PSFB 10kW 100kHz Design and Optimization

This Magnetics Note focuses on the design and optimization of a 10kW PSFB with an input voltage of 800V and 500V output, developed by Frenetic’s Field Application Engineer Pablo Blázquez. Follow him in this optimization process:

For this project, I decided to start from an original design by Texas Instruments that I uploaded in Frenetic Online quite a long time ago, and I managed to optimize it. The design is a 10kW PSFB with an input voltage of 800V and 500V output.

In the initial design I used an E71/33/32 core, but, after looking at it carefully, I saw that the losses were quite high and not well distributed. Therefore, I decided to give it a go and select a better core for this application with the support of Frenetic’s Core Optimizer ™.

Initial design

In my previous design, as I mentioned above, I used the E71/33/32. As you can see in Figure 1, the design doesn´t look very balanced:

Figure 1. E71 Results

The initial volume of this design is 70.05 mm Width x 66 mm Height x 58 mm Depth = 268.151,4 mm³ with 610.5 g of weight. We also get a very high temperature, even though we have some cooling.

It is a pretty big transformer, but I really think we can do better than that!

Transformer Optimization

I decided to divide this optimization process in two parts: the first one dedicated to the core and the second one to the windings.

First step: the Core

With the support of the Core Optimizer™, I’ve checked how different core shapes would work for our design. For example, I tested PQ and E cores, looking for a higher power density in our designs in order to reduce the volume and the losses.

Then I’ve checked several materials for the different core shapes that could work at this power levels. I´ve tested 3C95 and 3C97, and in Figure 2 you can see the results of the simulation.

Figure 2. Volume vs Nº Turns

The PQ cores are closer to the origin, and the PQ65/44 chosen in the picture seems like a good option, although I think that we could use the other options in the PQ shapes. For this reason, I’ve tested all the different PQ core shapes that appear in Figure 2 with materials 3C95 and 3C97, in order to see which one performs better at this operation point.

Figure 3. Core Losses vs Nº Turns

As you can see in the results shown in Figure 3, the PQ65/44 with 3C97 looks like the best option for our core.

Once the core is chosen, we need to define the number of turns that it will have: for a balanced design, we should be somewhere close to the 15-20W of losses in the core. So, I chose 20 turns with 15.23 watts of losses.

Then we move on to the windings, we select the optimal winding arrangement and we find wires that would fit in the winding window. Further details on this step comes in the second part.

If we find out that we have more losses in the windings than in the core, we can do more iterations using our inductance calculator. By doing so, we can see how by changing one or several of these parameters (number of turns, inductance, gap or number of gaps) the other parameters will be affected. This will also influence how the losses will be distributed between the core and windings.

As the losses initially were not very well distributed, and the wire that we had chosen was hard to fit in the winding window, I’ve decided to reduce the number of turns. After a couple of iterations, I was able to find the best solution for this design: I ended up with 18 turns on the primary side and with an ungapped core, as we don’t need a specific magnetizing inductance in this design. You can check the results in Figure 4.

Figure 4. Inductance calculator

After all these iterations we were able to make a thought decision as we had access to the information on how our transformer would perform with the different turns. Moreover, we will know if the core is saturated in order to make the best decision possible.

Figure 5 - Core Losses

What we´ve learned in this first steps

When designing magnetic components, we always need to iterate and by doing it in Software with very accurate results we are able to:

• Reduce the design time;
• Waste less material used for building samples;
• Spend less money on the design process.

Second step: the Winding

We can focus now on the optimization of the Windings and all the necessary iterations to find the best number of turns, winding arrangement and wire.

The core has already been selected with the primary number of turns defined in the first part, and the next step now is to check which wire fits in the winding window.

In the initial design, we had 24 turns on the primary side, as you can see in Figure 6. The wire fit in the winding window and the current density worked correctly.

Figure 6. E71 Windings

But now that we have considerably reduced the core size, we need to fit new wires accordingly.

The old design dimensions were 70.5 mm x66.4 mm x58 mm = 271509.6 mm³, while the size of the new one has been drastically reduced to 65 mm x43.5 mm x55 mm =155512.5 mm³. We managed to reduce the dimension by 43%, which is huge.

As a consequence, we need to find a much smaller wire or to reduce the number of turns in the design. If we try to use the same amount of turns as in the old design in the new one, we won’t be able to fit the wires in the winding window.

The solution is to perform several iterations, going back and forth between the Core tab, with the inductance calculator shown in Figure 7, and the Windings tab. This way we are able to find the best wire and number of turns for our design. At every step, we can see if the wires fit in the winding window and if their current density is feasible and doesn’t burn the design. After several tests in the software to meet these requirements, we've ended up with 18 turns on the primary side.

Figure 7. Inductance Calculator

During all the iterations we even tested different wires, Litz round or foil, as well as the winding arrangements, such as PS, PSP, PSPS, Two chambers, or the customized option. If we had to do all these iterations in hardware instead of software, building different prototypes for each possibility, that would take a huge amount of time and money.

However, with Frenetic Online, we were able to find the best winding arrangement option with a considerable reduction in losses.

For this specific case, I've started the design with a very good input: the “Suggest wire” option. This tool provides suggestions of a wire that fits in the winding window and has a current density that works for our design. Once I have this reference, I can select this wire and change the number of parallels, turns and the type of wire.

We also need to take into account the safety requirements, such as insulation. We can add a triple insulated wire on the primary side to meet this parameter, although we have the option of using margin tape to meet the creepage for our design.

The results show that a PSP winding arrangement would give us the best performance for a very simple design process. While PS, which is very simple to build, would provide a higher leakage inductance. Same for PSPS, which is more complex to build and wouldn’t give much better results than PSP.

As we are going to use an external inductor for this design, we don’t need to reach a specific leakage inductance, which is why I've chosen this specific winding arrangement.

For this design, I have decided to use Litz wire on the primary and secondary sides. We have selected triple insulated wire with 0.05 strand diameter and 900 strands, which in the end is going to fit in the winding window. This wire fits in the winding window and has a current density that works for our specifications.

Figure 8. Final design

Now we can compare the optimized design with the initial Transformer at the same operation point, see Figure 9. In the new design, the losses have been reduced by 33% in a 43% smaller design, and are now much better distributed between the core and windings.

Figure 9. Old VS New design performance

We were able to make a thought decision because we had access to the information on how our transformer would perform with the different turns. We also knew whether the core is saturated, that gave us the power to make the best decision possible.

Figure 10. Core Losses

What we´ve learned through the process

Here are my conclusions after performing all these iterations in software:

• By changing the number of turns, we can redistribute the losses between core and windings;
• Doing the iterations in software help us find the best possible options;
• Several wires can be tested to find the optimal solution.

We always need to test several options in order to find the perfect solution for our application, and making the iterations in hardware means longer time and higher costs. The best solution is to work with the software and to reduce the number of hours that our engineers spend designing magnetics.