FR4, FPC, Aluminum Core and Copper Core PCBs: Four Names, Three Different Axes

These four boards are usually presented as alternatives, which is the source of most confused specifications. FR-4 is the name of a laminate. FPC describes a structure that flexes. Aluminum core and copper core describe a thermal function. Comparing them directly is like comparing a material, a shape and a heatsink — the answer changes depending on which axis the design is actually constrained by. This article separates the three axes, then puts numbers on the part that matters most in practice: in a metal-core board the metal is not the bottleneck. A 0.1 mm thermally conductive dielectric contributes about 0.5 K/W per square centimetre, while 1.6 mm of aluminium contributes 0.096 and the same thickness of copper 0.042.

A specification that reads “FR4 or aluminium, whichever is cheaper” cannot be answered, because the two options are not on the same axis. FR-4 answers the question what is the dielectric. FPC answers does the board have to bend. Metal core answers where does the heat go. Each of the four names in this article fixes a different variable, and two of them can be true at once — a rigid-flex board is an FPC with FR-4 stiffeners, and an aluminium board is an FR-4-like circuit layer on a metal base.

Once the axes are separated, the practical questions become sequential: does the board bend, does it have to move heat, and only then which dielectric. Answering them in the other order is how a design ends up with a flexible circuit specified for its D_k and a metal core specified for its thickness.

FR-4 designates a flame-retardant woven-glass epoxy laminate. It is a material specification that a board can be made from, not a category of product. Its properties are ranges rather than nominal values, and that is the first thing to establish because it governs every downstream tolerance.

The 13 K/W figure is worth internalising, because it is the number a metal-core board is competing against. One watt through a square centimetre of ordinary FR-4 raises the far side by thirteen degrees. That is tolerable for a logic board and impossible for an LED array, and it explains why metal core exists at all — not because FR-4 is a bad material, but because heat is not what it was designed to move.

FR-4 remains the default for everything else. It has no practical layer limit, holds fine lines, supports plated through holes, and is the cheapest thing a fabricator can put on a press. The electrical limits of the material are covered in where FR-4 stops working.

A flexible circuit uses a polymer film instead of glass-reinforced epoxy, most often polyimide, sometimes polyester for low-temperature, low-cost parts. The film is 12.5 µm to 50 µm thick, the copper is bonded to it with an adhesive or cast directly, and the whole construction is covered with a coverlay or a solder mask that also has to flex.

Three properties decide whether an FPC will survive: the bend radius relative to total thickness, the copper type, and the number of cycles. The usual design rule is a static bend radius of six to ten times the total thickness, and twenty to forty times for a dynamic bend. On a 0.2 mm construction that is 1.2 mm static and 4 mm dynamic — numbers that quickly rule out the “just fold it” layout that motivated the flexible board in the first place.

Copper choice follows from the same arithmetic. Rolled-annealed copper tolerates repeated bending; electrodeposited copper is cheaper and fine for a board that is folded once during assembly. And the electrical properties are a secondary consideration: polyimide has a dissipation factor of roughly 0.008–0.02, no better than FR-4, so an FPC is not automatically a high-frequency solution. It is a mechanical solution that also carries signals, which is why a 5 Gbps flex needs the same loss arithmetic as any other line, as set out in the three divides.

Aluminium and copper core boards are the same construction with a different base metal. A circuit layer of copper foil is bonded to a metal plate through a thin, electrically insulating but thermally conductive dielectric. That middle layer is the engineering, and it is where the cost and the limits live.

Thermal resistance through a layer is thickness divided by conductivity and area: R = t/(k·A). For one square centimetre the area cancels to 10⁻⁴ m², so the numbers are easy to compare directly.

The table answers the aluminium-versus-copper question before it is asked. Going from aluminium to copper changes the through-thickness path from 0.60 to 0.54 K/W — about ten percent — while halving the dielectric thickness and improving its conductivity takes the same stack to 0.35 K/W, a forty percent improvement. The middle layer is the bottleneck, and it is the cheapest lever to pull.

So why does copper core exist? Because through-thickness conduction is only half of spreading. A small die on a large plate also has to spread heat sideways, and that spreading resistance scales with the metal’s conductivity rather than with the dielectric. For a 5 mm square source, an approximate spreading resistance of 1/(4ka) gives about 0.60 K/W on 1.6 mm aluminium and 0.26 K/W on copper — a gain larger than everything the through-thickness comparison offered. Copper is chosen for concentrated heat sources, not for uniform heating.

The practical comparison is not which metal conducts better, but which one pays for itself. Four properties separate them.

The selection rule that follows is narrower than the marketing. Aluminium handles the overwhelming majority of LED and power-converter work, where the heat source is distributed over square centimetres. Copper earns its multiple when the source is confined to a few square millimetres — a laser diode, a GaN die, a high-power RF transistor — or when weight is not a constraint and thermal cycling is severe enough that the CTE difference of six ppm matters.

Four limits come from the construction rather than from the materials, and they decide more designs than the thermal numbers do.

No plated through holes through the base. The base is a conductor. A via that reached it would short the circuit, so metal-core boards are single-sided or, at most, have plated holes confined to added dielectric layers. Interconnection goes around the edge or through connectors, not through the board — which removes multilayer routing as an option.

Dielectric breakdown sets the voltage ceiling. The insulating layer is 0.075–0.15 mm thick with a breakdown of roughly 3–6 kV depending on grade. That is ample for 48 V systems and marginal for mains-referenced LED drivers unless the creepage at edges and slots is designed with the same care as the layer itself.

