ProjekinAja. Embedded Systems Stage 2 · Circuits Lesson 2.6
● Stage 2 · Lesson 2.6

Protection: ESD, Reverse Polarity, Overcurrent, and Inrush

Your board will be plugged in backwards, zapped by a fingertip, and shorted by a stray screw. Design for the user you actually have.

⏱ ~26 min read ⚡ Survivability 🧮 kV, amps, milliohms

🎯 By the end you'll understand

What an electrostatic discharge physically is, why the two standard models — the human body model and IEC 61000-4-2 — are not interchangeable, and why a part "rated 2 kV" is not rated for a finger. How a TVS diode works, what its standoff, breakdown and clamping voltages mean, and the two costs it charges you: capacitance on the signal and leakage on a high-impedance node. Why a P-channel MOSFET beats a Schottky diode for reverse-battery protection, including the counter-intuitive orientation that makes it work. What a PTC fuse can and cannot do, why it is useless against a microsecond event, and how inrush into a bulk capacitor trips a bench supply that is nowhere near its limit. And the protection set Beacon actually carries, with the reason for each part and the reason for each part left out.

KEY INSIGHTProtection is a set of bets about how the board will be abused

Every protection component is an insurance premium: it costs money, board area, signal quality or standing current, every single unit, forever, against a failure that may happen to one board in fifty. Adding all of it is not caution, it is a refusal to think. The engineering is in deciding which bets are worth making for this product and this user.

Two boards on the same bench. The first is a sensor node that died during assembly: an operator picked it up by the edge connector on a dry January morning, felt nothing at all, and the I2C bus has been intermittently wrong ever since — not dead, worse than dead, because it works on the bench and fails in the field. The second is a prototype whose designer added every protection part in the catalogue: TVS arrays on every line, a polyfuse in the battery path, a Schottky for reverse polarity. It never gets damaged. It also cannot enumerate over USB, because 50 pF of clamp capacitance on each data line has rounded the edges past recognition, and it runs 40 minutes short on battery life, because the Schottky drops 0.3 V that the regulator no longer has. Both boards are failures, and they failed in opposite directions from the same missing step: nobody counted the cost of the insurance against the probability of the claim.

🔑 Hold onto this

For every protection decision, answer four questions in writing: what event (with a standard and a number, not "static"), at which entry point (only exposed conductors need protecting), what does the unprotected part survive (a datasheet number, not a guess), and what does the protection cost (picofarads, millivolts, microamps, rupiah). A protection scheme with all four answered is engineering. Without them it is decoration that sometimes works.

CORE CONCEPTESD: two models that mean completely different things

Electrostatic discharge is charge separation looking for a way home. Walk across a synthetic carpet in dry air and triboelectric charging can leave your body at several kilovolts relative to everything around you; the threshold at which a person feels a discharge is somewhere around 3 kV, and semiconductors are damaged well below that. This is the important asymmetry: the events you notice are the large ones, and the ones that quietly degrade a gate oxide happen without any sensation at all. "I didn't zap it" is not evidence.

There are two standard ways to describe a discharge, and confusing them is the single most common error in this subject.

Human Body Model (JEDEC JS-001)IEC 61000-4-2 (system level)
What it representsA charged person touching a loose component during manufacture or handlingA charged person touching a finished, assembled product in use
Network100 pF through 1500 Ω150 pF through 330 Ω
Rise timeAround 10 ns0.7–1 ns — an order of magnitude faster
At 2 kVPeak ≈ 2000/1500 = 1.33 A, decaying with τ = 150 ns; energy ½CV² = 200 µJNot the usual test level; the standard's levels are 2, 4, 6 and 8 kV contact
At 8 kVPeak ≈ 5.3 AThe standard specifies a first peak of 30 A with a 0.8 ns rise, 16 A still flowing at 30 ns and 8 A at 60 ns; energy ½CV² = 4.8 mJ
Who is testedThe chip, by its manufacturer, on a handlerThe product, in an EMC lab, with an ESD gun (Lesson 6.4)

Read the 8 kV row again. The IEC event carries 24 times the energy of a 2 kV HBM event and delivers its first peak in under a nanosecond, through a source impedance nearly five times lower. A chip rated "±2000 V HBM" — which is JEDEC class 2 (class 1C stops just below it, at 1000 V), and is roughly what the ESP32-C3's datasheet claims for its pins — has been qualified against the first column. It has said nothing whatsoever about the second. That is precisely why external protection exists: not because the silicon has no on-die clamps, but because the on-die clamps were designed for a factory handling event and your user is holding a USB connector.

