Thermal management in PCB manufacturing refers to three different problems that share one name, and the confusion between them is why thermal discussions so often end without a number. Keeping a junction below its limit is a resistance problem measured in K/W. Spreading heat out of a small area is a conductance problem measured in W/m·K. Surviving a reflow or a thousand cycles is a strain problem measured in ppm/°C. This article separates the three, then does the arithmetic that decides each one: a 0.254 mm dielectric is thermally equivalent to about 143 mm of copper, a single 0.3 mm plated via is about 185 K/W, and 10 W/cm² through that dielectric raises the far side by 36.8 K on RO4350B against 17.6 K on RO6035HTC.
Three Problems, One Name
The question “what does thermal management refer to” has a short answer and a useful one. The short answer is: controlling the temperature of a board and everything on it. The useful answer is that three separate engineering problems are hiding inside that sentence, and they are solved with different numbers, by different people, at different stages.
| What is meant | The question being asked | The number that decides it | Who owns it |
|---|---|---|---|
| Junction temperature control | Does this part stay below its maximum junction temperature at the worst-case ambient and load? | Thermal resistance, K/W, summed from junction to air | Circuit and mechanical design |
| Heat spreading | Can a 3×3 mm source dissipate its power into a board without a hot spot? | Thermal conductivity W/m·K, and the thickness it acts over | Layout and laminate selection |
| Thermal reliability | Does the assembly survive assembly and service without cracking a joint or a barrel? | CTE in ppm/°C, and the ductility of the plating | Process and materials |
An article that lists “thermal conductivity, thermal resistance, heat capacity, CTE and thermal stability” as five peer properties — which is what the earlier version of this page did — invites the reader to average five different things into one impression. The rest of this article takes them one at a time and attaches a number to each. The performance side of the same subject, that is what a thermally good board can do, is covered in the performance and role of thermal management; this page is about what the term refers to and how to calculate it.
Thermal Resistance Is the Only Currency
Steady-state heat flow obeys a law with the same shape as Ohm’s: temperature difference equals power times resistance. For a slab of thickness t and conductivity k, the resistance per unit area is R″ = t / k, conventionally quoted in K·cm²/W, and the absolute resistance is R = t / (k·A). Every thermal number in PCB design is one of those two, and both are the thickness divided by a conductivity.
Putting numbers in makes the laminate look different from the way it is usually described. Copper has a conductivity of about 400 W/m·K; the best filled high-frequency laminate is 1.44; unfilled PTFE is 0.20. Converting a 0.254 mm dielectric into the thickness of copper that would offer the same resistance gives the column that matters, because it says how much of a heat spreader the dielectric is not.
| Laminate | k, W/m·K | R″ at 0.254 mm, K·cm²/W | Copper equivalent thickness | Rise through it at 10 W/cm² |
|---|---|---|---|---|
| Copper | 400 | 0.0006 | — | 0.06 K |
| RO6035HTC | 1.44 | 1.76 | 70 mm | 17.6 K |
| RO3010 | 0.95 | 2.67 | 107 mm | 26.7 K |
| RO4003C | 0.71 | 3.58 | 143 mm | 35.8 K |
| RO4350B | 0.69 | 3.68 | 147 mm | 36.8 K |
| FR-4 | 0.30 | 8.47 | 339 mm | 84.7 K |
| RT/duroid 5880 | 0.20 | 12.7 | 508 mm | 127 K |
The copper-equivalent column is the one to quote in a design review, because it settles the argument that a thin dielectric cannot be a problem: 0.254 mm of RO4003C insulates as well as 143 mm of copper, so any heat that has to cross it is crossing a brick wall. The dielectric is not the heat path. This is the single most important consequence of reading thermal conductivity as a resistance term rather than as a merit badge.
So Where Does the Heat Actually Go
Down, through copper that was deliberately placed, and sideways, through copper that happens to be there. Three paths exist in a normal high-frequency stack and they are used in this order of effectiveness.
