The shift toward software-defined vehicles (SDVs), electric vehicles (EVs), and ultimately autonomous vehicles (AVs) is proving the value and exposing the weaknesses in simulating individual components and complete vehicles.

The ability to model this intensely complex maze of real-world interactions and possible scenarios is improving, and it’s happening faster than comparable road-testing of vehicles over millions of miles. Ultimately, this will make vehicles safer and enable them to respond better to unusual situations, which will be required for full autonomy. But there is still a learning curve when it comes to virtual automotive design.

Vehicles are safety-critical systems of systems, and each new generation has entirely new features, enhancements, architectural changes, not to mention nearly perpetual updates to software and firmware. There are advanced driver assistance systems (ADAS), telematics/infotainment, functional safety, and IoT content. Just keeping track of them all, let alone understanding and accounting for all the possible interactions, is nearly impossible to track without a mammoth amount of simulation horsepower.

“The automotive industry is witnessing transformative changes driven by electrification,” said Kumar Srinivasan, senior technical director for CFD solutions at Cadence. “OEMs that historically focused on internal combustion engine (ICE)-based powertrains are now transitioning their product lines to offer a blend of ICE, mild hybrid electric vehicle (mHEV), plug-in hybrid electric vehicle (PHEV), and fully battery electric vehicle (BEV) powertrains.  The additional investment needed to facilitate electrification has required OEMs to optimize cost of development by replacing physical testing with simulation technology more aggressively.”

These changing dynamics create new challenges for the simulation community, as well. There are new use cases and design considerations that need to be assessed, and increasing complexity involving electronics integration, thermal systems, software controls and calibration, cross-functional design optimization, and new regulations. “The laser focus on improving the energy efficiency of all subsystems is driving more collaborative and cross-functional simulations,” Srinivasan said.

The goals of simulation
Defining goals is the starting point for any effective simulation.

“The use of simulation can be powerful tool to accelerate the validation of software-defined vehicles,” said Marc Serughetti, senior director of product line management at Synopsys. “It enables development to start earlier, before any physical systems are available, and be more productive, identifying root cause analysis more easily or executing large set of tests in parallels. When planning for deployment of simulation, the first focus of OEMs should be on defining its intended objectives and use cases. Such understanding will enable the creation of a model build plan (MBP), resulting in faster model availability and as a result optimized benefits from using the simulation to start earlier, validate faster with lower cost and ultimately deliver better and safer products.”

However, to be able to design a fully functional virtual vehicle without prototyping is still a work in progress.

“It is ideal, of course, to simulate everything about the entire vehicle,” said David Fritz, vice president of hybrid and virtual systems at Siemens Digital Industries Software. “But let’s look at what’s practical. What’s being done today is simulating individual electrical control units (ECUs), which is a very high priority, but isolated and limiting. ECUs transmit and receive data over the car’s network between other ECUs, sensors, and actuators. Inside the chips — within ECUs, which are being completely changed to meet the autonomous and ADAS requirements — there is another level of communication happening within the chips, and external to the ECU. This system of systems is a perfect area for simulation. There are many things to consider within an electrical system simulation. Can we explore the architecture of that electrical system? Do we want one centralized monolithic control unit and distribute the data, or do we want to create a zonal architecture, which will have some central compute and some compute at the edges of the vehicle itself? There’s ongoing debate about these options. Instead of making arbitrary decisions, wouldn’t it be great to simulate those alternatives and get empirical data for making good decisions?”

Fritz explained once there is a system of systems simulation and information on how the system ought to behave, it’s possible to understand what kind of SoC design will be needed, as well as what its impact will be on the PCB design for the ECU. “With all that data, you can determine how everything should fit into the chassis. Do I need a special enclosure for cooling? Is it water cooled? Is it air cooled? Is it passive or actively cooled? The electrical functionality is going to drive the vehicle architecture. It will intersect with mechanical simulations, the friction of the tires on the road, the transmission, steering control, the braking control, the trunk opening and closing. Those are dynamic mechanical models, which will feed data into and receive data from the electrical system models. They all need to be tied together to have a complete solution, which in the virtual world will behave like the vehicle will behave in the production world.”

