(image) (image) [ Home ]

Emulation and Simulation of
Microelectronic Fabrication Processes

6 Summary and Outlook

State of the art simulation and emulation techniques for process technology computer aided design (TCAD) were presented and their implications on the modelled processes were discussed. In particular, modelling strategies for materials and material interfaces were presented and described, focussing on the underlying numerical methods and their relative advantages. A modelling framework for the highly physical simulation and for the efficient empirical emulation of semiconductor fabrication processes was designed and implemented in a C++ high performance toolchain. This framework allows for high flexibility in the modelling capabilities due to a careful consideration of the numerical methods employed. Hence, a number of existing modelling techniques were reviewed, their relative advantages discussed, and several additional algorithms developed to create a powerful and computationally efficient modelling framework for process TCAD.

Continuum modelling was identified as the most appropriate technique for the description of materials involved in semiconductor fabrication. Specifically the sparse-field level set method was shown to provide all the necessary functionality to describe, analyse and manipulate complex material interfaces with minimal computational requirements. Furthermore, the volume representation of materials using a cell-based approach, developed during the course of this work in the C++ ViennaCS library, was shown as a viable data structure complementing the topographic level set description. Using a sparse data approach on the same grid as the level set, allows for efficient access of volume and topography data with minimal memory requirements.

The robust description of material evolution through surface advection is key for the modelling of fabrication processes. Several numerical techniques for the manipulation of materials were reviewed. Due to the robust handling of complex topographic changes, the level set provides better computational efficiency compared to explicit methods.

The iterative advection of level sets and numerical schemes for the solution of the level set equation were presented and their limitations for process simulation were discussed. For highly anisotropic velocity fields, the Stencil Local Lax-Friedrichs integration scheme was developed which reduces numerical errors by setting appropriate dissipation factors and limiting the integration time to robust lengths. The evolution of materials through volume properties can best be achieved using the proposed volume data structure, extracting a velocity field from this structure, and modelling the topographical change using the level set engine.

Geometric advection provides an entirely different way to evolve interfaces in a highly efficient manner. The material is not moved in time, but rather geometric considerations about the evolved surface at the end of a fabrication step are described using geometric distributions which are applied to the initial surface in order to generate the final structure directly. Therefore, material interfaces can be advanced in a single step without any limitations on advection distance or resolution, as is the case when using iterative advection.

The exact properties of material evolution are a complex result of the interaction of atoms and molecules with material interfaces. Therefore, physical models rely heavily on transport mechanisms and surface reactions to be modelled appropriately. Top down Monte Carlo ray tracing was shown to be the most appropriate for this application, as it provides great flexibility in the physical description of manufacturing processes. Using this computational technique, a general modelling approach for ion-enhanced plasma etching was presented.

For the implementation of the sparse-field level set, hierarchical run-length encoding was implemented in a specialised data structure ViennaHRLE which provides fast sequential data access. All presented algorithms operating on the level set were implemented in the ViennaLS library, which provides a comprehensive tool set for the creation, manipulation and analysis of level set surfaces in multiple dimensions. The ViennaRay library provides the Monte Carlo ray tracing functionality using the open-source library embree which is used to perform efficient surface intersection tests. Finally, all computational capabilities are combined in the ViennaPS library, which provides a highly flexible modelling framework encompassing numerous types of process models. Due to the combination of these different computational techniques in one single process simulation library, highly efficient emulation models can be carried out on the same data structure as sophisticated physical simulations, which allows for maximal flexibility when modelling process flows for semiconductor fabrication.

Finally, numerous fabrication steps and models developed using the implemented process TCAD framework are presented. Empirical and physical process models are provided and their relative accuracy and run times are compared. These process models were then applied in full device process flows to generate process-aware structures highly efficiently. A full SRAM circuit at the 5 nm technology node was created using a combination of emulation and simulation techniques to provide a realistic description of crucial sections of the circuit while drastically reducing simulation time by emulating all other sections.

Due to the high flexibility of the implemented simulation framework, numerous additional process models can be developed and implemented straight-forwardly. Especially fabrication processes which strongly depend on volumetric information, such as ion implantation, diffusion, or oxidation could provide great insights into physical mechanisms and allow for an in-depth evaluation of modern process steps and how they influence each other. This simulation of such volumetric properties with sophisticated physical models is crucial in achieving a fully encompassing description of the fabricated structures and has been made possible by the flexibility of the process TCAD toolchain developed during the course of this work.

In order to make the simulator more comparable to experiments, reactor simulations can be used to extract the input parameters to the physical models and thus allow for better integration with processing equipment. However, coupling the simulation framework with fabrication equipment requires deep knowledge of the used reactors, their construction and their operation. Nonetheless, the flexible software design of the implemented framework allows for the integration with equipment-specific software straight-forwardly, making the full simulation of reactors and their effect on the fabricated structures possible.

The structures generated by the process simulation framework ViennaPS can already provide insights into the effects of certain processing techniques. However, the extraction of electrical properties using device simulations still relies heavily on manual conversions to data formats appropriate for device simulators. In order to make full use of the presented process modelling capabilities, a direct integration with device and circuit simulations is indispensable. Therefore, algorithms which allow for the generation of meshes compatible with commonly used device simulators, such as those developed at the Institute for Microelectronics, are crucial and present the logical next step in the development of a full TCAD toolchain.