Board-level DC/DC Converters for Embedded Design

At some point in most product designs the question arises of whether to design the power conversion or buy it. Discrete conversion is cheaper in component cost and is the right answer at volume. A module costs more per unit and eliminates the engineering, the EMC iteration and — where it carries approvals — a meaningful part of the certification effort.

The trade is usually decided by volume and by schedule. What follows assumes the decision has gone toward a module, and covers what then has to be checked.

Isolated or non-isolated?

An isolated converter has no electrical connection between input and output; a transformer transfers the energy and the two sides can sit at entirely different potentials. A non-isolated converter shares a common reference and is smaller, cheaper and more efficient for the same output.

Isolation is required for genuinely distinct reasons, and they are worth separating because they lead to different specifications. Safety isolation, where the barrier protects a person from a hazardous voltage, is defined by the applicable safety standard and by the isolation rating the module is tested to. Functional isolation breaks a ground loop or lets a measurement circuit float relative to its surroundings, and is a circuit requirement with no safety dimension. Interface isolation protects a communication port whose cable leaves the enclosure and may arrive at a different ground potential.

Where none of these applies — dropping a 24 V input rail to 5 V for local logic inside one earthed enclosure — a non-isolated module is the better engineering answer. The non-potted on-board category carries both: the NSD family is isolated, while NID and SPOL are non-isolated point-of-load parts.

Package, footprint and the board you already have

Module packages follow long-established footprints, and that standardisation is worth exploiting. A board laid out for a common footprint can accept parts from several manufacturers, which is worth a great deal when a single-sourced part goes on allocation.

The potted module category is organised by exactly this: families grouped by footprint, including the 1" × 1", 1.25" × 0.8" and 2" × 1" DIP outlines, the 2" × 2" size for higher power, SIP packages where board area is tighter than board length, and SMD packages for reflow assembly. Series such as the SKM, SKA, RSDW and SCW families populate these outlines across a range of powers and input windows.

Three physical dimensions constrain the choice as firmly as the electrical specification. Footprint and pinout decide whether the part fits the layout. Height decides whether the assembly fits its enclosure, which is often the binding constraint in a slim product. And assembly process matters: a through-hole DIP module on a board otherwise built for reflow adds a hand-soldering or wave step to manufacturing that the unit price does not show.

Potting is a real trade rather than a free improvement. An encapsulated module resists moisture, vibration and mechanical shock, and conducts heat out through its body. It is also heavier, sometimes taller, and cannot be repaired or inspected.

Regulated, unregulated, and the minimum load trap

Regulated modules hold the output constant across their input range and load range, which is what most designs want. Unregulated modules — usually fixed-ratio converters — are cheaper and more efficient, and their output tracks the input, so a 5 V input gives roughly 5 V out and a 5.5 V input gives roughly 5.5 V.

Unregulated parts are appropriate where the input rail is already well regulated and the load is stable. They are inappropriate where the input moves, and the mistake is easy to make because the nominal specification looks right.

The specific trap is minimum load. Many unregulated modules specify a minimum load, often around ten percent of rating, below which the output voltage rises — sometimes well above the nominal figure. A design that powers a circuit which spends most of its time in a low-power sleep state can sit below that threshold for most of its life, and the resulting overvoltage is applied continuously to whatever it feeds. Where a design has a genuine light-load state, either a regulated module or a deliberate preload is required, and the preload wastes exactly the power the sleep state was meant to save.

Input window is the other specification to check against reality rather than nomenclature. A part described as a 24 V input may have a narrow window around it, while a wide-input part covers a 2:1 or 4:1 range. The rail's tolerance, its ripple, and any transient it sees all have to fit inside that window — not just its nominal value.

Thermal design with no fan and no heatsink

A board-mounted module has no fan and usually no heatsink. It sheds heat into the board through its pins, into the surrounding air by convection, and by radiation, and all three are weak compared with forced air.

Derating curves are therefore central rather than a footnote. A module rated at its full output in free air at a moderate temperature will be rated at considerably less inside a sealed enclosure, and the ambient that applies is the air inside that enclosure — which is above room temperature by whatever everything inside it is dissipating.

Copper is the main tool available. Connecting the module's pins to substantial copper pours spreads heat into the board and makes the whole PCB part of the thermal path, which is far more effective than the module's own surface area. Placement matters for the same reason: a module downstream of a processor in the airflow, or sitting against a wall that blocks convection, runs hotter than the same part in open board area.

Worth checking against the actual product rather than the prototype. An enclosure closed for EMC reasons after the thermal work was done is a common way for a design to pass both tests separately and fail in the field.

AC/DC on an open-frame board

Where the product plugs into the mains, an open-frame AC/DC board mounts inside the equipment enclosure and gives the whole conversion in one bought part. The open-frame category divides along lines that reflect what the end product has to satisfy.

The industrial PCB-type families — ELP, EPP, LPP and LPS among them — and the LOP family cover general equipment. Separate medical families, including the RPS, RPD and MPM ranges, exist because medical equipment is held to different isolation and leakage current requirements, and a supply meeting the general standard does not necessarily meet them. On-board types such as the IRM and IOB families are small enough to solder directly onto the host board, which removes the mounting hardware and the internal wiring harness.

Two considerations belong to this format specifically. An open-frame board exposes mains potential on its own surface, so creepage and clearance to everything around it — and the enclosure's own protection against contact — become part of the host product's design rather than the supply's. And EMC is only ever certified for a complete product: a board that meets a standard in the manufacturer's test setup may not in a different enclosure with different cable routing. Leaving room in the schedule for filtering after the first EMC visit is the realistic plan.