Skip to main content
DD-CAT IDE logoDD-CAT IDE

Article · Tech Insights

A Fairy-Tale Guide to TVS Selection: How Low Should the Castle Gatekeeper Crouch?

Favorite

TVS selection is never as simple as "a 24 V system just needs a 24 V part." This piece uses a black-box castle's gatekeeper to explain the whole thing: how to read six parameters on one I-V curve (VRWM, VBR, VC, R_dyn, IR, Cj), how to draw three red lines, how to convert an 8/20 µs waveform into joules and junction temperature, and why that easily ignored one millimeter of trace equals 30 volts. The focus is deep versus shallow snapback: the same part behaves as a self-resetting gatekeeper on a high-speed data line, but becomes a dead short welded in place on a power rail—with the mechanism (parasitic pnpn positive-feedback condition), a pros-and-cons comparison table, and a "usable / not usable" scenario checklist.

ShareWeiboXLinkedIn

Key points

TVS selection is not about buying a diode rated for the system voltage—it is about keeping the reverse I-V curve between two limits: VRWM ≥ 1.2× the maximum operating voltage, VBR(min) (including cold drift) above the operating voltage, and the clamping voltage at the actual surge current, VC(I)=VZ0+I·Rdyn, below 80% of the protected device's absolute maximum rating. An 8/20µs surge carries roughly VC·Ipk·28.9µs of energy; repetitive surges must be checked against average power and junction temperature. Snapback TVS devices use parasitic pnpn positive feedback to clamp very low, but if the holding voltage Vh drops below the rail voltage and the source can supply the holding current Ih, the device latches—so they belong on weak-source nodes like high-speed data lines, never on power rails. In layout, each millimeter of trace is about 1nH, which adds roughly 30V at a 30A/ns ESD edge, so the TVS must sit right at the connector.

This is a side story in the "Black-Box Castle" series. The main chapters dealt with skills on paper—folding a network into two elements (Thévenin, Norton), winding a loop into equations (Kirchhoff), and splitting multiple sources into single ones (Superposition). This chapter changes the subject: the line of defense at the castle gate.

The formulas are the same, but the traps are not on paper. They hide in the fine print of datasheets, on package pins, and in that trace you assume is "only two millimeters anyway."

Story: The Gatekeeper of Thunderstorm Valley

North of the Black-Box Castle in Fog Valley lies a thunderstorm valley. Every summer, three kinds of water rush down the valley:

  • A crack—an extremely short, violent jet of water—lasting one nanosecond, peaking at thirty amps. This is electrostatic discharge (ESD).
  • A flood crest lasting a dozen-odd microseconds—rising to its peak in eight microseconds, falling to half in twenty. This is the 8/20 µs surge.
  • Water flowing back through the bellows when the neighboring smithy shuts down—this is the half LI2LI^2 when a relay or solenoid de-energizes.

The castle gate originally had no gatekeeper. The door knocker was smashed three times, the bookkeeper's abacus was soaked twice, and once, three days after the rain stopped, the guard dog was still barking at empty air.

Grandpa Copper-Nail posted a rule by the gate: a gatekeeper stands at the door. Normally he dozes in his chair—reverse high impedance, leaking only a few nanoamps. The moment the water level at the gate crosses a certain mark, he jumps up, diverts all the water into the moat, and pins the gate's water level down. When the water recedes, he goes back to sleep.

This gatekeeper is called TVS (Transient Voltage Suppressor).

Here's the problem: the gatekeeper has three brothers, with completely different temperaments.

  • The eldest (non-snapback): once he stands, he stays put, only slowly sitting back down after the water recedes. Steady as an old dog, but tall—the doorway needs a very high frame.
  • The second (shallow snapback): when bumped, he crouches a little, then stands right back up.
  • The third (deep snapback): one bump and he crouches all the way down, lower than anyone, so no wave can top the threshold. But once he crouches, if the water level at the gate is already higher than his head, he can never stand up again.

Figure 1: Three reverse I-V curves side by side—left, the non-snapback type rises monotonically after breakdown, clamping about 20 V at 30 A; middle, the shallow snapback type triggers at 16 V then dips slightly to 14 V; right, the deep snapback type triggers at 20 V then plunges to an 8 V holding voltage, clamping only about 9 V at 30 A, but it latches when the holding voltage is below the operating voltage

Figure 1: Snapback depth = (Vt−Vh)/Vt(V_t - V_h)/V_t. The eldest is "no snapback", the second "shallow snapback" (10%~20%), the third "deep snapback" (up to 40%~60%). Deep snapback gives the lowest clamping and the smallest capacitance, but once VhV_h drops below the circuit's operating voltage, it loses even the freedom to turn off—Section 5 covers this in detail.

