power dissipation is a must, thin die and very thin die attach solder are very important factors.
Second Breakdown: This is tough to determine and normally is determined by testing devices in a SOA tester by forcing power between collector and emitter and measuring the power dissipation time to secondary breakdown. Vertical structure of the device (collector and base thickness and resistivity) are important device design parameters, as well as die design geometry.
Current Gain: Emitter area determines the maximum current gain at high current levels of a device, too high peak hFE may result in lower BVCEO (Breakdown Voltage - collector to emitter, base open), higher hFE results in lower VCE(sat), VBE(on). As mentioned before, for good current gain linearity a perforated emitter design is best.
fT (Current Gain bandwidth Product): This is directly related to device gain and also to the device physical base width (wb). Most of the audio transistors in the industry have high fT (~30MHz), the trade-off is SOA performance with high voltage conditions. ON Semi Power Base Technology (which is unique in the market) has low/medium fT devices (8 to 12MHz) devices like the MJL21193/94 which have excellent SOA above 100V, these devices have wider Bases and also some unique 'base spreading resistor' design which make them extremely rugged, used by most high end audio manufacturers.
FBSOA (Forward Biased Safe Operating Area): Die size, power dissipation, die geometry and base width are some important parameters.
(The above information kindly provided by ON Semiconductor)
The following photos show a typical (functional) die, and two shots demonstrating destruction. The shiny sections are melted silicon! This is also a good example of the 'interdigitised finger' type of emitter and base construction referred to above.
Figures 4.1, 2, 3 - Transistor Dies (click for full size image)
The functional die (left) shows clearly what a typical transistor looks like. The emitter and base sections are clearly visible, with the emitter having the thicker 'fingers' for best current carrying ability. This is not one of the new ON Semi transistors - the photo is representative only.
The damage in the failed die is quite obvious (centre), and there is a section of melted silicon where the transistor failed. As is the case almost 100% of the time, the transistor is shorted. Open transistors normally are the result of a bonding wire failure after the short has caused excessive current. This failed die would (probably) show the base junction as intact in a resistance test.
A close-up view (right) with greater damage. A large section of the die has exploded from the failure point outwards, and molten silicon has been sprayed all over the die. This failure would almost certainly indicate a short on all terminals (provided bonding wires are intact).
It is a sobering thought that these failures would have taken place in a matter of milliseconds - once the second breakdown region has been reached, the transistor will enter a negative resistance state, and there is nothing that will prevent total failure once the process has started. (Negative resistance is probable, but not a certainty - it depends to some extent on the fabrication method.)
(The above photographs kindly provided by ON Semiconductor)
A great many protection systems have been used over the years, in the hope of protecting the transistors from damage under all conditions. Power opamp ICs have the most comprehensive protection, but often at the expense of sound quality. By necessity, the protection must operate while the transistors are still quite safe, so their maximum power is never available. For more information, see VI Limiters In Amplifiers, which discusses SOA and how limiters function.
Discrete amplifiers usually employ a simplified system, that will afford protection from most mishaps. A completely foolproof system is usually quite complex, and considerable care is needed to ensure that it does not activate during normal operation. This also applies to simplified systems, and a great many that I have seen do not provide complete protection at all - some are incapable of protecting against short circuits, unless at the end of a length of speaker cable (i.e. the cable's resistance forms part of the protection).
A further problem is that a full protection system will switch the power transistors off very fast, and given that the loudspeaker load is reactive, a 'flyback' voltage can be developed that can easily destroy the transistors anyway. Many amps use a pair of diodes from the output to each supply - they are designed to ensure that the output voltage can never exceed the supply rails (except by the diode voltage drop).
In nearly all cases, it is necessary to either use additional output transistors, or tailor the protection circuit to ensure that excessive fault currents are not possible. This invariably means that there will be regions of the signal waveform where the protection circuit will operate when it should not do so. It is thought by a great many people that protection circuits degrade sound quality, and from tests I have done, this is certainly the case if (when?) they operate on any normal loudspeaker load. The only way to avoid problems is to use more output devices than you planned to, and even this is no guarantee that the amplifier will survive every abuse that it will face in a typical domestic or professional application.
Few discrete protection circuits monitor the temperature of the output devices (or the average power level) and thus adjust to suit the conditions. This means that a hot amplifier has a lower level of protection than a cold one, and it is no surprise that amplifiers fail most when driven hard for long periods (commonly as a result of a speaker failure). Thermal tracking is 'automatic' in IC power amplifiers, since all devices are on a common piece of silicon.
