The automotive industry is undergoing a significant transformation as OEMs grapple with dramatic increases in the compute requirements for modern vehicles. With ADAS (advanced driver-assistance systems), sophisticated IVI (in-vehicle infotainment) systems, and even on-board AI (artificial intelligence), modern vehicles are quickly becoming data centers on wheels.

To deliver these advanced systems and carve out competitive differentiation, automotive companies need to accelerate software development and concern themselves more directly with selecting—or designing—the right silicon to run those workloads, resulting in tectonic shifts in the automotive value chain.

 

The shifting automotive value chain

As vehicles take on data-center characteristics, it creates a tremendous demand for advanced silicon to process massive volumes of in-car data. Leading semiconductor foundries like TSMC are working to deliver automotive standard nodes to address these challenges. At the same time, automakers themselves are diversifying and investing in custom silicon development to optimize costs, increase market differentiation, and gain a competitive advantage. Tesla’s FSD and Dojo chips are well-known examples.

Consequently, legacy OEMs increasingly prefer to work directly with silicon suppliers, ASIC design providers, and software vendors—taking more direct control over the supply chain and increasing the pace of product innovation. For many OEMs and suppliers, this is a significant departure from business as usual, leading to the big shifts we see in the automotive value chain.

 

Increasing in-house silicon competence

An important question all OEMs must ask is, “How much silicon competence does my organization need?” The answer will determine the degree of verticalization they can achieve across the silicon-to-systems value chain. OEMs that have no verticalization today must decide whether to progress to shallow or partial verticalization or full verticalization. Full verticalization requires the highest NRE (non-recurring engineering) effort but offers the greatest differentiation.

Successful silicon-to-system verticalization will require competencies in four key areas:

  • Hardware: From designing electrical/electronic (E/E) architectures and subsystems down to silicon selection.
  • Software: Creating and validating basic software and application layers.
  • Hardware/software co-design: Accelerating development and optimizing both cost and performance.
  • Feedback loop from in-field vehicle fleet: Monitoring that facilitates reliability, availability, and serviceability (RAS) of silicon components.

 

Hardware/software co-design

Automotive OEMs traditionally designed vehicle hardware and software using a hardware-centric, siloed approach. Distinct electronic control units (ECUs) were created for specific vehicle functions, each containing dedicated hardware and embedded software.

As the number of electronically enabled functions increased over time, so did the number of ECUs in each vehicle. Today’s vehicles now contain anywhere from dozens to more than 100 ECUs sourced from a variety of suppliers.

This has resulted in tremendous architectural complexity as well as limited flexibility. With software tightly coupled to discrete hardware, long design cycles (2 to 5 years) became the norm. And because requirements had to be defined upfront, there was little room for iteration or updates between cycles.

Hardware/software co-design techniques help address these limitations. With software and hardware requirements planned in parallel, OEMs can ensure the hardware better satisfies software requirements and software maximizes the capabilities of the hardware.

Transitioning from a hardware-centric, siloed approach to hardware/software co-design is not easy. OEMs face challenges in several areas.

Selecting silicon for future E/E architectures: Over the last 30+ years, multiple industries have reached a tipping point in which transforming the value chain with custom silicon has become highly desirable, if not essential. When it comes to designing next-generation E/E architectures, the automotive industry now faces a similar tipping point.

OEMs now must decide whether to select off-the-shelf (OTS) systems-on-chip (SoCs) or design custom SoCs—or a combination of the two—to power their vehicles. Decisions must be made on a case-by-case basis, with careful consideration and evaluation of priorities, trade-offs, and numerous variables.

For OEMs that decide to design custom SoCs, safety, reliability, quality, security, and PPA (power, performance, and area) all become their responsibility. PPA, in particular, is a critical consideration, especially in EV (electric vehicle) designs where power-hungry devices can impact range.

Selecting the best AI solution: Many OEMs are also beginning to evaluate AI solutions for in-car needs. Most OEMs recognize AI as a critical part of future differentiation, and many want to understand in advance what NPUs (neural processing units) are available and how they benchmark against each other before deciding whether to buy or build a solution.

Optimizing software for existing E/E architectures: While much of the focus in hardware/software co-design is on future systems, legacy OEMs must also support multiple existing E/E architectures. This includes improving the functionality of—or even adding new features to—existing architectures with limited computing power. This can sometimes be accomplished by shifting software workloads, or parts of workloads, from one ECU to another within the vehicle, or by moving the work from a satellite unit to an ECU with more computing capacity.

 

Continuous development, integration, and validation

The automotive industry is still dominated by “big bang” integrations, where the majority of software bugs are discovered after physical prototypes are built. This legacy approach leads to costly delays and untenable R&D expenses.

Static development models no longer work in a world of persistent and dynamic change. The future demands continuous development, integration, and validation.

From ADAS to IVI to powertrain, development teams are searching for ways to make development more efficient so they can get to the validation stage faster, validate more software variants at the same time, and deliver better software quality. There are a number of areas where OEMs are seeking to improve.

Starting software development earlier: In the past, a lot of software development work was dependent on the availability of physical hardware prototypes. Virtual prototyping and verification can reduce the dependency on physical hardware, enabling critical software development work to get started earlier. With the right tools, teams can simulate electronic components and designs and begin testing software before physical hardware or test benches are available, including testing of ECU software and embedded control systems.

Replacing hardware-in-the-loop with software-in-the-loop: Traditionally, a lot of automotive software testing has been performed via hardware-in-the-loop (HiL) approaches that use real, physical components. HiL fixtures are costly and often in short supply, especially during the most critical phases of development, bottlenecking testing and validation. Replacing HiL systems with software-in-the-loop (SiL) systems has significant benefits. The biggest advantage of SiL testing is the ability to identify system bugs and errors earlier. This not only allows for faster fixes but also reduces development time and cost.

Reducing the need for regional prototypes with cloud SiL: Most OEMs maintain a large number of regional software variants to meet regulations and address requirements for specific markets. Testing these regional variants also traditionally relied on HiL systems. By moving from HiL to SiL and moving SiL testing into the cloud, regional development centers can do their development and testing work in parallel, eliminating bottlenecks and reducing costs.

 

Silicon lifecycle management

As automotive OEMs move toward greater control over their silicon needs, they also need to be more concerned with the reliability of that silicon. How can they be certain that selected chips in the field are behaving the way they expect them to behave?

This is where silicon lifecycle management (SLM) comes into play. SLM provides the ability to monitor silicon in the field to ensure the chips are operating as expected. Using SLM, it’s possible to identify and correct minor problems in silicon through firmware updates, and SLM also provides an avenue for feedback into the development process, allowing for changes and refinements to be made to the next generation of silicon.

 

Accelerating the automotive revolution

As vehicles become increasingly defined by their computing capabilities, automotive companies must invest in new silicon and software expertise, build stronger partnerships across the value chain, and modernize their development and validation processes. By embracing these shifts and prioritizing innovation in both hardware and software, OEMs can accelerate time to market, ensure greater reliability and adaptability, and deliver vehicles that are safer, smarter, and more distinctive.

 

Walter Wottreng, VP of Automotive Strategy & Business Development at Synopsys, wrote this article for Futurride.