1. A TVS Is a Floodgate, Not a Current Limiter

Get the concept straight and you'll eliminate half of all selection mistakes:

A TVS does not destroy energy; it only routes energy to ground. It sits in parallel between the protected node and ground—high impedance normally, conducting the instant an overvoltage appears, stealing the surge current away from the chip. It has no "current limiting" ability at all—the surge current is pushed by the source, and the TVS can only decide how much it takes.

Three iron rules follow:

  1. It must be in parallel. Putting it in series is like installing a fuse for yourself—and the exploding kind.
  2. It must be close by, because there is inductance in the path (Section 6 will prove "1 mm = 30 V").
  3. There must be impedance or a disconnect device upstream (fuse, PPTC, current-limiting resistor, connector/cable inductance). Without impedance, the surge source is an infinite water tower—however much the TVS takes, it pours more.

Energy sharing is also a clean formula. Let the surge source open-circuit voltage be VSV_S, the series impedance RSR_S, and the TVS clamping voltage VCV_C:

I=VS−VCRS,ETVS=∫VC i dt≈VC⋅VS−VCRS⋅τI=\frac{V_S-V_C}{R_S},\qquad E_{TVS}=\int V_C\,i\,dt\approx V_C\cdot\frac{V_S-V_C}{R_S}\cdot\tau

Note that ratio VCVS\dfrac{V_C}{V_S}: the larger the series impedance, the more energy lands on the resistor and the easier life is for the TVS. That's why a 10 Ω series resistor lets a small SMB-packaged device survive a surge that would otherwise require an SMC package. Conversely, when RSR_S approaches 0 (say, hanging directly on a battery), every bit of energy lands in the TVS's lap.

While we're at it, let's line up a few of the TVS's "colleagues" so you don't buy the wrong specialist:

Device Response Current capability Characteristics One-liner
TVS diode ps~ns Medium Precise clamping, repeatable, no aging The fastest gatekeeper
MOV varistor ns High Cheap, high energy, ages A guard who eats well but wears out
GDT gas discharge tube µs Very high Huge current, follow-current issues, limited life Thor's hammer—each swing counts
Zener diode µs Low High precision, low power An old gatekeeper, only good for small thieves

The TVS's strengths are fast + precise + no aging; its weakness is it can't swallow too much energy at once. That's why industrial sites commonly use a two-stage "GDT/MOV leads, TVS finishes" structure—that's Section 7's topic.

2. Reading the Curve: Six Parameters

90% of selection errors come from "only looking at the two numbers VRWM and VC." A reverse I-V curve actually holds six parameters.

Figure 2: A TVS reverse characteristic curve, marking VRWM, VBR, VC, the Rdyn triangle, IR, and the absolute maximum rating red line, with six parameter definitions listed on the right

Figure 2: The green zone on the left is "normal operation, high impedance," the blue zone in the middle is "surge arrives, conduct and divert," and beyond the red line on the right is "the protected device's absolute maximum rating." The entire curve must stay left of the red line—that is the whole goal of selection.

Parameter Meaning How to use
VRWMV_{RWM} Reverse standoff voltage—the maximum DC voltage it can withstand long-term without conducting ≥1.1∼1.2×\ge 1.1\sim1.2\times maximum operating voltage
IRI_R Leakage current at VRWMV_{RWM} A must-check for battery devices; roughly doubles every 10 °C
VBRV_{BR} Breakdown voltage, measured at a specified test current (commonly 1 mA) Look only at the minimum, and account for cold drift
VCV_C Clamping voltage, measured at the single point IPPI_{PP} Must be converted to your own current, see below
RdynR_{dyn} Dynamic resistance—the slope after turn-on Rdyn=VC@IPP−VC@0.5IPP0.5 IPPR_{dyn}=\dfrac{V_C@I_{PP}-V_C@0.5I_{PP}}{0.5\,I_{PP}}
CjC_j Junction capacitance Look at this for high-speed lines, see Section 8
PPPP_{PP} Peak pulse power (usually specified at 10/1000 µs) Only meaningful with the real waveform

The third row is where people crash most often. A datasheet saying "VC=19.9V_C=19.9 V @ IPP=30I_{PP}=30 A" does not mean the line is always 19.9 V. The real clamping voltage is

