The Principles of Switching Power Conversion--Overview and Basic Terminology

Efficiency:

Any regulator carries out the process of power conversion with an 'efficiency,' defined as:

? = PO/PIN

… where PO is the 'output power,' equal to …

PO = VO × IO

...and PIN is the 'input power,' equal to PIN = VIN × IIN Here, IIN is the average or dc current being drawn from the source.

Ideally we want ? = 1, and that would represent a "perfect" conversion efficiency of 100%.

But in a real converter, that is with ? <1, the difference 'PIN - PO' is simply the wasted power "P_loss," or 'dissipation' (occurring within the converter itself). By simple manipulation we get:

P_loss = P_IN - PO P_loss = PO

- PO P_loss = PO ×

This is the loss expressed in terms of the output power. In terms of the input power we would similarly get P_loss = PIN ×

The loss manifests itself as heat in the converter, which in turn causes a certain measurable 'temperature rise' ?T over the surrounding 'room temperature' (or 'ambient temperature').

Note that high temperatures affect the reliability of all systems - the rule-of-thumb being that every 10fic rise causes the failure rate to double. Therefore, part of our skill as designers is to reduce this temperature rise, and thereby also achieve higher efficiencies.

Coming to the input current (drawn by the converter), for the hypothetical case of 100% efficiency, we get:

I_IN_ideal = IO × _ VO V_IN

So, in a real converter, the input current increases from its "ideal" value by the factor 1/?.

I_IN_measured = 1 × I_IN_ideal

Therefore, if we can achieve a high efficiency, the current drawn from the input (keeping application conditions unchanged) will decrease - but only up to a point. The input current clearly cannot fall below the "brickwall" that is "IIN_ideal," because this current is equal to PO/VIN - that is, related only to the 'useful power' PO, delivered by the power supply, which we are assuming has not changed.

Further, since VO × IO = VIN × IIN_ideal by simple algebra, the dissipation in the power supply (energy lost per second as heat) can also be written as P_loss = VIN × _ IIN_measured - IIN_ideal

This form of the dissipation equation indicates a little more explicitly how additional energy (more input current for a given input voltage) is pushed into the input terminals of the power supply by the applied dc source - to compensate for the wasted energy inside the power supply - even as the converter continues to provide the useful energy PO being constantly demanded by the load.

A modern switching power supply's efficiency can typically range from 65 to 95% - that figure being considered attractive enough to have taken switchers to the level of interest they arouse today, and their consequent wide application. Traditional regulators (like the 'linear regulator') provide much poorer efficiencies - and that is the main reason why they are slowly but surely getting replaced by switching regulators.

Linear Regulators

'Linear regulators,' equivalently called 'series-pass regulators,' or simply 'series regulators,' also produce a regulated dc output rail from an input rail. But they do this by placing a transistor in series between the input and output. Further, this 'series-pass transistor' (or 'pass-transistor') is operated in the linear region of its voltage-current characteristics - thus acting like a variable resistance of sorts. As shown in the uppermost schematic of Fgr. 2, this transistor is made to literally "drop" (abandon) the unwanted or "excess" voltage across itself.

The excess voltage is clearly just the difference 'VIN - VO'- and this term is commonly called the 'headroom' of the linear regulator. We can see that the headroom needs to be a positive number always, thus implying VO < VIN. Therefore, linear regulators are, in principle, always 'step-down' in nature - that being their most obvious limitation.

In some applications (e.g. battery powered portable electronic equipment), we may want the output rail to remain well-regulated even if the input voltage dips very low - say down to within 0.6 V or less of the set output level VO. In such cases, the minimum possible headroom (or 'dropout') achievable by the linear regulator stage may become an issue.

Fgr. 2: Basic Types of Linear and Switching Regulators

No switch is perfect, and even if held fully conducting, it does have some voltage drop across it. So the dropout is simply the minimum achievable 'forward-drop' across the switch.

Regulators which can continue to work (i.e. regulate their output), with VIN barely exceeding VO, are called 'low dropout' regulators, or 'LDOs.' But note that there is really no precise voltage drop at which a linear regulator "officially" becomes an LDO. So the term is sometimes applied rather loosely to linear regulators in general. However, the rule-of thumb is that a dropout of about 200 mV or lower qualifies as an LDO, whereas older devices (conventional linear regulators) have a typical dropout voltage of around 2V. There is also an intermediate category, called 'quasi-LDOs' that have a dropout of about 1 V, that is, somewhere in between the two.

