| Episodes | 65 |
| Mentions | 86 |
| Cited here | 38 |
| First — last | #5 — #684 |
| Top guests | Mike Engelhardt, Trey German, Jeff Keyzer |
| Related | power supply · transformer · inductor · digital to analog converter · fpga |
A switching power supply (also switched-mode power supply, SMPS) is an electronic power converter that regulates its output by interrupting the input tens of thousands of times per second, with the ratio of on-time to off-time setting the output voltage.[361] The technique matters because operating the switch at high frequency allows the magnetics to be made very small — small enough that a mains charger can occupy a cubic inch rather than the brick-sized enclosure typical of 1980s equipment.[361] Switching conversion displaced the linear regulator in consumer products, early portable computers, and eventually most electronic equipment, on grounds of cost, heat, weight, and size.[222][684]
Operating principle
The core principle is interruption: the input is switched on and off at rates from tens of kilohertz to several megahertz, and the duty ratio between the on and off intervals determines the output voltage.[361][61] Because energy is transferred in discrete packets at high frequency, the transformer or inductor that stores and transfers that energy can be physically small; at these frequencies a transformer tiny enough to fit inside a plug body suffices.[361]
Internally, the converter functions as a small control system rather than a passive power component. Output current is sensed across a resistor, the sensed value is fed back, and the controller uses it to adjust the drive and the pulse width.[62] Controllers also change strategy across the operating range, including skipping pulses entirely under light load.[62]
Failure behaviour
The failure behaviour of a switching supply differs fundamentally from that of a linear supply, and this shapes how the surrounding system must be designed. A linear supply that is overloaded sags and carries on; a switching supply pushed past its limit shuts down.[222] In consumer computers this hard limit turned an accessory drawing six hundred milliamps into a supply-sizing problem for every unit shipped, and integrating the accessory’s function — which required only a hundred milliamps — became the cheaper answer.[222]
Frequency, magnetics, and switching devices
Switching frequency sets the size of the magnetics. At fifteen kilohertz the inductors are large components, often hand-wound, occupying a serious fraction of the board.[61] Commercial designs mostly operate between a few hundred kilohertz and a few megahertz, and at the top of that range the inductor shrinks to the point where a loop of copper trace on the board itself can serve as one.[61]
There is, however, a ceiling. As frequency rises, the parasitic capacitance and inductance of the layout and of the components themselves come to dominate, so the efficiency curve peaks and then falls — there is a sweet spot rather than a monotonic gain from ever-higher frequency.[61]
Wide-bandgap semiconductors
Wide-bandgap devices — gallium nitride and silicon carbide — move that ceiling rather than removing it, switching fast enough (on the order of 100 volts per nanosecond) that the transistor spends essentially no time in its dissipative linear region.[61] Their significance lies in breaking the usual trade-off: raising a conventional transistor’s breakdown voltage tends to make it slower, whereas these materials retain both high breakdown voltage and high speed.[553] Sharp switching edges combined with high breakdown produce very efficient DC–DC converters, and efficiency is what makes a supply physically small: higher efficiency means less waste heat, less waste heat means no large internal or external heatsink, and that chain is what allows sixty watts to be delivered from a plug-sized package.[553]
Power silicon in general does not follow the leading edge of process technology; such parts are built on mature nodes such as 90 or 45 nanometres rather than on the processes attracting fabrication investment.[582]
Control
Control of a switching converter has traditionally been implemented by a dedicated analogue integrated circuit closing the feedback loop. The digital alternative reads the output voltage and current and adjusts the pulse-width-modulated duty cycle in software.[212]
The argument for digital control is iteration speed: an incorrect filter value in an analogue controller means recalculating the resistor and capacitor network, unsoldering parts, and possibly reordering them, whereas the digital equivalent is a coefficient change, a recompile, and a reflash.[212] The argument against is that a firmware image then has to be maintained, which is a poor trade against a fixed resistor and capacitor in a static application such as a plain 5 V to 3.3 V regulator.[212]
What makes digital control workable in practice is a hardware linkage between peripherals: the comparison events that shape the pulse width can also trigger the analogue-to-digital converter, so the measurement always lands at the same point in the switching cycle.[212] That determinism — the sampling instant relative to the switching edge — matters more than the raw speed of the converter.[212]