VC(I)=VZ0+I⋅Rdyn,VZ0=VC,spec−IPP,spec⋅RdynV_C(I)=V_{Z0}+I\cdot R_{dyn},\qquad V_{Z0}=V_{C,\text{spec}}-I_{PP,\text{spec}}\cdot R_{dyn}

Take a device with VBR(min⁡)=13.3V_{BR(\min)}=13.3 V and VC=19.9V_C=19.9 V @ 30 A: Rdyn≈(19.9−13.3)/30≈0.22 ΩR_{dyn}\approx(19.9-13.3)/30\approx0.22\ \Omega. If your surge is only 5 A:

VC(5 A)≈13.3+0.22×5≈14.4 VV_C(5\,\mathrm{A})\approx13.3+0.22\times5\approx\mathbf{14.4\ V}

Don't scare yourself with 19.9 V, and don't fool yourself with 13.3 V. The former is the "retail price at the 30 A moment," the latter is the "wholesale price at the 1 mA moment"—what you actually pay is the price in between.

Accuracy note: this linear formula has significant curvature near the breakdown region, so treat it as an engineering estimate. If you care, ask the vendor for a TLP curve or I-V plot instead of extrapolating from two points.

Temperature must be considered in two directions:

  • VBRV_{BR} has a positive temperature coefficient, roughly +0.06%∼+0.1%+0.06\%\sim+0.1\% per degree Celsius. Breakdown voltage rises when hot (no false triggering), and is lowest when cold—so when checking "VBR(min⁡)V_{BR(\min)} above maximum operating voltage," use the −40 °C point.
  • IRI_R grows exponentially, roughly "doubling every 10 °C." A device specified at 1 µA at 25 °C could be 30~60 µA at 85 °C. With a battery on standby for three months, this alone decides whether customers praise your product or return it.

3. Three Red Lines: Selection Is Squeezing a Window

The protected chip has only one hard limit: Absolute Maximum Ratings. The TVS's entire job is to ensure that during a surge, the voltage on the chip's pins never exceeds it.

Hence three inequalities that must all hold at once:

(1)VRWM ≥ 1.2×Vop,max⁡(2)VBR(min⁡)(Tmin⁡) > Vop,max⁡(3)VC(Isurge) ≤ 0.8×Vabs,max⁡\begin{aligned} (1)\quad & V_{RWM}\ \ge\ 1.2\times V_{op,\max}\\[2pt] (2)\quad & V_{BR(\min)}(T_{\min})\ >\ V_{op,\max}\\[2pt] (3)\quad & V_C(I_{surge})\ \le\ 0.8\times V_{abs,\max} \end{aligned}

  • ① ensures "no conduction in normal operation"—don't let the TVS leak, partially conduct, or heat up at normal voltage.
  • ② ensures "the breakdown point is above the operating voltage," including cold drift—otherwise it breaks down the moment power is applied in winter.
  • ③ ensures "it clamps during the surge"; the 0.8 leaves margin for temperature, lot-to-lot variation, and "that bit of inductance you didn't account for."

How wide the window is decides whether the device works. Let's run through a real 24 V industrial power rail:

  • Nominal 24 V, ±10% → 26.4 V; switching ripple 0.8 Vpp0.8\ \mathrm{V_{pp}} → 26.8 V; power-on overshoot 1 V → take the maximum operating voltage as 28 V.
  • The protected Buck chip has an absolute maximum VINV_{IN} of 60 V.
  • Measured surge (8/20 µs, with a PPTC and one meter of cable upstream): 5 A.

Put three candidates side by side and actually compute VC(5 A)V_C(5\,\mathrm{A}):

Candidate VRWMV_{RWM} VBR(min⁡)V_{BR(\min)} VCV_C @ IPPI_{PP} RdynR_{dyn} VC(5 A)V_C(5\,\mathrm{A}) Verdict
SMBJ30A 30 V 33.3 V 48.4 V @ 12.4 A 1.22 Ω 39.4 V ① barely passes (30 vs 33.6), margin too thin
SMBJ33A 33 V 36.7 V 53.3 V @ 11.3 A 1.47 Ω 44.1 V ✓ all three lines pass
SMBJ36A 36 V 40.0 V 58.1 V @ 10.3 A 1.76 Ω 48.8 V ③ on the line (48 vs 48), too risky

SMBJ33A's three red lines: 33≥1.2×28=33.633\ge1.2\times28=33.6 (well, this one actually misses by 0.6 V—with 1.2 it fails, with 1.1 it passes. In this kind of "on the line" situation, go look up the chip's measured ripple instead of using estimates); 36.7>2836.7>28 ✓; 44.1≤0.8×60=4844.1\le0.8\times60=48 ✓.