The science behind simulation
Simulation has been used in critical component development. It offers information on which to base analysis and design. It can be used to examine aerodynamism and structural integrity, thermal behavior, and load distribution. In addition, vibration, stress, and the strength of various materials such as steel, aluminum, and plastic can be studied. So can crash impact in different situations and emergency braking effectiveness. Everything can be simulated, from body and chassis design to SoCs, electronic modules, in-cabin driver and passengers monitoring, infotainment systems, sensors, and more recently, ECUs and ADAS.

“When it comes to the electronic component in the car, there are two main elements,” noted Christophe Bianchi, chief technologist at Ansys. “The first is safety and safety standard compliance. For example, when you have an inverter, which is usually silicon carbide, and big transistors that convert the current of the battery to drive the EV engine, or when you charge the battery — that inverter is very complex. You have electromagnetic, thermal, electrical, and mechanical behavior, all the physics playing together. We model those elements to create a virtual model of the complete system, and push the limit of the use cases to check that it is safe and to ensure that the battery is not going to overheat and create thermal runaway. This is one of the issues we have with electric batteries.”

The second part of simulation is ADAS simulation. “It would take far too many millions of kilometers of driving to test out the many different scenarios,” Bianchi said. “It is now understood that to get a certain level of certification, like L3, L4, or L5, you need virtual validation. And that simulation is very specific. It’s guaranteeing that the intended function is operating safely.”

The simulation algorithm will provide real-time feedback and diagnosis in self tests, so design optimization via modifying certain variables can take place.

Two common testing methods used in simulation are software in the loop (SIL) and hardware in the loop (HIL). SIL refers to the process of validating the software codes in a simulation environment to ensure the codes perform according to the specification. HIL refers to testing the hardware by simulating real-life environment(s).

Fig. 1: Hardware in the loop test system enables virtual vehicle road testing by stimulating real-life input signals from sensors. Source: Keysight

Fig. 1: Hardware in the loop test system enables virtual vehicle road testing by stimulating real-life input signals from sensors. Source: Keysight

In software-defined vehicles, the software increasingly is used to control the various parts of the vehicle, such as an ECU. Testing the full capability of the ECU is a non-trivial task. If road tests alone are conducted, the variety of scenarios tested will be limited.

To overcome this challenge, with HIL testing the ECU can be connected to sensors, such as cameras, through which data will be generated and sent to the ECU with many possible scenarios. This method ensures the ECU will react in real-time and make the correct decisions. In other words, the HIL test system generates/simulates real-life input signals, as if the ECU is actually on board a vehicle doing physical road tests. Other benefits of HIL include repeatability and cost-effectiveness for testing various scenarios.

HIL makes it easy to try out new software releases, and that is reflected in its market growth. ReportLinker projects the global HIL market will reach $1.5 billion by 2028.

What is simulation
Definitions matter when it comes to simulation. “The term ‘simulation’ encompasses various meanings and approaches, often accompanied by terms like virtualization and emulation,” said  Chris Thibeault, senior manager, partner and ecosystem management at Infineon Technologies Americas. “When considering Original Equipment Manufacturers’ (OEMs) diverse applications, these concepts play vital roles.”

This is especially evident with vehicle development. “At its core, virtualization is about creating digital representations of physical components,” Thibeault said. “At Infineon we construct virtual models of chips that are essentially software-based replicas of actual hardware. This practice requires collaboration with our partners to ensure accuracy. As these virtual models for various components accumulate, they can be integrated into a virtual ECU. Emulation enters the scene when deciding the depth of testing required for the ECU’s behavior — in essence, abstracting lower system layers, such as driver-level software, which demands substantial computational resources. Emulation of ECU behavior becomes essential to expedite testing and development without sacrificing the functional performance aspect. This process helps streamline development by focusing resources on areas of the system that need the most scrutiny.”

