Heatsink calculator
Junction temperature from the whole thermal stack, and the heatsink rating your target actually needs — plus the two things a fixed-power calculator cannot tell you: the temperature no heatsink on earth gets you below, and whether your dissipation heats itself faster than the heatsink can carry it away.
Heatsink ratings are catalogue figures for natural convection; the airflow setting derates them the way a manufacturer's curves do. Package and interface figures are typical — pick Custom to enter a datasheet's. The temperature coefficient is for whatever dissipates: about 0.8 %/°C for a silicon MOSFET's on-resistance, 0 for a linear regulator or a resistor.
| Where the temperature rise goes | θ (°C/W) | Rise (°C) | Share |
|---|
| Heatsink (°C/W) | Junction (°C) | Margin (°C) | Max power (W) |
|---|
Catalogue ratings, derated by the airflow you chose. The highlighted row is your design. Max power is what this stack allows at the junction rating, stated at 25 °C the way a datasheet states it.
How this is calculated
Heat leaves a semiconductor through a chain of resistances in series, and temperature is what develops across them. It is the same arithmetic as a resistor chain carrying a current, with watts in place of amps and degrees in place of volts:
Tj = Tambient + P · (θJC + θCS + θSA)
Junction to case is fixed by the package and the die inside it, and you cannot change it without changing the part. Case to sink is the mounting interface — grease, a pad, a mica washer, or nothing. Sink to air is the heatsink, and it is the only term you can buy. With no heatsink fitted the whole chain collapses into one figure, the junction-to-air resistance, which for a plastic power package standing in still air is enormous.
Because the terms add, the rise divides between them in proportion, which is what the breakdown table shows. It is worth reading before buying anything: on a package with a high junction-to-case figure, a better heatsink moves a number that was never the problem.
The junction temperature no heatsink can beat
Set the heatsink to a perfect zero and the chain still contains the package and the interface. Whatever else you do, the junction sits at
Tfloor = Tambient + P · (θJC + θCS)
and no heatsink, fan, cold plate or amount of money gets it below that. At the defaults — a 20 A MOSFET with 8 mΩ of on-resistance in a TO-220, greased to its sink, in 40 °C air — the floor is 45.6 °C while the design as entered runs at 67.9 °C. There are 22 degrees of heatsink in play, and every degree below 45.6 is unreachable.
That single number decides which kind of problem you have. If the floor is comfortably under the rating, a better heatsink is a real lever and the page names the rating to buy. If the floor is already near the rating, the heatsink is irrelevant and the answer is a different package, a paralleled pair, or less dissipation — and no amount of shopping will find that out for you.
Self-heating is a feedback loop, and it has a loop gain
Every other heatsink calculator asks for a dissipation and treats it as a constant. For a resistor or a linear regulator that is fair. For a MOSFET it is not: on-resistance climbs with temperature, roughly 0.8 % per degree for silicon, so the part dissipates more as it warms, which warms it further. That is a loop, and its gain is
G = θtotal · dP/dTj = α · P25 · θtotal
Solving the loop instead of ignoring it gives the junction temperature exactly:
Tj = [ Tambient + P25 · θ · (1 − 25α) ] / (1 − G)
and with it the figure in the results above: 1 / (1 − G) is exactly how many degrees the junction rises per degree the ambient rises. At the defaults G is 0.166 and that factor is 1.20, so the ordinary fixed-power answer of 60.8 °C is 7.1 degrees optimistic, and ten degrees of hot day inside the enclosure costs twelve at the die. It is the same quantity as the loop gain on the op-amp gain calculator, doing the same job: what matters is not the output, it is how much the loop multiplies whatever disturbs it.
The interesting part is what happens as G approaches one. There the factor goes to infinity, and past it the equation has no solution at all — not a part that runs hot, a part with no temperature at which it settles. The threshold is a thermal resistance:
θcrit = 1 / (α · P25)
At the defaults that is 39.1 °C/W. A bare TO-220 in still air is about 62 °C/W, so this MOSFET without a heatsink is not merely two-hundred-and-something degrees hot, as a fixed-power calculator would cheerfully report. It has no operating point. In the real part the climb stops when something gives — the on-resistance curve bends, the gate drive runs out, the die fails — and which of those happens is not a design decision anyone should be making by accident.