What if the Buck has a 40 V absolute maximum instead? Red line ③ becomes 44.1≤3244.1\le32—instant failure. This is the truth behind so many "I installed a high-power TVS and the chip still blew up" stories: it's not that the TVS isn't trying; the window simply isn't wide enough. At that point there are only three paths:

  1. Make the surge smaller: add a 10 Ω series resistor / PPTC / ferrite bead to shave it first (remember to calculate the resistor's power dissipation);
  2. Go two-stage: GDT or MOV leads, TVS finishes, with decoupling elements in between;
  3. Switch to a bigger chip or a lower-clamping technology—but option 3 has a prerequisite on power rails, see Section 5.

Conversely, there's an even more fundamental path: swap in a DC-DC with Vabs,max⁡=100V_{abs,\max}=100 V. Half of selection problems are on the TVS side; the other half come from choosing a protected device with specs that are too tight.

4. Energy and Junction Temperature: How Many Joules Can It Really Eat

Getting the voltage right doesn't mean it survives. A TVS basically dies of heat: short-term energy burns through the junction, or long-term average power pushes the junction temperature up.

Figure 3: An 8/20 microsecond waveform, reaching its peak at 8 microseconds and decaying to half at 20 microseconds, with the area under the curve labeled as charge Q; on the right, five conversion formulas for Q, E, P_avg, T_j, and I²t, plus a 50 A / 40 V example

Figure 3: For 8/20 µs, "20 µs" means the half-value time, so the exponential tail's time constant is τ=thalf/ln⁡2≈28.9 μs\tau=t_{half}/\ln2\approx28.9\ \mu s. Don't use 20 µs directly as τ, or your energy calculation will be 30% too low.

Four formulas, one per line:

Q=∫i dt≈Ipk τ≈28.9 μs⋅IpkQ=\int i\,dt\approx I_{pk}\,\tau\approx 28.9\ \mu s\cdot I_{pk}

E≈VC⋅Q,Pavg=E⋅fsurge,Tj=Ta+Pavg Rth(j−a)E\approx V_C\cdot Q,\qquad P_{avg}=E\cdot f_{surge},\qquad T_j=T_a+P_{avg}\,R_{th(j-a)}

Example: Ipk=50I_{pk}=50 A, VC≈40V_C\approx40 V.

  • Q≈50×28.9 μs=1.44Q\approx50\times28.9\ \mu s=1.44 mC
  • E≈40×1.44 mC≈58 mJE\approx40\times1.44\ \text{mC}\approx\mathbf{58\ mJ}
  • If this is a once-a-day event, PavgP_{avg} is small enough to ignore and the "single-shot energy" is what matters;
  • If this is a continuous 10 Hz surge, Pavg=0.58P_{avg}=0.58 W, and with the SMB package's Rth≈75∼100R_{th}\approx75\sim100 ℃/W, junction temperature rises 40~60 ℃—dangerous in summer.

Single events and repetitive events are two different worlds. The datasheet's PPPP_{PP} is a single pulse; for repeated hits, look at the derating curve or simply calculate by TjT_j. Many people read "can withstand 600 W" as "can eat 600 W continuously"—that's using a fire hose as a water main.

Fuse coordination is settled with I2tI^2t—that's the referee for "who dies first":

I2t=∫i2dt≈12Ipk2τI^2t=\int i^2dt\approx\frac{1}{2}I_{pk}^2\tau

In the example, I2t≈0.036 A2sI^2t\approx0.036\ \mathrm{A^2s}. The design goal: during a normal surge, neither acts (the TVS blocks, the fuse doesn't blow); during a sustained fault, the fuse dies first—don't let the TVS hold on until it becomes a carbonized little block.

Scenario Energy Who handles it
Relay/solenoid de-energizing, L=0.3L=0.3 H, I=50I=50 mA 12LI2=0.375\frac12LI^2=0.375 mJ Easy for a TVS; it doubles as a freewheeling path
ESD, IEC 61000-4-2 contact ±8 kV mJ-level per strike, but 30 A / 1 ns The TVS's home turf
8/20 µs, 50 A, clamping at 40 V 58 mJ Comfort zone for SMB/SMC
Hot-plug, 1000 µF charged to 24 V 12CV2=288 mJ\frac12CV^2=\mathbf{288\ mJ} Not the TVS's job—use soft-start / hot-swap controllers

The last row deserves to be carved into your desk: hot-plug inrush is a "long-duration, high-energy" event, not a surge. Asking a TVS to handle it is like asking a gatekeeper to block an entire river—he can hold back a moment, but not three milliseconds.

