Защита плат DSP и MCU от перенапряжения при зажигании плазмы.

Защита плат DSP и MCU от перенапряжения при зажигании плазмы.

Резюме

The control board of a plasma cutter sits centimetres from circuitry that fires kilovolt pulses with 20 ns edges, and it connects to that circuitry through bus, sensor, gate and ground paths. This article sizes the four coupling paths, the clamp energy per event, the decoupling self-resonance that only works below 15.9 MHz, the filter resonance that amplifies what it should suppress, and the brownout arithmetic that resets the processor mid-sequence.

1. Введение

The digital control unit of a plasma cutter supervises the power stage, the ignition sequence, gas flow and the safety interlocks. It is the most expensive board in the machine and the least tolerant of disturbance. It also sits centimetres from the ignition stage, joined to it by sensors, gate drives and shared ground planes.

When the igniter fires, the control board's rails and pins see voltages far above their absolute maximum ratings. The outcomes range from latch-up to firmware corruption to a dead processor, and because the processor is dead the machine reports nothing. The field symptom is a power-supply death that no measurement confirms.

Failures of this kind cost a board swap plus the customer's downtime, and repetition costs the brand. In a robotic cell a control glitch during ignition can leave the torch energised without supervision, which counts as a safety event, not a reliability event.

This article sizes the four coupling paths, the three protection layers, and the two resonances that decide whether the protection works or makes the problem worse.

2. Основной механизм

2.1 Path 1: the bus rail

Current changing in the bus wiring develops a voltage across the wiring inductance, and that overshoot rides on the rail.

V_bus_spike = V_bus + L_bus × di/dt

where V_bus_spike is the peak rail voltage, V_bus is the nominal rail, L_bus is the bus wiring inductance and di/dt is the current slope. With 2 µH and 30 A/µs the overshoot reaches 60 V, so a 600 V rail peaks at 660 V and the regulator that feeds the control board has to reject it.

2.2 Path 2: capacitive coupling into signal wiring

Ignition leads couple into gate and sensor wiring through mutual capacitance, and the coupled voltage follows the ignition edge rate.

V_coupled = C_m × Z_src × dV_ign/dt

where V_coupled is the voltage that appears on the victim line, C_m is the mutual capacitance, Z_src is the source impedance of that line and dV_ign/dt is the ignition edge rate.

A 0.5 pF mutual capacitance into a 10 kΩ sensor input at 50 kV/µs delivers 250 V to the pin. On a line driven by a 50 Ω gate driver the same coupling delivers 1.25 V, which is why high-impedance sensing lines are the ones that die.

2.3 Path 3: ground bounce

Ignition return current shares the ground plane with the control section, and the shared inductance converts that current into a reference shift.

V_gb = L_gnd × di_ign/dt

where V_gb is the ground bounce, L_gnd is the inductance shared between the ignition return and the control ground, and di_ign/dt is the ignition current slope.

With 20 nH of shared path and 3.64 × 10⁹ A/s from a 480 A ignition discharge, the control ground moves 73 V relative to the power ground. Every clamp that references that ground then clamps to a moving target, and the processor sees an overvoltage relative to its own reference even though the clamp worked.

2.4 Path 4: rail droop and brownout

The ignition charge current comes out of the control rail's bulk capacitance, and the droop follows the capacitance and its ESR.

ΔV_rail = I_surge × (ESR_bulk + t_surge / C_bulk)

where ΔV_rail is the rail droop, I_surge is the surge current, ESR_bulk is the bulk capacitor resistance, t_surge is the surge duration and C_bulk is the bulk capacitance.

A 5 A surge lasting 200 µs from a 470 µF bank with 20 mΩ of ESR droops the 5 V rail by 2.23 V, which puts it at 2.77 V. Below a 3.0 V brownout threshold the processor resets in the middle of the ignition sequence.

2.5 Sizing the decoupling at the pin

The third protection layer has to hold the pin voltage while the spike passes.