Noise
The defining drawback of the topology is inherent rather than incidental: pushing pulses through a system is what a switching converter does, so electrical noise accompanies it, and a genuinely quiet system still requires a linear stage or a transformer somewhere in the chain.[9] Cascading two switching converters with no linear stage anywhere between them leaves nothing to attenuate the high-frequency content, and in such a chain the output capacitance is also what holds the loop stable when there is no load at all.[360]
The noise is not only electrical. Magnetics hum and vibrate, so a “quiet” supply may need to be quiet in the mechanical sense as well.[9] Audible noise has three distinct sources. First, an inductor wound loosely — a risk with hand-wound parts, as large power magnetics often are — vibrates under load.[127] Second, operating the supply outside its intended range can drive the controller into a very low duty-cycle mode that whines.[127] Third, multilayer ceramic capacitors are microphonic and audibly sing; because the effect is reciprocal, tapping the board makes them generate a voltage.[127]
Because switching behaviour, output-capacitor effectiveness, and sometimes the switching frequency itself all change with load, noise must be characterised across the load range rather than at a single operating point.[360]
Electromagnetic compliance
Including a switching converter in a product commits the design to electromagnetic-compatibility work, because the switching action is a deliberate source of emissions.[184] One practical response in noise-sensitive analogue products is to buy the problem rather than design it: an off-the-shelf external brick with switching content in the tens of kilohertz (roughly 30–50 kHz) is straightforward to filter and has small leakage currents.[513] The same choice keeps mains wiring out of the product entirely.[513]
Design practice
Layout dependence
Switching converters are heavily layout-dependent, so a component substitution is not a like-for-like swap the way replacing a linear regulator is.[601] The cost cascades: new passives must be selected, qualification effort grows, and a replacement part in a smaller package can force the board into blind and buried vias.[601] All of this can be triggered by an unremarkable part making an unremarkable rail, which is why a switcher is a poor place to carry sourcing risk.[601]
Controller selection and sourcing risk
Implementing the control loop on a general-purpose microcontroller using its own timers and converters is considered the wrong instinct; a dedicated controller chip is known to work.[88] Dependence on a specific vendor part, however, creates supply exposure: an end-of-life notice or an allocation problem leaves the design stranded.[88] When such a part becomes scarce, the decision is binary — pay as much as a hundred times the original price or design the part out — and there is no honest probability to attach to either branch.[601]
Modelling and simulation
Device modelling for switching-converter simulation is difficult. For bipolar transistors it is essentially impossible to build a model from a datasheet, because the quasi-saturation region is never documented well enough.[196] For field-effect devices it is possible and adequate for the application, but laborious: test fixtures are built that curve-trace the device to reproduce the datasheet characteristics — output curves, on-resistance against gate voltage, gate charge — and the model parameters are adjusted until they match.[196] Even then, a trap remains: often no single physical transistor matches the whole datasheet, because different curves were extracted from different devices, so no model can fit all of them at once.[196] Beginner-oriented browser-based simulators are likewise unsuited to the problem, since convergence on a switching supply is exactly the case they are not designed to handle.[210]
Modules and bought-in converters
At small production volumes the calculation favours bought-in conversion: for a run of ten units, an off-the-shelf module avoids selecting the transistor and inductor, working through the calculations, and discovering afterwards that the loop oscillates.[604] Complete modules go further, packaging the converter with a panel interface and control knob so that a bench supply is only an enclosure and an input source away.[408] That reusability once underpinned a broader claim — that since every product contains a converter and the layout had been done many times, nobody would lay out a converter board from scratch again — but the claim did not survive contact with practice.[565] At the low end of the market a gap remains: adjustable modules are widely available and noisy, whereas much of the time a fixed, properly regulated output with no adjustment is what is wanted.[406]
Organisation and review