Thermal cycling is the failure mode. Aluminium at 23 ppm/°C against a filled epoxy dielectric and a copper foil puts shear on the bond line every cycle. A 1.6 mm board over 80 mm of length moves about 0.03 mm for a 100 °C excursion, which the dielectric must absorb in shear. This is why cycling tests, not one-off thermal measurements, separate good metal-core suppliers.

No fine lines. The circuit layer is usually 1–2 oz copper on a roughened dielectric, etched with wider minimum features than an FR-4 inner layer. A 0.15 mm track is routine; 0.075 mm is not, and should not be assumed when the layout is transferred from an FR-4 design. The etch arithmetic behind that floor — undercut scaling with copper thickness — is the same one described in trace width and spacing.

Heat flux, in watts per square centimetre, is the variable that separates the four options more cleanly than any product category. The bands below are engineering approximations, not standards, but they order the decision correctly.

The via-array row is the one that saves the most money in practice. A 0.3 mm plated via with 25 µm of copper in a 1.6 mm board has a barrel resistance around 12 mm of equivalent copper length, which is why arrays are used rather than single vias: twenty in parallel, under a thermal pad, deliver the same order of thermal resistance as a metal core for a fraction of the cost — as long as the heat can reach them.

Absolute prices move too often to be useful, but the multiples are stable enough to plan with. They refer to the bare board at comparable volumes and exclude assembly.

Two of those multiples are frequently underestimated. A single-sided metal core looks cheap per square metre until the design needs a second circuit layer, which it cannot have; the redesign cost usually exceeds the material saving. And rigid-flex is priced by lamination cycles, not by area, so a small board with three press cycles can cost more than a large one with two.

Before choosing a metal core, three hybrid constructions solve the same problem with less disruption.

FR-4 with a copper coin or a press-fit insert keeps the multilayer routing and the fine lines, and routes the concentrated heat through a solid copper pillar to the chassis. Thermal performance rivals copper core at the source, without giving up vias.

FR-4 with a thermal via array and 2 oz copper covers the 0.05–0.3 W/cm² band, which is where most power electronics actually sit. It is the least expensive option that is also the least risky.

Rigid-flex with FR-4 stiffeners handles the space-constrained assemblies that a metal core cannot, and the stiffeners can be shaped to press against a chassis for conduction cooling.

Each of these is a stack decision rather than a material decision, which is why the stack-up documentation matters as much as the laminate part number — the arithmetic behind that is in high-frequency stack-up design and in copper thickness in high-frequency PCBs.

Most metal-core problems are inherited, not new. A layout that worked on FR-4 arrives with features the construction cannot support, and the changes are cheaper to make on the drawing than after a first article.

Every via disappears. Not only the through-holes but the via stitching that tied pours together. Ground continuity has to be re-established on the single circuit layer, which usually means a redesign of the return path rather than a port of the routing.

The second side disappears with it. A metal-core board has one populated face. Components that lived on the bottom of the FR-4 stack — decoupling, protection diodes, connectors — have to be found a place on top, and the layout grows accordingly.

Minimum features widen. One to two ounce copper on a roughened dielectric etches with wider minimum width and spacing than an FR-4 inner layer. Traces at 0.075 mm, routine on FR-4, become 0.15 mm; fine-pitch footprints that depended on them need re-checking.

Copper balance starts to matter mechanically. The circuit layer is on one side of a 1.6 mm metal plate, and asymmetric copper changes how the panel moves through reflow. Large solid areas are usually thieved or cross-hatched, which the original layout never needed.

Thermal relief replaces thermal vias. A pad connected directly to the base is a very effective heatsink for the iron as well: hand assembly on a metal core needs relief spokes or a preheat profile, otherwise the joint never reaches temperature.

None of these is exotic, but together they explain why a design handed over as “same board, aluminium instead of FR-4” typically comes back with a different outline, a reduced component count on the bottom side and a revised minimum feature table.

Five items turn any of these four choices into something a fabricator can quote and a test can verify: the base construction and its thickness; the dielectric layer thickness and thermal conductivity with the test method; for FPC, the bend radius and the number of flex cycles with the copper type; the maximum operating temperature and the cycling requirement; and the fire rating the end product needs. Without the second item a metal-core quote is not comparable between suppliers, and without the third an FPC quote is not comparable either.

XCE PCB builds rigid FR-4 from 2 to 72 layers, metal-core and hybrid stacks, and rigid-flex assemblies: panels to 600×1200 mm, impedance to ±10 %, samples in as fast as 24 hours, inspected to IPC-A-600G and IPC-6015. Where the same board also has to carry RF, the laminate questions are answered in high-frequency versus standard PCB.

FR-4, FPC, aluminium core and copper core are not four points on one line. FR-4 names a dielectric, FPC names a structure that bends, and the two metal-core families name a thermal function. The comparison that actually decides a design is the thermal one, and there the metal is not the limiting layer: a 0.10 mm dielectric at 2 W/m·K contributes about 0.50 K/W per square centimetre, against 0.096 for 1.6 mm of aluminium and 0.042 for copper. Going from aluminium to copper therefore buys about ten percent through the thickness and a factor of 2.3 in lateral spreading; halving the dielectric buys forty percent.

Use the order of the axes: does it bend, does it have to move heat, and which dielectric. A flexible board is a mechanical answer, not a low-loss one. A metal core is a thermal answer, and it costs the vias and the second circuit layer. And an FR-4 board with a via array or a copper coin is often the cheapest answer that keeps both.