Besides being step-down in principle, linear regulators have another limitation - poor efficiency. Let us understand why that is so. The instantaneous power dissipated in any device is by definition the cross-product V × I, where V is the instantaneous voltage drop across it and I the instantaneous current through it. In the case of the series-pass transistor, under steady application conditions, both V and I are actually constant with respect to time - V in this case being the headroom VIN - VO, and I the load current IO (since the transistor is always in series with the load). So we see that the V × I dissipation term for linear regulators can, under certain conditions, become a significant proportion of the useful output power PO. And that simply spells "poor efficiency"! Further, if we stare hard at the equations, we will realize there is also nothing we can do about it - how can we possibly argue against something as basic as V × IFI For example, if the input is 12 V, and the output is 5 V, then at a load current of 100 mA, the dissipation in the regulator is necessarily:

?V × IO = (12 - 5) V × 100 mA = 700 mW. The useful (output) power is however VO × IO = 5V × 100 mA = 500 mW. Therefore, the efficiency is PO/PIN = 500/(700 + 500) = 41.6%. What can we do about that?!

On the positive side, linear regulators are very "quiet" - exhibiting none of the noise and EMI (electromagnetic interference) that have unfortunately become a "signature" or "trademark" of modern switching regulators. Switching regulators need filters - usually both at the input and the output, to quell some of this noise, which can interfere with other gadgets in the vicinity, possibly causing them to malfunction. Note that sometimes, the usual input/output capacitors of the converter may themselves serve the purpose, especially when we are dealing with 'low-power' (and 'low-voltage') applications. But in general, we may require filter stages containing both inductors and capacitors. Sometimes these stages may need to be cascaded to provide even greater noise attenuation.

Achieving High Efficiency through Switching

Why are switchers so much more efficient than "linears"?

As their name indicates, in a switching regulator, the series transistor is not held in a perpetual partially conducting (and therefore dissipative) mode - but is instead switched repetitively. So there are only two states possible - either the switch is held 'ON' (fully conducting) or it’s 'OFF' (fully non-conducting) - there is no "middle ground" (at least not in principle). When the transistor is ON, there is (ideally) zero voltage across it (V = 0), and when it’s OFF we have zero current through it (I = 0). So it’s clear that the cross-product 'V × I' is also zero for either of the two states. And that simply implies zero 'switch dissipation' at all times. Of course this too represents an impractical or "ideal" case. Real switches do dissipate. One reason for that they are never either fully ON nor fully OFF. Even when they are supposedly ON, they have a small voltage drop across them, and when they are supposedly "OFF," a small current still flows through them. Further, no device switches "instantly" either - there is a always definable period in which the device is transiting between states. During this interval too, V × I is not zero, and some additional dissipation occurs.

We may have noticed that in most introductory texts on switching power conversion, the switch is shown as a mechanical device - with contacts that simply open ("switch OFF") or close ("switch ON"). So a mechanical device comes very close to our definition of a "perfect switch" - and that is the reason why it’s often the vehicle of choice to present the most basic principles of power conversion. But one obvious problem with actually using a mechanical switch in any practical converter is that such switches can wear out and fail over a relatively short period of time. So in practice, we always prefer to use a semiconductor device (e.g. a transistor) as the switching element. As expected, that greatly enhances the life and reliability of the converter. But the most important advantage is that since a semiconductor switch has none of the mechanical "inertia" associated with a mechanical device, it gives us the ability to switch repetitively between the ON and OFF states - and do so very fast. We have already realized that that will lead to smaller components in general.

We should be clear that the phrase "switching fast," or "high switching speed," has slightly varying connotations, even within the area of switching power conversion. When it’s applied to the overall circuit, it refers to the frequency at which we are repeatedly switching: ON OFF ON OFF and so on.

This is the converter's basic switching frequency 'f' (in Hz).

But when the same term is applied specifically to the switching element or device, it refers to the time spent transiting between its two states (i.e. from ON to OFF and OFF to ON), and is typically expressed in 'ns' (nanoseconds). Of course this transition interval is then rather implicitly and intuitively being compared to the total 'time period' T (where T = 1/f ), and therefore to the switching frequency - though we should be clear there is no direct relationship between the transition time and the switching frequency.

We will learn shortly that the ability to crossover (i.e. transit) quickly between switching states is in fact rather crucial. Yes, up to a point, the switching speed is almost completely determined by how "strong" and effective we can make our external 'drive circuit.' But ultimately, the speed becomes limited purely by the device and its technology -- an "inertia" of sorts at an electrical level.

