Robust Design Using Saber Simulator to Improve System Reliability

Robust design concept

　　"Robust Design" is a common and proven development concept that is committed to improving the reliability of a process or product. In order to improve reliability, robust design principles must be an integral part of the design cycle in the early stages, with the goal of protecting the final product from parameters that may adversely affect reliability. As shown in Figure 1, the general robust design approach requires that the four parameters of signal, response, noise, and control be considered during the design process.

　　In the current system design environment, these four parameters have specific meanings.

　　1. Signal parameters

　　Signal parameters refer to the characteristic parameters of the input signal to the system. There are many parameters to consider in this regard, including the type of signal (analog, digital, etc.), amplitude, frequency, spectrum, etc. Designers must understand these characteristics before creating an effective system design. Signal parameters determine the structure of the system input stage, which prepares the input signal and allows the system to process it.

　　2. Response parameters

　　Response parameters are requirements placed on the output of a system. Similar to signal parameters, there are many response parameters to consider. The system must process the input signal in such a way that the output meets the performance requirements. Therefore, the response parameters determine the structure of the system output stage.

　　3. Noise parameters

　　Noise parameters are disturbances that cause the relationship between the system signal and response to drift. They can be internal or external to the system, and most of them cannot be directly controlled by the designer. Often, the designer's only option to eliminate the disturbance is to incorporate noise compensation into the system design. To do this, the designer must first identify and quantify all noise parameters that can adversely affect the system. Then, the designer must choose which parameters need to be compensated.

　　4. Control parameters

　　Control parameters are used to compensate for noise parameters and can be directly controlled by the designer. Its goal is to predict and compensate for those noise parameters that can significantly affect the system and cause it to deviate from the nominal performance. When multiple compensation solutions for a noise parameter are possible, the principle of robust design advocates the use of the simplest and most cost-effective method. To meet this goal, the designer must often choose control parameters that can mitigate multiple noise parameters.

　　Design Example - Automobile Braking

　　To illustrate how robust design parameters apply to system design, let's take a basic automobile braking system as an example. In this example, the car is assumed to have disc brakes and the goal is to completely stop its rotational motion.

　　In theory, the operation of the brake system is quite simple. The driver applies pressure to the brake pedal, which is transmitted hydraulically or electrically to the brake caliper, which pushes the brake pads against the rotating brake rotor. The force applied by the brake pads to the brake rotor ultimately slows the vehicle down to a complete stop.

　　Referring to Figure 1, the input signal to the system is the pressure applied to the brake pedal. The main signal parameter is the amount of pressure applied. The system responds to the brake pedal pressure by reducing the vehicle speed. The main response parameter is the length of time required to bring the vehicle to a complete stop.

　　Figure 1: Parameters in a robust design.

　　There are many noise parameters that can affect the braking system's ability to stop a vehicle. Common noise parameters include the weight of the vehicle, the condition of the tires, the type of surface the vehicle is driving on, the condition and temperature of the braking surfaces, and weather conditions, all of which are present at any time while the vehicle is in motion. Designers must understand all of these parameters and prioritize them based on their impact on braking system performance.

　　Designers can choose from several control parameters to compensate for brake system noise parameters. Common control parameters include the size of the braking surface, computer control of braking force, suspension stiffness, and increased brake assist. Designers must select the combination of control parameters that best meets the system performance specifications.

　　Once the critical noise parameters are identified and the control parameters are selected, the robust design process can be used to implement and analyze the design to ensure the reliability of the brake system. The goal of the robust design process is to meet the performance requirements with the highest system reliability and the most reasonable system cost.

Robust design process

　　In the modern system design environment, applying robust design principles to improve reliability means making the system's performance independent of changes in design technology, component parameters, manufacturing processes, and environmental conditions. In a robust design flow, these changes become noise parameters that affect system performance. System designers must find control methods that help compensate for each change. Control methods can be as simple as selecting high-precision devices or involve implementing new control algorithms. However, the traditional design-prototype-test process is no longer practical because the matrix of possibilities has become too complex. Designers must move their design activities into the virtual world, where powerful simulation tools such as Synopsys' Saber simulator can support the design and verification of the entire system using robust design principles.

　　The robust design process is often customized according to the company's specific requirements and system applications. There is no one-size-fits-all approach. However, even with a customized approach, there are still some common elements in every robust design process. A complete development process based on robust design techniques should include some of the steps shown in Figure 2.

　　Figure 2: Steps in the robust design process.

　　1. Nominal design

　　The first step in the robust design process is to complete the nominal design of the system. The system must be able to achieve performance that meets the technical specifications under nominal conditions. The results of the nominal design become the response targets for the remaining analysis steps in the robust design process.

　　The Saber simulator supports nominal designs with standard analyses (operating point, time domain, frequency domain) as well as a large library of behavioral and characterized models.

　　2. Sensitivity analysis

　　After the nominal design phase is completed, the system must be subjected to a sensitivity analysis. The designer must determine which design parameters have the greatest impact on system performance. The purpose of this analysis is to determine how much the system performance changes when various parameters are changed. In a sensitivity analysis, the impact of each parameter is calculated separately. By analyzing the data, the designer can determine which parameters have the greatest impact on system performance and determine which parameters to focus on in the subsequent design process.