Thermal vias under the source. A via is a copper cylinder through the dielectric, and it is the only structure that crosses the 143 mm-of-copper barrier in a straight line. Planes and thick copper. A ground plane directly under a hot component spreads heat laterally at a conductivity three orders of magnitude better than the laminate, at the cost of the dielectric thickness that sets the impedance. Edge and connector conduction. Small but real, and the reason a board bolted to a metal chassis behaves differently from the same board on standoffs.
What does not work is expecting the copper of the trace carrying the current to act as a heat sink. The trace is where the heat is generated; it can spread it, but it cannot remove it. This distinction between generating and spreading is the second correction worth making to the common list, and it is why “use heavier copper” is a partial answer: 2 oz copper doubles the cross-section and halves the resistance per square, but it also doubles the etch undercut and changes the impedance, as set out in the performance and role of PCB copper thickness.
The Via Calculation Nobody Runs
Thermal vias are usually added in arrays because “more is better”, which is true and not quantitative. The resistance of one plated via is t / (k·A), where A is the copper annulus, not the hole. A 0.30 mm finished hole with 25 µm of plating has an annulus area of π/4 × (0.30² − 0.25²) = 0.0216 mm². Through a 1.6 mm board that is 0.0016 / (400 × 2.16×10−8) = 185 K/W.
| Via array | Parallel resistance | Rise at 2 W through the array | Comment |
|---|---|---|---|
| 1 × 0.30 mm | 185 K/W | 370 K | Useless on its own |
| 3 × 3 (9 vias) | 20.6 K/W | 41 K | Adequate for a small LDO |
| 5 × 5 (25 vias) | 7.4 K/W | 15 K | Typical for a 3 W PA |
| 8 × 8 (64 vias) | 2.9 K/W | 5.8 K | Diminishing returns below the spread resistance |
Three details change the result by more than the array size does. The annular area, not the drill. Doubling the plating from 25 to 50 µm more than doubles the copper area, because the annulus grows from (0.30² − 0.25²) to (0.30² − 0.20²) — a factor of 2.2, not 2. Whether the via is filled. An unfilled via has air in the middle, and air at 0.026 W/m·K contributes nothing; a resin plug at 0.3 is also negligible; a thermally filled via comes close to making the whole drill area work. Where the array ends. Once the array is larger than the source, adding vias spreads heat into laminate that cannot take it away, and the board’s own spreading resistance sets the floor.
The same arithmetic explains why a via-in-pad design beats a ring of vias around a pad at the same drill count: the distance from the source to the copper is shorter, and the spreading resistance between them, which is the part that cannot be bought with more vias, is smaller. The HDI context for these structures is in performance and functions of HDI PCB.
What a Filled Laminate Buys, in Kelvin
The difference between 0.69 and 1.44 W/m·K sounds like a datasheet detail until it is converted into a temperature. At a heat flux of 10 W/cm² through 0.254 mm — a realistic figure for a power amplifier die under a thermal pad — RO4350B gives 36.8 K and RO6035HTC gives 17.6 K. The filled material removes 19 K from the stack, which is often the difference between a junction at 105 °C and one at 125 °C.
The comparison has a second order term that is usually ignored: the same ceramic filling that raises k raises Dk to 3.50 and, more importantly, changes the thickness window the laminate is supplied in. A thermal decision therefore becomes a stack decision, and the stack decision becomes an impedance decision. This is the normal shape of thermal work on a high-frequency board — you almost never buy conductivity without buying something else — and it is why the material choice for a thermally loaded RF board is usually made once, early, and then treated as fixed.
The Chain From Junction to Air
Component-level thermal management — the first of the three meanings — is a series sum: Tj = Ta + P × (θjc + θcs + θsa). The board owns the middle term and part of the third. Four consequences follow directly from the arithmetic.
