APPLICATION NOTE
LATERAL MOSFET DESIGN RECOMMENDATIONS FOR AUDIO AMPLIFIERS
Semelab Coventry Road Lutterworth Leicestershire LE17 4JB Telephone +44 (0) 1455 554711 Fax +44 (0) 1455 558843 Email: [email protected] Website: http://www.semelab-tt.com
222
Introduction Semelab’s range of lateral mosfets has been specifically designed for audio power amplifiers and there are a number of advantages that they exhibit over bipolar transistors in these applications. This application note is intended to highlight these differences and give design guidelines in using the mosfets for high power ` audio products.
Semelab’s new range of Alfet devices are Figure 1 – Complimentary N & P channel device available in both single (8Amp) and double-die (16Amp) versions with complementary N- channel and P-channel parts rated at both 160V Lateral Mosfet AdvaAdvannnntagestages and 200V. Table 1 outlines the currently The key advantages of lateral mosfets over available parts. bipolar junction transistors (BJTs) and vertical mosfets are: Table 111 • Complete absence of secondary breakdown Part Number Pol Voltage Current Package when compared to BJTs. This results in very ALF08N16V/ALF08P16V N/P 160V 8A TO-247 rugged performance and simplified ALF08N16K/ALF08P16K N/P 160V 8A TO-3 protection. ALF08N20V/ALF08P20V N/P 200V 8A TO-247 • Self limiting current characteristic. No ALF08N20K/ALF08P20K N/P 200V 8A TO-3 matter how much drive is given to a lateral ALF16N16W/ALF16P16W N/P 160V 16A TO-264 mosfets they will get to a point where they will not deliver further current. This factor, ALF16N16K/ALF16P16K N/P 160V 16A TO-3 contributed with the previous advantage, ALF16N20W/ALF16P20W N/P 200V 16A TO-264 gives the devices their bullet-proof ALF16N20K/ALF16P20K N/P 200V 16A TO-3 reliability. ALF08NP16V5 N&P 160V 8A TO-247-5 • In rare cases that a device does fail the ALF08NP20V5 N&P 200V 8A TO-247-5 resulting damage in the amplifier is usually limited to the output stage and does not Double-die devices incorporate 2 parallel tend to cause the typical chain reaction connected die that are taken from adjacent back through the amplifier that is often positions on the silicon wafer and this seen in bipolar designs. • guarantees a very high degree of matching of Simple and stable bias because of the low characteristics such that the 2 pieces of silicon current point where the temperature behave as a single device with twice the current characteristics cross from positive to capability. Other highlights in this range include negative coefficient. This is an advantage a new N and P-channel complementary part in a when compared to both vertical mosfets 5-pin power package that has a symmetrical and bipolars that many people are not fully pinout. This device allows a 100W-150W aware of! • amplifier to be achieved with a single output No charge storage effects like bipolar device resulting in a very small footprint that can transistors meaning that they are capable of prove beneficial in applications requiring multi- higher frequency operation and do not channel output in a small box. Figure 1 shows exhibit crossover distortion switching at this new 5-pin version of the TO-247 package. higher frequencies. • No beta droop characteristic of gain at As can be seen in the associated internal higher currents that BJTs exhibit. This diagram, the pinout is totally symmetrical contributes to consistent distortion allowing an optimised pcb layout to be performance across wide ranges of load achieved. impedance. • Easy to drive compared to BJTs due to the high impedance gate. This not only simplifies the drive stages of the amplifier that are required (smaller devices and lower cost) but also results in the voltage amplifier stage generating less distortion.
333
• Integral anti-parallel protection diode. The distortion (most readily observed with a 20kHz intrinsic mosfet parallel body diode means audio signal into load at high amplitude) and that an external diode does not need to be therefore the designer needs to make tradeoffs added as in BJT amplifiers. based on the requirements of the design.