Basic Types of Semiconductor Switches

Historically, most power supplies used the 'bjt' (bipolar junction transistor) shown in Fgr. 2. It’s admittedly a rather slow device by modern standards. But it’s still relatively cheap! In fact its 'npn' version is even cheaper, and therefore more popular than its 'pnp' version. Modern switching supplies prefer to use a 'mosfet' (metal oxide semiconductor field effect transistor), often simply called a 'fet' (see Fgr. 2 again). This modern high-speed switching device also comes in several "flavors" - the most commonly used ones being the n-channel and p-channel types (both usually being the 'enhancement mode' variety). The n-channel mosfet happens to be the favorite in terms of cost-effectiveness and performance, for most applications. However, sometimes, p-channel devices may be preferred for various reasons - mainly because they usually require simpler drive circuits.

Despite the steady course of history in favor of mosfets in general, there still remain some arguments for continuing to prefer bjts in certain applications. Some points to consider and debate here are:

a) It’s often said that it’s easier to drive a mosfet than a bjt. In a bjt we do need a large drive current (injected into its 'base' terminal) - to turn it ON. We also need to keep injecting base current to keep it in that state. On the other hand, a mosfet is considered easier to drive. In theory, we just have to apply a certain voltage at its 'gate' terminal to turn it ON, and also keep it that way. Therefore, a mosfet is called a 'voltage-controlled' device, whereas a bjt is considered a 'current controlled' device.

However, in reality, a modern mosfet needs a certain amount of gate current during the time it’s in transit (ON to OFF and OFF to ON). Further, to make it change state fast, we may in fact need to push in (or pull) out a lot of current (typically 1 to 2 A).

b) The drive requirements of a bjt may actually turn out easier to implement in many cases. The reason for that is, to turn an NPN bjt ON For example, its gate has to be taken only about 0.8 V above its emitter (and can even be tied directly to its collector on occasion). Whereas, in an n-channel mosfet, its gate has to be taken several volts higher than its source. Therefore, in certain types of dc-dc converters, when using an n-channel mosfet, it can be shown that we need a 'drive rail' that is significantly higher than the (available) input rail VIN. And how else can we hope to have such a rail except by a circuit that can somehow manage to "push" or "pump" the input voltage to a higher level? When thus implemented, such a rail is called the 'bootstrap' rail.

Note: The most obvious implementation of a 'bootstrap circuit' may just consist of a small capacitor that gets charged by the input source (through a small signal diode) whenever the switch turns OFF. Thereafter, when the switch turns ON, we know that certain voltage nodes in the power supply suddenly "flip" whenever the switch changes state. But since the 'bootstrap capacitor' continues to hold on to its acquired voltage (and charge), it automatically pumps the bootstrap rail to a level higher than the input rail, as desired. This rail then helps drive the mosfet properly under all conditions.

c) The main advantage of bjts is that they are known to generate significantly less EMI and 'noise and ripple' than mosfets. That ironically is a positive outcome of their slower switching speed!

d) Bjts are also often better suited for high-current applications-because their 'forward drop' (on-state voltage drop) is relatively constant, even for very high switch currents.

This leads to significantly lower 'switch dissipation,' more so when the switching frequencies are not too high. On the contrary, in a mosfet, the forward drop is almost proportional to the current passing through it - so its dissipation can become significant at high loads. Luckily, since it also switches faster (lower transition times), it usually more than makes up, and so in fact becomes much better in terms of the overall loss - more so when operated at very high switching frequencies.

Note: In an effort to combine the "best of both worlds," a "combo" device called the 'IGBT' (insulated gate bipolar transistor) is also often used nowadays. It’s driven like a mosfet (voltage-controlled), but behaves like a bjt in other ways (the forward drop and switching speed). It too is therefore suited mainly for low-frequency and high-current applications, but is considered easier to drive than a bjt.

Semiconductor Switches Are Not "Perfect"

We mentioned that all semiconductor switches suffer losses. Despite their advantages, they are certainly not the perfect or ideal switches we may have imagined them to be at first sight.

So For example, unlike a mechanical switch, in the case of a semiconductor device, we may have to account for the small but measurable 'leakage current' flowing through it when it’s considered "fully OFF" (i.e. non-conducting). This gives us a dissipation term called the 'leakage loss.' This term is usually not very significant and can be ignored. However, there is a small but significant voltage drop ('forward drop') across the semiconductor when it’s considered "fully ON" (i.e. conducting) - and that gives us a significant 'conduction loss' term. In addition, there is also a brief moment as we transition between the two switching states, when the current and voltage in the switch need to slew up or down almost simultaneously to their new respective levels. So, during this 'transition time' or 'crossover time,' we neither have V = 0 nor I = 0 instantaneously, and therefore nor is V × I = 0. This therefore leads to some additional dissipation, and is called the 'crossover loss' (or sometimes just 'switching loss'). Eventually, we need to learn to minimize all such loss terms if we want to improve the efficiency of our power supply.