C_dec = I_spike × Δt / ΔV

where C_dec is the required decoupling capacitance, I_spike is the spike current, Δt is its duration and ΔV is the voltage the pin is allowed to gain. Five amps for 50 ns across a 3.3 V allowance needs 75.8 nF, so a 100 nF part covers it.

That part only works while it is capacitive, and its self-resonance sets the ceiling.

f_srf = 1 / (2π √(ESL × C))

where f_srf is the self-resonant frequency, ESL is the equivalent series inductance of the capacitor and its mounting, and C is the capacitance. A 100 nF part with 1 nH of loop inductance self-resonates at 15.9 MHz, and a 20 ns edge carries content to 17.5 MHz. The decoupling therefore stops working across the part of the spectrum that carries the edge, which is the argument for a low-ESL part and not simply a larger one.

2.6 The filter resonance that amplifies transients

The series element and the decoupling capacitor form a resonant circuit, and an undamped one amplifies at its own frequency.

Q = (1 / R) × √(L / C)

where Q is the quality factor of the filter, R is the series resistance, L is the series inductance and C is the shunt capacitance. A 1 µH ferrite with a 100 nF capacitor resonates at 503 kHz with a characteristic impedance of 3.16 Ω. With 0.1 Ω of loss the Q reaches 31.6, and a 250 V transient at that frequency rings across the capacitor at 7.9 kV. A 3.3 Ω damping resistor brings the Q to 0.96 and removes the amplification.

Fig. 1 — Bus, coupled, ground and rail paths each cross three protection layers before reaching a DSP limited to 3.2 V.

3. Проблемы проектирования

Pain Point 1: A TVS that clips but does not survive

A single TVS across the processor rail clips the spike and absorbs everything the spike carries. The ignition discharge stores 2.3 mJ in the shared ground inductance on every event, and at 400 pulses per second with two edges each that is 1.84 W into one device. An SMB package rated near 1 W continuous runs hot, drifts, and then fails short, taking the rail with it.

Pain Point 2: Clamping to a moving ground

When the protection diodes reference a ground that moves 73 V on every ignition, the clamp level moves with it. The pin sees an overvoltage relative to its own reference even though the diode did its job, and the failure looks like a clamp that never worked. A star connection to one point drops the shared inductance from 20 nH to 2 nH and the bounce to 7.3 V.

Pain Point 3: Filter resonance between the protection layers

A ferrite bead and a decoupling capacitor form a resonator. With 1 µH and 100 nF it sits at 503 kHz with a characteristic impedance of 3.16 Ω, and 0.1 Ω of loss leaves a Q of 31.6. A transient at that frequency then rings up 30 dB across the capacitor, so the protection network delivers 7.9 kV where it was meant to deliver 250 V.

Pain Point 4: The brownout and glitch sequence

A processor that resets mid-ignition restarts into a half-executed state. Without a supervised recovery state machine the machine either hangs or fires the ignition again in an unverified state, and that second outcome is a safety event. The rail droop that causes it is 2.23 V on a 5 V rail, which is a passive problem with a passive fix.

4. Этапы проектирования

Strategy 1: Give each layer a defined share of the energy

Layer 1 is a TVS array at the board edge that absorbs the bulk of the transient. Layer 2 is a series impedance, either a ferrite bead or a толстопленочный резистор, that limits the current reaching the processor. Layer 3 is a low-ESL decoupling capacitor at the pin that slows what is left. Size each layer so no element carries more than its own rating: four edge TVS devices split the 1.84 W to 0.46 W each.

Strategy 2: Stabilise the ground before clamping

Run a dedicated low-inductance control ground and star-connect it to the power ground at exactly one point. Then place every clamp against the control ground so its reference does not move with the ignition current, and the 73 V bounce becomes 7.3 V of common-mode shift that the clamps can ignore.

Strategy 3: Damp the filter on purpose

Add 3.3 Ω in series with the 1 µH element so the filter Q falls to 0.96 and the 503 kHz resonance stops amplifying. Use an N4700 Class I decoupling part with a dissipation factor below 0.1% and self-inductance below 8 nH, so the capacitor does not add a second resonance of its own in the band the edge occupies.