This application note will look at some of these This is one of the key characteristics that make advantages in a lot mor e detail as well as lateral mosfets so well suited to class -AB audio outlining some good practices to use when amplifier applications. Stable bias against designer audio power amplifiers with lateral temperature results in a number of significant mosfets. advantages:
Biasing Considerations • Complete absence of thermal runaway risk It is commonly known and quoted that bipolar • Simple bias circuitry . An adjustable or fixed transistors have a positive temperature value resistor is adequate and no thermal coefficient (ptc) and that mosf ets have a coupling between bias devices and the negative temperature coefficient (ntc) in their heatsink is required . Transistor based diode relationship of drain current to temperature. multipliers with thermal coupling are not Whilst this ntc characteristic of mosfets is needed. certainly true, it does, however, only occur • No requirement for source resistors for above a particular operating drain current. For single device or parallel device operation lateral mosfets from Semelab this current level is (more on this in a later section) around 100mA , which it just so happens is the • Devices cool faster and return to a stable ideal point for optimum biasing of a cl ass-AB temperature faster after high power use audio amplifier! Figure 2 shows a typical • Much less critical bias point setting than transfer characteristic for the ALF 08N20V and bipolar amplifiers that does not drift with demonstrates a thermal coefficient crossover temperature allo wing easier setup and point at around 110mA. At this point the consistency in production thermal coefficient is in fact zero and therefore completely stable with fluctuations in There are other mosfets on the market aimed at temperature as the output stage warms up at audio applications that are not lateral types but turn on or is hot after high pow er usage. instead of a vertical structure. These do not exhibit the same thermal characteristics and a Figure 2 - ALF16N20W Transfer transfer curve fo r a typical device in shown in Characteristic Figure 3. 1 0.9 0.8 (A) (A) (A) (A) 0.7 D D D D 0.6 0.5 0.4 0.3 0.2 DrainDrain Current Current I I DrainDrain Current Current I I 0.1 0 0 0.5 1 1.5 2 Gate Source Voltage V (V) Gate Source Voltage V GSGSGS (V)(V) Figure 2 – ALF08N20 Transfer Characteristics around temperature coefficient crossover point Figure 3 – Transfer characteristic of typical vertical Although this is the optimum bias point for mosfet stable thermal operation some high power designs that need to minimise the power As can be seen from this curve, the temperature dissipated under quiescen t bias conditions often crossover point does not occur until the drain run the devices at a lower current. Although current is at around 8A! Therefore, for there will be a small shift of the operating point operation at bias levels of 100mA, source with temperature it is still quite small and resistors and thermally coupled transistor bias acceptable and 50mA will work well. Taking the circuits like those used for bi polar amplifiers are bias too low will eventually result in cross over necessary for stable thermal operation to control
444
runaway and for paralleled applications. The Clipping Mechanism in Mosfet Switching mosfets also used in some amplifier Amplifiers designs also exhibit similar characteristics (with The output voltage swing capability and thermal crossovers often at even higher resulting clip point is a result of different currents) and similar steps in the design to parameters than the V induced clipping that provide stable thermal operation are also ce(sat) required. As a result lateral mosfets amplifiers occurs in bipolar power amplifiers. There are 2 are far simpler to set-up with reduced potentially contributing factors with a mosfet: • component requirements for stable bias Rds(on) induced clipping. When the device is operation that results in a smaller footprint and fully turned on it behaves as a resistor with the eradication of these source resistors. an approximate value of 1 Ω per die. Simplified bias circuitry and reduced pcb area Therefore 6A of current per die would mean mean a cost reduction in both materials and that it exhibits a saturation voltage of 6V labour that should be considered in the total and the amp would only swing within 6V of solution cost comparison. This exceptional the power supply rail. thermal stability also enhances reliability. • Gate voltage (Vgs) induced clipping. To be Figure 4 shows a typical lateral mosfet output able to deliver a give drain current the stage and the recommended bias arrangement. mosfet will need a corresponding value of The values of RBF and RBA are dependent on gate-source voltage. For the 6A example, the rest of the design and the current that is approximately 5V would be required for a present in the preceding stage. O single die device at 25 C (see figure 5) .
Figure 5 – ALFET transconductance characteristic
Which of these mechanisms happens first Figure 4 – Typical Lateral Mosfet Output Stage depends on a number of factors in the overall design of the amplifier circuit. In the typical As a guideline to the nominal value of this output stage shown in figure 6 the driver stage resistance the following equation can be used: can swing to within about 1V of the rail (the driver stage bias current plus driver transistor saturation). This means that the voltage Ω resulting from the R ds(on) and the voltage required to drive the mosfet are around the same so the clip point would be about 6V from the power Where: Vbias(total) = the required bias voltage of the N & P channel mosfets added together rail.