However, we must remember that power supply design is by its very nature full of design tradeoffs and subtle compromises. For example, if we look around for a transistor with a very low forward voltage drop, possibly with the intent of minimizing the conduction loss, we usually end up with a device that also happens to transition more slowly - thus leading to a higher crossover loss. There is also an overriding concern for cost that needs to be constantly looked into, particularly in the commercial power supply arena. So, we should not underestimate the importance of having an astute and seasoned engineer at the helm of affairs, one who can really grapple with the finer details of power supply design. As a corollary, neither can we probably ever hope to replace him or her (at least not entirely), by some smart automatic test system, nor by any "expert design software" that we may have been dreaming of.

Achieving High Efficiency through the Use of Reactive Components

We have seen that one reason why switching regulators have such a high efficiency is because they use a switch (rather than a transistor that "thinks" it’s a resistor, as in an LDO).

Another root cause of the high efficiency of modern switching power supplies is their effective use of both capacitors and inductors.

Capacitors and inductors are categorized as 'reactive' components because they have the unique ability of being able to store energy. However, that is why they cannot ever be made to dissipate any energy either (at least not within themselves) - they just store any energy "thrown at them"! On the other hand, we know that 'resistive' components dissipate energy, but unfortunately, can't store any!

A capacitor's stored energy is called electrostatic, equal to 1 2 × C × V2 where C is the 'capacitance' (in Farads), and V the voltage across the capacitor. Whereas an inductor stores magnetic energy, equal to 1 2 × L × I 2, L being the 'inductance' (in Henries) and I the current passing through it (at any given moment).

But we may well ask - despite the obvious efficiency concerns, do we really need reactive components in principle? For example, we may have realized we don't really need an input or output capacitor for implementing a linear regulator - because the series-pass element is all that is required to block any excess voltage. For switching regulators however, the reasoning is rather different. This leads us to the general "logic of switching power conversion" summarized below.

+++ A transistor is needed to establish control on the output voltage, and thereby bring it into regulation. The reason we switch it’s as follows - dissipation in this control element is related to the product of the voltage across the control device and the current through it, that is V × I. So if we make either V or I zero (or very small), we will get zero (or very small) dissipation. By switching constantly between ON and OFF states, we can keep the switch dissipation down, but at the same time, by controlling the ratio of the ON and OFF intervals, we can regulate the output, based on average energy flow considerations.

+++ But whenever we switch the transistor, we effectively disconnect the input from the output (during either the ON or OFF state). However, the output (load) always demands a continuous flow of energy. Therefore we need to introduce energy storage elements somewhere inside the converter. In particular, we use output capacitors to "hold" the voltage steady across the load during the above-mentioned input-to-output "disconnect" interval.

+++ But as soon as we put in a capacitor, we now also need to limit the inrush current into it - all capacitors connected directly across a dc source, will exhibit this uncontrolled inrush - and that can't be good either for noise, EMI, or for efficiency.

Of course we could simply opt for a resistor to subdue this inrush, and that in fact was the approach behind the early "bucket regulators" (Fgr. 2).

+++ But unfortunately a resistor always dissipates - so what we may have saved in terms of switch dissipation, may ultimately end up in the resistor! To maximize the overall efficiency, we therefore need to use only reactive elements in the conversion process. Reactive elements can store energy but don’t dissipate any (in principle).

Therefore, an inductor becomes our final choice (along with the capacitor), based on its ability to non-dissipatively limit the (rate of rise) of current, as is desired for the purpose of limiting the capacitor inrush current.

Some of the finer points in this summary will become clearer as we go on. We will also learn that once the inductor has stored some energy, we just can't wish this stored energy away at the drop of a hat". We need to do something about it! And that in fact gives us an actual working converter down the road.

Early RC-based Switching Regulators

As indicated above, a possible way out of the "input-to-output disconnect" problem is to use only an output capacitor. This can store some extra energy when the switch connects the load to the input, and then provide this energy to the load when the switch disconnects the load.

But we still need to limit the capacitor charging current ('inrush current'). And as indicated, we could use a resistor. That was in fact the basic principle behind some early linear-to-switcher "crossover products" like the 'bucket regulator' shown in Fgr. 2.

The bucket regulator uses a transistor driven like a switch (as in modern switching regulators), a small series resistor to limit the current (not entirely unlike a linear regulator), and an output capacitor (the "bucket") to store and then provide energy when the switch is OFF. Whenever the output voltage falls below a certain threshold, the switch turns ON, "tops up" the bucket, and then turns OFF. Another version of the bucket regulator uses a cheap low-frequency switch called an SCR ('semiconductor controlled rectifier') that works off the secondary windings of a step-down transformer connected to an ac mains supply, as also shown in Fgr. 2. Note that in this case, the resistance of the windings (usually) serves as the (only) effective limiting resistance.

