The automotive industry is moving towards centralized computing architectures, software-defined vehicles and – increasingly – AI-enabled systems. The concept of the AI-defined vehicle, morphing from the software-defined vehicle (SDV) and seen as a move from ‘code to cognition’, is one which seems inevitable. Yet there are key challenges to think of beyond integration.  

Abdul Salam Abdul Karim, ADAS platform hardware systems engineer at Ford, has more than 20 years of experience in the automotive electronics industry, with a primary focus on ADAS (advanced driver assistance systems), embedded hardware platforms, and safety-critical automotive electronics. Ahead of his speaking appearance at Microelectronics US in Austin on April 22-23, he discusses the validation, certification and explainability issue with AI-defined vehicles, as well as his research on hardware software co-design frameworks. 

--- 
 
Hi, Abdul Salam. Could you tell us about your role at Ford and the areas you are focused on? 

At Ford, I work in ADAS platform hardware systems engineering, where my focus is on integrating advanced semiconductor platforms into safety-critical automotive applications. 

My role sits at the intersection of hardware architecture, functional safety under ISO 26262, and overall system reliability. A key part of my work involves evaluating and integrating modern semiconductor technologies such as system-on-chip platforms and high-performance microcontrollers into next-generation vehicle systems. 

One of the most interesting developments in the industry today is the transition toward centralized compute architectures. In these systems, semiconductor platform decisions directly influence software behavior, safety strategies, and the overall scalability of the vehicle platform. From a platform perspective, my responsibility is to ensure that the hardware and software are integrated effectively so the system can support advanced driver assistance features in a reliable and safe way. 

This work also involves interfacing with multiple sensor technologies used in ADAS, including radar, ultrasonic sensors, and driver monitoring systems. These sensors generate large amounts of data, and the system architecture must process and combine this information to support intelligent decision-making for the vehicle. 

Ultimately, the goal is to build a robust and scalable ADAS platform that enables safer and smarter vehicle functions. 

Your role seems highly cross-functional. How do you collaborate with different teams and disciplines within the organization? 

My role is highly cross-functional because modern automotive platforms require many different technologies to work together within a single system architecture. 

From a platform engineering perspective, I work closely with multiple engineering teams, including software development, software architecture, functional safety, cybersecurity, and system validation teams. Each of these groups is responsible for a different part of the vehicle platform, so strong coordination across disciplines is essential. 

I am also involved in working with teams responsible for perception systems and sensor technologies, such as radar, ultrasonic sensors, and driver monitoring systems. These systems generate large amounts of data that must be processed and integrated correctly in order to support reliable ADAS functionality. 

Another important part of the role is collaborating with semiconductor suppliers and automotive component suppliers. Since modern ADAS platforms rely heavily on advanced semiconductor technologies, it is important to align hardware capabilities, software behavior, and safety strategies across the entire ecosystem. 

From a system architecture perspective, one of the biggest challenges is integrating different technologies and system behaviors into a single platform that can make reliable decisions under many operating conditions. 

Ultimately, this work goes beyond performance or feature development. Every architectural decision in a safety-critical automotive platform can directly influence vehicle safety and human life. That is why collaboration across engineering domains is essential when designing modern automotive systems. 

You are also very active in research and technical publishing, with several recent papers in journals and conferences. Could you share one or two of your recent research contributions and explain the key ideas behind them? 

Research is something I am personally very passionate about. Along with my industry work, I regularly collaborate with other researchers and engineers and publish technical papers in international journals and conferences such as IEEE and SAE. Over the past few years I have published a number of papers focused on automotive electronics architecture, functional safety, and reliability of advanced driver assistance systems. Many of these topics are directly connected to the challenges the industry is facing as vehicles become more software-defined and increasingly complex. 

One of the recent papers that is closely related to this conference is titled “Cyber-Physical Co-Design Reliability Framework for ASIL-D Automotive Sensor ECUs.” The main idea of this research is that modern automotive systems can no longer treat hardware and software as two separate development processes. In the past, hardware was designed first and software integration happened later. But with today’s centralized compute platforms and sensor-rich ADAS systems, that sequential approach is no longer sufficient. 

What we are proposing in the paper is a hardware–software co-design framework, where both hardware architecture and software safety mechanisms are developed together from the very beginning of the system design. The framework integrates hardware redundancy, software-based fault detection, and cybersecurity monitoring within a unified architecture. 

In our analysis we found that this co-design approach can significantly improve system reliability. For example, the architecture can achieve over 90% diagnostic coverage, with fault detection accuracy close to 95%, and recovery times in the range of a few milliseconds, which is very important for ASIL-D safety-critical systems. 

The key takeaway is that reliability in modern automotive platforms cannot be achieved by hardware or software alone. Safety, security, and performance must be designed together at the semiconductor level, the system architecture level, and the software level. This type of integrated approach will become increasingly important as vehicles move toward software-defined and AI-enabled architectures. 