The largest term dominates, and it is usually not the board. A TO-220 with a clip and no heat sink has θsa of 40–60 K/W, which swamps whatever the board contributes. Interface materials are not optional. A 0.1 mm gap filled with air instead of thermal grease is 385 K·cm²/W against 1–2 for grease — a factor of two hundred on the term that is easiest to control. Parallel paths matter more than the best path. Heat leaving through 25 vias and through the ground plane in parallel is better than either alone, and the arithmetic for parallel resistances is unforgiving of the weakest link. Ambient is a specification, not a measurement. Declare the worst case: 70 °C inside an enclosure is not 25 °C in a lab.
Heat Capacity and Thermal Stability Are Not the Same as Thermal Resistance
Two of the five entries in the usual list are mislabelled, and the confusion is worth clearing because it changes what you measure.
| Term as usually written | What it actually is | When it matters | When it does not |
|---|---|---|---|
| Heat capacity | Energy needed to raise the temperature of a mass, J/(g·K) | Transients: a pulsed load, a warm-up, a short circuit | Steady state, which depends on conductivity and geometry only |
| “Thermal resistance” as stability | Thermal stability: the temperature at which the resin decomposes, Td | Reflow exposure, repeated rework, high-temperature service | Heat flow, which is what thermal resistance actually measures |
| Thermal resistance | t/k, K·cm²/W, or K/W for a defined area | Every steady-state calculation in this article | Nothing — it is the currency |
The earlier version of this page listed “thermal resistance” twice, once correctly and once as “the stability and durability of a material in a high temperature environment”. The second entry is thermal stability, and its numbers are Td and Tg, not K/W: the high-frequency laminates in this family typically decompose at 500 °C, RO4000 series at 390–425 °C, which is why a hydrocarbon board survives lead-free reflow without the resin degrading even though its conductivity is a third of a filled PTFE. Mixing the two numbers up produces the strange specification in which a thermally poor material is rejected for a reflow reason that does not apply to it.
Thermal Cycling: The Reliability Half
The third meaning of thermal management is the one that fails in the field rather than on the bench. A joint or a plated barrel is a composite of materials with different expansion coefficients, and every temperature excursion loads it. The dielectric’s contribution is the z-axis CTE, and the spread across laminate families is large: 16 ppm/°C for RO3010, 46 for RO4003C and FR-4, 217 for TLY-5 and 237 for RT/duroid 5880.
Translated into a reflow, a 1.6 mm board grows about 82 µm through its thickness on 5880 against 11 µm on a hydrocarbon laminate, and the plating has to stretch by the difference. That is a requirement on plating ductility — 20 % elongation or better — and on aspect ratio, and it is measured in a thermal cycling test rather than inferred from a conductivity figure. The measurement side of the same property, including why CTE is not an electrical characteristic, is in electrical characteristics in high-frequency PCB manufacturing, and the glass transition, which governs how much of that expansion happens in the resin, in glass transition temperature in PCB manufacturing.
The Conductivity Ladder Across the Material Families
For reference, here is the thermal conductivity of the laminates that appear most often in high-frequency work, with their dielectric constants, because the two are chosen together.
| Laminate | k, W/m·K | Dk | Thermally notable for |
|---|---|---|---|
| RO6035HTC | 1.44 | 3.50 | Highest k in the family; 17.6 K at 10 W/cm² |
| RO3010 | 0.95 | 10.2 | High k with a high Dk; small features carry the heat |
| RO3210 | 0.81 | 10.2 | Glass-reinforced version, lower tolerance |
| RO3006 | 0.79 | 6.15 | Mid-Dk with useful conductivity |
| RO4003C | 0.71 | 3.38 | The default RF laminate; 143 mm copper equivalent |
| RO4350B | 0.69 | 3.48 | FR-4 process compatible |
| RO4835 | 0.66 | 3.48 | Oxidation-resistant version of 4350B |
| RO3003 / RO3035 | 0.50 | 3.00 / 3.50 | Low loss, poor conductor of heat |
| RT/duroid 5880 | 0.20 | 2.20 | Best electrical, worst thermal; 127 K at 10 W/cm² |
The spread is a factor of seven, which is larger than the spread in Df across the same list and is the reason a thermally loaded design cannot be done by starting from the lowest-loss material and working backwards. The comparative advantages of the filled grades are set out in the unique advantages of Rogers RT6035HTC; the design question is whether the application needs conductivity more than it needs Df, and that question has a numeric answer once the flux is known.
