| Episodes | 54 |
| Mentions | 64 |
| Cited here | 30 |
| First — last | #18 — #727 |
| Top guests | Mike Harrison, Ronald Sousa, Carmen |
| Related | fpga · component sourcing · usb · capacitor · dc dc converter |
A voltage regulator is an electronic circuit or component that maintains a constant output voltage despite variation in its input supply or load.[18][120] Regulators fall into two broad families—linear regulators, which dissipate the difference between input and output as heat, and switching converters, which use an inductor to transfer energy—and the two can be distinguished on a board by eye, since an inductor adjacent to the power device is a definitive indicator of the switching type.[120][591] The category spans an enormous range: a simple three-terminal linear part costing roughly one cent and a modern processor power regulator that amounts to a complete mixed-signal system in a single package are both technically voltage regulators, despite a difference in specification and performance comparable to that between two very different vehicles that are both technically cars.[493][566] Because fixed-output regulators in the common voltages cost around a cent, one appears in almost every design, and the engineering decision is typically which regulator to use rather than whether to use one.[493]
Types and identification
Regulator type can be read off a board visually: a small three-terminal package surrounded by capacitors could be either a linear regulator or a switching converter, but an inductor sitting next to it settles the question in favour of the switching type.[591] Recognising circuit blocks by their physical layout in this way is a learned skill rather than an innate one; the shapes and arrangements mean nothing at first and become obvious with exposure, which is why experienced engineers can identify a function from a photograph.[591]
Linear regulation is the cruder approach but remains useful when a rail is needed immediately: a linear regulator from the parts drawer and a stack of batteries will produce five volts, at the cost of dissipating the entire input-to-output difference as heat.[120] Switching converters dominate where efficiency matters, and a modern processor power regulator is best understood as a complete mixed-signal system inside one package, carrying sensitive analog lines, digital buses, and power transistors within a few square millimetres.[566] Switching converters intended to sit near a radio are designed around the spectrum as well as efficiency: pushing the switching frequency to two megahertz places it above the AM band, and spreading the switching frequency reduces peak emissions, both measures existing to ease the customer’s compliance testing.[635]
The distance between a three-terminal linear part and modern processor power delivery is reflected in the engineering roles around them: there is an architecture function between applications and circuit design that specifies what a regulator’s control loop requires—error amplifier gain and bandwidth—then negotiates against what the designers can deliver on schedule and judges whether the regulation specification still holds with what comes back.[566]
Operating principles
Adjustable regulators and the reference pin
A widespread misunderstanding concerns the adjustable three-terminal regulator: its reference is a fixed voltage developed between the output pin and the adjust pin, not a feedback node held at that voltage relative to ground.[574] Tying the adjust pin to the output through a series resistor therefore turns the part into a current source, with the reference voltage divided by that resistor setting the output current.[574]
Shutdown, quiescent current, and wake-up
Shutdown current and quiescent current describe different states and are frequently confused.[635] In shutdown the regulator is off and the output is not regulated at all; quiescent current is what the part draws while holding the output in regulation with no load, so that the rail is ready the instant the load wakes up.[635] Wake-up time is specified alongside that current, typically requiring the rail to return to full regulation within microseconds to milliseconds, because a low-power state is only useful if leaving it is fast.[635]
Control loops
A supply can regulate voltage and current simultaneously by running two control loops in parallel and letting whichever loop demands the smaller pulse width take control.[522] That arrangement handles a load whose resistance changes during operation, such as a plating bath being loaded and unloaded.[522]
Rails and the proliferation of regulators
Modern boards often require many regulators. As process geometry has pushed core voltages down to around 0.9 volts while input and output pins remain near 3.3 volts, whether a part integrates its own core regulator has become one of the first selection questions rather than a detail.[217] A cheaply priced 32-bit microcontroller can carry a hidden cost here: a part needing a separate low core voltage cannot run directly from an existing rail, so an extra regulator joins the design and the system cost stops resembling the part price.[54] The economics underneath are that cost tracks silicon area, so with all else equal an 8-bit part remains cheaper than a 32-bit one regardless of headline prices.[54] Some parts remove the problem with an on-chip core regulator: the die runs internally at 1.5 or 1.8 volts while the package is powered from an ordinary rail, and the designer’s only obligation is a bypass capacitor.[53]
