Artificial intelligence (AI) and machine learning (ML) require vast amounts of computing resources, and subsequently the chips used in these applications consume more current than ever. This raises the need for more detailed electrical and thermal evaluations at every level of the system design, including interconnects, which often need to cumulatively carry thousands of amperes of current. This article explores the various aspects of electrical-thermal co-simulation of multi-pin interconnects.  With many independent pieces, it is challenging to set up the simulations correctly. This article explores options for different environmental and boundary conditions on a multi-net interconnect, various levels of forced air cooling, as well as the impact of the thermal properties of the connected environment (for instance a PCB and a package) on both the electrical and thermal results.

Ansys Q3D and Icepak (version 2025R1) were used for the simulations. During analysis, some of the geometries created were too big and complex for the solver to handle, so settings were adjusted to drive the tools in the right direction and achieve the required results. This article reviews some important considerations to keep in mind during this type of analysis, such as parameterization, matrix reduction, and mesh settings in Q3D and Icepak.

Model Setup

For the purpose of this article, a 20 x 20 array of socket pins, 30 mm tall on a 1mm grid will be modeled. This could be used to connect high current devices with low inductance, using a checkerboard pattern of pin assignments. Figure 1 shows the 3D view of the socket. Simulation and modeling like described above is a significant advantage during interconnect R&D. Recently, similar techniques were used by Samtec engineers in the development of the mPower UMPx ultra-micro power connectors that handle up to 18 A per power blade, enabling thermal and electrical simulations during the design and development phase.

Figure 1. Isometric view of the 20 x 20 array of pins.

When creating any kind of model in a CAD tool, it is generally recommended to use parameterization. With parameterization, it becomes easier for the designer to make changes. By simply entering a new value for the variable, the entire design is updated automatically. This also includes duplicates created from the original. Parameterization reduces errors that would be caused by manual edits, allowing for consistency across all dimensions of features. Something else to keep in mind is that equations involving other variables can also be used as values for the parameters, as seen in Figure 2. This link between variables allows them to scale proportionally together, keeping the project functionally consistent.

Figure 2. List of parameters used in the project. The parameters are not limited to numerical values, as they also include equations involving other variables.

Matrix Reduction

Ansys Q3D enables designers to reduce the number of terminals by port reduction. This reduces the amount of individual signal nets. When solving large projects with many individual signal nets, the solver may fail due to not having enough memory for AC-RL simulations. (As a side note, to get even more accurate results, enable DC-RL as well). To prevent this failure, operations such as “Reduce Matrix” can be performed, as seen in Figure 3. A Reduce Matrix operation such as “join in parallel” will merge multiple ports, creating a single equivalent port. To do so, first design the conductors and then assign the sinks and sources. Next, select the desired ports and perform the reduced matrix operation by right clicking “join in parallel.”

Figure 3. The signal nets were divided into two groups by matrix reduction. This creates one combined net for the power pins and one combined net for the ground pins.

Q3D Mesh Settings

Using and adjusting the mesh settings is one of the most important steps to steering the solver in the right direction. In the initial mesh settings, there are three options: TAU, classical, or automatic. Typically, TAU would be used in signal integrity applications, as it is recommended for high frequency simulations and tight gaps between geometries. Classical is better suited for projects with large meshes and longer solve times for intricate designs. Because this project was focused on the power integrity of the connector, classical was used when using auto. Along with the classical mesh, Ansys Prime Mesh™ was also enabled. This created uniform meshes better equipped for the moment of methods type solve. This, however, caused the mesh to be unnecessarily dense in certain areas, as seen in Figure 4. As a result, we used mesh seeding to better refine the mesh, as seen in Figure 5. By adding a length-based mesh operation in Ansys Q3D, the mesh can become finer and denser than it is in the default setting.

Figure 4. Ansys Prime Mesh™ creates over 500,000 tetrahedral, creating a very dense mesh. 

Figure 5. Classical Mesh with Mesh Seeding is still dense, but more refined.

Adjusting the mesh seeding controls the length of the tetrahedral and thus the density of the mesh. The goal is to ultimately find a length that allows for enough tetrahedral to form, creating a concentrated enough mesh for the solver, but not too heavy, that the solver takes too long and fails. This may take a few tries before the right value is found; in our case it resulted in a run time of 30 minutes when solving for DC only, but with save fields enabled to be used by Icepak.

Icepak Mesh Settings

If importing an Ansys Q3D project into Ansys Icepak using a target design, the mesh settings will not transfer over. Since Icepak is used for thermal and fluid analysis, it has completely different mesh settings than Q3D. To change the mesh density in Icepak, the user creates a mesh region by highlighting the geometry. Next, select and enter the desired padding settings. This will create a non-model region, as seen in Figure 6. Under the advanced tab, more adjustments can be made through user specifications such as maximum element size, minimum gaps, mesh parameters, and even multi-level meshing.

Figure 6. Custom mesh region of 2 mm absolute offset padding in all directions similarly. This means that the region is 2 mm extended from the furthest part of the geometry in all directions.

Enabling the “Perform Minimal Validations” feature in Ansys Icepak will also help speed up the simulation when dealing with many nets. Like Q3D, it might take trial and error to find the mesh settings for the project to solve. If one ensures that “save fields” are enabled in the Q3D solve setup, Icepak can import that data. When creating the target design in Icepak there are two options: “forced convection” and “natural convection” to specify the size of automatically generated bubbles (see Figure 7). “Forced convection” specifies a rectangular prism with a longer horizontal side with two openings (depending on where the air is blowing). “Natural convection” is also rectangular prism, but the vertical side is longer, and there are openings on all six sides. One can also assign heat with a stationary wall under “thermal” by selecting the face of the object. The pins were assigned uniform current with amounts ranging from 1 A to 5 A. The heat of the interconnect was measured, and if it got past 65°C, a stationary heat wall of 65°C was applied. The maximum temperature the interconnect could reach was based on customer constraints. 

Figure 7. Thermal profile of interconnect with natural convection on the left and forced convection with 2 m/s of air flow on the right.

Conclusion

Matrix reduction, mesh seeding, or assigning mesh regions in the Ansys tools will simplify large projects and help the tool solve faster with less resources when creating complex geometries.

ACKNOWLEDGMENTS 

A special thanks to Dane Thompson and Juliano Mologni from Ansys for providing their assistance. Much of the information gathered from this project was obtained through their guidance and suggestions. Thanks to Istvan Novak of Samtec as well for the mentorship given during the project and contributions to this publication. Thanks to Sandeep Sankararaman of Samtec for spearheading the project.