The increasing complexity of semiconductor manufacturing has shortened time to market and significantly increased research and development costs. The world of AI is increasingly focused on efficiency to address these issues simultaneously. Wafer-based learning, which is an iterative and linear process, is a major contributor to increasing semiconductor development time and costs. Computer-aided design (TCAD) technology has been used throughout the semiconductor industry to supplement wafer-based learning. TCAD uses simulation to develop and optimize semiconductor process technologies and devices. TCAD models the effects of real silicon during the design phase to ensure that the manufactured device behaves as expected. As a result, engineers need fewer silicon wafers for development. Beyond R&D and toward product launch, TCAD’s accuracy on experimental data will be critical to supplementing an even larger portion of wafer-based learning with simulation.
For early technology development, TCAD tools use physically-based models to predict the behavior of devices in new architectures such as Si C-FETs, vertical GaN FETs, and 3D CMOS image sensors. Wafer processing during technology development occurs within process conditions captured by specific physical TCAD model parameters. TCAD calibration is the process of adjusting these physical model parameters so that simulation results match measured data from real devices. Default model parameters are often unable to capture process and technology-specific variations to the extent that would lead to quantitatively accurate and predictive results. Proper calibration ensures that TCAD can be used not only for qualitative device understanding, but also for design optimization and technology development. Calibration, which uses this real-world feedback from test wafers to fine-tune TCAD parameters, is a critical step in maximizing the efficiency of modern technology development.
TCAD calibration solutions
Synopsys provides a comprehensive end-to-end TCAD calibration solution through the Sentaurus Calibration Workbench (SCW). SCW is part of a broader TCAD suite of advanced process and device simulation tools, complemented by a robust graphical user interface (GUI) that streamlines simulation management and facilitates detailed analysis of simulation results. Additionally, Synopsys TCAD provides tools for interconnect modeling and extraction, providing critical parasitic information to optimize chip performance.
This blog post describes two new developments in SCW that will significantly accelerate the calibration workflow for TCAD engineers. These include an expert proofing module that increases user productivity by 5x by allowing users to jumpstart their proofing workflows. In addition, SCW’s new companion product, the Sentaurus ML Calibration Accelerator, reduces calibration time by more than 5x. This post also describes new ML-focused enhancements in SCW.
The value of TCAD increases with the accuracy of modeling against hardware data. Initial models with a wide range of uncertainties are sufficient during the early stages of process development and technology pathfinding as manufacturing plants experiment with different device architectures. As the technology matures and there is a large amount of silicon test data available for promising technology candidates, continuous TCAD calibration helps bridge the gap between initial model predictions and hardware data, which can be used to refine subsequent processes and improve device characteristics.
However, the process of TCAD calibration is far from simple. Users often lack calibration expertise, reducing productivity and creating difficulties in selecting appropriate parameters, calibration sequences, and parameter ranges. If manual calibration takes too long, the model may not recalibrate against the test chip and subsequently predict the wafer quickly enough. In that case, modeling has little value because the process engineer cannot afford to wait for calibration.
The goal is to achieve automatic calibration fast enough so that measurements from each test can be used to update the TCAD model. The calibrated TCAD model can then be used to perform virtual design of experiments (DOE) and make informed decisions about the next wafer run. Artificial intelligence (AL) and machine learning (ML) can automate TCAD calibration and reduce the expertise required by users.
Expert module increases productivity by 5x
SCW is the primary TCAD calibration tool that meets all the requirements outlined above. Unlike manual approaches where engineers run one simulation at a time to fine-tune the model, SCW runs multiple parallel simulations upfront to create a TCAD-based ML model. Use fast inference from this ML model to precisely tune your device in a very short time.
Setting up a proofing workflow requires expertise and time. Synopsys offers expert-curated calibration modules to help you jump-start your calibration workflow. The calibration module includes targeted use cases and a wide range of calibration scenarios (see Figure 1). Users can start with 80% of pre-built workflows, increasing productivity by 5x. These modules must be customized to the customer’s specific needs, and the Synopsys team works closely with users to achieve this. Once developed and customized, these models serve as internal baselines that can be used across technologies or shared and reused across different teams. As shown in Figure 1, SCW comes with a specific set of targeted calibration modules as well as broader modules such as CMOS and SIMS (Secondary Ion Mass Spectrometry) calibration. Synopsys continues to develop new modules and works with users to help them customize them to fit their internal workflows.
Figure 1: The Expert Calibration module increases TCAD engineers’ productivity by 5x as a quick start-up solution.
ML enhancements allow users to create their own modules and workflows
SCW continues to invest in ML capabilities, allowing professional users to create their own modules. TCAD-related ML technology blocks can be quickly reused by TCAD engineers with limited ML expertise. The focus is on optimizing the DOE required to create ML models. Active learning leverages committee queries to interactively populate the DOE and optimize the accuracy of the ML model. Active learning uses up to 50% less DOE compared to random sampling to achieve similar accuracy. Two additional features, weighted and hyperspherical sampling, allow users to focus on high-value DOEs and quickly remove outliers.
For the first time, TCAD uses Uncertainty Aware Neural Network (UANN) calibration. These are ideal for the semiconductor world, where data may not be uniformly distributed in parameter space. Among the many advantages of UANN is that it provides more than five times faster training compared to Gaussian process networks. Because UANN does not go through the training, testing, and validation process of ML model development, it is easy to set up and helps simplify the calibration workflow for TCAD engineers. Additionally, it achieves similar accuracy to parametric ML models while reducing DOE consumption from 15% (Sentaurus devices) to 70% (Sentaurus Topography).
Sentaurus ML Calibration Accelerator: Scaling calculations to speed up calibration
Even with ML-based approaches, calibration time remains a major bottleneck for engineers looking to reduce wafer dependence and R&D costs. To address this challenge, the Sentaurus ML Calibration Accelerator was introduced as a companion feature to SCW and is designed to help reduce calibration time for ML-based workflows. This acceleration is provided through a built-in stackable licensing model, giving users a scalable knob to allocate additional compute to demanding use cases such as 3D calibration (see Figure 2). This ML accelerator reduces calibration time by more than 5x. It is ready-to-use in existing calibration workflows, is calibration use case agnostic, and is a key enabler that brings TCAD closer to the production ramp in semiconductor fabs.
Figure 2: Sentaurus ML Calibration Accelerator is a companion product that helps users accelerate their calibration workflows.
In summary, automatic calibration is essential to improve TCAD accuracy while minimizing expensive wafer costs. With ML-focused enhancements, expert modules, and calibration acceleration, SCW provides a one-stop shop for your TCAD calibration needs. Recent updates increase productivity by more than 5x when creating calibration workflows and make these workflows an additional 5x faster, enabling TCAD calibration within days of acquiring new test data.
