An Introduction to Semiconductor Device Modeling
The semiconductor integrated circuit (IC) design industry is witnessing a monumental shift with the emergence of chips harboring over 1 trillion transistors. This advancement is largely due to the development of advanced production nodes, which also introduces a new realm of complexities in semiconductor device modeling.
Device modeling engineers encounter novel effects and massive datasets that challenge traditional, manual modeling approaches. The accuracy required in semiconductor device modeling has also reached unprecedented levels. Device modeling engineers are responsible for creating accurate SPICE models and process design kits (PDK) for IC designs based on both silicon and III-V semiconductors under various operating conditions such as radio frequencies (RF), voltages, and temperatures. To address these challenges, Keysight has empowered semiconductor foundries and design companies with a highly flexible, automated device modeling solution from measurement to verification for over two decades.
This article will delve into the following key areas.
- Current trends and challenges faced in semiconductor device modeling
- The components of a device modeling flow
- The role of a device model in the IC design ecosystem
- Key requirements for device modeling
Readers will acquire a foundational understanding of the basics of device modeling with practical solutions to enhance device modeling accuracy and efficiency in the post Moore’s Law era.
What is device modeling?
Device modeling is the creation of mathematical models and formulas that describe the characteristics of semiconductor devices. It is a critical aspect of the semiconductor integrated circuit (IC) design process, enabling design engineers to predict the behavior of semiconductor devices like transistors, diodes, and capacitors before physical prototyping.
At its core, a semiconductor device model consists of a set of rigorously defined equations that describe how characteristics of a transistor or diode change under various conditions. Users can externally access and set the values of parameters within these equations to reflect different operating environments.
Some common characteristics of a semiconductor device include:
- Electronic characteristics: Voltages (i.e., threshold voltage and voltage drop), currents (i.e., base current and leakage current), and capacitances (i.e., drain-gate and source-gate capacitances) under direct current (DC) and alternating current (AC) signals
- Electromagnetic (EM) and radio frequency (RF) characteristics: S-parameters and noise (i.e., 1/f noise and random telegraph noise)
Figure 1. Sample BSIM model equations
The determination of parameter values is a meticulous process that involves either direct extraction from empirical data or optimization techniques to ensure the model equations closely fit the observed data from experimental results, including varied sizes, temperatures, and operating voltages.
Usage of a device model in IC design
Design engineers use models in circuit simulators such as Spice3, ADS (Advanced Design System), Spectre, HSPICE, PSPICE, Eldo, LTSpice, among others.
The process unfolds as follows:
- Sending a netlist (a detailed description of the device’s parameter values and stimulus) to the simulator.
- Ensuring the netlist syntax conforms with the selected simulator.
- Running the simulator with the model and netlist to predict the behavior of the device or circuit.
- Verifying model simulation results are consistent across simulators.
Accurate device modeling serves as the foundation for developing accurate simulation models and process design kits (PDKs). Design teams can significantly reduce the time and cost associated with the design and testing of ICs.
Types of device models
Device modeling falls into two broad categories: the physical model and the compact model.
Physics-based models
Engineers recognize physics-based device models or their accuracy as they derive from fundamental physical laws. But they are not always fast enough for circuit simulators.
Compact models
Compact models find wider application in high-level electrical simulations of integrated circuits (ICs) and circuit boards. Some are derived from or simplified versions of physics-based models. Others are empirically determined from experimental data. Compact models offer faster simulations due to their simpler nature.
Standardization of compact models enables all stakeholders to agree on the same modeling theories, equations, model parameters, mathematical attributes, definitions, and names. This facilitates the interoperability of different foundries’ PDKs, modeling and design engineers, and modeling/simulation software vendors. The Compact Model Coalition (CMC) is a standards body that publishes a set of standard models for various devices.
Table 1. Standard compact model examples
Example: GaN device model
For a better understanding of device models, we will use a mode of Gallium Nitride (GaN) high electron mobility transistors (HEMTs) as an example.
Figure 2. Core surface-potential-based drain current model
The ASM-HEMT is a physics-based model. Its formulation consists of three parts:
- Calculations of surface-potential and channel-charge.
- Terminal current and charge model.
- Various device secondary effects such as trapping and self-healing.
The core model formulations calculate the drain current, considering the current continuity and the integration of surface potential along the channel under the gate.
Next, various real device effects are added to the core model, including velocity saturation effect, mobility field dependence, sub-threshold slope degradation, non-linear series resistances, channel-length modulation, drain-induced barrier lowering, and self-heating effect temperature dependence.
App note: Extracting ASM-HEMT Model for GaN RF Devices
How does device modeling work
The typical semiconductor device modeling flows require four main steps.
Step 1. Device characterization
Device characterization involves measuring the physical and electrical properties of the devices. This step involves collecting a massive amount of measured data from different wafers over several temperatures. Therefore, it is important for device modeling teams to automate wafer-level measurements for efficient device characterization.
One of the major challenges is automated measurements over temperature. Due to wafer and hardware expansion (or contraction), the wafer mapping software may lose control of the alignment in the X and Y directions. Pattern recognition technology is critical to overcome these problems.
Step 2. Parameter extraction
In this step, specific parameters such as DC, CV, S-parameters, and noise that define the behavior of the device are extracted from the characterization data. The accuracy of the model largely depends on how precisely these parameters are extracted, which is impacted by the type and accuracy of the available device characterization data.
Step 3. Modeling
This step involves creating the mathematical model that represents the device's behavior based on the extracted parameters. The merits of the model are partially determined by the quantity and the type of characterization required.
Step 4. Model validation
This final step is important for modeling teams to ensure the quality of model libraries before they can be deployed by internal or external customers.
