Skip to content
Tutorials Center
My learning

Topics

Lesson 34 of 45 · Magnetics and Thermal Design

Thermal design: losses, junction temperature and heat sinks

Estimate MOSFET conduction and switching loss, use the thermal resistance ladder to find junction temperature, choose a heat sink, and see how pulses and thermal feedback behave.

Every watt that a converter loses becomes heat, and heat must flow from the tiny silicon die to the room. Semiconductors are specified for a maximum junction temperature, typically 150 °C for silicon MOSFETs and IGBTs and up to 175 or 200 °C for silicon carbide. Operate above it and the device degrades or fails; operate hot and life shortens: a common rule of thumb is that reliability, and the life of electrolytic capacitors in particular, roughly halves for every 10 °C of extra temperature.

Thermal design is therefore a two-step calculation: first how much power is lost and where, then how large a temperature rise that power causes along the heat path. The second step is just Ohm's law with different units.

Electrical analogy: an 8 W current source feeds three thermal resistances in series to a 40 degree ambient; node temperatures are 88, 80, 76 and 40 degrees Celsius
Temperature is voltage, heat flow is current and thermal resistance is resistance: Tj = Ta + P (θJC + θCS + θSA).

What you will learn

  • Compute the conduction, switching, gate and output-capacitance losses of a MOSFET.
  • Use the thermal resistance ladder to find junction, case and heat-sink temperatures.
  • Choose a heat sink by calculating the maximum allowed θSA.
  • Explain thermal feedback (loss that rises with temperature) and solve it iteratively.
  • Use transient thermal impedance for short pulses.

Key terms

Junction temperature (Tj): the temperature of the silicon die; the number that must stay below the limit.

Thermal resistance (θ): temperature rise per watt along a path, in °C/W (equivalently K/W).

θJC, θCS, θSA: junction-to-case, case-to-sink (the interface material) and sink-to-ambient resistances.

Transient thermal impedance (Zth): the thermal resistance seen by a pulse of a given length; it is smaller than the steady-state value for short pulses.

Derating: reducing power at high ambient temperature so that Tj stays below the limit.

Where the heat comes from

For a MOSFET in a hard-switched converter the main loss terms are:

Loss Formula Comment
Conduction Irms² · Rds(on)(Tj) Rds(on) rises about 0.6 to 1 % per °C for silicon, so about 1.5 to 2 times at 125 °C
Switching (overlap) ½ · Vds · Id · (tr + tf) · fsw voltage and current overlap during each transition
Output capacitance ½ · Coss · Vds² · fsw energy of Coss dissipated each turn-on
Gate drive Qg · Vgs · fsw dissipated in the gate driver and gate resistor, not mostly in the die
Body diode and reverse recovery Vf · Iavg + Qrr · Vds · fsw during dead time and with synchronous rectifiers

Other parts add their share: the inductor's copper and core losses, the diode's Vf·Iavg, the capacitors' Irms²·ESR, the sense resistor and the PCB traces. The sum of all of them is the efficiency loss.

Worked example (the picture). A 48 V to 12 V, 20 A buck at fsw = 200 kHz has a duty cycle D = 0.25. The high-side MOSFET carries Irms ≈ Io·√D = 10 A. With Rds(on) = 8 mΩ at 25 °C and 12 mΩ at 100 °C the conduction loss is 10² × 0.012 = 1.2 W. With tr + tf = 40 ns the overlap loss is ½ × 48 × 20 × 40e-9 × 2e5 = 3.8 W, and the Coss loss is another 0.18 W. Switching loss, not conduction loss, dominates here: a faster device or a lower frequency would help far more than a lower Rds(on).

The thermal resistance chain

In steady state heat flows from the junction through the case, the interface and the heat sink into the ambient air. Every step has a thermal resistance, and the resistances add in series:

Text
Tj = Ta + P · (θJC + θCS + θSA)

The figure shows an 8 W loss with θJC = 1.0 °C/W (from the datasheet), θCS = 0.5 °C/W (a thermal pad), θSA = 4.5 °C/W (a small extruded heat sink with a little airflow) and Ta = 40 °C. The temperatures are Ts = 76 °C, Tc = 80 °C and Tj = 88 °C. Notice that most of the temperature rise, 36 of 48 K, is across the heat sink: cooling is usually won or lost outside the package.