The cumulative effect is substantial: five separate regulators can be needed just to run one processor, each with its own passives and board area.[326] Some processors add further requirements—a supply that varies with clock speed, in one case between 1.2 and 1.35 volts, which cannot be met by sharing an existing rail and forces a dedicated adjustable regulator.[325] Conversely, removing a required rail simplifies a board disproportionately: when one FPGA family dropped its 2.5-volt requirement, the design could run from USB five volts with a couple of regulators and nothing else.[181] That rail had existed only for auxiliary configuration circuitry, making it pure overhead—a whole regulator and its passives supporting something incidental to the part’s actual function.[181]
Distributed and local regulation
Local regulation is what lets an awkward part live on a board whose supply does not suit it: placing a small regulator next to a device that needs an unusual rail means the board can be fed from whatever is available.[18] Older instruments distribute regulation widely rather than centrally, taking plus and minus 18-volt rails down to plus and minus 15 for the analog sections and eight volts down to five for the digital, which can put twenty regulators on a single board.[168] Distributing power across a network of interconnected boards turns the supply into a resistor-grid problem that is tedious to analyse; putting a regulator on each device absorbs about a volt of drop and removes the need for the analysis entirely.[330] Specifying generously at the supply removes a class of user problems: an eight-amp five-volt regulator on a USB hub lets every downstream port deliver two amps, which adds cost but means anything plugged in will power up.[425]
Design considerations and failure modes
Capacitors and stability
A classic stability failure comes from reading only half the capacitor specification: a datasheet calling for a minimum output capacitance may also assume a minimum equivalent series resistance, so substituting a modern ceramic of the correct value provides too little resistance and the regulator oscillates.[188]
Minimum load
Some regulators require a minimum load to stay in regulation; a crude but effective production fix is a power resistor in parallel with the output, guaranteeing the part always draws its minimum current whatever the real load is doing.[222]
Input rating and margin
A systematic field failure can be caused purely by margin: a five-volt regulator rated to 21 volts input running from a 24-volt supply will fail at scale, and the fix is simply fitting a properly rated part.[490] The absolute maximum input must be checked against the actual supply rather than the nominal one.[490]
Feedback networks
A wrong resistor value in a regulator’s feedback network produces a plausible but incorrect output rather than an obvious failure: the board comes up, the rail is simply not what it should be, and nothing points at the cause.[561]
Control pins and part selection
Parametric search does not cover pin function, so there is no way to filter regulators by whether they have a shutdown pin; vendors name and implement those pins differently, leaving manual reading of datasheets as the only method.[68] An unwanted control pin is a small liability rather than a free feature—it must be tied off correctly, usually through a resistor, and is one more thing to get wrong—so designers actively filter such parts out when the function is not needed.[68]
Supply risk
Supply risk concentrates in regulators despite their low cost: regulator packages are often specialised enough that an unavailable part is not footprint-compatible with anything else, so a shortage forces a redesign rather than a substitution.[530]
Omission of regulation
A product built to an extreme price may omit regulation entirely, which is visible in use: the display fades whenever the device starts a high-current operation, because the supply sags and nothing is holding the rail up.[633] At the other extreme, a single defective part can end a whole product: adding external Schottky diodes in parallel with a regulator brought a dropout from 0.6 to 0.3 volts and back inside specification, but the unresolvable question was how to prove that a one-in-a-hundred failure rate did not remain.[140]
Assembly and troubleshooting
The single most common kit assembly failure is a three-pin regulator fitted backwards, and it arrives reported as a design defect after the builder has checked everything else; the part, such as a 78L05 in a TO-92 package, is identifiable at a glance on a returned board.[477] The design response is clearer silkscreen marking, because orientation is genuinely not obvious to a beginner looking at a three-pin package, so the marking does real work rather than decorating the board.[477]
When hardware is reported broken, the discipline is to resist the reflex that it must be the firmware or the other party’s fault, and to check all the supply voltages first; the reflex is strong and usually wrong.[176] Remote troubleshooting of such problems is impractical: every measurement is a round trip of hours by email, so either the person has an instrument and knows how to use it interactively, or the exchange goes nowhere.[82]
Powering a precision measurement circuit directly from batteries lets its offset drift as the cells discharge, which shows up as a slowly changing result rather than an obvious fault; adding a regulator is what makes the measurement stable over the life of the battery.[146]
Protection and improvised uses
Regulators frequently act as sacrificial protection: when something upstream goes wrong they die and shield the circuitry behind them, which is why a repair on old equipment can mean replacing several regulators before the instrument comes back.[168]
Regulators have also been pressed into service outside their intended function. With no optocoupler available in the field, a five-volt regulator fed from a 24-volt supply has been used to provide level conversion, the current involved being low enough and a couple of extra passives handling the switching; in one case the temporary fix stayed in service for about two years.[266]