Advisor: Simulation Ask Schulz! Simulation and reality in harmony

Related Vendor

FAULHABER Drive Systems

A diode designed for 500 A with a 2000V blocking capability is to be specified. In combination with the cooling system, the temperature rise on the chip must not exceed 85K. Simulation is to show whether this is achievable.

Simulation is a way to estimate with numerical means what happens under given conditions in an application. Virtually all physical processes can be encapsulated in equations that can be used to calculate how a system behaves. When used correctly and the results of the simulation are correctly interpreted, it represents a powerful tool, even in the design of power electronic systems.

But who isn't familiar with the story of the aeronautical engineer whose simulation clearly showed that bumblebees can't fly? The GIGO principle of simulation remains: Garbage in – garbage out. What and how precisely you feed the simulation, and how you interpret the simulation results, have a significant impact on the statement you make using the simulation in the end.

The backstory

In the recent past, the task arose to specify a diode for 500 A with a blocking capability of 2000 V. Of course, this is no longer an SMD component, and even with only 1 V forward voltage, 500 W of power loss occur at the diode. In combination with the cooling system and the requirement for lifespan, the temperature rise at the chip should not exceed 85K.

Before deciding on the proposed component, a simulation was to show whether the set goals could be achieved. The component, a power semiconductor in module form with a copper base plate, has a layered structure as schematically shown in Image 1.

Unlike with current IGBT modules, a comprehensive copper flake is soldered onto the front side of the diode chip. In this assembly and interconnection technology, the connection to the environment is not made through a multitude of thin bond wires but through soldered clips.

This technique has proven to be particularly robust in demanding applications and has been in use with power semiconductors such as diodes and thyristors for decades in the field.

For simulation engineers, such a structure seems to be an easy task, as both the material properties and the layer thicknesses are very well known. Thus, all required parameters are available for a simulation.

The Simulation

In a first step, a 3D model suitable for the simulation software of the setup was designed, as shown in Figure 2. To save simulation time, it is common practice not to include more elements than strictly necessary and to utilize symmetries. Thus, a simplification involves initially simulating only one of the two chips and assuming that the power dissipation is distributed evenly across both components.

Since the chip is symmetrical in both the X and Y directions, further simplification is possible, which involves considering only half or even just one quarter of the chip area. The decision was made to perform the calculation with half the chip area, or a quarter of the semiconductor module, and the underlying model is shown in Image 3.

From the simulation and the shown result, it was concluded that the component reaches "a chip temperature" of 160°C and is therefore not suitable for the planned use. However, a closer look reveals weaknesses in the simulation and the interpretation of the results.

1) Only the chip was simulated as a heat source. The soldered copper plate and the connection clip are not part of the simulation. However, the clip provides a thermal path with a significant cross-section, whose second foot is mounted on a well-cooled surface. The resulting dissipation of thermal energy was neglected.

2) The transition resistance of the thermal interface material – the thermal paste – is not well known, and its influence on the chip temperature is significant. Assuming a uniform layer thickness with a homogeneous thermal conductivity according to the datasheet of the thermal conductive medium usually does not yield a realistic result.

3) "The chip temperature" was not correctly determined here, and instead, a maximum value in the middle of the chip was assumed.

All three factors influence the simulation in the same direction, towards a too conservative statement.

Already a correction in the consideration of chip temperature provides relief [reference to the article: The Mystery of Chip Temperature]. The so-called virtual chip temperature T_vj results from an averaging of the temperature in the center of the chip and a corner. From the data of the simulation, the 160°C already reduce to T_vj = (2·T_center+T_corner)/3 = (320+130)°C/3 = 150°C.

The influence of the missing metal parts can be estimated at about 5% temperature rise, and a similar value is given for the thermal conductive material. Instead of a 95K rise in the simulation, only 55K? Even a reduction of the temperature rise by only 10K would bring the semiconductor into the desired situation.

The Reality

To find a reliable answer, the semiconductor was mounted on a cooling system and powered by a voltage source that can deliver an adjustable current of up to 1000A at a maximum output voltage of 10V. An IR camera above the semiconductor provides the desired data.

Since the experiment is conducted using protective low voltage, no elaborate protective measures are necessary, as would otherwise be mandatory when working with high voltage. The setup is shown in Image 4.

The measurement captures the current, the voltage across the components, and their temperature in a steady state, which is established within a few seconds due to the liquid cooling. Image 5 is a picture from the IR camera, recorded at a current of 500 A.

The measurement was carried out in 50 A steps from 0-500A and the collected data summarized in Image 6. The data series DT_vj=f(I_D) was a first estimate of the temperature rise relative to the cooling body temperature, which was assumed to be constant.

However, the setup includes an adapter plate for mounting the semiconductor. This in turn introduces another temperature rise. To correct for this, the temperature of the adapter plate, also determined in the image, was used to calculate the corrected data series DT_{vj_corrected}=f(I_D).

The completed measurement shows that the semiconductor experiences a temperature rise of significantly less than 60K in the application, making the estimate of 55K actually a good one.

The conclusion

Simulation, when correctly fed and correctly interpreted, is an excellent tool for engineers to examine a non-existent system without costly experiments. Being aware of the weaknesses and pitfalls allows for a critical view of the results.

With increasing experience and repeated optimization by aligning the simulation with measured results, simulations become more precise and the difference to the measured result decreases. (jw)