The two cheap levers, measured
The interface is the one people skip. At the defaults, mounting the same part dry instead of greased adds 0.7 °C/W and takes the junction from 67.9 °C to 71.6 °C — nearly four degrees for a smear of compound that costs nothing. An insulating silicone pad costs more again, which is worth knowing before choosing an isolated mounting you may not need.
Airflow is the one people underestimate. A heatsink's catalogue rating is a natural-convection figure, and moving air past it is worth roughly a factor of two: at the defaults, 2.5 m/s over the existing 5 °C/W sink brings the junction to 54.9 °C, while halving the heatsink's thermal resistance — a far larger and more expensive extrusion — only reaches 56.0 °C. A small fan beats doubling the heatsink, and it does so in the same board area. The airflow selector derates the rating the way a manufacturer's curves do, so the heatsink figure the page tells you to buy is always the catalogue one.
Surface-mount parts: the board is the heatsink
A DPAK, D²PAK, SOT-223 or thermal-pad QFN has no heatsink to bolt on. Its junction-to-air figure is a property of the copper it is soldered to, which is why datasheets quote it against a stated pad area — often one square inch of one-ounce copper, worth very roughly 50 °C/W on its own. Pour more copper, use both sides with vias under the pad, and it improves, with strongly diminishing returns past a few square centimetres.
Treat a quoted junction-to-air figure as valid only for the pad the datasheet used, and remember that the same copper is carrying the current: a pour sized for heat is usually far wider than the PCB trace width calculator would demand for the current alone, and the trace's own resistive heating adds to the package's. For a part that dissipates because it is dropping voltage, the arithmetic is the linear-regulator mode above, and the reason to care is usually runtime rather than temperature — that side of it is on the battery life calculator.
FAQ
How do I calculate junction temperature?
Multiply the dissipation by the total thermal resistance from junction to ambient and add the ambient temperature. The total is the package's junction-to-case figure plus the mounting interface plus the heatsink, or the single junction-to-air figure if there is no heatsink. If what dissipates gets worse as it heats — almost anything carrying current through silicon — the fixed-power answer is optimistic and the loop solution above is the right one.
What size heatsink do I need?
Work backwards from the junction temperature you will accept. The total resistance allowed is the temperature budget divided by the dissipation, and the heatsink rating you need is that total minus the package and interface terms. This page does it for your target and reports the catalogue rating to buy at the airflow you chose. If the answer comes out at or below zero, the floor is already above your target and no heatsink exists that will do it.
What is a good target junction temperature?
Well below the rating. Silicon is usually rated to 150 °C and most failure mechanisms roughly double in rate every ten degrees, so designing to 150 buys no life at all. Forty degrees below the rating is a common default and is where the target field starts; safety-critical or long-life designs go further. The rating is where the datasheet's guarantees stop, not where the part is happy.
Does thermal grease actually matter?
Yes, and not because the compound conducts well — it barely does. Two machined surfaces touch across a small fraction of their apparent area and the rest is air, which is an excellent insulator. The grease's job is to fill that air. At the defaults it is worth almost four degrees. More is not better: a thick layer is worse than a thin one, because then the compound itself is the gap.
What is the difference between junction-to-case and junction-to-air?
Junction-to-case is a property of the part — die to mounting surface — and it means something only when the mounting surface has something attached to it. Junction-to-air is the whole path with nothing attached, measured in still air on a stated board, and it is a property of the test setup as much as of the part. Do not add them. Use junction-to-case when a heatsink is fitted and junction-to-air when it is not.
What causes thermal runaway in a MOSFET?
On-resistance rising with temperature while the current stays the same. Dissipation and temperature then drive each other, and when the loop gain reaches one there is no stable temperature. It is a property of the whole design, not of the transistor: the same part at the same current is stable on an adequate heatsink and unstable without one, and the boundary is a thermal resistance this page prints.
Does this cover short pulses?
No. Everything here is the steady state, which is the right question for anything dissipating continuously for more than a few seconds. A brief pulse is limited by the transient thermal impedance curve in the datasheet instead, which is far below the steady-state figure and is why parts survive fault currents that would destroy them if sustained.