Inrush Current Limiting: Taming the Turn-On Surge

Why a Capacitor Looks Like a Short Circuit

The trouble starts with the front-end bulk capacitor in almost every switch-mode power supply. Before the switch closes, that capacitor sits at 0 V. At the instant of connection, the full source voltage appears across whatever resistance is in the loop — the wiring, the rectifier bridge, the capacitor's own ESR — and Ohm's law is merciless: I_pk = V / R_series. On a rectified 230 V AC line the DC bus peaks near 325 V; with only 100 mΩ of loop resistance that is a theoretical 3,250 A initial spike.

The transient is the classic RC charging response. Solving the loop equation V = i·R + (1/C)∫i dt gives:

• i(t) = (V / R)·e^(−t / RC) — current starts at V/R and decays exponentially.
• v_C(t) = V·(1 − e^(−t / RC)) — the capacitor charges toward the bus voltage.
• The time constant τ = RC sets the pace; the capacitor is ~99% charged after 5τ.

With V = 325 V, R = 0.1 Ω, C = 470 µF, τ = 47 µs and the surge is essentially over in ~250 µs. Short — but the energy delivered in that window (½CV² = 24.8 J into the cap, plus the I²R burned in the loop) is what damages contacts, fuses, and semiconductors. The whole design task is raising that R during charge without paying for it forever afterward.

The NTC Thermistor: A Resistor That Gets Out of the Way

The cheapest and most common limiter is a negative-temperature-coefficient (NTC) thermistor placed in series with the AC line or DC bus. Cold, it presents a high resistance — a typical part is rated R₂₅ = 10 Ω. That 10 Ω caps the 325 V surge at ~32 A instead of thousands. As the inrush current flows, the thermistor's own I²R self-heating drives its temperature up, and its resistance collapses by a factor of 20× to 50× — down to perhaps 0.3–0.5 Ω — where it stays while the supply runs.

The physics is exponential. NTC resistance follows the Steinhart–Hart / beta relation:

• R(T) = R₂₅ · exp[β·(1/T − 1/298.15)], with T in kelvin.
• Typical β = 2,800–4,000 K; a thermistor at 150 °C sits at a small fraction of its 25 °C value.

The catch is the residual dissipation: even 0.4 Ω at 3 A rms burns 0.4 × 3² ≈ 3.6 W continuously, and that heat keeps the part hot — which is the whole point, because a cooled-off NTC would re-limit and starve the load. The fatal weakness is repeat cycling: after a brief power interruption the thermistor is still hot and low-resistance, so it offers little protection on the very next turn-on. Datasheets (Ametherm, EPCOS/TDK) specify a maximum steady-state current, a maximum capacitance it can charge, and a mandatory cool-down time of 30–60 s before full protection returns.

Resistor-and-Relay Pre-Charge: Zero Residual Loss

When the load can't tolerate ongoing thermistor loss — variable-frequency drives, servo amplifiers, or an EV traction inverter with a 100–800 V DC-link bank — the standard answer is a fixed pre-charge resistor bypassed by a relay or contactor. At power-on the resistor alone carries the charging current; once the bus reaches ~90–95% of the source voltage, a controller closes the bypass contactor and shorts the resistor out, dropping its steady-state loss to essentially zero.

Sizing the resistor is a two-constraint problem:

• Peak current constraint: R ≥ V / I_limit. To hold a 400 V EV bus below 60 A, R ≥ 6.7 Ω.
• Energy constraint: the resistor absorbs the full ½CV² of the capacitor bank in one shot, independent of R. For C = 2,000 µF at 400 V that is 160 J per charge — the resistor's pulse-energy rating (not its average wattage) governs survival.
• Timing constraint: charge time ≈ 3–5τ = 3–5·RC must be short enough for the sequencer but long enough to stay under I_limit.

Wirewound or thick-film power resistors with pulse ratings of tens to hundreds of joules are used. The controller closes the contactor on either a fixed timer or, more robustly, a bus-voltage comparator — closing early into a not-yet-charged bus dumps a second, larger inrush straight through the contactor and welds it, one of the most common field failures on EV and drive hardware.

Active MOSFET Soft-Start: Ramping the Slope

The most elegant modern approach controls the inrush directly with a series MOSFET operated briefly in its linear (saturation) region. Instead of a fixed resistor, the FET's gate is driven so the drain-source voltage falls in a controlled ramp, forcing a constant dV/dt across the load capacitor. Because a capacitor's current is i = C·dV/dt, controlling the slope directly sets the inrush current — a flat-topped, predictable pulse rather than a decaying spike.