(typically 2V) and Isource = the driver stage quiescent current. For example, if the driver It should also be noted that both these stage has 20mA quiescent current then the total parameters, the R ds(on) and the required V gs to bias resistance should have a central, nominal deliver a given current, will increase with value of 100 Ω. temperature. This means that the onset of clipping will happen earlier at higher operating temperatures. These mechanisms form part of the action of self protection in the devices
555
where a point will be reached on a given Yes, the effective gain is lower resulting in a heatsink where the amplifier will not deliver higher apparent distortion figure of a source more than a certain amount of power. follower compared to a bipolar emitter follower VRAIL+ in an open loop configuration but there are some other factors that come into effect where R4 R3 the mosfet has advantages: 100 100
5mA • Mosfets do not exhibit a roll-off of gain at Q2 higher currents beta drop) as bipolars do. Vds = Rds(on) x I Figure 7 shows a graph of the transconductance of a single die lateral R1 Vgs mosfet where this can be clearly seen. As a 6.8nF R7 C1 12k 1K result the amplifier distortion specification is RLOAD very similar into different impedances 27pF C2 whereas a bipolar amplifier will always see a significant distortion increase into lower Q1 impedance loads. This make lateral mosfets especially suitable for amplifiers designed to R6 drive 2 & 4 Ω loads. 100
VRAIL-
Figure 6 – Typical Output Stage
Elevated Power Rails Because of this voltage driven characteristic and the fact that it can become the controlling factor in output voltage swing, some amplifier designs utilise higher voltage driver stage rails about 10V high than the main power rails. This is to ensure that there is plenty of available gate drive to take the output devices all the way to their R limit and use all the available voltage dsON swing of the power supply rails. Ultimately the Figure 7 – Typical Output Stage limiting factor on the output voltage swing will need to be determined for a given design but • Because of the stability of the bias point Semelab’s Amplifier Power Calculator with temperature and less critical setting, spreadsheet will allow simulation of various lateral mosfets typically exhibit better scenarios to determine what limits the output crossover distortion characteristics. Bipolars swing. It is often stated that mosfets must have can exhibit excellent specifications in this a higher voltage drive rail to utilise the full respect but because of the difficulties of output capability but this is not the case with stabilising this against temperature, the laterals that have a relatively high R dsON and will optimum bias point is usually only there only become important in designs with lots of under quiescent conditions. parallel devices (resulting in low total combined • As a result of the mosfet’s high bandwidth resistance) and topologies which do not allow capabilities both high open loop gain and the drivers to swing close to the power rails. wide open loop bandwidth are achievable. This not only makes a mosfet amplifier Lateral Mosfet Linearity easier to compensate but means that the It is often stated that mosfets are not as linear high open loop bandwidth typically as bipolar transistors. In simple terms this achievable when compared to bipolar statement is true and mostly can be attributed amplifiers means that the resulting closed- to the lower gain exhibited by the devices (in loop distortion figure can be similar. terms of their transconductance). However, as is • The absence of stored charge that occurs in often the case with comparing different BJTs means that they do not exhibit any technologies, it is not quite as clear cut as this! switching distortion that means high frequency crossover distortion is lower.