Note also that in either of these RC-based bucket regulator implementations, the switch ultimately ends up being toggled repetitively at a certain rate - and in the process, a rather crudely regulated stepped down output dc rail is created. By definition, that makes these regulators switching regulators too! But we realize that the very use of a resistor in any power conversion process always bodes ill for efficiency. So, we may have just succeeded in shifting the dissipation away from the transistor - into the resistor! If we really want to maximize overall efficiency, we need to do away with any intervening resistance altogether.

So we attempt to use an inductor instead of a resistor for the purpose - we don't really have many other component choices left in our bag! In fact, if we manage to do that, we get our first modern LC-based switching regulator - the 'buck regulator' (i.e. step-down converter), as also presented in Fgr. 2.

LC-based Switching Regulators

Though the detailed functioning of the modern buck regulator of Fgr. 2 will be explained a little later, we note that besides the obvious replacement of R with an L, it looks very similar to the bucket regulator - except for a "mysterious" diode. The basic principles of power conversion will in fact become clear only when we realize the purpose of this diode.

This component goes by several names - 'catch diode,' 'freewheeling diode,' 'commutation diode,' and 'output diode,' to name a few! But its basic purpose is always the same - a purpose we will soon learn is intricately related to the behavior of the inductor itself.

Aside from the buck regulator, there are two other ways to implement the basic goal of switching power conversion (using both inductors and capacitors). Each of these leads to a distinct 'topology.' So besides the buck (step-down), we also have the 'boost' (step-up), and the 'buck-boost' (step-up or step-down). We will see that though all these are based on the same underlying principles, they are set up to look and behave quite differently. As a prospective power supply designer, we really do need to learn and master each of them almost on an individual basis. We must also keep in mind that in the process, our mental picture will usually need a drastic change as we go from one topology to another.

Note: There are some other capacitor-based possibilities - in particular 'charge pumps'- also called 'inductor-less switching regulators.' These are usually restricted to rather low powers and produce output rails that are rather crudely regulated multiples of the input rail. In this book, we are going to ignore these types altogether.

Then there are also some other types of LC-based possibilities - in particular the 'resonant topologies.' Like conventional dc-dc converters, these also use both types of reactive components (L and C) along with a switch. However, their basic principle of operation is very different. Without getting into their actual details, we note that these topologies don’t maintain a constant switching frequency, which is something we usually rather strongly desire. From a practical standpoint, any switching topology with a variable switching frequency, can lead to an unpredictable and varying EMI spectrum and noise signature. To mitigate these effects, we may require rather complicated filters. For such reasons, resonant topologies have not really found widespread acceptance in commercial designs, and so we too will largely ignore them from this point on.

The Role of Parasitics

In using conventional LC-based switching regulators, we may have noticed that their constituent inductors and capacitors do get fairly hot in most applications. But if, as we said, these components are reactive, why at all are they getting hot? We need to know why, because any source of heat impacts the overall efficiency! And efficiency is what modern switching regulators are all about! The heat arising from real-world reactive components can invariably be traced back to dissipation occurring within the small 'parasitic' resistive elements, which always accompany any such (reactive) component.

For example, a real inductor has the basic property of inductance L, but it also has a certain non-zero dc resistance ('DCR') term, mainly associated with the copper windings used.

Similarly, any real capacitor has a capacitance C, but it also has a small equivalent series resistance ('ESR'). Each of these terms produces 'ohmic' losses - that can all add up and become fairly significant.

As indicated previously, a real-world semiconductor switch can also be considered as having a parasitic resistance "strapped" across it. This parallel resistor in effect "models" the leakage current path, and thus the 'leakage loss' term. Similarly, the forward drop across the device can also, in a sense, be thought of as a series parasitic resistance - leading to a conduction loss term.

But any real-world component also comes along with various reactive parasitics. For example an inductor can have a significant parasitic capacitance across its terminals - associated with electrostatic effects between the layers of its windings. A capacitor can also have an equivalent series inductance ('ESL') - coming from the small inductances associated with its leads, foil, and terminations. Similarly, a mosfet also has various parasitics - For example the "unseen" capacitances present between each of its terminals (within the package). In fact, these mosfet parasitics play a major part in determining the limits of its switching speed (transition times).

In terms of dissipation, we understand that reactive parasitics certainly cannot dissipate heat - at least not within the parasitic element itself. But more often than not, these reactive parasitics do manage to "dump" their stored energy (at specific moments during the switching cycle) into a nearby resistive element - thus increasing the overall losses indirectly.