Many different simulations happen at various levels of scope. “At the ‘systems-of-systems level, OEMs simulate car-to-car communication and car behavior on streets to validate ADAS-relevant situations, for instance,” noted Frank Schirrmeister, vice president solutions and business development at Arteris. “At the ‘system level’ of cars, OEMs assess EM, thermal, electrical, and mechanical effects in the virtual environments to simulate the physical environments — often called digital twins. ECUs, chips, and IP have complex design chain interactions. To simulate these interactions, vendors use specific IP blocks for SoC or chiplet-based designs to verify functionality, performance, power, safety, and security.”

The tools used here are traditional EDA tools – like Verilog and SystemC simulation – augmented with tools for SystemC-based performance analysis, and safety simulation of stuck-at faults, among other things. System-simulation tools for EM, thermal, and fluidity aspects extend the traditional arsenal of EDA tools. Different types of simulation are available to meet different  component testing needs.Fig. 2: Using simulation in automotive design, and enables OEMs to go through most of the design process using software. Source: Ansys

Fig. 2: Using simulation in automotive design, and enables OEMs to go through most of the design process using software. Source: Ansys

Additionally, Cadence’ Srinivasan noted some new areas for simulation:

  • Custom chip design;
  • Model-based engineering to facilitate SIL/HIL model in the loop development and validation;
  • Simulations to develop and validate ADAS and human-machine interface (HMI) systems;
  • Controls and calibration development, and validation during the virtual phase to reduce physical testing costs and improve quality;
  • Battery thermal management, thermal runaway, etc.;
  • Increased focus on Noise Vibration and Harshness (NVH);
  • Electronic systems integration and validation in the vehicle environment, and
  • Multi-disciplinary optimization during the early architecture phase.

Developing fully functional virtual vehicles
The ultimate goal for OEMs is to develop virtual vehicles operating in a virtual environment running multiple billions of miles. Realizing this goal would enable simulating nearly all of the possible scenarios without prototyping. If everything worked, the first pilot unit would be 99% perfect with minimum fine-tuning. But while the industry is heading in the right direction, it is not there yet. Additionally, there are still gaps in today’s simulation tools.

One of the simulation tools in designing fully functional virtual vehicles is the “digital twin.”

“The goal is to be able to simulate the entire vehicle, but the journey has only just begun,” said Hirofumi Kawaguchi, vice president of Renesas‘ HPC Software Solution Division. “Renesas provides system or ECU-level simulation beyond chip-level simulation, which consists of multiple device models. For example, by integrating and connecting multiple simulators that were previously used for single-chip individual devices, such as SoCs and microcontrollers, Renesas is delivering new simulation environments for multi-device operation, in the form of co-simulation environments for multi-devices facilitating optimal system design. Designs can now be optimized by balancing different application functions and incorporating software verification at the systems level.”

Mercedes-Benz uses various simulation tools for different purposes from the development process to the production of vehicles. Crash simulations, for example, are used to explore more scenarios and create a higher maturity before going into the physical testing phase of prototypes. Occupant simulation is used to develop the restraint systems at an early stage. Additionally, the advanced driver assistance systems rely on simulation tools to improve comfort and active safety functions.

Digital twins
Some OEMs envision creating a complete virtual vehicle in digital form using what’s called a digital twin. The concept has gained in popularity in recent years. Theoretically, the digital replica of the vehicle or virtual vehicle will behave exactly like the real thing. Today, OEMs have achieved various degrees of success.

“Digital twins increasingly are being developed to replicate physical tests more closely, such as wind tunnel tests, crash tests, ADAS system validation duty cycles, etc.,” said Cadence’s Srinivasan. “The cost of simulation needs to be carefully balanced in all these instances. Reduced-order models that can capture the behavior of the system/subsystem to the degree needed to replicate physical tests in the simulation models are also increasingly being adopted. For each subsystem, several pre-defined simulation load cases are identified based on design specifications. The specifications typically are a combination of requirements defined by OEMs, as well as supplier-identified design envelopes.”

Further, digital twins encapsulate a comprehensive digital representation of a physical entity, which could range from individual components to an entire vehicle, even extending to broader constructs like buildings or cities.