Strategy 4: Hold the rail through the surge

Raise the bulk capacitance from 470 µF to 2200 µF and cut the ESR. The same 5 A surge lasting 200 µs then droops the 5 V rail by 0.56 V instead of 2.23 V, which keeps it at 4.44 V and well above any 3.0 V brownout threshold. Monitor the rail with a резистор выборки divider so the trend is visible before the first reset.

Strategy 5: Isolate the interface wiring

Use isolated gate drivers and differential sensing with common-mode rejection at the ignition frequency, and confirm that the isolation barrier's transient immunity exceeds the worst-case ignition edge rate of 50 kV/µs. Compare the alternative against a пленочный конденсатор barrier if the working voltage exceeds what a ceramic coupling part can hold.

Strategy 6: Supervise the brownout recovery in firmware

Pair the processor's brown-out reset with a state machine that aborts the ignition sequence cleanly on rail loss, returns the torch to a safe state, and restarts through a verified pre-charge. A recovery that re-arms the igniter without checking the torch state turns a reset into a safety hazard.

Параметр компонента	Типичное значение	Условие теста
Номинальное напряжение (HVCT8G серии)	10 кВ постоянного тока	высоковольтный режим 1.5× в течение 60 с
емкость	4700 пФ ±20% (472М)	1 МГц, 1 В среднеквадратичное значение
Развязывающий конденсатор	100 nF, ESL below 1 nH	собственный резонанс выше 15.9 МГц
Объемная емкость	2200 µF, ESR below 20 mΩ	5 A surge for 200 µs
Элемент серии	1 µH with 3.3 Ω damping	filter Q below 1
Частичная разрядка	ниже 5 пС	при номинальном напряжении
Pin voltage limit	3.2 V (absolute max 4.0 V minus 20%)	1,000 ignition events

5. HVC N4700 (T3M) capacitors for the control interface

A protection network holds only if its elements hold their impedance. HVC builds the N4700 (T3M) Class I family for that duty, with the divider and damping parts that complete the control interface, from a 6,000 m² owned factory in Dongguan across a 1 kV to 100 kV range.

Шесть параметров, которые здесь имеют значение.

• Low-ESL decoupling. A self-inductance below 8 nH and a short mounting loop keep the layer-3 part capacitive through the part of the spectrum that carries the 20 ns ignition edge, where a standard part has already turned inductive.
• Predictable filtering across temperature. A ±30 ppm/°C coefficient holds the filter cutoff in place, so the protection measured on the bench is the protection delivered in a closed cabinet at 70 °C.
• Loss that damps. A dissipation factor below 0.1% lets the filter elements absorb transient energy instead of ringing, which is what stops the 503 kHz resonance from amplifying.
• Isolation elements for signal interfaces. Ratings from 1 kV to 100 kV and discharge below 5 pC support the coupling and barrier parts that face ignition potential on the sensor side.
• Stability over years. The Class I formulation resists ageing and moisture, so the margins set at commissioning are still the margins after a million ignition events.
• Proven where transients are the design driver. The 40 kV 441-series doorknob capacitors replaced first-tier brand parts in high-voltage field-service programs, and a Перекрестная ссылка against those parts shows the drift a protection network would otherwise have to absorb.

Доставка и поддержка

Standard parts ship in 4 to 8 weeks and matched protection sets in 6 to 10 weeks. HVC applications engineering supports clamp sizing, damping calculation, decoupling selection and ignition injection test setup. Send the rail voltages, the ignition edge rate and the control ground layout, and HVC returns a selection sheet with impedance curves and PD reports within five working days.

6. Часто задаваемые вопросы

Q1: Why does a single TVS fail on the rail?

Because it absorbs every event on its own. The ignition discharge stores 2.3 mJ in the shared ground inductance per event, and at 400 pulses per second with two edges each that is 1.84 W into one device. An SMB package rated near 1 W continuous runs hot, drifts and fails short.

Q2: Why does ground bounce defeat a clamp that works?