📖 What actually breaks

Three failure mechanisms, and only the first is obvious. Gate-oxide rupture: the field across a few nanometres of oxide exceeds its breakdown strength and punches a permanent conductive path — the pin is dead, and this is the failure everyone pictures. Junction damage: localised heating melts a small volume of silicon and creates a leakage path — the pin still works but now leaks microamps, which on a battery device is a mystery current draw and on a high-impedance node is a measurement offset. Latent damage: the part survives with reduced margin and fails weeks later, in the field, under temperature. The third one is why "it still works" is not a test result, and why a 25-unit pilot with an unprotected exposed connector is a slow-motion warranty problem.

HOW IT WORKSThe TVS diode, and the two bills it sends you

A transient voltage suppressor is a diode engineered to break down at a known voltage, conduct enormous current for a very short time, and survive the experience. Placed from a signal to ground, it does nothing until the line exceeds its breakdown voltage, then becomes a low-impedance path that steers the strike current to ground instead of into your chip. Three voltages describe it, and they are frequently confused:

1

Working (standoff) voltage, VRWM

The highest voltage the line may reach in normal operation without the TVS conducting. It must be above your signal's maximum, including tolerance and any legitimate transient. A 3.3 V TVS on a line that occasionally reaches 5 V is not protection; it is a permanent short with a delay.

2

Breakdown voltage, VBR

Where the part starts to conduct meaningfully, typically 10–20 % above VRWM. Specified at a small test current, usually 1 mA, which is nothing like the conditions of a real strike.

3

Clamping voltage, VC

The voltage across the TVS at its rated peak pulse current — the number that actually matters, because it is what your chip's pin will see. It is always well above VBR, because the part has dynamic resistance and tens of amps flowing through even half an ohm is a lot of volts. A "5 V TVS" may clamp at 9–15 V during an 8 kV strike. Your chip has to survive that, briefly, which is why a series resistor after the TVS is such a cheap improvement.

WHERE THE CURRENT GOES ESD 8 kV 330 Ω 24 A TVS clamps at ≈ 12 V 23.7 A here 22 Ω ESP32-C3 on-die clamp 0.28 A 99 % of the strike leaves through the TVS; the series resistor decides how much of the rest reaches the die VOLTAGE AT THE PIN 500 V 50 V 5 V oxide damage threshold (illustrative) unprotected: 463 V TVS + 22 Ω: 11.7 V log voltage scale, first 200 ns of an 8 kV contact discharge
A TVS does not absorb the strike, it redirects it. Because the clamp still develops volts across its own dynamic resistance, the series resistor that follows is what turns "clamped" into "safe" — and it costs a fraction of a rupiah.

The first bill: capacitance. A TVS is a reverse-biased junction, and junctions have capacitance. A general-purpose 5 V TVS may present 50 pF or more per line. On a power rail that is harmless. On USB data lines it is fatal: 50 pF against the driver's source impedance rounds the edges so badly that even full-speed 12 Mbit/s enumeration becomes unreliable, and high-speed is out of the question. This is why low-capacitance ESD arrays exist. Beacon's PRTR5V0U2X uses a steering-diode topology — small, fast diodes route the strike up to the rail or down to ground, where a single larger Zener does the clamping — which gets the line capacitance down to about 1 pF. Its datasheet claims IEC 61000-4-2 level 4 protection, ±8 kV contact and ±15 kV air, which is the right rating to ask for on a user-touchable port.

⚠ The second bill: leakage, and the diode that ruins the measurement you needed

Every clamp leaks. A microamp of leakage is invisible on a 5 V rail and catastrophic on a high-impedance node. Put a TVS on Beacon's battery-sense node — a reasonable-sounding instinct, since the divider connects to the battery — and its 1 µA of leakage flows through 500 kΩ of Thevenin resistance, producing 0.5 V of error: an 820-LSB offset on a measurement whose entire budget was 45 mV. The protection would have worked perfectly and destroyed the product's battery gauge.

