System Integration for Industrial Boards: Galvanic Isolation Done Right

Galvanic isolation gets specified on industrial boards for three different reasons, and the design that satisfies one of them often fails the others. Before any layout starts, the design team needs to be explicit about which reason — or which combination — applies, because the implementation diverges fast after the first conversation.

Reason 1 — Safety isolation

A path from a hazardous voltage to a touchable surface must be interrupted by reinforced insulation. This is the IEC 61010 / IEC 60601 / UL conversation, driven by certification, with mandated creepage and clearance distances that scale with mains voltage and pollution degree. The isolation here is non-negotiable; it is documented and audited.

Reason 2 — Functional ground-loop isolation

A signal travels between two boards that don't share a clean reference. The voltage difference between their grounds — caused by power-conductor IR drop, capacitive coupling, or just being in different cabinets — corrupts the signal or, in the bad case, drives a destructive current through it. The fix is to break the DC ground path while preserving signal transfer. The classic case: a 24V industrial sensor reporting to a controller across 20 metres of cable.

Reason 3 — Noise rejection at high-impedance inputs

A high-impedance measurement (thermocouple, strain gauge, isolated ADC for an analyser) must reject common-mode noise that would otherwise dominate the small differential signal. Isolation here is about CMRR, not about hazard.

"The first question is always: is this isolation for the auditor, for the link, or for the noise? Three different answers, three different topologies, three different bills of material." — Pioneer Horizon OEM lead

Most industrial designs we see need all three boundaries in different places on the same board. The skill is in identifying which boundary serves which purpose and not over-isolating where it isn't needed — every isolation barrier adds cost, latency, and a power-conversion step.

Once you've decided you need a safety-rated isolation boundary, the layout is mostly arithmetic. The mistake is treating the arithmetic as an afterthought; it should be the first thing on the layer-1 keep-out drawing, before any component is placed.

Working voltage and pollution degree

The two numbers that fix every distance on the board are the working voltage across the barrier (peak, including the highest expected transient that the design must survive without isolation breakdown) and the pollution degree (typically PD2 for indoor industrial, PD3 for outdoor or condensing).

The numbers we hold for common industrial cases

• 24V to 0V signalling, PD2: 0.4mm creepage minimum, but we hold to 1.0mm for noise-immunity comfort.
• 230V mains to SELV, PD2, reinforced: 6.4mm creepage, 5.0mm clearance per IEC 61010.
• 400V three-phase to SELV, PD2, reinforced: 8.0mm creepage, 5.5mm clearance.
• Internal high-voltage rails (e.g., 800V EV propulsion DC bus) — different rule set entirely, see IEC 61800-5-1 and our automotive design notes.

The geometry rules that get missed

1. Creepage is along the surface — it follows the PCB topography. A slot cut into the board between primary and secondary increases the creepage distance without consuming X-Y board area. We use this trick on space-constrained 230V designs to recover 20–30% of board area.
2. Clearance is through air — it's the shortest straight-line distance, even if the surface routing is longer. Components sticking up tall reduce clearance, including capacitor cans and connector bodies.
3. The barrier extends to inner layers — no inner-layer plane crosses the isolation boundary. We carve a keep-out polygon through every plane layer that spans the primary-secondary gap.
4. Vias near the barrier — minimum 1mm offset from any creepage edge, and never a via that tunnels under the barrier on an inner layer.

Print these rules as a keep-out drawing for the layout engineer before routing begins. Re-spinning a board to add creepage after the fact is, in our experience, the single most expensive class of board re-spin — the changes often cascade into connector position, mounting hole position, and enclosure cutouts.

Three families of isolation components dominate industrial boards. They look interchangeable on paper and produce very different reliability profiles in the field.

Optocouplers

Optocouplers are the oldest, cheapest, and slowest. They couple via an LED + photodiode pair, which gives them excellent isolation voltage (5–10kVrms reinforced is common) and a serious aging problem — LED output decays with time and current, so the current transfer ratio (CTR) drops over the part's life. A design that worked at CTR 100% on day one needs to still work at CTR 30% on year ten.

• Best for: low-speed digital signals, sensor inputs, occasional output triggers.
• Worst for: anything you want to forget about for ten years without a margin recalculation.
• Failure mode: gradual loss of CTR until the receiver no longer triggers.