Probably the best protection of all is a monitor that will operate if the output transistor SOA is approached, and removes the input signal. This is unfortunately much more difficult than it may seem at first, and the signal switching circuit is another candidate for sound quality degradation. Relays cannot be used, as they are not fast enough - remember, faults lasting only one millisecond can be sufficient to cause failure.
Fuses are used in amplifiers to prevent fire and further damage - no fuse is fast enough to protect an amplifier against fault currents, unless it is so low in value that it will blow during normal use (and even that is very doubtful).
Some 'high end' amplifiers where cost is no object use a vast number of transistors, and ensure that at no normal (or abnormal) load will they ever exceed perhaps 1/2 their maximum rating. For typical consumer amplifiers (and most professional amps as well) the cost is a primary consideration, and transistors are run to their limits - to do otherwise would make the amplifier uncompetitive in the market.
A typical protection circuit is shown in Figure 5.1 and it is representative of the majority of those in common use. By sensing the current through the emitter resistors (R1 and R4), the circuit detects an over-current fault, and removes base drive from the driver transistors (Q2 and Q5). As the voltage across the output transistor(s) is reduced and more voltage is applied to the load, the sensitivity of the protection circuit is also reduced, allowing the maximum current at lower collector-emitter voltages where second breakdown is not a problem. One of the critical areas is when the voltage across the transistor is at the maximum, but due to a reactive load, it is also expected to deliver high current. This condition is almost as bad as a short-circuit on the speaker lead!
Figure 5.1 - Typical Output Protection Circuit
How does the circuit work? It is fairly straightforward to explain, and the lower section is in grey because we won't be using it for this explanation. Looking only at the upper section, R1 is used to sense the current through Q3, and if it exceeds about 0.65V, Q1 will turn on, 'stealing' base current from Q2 (and thence Q3). D1 isolates the circuit from the drive circuit under normal operation. At zero volts output the full positive supply voltage is across Q3, so the current through Q3 must be limited to remain below the danger level on the SOA curve.
As the output voltage increases, R3 (via D2) shunts some of the current sense voltage to ground, reducing the effect, and allowing more current. When Q3 is fully on, the voltage across it is very low (typically less than 1.5V), and R3 is selected to limit the current to the maximum allowable collector current. The design of a good protection circuit is not trivial, and requires many interactive calculations to get right.
The danger zone (and the cause of most of the problems) remains at around 0V though - with a typical loudspeaker reactive load, a significant current is still needed, even when there is no voltage across the speaker. With 35V supplies (as described above), we may need as much as 6.5A when the output voltage is at zero volts (see Fig 2.3). The problems caused by phase angle are such that it is almost impossible to design a current limit circuit that will allow maximum power, but still provide protection for shorted speaker leads, unless there are more output devices than appear to be required. A short circuit across the speaker leads is especially dangerous, and the protection needs to limit the output transistor dissipation so it remains within the SOA curve of the devices. This isn't easy to achieve!
With supplies of 35V or less there are no major problems, but above ±35, the SOA becomes more and more limited. For example, at 50V, the maximum current is 4A, so R1 must be chosen to provide 0.65V at (or below) 4A, so the value should be 0.16 ohms. This is not available, so 0.22 ohms could be used, along with a resistor between the base and emitter of Q1 to reduce the voltage slightly. Remember that the SOA is limited further with increasing temperature, so 2A would be safer, or you could omit the resistor between base and emitter of Q1, giving a current limit of just under 3A with 0.15 Ohm emitter resistors for Q3 (and Q6). The resistor in series with the base (R2) allows the voltage sensing circuit (R3 & D2) to function and protects Q1 from excess base current caused by instantaneous fault conditions.
Any SOA protection scheme needs to be very well thought out, or major problems will be experienced with some loads. With the arrangement shown above, protection is virtually instant. With a fairly typical reactive load, this will cause the amplifier to switch off the transistor that's supplying current, generating gross distortion and high voltage spikes (think in terms of National Semiconductor's 'SPiKe™' protection, used in many of their IC power amps). Yes, the transistors are protected, and the 'catch' diodes (D3 & D6) get a good workout as they dissipate the back EMF from the speaker into the supply lines. Many professional high-power amps will have capacitors installed between the base and emitter (or between base and collector in some cases) on Q1 and Q4 to slow down the reaction and prevent very short-term overloads from triggering the protection circuits.