In organisations that treat converter design seriously, the work is partitioned to a specialist: one designer takes the converter half of the board and another takes the digital and analogue remainder, with the two merged at the end.[230] The converter is also a classic subject for a design review to catch, and the feedback loop in particular is where an inexperienced designer is found out — Dave Jones has described being caught out on exactly this point early in his career.[138] A development circuit of any kind should expose its intermediate signals deliberately, because that access is exactly what disappears once the function is integrated into production hardware.[309]
Applications
Bench and laboratory supplies
Laboratory instruments are not exempt from the trade-offs. A bench supply can be switch-mode internally, with a quasi-linear post-regulator and substantial filtering behind it, and still be noisier than a classic linear design.[169] Above a certain current there is no choice: a linear supply delivering forty amps would be unmanageably heavy, so high-current bench supplies are switch-mode by necessity.[277]
Commodity and consumer supplies
A commodity computer supply is a cheap source of serious current — a few hundred watts for around twenty dollars — and quieter than its reputation suggests, because low-cost manufacturers have adopted better offline switching techniques.[199] Its low-voltage rail carries tens of amps (a 5 V rail can supply on the order of 35 A), making it a practical starting point for generating other rails by boosting rather than by building a supply from scratch.[198] Converters built to modern efficiency standards draw around a tenth of a watt in standby.[5] Universal input — the ability of a switching front end to accept any regional mains voltage — is what allows a single stock item to be sold across regions with interchangeable plug pins, though the saving in overhead is pushed onto the buyer as a drawer of unused adapters.[523]
Front-panel switching and standby behaviour
A front-panel power switch is often not a mains switch at all but a logic input to the converter’s enable pin. This is a legitimate design, and it is also the reason an instrument can draw six and a half watts while apparently switched off.[39]
Mains-side failure behaviour
A mains-side failure inside a switching supply can put current to earth and trip a residual-current breaker while leaving no visible damage and no equipment that has stopped working, which makes the fault very hard to localise afterwards.[81]
Interaction with dimmers
Lamp dimming interacts with switching supplies in specific ways. A conventional (leading-edge) dimmer waits past the mains zero crossing, then triggers a latching device that stays on until the current falls to zero at the end of the half cycle.[524] Modern lamps, whose front end is a rectifier feeding an electrolytic capacitor, prefer the opposite arrangement, because switching the rectifier-capacitor input on partway up the waveform draws a large current surge — the reason conventional dimmers contain a sizeable choke to slow the edge.[524] Switching off late in the cycle has its own hazard: interrupting several amps drawn through the inductance of a transformer secondary can generate a spike on the order of 800 V, sufficient to destroy a small internal mains supply while leaving the output devices under test unharmed.[524] The remedies are to avoid the transformer or to place a few microfarads of capacitance across its output; the diagnosis requires putting an oscilloscope on the incoming mains rather than on the suspected circuit.[524]
High-voltage and industrial converters
At extreme voltage the topology is unchanged but everything around it is different. A converter taking mains in and producing ten kilovolts out means creepage and clearance distances measured in inches and circuit boards floating at the full output voltage.[438] Potting the assembly solves the insulation problem and creates a thermal one, because the heat — up to a kilowatt of it — then has to cross the potting compound to escape.[438] The way out is materials that conduct heat without conducting electricity, such as alumina and other technical ceramics, supplemented by heat pipes spreading the thermal load to several points.[438]
Industrial rectifiers are the same topology at scale: despite a name that suggests four diodes, they are switch-mode supplies built for high current and high voltage.[522] The same logic appears in power distribution. Running low-voltage direct current over any distance loses too much to I²R resistive drops, so the converter belongs local to the load rather than centralised.[25] The grid-scale version of the idea, a solid-state transformer operating at several kilovolts (for example 7.2 kV), is essentially the switching-supply topology scaled up.[583]
History
The transition from linear to switching supplies in consumer products was driven by cost alone.[222] In early portable computers the attraction was that the topology addressed three constraints at once — heat, weight, and size — rather than any one of them.[684] The transition also raised system complexity: instruments from the 1980s are markedly harder to service than those from the 1970s, and the switching supply is among the components responsible for the increase.[655]