The general form of this failure is worth stating plainly: a clamp protects the pin by constraining the signal, so it must be chosen against the signal's legitimate range and the node's impedance, not just against the threat. A 3.3 V TVS on a line that legitimately reaches 5 V during USB charging conducts continuously and cooks. A clamp with 30 pF on a 400 kHz I2C bus adds to the bus capacitance you sized pull-ups against in Lesson 2.1. Protection is never free, and the price is always paid on the signal.

The series resistor is the most underrated protection component there is. It costs a fraction of a rupiah, occupies an 0402, and does three jobs at once: it limits the current that reaches the die after the clamp, it limits fault current if the pin is shorted to a rail, and it damps reflections. Its cost is a voltage drop proportional to the signal current and, on a fast line, an RC with the downstream capacitance. On a slow GPIO going to a connector, 100 Ω to 1 kΩ is nearly free protection. On a USB data pair it must be small enough to keep the driver's total output impedance inside USB 2.0's 28–44 Ω window, which is exactly the argument Beacon's schematic has to settle.

COMPARISONReverse polarity: the diode everyone uses and the MOSFET they should

A two-pin JST connector can be inserted the wrong way round by anyone with confidence and poor lighting. When it is, 4.2 V appears backwards across a board whose ICs all contain substrate diodes pointing the other way, and the cell — a lithium polymer with an internal resistance of a fraction of an ohm — is entirely capable of supplying several amps into that path. The regulator, the sensors, and the module can all be destroyed in the time it takes to say "oh".

ApproachHow it blocksCost when everything is normalVerdict
Series Schottky diodeReverse-biased, blocks (leaking tens to hundreds of µA, worse hot)0.25–0.35 V forward drop. At 500 mA charging that is 150 mW of heat; on discharge it permanently removes 0.3 V from the cell's usable range, which near the 3.3 V knee is a large fraction of the remaining capacityFine on mains-powered boards. On a battery product it is expensive in exactly the currency you cannot spare.
Parallel ("crowbar") diode across the inputConducts hard on reverse and blows the fuseNothing in normal operationNeeds a real fuse to work, and a lithium cell that is shorted through a diode is a fire risk. Not appropriate here.
Series P-channel MOSFETBody diode reverse-biased on wrong polarity, and VGS = 0 keeps the channel offI × RDS(on). A logic-level part at 50 mΩ drops 25 mV at 500 mA and 2 mV at Beacon's 40 mA active current — around 12 mW at worstThe right answer for anything battery-powered. Two extra pins of board area and a part that costs a few hundred rupiah.
📖 The orientation that trips everyone up

For reverse-polarity protection the P-channel MOSFET is fitted backwards compared with a load switch: the drain connects to the battery and the source to the load, with the gate tied to ground (through a resistor, typically 100 kΩ to 1 MΩ).

Connect the battery correctly and the body diode — anode at the drain, cathode at the source for a P-channel device — conducts first, pulling the source up to roughly VBAT − 0.4 V. The gate is at 0 V, so VGS is about −3.4 V, the channel turns fully on, and the drop collapses to I × RDS(on). Connect it backwards and the body diode is reverse-biased; nothing pulls the source anywhere; VGS stays at zero and the channel stays off. Fit it the other way round — source to the battery, the "obvious" load-switch orientation — and the body diode conducts on reverse polarity and the protection does nothing at all. This is one of the most common single-part errors on hobby and early-production boards, and it passes every electrical rule check because the netlist is identical.

One check before you commit: VGS must stay inside the part's rating, usually ±20 V. For a single lithium cell at 4.2 V there is no issue, so Beacon needs no gate Zener. On a 24 V input you would need one, plus the resistor divider to go with it.

IN PRACTICEOvercurrent, PTC fuses, and the inrush nobody budgets for

A PTC resettable fuse is a polymer filled with conductive particles. Current heats it; heat expands the polymer; the conductive paths break; resistance jumps by orders of magnitude and stays there until the current stops and it cools. Three specifications matter and two of them surprise people:

1

Hold current and trip current are far apart

A part specified to hold 500 mA typically does not guarantee a trip until about 1 A, and the ratio is roughly 2:1 by construction. There is a wide band of currents in which the fuse neither carries nor interrupts reliably. If you need protection at 600 mA, a 500 mA PTC does not provide it.

2

It is a thermal device, so it is slow

Time to trip at twice the hold current is measured in seconds; at eight times, in milliseconds. Nothing it does is relevant to a microsecond ESD strike or an inrush spike. A PTC protects the wiring, the connector and the battery from a sustained short. It does not protect a chip from anything.

