After the chain works: where a supply bites back
By the end of this rung you can build the whole chain from guide 1: a transformer steps the wall voltage down, a bridge rectifier flips it all positive, a reservoir capacitor fills the gaps into a lumpy DC, and a linear regulator (or the efficient switcher of guide 4) shaves that down to a clean, steady rail. On paper, it works. On the bench, three things still reach out and bite: it runs hot, the rail is never perfectly flat, and the real world throws surges and shorts at it. This guide is about those three — heat, ripple, and protection — because they are where supplies actually fail.
Each of the three traces straight back to an honest physical cost you have already met. Heat is the linear regulator's tax: it can only throw away the voltage it does not pass, and thrown-away voltage times current is power dissipated as heat. Ripple is the reservoir capacitor admitting it cannot hold a perfectly flat voltage while it feeds a load between charging peaks. And protection exists because the same big capacitor that smooths the rail looks like a dead short at the instant you switch on. None of these is a flaw to engineer away — they are the price of the parts, and good design budgets for them up front.
Heat: the linear regulator's honest tax
Recall the heart of guide 2: a linear regulator holds its output steady by dropping the rest across a series-pass transistor, like a tap throttling a hose. Whatever voltage it drops, the full load current still flows through it, so it must burn (Vin - Vout) times I as heat. Take a common 5 V regulator fed from 12 V delivering 0.5 A: it dissipates (12 - 5) times 0.5 = 3.5 W while delivering only 5 times 0.5 = 2.5 W to the load. That is a efficiency of 2.5 / 6 = about 42 percent — more than half the energy becomes heat. The wider the gap from input to output, the worse the tax.
SAME OUTPUT, TWO SUPPLIES (5 V @ 0.5 A from a 12 V input)
-----------------------------------------------------------------
useful power out = 5 V x 0.5 A = 2.5 W
LINEAR (e.g. 7805)
wasted as heat = (12 - 5) x 0.5 = 3.5 W
efficiency = 2.5 / (2.5 + 3.5) = 42 %
junction rise = 3.5 W x theta
no heat sink ~ 3.5 x 50 C/W = 175 C -> COOKS, shuts down
good sink ~ 3.5 x 20 C/W = 70 C -> ok (ambient + 70)
SWITCHER (buck, ~90% eff)
input power = 2.5 / 0.90 = 2.78 W
wasted as heat = 2.78 - 2.5 = 0.28 W (12x less)
cost = switching NOISE on the rail, more parts
-----------------------------------------------------------------
rule: a linear wastes (Vin - Vout) x I as heat;
a switcher trades that heat for noise.Where does 3.5 W go? Into the regulator's package, and from there it must escape to the air or the silicon cooks itself. Heat flow obeys its own Ohm's law: a temperature rise equals power times a thermal resistance (theta, in C per watt), exactly as voltage equals current times resistance. A small package alone might have theta around 50 C/W, so 3.5 W lifts the junction about 175 C above ambient — far past the roughly 125 C where silicon fails. Bolt it to a heat sink and you drop theta toward 20 C/W, holding the rise near 70 C, which survives. And if it ever does overheat anyway, a modern regulator's built-in thermal shutdown quietly switches it off before it dies — a safety net, not a license to skip the heat sink.
- Find the worst-case heat in the pass device: P = (Vin_max - Vout) times I_max. Use the highest input and the full load — both at once, because that is the moment it runs hottest.
- Set the junction limit and the ambient: take the datasheet's max junction temperature, derate it (say to 110 C for margin), and use the hottest ambient your enclosure will actually see.
- Compute the total thermal resistance you are allowed: theta_total = (Tj_limit - Tambient) / P. This is the whole budget from silicon to air.
- Subtract the fixed junction-to-case theta from the datasheet; what remains is the budget for the heat sink plus its thermal interface (grease or pad).
- Pick a heat sink whose theta sits at or below that budget, add thermal compound, and leave margin — still air is optimistic and dust only makes things worse over time.
- If the budget comes out tiny or negative, stop fighting heat: drop Vin closer to Vout, or switch to a switching regulator — no heat sink can rescue a hopeless gap-times-current.
Ripple and noise: keeping the rail flat
Before the regulator even sees it, the rail is bumpy. The reservoir capacitor charges to the peak each time the rectifier pushes, then discharges into the load until the next push, sagging a little in between — that sag is ripple. Its size follows the capacitor's own rule, V = I times t / C: with 0.5 A drawn, a full-wave rectified 60 Hz mains refilling every t = 1/120 = about 8.3 ms, and C = 2200 uF, the ripple is 0.5 times 0.0083 / 0.0022 = about 1.9 V peak-to-peak. Almost two volts of wobble — which is exactly why we do not feed raw rectified DC to a circuit, and why the regulator earns its place.