A gate soft-start capacitor C_gs charged through a resistor from a current-limited source produces that ramp. If the design targets I_inrush = 5 A into C_load = 1,000 µF, the required ramp is dV/dt = I/C = 5,000 V/s, and the load reaches 48 V in ~9.6 ms. Key design points:

• The MOSFET dissipates P = V_DS · I_D during the ramp; the worst instant is early, when V_DS is high. The transient must stay inside the FET's Safe Operating Area (SOA) — the single biggest active-limiter failure mode.
• Integrated hot-swap controllers (TI LM5069, Analog Devices LTC4234, ADM1270) automate this with programmable dV/dt, a circuit-breaker function, and an SOA-timing fault that latches off if the ramp lingers.
• Steady-state loss is only I²·R_DS(on) — a few milliohms — so it dwarfs the NTC on efficiency and re-arms in microseconds.

This is the default for hot-swappable rack cards, 48 V telecom, and Li-ion battery pack pre-charge, where boards are inserted into a live backplane and both inrush and short-circuit protection are mandatory.

Sizing the Limiter: The I²t Number That Matters

Every downstream protective and switching element is rated by I²t — the amp-squared-seconds of let-through energy a fuse can pass or a semiconductor can survive. The inrush limiter's job is to keep the surge's I²t below every one of those ratings. A quick worked example for a 230 V AC / 325 V DC input SMPS with C = 680 µF:

• Unlimited peak: with 100 mΩ loop, I_pk ≈ 3,250 A — instantly over any line fuse's I²t.
• With a 10 Ω cold NTC: I_pk = 325 / 10.1 ≈ 32 A, τ = 6.9 ms, and the integrated I²t over the decay is roughly (I_pk²·τ)/2 ≈ 3.3 A²s — safely below a slow-blow 6.3 A fuse's ~150 A²s rating.

The sizing sequence engineers actually run:

• 1. Set I_limit from the weakest link — the relay contact rating, fuse I²t, or bridge-rectifier surge rating (I_FSM, often 50–100 A for one cycle).
• 2. Compute R_min = V_peak / I_limit and pick the limiting element to exceed it.
• 3. Verify pulse energy ½CV² against the element's joule rating (resistor) or SOA (FET), or the NTC's rated max capacitance.
• 4. Check charge time 5τ against sequencing and hold-up requirements.
• 5. Confirm compliance with IEC 61000-3-3 (flicker) and any agency inrush ceiling — often expressed as a multiple of steady-state input current.

Failure Modes, Limits, and Best Practice

Inrush limiters fail in characteristic and instructive ways:

• Hot re-strike (NTC): a momentary brownout leaves the thermistor hot and near-zero-ohm; the immediate re-connection delivers a nearly unlimited surge. Datasheets exist specifically to warn against fast power-cycling.
• Welded contacts (resistor-relay): closing the bypass into a partially-charged bus dumps a second inrush through the contactor, welding it. Always gate the relay on a bus-voltage threshold, not a blind timer, and add hysteresis.
• SOA blow-out (active MOSFET): too slow a ramp holds the FET at high V_DS and high I_D simultaneously, exceeding the transient SOA and cracking the die. Hot-swap controllers enforce a maximum ramp time and latch off.
• Nuisance limiting in cold climates: an NTC at −40 °C can be several times its R₂₅, over-limiting startup and preventing the supply from ever reaching operating voltage.

Best-practice hardening: parallel a bypass relay across an NTC to get cheap inrush limiting plus zero residual loss and cold re-strike immunity; place a small series line inductor to blunt the initial di/dt for EMI compliance; and for multi-kW systems, always prefer active or resistor-relay pre-charge over thermistors, whose self-heating cannot scale. The governing trade-off never disappears — protection during turn-on always costs either efficiency, response time, or part count, and the designer's task is choosing which price to pay.

Three dominant inrush-limiting strategies and where each wins

Method	Limiting element	Steady-state loss	Repeat cycling	Best for
NTC thermistor	Self-heating resistor (R drops ~20–30× when hot)	0.5–3 W residual	Poor — needs cool-down 30–60 s	Low-cost SMPS ≤ 500 W
Fixed resistor + bypass relay	Power resistor, shorted after charge	≈ 0 W (relay closed)	Excellent	VFDs, servo drives, EV pre-charge
Active MOSFET soft-start	MOSFET in linear region, dV/dt ramp	≈ conduction I²·R_DS(on)	Excellent, fast re-arm	Hot-swap boards, 48 V telecom, Li-ion packs
Series inductor / choke	L opposes di/dt	Copper loss only	Excellent	AC line filters, EMI-sensitive gear