3

Its hold current falls with temperature, and its resistance rises after every trip

The 500 mA rating is at 20–25 °C; at 60 °C the same part may hold only 350 mA, so a device that works in an air-conditioned office nuisance-trips in a warm warehouse. And a tripped PTC never returns quite to its original resistance — after several events, the series resistance it adds to your power path is measurably higher than the day it shipped.

Inrush is the mirror image: enormous current, over almost instantly, that no thermal device can see. Plug a 5 V source into a board carrying 22 µF of bulk capacitance and the capacitors are, for the first microseconds, a short circuit. The peak current is limited only by the cable, the connector contacts and the capacitors' own ESR:

🧪 Worked example: why the bench supply trips and the USB port does not

Take a path resistance of 0.15 Ω (cable plus connector plus ESR). Peak inrush = 5 V / 0.15 Ω ≈ 33 A, decaying with τ = R·C = 0.15 Ω × 22 µF = 3.3 µs. The total charge moved is C·V = 110 µC and the energy dissipated in the resistance is ½CV² = 275 µJ — trivial energy, enormous current, over in about 15 µs.

A bench supply in constant-current mode has a control loop that responds in tens of microseconds. The inrush is finished before the loop reacts; what the loop then sees is an output that has been dragged to near zero, so it folds back into current limit and sits there, and you get "my supply won't come up" on a board with no fault at all. A USB host, by contrast, has milliamp-scale current sensing with its own filtering and simply does not notice — which is why the same board that trips your lab supply enumerates happily on a laptop. The fix is not a bigger supply; it is either to raise the supply's current limit above the inrush, or to soft-start the board.

Soft-start means slowing the voltage ramp deliberately, usually with a series MOSFET whose gate is charged through an RC. Ramp the rail over 1 ms instead of 15 µs and the inrush becomes I = C·dV/dt = 22 µF × 5 V / 1 ms = 110 mA — a 300× reduction, from two passives and a FET.

The last piece of context: USB-C. A device that draws power must present 5.1 kΩ pull-downs (Rd) on CC1 and CC2 — one resistor on each, never one shared between them. That is how the source detects an attached sink and which way round the cable is; a single shared resistor makes orientation detection fail and is a recurring bug in first-revision boards. The source advertises its capability by its own pull-up: roughly 56 kΩ for default USB power, 22 kΩ for 1.5 A, 10 kΩ for 3.0 A. Beacon reads none of this — it takes the default 500 mA and its charger is configured for it — but the resistors are still mandatory, because without them a USB-C source never turns VBUS on at all and the board simply does not charge. Lesson 4.4 takes the charging path further.

SEE ITThe strike test: fire an event at Beacon and see what survives

Choose an abuse event and a protection set. The plot shows the current flowing into the vulnerable part over time against the limit it can survive; the readouts show what the protection cost you in signal quality, forward drop, and standing current — because the goal is not to survive everything, it is to survive what is likely at a price the product can pay.

⚙ Interactive · Strike test

Find the cheapest protection set that survives all five events. Then check the cost panel: at least one "safe" configuration breaks USB and one wrecks the battery gauge.

— 0 — time → —
current into the vulnerable partsurvival limitover the limit
Peak into the part
—
Voltage at the pin
—
Energy delivered
—
Outcome
—
Added line capacitance
—
D± edge · USB needs < 20 ns
—
Drop at 500 mA charge
—
Extra BOM at 25 units
—
—

Work through it in this order. Start unprotected and fire the 8 kV contact discharge: 463 V at the pin and 23 A through an on-die clamp built for 1.3 A. Now add the low-capacitance TVS alone — survivable, but the die still takes 590 mA. Add 22 Ω and it drops to 284 mA with room to spare, which is why the pair is the standard answer rather than either alone. Then switch to 1 kΩ: the die takes almost nothing, and the cost panel shows the USB edge has gone past 20 ns and the port will not enumerate. Then the battery events: "straight to the load" and reverse polarity puts amps through every substrate diode on the board; the Schottky stops it and quietly bills you 0.30 V forever; the MOSFET stops it for 25 mV. Finally hot-plug with the PTC fitted and watch the 33 A spike sail straight through it, because a thermal fuse cannot see a three-microsecond event — only the soft-start ramp can.