Three Statements About Thermal Management That Are Wrong
“Thermal management means adding a heat sink.” A heat sink is the last term of a series sum whose largest terms are usually the interface and the spread. Adding one to a board that cannot get heat out of a 3 mm package changes little.
“A thicker dielectric insulates the heat, so a thinner one runs cooler.” Half true and misleading. Halving 0.254 mm to 0.127 mm halves R″ from 3.58 to 1.79 K·cm²/W, which is real, but the same change halves the impedance width and doubles the etch sensitivity. The heat path that actually matters is still the vias, and their resistance does not change at all.
“Thermal vias solve the problem once there are enough of them.” They solve it once the array resistance is below the spreading resistance of the copper it feeds. Beyond that point, the return on more vias is zero, and the limit moves to the plane thickness and the chassis connection. The table in section 4 reaches that point at around 25–64 vias for a 3 W source, which is a smaller number than most designs use.
A Six-Step Sequence That Produces a Number
1. State the flux, not the power. Watts divided by the source area, in W/cm². 2. Pick the stack and compute R″ as t/k for each dielectric layer in the path. 3. Size the via array from t/(k·A) per via and the parallel sum, then check it against the spreading resistance. 4. Add the interface and the sink in series and compare the total with the allowed rise. 5. Check the thermal reliability side — z-axis CTE against plating ductility and aspect ratio. 6. Write the result on the drawing as an allowed rise and a via pattern, not as “good thermal management”.
Each step ends in a number that someone else can check. That is the practical difference between the three meanings of the term: junction temperature is a sum in K/W, spreading is a division in W/m·K, and reliability is a strain in ppm/°C. A thermal review that keeps them apart finishes in an hour; one that mixes them finishes in a meeting.
Conclusion
Thermal management in PCB manufacturing refers to keeping a board and its parts inside their temperature limits, but the term covers three problems that are calculated differently. Junction control is a series of thermal resistances in K/W. Spreading is conductivity over thickness, and the comparison that makes it concrete is that 0.254 mm of RO4003C insulates as well as 143 mm of copper while 0.254 mm of 5880 insulates as well as 508 mm. Reliability is thermal expansion against plating ductility, where a 1.6 mm PTFE board grows 82 µm across a reflow against 11 µm for a hydrocarbon one.
The single design consequence is that the dielectric is not the heat path. Heat leaves through plated copper placed for the purpose, and the arithmetic for that copper is t/(k·A): 185 K/W for one 0.30 mm via through 1.6 mm, 7.4 K/W for twenty-five of them in parallel. Two entries on the usual property list — heat capacity and thermal stability — are real but belong to transients and to reflow respectively, not to the steady-state calculation that decides whether the part survives.
XCE PCB builds thermal structures into high-frequency boards as a stack decision rather than a postscript: 2 to 72 layers, filled and plated via arrays, mixed laminates including filled PTFE for hot spots, panels to 600×1200 mm, and samples in as fast as 24 hours. The via and plating windows that make the arithmetic above achievable are covered in HDI performance and functions.
Frequently Asked Questions
What does thermal management refer to in a PCB, in one sentence?
Why is the dielectric described as a heat insulator?
How many thermal vias are needed under a power device?
Which laminate should be used for a hot RF board?
Is heat capacity part of thermal management?
How does thermal expansion cause failures in high-frequency boards?
Need the Heat Path Designed, Not Assumed?
Send us your Gerber files and requirements for a manufacturability review and a tailored quotation.
sales1@xcepcb.com
+86 180 3341 4657