666
Heatsink Design Considerations The maximum junction temperature of the o The ‘on’ resistance of mosfets increases with mosfet silicon is rated at 150 C. Therefore, the temperature and although this is part of the heatsink size and any fan cooling must be mechanism that allows them to be very rugged calculated to keep the junctions temperatures in high power audio applications it will also inside this rating. To properly calculate the decrease the obtainable output power at required heatsink and indeed how many devices elevated junction temperatures. Lateral mosfets should be selected a number of design will tend to self limit at high temperatures parameters need to be known for the results to through this phenomenon which is beneficial be valid in the real world and therefore ensure from a protection point of view but good reliable operation. The key parameter is in thermal design is required to ensure that the understanding the worst case power dissipation required output power can be delivered with an of each output device. This is not always easy to optimal number of paralleled mosfets. calculate on a theoretical basis and experienced The absence of source resistors creates some designers tend to use rules of thumb they have additional extra benefits for designers when gained over the years A number of factors will combined with another factor that is specific to effect this figure. lateral mosfets. Vertical structure mosfets’ silicon die back connection is the drain and it is • The regulation of the power supply rails. that part of the silicon that is electrically bonded Depending on the size of smoothing to the metal back of the package. This means capacitors and rating of the mains that the power devices back metal is at opposite transformer, the rails will move up and rail polarities for the N and P devices and that if down with power output. a common heatsink is used the devices must be • The nature of the load reactance. An electrically isolated from it. Live heatsinks can be inductive load (as is the case with a used but a separate heatsink for positive and loudspeaker) will always result in increased negative sections of the amplifier are required. dissipation over that calculated for a purely This is an identical situation to amplifiers using resistive load due to phase shift. In addition power bipolar transistors. a drivers impedance rating will always be lower at low frequency as the inductive With lateral mosfets, however, the back of the reactance decreases where the impedance die is the source connection and therefore that will become the dc resistance of the voice back metal of the package is also electrically coil. Many commercial loudspeaker cabinets connected to the source. This factor combined will also present dips in the impedance with the fact that source resistors are not through the crossover points and these required for stable biasing and sharing with factors should also be included in paralleled devices means that all the sources of calculations and safety margins. the devices directly connect together and • It should always be remembered in these therefore the metal backs can as well. This is calculations that the maximum output significant and means that a single live internal device power dissipation does not occur at heatsink can be used in the product (one for maximum power output! each amplifier channel) that is at the potential • Remember to include the bias current of the amplifier output. By doing this the induced power dissipation. This figure is isolating pad of mica or silicon material can be always in addition to that dissipated by removed and thermal resistance is considerable delivering power into the load. ±50V power lowered resulting in both potentially smaller rails with 50mA of current results in 2.5W heatsinks and improved reliability. The speed of per output device at idle. response of die temperature increase to a high power transient is significantly improved If we look at an example of a class-AB amplifier without an electrically isolating pad and many designed to deliver 150Wrms output power into existing lateral mosfets customers exploit this a 4 Ω resistive load with a bias current of 50mA benefit of features within their designs. There is and a power supply regulation of 6% (typical also another potential benefit in this technique for a toroidal supply), we can plot the power in that the individual source connections do not dissipated against output power: need to be routed on the pcb and can instead be taken from a single point on the heatsink allowing for simplified pcb routing.
777
VOUT(rms) VOUT(rms) 2 5 7 10 12 15 17 20 22 25 2 5 7 9 11 14 16 18 21 23 140 160 120 140 100 120 80 100 80 60 60
PDISS(W) 40 PDISS(W) 40 20 20 0 0 2 6 2 6 14 24 38 54 74 97 15 26 40 58 79 122 151 103 131 162
Output Power(rms) Output Power(rms)
Figure 8 – Output stage total dissipated power Figure 10 – Power dissipated at 50Hz plotted against output power for 150W rms into 4Ω The same setup again but at 1000Hz input: The P is the total output stage dissipation and DISS 60 8 includes the bias power. For the single pair of output devices this was simulated for the 40 6 dissipation in each one is half the figure on this 20 graph. This theoretical plot shows us that we 4 would have a worst case power dissipation of Volts 0 IOUT(A) 115W (57.5W per device) that occurs around -20 2 100W output. OK, now we understand the behaviour into a purely resistive load, what will -40 0 happen if we hook up a real loudspeaker that Voltage VRAIL Current includes some inductance? As an example if we run the same simulation with a typical Figure 11 – Output stage Voltage and current at 1kHz loudspeaker ( 2 x 8 Ω units in parallel ), we can produce the following results at 50Hz input: The phase shifted relationship between the output stage voltage and current can clearly be 50 12 observed and although this results in lower power output than the resistive load, the power 40 10 dissipation is higher. 30 8 VOUT(rms) 20 6