5. Deep Snapback vs Shallow Snapback: How Low the Gatekeeper Crouches

This is the hottest and most easily misbought category of recent years. Many domestic/Taiwanese vendors put it in the title as a selling point ("deep snapback", "Snapback", "ultra-low clamping"), but few selection guides spell out: why it can crouch this low, and what lets it stand back up afterwards.

Mechanism: After Avalanche, the Parasitic Transistor Takes Over

A normal TVS only has avalanche breakdown: when the voltage reaches VBRV_{BR}, carriers multiply, current surges, and the voltage barely changes—that's the "no snapback" on the left of Figure 1.

Deep snapback adds one more layer of structure: there is a parasitic four-layer pnpn inside the device (essentially a small SCR). When avalanche current flows through the body resistance RBR_B, it creates a voltage drop that forward-biases the parasitic NPN's emitter junction, so:

avalanche current → VBE↑ → parasitic BJT turns on → more current → VBE higher\text{avalanche current}\ \rightarrow\ V_{BE}\uparrow\ \rightarrow\ \text{parasitic BJT turns on}\ \rightarrow\ \text{more current}\ \rightarrow\ V_{BE}\text{ higher}

Once the positive feedback is established, the criterion is the classic thyristor turn-on condition:

αnpn(I)+αpnp(I) ≥ 1\alpha_{npn}(I)+\alpha_{pnp}(I)\ \ge\ 1

Once satisfied, the device jumps from the "avalanche state" into a "low-impedance state," the voltage collapses to the holding voltage VhV_h, and current can keep rising—that's the negative-resistance region where the curve bends back (right side of Figure 1).

The negative-resistance region (dV/dI<0)(\mathrm{d}V/\mathrm{d}I<0) also brings a hidden property: two deep-snapback devices in parallel on the same node will "fight for current"—the one that triggers first pulls the voltage down, and the other can never trigger. In multi-channel protection, don't expect them to share evenly.

Pros and Cons: One Table to See All Three Brothers

Dimension No snapback (avalanche/Zener) Shallow snapback Deep snapback
Curve Monotonic rise after breakdown Triggers, dips 10~20%, then rises Triggers, plunges 40~60%, then rises
VtV_t trigger point =VBR=V_{BR} Slightly above operating voltage A "design point" clearly above operating voltage
VhV_h holding voltage ≈VBR\approx V_{BR} Slightly below VtV_t Far below VtV_t (can go down to 2~8 V)
Dynamic resistance Medium Medium Low
High-current clamping Highest Medium Lowest
Junction capacitance Large (several pF to tens of pF) Medium Small (down to 0.3~0.5 pF)
Latch-up risk None Low Yes, and it's a design-level risk
Temperature/spread Good Good VtV_t, VhV_h spread widely, and sensitive to dV/dt
Datasheet data Complete Average Often only typical curves; VhV_h, IhI_h may not be specified
Typical use Power rails, automotive, industrial General IO, low-voltage rails High-speed data lines, ESD entrances

One Criterion Decides Whether Deep Snapback Is Usable

The key isn't how low VhV_h is, but who is powering this device, and how much they can supply.

The complete latch-up condition is both of these at the same time:

Vsource>VhandIsource,max ≥ IhV_{\text{source}}>V_h\quad\text{and}\quad I_{\text{source,max}}\ \ge\ I_h

Here IhI_h is the holding current—the minimum current that keeps the "αnpn+αpnp≥1\alpha_{npn}+\alpha_{pnp}\ge1" positive feedback from failing. It's typically in the 10~50 mA range, but unfortunately half of datasheets don't specify it.

Figure 4: Left, on a high-speed data line, 3.3 V through a 1.5 kΩ pull-up can supply at most 2.2 mA, far below the holding current, so the device turns itself off after the surge; right, on a 24 V 3 A power rail, the holding voltage after triggering is only 8 V, the supply keeps feeding 3 A, and 24 W of dissipation continues until it burns up

Figure 4: On the left, the deep-snapback TVS hangs on USB D+, and the 1.5 kΩ pull-up can only supply 2.2 mA—less than the holding current—so once the surge passes, it turns itself off. On the right, the identical device hangs on a 24 V/3 A power rail: Vh=8V_h=8 V is below 24 V and the source can pump 3 A, so it gets pinned at 8 V × 3 A = 24 W until the fuse acts or it carbonizes itself.