Therefore we see that to improve efficiency, we generally need to go about minimizing all such parasitics - resistive or reactive. We should not forget they are the very reason we are not getting 100% efficiency from our converter in the first place. Of course, we have to learn to be able to do this optimization to within reasonable and cost-effective bounds, as dictated by market compulsions and similar constraints.

But we should also bear in mind that nothing is so straightforward in power! So these parasitic elements should not be considered entirely "useless" either. In fact they do play a rather helpful and stabilizing role on occasion.

+++ For example, if we short the outputs of a dc-dc converter, we know it’s unable to regulate, however hard it tries. In this 'fault condition' ('open-loop'), the momentary 'overload current' within the circuit can be "tamed" (or mitigated) a great deal by the very presence of certain identifiably "friendly" parasitics.

+++ We will also learn that the so-called 'voltage-mode control' switching regulators actually rely on the ESR of the output capacitor for ensuring 'loop stability'- even under normal operation. As indicated previously, loop stability refers to the ability of a power supply to regulate its output quickly, when faced with sudden changes in line and load, without undue oscillations or ringing.

Certain other parasitics however may just prove to be a nuisance and some others a sheer bane. But their actual roles too may keep shifting, depending upon the prevailing conditions in the converter. For example:

+++ A certain parasitic inductance may be quite helpful during the turn-on transition of the switch - by acting to limit any current spike trying to pass through the switch.

But it can be harmful due to the high voltage spike it creates across the switch at turn-off (as it tries to release its stored magnetic energy).

+++ On the other hand, a parasitic capacitance present across the switch For example, can be helpful at turn-off - but unhelpful at turn-on, as it tries to dump its stored electrostatic energy inside the switch.

Note: We will find that during turn-off, the parasitic capacitance mentioned above helps limit or 'clamp' any potentially destructive voltage spikes appearing across the switch, by absorbing the energy residing in that spike. It also helps decrease the crossover loss by slowing down the rising ramp of voltage, and thereby reducing the V-I "overlap" (between the transiting V and I waveforms of the switch). However at turn-on, the same parasitic capacitance now has to discharge whatever energy it acquired during the preceding turn-off transition - and that leads to a current spike inside the switch. Note that this spike is externally "invisible" - apparent only by the higher-than-expected switch dissipation, and the resulting higher-than-expected temperature.

Therefore, generally speaking, all parasitics constitute a somewhat "double-edged sword," one that we just can't afford to overlook for very long in practical power supply design.

However, as we too will do in some of our discussions that follow, sometimes we can consciously and selectively decide to ignore some of these second-order influences initially, just to build up basic concepts in power first. Because the truth is if we don't do that, we just run the risk of feeling quite overwhelmed, too early in the game!

Switching at High Frequencies

In attempting to generally reduce parasitics and their associated losses, we may notice that these are often dependent on various external factors - temperature for one. Some losses increase with temperature - For example the conduction loss in a mosfet. And some may decrease - For example the conduction loss in a bjt (when operated with low currents).

Another example of the latter type is the ESR-related loss of a typical aluminum electrolytic capacitor, which also decreases with temperature. On the other hand, some losses may have rather "strange" shapes. For example, we could have an inverted "bell-shaped" curve - representing an optimum operating point somewhere between the two extremes. This is what the 'core loss' term of many modern 'ferrite' materials (used for inductor cores) looks like - it’s at its minimum at around 80 to 90fic, increasing on either side.

From an overall perspective, it’s hard to predict how all these variations with respect to temperature add up - and how the efficiency of the power supply is thereby affected by changes in temperature.

Coming to the dependency of parasitics and related loss terms on frequency, we do find a somewhat clearer trend. In fact it’s rather rare to find any loss term that decreases at higher frequencies (though a notable exception to this is the loss in an aluminum electrolytic capacitor - because its ESR decreases with frequency). Some of the loss terms are virtually independent of frequency (e.g. conduction loss). And the remaining losses actually increase almost proportionally to the switching frequency - For example, the crossover loss.

So in general, we realize that lowering, not increasing, the switching frequency would almost invariably help improve efficiency.

There are other frequency-related issues too, besides efficiency. For example, we know that switching power supplies are inherently noisy, and generate a lot of EMI. By going to higher switching frequencies, we may just be making matters worse. We can mentally visualize that even the small connecting wires and 'printed circuit board' (PCB) traces become very effective antennas at high frequencies, and will likely spew out radiated EMI in every direction.

This therefore begs the question: why at all are we face to face with a modern trend of ever-increasing switching frequencies? Why should we not decrease the switching frequency? The first motivation toward higher switching frequencies was to simply take "the action" beyond audible human hearing range. Reactive components are prone to creating sound pressure waves for various reasons. So, the early LC-based switching power supplies switched at around 15-20 kHz, and were therefore barely audible, if at all.