You mentioned the hardware–software co-design framework. What are the main challenges in implementing it, and how far are we from seeing this approach become mainstream in automotive design?  

Hardware–software co-design is not just a theoretical idea anymore. The goal of my research is to propose a systematic framework that engineers can actually use when designing modern automotive platforms. 

When we design a new ECU or a centralized compute platform, the first step is always system architecture. Engineers must decide how the system will behave under normal conditions and also how it will behave when something fails.  In the automotive industry, safety is extremely important. According to functional safety principles such as ISO 26262, every system must be designed so that if a failure occurs, the system moves into a safe state. This concept is commonly called fail-safe behavior. For example, if an autonomous driving function stops working while the vehicle is in operation, the system must inform the driver and provide enough time for the driver to safely take control of the vehicle. So, in traditional designs the goal was mainly fail-safe operation. If something fails, the system detects the failure and transitions to a safe condition. 

What we are trying to explore in our research is the next step beyond fail-safe design. Instead of waiting for a failure to happen, modern systems should be able to predict failures earlier and take mitigation actions automatically. This can be achieved by combining hardware–software co-design with advanced algorithms and data-driven system monitoring. 

In our framework we integrate hardware redundancy, software fault detection mechanisms, and predictive analytics so that the system can detect abnormal behavior early and respond before a critical failure occurs. This approach can improve reliability and extend the operational capability of safety-critical automotive systems. However, one of the biggest challenges is validation and certification, especially when artificial intelligence or advanced algorithms are involved. 

Modern vehicles collect large amounts of sensor data from cameras, radar, ultrasonic sensors, and other perception systems. Software algorithms analyze this data and make decisions about vehicle behavior. 

If these algorithms are responsible for system control, we must ensure that they behave correctly under all operating conditions. That means engineers must develop very robust validation and certification processes to verify these systems before they can be deployed in real vehicles. 

The key challenge going forward is not only developing advanced algorithms, but also building trustworthy validation frameworks that can certify these systems for safety-critical automotive applications. 

As these validation methods mature, hardware–software co-design will gradually move from research concepts into mainstream automotive system design. 

At CES, many companies such as Qualcomm and NVIDIA were talking about the transition from the software-defined vehicle to the AI-defined vehicle. From your perspective, how do you assess this shift and what does it mean going forward for the industry? 

Over the past 10 to 15 years, the automotive industry has been evolving toward what we call the software-defined vehicle. If we look back about one or two decades, vehicle electronics followed a distributed architecture. Each function had its own ECU. For example, one ECU for braking, one for lighting, another for driver assistance. As vehicles added more features, the number of ECUs kept increasing. The concept of the software-defined vehicle changed this model. Instead of having many dedicated ECUs, we move toward centralized compute platforms where powerful processors control multiple vehicle functions through software. 

One of the key benefits of the SDV is flexibility. By abstracting functionality from the hardware, we can update vehicle features even after the vehicle is released. For example, through over-the-air software updates, manufacturers can improve functionality or add new features without changing the hardware.  

However, this transition also introduces several engineering challenges. First is computing capability. When many vehicle functions run on the same centralized platform, we need very powerful semiconductor devices such as high-performance SoCs and advanced microcontrollers, along with large memory resources such as DDR. 

Second is data bandwidth. Modern ADAS and automated driving systems generate massive amounts of sensor data from cameras, radar, ultrasonic sensors, and driver monitoring systems. Traditional vehicle networks like CAN are no longer sufficient for these data rates, so we increasingly rely on automotive Ethernet communication. 

Third is power and thermal management. High-performance centralized compute platforms require significantly more power, so engineers must design robust power architectures and efficient system-level energy management. 

Now the industry is moving toward the next stage, which many companies at CES described as the AI-defined vehicle. In this architecture, artificial intelligence becomes part of the decision-making layer of the vehicle. AI algorithms process large volumes of sensor data and support perception, prediction, and system control functions. In many ways, AI becomes the new software layer of the vehicle. In the past, software defined the functionality of the vehicle. Going forward, AI will increasingly influence how the vehicle perceives its environment, learns from data, and adapts its behavior over time. 

But the biggest challenge is not simply integrating AI into the vehicle. The real challenge is validation, certification, and explainability. Automotive systems are safety-critical, so AI must operate within a deterministic safety framework. It cannot define its own safety boundaries. Those boundaries must be defined by engineering principles such as ISO 26262 functional safety architectures, and the AI must operate within those constraints. 

Another important aspect is trust. For drivers and end users to accept AI-enabled vehicle systems, the industry must be able to clearly explain how these systems behave and why they make certain decisions. 

So going forward, the future vehicle will combine centralized compute architecture, software flexibility, and AI capability. But the companies that succeed will be those that combine AI innovation with strong engineering discipline in safety, validation, and system architecture. 

Only then can we safely move from software-defined vehicles towards AI-defined mobility systems. 

How far are we from having robust certification and validation mechanisms for these increasingly complex automotive systems? 