Usable / Not Usable: a Scenario Checklist

Scenario Deep snapback Why
USB 2.0 D+/D−, HDMI/Ethernet differential pairs ✅ A must Weak source (pull-up/termination resistors) + needs tiny CjC_j + needs very low clamping
Sensor inputs, antennas, buttons, ADC front-ends ✅ Good High source impedance can't supply IhI_h
3.3 V / 5 V IO pins (CMOS push-pull drivers) ⚠️ Use with care The driver can supply tens of mA, possibly more than IhI_h; requires Vh>V_h> rail voltage
5 V / 12 V / 24 V power rails ❌ Don't use Low-impedance strong source; triggering means latch-up
Direct battery nodes, charging ports, hot-swap entrances ❌ Never use Inrush di/dt lights it up first, then the source keeps supplying
Automotive 12 V/24 V, load-dump nodes ❌ Don't use Same as above, plus automotive specs require a monotonic curve with no negative resistance
AC 220 V input ❌ Needs bidirectional devices + high energy—that's MOV/GDT territory

If you insist on using deep snapback on a "semi-strong source" (say, some IO pin), remember four amulets:

  1. VhV_h must be higher than the node's maximum operating voltage, with 20%+ margin;
  2. Add a series current-limiting resistor so the source can supply at most ≤0.3 Ih\le 0.3\,I_h;
  3. Get VhV_h, IhI_h, and the VtV_t-vs-dV/dt curves from the vendor—don't treat typical values as guaranteed ones;
  4. Verify power-on transients separately: soft-start dV/dt and hot-plug inrush are both "invisible triggers" for deep snapback.

Shallow Snapback: the Middle Child Nobody Advertises Loudly

Shallow snapback (only a 10~20% dip after triggering) has far less fame than deep snapback, but in engineering it's a very comfortable option:

  • The curve is nearly monotonic, and the latch-up risk is low enough to ignore;
  • Temperature and lot-to-lot spread are small, datasheets are complete, and you can safely write it into design rules;
  • The price is higher clamping than deep snapback at high currents—on nodes like 3.3 V where "the window is only 1 V," it may fail to protect.

One sentence to summarize the three brothers: deep snapback is the highest-performing gambler, no snapback is the safest straight shooter, and shallow snapback is the middle option that can actually go to volume production.

6. Layout: One Millimeter of Trace Equals Thirty Volts

All the previous calculations rest on one premise: "the voltage the chip pin sees equals the TVS's clamping voltage." That statement holds in exactly one case: trace inductance is zero. In reality it isn't.

Vpin=VC+Lloop⋅didtV_{\text{pin}}=V_C+L_{\text{loop}}\cdot\frac{di}{dt}

Ltrace≈1 nH/mm,didt∣ESD≈30 A/nsL_{\text{trace}}\approx1\ \mathrm{nH/mm},\qquad \left.\frac{di}{dt}\right|_{\text{ESD}}\approx30\ \mathrm{A/ns}

1 mm ⇒ 1 nH×30 A/ns=30 V1\ \mathrm{mm}\ \Rightarrow\ 1\ \mathrm{nH}\times30\ \mathrm{A/ns}=30\ \mathrm{V}

This is not fear-mongering; it's dimensional analysis. A 10 mm branch, two vias, and a narrow pad add up to 12~15 nH—an extra drop on the order of 400 V. The 44 V clamp you carefully calculated already reaches over four hundred volts at the chip pin.

Figure 5: Top, the wrong way—the TVS placed midway, the ground trace routed 10 mm around two vias, loop inductance about 32 nH, the chip sees 980 V; bottom, the right way—the TVS right at the connector, vias in the pads, loop about 3 nH, the chip sees about 110 V

Figure 5: Note the line in the correct drawing—"Still high? That's why the chip's own on-die ESD protection can't be omitted"—the TVS is an outsourced bodyguard; the chip's internal ESD is the close-protection bodyguard. They don't replace each other.

Five layout rules—memorize them; they can save your life:

  1. Put the TVS at the connector, don't let the surge travel through 20 mm of trace before meeting it;
  2. Keep the ground return as short and wide as possible, put vias directly in the TVS's ground pad, don't route to the other end of the board;
  3. The protected trace must pass the TVS before entering the chip, the order can't be reversed (the trace should pass "through" the node, not branch off to it);
  4. Don't leave long stubs—a stub is an antenna plus an inductor;
  5. Be careful with shared grounds: the path surge current takes through the ground plane couples into other signals through common impedance.

A commonly ignored detail: where you measure decides what you see. Measure across the TVS and you're verifying the device; measure from chip pin to ground and you're verifying the design. The former always looks pretty; the latter decides whether the product survives.