BEACONBeacon's protection set, and what was deliberately left out

Beacon is an indoor asset tag on a 25-unit pilot, handled by warehouse and lab staff, charged over USB-C from whatever cable is nearby, with a battery on a JST-PH connector that someone will eventually reverse. R9 fixes the environment as indoor and 0–45 °C; R5 caps the unit BOM at IDR 150,000. That combination decides every line below.

PartWhereThe event it stopsWhat it costs
PRTR5V0U2XUSB-C VBUS, D+, D−IEC 61000-4-2 contact discharge into the connector — the one exposed metal a user touches every time they charge the tagAbout 1 pF per line, well inside full-speed USB's budget; sub-microamp leakage on a low-impedance node where it does not matter; a few thousand rupiah
22 Ω on D+ and D−Between the connector and GPIO18/GPIO19Limits the residual current the TVS lets through, and limits fault current if a pin is shorted to VBUSSeries impedance in a 90 Ω differential pair. Design review item: USB 2.0 requires the driver's total output impedance to land between 28 and 44 Ω, so if the C3's own PHY already sits near 40 Ω, these must be populated as 0 Ω links. The footprints go on the sheet; the value is confirmed against the transceiver's specification in Lesson 4.4.
Reverse-battery P-MOSFETBattery positive, drain to the cell, source to the load, gate to groundA JST-PH connector inserted backwards50 mΩ: 2 mV at Beacon's 40 mA active current, 25 mV at 500 mA charging. Negligible against the 0.30 V a Schottky would have taken permanently.
DW01A + FS8205 pairIn series with the cell's negative terminalOver-charge above ≈4.25 V, over-discharge below ≈2.4 V, and over-current or short-circuit at the cell — the failures that make a lithium cell dangerous rather than merely brokenA few tens of milliohms and a couple of microamps of quiescent draw, counted inside the canon's 8 µA sleep budget. Note that many JST-PH cells already ship with this exact pair on a tab inside the pack; carrying it on the board too is harmless but should be a deliberate choice, not an accident.
5.1 kΩ × 2 on CC1 and CC2USB-C receptacleNot protection — without them a USB-C source never applies VBUS and the tag simply never chargesTwo 0402s. They are listed here because the most common USB-C first-revision bug is one shared 5.1 kΩ instead of two.
🔑 What Beacon deliberately does not carry, and why

No TVS on the battery-sense node. A clamp's microamp of leakage through 500 kΩ of Thevenin resistance would inject 0.5 V of error into a measurement whose whole budget is 45 mV (Lesson 2.4). The node is internal, never exposed, and needs no protection.

No PTC in the battery path. The DW01A/FS8205 pair already interrupts a cell short in milliseconds, which is faster and more appropriate than a thermal device; adding a PTC would put tens of milliohms and a temperature-dependent trip point in series with a rail that already has a 6-month budget to meet.

No soft-start on VBUS. Beacon's total bulk capacitance is around 22 µF, its charge current is limited to 500 mA by the MCP73831 anyway, and its user plugs it into a USB port rather than a bench supply. The inrush is real but nobody is harmed by it. This is written on the sheet as a decision rather than an omission, so that the first person who trips a lab supply during bring-up (Lesson 6.1) knows it was expected.

Nothing on the pogo pads or the Tag-Connect. This one is a considered risk, not an oversight: those are exposed conductors touched by a test jig and by human fingers during bring-up, which makes them a genuine ESD path straight to unprotected GPIOs. The mitigation is procedural — the jig is grounded and the operator is wristbanded (Lesson 6.3) — because clamping them would put capacitance and leakage on lines that are also used for programming. If the pilot produces a single unexplained dead unit, this is the first entry on the Rev B list.

WEAKNESSWhere this bites

✗ Reading a component's HBM rating as system-level protection

"±2000 V ESD" on a datasheet's absolute-maximum table means the die survived a 100 pF / 1500 Ω discharge on a handler. It is a manufacturing-handling specification. Your user's fingertip on a USB shell is an IEC 61000-4-2 event with 150 pF, 330 Ω and a sub-nanosecond rise, and at the standard's 8 kV level it carries 24 times the energy at 20 times the peak current. Designing a product's external protection from a component's HBM number is the most common way boards pass every bench test and fail in the field. If the port is touchable, specify against IEC 61000-4-2 and put the level in the requirements document.