　　The Saber simulator supports detailed sensitivity analysis. Designers can include all design parameters in the analysis process, or specify a list of parameters that are most likely to affect system performance. Each run changes only one of these parameters, and the designer can specify the amount by which the parameter will change.

　　3. Parameter Analysis

　　Parameter analysis allows designers to fine-tune the device parameters that most affect system performance. The purpose of this analysis is to determine a set of parameters that best meet performance specifications by varying a specific parameter within a certain range. Once the parameter values ​​are determined, the focus is on verifying performance within a certain range of environmental conditions.

　　The Saber simulator gives designers access to all system parameters. Parameter values ​​can be swept over a range in a variety of ways, including linear stepping, logarithmic stepping, fixed stepping, or a fixed set of values. Parameter sweeps can be nested within each other to cover all possible combinations of values. Environmental parameters such as temperature can also be swept.

　　4. Statistical Analysis

　　Statistical analysis is used to study how random combinations of parameter values, calculated based on tolerances and statistical distribution information, affect system performance and reliability. A series of simulation runs are performed, with random variations in parameter values ​​in each simulation run. Depending on the system, hundreds or even thousands of runs may be required to obtain statistically significant results. These results are then statistically analyzed to better understand the reliability profile of the system.

　　It is worth noting that the computational workload of statistical analysis can be extremely heavy. Performing performance simulations of complex systems thousands of times or more will consume considerable computing resources. We can use tools that support distributed computing to alleviate this resource demand.

　　The Saber simulator supports advanced statistical analysis. The parameter values ​​of the behavioral model can be assigned tolerances with a variety of statistical distributions, ranging from predefined distributions to user-defined distributions. Many of Saber's characterization models contain tolerance and distribution information. These tolerances and distributions, when analyzed by Saber's Monte Carlo, can provide an accurate statistical overview of the system. The Saber simulation environment supports textual and graphical statistical data analysis.

　　5. Stress analysis

　　In stress analysis, the system is simulated to see if it will cause some devices to exceed their safe operating range when meeting performance indicators. All device parameters are given maximum ratings to see if their operating parameters exceed the maximum ratings, which is considered an overstress condition. Stress analysis requires characterization of the device using performance rating data.

　　Many models in the Saber library have built-in performance ratings or allow rating information to be added during the model characterization process. With the rating information, Saber's stress analysis can analyze the stresses to which the model is subjected during operation. Saber then generates a report detailing the stress conditions to which each device is subjected.

　　6. Failure mode analysis

　　The final step in the robust design process is to determine the behavior of the system when individual components fail. Depending on the type of system and the technology used in the system, the failure of a single component can cause the entire system to fail, or the system can continue to operate but not meet the design requirements, or the system can recover from this failure and continue to meet the performance indicators. The failure mode requirements are usually stated in the design technical specification and must be verified during the design process.

　　Saber's Testify Failure Mode Analysis tool helps designers set up and run failure mode experiments in their system designs. During failure mode analysis, components can fail in a variety of ways and at a specified time. When a component fails, Saber can continue to perform simulations, allowing designers to study how this failure affects system performance.

　　Choosing the Right Tools

　　Implementing an effective and efficient robust design process requires the use of simulation tools with special capabilities. The key requirements for the tools are simulation support, model library support, modeling language support, and advanced data analysis.

　　A robust design process cannot be established with just a few standard analyses. The simulator must have specific, built-in capabilities for each step of the robust design process: nominal design, sensitivity analysis, parametric analysis, statistical analysis, stress analysis, and failure mode analysis. Simple support for these advanced analyses is not enough, and designers must be able to customize the models and analyses to meet specific system design goals.

　　In addition to advanced analysis capabilities, simulators must be supported by accurate model libraries. A robust design flow requires both behavioral and characterized device models. To ensure accuracy, models should be based on equations that define device behavior. Behavioral models allow designers to easily access key parameters. Characterized models should be created using data collected from bench tests rather than from device manuals.

　　No matter how extensive the model library is, there will always be a situation where the required model is not available. Therefore, the simulator used in a robust design flow must support a variety of standard modeling languages ​​that allow designers to create models based on real device formulas and that are well used and proven in the designer's industry.

　　Finally, the simulator must be supported by powerful post-processing tools for analyzing simulation data. These tools should allow designers to drill down into the details of the design and enable the measurement, combination, and transformation of design data so that designers can gain a complete and accurate understanding of system performance.

　　As mentioned in the robust design flow above, the Saber simulator enables an effective and efficient robust design flow because it supports both advanced analysis and model libraries. Saber also supports the MAST (de facto standard) and VHDL-AMS (IEEE standard) modeling languages ​​that are well used and recognized in the field of system design. In terms of data analysis, the Saber design environment includes the CosmosScope tool, a feature-rich post-processing tool that designers can flexibly control when analyzing design data.