Due to smaller device sizes and more complicated models, validating SPICE models can be time-consuming. However, modeling engineers and model users still want the models to be thoroughly checked and the model characteristics to be easily obtained. Keysight’s Model QA Software (MQA) addresses this pressing need by rigorously checking the model quality, plotting model characteristics, and automating model quality assurance and reporting procedures for both silicon and III-V technologies.
Figure 3. The typical device modeling flow
Keysight’s device modeling products have empowered companies like KIOXIA to accelerate PDK development with an end-to-end solution from semiconductor model extraction to verification.
See how KIOXIA Improves Total Modeling Efficiency by 10X
Why is semiconductor device modeling important
To cope with increasingly complex design requirements, design teams rely on device modeling software to thoroughly understand, predict, and optimize the behaviors of devices. The ability to model and simulate device behavior under various conditions allows design engineers to reduce the dependency on costly physical prototypes, saving both time and resources.
To understand the essential role of device modeling in IC design, consider a simplified resistor model. In this scenario, you have a set of measured data points for current (I) and voltage (V). The relationship between current and voltage in a resistor is linear, described by the equation I = V/R, where R is the resistance. By adjusting the parameter R, the best-fit results with the data points can be obtained when R=R0.
Figure 4. A simplified version of the resistor model
This example illustrates two crucial considerations for device modeling:
- Identify the correct mathematical relationship that describes the behavior of the device.
- Adjust the parameters in the equation to best fit the measured data.
Key objectives of device modeling
Accurate semiconductor device modeling is an extremely important procedure to design higher speed and higher frequency integrated circuits and subsystems, enabling the following benefits.
More design flexibility
A robust device model provides a clear physical description of the model’s behaviors and helps design engineers simulate different operating conditions to meet diverse application-specific requirements and push the IC design innovation envelope.
For example, leveraging Keysight’s device modeling and characterization solutions, Professor Yogesh Chauhan from the Indian Institute of Technology (IIT) in Kanpur and Dr. S. Khandelwal were able to develop the industry-first ASM-HEMT model for GaN transistors. The model offers the right amount of compactness, but also a high degree of accuracy and minimal simulation time. It enables faster and more accurate simulation for the growing range of 5G communication and wide bandgap semiconductor power applications.
Reduced design cost
Although foundry models have achieved impressive quality, they may not be sufficient for certain specific applications not characterized during extraction. Therefore, fine-tuning foundry libraries in-house becomes critical for design houses to speed up the model/design iterations, saving IC design costs.
Faster time-to-market
As semiconductor process specifications mature during its life cycle, modeling teams must quickly turn around updated model libraries whenever new specs are available. To enhance efficiency, Keysight’s model re-centering tool enables modeling engineers to align the model to the new process targets and reduces the overall extraction time by 70% compared to traditional model extraction.
Emerging challenges in semiconductor device modeling
The semiconductor industry faces continuing challenges to maximize product performance and yield, decrease time-to-market, and reduce production costs.
As technology nodes get smaller, the need to use accurate device models and control statistical variations in device processing performance becomes ever more critical.
With 5G applications, typical circuit operating frequencies continue to advance well into the RF and microwave frequency range. Design teams need models that can accurately predict device behaviors at DC, RF, and millimeter-wave regions.
Moreover, the amount of data measured for device modeling purposes has been increasing exponentially. With measurements taking several hours or even days, it is essential to be as efficient as possible without compromising measurement accuracy. Measurement control software must work in conjunction with the prober's native control software and each instrument to allow automated measurements across temperature.
Enhancing device modeling accuracy and efficiency with Keysight
Today's most advanced semiconductor foundries and Integrated Device Manufacturers (IDMs) rely on Keysight for modeling silicon CMOS, Bipolar, compound gallium arsenide (GaAs), gallium nitride (GaN), and many other device technologies.
Figure 5. Keysight Device Modeling (IC-CAP)
Keysight’s IC-CAP device modeling software offers modular products so that users can choose precisely the modules required for modeling scenarios. It supports graphical analysis, programming via both parameter extraction language (PEL) and Python, and custom model and user interface development. The analysis module enables simulation, optimization, statistical analysis, and interfacing to external simulators. IC-CAP supports an extensive list of measurement instruments, including industry-standard DC Analyzers, LCRZ Impedance Analyzers, and Network Analyzers.
Within a single environment, modeling teams can use IC-CAP to automate measurements, simulate device performance, extract data, optimize model parameters, perform statistical analysis, and generate best-case and worst-case models. IC-CAP provides extraction routines for industry-standard and Keysight proprietary models such as diodes, BJT, MOSFET, MESFET, HEMT, noise, thermal, and more.
Here are several examples of using IC-CAP.
RF and microwave modeling
Building on proven strengths in RF and microwave test and measurement, Keysight EDA provides configurations for various RF instruments such as the Keysight PNA, PNA-X, and ENA series.
IC-CAP RF extraction modules for proprietary and industry-standard models include RF-dependent parameter extraction, ensuring models are suitable for high-frequency circuit simulation.
CMOS modeling
IC-CAP provides powerful turnkey extraction packages for CMC industry-standard CMOS models:
- BSIM3
- BSIM4
- BSIMSOI
- PSP
- HiSIM
- BSIM-BULK (formerly BSIM6)
- HiSIM_HV
The CMOS Extraction packages share the same architecture, making it possible to use the same measured data to extract different CMOS models.
GaN modeling
IC-CAP offers complete coverage for GaN devices, from the traditional, empirical-based, Angelov-GaN model to the most recent physics-based models recently promoted to industry standards by the Compact Modeling Council, the ASM-HEMT, and MVSG models.
For maximum accuracy, Keysight offers the Artificial Neural Network (ANN)-based DynaFET model.
In addition, Keysight provides device modeling services and special support services for our device modeling products.
Explore our full range of device modeling solutions, and contact us if you have any questions.