Because the clamp references the ground it sits on. With 20 nH shared between the ignition return and the control ground, the 3.64 × 10⁹ A/s ignition slope moves that reference by 73 V. A star connection to one point cuts the shared inductance to 2 nH and the bounce to 7.3 V.

Q3: How do I know whether my decoupling still works?

Compare its self-resonance with the edge bandwidth. A 100 nF part with 1 nH of loop inductance self-resonates at 15.9 MHz, and a 20 ns edge carries content to 17.5 MHz. Above 15.9 MHz the part is inductive, so the fix is lower ESL, not more capacitance.

Q4: Why would a filter make a transient worse?

Because an undamped L-C pair resonates. A 1 µH element with 100 nF sits at 503 kHz with a characteristic impedance of 3.16 Ω, and 0.1 Ω of loss leaves a Q of 31.6. A transient at that frequency then reaches 7.9 kV across the capacitor instead of 250 V. A 3.3 Ω damping resistor removes it.

Q5: How much bulk capacitance does the rail need?

Enough that the surge droop stays above the brownout threshold. A 5 A surge lasting 200 µs droops a 470 µF bank by 2.23 V, which puts a 5 V rail at 2.77 V. Raising the bank to 2200 µF cuts the droop to 0.56 V and holds the rail at 4.44 V.

Q6: What test proves the protection works?

Fire the igniter at full energy while monitoring the processor rails and pins with a fast scope, and require every pin to stay inside the absolute-maximum window minus 20%, which is 3.2 V on a 3.3 V part, across 1,000 consecutive ignition events. Testing a single event finds nothing, because the failures accumulate.

7. Рекомендации

1. Э. Куффель, В.С. Заенгль, Дж. Куффель, Основы высоковольтной техники, 2-е изд., Ньюнес, 2000.
2. Мохан, Н., Унделанд, Т.М., и Роббинс, В.П. Силовая электроника: преобразователи, применение и дизайн, 3-е изд. Хобокен, Нью-Джерси: John Wiley & Sons, 2003.
3. Уведомление 2 MIL-HDBK-217F. Прогнозирование надежности электронного оборудования, 1995.
4. МЭК 60384-1:2021, Фиксированные конденсаторы для использования в электронном оборудовании — Часть 1: Общие технические характеристики, 6-е изд. Женева: МЭК, 2021.
5. МЭК 61010-1:2010, Требования безопасности к электрооборудованию для измерений, контроля и лабораторного применения — Часть 1: Общие требования, 3-е изд. Женева: МЭК, 2010.
6. Д. Ли, Ю. Чжун, С. Чжоу и Г. Чжай, Метод анализа разрушения многослойного керамического конденсатора с помощью фазового поля.IEEE Transactions on Components, Packaging and Manufacturing Technology, vol. 14, no. 2, pp. 211-220, 2024. https://doi.org/10.1109/tcpmt.2024.3359598.
7. А. Марискотти и Л. Ваккаро, Катушка Роговского для высоковольтных применений, в сборнике трудов конференции IEEE Instrumentation and Measurement Technology Conference 2008 года, стр. 1069-1073, 2008. https://doi.org/10.1109/imtc.2008.4547197.
8. Заряжая энергией будущее: как высоковольтные многослойные керамические конденсаторы повышают эффективность современной электроникиМурата, https://www.murata.com/en-eu/s/blog/powering-the-future-how-high-voltage-mlccs-drive-efficiency-in-modern-electronics.html .
9. США 6,937,027, Гибридный датчик для обнаружения высокочастотных частичных разрядов, Hanyang Hak Won Co., Ltd., выдано в 2005 г., https://www.freepatentsonline.com/6937027.html.
10. ГБ/Т 6346.1-2024, 电子设备用固定电容器 第1部分:总规范 (Постоянные конденсаторы для использования в электронном оборудовании – Часть 1: Общая спецификация). Пекин: 国家市场监督管理总局, 2024 г.
11. Команда инженеров по разработке приложений HVC, Керамический конденсатор высокого напряжения Тeхничeскиe рeсурсыhttps://www.hv-caps.com