The regulator's job is to swallow that wobble, and a linear one is good at it: a typical part rejects ripple by 60 dB — a factor of 1000 — turning 1.9 V of input ripple into under 2 mV at the output. But there is a catch tied straight to guide 2's dropout: the regulator only works while its input stays at least the dropout voltage above its output. The ripple dips, and at the bottom of each dip the input must still clear Vout + dropout, or the output collapses for that instant. So you size the reservoir cap and transformer not for the average input but for the trough — and this is where a low-dropout regulator shines, because its tiny headroom requirement leaves room for more ripple and a smaller, cheaper cap.
Even a perfect regulator cannot keep the rail flat at the far end of a long wire, because a fast-changing load — a logic chip switching, say — demands a gulp of current faster than the regulator can respond down a length of inductive trace. The fix is local: a decoupling capacitor, a small ceramic placed right at the chip's power pins. Think of it as a little water tank beside a thirsty machine, so the long pipe back to the pump never sags during a sudden sip. A common pairing is a 100 nF ceramic close in for the fast transients plus a larger 10 uF nearby for the bigger, slower gulps.
Protection: surviving the real world
Switch the supply on and, for an instant, that big reservoir capacitor is empty — and an empty capacitor looks like a dead short. The result is inrush current: a brief, violent spike as the cap charges, limited only by whatever tiny resistance happens to be in the path. It can weld relay contacts, blow fuses on power-up, and stress the rectifier diodes far past their steady rating. The classic cure is an NTC thermistor in series — cold at switch-on it has high resistance to throttle the surge, then it self-heats and drops to near-nothing so it wastes little in normal running. Bigger supplies add an active soft-start that ramps the input up gently instead.
Downstream, the danger is the opposite: a short on the output. Every decent regulator has a built-in current limit that caps how much it will deliver, folding back or simply clamping the output so a slipped probe or a solder bridge cannot let unlimited current flow. That limit works hand in hand with the thermal shutdown from earlier — limit the current first, and if the resulting heat still climbs too high, shut down entirely. A fuse is the last, crudest line: it protects the wiring and the source from a catastrophic fault, but it blows far too slowly to save a semiconductor. Never rely on a fuse to protect a chip; rely on it to stop a fire.
Two more faults are worth designing against because they are so common and so fatal. Reverse polarity — plugging the input in backwards — can destroy a board in microseconds; a series Schottky diode blocks it at the cost of a small forward drop, while a P-channel MOSFET does the same job with almost no loss. And overvoltage — a regulator failing short and dumping the full input onto a 5 V rail — is guarded by a crowbar: a Zener sensing the over-voltage fires a thyristor that deliberately shorts the rail and blows the fuse, sacrificing the supply to save the expensive circuit it feeds. Protection is always a small, deliberate cost paid to avoid a large, accidental one.
Batteries, and choosing the whole supply
Not every supply starts at the wall. A battery is a voltage source that lies a little: its voltage sags as it discharges and droops further under load through its internal resistance, so a 'nominal' 3.7 V lithium cell really swings from about 4.2 V full down to 3.0 V empty. That moving input is exactly why dropout matters so much on battery power — a regulator with high dropout stops working while there is still useful charge in the cell, throwing away runtime. A low-dropout regulator needing only a hundred millivolts of headroom squeezes far more life out of the same battery.
But notice the trap: that 3.7 V cell starts above a 3.3 V rail and ends below it. A plain buck (step-down) cannot help once the battery sags under 3.3 V, and a boost (step-up) is wrong while it is above — the answer is a buck-boost converter, which steps either way and holds 3.3 V across the whole discharge. This is the kind of decision that ties the rung together: the shape of the input relative to the output decides the topology, before efficiency or noise ever enters the conversation.
Step back and the whole rung becomes one habit: choose the supply for the job, then budget for its honest costs before you build. Pick linear for quiet and simplicity or a switcher for efficiency; size the reservoir cap for the ripple trough, not the average; compute the worst-case heat and give it a path out; and add the cheap protections before the expensive fault arrives. Above all, derate — run every part well inside its limits and leave margin, because the datasheet's numbers assume a kinder world than your enclosure on a hot day. A power supply that is merely correct on paper is not finished; one that stays correct hot, loaded, surged, and abused is.