Choosing a heat sink

Rearranging, the heat sink must satisfy

Text
θSA ≤ (Tj,max − Ta) / P − θJC − θCS

where you pick Tj,max with a margin (120 to 125 °C for a 150 °C part). For P = 5.4 W, Ta = 50 °C, Tj,max = 125 °C, θJC = 1, θCS = 0.5, the sink needs θSA ≤ 75/5.4 − 1.5 = 12.4 °C/W: a modest sink in still air. Typical orders of magnitude are: a TO-220 in free air, about 60 °C/W (junction to ambient); a small clip-on sink 15 to 25 °C/W; an extruded aluminium sink with natural convection 2 to 8 °C/W; the same with forced air 0.5 to 2 °C/W. A PCB with large copper areas can give 30 to 50 °C/W for a surface-mount power package.

Thermal feedback

Because Rds(on) increases with temperature, the loss depends on Tj, and Tj depends on the loss. Solve it by iterating: guess Tj, compute P(Tj), compute the new Tj, repeat. The iteration converges when the product of the loss's temperature sensitivity and the thermal resistance is less than one. A positive temperature coefficient of this kind is safe for a single MOSFET (hotter means more loss, but the loss grows less than the cooling), but if a heat sink is too small the iteration diverges: that is thermal runaway.

Short pulses

The steady-state θ applies only to long loads. A heat path has thermal capacitance as well, so a short pulse sees a smaller transient thermal impedance Zth(t) = Σ Ri (1 − e^(−t/τi)). A 100 W pulse lasting 1 ms heats the junction far less than 100 W applied continuously (about 40 % as much in the illustrative network of the second example below). Datasheets give Zth curves for single pulses and for various duty cycles; use them for start-up, fault and motor-stall conditions.

Heat spreads. A device on a large copper plane or a thick aluminium plate sees a lower effective θ because the heat fans out. Thermal interface material between the case and the sink should be thin and without air gaps; a dry, bare joint can add several °C/W.

Common misconceptions

"A bigger heat sink always fixes it." Past a point the interface and package resistances dominate; the junction cannot be cooler than Ta + P·(θJC + θCS).

"The case temperature is the junction temperature." The junction is hotter than the case by P·θJC. Measure the case and add that offset.

"Efficiency is a fixed number." Conduction loss rises with temperature and with load, so efficiency at a hot ambient is lower than the datasheet curve at 25 °C.

"Thermal resistance is a property of the device alone." θJA depends on the board, the airflow and the neighbouring parts.

Try it

The first example calculates the loss budget of the 48 V to 12 V buck's MOSFET, solves the thermal feedback loop and draws the graphical solution: the loss curve P(Tj) against the cooling line (Tj − Ta)/θ. Where they cross is the operating temperature. It uses an assumed Rds(on) temperature model.

Python Loss budget and thermal feedback

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The second example shows the transient thermal impedance of an illustrative four-stage Foster network (values chosen to sum to 1.0 °C/W, not from a datasheet) and the junction temperature rise of a 100 W pulse for several pulse lengths.

Python Transient thermal impedance

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The third example is a heat-sink selector. Enter the loss, ambient temperature and thermal resistances, and it reports the maximum allowed θSA and what kind of cooling that suggests. The guidance ranges are rough.

Heat-sink selector

Check your understanding

Quick check · 5 questions

of 5 answered You scored / 5 · perfect!

  1. 1. A device dissipates 6 W with θJC = 1.2, θCS = 0.4 and θSA = 3.0 °C/W at an ambient of 35 °C. What is Tj?

    Tj = 35 + 6 × (1.2 + 0.4 + 3.0) = 35 + 27.6 = 62.6 °C.
  2. 2. What is the maximum θSA for P = 10 W, Ta = 45 °C, Tj,max = 125 °C, θJC = 1.0 and θCS = 0.5 °C/W?

    θSA ≤ 80/10 − 1.5 = 6.5 °C/W.
  3. 3. A MOSFET at 400 V, 5 A with tr + tf = 30 ns and fsw = 100 kHz has an overlap loss of about

    ½ × 400 × 5 × 30e-9 × 1e5 = 3 W.
  4. 4. Why can a 1 ms pulse of high power be tolerated although the same power continuously would not?

    Transient thermal impedance rises with pulse length toward the steady-state value.
  5. 5. Which statement about Rds(on) is true for silicon MOSFETs?

    The positive temperature coefficient (about 0.6 to 1 % per °C) raises the loss when hot.

Summary

  • Losses: conduction Irms²·Rds(on), switching ½ V I (tr+tf) fsw, Coss and gate drive, plus the magnetics and capacitors.
  • Steady state: Tj = Ta + P·(θJC + θCS + θSA); the heat sink must satisfy θSA ≤ (Tj,max − Ta)/P − θJC − θCS.
  • Loss depends on temperature, so solve by iteration; too little cooling means thermal runaway.
  • Short pulses see the transient impedance Zth(t), much lower than the steady-state resistance.
  • Derate at high ambient temperature; reliability falls quickly with temperature.

Next: Chapter 7 review