“In the automotive sector, the aspiration is to develop a complete digital twin of an entire vehicle,” Infineon’s Thibeault said. “This is an ambitious objective that aims to replicate not only the physical attributes, but also the functional and behavioral aspects of the vehicle. A complete automotive digital twin includes a multitude of components, ECUs, and networking architecture, and even extends to simulating vehicle performance across diverse environmental and weather conditions. Safety is also a concern, and simulation offers the opportunity to inject faults into a system to prove out fail-operational functionality, which is especially needed for autonomous driving.”

Creating digital twins involves multiple facets. “Virtualization tools, offered by different suppliers, play a role in generating models of individual components and ECUs,” Thiebeault said. “Solutions like environmental driving simulators come into play, too. The environmental driving simulators consider the digital twin within a broader context, such as road conditions, traffic scenarios, and environmental variables.  The range of functions that a digital twin can simulate is wide and varied, depending on the desired level of abstraction and performance analysis. What’s important is to strike a balance between computational demands for verification and validation, and the assurance of high-quality results.”

Big picture, these software simulation tools are used early in the development cycle, as Sven Kopacz, autonomous vehicle section manager at Keysight Technologies noted. “Ranging from subcomponent levels to entire vehicles, dedicated software makes simulation of the underlying sensors’ behavior, vehicle dynamics, and weather conditions possible. Specifically, using digital twin technology, automakers can use the vast quantities of data collected from autonomous cars’ drive tests to build complex simulations of how an autonomous vehicle’s artificial intelligence (AI) will respond to some of those unpredictable situations, such as weather conditions like hail and snow, or traffic jams. On the right side of the V model for software development there is the physical system. Continuing with this example, a prototype car is built and taken to the test track, or on public roads, to create those same software-simulated situations. This testing methodology is expensive, impractical, inefficient, and potentially dangerous. A digital twin is not, of course, replacing tests of a physical product. It only reduces the amount of physical testing required.  This also means that industry testing solutions must evolve to meet more complex testing requirements.”

Hybrid models
As OEMs work toward designing complete virtual vehicles with simulation, the use of hybrid models will steadily rise.

“We see more and more hybrid simulation approaches in which different levels of fidelity connect to build a representation of subsystems,” said Arteris’ Schirrmeister. “Think hardware-based simulation of accurate accelerator representations with abstracted models for processors and networks-on-chips. In testing autonomous driving, how would simulation work in terms of testing? How will the sensors detect moving objects? Will the virtual vehicle be presented with some moving objects in the form of digital signal input to see how fast the emergency brake would be applied? This harkens back to the incident earlier this year when a Cruise Chevy Bolt robotaxi rear-ended a San Francisco Muni bus. If this scenario had been simulated, that would never have happened.”

Hybrid models combine the best of both worlds. While simulation is a software model of the real thing, it is subject to the quality and accuracy of the input algorithm. AI does not seem to be learning fast enough. There have been multiple robotaxi accidents in the past few years. In August 2023, a Cruise robotaxi collided with a San Francisco fire department vehicle. During the development cycle, the simulation and testing processes should have included this scenario to prevent the collision from occurring. Improving simulation algorithms remains the biggest challenge.

Additionally, the simulation process, including the digital twin, requires enormous computing power and energy. That is why some simulation solution providers are teaming up with cloud computing companies to offer cloud-based, on-demand computing resources so OEMs do not need to have super-computers on site. Finally, as digital twins scale to handle more complex automotive design, the cost of development also will increase. OEMs will be forced to make the decision of balancing 100% simulation versus using the hybrid model of combining simulation with some physical tests.

The notion of 100% simulation with zero prototyping is very attractive, and as such it is not a surprise why so many OEMs are pursuing simulation as a mean to stay competitive. But there are still questions to be answered. A tremendous amount of data will be generated in simulation, including that provided by Tier One suppliers and the OEMs. Where will the data be stored and who will own the data? What standards and regulations will be needed to regulate the industry?

Regardless, for the foreseeable future the demand for automotive design simulation tools will continue to grow and improve.