Which HAL does the car of the future need: a diplomatic protocol between chips and code ✛ ✎
Context
Today's automotive electronics resemble a Tower of Babel: each ECU maker uses unique interfaces and data formats. Software is soldered to a specific chip, blocking updates, complicating diagnostics, and stifling innovation. The SDV paradigm demands the opposite: software should live its own life, like an app on a smartphone, independent of which processor is under the hood. The key to this separation is the hardware abstraction layer (HAL), a model of diplomatic protocol between the world of chips and the realm of programs. Without it, you can't achieve safe driving with instant response, integrate video streams from cameras, deploy self-learning driver assistance algorithms, or seamlessly connect cloud services. Current HAL solutions, born in the automotive industry, bear the imprint of many years of evolution and are ill-suited to multi-supplier ecosystems and cloud-to-edge computing.
Methods
A team of von Neumann–trained researchers from the IAS in Stuttgart conducted a twofold analysis. First, they examined non-automotive HALs: Android Treble with its HIDL, the Asset Administration Shell for Industry 4.0, network virtualization via NFV, and cloud interfaces like VirtIO. From these, they distilled two main mechanisms: the API/middleware approach and hypervisor-based virtualization. Then they classified automotive solutions along the same lines—AUTOSAR Adaptive, Android Automotive, COVESA, SOAFEE, Apex.OS (first group) and QNX, EB Corbos, Xen, KVM, RTA-LWHVR (second). Evaluation used 14 criteria, categorized by the MoSCoW method into must-haves, should-haves, and could-haves. Each HAL scored points: 5 for a must-have, 2 for a should-have, 1 for a could-have, with a maximum of 45. Data were drawn from certification documents, papers, benchmarks, and public design specifications.
Results
The core finding crystallized quickly: hypervisor-based HALs (QNX, EB Corbos, RTA-LWHVR) win by a wide margin on the mandatory criteria—safety, OTA support, low latency, and hardware efficiency (Fig. 3). For instance, QNX runs with less than 2% overhead, while RTA-LWHVR hardware-separates critical and non-critical functions. Open-source hypervisors Xen and KVM fell 15 points short due to lack of ISO 26262 certification, but their communities are moving in that direction. The middleware camp is mixed: AUTOSAR Adaptive, an industry standard, fails on resource efficiency and scores 37 points; Apex.OS, ASIL D–certified, rises to 43, losing points only on toolchain costs. Android Automotive, lacking functional safety artifacts, scores only 29. COVESA, being more of a signal standardization bus, is not competitive. SOAFEE, still under development, promises a perfect 45 points by combining hypervisor isolation with cloud-native middleware. The numbers speak clearly: no single existing HAL meets all the demands of an SDV.
Implications
The scoring isn't just a ranking—it exposes an architectural divide. Hypervisors are great for isolation and consolidation (one SoC instead of multiple ECUs), but they're expensive and proprietary; middleware offers flexibility and development speed, but lags in determinism and safety. Tomorrow's SDV can't pick just one. The paper concludes: a hybrid HAL is needed, where the critical loop (braking, steering) is protected by a hypervisor, and everything else—from infotainment to cloud telemetry—communicates via standardized middleware. This diplomatic protocol will prevent a navigation glitch from affecting the brakes, while still allowing OTA updates to the infotainment system without rebuilding the entire firmware.
Future development
The next step is to implement the hybrid scheme in software and test it on a real SDV prototype. Integration with the Eclipse SDV toolchain promises openness and faster industry adoption. In the long run, this could lead to a single HAL standard, much like HTTP became the universal protocol of the web. The car would then become a digital twin of itself, updatable on the fly and seamlessly adapting to new hardware.
Impact
The results hit right at the heart of automotive standardization—AUTOSAR, ISO 26262, future SDV profiles. Manufacturers will gain a tool to cut development costs, while suppliers get clear rules of the game.
Next steps
Validation of the hybrid HAL on real hardware with an open codebase and publication of a reference implementation.
Key open problems
The work addresses a fundamental problem in complex systems engineering: how to combine hard real-time predictability with evolutionary flexibility. The SDV is a proving ground where this problem is most acute, due to safety requirements, supplier diversity, and the long product lifecycle. Failure here will set the concept of connected mobility back by years.
The knowledge machine's view
This section is addressed to the author of the paper and their colleagues. Neighbouring works were found by meaning-based search across our archive, not by word overlap. We do not review: we show the field around the work.
What the work does
This paper presents a comparative analysis of hardware abstraction layers (HAL) for software-defined vehicles. The authors examined non-automotive approaches (Android Treble, Asset Administration Shell, NFV, VirtIO) and classified automotive solutions (AUTOSAR Adaptive, Android Automotive, COVESA, SOAFEE, Apex.OS, QNX, EB Corbos, Xen, KVM, RTA-LWHVR) against 14 criteria using the MoSCoW method. The results highlight the advantage of hypervisor-based HALs on mandatory safety and efficiency criteria, as well as the potential of hybrid approaches.
Where its strength is
The strong point is the systematic comparison using a unified methodology with clear quantitative assessments, which reveals trade-offs between safety, flexibility, and performance. This provides a foundation for informed HAL selection or design in the automotive industry.
Where it could go next
- It might be interesting to see how other fields solve the problem of hardware abstraction to ensure system survivability and autonomy—for example, in space modules that heal themselves without commands. They use self-healing and decentralized control mechanisms that could inspire more autonomous HALs for vehicles.based on: Space modules heal wounds without commands arXiv:2607.13444
- A related work explores how chaos in stellar memory stores the history of the Galaxy. Although the topic is far from automotive, it investigates how complex systems retain information under instability. Perhaps analogies with such mechanisms could suggest new approaches to state management in dynamic automotive environments.based on: Symphony of Stellar Memory: How Chaos Preserves Galactic History arXiv:2607.10863
Worth trying
- Test the self-healing concept from space modules to create a HAL that automatically isolates failures and restores critical functions without driver intervention.
- Investigate whether principles of distributed information storage from astrophysics can be applied to enhance data fault tolerance in automotive systems.
- Check how decentralized control methods from other fields could reduce dependence on a central hypervisor and increase update flexibility.
Nearby works in our archive
- How to Stop Molecular Motors from Spinning Their Wheelssimilarity 0.55
- A Quantum Cat on Guard for Computingsimilarity 0.55
- The Metal Trace That Won't Vanishsimilarity 0.54
- Quantum Networks: How Optical Switches Link Processorssimilarity 0.53
- Quantum Memory in Diamondsimilarity 0.53
- Space modules heal wounds without commandssimilarity 0.53
- Symphony of Stellar Memory: How Chaos Preserves Galactic Historysimilarity 0.53
The archive is sparse around this work: the closest review is only 0.55, against a usual 0.64. It looks like few have drilled nearby.
The QNX Hypervisor uses less than 2% of the CPU for its own needs, like a diplomat at negotiations spending only a couple of seconds per hour on translation.
In the series Knight Rider, KITT could receive updates and new capabilities over the air — decades before engineers seriously tackled OTA. Today's paper is about making that magic an everyday reality for millions of production cars.
Key numbers
- SDV market size by 2030: >650 billion $
- QNX Hypervisor overhead: <2%
- maximum HAL evaluation score: 45
- AUTOSAR Adaptive score: 37
- Apex.OS score: 43