In other cases, the protection circuits are much more complex, and follow the transistor load-lines very accurately. However (and this is important), the majority do not compensate the protection transistors against output device temperature! As output devices get hot, the allowable safe area reduces. Unless there is comprehensive temperature compensation, the output devices need to have reserve capacity or an overload on an already hot amplifier can still cause failure. A good example would be an amp driving a speaker that fails after being driven hard for a few hours. The amp is probably at its thermal limit, and if the protection circuits have no thermal feedback the amplifier may fail anyway - despite the fact that it's supposed to be short-circuit-proof. As noted above and by default, IC power amps have all transistors on the same silicon die, and the protection circuits will be thermally coupled to the output transistors. (The 'K' in SPiKe stands for Kelvin - as in temperature.)
Note: At ±35V, an amplifier such as P3A is completely happy, and with high power (200W) transistors is operating within the SOA curve at all times and with any load (down to a typical 4 ohm nominal impedance). There is sufficient reserve power capacity to enable the amp to withstand full power into 4 ohm loads even at reasonably elevated temperatures.
However, once the supply voltage is increased, much of the reserve will be used up, making the amp liable to failure. This applies to any amplifier of similar ratings operating into typical loads - P3A has been used as an example, but the same constraints apply to any other design used in the same way.
With careful component selection, circuits such as the above can work quite well. A good design will be conservative (and will therefore need additional output transistors), and will be reasonably effective in all cases. If the designer tries to get as close as possible to the transistor's ratings, the safety margin is reduced, and protection is less effective, Some legitimate signals will cause limiting, and other loads (especially at elevated temperatures) are likely to cause the transistors to exceed their ratings. There is a very good chance that the amp will survive regardless for many years, as the danger point may never be reached in some installations - other applications will destroy amp after amp, until one is found that can handle the abuse (or the abuse is removed).
It is obviously imperative to avoid second breakdown, and there are many ways that various designers have selected to do so. Protection circuits, Class-G (using two or more supply rails of each polarity), variable supply voltages, and even switched supply voltages - these are common in many home theatre amps, and a switch is sometimes used to select the voltage to suit the load impedance (which simply reduces the supply voltages when 'low impedance' is selected).
There is also the '"brute force' method, where there are so many power transistors that the cables will melt before any one transistor's ratings are exceeded, but this is uncommon except in extreme high end amps where the added cost is not considered a problem. Many amps provide no protection at all, other than ensuring that dissipation limits are observed, but a shorted speaker lead (or a lower than recommended load impedance) can cause the amp to fail.
Regardless of the method used, it is important to ensure that the designers' recommendations are followed - good output transistors are expensive, and few of us can afford the luxury (??) of blowing up amplifiers for the hell of it. While a design that exceeds the transistor ratings may last for many years, there will eventually be a combination of circumstances that will cause failure. Parties are a prime cause of blown amps and speakers, and if they feature regularly in your activities, a cheap system (that can play loud, but is very basic and has passable fidelity) is highly recommended. Its failure is not something you would cry over, and the main hi-fi system remains intact.
Finally, it is important to stress the importance of the SOA curve for any transistor used in an output stage (including driver transistors !). Any design that appears to be able to get more power from smaller transistors has almost certainly pushed the devices to (or beyond) their limits, and when driven hard into a difficult load, it will most probably fail - this is an expensive exercise if it takes the loudspeaker with it (not at all uncommon). Ultimately, a 'worst case' design procedure assumes that the amp will be driven hard into a difficult load, and with undersized or barely adequate heatsinks. Such a design will survive - others will not.
I am indebted to ON Semiconductor for reference material, die fabrication details, photographs, semiconductors and data sheets used in preparation of this article. Photos and other material provided by ON Semiconductor are used with their permission.
Further information is available in the ON Semiconductor application note AN1628-D in PDF format.
|Copyright Notice. This article, including but not limited to all text and diagrams (other than that material which is Copyright © ON Semiconductor), is the intellectual property of Rod Elliott, and is Copyright © 2003. Reproduction or re-publication by any means whatsoever, whether electronic, mechanical or electro- mechanical, is strictly prohibited under International Copyright laws. The author (Rod Elliott) grants the reader the right to use this information for personal use only, and further allows that one (1) copy may be made for reference. Commercial use is prohibited without express written authorisation from Rod Elliott.|