7. Two-Stage Protection: Let the Big Guy Go First

When energy exceeds a single TVS's capability (lightning, long-line induction, AC entrances), split into two stages. The typical arrangement is "GDT or MOV + TVS," with a decoupling element in between—otherwise, in the few microseconds before the first stage conducts, all the pressure lands on the second stage.

The decoupling element is usually an inductor or a stretch of trace inductance that staggers the two stages' response times:

Δt=LR  or  12πf,V1−V2=Ldidt\Delta t = \frac{L}{R}\ \ \text{or}\ \ \frac{1}{2\pi f},\qquad V_1-V_2 = L\frac{di}{dt}

The selection order is back to front: first pin down what the chip can tolerate (VabsV_{abs}), then set the second-stage TVS's clamping, then calculate how much the first stage must absorb, and only then the GDT/MOV's surge rating. Choose in reverse and you'll almost certainly crash.

8. Junction Capacitance: the Invisible Speed Limit on High-Speed Lines

CjC_j in parallel with the signal line is like hanging a capacitor on the trace. It rounds off the rising edge:

tr′=tr2+(2.2ReqCj)2t_r'=\sqrt{t_r^2+\left(2.2R_{eq}C_j\right)^2}

ReqR_{eq} is the parallel combination of the driver source impedance and the termination resistance. Rough rule of thumb: every time CjC_j doubles, the eye height drops a notch and jitter ticks up. The approximate "tolerance ceilings" for common interfaces are:

Interface Speed Recommended CjC_j Notes
USB 2.0 High Speed 480 Mbps < 3 pF Match the differential pair
USB 3.x / HDMI 2.0 5~6 Gbps < 0.5 pF Deep snapback's home turf
Ethernet 1000BASE-T 1 Gbps < 3 pF Watch pair-to-pair matching
RS-485 / CAN ≤ 10 Mbps < 50 pF Much more forgiving
ADC front-end / sensors DC~kHz Doesn't matter Can even double as a filter cap

This is where deep snapback is truly irreplaceable: its low clamping is "performance," and its low capacitance is the "admission ticket." On these interfaces, a non-snapback device's CjC_j can't even get through the door.

9. The Five-Step Process (Run Through Before Layout)

Figure 6: A five-step selection flow—draw the threat diagram, define the window, make an initial VRWM/VBR pick, convert clamping at the actual current, re-check energy and layout; on the right, the three red lines and three remedy paths

Figure 6: Order matters. Those who draw the threat diagram first finish selection in ten minutes; those who flip through selection tables first spend three days—and then it blows up.

10. Ten Crash Sites

All real sightings, sorted by how often people get bitten:

  1. Treating VCV_C@IPPI_{PP} as the actual clamp. Your surge is only 5 A, yet you scare yourself with the 30 A price list—or the reverse, treating the 1 mA VBRV_{BR} as the clamp and comforting yourself.
  2. Comparing the typical VBRV_{BR} against the operating voltage. Look at the minimum, and factor in the −40 °C drift.
  3. Ignoring temperature's effect on leakage. Spec'd at 1 µA at 25 °C, it's sixty times that at 85 °C—battery standby current blows the budget across the board.
  4. Deep snapback on a power rail. One trigger, one burned device—and then you question your life choices.
  5. Measuring deep-snapback VtV_t with a slow dV/dt source. The static curve looks great; the real trigger point is completely different.
  6. The TVS sitting 20 mm from the connector. The chip dies for no apparent reason—until you probe across the TVS and see it's "perfectly fine."
  7. Only a TVS, no upstream impedance. The whole surge lands on it, and the P-type package blows into an N-type.
  8. Using a TVS as OVP. Wire 48 V into a 24 V port by mistake and the TVS burns first, then the downstream follows. Sustained overvoltage needs an eFuse / hot-swap controller—a TVS only blocks transients.
  9. Mixing up bidirectional and unidirectional. A unidirectional TVS on an AC or bidirectional differential signal gives the surge a one-way street—the reverse direction takes everything.
  10. Forgetting the chip's own on-die ESD layer. Inner and outer layers form a series window—calculate them together, don't let them blame each other.