The next impetus toward even higher switching frequencies came with the realization that the bulkiest component of a power supply, that is, the inductor, could be almost proportionately reduced in size if the switching frequency was increased (everybody does seem to want smaller products, after all!). Therefore, successive generations of power converters moved upward in almost arbitrary steps, typically 20 kHz, 50 kHz, 70 kHz, 100 kHz, 150 kHz, 250 kHz, 300 kHz, 500 kHz, 1 MHz, 2 MHz, and often even higher today. This actually helped simultaneously reduce the size of the conducted EMI and input/output filtering components - including the capacitors! High switching frequencies can also almost proportionately enhance the loop response of a power supply.

Therefore, we realize that the only thing holding us back at any moment of time from going to even higher frequencies are the "switching losses." This term is in fact rather broad - encompassing all the losses that occur at the moment when we actually switch the transistor (i.e. from ON to OFF and/or OFF to ON). Clearly, the crossover loss mentioned earlier is just one of several possible switching loss terms. Note that it’s easy to visualize why such losses are (usually) exactly proportional to the switching frequency - since energy is lost only whenever we actually switch - therefore, the greater the number of times we do that (in a second), the more energy is lost (dissipation).

Finally, we also do need to learn how to manage whatever dissipation is still remaining in the power supply. This is called 'thermal management,' and that is one of the most important goals in any good power supply design. Let us look at that now.

Reliability, Life, and Thermal Management

Thermal management basically just means trying to get the heat out from the power supply and into the surroundings - thereby lowering the local temperatures at various points inside it. The most basic and obvious reason for doing this is to keep all the components to within their maximum rated operating temperatures. But in fact, that is rarely enough. We always strive to reduce the temperatures even further, and every couple of degrees Celsius may well be worth fighting for.

The reliability 'R' of a power supply at any given moment of time is defined as R(t) = e-?t.

So at time t = 0 (start of operational life), the reliability is considered to be at its maximum value of 1. Thereafter it decreases exponentially as time elapses. '?' is the failure rate of a power supply, that is, the number of supplies failing over a specified period of time. Another commonly used term is 'MTBF,' or mean time between failures. This is the reciprocal of the overall failure rate, that is, ? = 1/MTBF. A typical commercial power supply will have an MTBF of between 100,000 hours to 500,000 hours - assuming it’s being operated at a fairly typical and benign 'ambient temperature' of around 25 C.

Looking now at the variation of failure rate with respect to temperature, we come across the well-known rule-of-thumb - failure rate doubles every 10 C rise in temperature. If we apply this admittedly loose rule-of-thumb to each and every component used in the power supply, we see it must also hold for the entire power supply too - since the overall failure rate of the power supply is simply the sum of the failure rates of each component comprising it (? = ?1 + ?2 + ?3 + ... .). All this clearly gives us a good reason to try and reduce temperatures of all the components even further.

But aside from failure rate, which clearly applies to every component used in a power supply, there are also certain 'lifetime' considerations that apply to specific components. The

'life' of a component is stated to be the duration it can work for continuously, without degrading beyond certain specified limits. At the end of this 'useful life,' it’s considered to have become a 'wearout failure'- or simply put - it’s "worn-out." Note that this need not imply the component has failed "catastrophically" - more often than not, it may be just "out of spec." The latter phrase simply means the component no longer provides the expected performance - as specified by the limits published in the electrical tables of its datasheet.

Note: Of course a datasheet can always be "massaged" to make the part look good in one way or another - and that is the origin of a rather shady but widespread industry practice called "specmanship." A good designer will therefore keep in mind that not all vendors' datasheets are equal - even for what may seem to be the same or equivalent part number at first sight.

As designers, it’s important that we not only do our best to extend the 'useful life' of any such component, but also account upfront for its slow degradation over time. In effect, that implies that the power supply may initially perform better than its minimum specifications.

Ultimately however, the worn-out component, especially if it’s present at a critical location, could cause the entire power supply to "go out of spec," and even fail catastrophically.

Luckily, most of the components used in a power supply have no meaningful or definable lifetime - at least not within the usual 5 to 10 years of useful life expected from most electronic products. We therefore usually don't, For example, talk in terms of an inductor or transistor "degrading" (over a period of time) - though of course either of these components can certainly fail at any given moment, even under normal operation, as evidenced by their non-zero failure rates.