Volts 3 5 8 10 13 15 18 21 23 26 10 4 IOUT(A) 160 0 2 140 120 -10 0 100 80 Voltage VRAIL Current 60 PDISS(W) 40 Figure 9 – Output stage Voltage and current at 50Hz 20 0 1 5 12 22 34 49 67 88 111 137
Output Power(rms)
Figure 12 – Power dissipated at 1000Hz
Please note that the maximum output power achieved in each case varies as the simulation calculates the maximum voltage swing that will
888
be achievable using real Alfet characteristics on sink and – sink to air. The classic thermal θsa the same power rails. Therefore, the conditions equation used in heatsinks is: differ in each case compared to the nominal T – T (OC/W) 150W rms output with a resistive load. It can be θ θ θ θ PDISS seen that at both low and mid frequencies the power dissipation is higher than for the resistive Ta Ambient - Ta load for this nominal 4 Ω rating. This drive unit Θ had a 6.5 Ω dc resistance and 1.15mH of sa inductance but clearly this situation just Ts becomes worse with increasing inductance. Θ These examples have demonstrated the need to cs Device case - Tc Heatsink - Ts weigh up these different considerations in Tc calculating the heatsink size and number of Insulator Θ devices. jc
In most commercial amplifier designs aimed at Device Junction - Tj Figure 13 – Semiconductor mounting thermal music reproduction most engineers make some resistances assumptions about the nature of the signal source. This assumption is based on the fact O The jc of the mosfets is 1 C/W and we have 2 that music does not represent a continuous full so the total result is 0.5 OC/W. If we assume a power sine wave and that through both the heatsink insulator (mica or silpad) has a of peak to average ratio and duty cycle of typical cs 0.4 OC/W, again we have 2 so their total is 0.2 music the resulting average power (and OC/W. We want our T to be 150 OC, our P is therefore heating effect) of the signal is lower. jmax DISS th 112W and we will assume the ambient Traditionally this figure was assumed to be 1/8 temperature, T to be 25 OC (although for rated power output and long term thermal tests a were measured at this level. There is, however, a heatsinks mounted internally and allowing for operations in hot countries, 50 OC may be a strong argument with modern music with high more realistic figure). Therefore: bass content to use a higher figure of average power. 150 – 25 0.5 0.2 θ Therefore, let’s assume we want to rate our 112 heatsinks such that the junction temperature of O O – C/W the mosfets does not go above 150 C with 0.5 0.2 θ 0.5 0.2 0.42 continuous operation at ¼ rated power output (in this case 37.5W. If we look at our 3 previous So we can now select a suitable heatsink or examples we can see the power dissipation at sink/fan combination. Interestingly, if we take this power point is: the approach outlined previously of running a heatsink live and we assume a thermal O Condition Dissipation at 37.5W mounting, , of 0.05 C/W then the required θ O output power heatsink thermal resistance would be 0.57 C/W Resistive load 98W (49W per device) – a significant difference. Alternatively, using Typical speaker at 50Hz 112W (56W per device) the previously calculated 0.42 OC/W heatsink Typical speaker at 1000Hz 99W (49.5W per device) without device insulators this would yield a Table 2 – Power dissipated at 1000Hz junction temperature of:
So assuming we want to take this worst case (OC) T PDISS θ θ θ T 133.6 112W figure for continuous average quarter power output dissipation we can now work out Taking into account that for approximately every our required heatsink thermal resistance. 10 OC rise in temperature that silicon lifetime is halved, this is a significant improvement in Figure 13 shows the thermal parameters of a junction temperature and potential long term heatsink mounted power semiconductor and reliability. the 3 thermal resistances that make up the total coupling from the silicon junction to the Often it is difficult with fan cooled heatsinks and ambient air: – junction to case, – case to fabricated designs to truly know the figure of θjc θcs heatsink thermal resistance. As a result it can be
999
useful to work backwards on a prototype design Parallel OperatiOperationon Considerations to verify the junction temperature under a given For higher power operation adding extra output set of conditions. Using our previous example, devices in parallel is an easy approach to scaling we know that if we set our amplifier up with power. One of the advantages of lateral mosfets 37.5W output then the output stage is is the ease of parallel operation. However, there dissipating 98W. Under these conditions, if we are a number of points that should be measure the heatsink temperature close to the considered to ensure both optimum output device we can approximate the junction performance and reliable operation: temperature and heatsink thermal resistance: • Current Sharing. We obviously want devices (OC) T PDISS θ θ T to share current as equally as possible. The