Capacitive / magnetic digital isolators

Silicon-based isolators (TI ISO7xxx, Analog Devices ADuMxxxx, Skyworks Si86xxx) couple via tiny on-die capacitors or transformers. They're fast (up to 150 Mbps), don't age, and have predictable propagation delays. The trade-off: lower peak isolation voltage than transformer-based parts, and a hard dependence on supply-side decoupling discipline.

• Best for: RS-485, CAN, SPI, fast digital signals, any path that needs deterministic timing.
• Worst for: very high transient voltage events without external clamping.
• Failure mode: rare — typically silicon damage from overvoltage transient. When it goes, it goes hard.

Transformer-coupled (analog) isolators

For analog signals across an isolation barrier, an isolation amplifier (e.g., AMC1300, ISO224) or a discrete transformer is the move. They preserve DC accuracy across the barrier and can handle high common-mode transients. The cost is higher and the layout is more constrained — return-path planning around the transformer matters.

For our default 24V industrial RS-485 design, we use a digital isolator with a small isolated DC-DC for the bus side. That topology has run cleanly across the last five industrial-automation programmes — see the ramp study for the field-failure baseline.

An isolated communications interface needs isolated power. The two common topologies are an isolated DC-DC module (e.g., RECOM RxxPxx, Mornsun, Murata MEU) and a discrete flyback or push-pull design driven by an integrated transformer. Each has a noise signature, and that signature ends up in your ADC if you're not careful.

Isolated DC-DC modules — what we look at

• Switching frequency — 200kHz to 400kHz is typical. Place the harmonics in mind when picking the module: a 400kHz module produces a 800kHz harmonic right in the band of some industrial sensors.
• Common-mode capacitance across the barrier — published in pF on the datasheet. Lower is better for common-mode current at the switching frequency. We target <15pF for clean analog work.
• Rated load vs. actual load — efficiency and noise both degrade below 20% load. Don't over-spec the module just to feel safe; pick one whose nominal load matches your actual current draw.

Layout discipline at the isolated side

1. A dedicated, small ground plane for the isolated side — sized just to cover the secondary components, not extended.
2. A bridge cap (Y-rated for safety boundaries) tying primary to secondary ground at exactly one point near the isolator, to provide a defined return path for common-mode currents.
3. Decoupling on the isolated supply: 10µF bulk + 100nF + 10nF stacked within 5mm of the digital isolator's secondary VDD.
4. Twisted-pair routing for the differential signal lines on the isolated side, with continuous reference plane immediately below.

What you'll measure if you skip this

A switching-frequency tone, sometimes 20–40dB above the broadband noise floor, riding on every measurement made on the isolated side. We see this on boards that come in for failure analysis — the design is functional, but the analog SNR is unusable because the DC-DC noise wasn't planned. The fix at that point is a re-spin; the fix at design time is a 90-minute review with the SI engineer.

Isolation is not just a PCB problem. The same boundaries that exist on the board exist (or have to exist) inside the enclosure, and the certification body will trace every one of them from the schematic through to the box. A board that's correctly isolated and an enclosure that isn't will cost you a certification cycle.

Enclosure-level rules that follow from board-level isolation

• Cable separation — primary-side and secondary-side cables exit through separate cable glands on the enclosure, with the same creepage rules applying to gland-to-gland distance.
• Connector segregation — primary-side connectors keyed differently from secondary-side connectors, so a field tech cannot accidentally swap them.
• Internal partition — for higher-voltage designs (mains-powered), a physical barrier inside the enclosure between primary and secondary zones, with the creepage distance honoured around the barrier opening for harness pass-through.
• Labelling — primary-side voltages marked on the enclosure interior in the language the certification body's market requires.

Documentation the certification body will want

1. The schematic with the isolation boundary marked as a heavy line, primary and secondary clearly labelled.
2. A creepage-clearance drawing showing measured distances on the actual PCB at every barrier crossing — not the theoretical layout distances, the measured ones.
3. Datasheets for every component bridging the barrier, with the reinforced/basic isolation rating highlighted and the certificate of compliance attached.
4. The enclosure drawing showing the partition, the gland positions, and the labelling.

"The certification body doesn't audit the schematic; they audit the trail from schematic to silicon to enclosure. Any link weaker than the others is the link they find." — Pioneer Horizon test team

If your next industrial design is in the 24V-to-mains range, share the schematic and the rough enclosure CAD and we'll mark up the isolation boundaries and the cascade implications in a one-hour review. The redline output is what we'd build the board to internally.