Note: Lifetime issues related to the materials used in the construction of a component can affect the life of the component indirectly. For example, if a semiconductor device is operated well beyond its usual maximum rating of 150 C, its plastic package can exhibit wearout or degradation - even though nothing happens to the semiconductor itself up to a much higher temperature. Subsequently, over a period of time, this degraded package can cause the junction to get severely affected by environmental factors, causing the device to fail catastrophically - usually taking the power supply (and system) with it too! In a similar manner, inductors made of a 'powdered iron' type of core material are also known to degrade under extended periods of high temperatures - and this can produce not only a failed inductor, but a failed power supply too.

A common example of lifetime considerations in a commercial power supply design comes from its use of aluminum electrolytic capacitors. Despite their great affordability and respectable performance in many applications, such capacitors are a victim of wearout due to the steady evaporation of their enclosed electrolyte over time. Extensive calculations are needed to predict their internal temperature ('core temperature') and thereby estimate the true rate of evaporation and thereby extend the capacitor's useful life. The rule recommended for doing this life calculation is - the useful life of an aluminum electrolytic capacitor halves every 10 C rise in temperature. We can see that this relatively hard-and-fast rule is uncannily similar to the rule-of-thumb of failure rate. But that again is just a coincidence, since life and failure rate are really two different issues altogether.

In either case, we can now clearly see that the way to extend life and improve reliability is to lower the temperatures of all the components in a power supply and also the ambient temperature inside the enclosure of the power supply. This may also call out for a better-ventilated enclosure (more air vents), more exposed copper on the PCB (printed circuit board), or say, even a built-in fan to push the hot air out. Though in the latter case, we now have to start worrying about both the failure rate and life of the fan itself! Stress Derating

Temperature can ultimately be viewed as a 'thermal stress'- one that causes an increase in failure rate (and life if applicable). But how severe a stress really is, must naturally be judged relative to the 'ratings' of the device. For example, most semiconductors are rated for a 'maximum junction temperature' of 150 C. Therefore, keeping the junction no higher than 105 Cin a given application represents a stress reduction factor, or alternately – a 'temperature derating' factor equal to 105/150 = 70%.

In general, 'stress derating' is the established technique used by good designers to diminish internal stresses and thereby reduce the failure rate. Besides temperature, the failure rate (and life) of any component can also depend on the applied electrical stresses - voltage and current. For example, a typical 'voltage derating' of 80% as applied to semiconductors means that the worst-case operating voltage across the component never exceeds 80% of the maximum specified voltage rating of the device. Similarly, we can usually apply a typical 'current derating' of 70-80% to most semiconductors.

The practice of derating also implies that we need to select our components judiciously during the design phase itself - with well-considered and built-in operating margins. And though, as we know, some loss terms decrease with temperature, contemplating raising the temperatures just to achieve better efficiency or performance is clearly not the preferred direction, because of the obvious impact on system reliability.

A good designer eventually learns to weigh reliability and life concerns against cost, performance, size, and so on.

Advances in Technology

But despite the best efforts of many a good power supply designer, certain sought after improvements may still have remained merely on our annual Christmas wish list! Luckily, there have been significant accompanying advances in the technology of the components available, to help enact our goals. For example, the burning desire to reduce resistive losses and simultaneously make designs suitable for high frequency operation has ushered in significant improvements in terms of a whole new generation of high-frequency, low-ESR ceramic and other specialty capacitors. We also have diodes with very low forward voltage drops and 'ultra-fast recovery,' much faster switches like the mosfet, and several new low-loss ferrite material types for making the transformers and inductors.

Note: 'Recovery' refers to the ability of a diode to quickly change from a conducting state to a non-conducting (i.e. 'blocking') state as soon as the voltage across it reverses. Diodes which do this well are called 'ultrafast diodes.' Note that the 'Schottky diode' is preferred in certain applications, because of its low forward drop (~0.5 V). In principle, it’s also supposed to have zero recovery time. But unfortunately, it also has a comparatively higher parasitic 'body capacitance' (across itself), that in some ways tends to mimic conventional recovery phenomena. Note that it also has a higher leakage current and is typically limited to blocking voltages of less than 100 V.

However we observe that the actual topologies used in power conversion have not really changed significantly over the years. We still have just three basic topologies: the buck, the boost, and the buck-boost. Admittedly, there have been significant improvements like 'ZVS' (zero voltage switching), 'current-fed converters,' and 'composite topologies' like the 'Cuk converter' and the 'SEPIC' (single ended primary inductance converter), but all these are perhaps best viewed as icing on a three-layer cake. The basic building blocks (or topologies) of power conversion have themselves proven to be quite fundamental. And that is borne out by the fact that they have stood the test of time and remained virtually unchallenged to date.

So, finally, we can get on with the task of really getting to understand these topologies well.

We will soon realize that the best way to do so is via the route that takes us past that rather enigmatic component - the inductor. And that's where we begin our journey now. . .