O best approach to achieving this is to make So if the heatsink measured 70 C sure that they share under bias conditions. Therefore, we ideally want parts that have (OC) T 112 0.5 0.2 100 148.4 the V characteristic matched. Again, the gs(th) best approach to this is to make sure that And the in this example would be: θ the V gs of the devices are as close as
O possible at the amplifiers intended bias T – T – C/W θ 0.402 current. Therefore, if the bias is 50mA per PDISS device, then match them at that current. This is a useful technique to validate thermal Luckily the turn-on characteristic of lateral resistance and junction temperature in a real mosfets is quite soft at these operating design. points and it is quite easy to find devices which match sufficiently well. In addition, Effects of Increasing Number of because of the ntc characteristic of the OOOutputOutput DDDevicesDevices devices above typical bias points no single device can runaway with the current and Clearly it can also be stated that increasing the they will balance themselves to some number of output devices will also result in extent. However, we need to make sure lower junction temperatures at a given power or that none of the devices are off and not a smaller heatsink. In our example, doubling up contributing at bias. on the output stage devices (i.e. 2 pairs of • This threshold does not need to be a set output devices rather than one) will half the θ voltage, it just needs to be matched in the in our equations to 0.25 and the to 0.1. In θ set of devices in a particular amplifier. In addition to this the equivalent total Rds(on) of the addition, it is not important that N-channel mosfets will half (and the required gate drive devices are matched to Ps but rather that all will reduce) so on the same power supply rails Ns are matched to each other and all Ps are the amplifier will be able to swing more output matched to each other. Many customers voltage. In our previous ±50V rail example, find that matching of 10-20% is quite adding an extra pair of output devices results in sufficient for reliable operation because at the amplifier being able to deliver 200Wrms higher currents the devices will very much output power – a significant increase. With the balance each other. O same C/W heatsink the mosfet junction • As devices are added in parallel the 0.42 O temperatures would now be around 122 C, capacitance presented to the driver stage even allowing for the increased dissipation that increases. Therefore, this driver stage needs results through the increased rail swing and to be able to deliver sufficient current to currents to deliver the higher 200W power. operate at the required bandwidth and slew rate of the amplifiers target specification. Therefore, there is always a trade-off in the • Semelab can supply devices pre-matched in decisions a designer makes against amplifier sets and colour coded to threshold value to efficiency, heatsink size, junction temperature a customer’s specification. Please enquire and the costs associated with the various about this service if it is required. options.
101010
Thermal Protection Considerations Gate Drive Power Requirements When used at their limits the mosfets will get Although a lateral mosfet does not require gate hot very fast and this will result in an increase in current to operate it does require current to
RdsON as well as a fall off in transconductance so charge and discharge the input capacitance (C iss a point will be reached where it is difficult to get is the lumped combination of C gs & C gd ). This further power from the devices and a self capacitance does vary with gate drive voltage limiting effect will happen. Although this and drain-source voltage but the datasheet characteristic contributes to the ruggedness of parameter does indicate a worst-case real value the devices the junctions will, however, get very to use. hot under these conditions and potentially exceed their maximum rating. Therefore it is In addition the p-channel parts use a larger die always recommend to include thermal size than the n-channel in order to match the on protection on the heatsink when relying on resistance ratings and therefore exhibit a larger protection by this mechanism. capacitance. As a result, to ensure we have sufficient current available and that the driver The example in the previous section stage is scaled to deliver this, it is wise to use demonstrates that if the average power were this higher figure. Therefore if we assume any high than 112W or if the product was around 700pF per single die device, we will exposed to higher ambient temperatures or any need a peak current of: cooling fans ceased to function that the junction temperature would exceed its 150 OC rating and 2 long term reliability would be compromised. Therefore, the techniques outlined previously This equation assumes delivering enough can be used to determine the junction current to reproduce a sine wave of frequency f temperature where this occurs and set the without generating slewing distortion rather thermal trip at this point. This ensures the amp than simply charging the capacitance in a time will both run continuously at its intended power of 1 ⁄ . Therefore, if we want to drive the gate level and be protected under adverse operating to 12V and operate to a frequency of 100kHz conditions for any period of time. we can calculate the required gate drive current:
Many power amplifier failures can be attributed 2 700 10 · 12 · 100 10 to these high temperatures that occur in designs with insufficient heatsinking or thermal 2 700 10 · 12 · 100 10 protection set at the wrong point.