BEOL Barricades: Navigating Future Yield, Reliability and Cost Challenges
December 15, 2016
Semiconductor Process Development: Finding a Faster Way to Profitability
February 16, 2017Nowadays, novel semiconductor technologies have brought complex process flows to the fab. These process flows are needed to support the manufacturing of advanced 3D semiconductor structures. It can be helpful to model process flows, and their effect on a novel device, prior to physical fabrication.
Process modeling is a technique that can predict the 3D structure of a device using an understanding of unit process steps. During process modeling, unit processes (such as deposition conformality, etch anisotropy, selectivity, etc.) interact with each other and with design data in a complex way to impact the final device structure, just as they would in the actual fab. Process modeling can identify process problems (such as variability issues) prior to actual fabrication, and can eliminate lengthy fabrication cycles needed to qualify a novel technology.
While process modeling is useful, a user might like to export 3D models (built during process simulation) out of the purely physical modeling domain and into other simulation domains, such as electrical device simulation. Predictive process models can be a powerful tool when used in conjunction with third party tools to simulate electrical behavior. For example, we might model the electrical performance of diodes, FinFets and other discrete devices using 3D process models that accurately reflect the physical and material properties of these devices.
Most device electrical simulation tools are TCAD-based tools, which are typically finite-element based. The finite element mesh structures used in these tools are designed to both model the physical structure and solve for the electrical characteristics of the device. Surfaces are modeled with mathematical equations or discrete polygons. This works for simple, well-defined models, but can fail or can be too mathematically complex for the very complex topologies common in semiconductor devices. One of the drawbacks of TCAD modeling is the computation time required to arrive at a solution – both process model solution and device electrical solution. A larger modeled area usually means longer simulation time, and modeling complex devices can become impractical using finite element techniques.
SEMulator3D is a process modeling platform that can interface to third-party TCAD tools to assess device electrical performance prior to fabrication. The major advantage of using SEMulator3D over typical finite-element based modeling tools is its ability to quickly create highly-accurate 3D models over large device areas, thanks to its unique voxel-based mesh and computational engine. A voxel is a 3D pixel, and is the basis of the product’s mesh and computation technology. The voxel-based, physics-driven models created in SEMulator3D are very tolerant and don’t fail due to small mask or model defects. This modeling technique is ideal for arbitrarily complex 3D models, and is highly accurate, fast and reliable.
In order to integrate process models with your TCAD simulation, a user must first input mesh information into the process modeling tool. Users can typically define the element size and other parameters within the mesh at any location. A volume mesh can then be exported to third party TCAD tools or other finite element solvers.
Creating and exporting the best mesh in SEMulator3D sometimes requires trial and error iteration. My own experience has shown that I usually obtain the best mesh results by using the default mesh parameter settings when exporting a mesh. If I can’t get the optimal mesh result using the default settings, I will modify the mesh settings or modify the structure and material regions of the meshed device. Of course, voxel resolution also needs to be considered, because any mesh will vary based upon the structure and dimensions of the underlying physical device. When setting the voxel resolution, I generally recommend using a voxel resolution equal to 25% of the thinnest layer of the device structure. For example, if the thinnest layer of the structure is 1nm, we will use a 0.25nm voxel resolution.
Defining and exporting mesh structures is a critical step when integrating process models with TCAD models. Your reward for a job well done – the ability to quickly and accurately predict changes in the electrical behavior of your device due to device process changes, without having to go through lengthy build-and-test cycles in the fab.
