Thermal Simulation: Design Rules and Process Limits

Thermal simulation is worth doing when it answers a question that the design cannot answer by estimation, and it is a waste of effort when it produces a colourful picture that nobody acts on. The useful questions are specific: will this junction stay below its limit at the worst case ambient, does this copper area move enough heat, and does adding a thermal via array change the answer by more than the manufacturing tolerance does.

What Simulation Can Answer

Simulation is good at comparing options and at locating the hot spots in a design. It is less good at predicting an absolute temperature accurately, because the answer depends on boundary conditions that are often poorly known: the ambient, the airflow, the conduction into the enclosure and the emissivity of the surfaces. A model that is validated against one measurement becomes far more useful than one that is not.

The comparison use is the most valuable. Two copper areas, two via patterns, two component placements or two board orientations can be compared under identical assumptions, and the difference between them is much more reliable than the absolute value. Design decisions rest on differences of this kind.

The absolute question still has to be answered, and the honest way to answer it is with margin. If the simulation says the junction will reach ninety degrees with a limit of a hundred and twenty five, the design has margin against the errors in the model. If it says a hundred and twenty, the model is not accurate enough to support the conclusion and the design should be changed or measured.

Building the Model

The model should include the sources, the paths and the boundary. The sources are the components that dissipate power, with their values taken from the datasheet at realistic operating conditions rather than at the nominal condition. The paths are the copper, the dielectric, the vias, the thermal interface material and the enclosure. The boundary is the ambient temperature and the airflow, which should be the worst case the product will see.

Simplification is necessary and should be deliberate. Representing every trace in a power plane as geometry is impractical, so the copper is usually treated as a lumped layer with an equivalent in plane and through plane conductivity. The simplification is acceptable when the copper is uniform and risky when it is not, which is exactly the case in a board with a fragmented ground plane.

The component models deserve attention because they often dominate the error. A package with an exposed pad has a very different thermal path from one without, and a model that omits the solder voiding under a thermal pad will overestimate the performance. Where the manufacturer provides a thermal model, it should be used in preference to a generic one.

Junction Temperature and Thermal Resistance

Junction temperature is the quantity that matters, and it is calculated as the ambient plus the power multiplied by the total thermal resistance from junction to ambient. The resistance is the sum of several terms in series: junction to case, case to board or to heat sink, and heat sink to ambient, with parallel paths through the board and the surrounding air.

The datasheet usually gives the resistance for a specified mounting condition, and applying it to a different condition is a common source of error. A value quoted for a device on a large copper area is not valid for one on a small pad, and the difference can be tens of degrees. Where the mounting differs from the reference, the resistance should be estimated from the geometry rather than taken from the table.

The thermal resistance of the board itself depends on the copper distribution and on the layer count. A four layer board with a solid ground plane spreads heat far better than a two layer board with fragmented copper, and this is one of the reasons that thermal performance improves with layer count even when the electrical design does not require it.

Heat Spreading and Thermal Vias

Heat spreading in a board happens in two directions. Laterally, the copper spreads heat away from the source and into a larger area that can pass it to the air. Through the thickness, heat passes to the opposite side and, where a plane or a heat sink is present, to a larger surface. Both paths should be considered, because a design that improves one and neglects the other may gain very little.

A thermal via array, or a via in pad structure, is the standard way to move heat through the thickness. The vias conduct heat far better than the dielectric, and their effect depends on the number, the diameter and the copper plating thickness. A dense array under a thermal pad typically produces a large improvement over a few vias, and the improvement saturates once the array covers the available area.

The vias also affect the electrical design, because they penetrate the planes and can create return path discontinuities. Where a thermal array sits under a device with high speed signals, the plane integrity should be checked. This is a case where the thermal and electrical designs interact, and both should be reviewed together rather than sequentially.

Boundary Conditions and Airflow

The ambient temperature is often specified by the customer or by the environment, and it should be the worst case rather than an average. A board that is comfortable at twenty five degrees in a laboratory may be at sixty in an enclosure with other equipment around it, and the difference is the whole margin.

Airflow is the second variable and the least predictable. Natural convection gives a low and orientation dependent heat transfer, while forced convection depends on the ducting, the obstructions and the fan curve. Where a fan is used, it should be modelled with the flow it will actually deliver in the installed arrangement rather than at its free air rating.

Conduction into the enclosure is often the largest path and is often omitted. Mounting posts, metal chassis, connectors and cables all conduct heat away from the board, and a design that ignores them can be substantially better in practice than the model suggests. Where an accurate answer is needed, the measurement and the model should be reconciled.

Validating the Model

Validation is what turns a simulation into a tool. A thermocouple on the actual board or a thermal camera image, compared with the model at the same operating point, shows whether the assumptions are reasonable. The comparison rarely agrees exactly, and the value is in understanding the direction and the size of the discrepancy.

Once the model is calibrated against one measurement, the differences between options become trustworthy, and the model can be used to make decisions without further measurement. This is the point at which thermal simulation begins to save money rather than consume it, because it prevents a redesign after the first thermal test.

The calibration should be recorded. The ambient, the airflow, the power level and the measured temperature together define the condition under which the model is valid, and the record allows the next engineer to reuse the work rather than repeat it.

Practical Rules

Use simulation to compare options and to locate hot spots, and use margin rather than precision when predicting an absolute temperature. Include the copper distribution, the thermal vias and the conduction paths into the enclosure.

Validate the model against one measurement and record the conditions with the design records. A thermal model that has been checked once is worth more than a detailed one that has never been compared with reality, and the test data from the first build is the measurement that makes the difference.

FAQ

How accurate is thermal simulation? Absolute values depend heavily on boundary conditions. Comparisons between design options are much more reliable than absolute predictions.

How many thermal vias are enough? As many as will fit under the pad. The improvement saturates once the array covers the available area, so a dense array is usually best.

Why validate a model? Because it turns an estimate into a tool. Once calibrated against one measurement, differences between options become trustworthy.