11. Quick Reference: Common Power Rails

System Vop,max⁡V_{op,\max} Suggested VRWMV_{RWM} Starting part Notes
5 V 5.5 V 6.5 V SMBJ6.5A (6.5CA if bidirectional) For 5 V USB VBUS, confirm the upstream can current-limit
12 V 13.2 V 15 V SMBJ15A Automotive 12 V must pass ISO 7637—separate calculation
24 V 28 V 33 V SMBJ33A Narrow window; compute VC(I)V_C(I) first
48 V PoE 57 V 58~64 V SMBJ58A On the DC side, watch voltage rating and leakage
3.3 V high-speed IO 3.6 V Deep-snapback only 0.5 pF-class ESD array Look only at CjC_j and VhV_h
220 V AC — — MOV + GDT + fuse Don't let a TVS take it all

Four Exercises for Yourself

  1. A 24 V industrial rail, maximum operating voltage 28 V, the protected LDO's VINV_{IN} absolute maximum is 36 V, and the measured surge is 8/20 µs at 8 A. Work it out: can a single SMBJ33A hold? If not, what parameters do the three remedy paths need? (Hint: VC(8 A)≈36.7+1.47×8≈48.5V_C(8\,\mathrm{A})\approx36.7+1.47\times8\approx48.5 V, 0.8×36=28.80.8\times36=28.8 V—far apart; the easiest moves are adding a series resistor or switching to a 60 V-rated device.)
  2. A deep-snapback TVS with Vt=12V_t=12 V, Vh=4V_h=4 V, Ih=20I_h=20 mA. What happens if it's hung on a 5 V/1 A supply? What about on a 5 V signal line through a 10 kΩ pull-up? Write out the current and power for each.
  3. A 1.5 pF TVS on a 480 Mbps differential pair, Req=50 ΩR_{eq}=50\ \Omega, original tr=0.3t_r=0.3 ns. Calculate the degraded rise time, and explain why HDMI 2.0 requires a 0.5 pF device.
  4. 8/20 µs, Ipk=100I_{pk}=100 A, clamping 30 V, twenty times a day. Compute EE, PavgP_{avg}, TjT_j and decide whether an SMB (Rth=90R_{th}=90 ℃/W) is enough. If not, would you change the device or the architecture first?

The Tail

One-sentence summary: the gatekeeper's job has three parts—stay still in normal times (VRWMV_{RWM} and IRI_R), don't let visitors slip in (VC(I)V_C(I) and RdynR_{dyn}), and don't linger afterwards (VhV_h and IhI_h). Add one rule that runs through the whole article: don't let him stand too far from the gate (1 mm = 30 V).

As for deep snapback, it's not mysterious and not a scam—it simply trades "performance" for "dependence on the source." On a weak source, it's a genius; on a strong source, it's a time-bomb resistor. There is only one test: after that device crouches down, who has the power to pull it back up?

Sister articles: Thévenin's Theorem · Norton's Theorem · Kirchhoff's Laws · Superposition Theorem · Femtoamp Current Measurement · MOSFET Parameter Quick Reference

Get new articles by email

Enter your email — we notify you about new articles and releases (unsubscribe anytime).

Related reading

Tech Insights

学习泡妞

学习这件事和泡妞是一回事:为什么不想学、学不会、记不住,蒙卦和逻辑学一句话就点破,一切科学简单到只剩一个「咔嗒」声。用一条条对比的方式,把自学讲成泡妞,看完就忘不掉。

Tech Insights

童话版偏置电路:从电阻分压到 ΔV/R 型,把放大器的「待命点」钉住

放大器的本事全看「待命点」钉在哪:钉得偏,信号一大就削顶;钉得飘,电源一抖就全歪。这篇用斜坡上的木桶讲清偏置是什么,再把六代做法一代代推出来:电阻分压 → 二极管接法当小电阻 → 恒流源当无穷大电阻 → 自偏置四管环(为什么必然「要么死要么不定」)→ 源极垫电阻的 ΔV/R 型 → 大电流的 V/R 型。全篇一套 1.8 V / μCox=200 µA/V² 的数字算到底,附四种结构的电源敏感性对照曲线、五个最容易踩的坑和一屏速记卡。

Tech Insights

童话版电流镜:12 种常用接法一次枚举(附公式、速记卡与选型树)

电流镜只有一句话——共用一个栅压,电流按宽长比被抄过去;但"怎么抄得更稳"有十二条常见路。这篇按"比例 / 叠管子 / 加反馈 / 自己造"四类把常用电流镜全部枚举:单管、源极退化、Widlar、共源共栅、高摆幅、多重共源共栅、威尔逊、改进型威尔逊、增益提升 RGC、β 倍乘、有源电流镜、BJT 版,每种都给 R_out、V_out,min、管数和一句"什么时候用它";再配三代摆幅对照、1.8 V·200 µA·±1% 的选型算例、失配与版图规矩、五个实战场景、七条踩坑和一屏速记卡。