✗ Protecting a node and destroying its signal

The catalogue-full-of-clamps board in the opening is not a caricature; it is the default outcome of treating protection as a checklist. Fifty picofarads per line kills USB. A microamp of leakage kills a high-impedance measurement. A 3.3 V standoff on a line that legitimately reaches 5 V kills the clamp. A kilohm of series resistance kills a fast edge. Every one of these is a protection part working exactly as specified, on a node where nobody checked what the signal needed. Before you place a clamp, write down the node's maximum legitimate voltage, its source impedance, and its highest frequency — then choose a part that respects all three.

✗ Expecting a fuse to protect a semiconductor

There is an old and comforting belief that a fuse in the supply protects the circuit. It does not, and it never did: fuses are thermal devices that take milliseconds at best, and a silicon junction fails in nanoseconds. The traditional joke is that semiconductors protect fuses by failing first, and it is not really a joke. A fuse or PTC protects the wiring, the connector, the battery and the building from a sustained fault — which are excellent things to protect and are why you fit one. If you want a chip protected from overcurrent, that is a current-limited supply, a series resistor, or a load switch with a fast electronic limit, and each of those has to be chosen on its response time, not its rating.

KEY INSIGHTHow this connects

  • Diodes, Transistors, and MOSFETs as Switches (1.5) — the body diode you learned to draw is the entire mechanism of reverse-polarity protection, and knowing which way it points is the difference between a MOSFET that blocks and one that does nothing at all.
  • Decoupling, Grounding, and Return Paths (2.5) — a strike is the fastest di/dt your board will ever see, so the inductance between the TVS and its ground via decides how well it clamps. A protection part with a long return path clamps late and high, which is a layout failure wearing a component's name.
  • Switching Things: Driving LEDs, Motors, and Relays (2.7) — next: the switches that drive real loads, the flyback diode that is protection by another name, and the addressable LED whose idle current is a bigger threat to Beacon than any ESD strike.

The thread continues into Stage 4, where USB-C's CC resistors and the charger's current limit become a full charging design (4.4), and into Stage 6, where these decisions get tested for real: the EMC lab fires an actual ESD gun at the enclosure (6.4), the bring-up procedure meets the inrush that was consciously left unmanaged (6.1), and the field-failure analysis (6.6) finds out whether the unprotected test pads were a reasonable bet or the first entry on the Rev B change list.

GLOSSARYKey terms introduced

Human Body Model (HBM)
A component-level ESD test using 100 pF through 1500 Ω, representing handling during manufacture. A 2 kV HBM rating means a 1.33 A peak decaying with a 150 ns time constant.
IEC 61000-4-2
The system-level ESD test: 150 pF through 330 Ω with a sub-nanosecond rise, applied to a finished product. At 8 kV contact the standard specifies a 30 A first peak — a far harsher event than any HBM rating covers.
Standoff (working) voltage, VRWM
The highest voltage a TVS may see continuously without conducting. Must exceed the line's legitimate maximum, including charging transients.
Clamping voltage, VC
The voltage across a TVS at its rated peak pulse current — what the protected pin actually sees, always well above the breakdown voltage because of the part's dynamic resistance.
Low-capacitance ESD array
A steering-diode topology that routes strike current to the rails through small fast diodes and clamps with one larger Zener, keeping line capacitance near 1 pF so high-speed signals survive the protection.
Body diode
The intrinsic diode between a MOSFET's drain and source. For reverse-polarity protection the P-channel device is fitted drain-to-battery so this diode blocks on the wrong polarity — the opposite orientation to a load switch.
PTC resettable fuse
A polymer device whose resistance jumps when heated by overcurrent. Hold and trip currents differ by roughly 2:1, response is seconds to milliseconds, hold current falls with ambient temperature, and resistance rises permanently after each trip.
Inrush current
The surge that charges a board's bulk capacitance at power-on, limited only by path resistance: tens of amps for a few microseconds. Invisible to thermal protection; reduced only by slowing the voltage ramp.

ACTIVE RECALLCheck your understanding

Answer in your own words first, then reveal.