Certification and validation are becoming increasingly challenging as automotive systems become more software-driven and AI-enabled. Traditionally, automotive safety systems have been developed and validated using established safety frameworks such as ISO 26262. These standards work well for deterministic hardware and software systems where system behavior can be clearly defined and verified. 

However, as we introduce more advanced software architectures and AI-based algorithms into vehicle systems, validation becomes more complex. These systems process large amounts of sensor data and make decisions in real time, which makes it more difficult to verify their behavior under every possible scenario. 

In practice, this validation often begins with simulation and laboratory testing, followed by large-scale real-world data collection across different driving conditions to ensure the system behaves reliably in diverse environments. 

This becomes even more challenging in modern ADAS and automated driving systems, where decisions are based on large volumes of real-time sensor data. So, the industry is currently working on extending existing safety frameworks and developing new validation approaches that combine traditional safety engineering with large-scale simulation, scenario-based testing, and continuous system verification. 

We are making progress, but the certification methodologies for AI-enabled automotive systems are still evolving. Over time, as these validation frameworks mature, these technologies will gradually move from research concepts into mainstream automotive system design. 

You will be speaking at the upcoming Microelectronics US event. Could you tell us a little about the topic of your session and what attendees can expect to learn? 

At Microelectronics US, I will be contributing in two ways. I will be delivering a technical presentation, and I will also be participating in a panel discussion with other industry experts. 

My presentation will focus on certifying embedded hardware and software in safety-critical automotive systems, particularly in the context of modern ADAS and centralized compute architectures. As vehicles become more software-defined and increasingly AI-enabled, the role of semiconductor platforms becomes much more critical. Modern automotive systems rely on powerful SoCs, high-speed communication networks, and complex sensor platforms to support advanced driver assistance and automated driving functions. In my session, I will discuss how hardware architecture, embedded software, and safety mechanisms must be designed together to ensure reliable and certifiable system behavior in these complex platforms. 

A key theme of the presentation is hardware–software co-design for safety-critical automotive electronics. I will explore how semiconductor-level design decisions — such as safety islands, redundancy strategies, memory partitioning, and fault supervision mechanisms — interact with system-level safety frameworks such as ISO 26262. 

Another important topic will be how emerging technologies, including AI-assisted monitoring and predictive fault detection, can help improve reliability in next-generation automotive platforms. The overall objective of the session is to bridge semiconductor innovation with production-grade safety architectures, helping engineers understand how advanced semiconductor technologies can be integrated into reliable and certifiable automotive systems. 

Microelectronics US brings together professionals from areas such as embedded systems, photonics, and semiconductor technologies. How important is it to bring these communities together to discuss the future of automotive electronics? 

I think it is extremely important because modern automotive systems are no longer developed within a single engineering discipline. 

Today’s vehicle platforms depend on collaboration across multiple domains. Semiconductor suppliers design the SoCs, microcontrollers, memory systems, and safety mechanisms that form the foundation of the computing platform. Hardware engineers integrate these components into reliable system architectures. Software engineers develop the algorithms and control logic that run on top of these platforms. And system integration and validation teams ensure that everything works together safely and reliably. 

What I am trying to emphasize through my session is that no single discipline can solve these challenges alone. Reliable ADAS and automated driving systems require close collaboration between semiconductor technology, embedded software development, system architecture, and safety engineering. In this ecosystem, every contributor plays an important role. Semiconductor suppliers, hardware engineers, software developers, and system architects all contribute to the overall reliability and safety of the platform. 

Events like Microelectronics US create an environment where these different communities can come together, share ideas, and discuss how their technologies interact with one another. That type of collaboration is essential if we want to successfully develop the next generation of intelligent and safety-critical automotive platforms. 

We’ve discussed hardware–software co-design and some of the research and industry work around it. How far can this journey go in automotive systems? What does the future hold for this approach? 

Hardware–software co-design is no longer just a theoretical concept. It is already evolving and being adopted across the automotive industry. 

In the past, development was mostly sequential. Engineers first designed the hardware platform, and only after that was the software developed and integrated. Today the industry is moving toward a parallel development approach, where hardware architecture and software functionality are designed together from the beginning of the project. 

What we are seeing now is the expansion of this concept. It is no longer just hardware and software working together. Modern automotive platforms must also integrate functional safety mechanisms, cybersecurity strategies, over-the-air update capabilities, and increasingly AI-based algorithms. 

Because of this, the most important change is happening in the early architecture phase of system design. Hardware engineers, software engineers, safety specialists, and system architects need to collaborate from the very beginning of the project. 

The goal is to define system architecture, safety strategies, redundancy concepts, and validation approaches early in the development process. So, the journey is far from finished. The future of hardware–software co-design will continue to evolve toward earlier collaboration, tighter system integration, and continuous validation across the entire development lifecycle. 

This will be essential for building the next generation of software-defined and AI-enabled automotive platforms.