1A microcontroller's datasheet lists "ESD (HBM) ±2000 V" in its absolute maximum ratings. A colleague concludes the USB connector needs no external protection. Rebut with numbers.›

The rating describes a 100 pF / 1500 Ω discharge with a roughly 10 ns rise — a handling event on a chip handler, at 2000/1500 = 1.33 A peak decaying with a 150 ns time constant, carrying ½CV² = 200 µJ. The event at a user-touchable USB shell is IEC 61000-4-2: 150 pF through 330 Ω with a 0.8 ns rise. At the standard's 8 kV contact level that is a 30 A first peak with 16 A still flowing at 30 ns, and ½CV² = 4.8 mJ — 24 times the energy at over 20 times the peak current, arriving 10 times faster.

The on-die clamps have neither the area nor the response time for that, and the failure is often latent rather than immediate: the pin keeps working with new leakage or reduced margin and fails weeks later in the field. The correct move is to specify the port against IEC 61000-4-2 at a stated level, fit a low-capacitance array rated for it, and add a series resistor to limit what the clamp still lets through.

2Why does adding a 22 Ω series resistor after a TVS reduce the current into the die by roughly half, and why not use 1 kΩ instead?›

The strike current divides between two parallel paths at the port node: the TVS branch (its clamp voltage plus a few tenths of an ohm of dynamic resistance) and the die branch (the on-die clamp's turn-on voltage plus its own tens of ohms of dynamic resistance). The series resistor adds directly to the die branch's impedance. With a die clamp of about 20 Ω, adding 22 Ω roughly doubles that branch's resistance and roughly halves its share — on an 8 kV strike, from around 590 mA to around 284 mA.

The reason not to use 1 kΩ is that the resistor is in the signal path all the time. On USB data lines it forms an RC with the line and load capacitance and must also keep the driver's total output impedance inside USB 2.0's 28–44 Ω window; a kilohm violates both and the port stops enumerating. On a slow GPIO going to a connector, 1 kΩ is excellent and nearly free. The resistor's value is set by the signal, not by the threat — which is the general rule for every part in this lesson.

3A reverse-polarity P-MOSFET has been fitted with its source to the battery and its drain to the load. What happens on correct polarity, what happens on reverse, and how would you find this on a board?›

On correct polarity it works, which is what makes it dangerous: the gate at ground gives VGS ≈ −4.2 V, the channel is on, and the drop is I × RDS(on). Everything measures fine. On reverse polarity the body diode — anode at the drain, cathode at the source for a P-channel part — is now oriented so that it conducts, so the reversed battery reaches the load through a 0.4 V drop and destroys whatever it finds. The protection does nothing. The correct orientation is drain to the battery, source to the load, so that on reverse the body diode blocks and VGS stays at zero.

You will not find this in the netlist, in an electrical rule check, or in any functional test, because on correct polarity the two orientations are indistinguishable. You find it by reading the schematic symbol and asking "which way does the body diode point when the battery is backwards?", and you find it on a built board by deliberately reversing a cell through a current-limited supply set to 50 mA — if current flows, the FET is in backwards. That test belongs in the bring-up procedure (Lesson 6.1), on the first unit, before the other twenty-four are assembled.

4Beacon's designer proposes adding a TVS to the battery-sense node "since it connects to the battery", and a 500 mA PTC on VBUS "to be safe". Evaluate both proposals.›

The TVS on the sense node: reject it, decisively. The node is internal — no user or jig can touch it — so there is no ESD path to protect against. Meanwhile the node's Thevenin resistance is 500 kΩ, so even 1 µA of clamp leakage develops 0.5 V of error, against a battery measurement whose entire error budget is about 45 mV. The protection would work perfectly and destroy the product's only battery gauge. It is a clean illustration of the rule that a clamp must be judged against the node's impedance, not just the threat.

The PTC on VBUS: probably reject, and say why in one line on the sheet. It cannot do the job people imagine. It will not see the 33 A, 3 µs inrush at hot-plug at all. It will not protect any semiconductor, because it needs milliseconds to seconds while a junction fails in nanoseconds. What it would genuinely do is interrupt a sustained VBUS-to-ground short — but the charge current is already limited to 500 mA by the MCP73831, and the cell's own DW01A/FS8205 pair handles cell-side faults far faster. Against that it charges 0.30 Ω in series with the charging path (150 mV and 75 mW at 500 mA), a hold current that falls in a warm warehouse and may nuisance-trip, and a resistance that rises permanently after every event. The right answer for this product is no PTC, recorded as a decision with these reasons, so that it is revisited if the pilot ever shows a VBUS fault.

Next: GPIO pins, optimism, and an LED that costs three years of battery →