Disruptive Integrated System Architectures and Complexity - A Multi-Billion Challenge

Why going fully integrated?

The promise of cost and wiring weight reduction, tighter integration of functions, and sharing of embedded system resources (by hosting many functions on few computers) was always attractive. Design of integrated platforms and architectures was a key topic in aerospace since early 1980s, from mid-2000s railway industry works on integrated architectures, and since mid-2015 this trend and considerations started in the automotive industry.

Fully integrated architecture is a different "beast"

When we discuss disruptive advanced integrated system architectures and platforms, it is not about gateway-based architectures or domain architectures with Ethernet. They are (relatively :-) ) simple to manage while the domains still use separate domain ECUs and resources for function hosting.

Real challenges start first with fully integrated architectures which can handle hard RT, real-time and soft-time application on very few computers, and where the applications can access any sensor or actuator in the system. They are best suited for reuse across different vehicle models, adaptations to variants, upgrades and extensions. And they can really optimize wiring and ECU count.

In its extreme, such system architectures can be compared to mini-datacenters which can be virtualized to host hard RT functions, and which allow simple integration of critical functions without the need to retest the whole system and all already integrated and tested functions. But there are no such systems in the field! Not because they are not needed, but because they are hard to design/make.

If we understand the associated costs to design a robust fully integrated embedded platform for L3/4/5, it is obvious that very few such integrated platforms will emerge in the market.

A Multi-Billion Dollar/EUR Challenge

There are few key questions for an OEM believing that virtually unlimited resources enable a reliable integrated platform, which can host safety relevant applications and all other less critical functions. Yes, it will be a 9-digit exercise for automotive industry (plus few hundreds of millions USD/EUR in aerospace, at least) to build a robust and integrated embedded platform capable of hosting all relevant SW functions for autonomous driving, vehicle controls, cockpit and other less critical functions, and consisting of:

1. One or more central computers with HPC SoC, including tailored SoC designs
2. Networking backbone and interfaces (PCIe, Automotive Ethernet, MIPI)
3. Remote IO units/Zone ECUs with computing capability
4. Methdology to integrate all distributed platform functions and high-level (20-100ms) and fast low-level (fast, 0.1-20ms) control loops
5. Platform as a SEooC (Safety Element out of Context) and a standalone systems, if viable

Note: The software applications (vehicle controls, AI/NN/ML, media, cockpit/HMI, in-car entertainment) are not included in this price tag - this is a separate budget.

Few Simple Questions to Ask

Technical Manager: The Nx 1Million USD/EUR questions is - if I try to add/modify a (single) function, how much is it going to cost? Do we need to retest all other already integrated functions, or I can do it without it (incremental certifiability and V&V)? Is my platfrom scalable? Does it guarantee non-interference and system-level partitioning? Is my team working around the platform limitations for 90% of their time to integrate a simple function, or it is just integrating it with no extra "platform workaround" effort? Note: This question can be multiplied by the number of functions you integrate ...

Advanced Technology VP & Vehicle E/E Group VP: The Nx100Million USD/EUR questions is - do I have organizational structure and a platform team which has experience to build such a novel, fully integrated platform? Are right means deployed - will they be able to make right design decisions? System availability, control/data integrity? What about tolerance to faults, fault propagation prevention and platform health management? Isolation of functions? Capability to integrate hard RT, real-time and soft-time at high utilization rates (using 5% of embedded resources is no-brainer, utilizing 60-70+% is a big deal and cost reduction factor)? Scalability? Reuse? Obsolescence management? Who will test all that?

C-Level: Nx 1 Billion EUR question is - we have managed to design an optimized integrated platform for AD L3/4/5 - this is probably the most complex piece of HW/SW, we have ever built and very few companies can do something similar. If we create such an integrated platform and V&V it, can we guarantee there are no systemic faults which will create a Boeing 737MAX type of distress for the company?

This is just a fraction of issues and questions one can ask about new disruptive integrated embedded platforms designed to reduce vehicle and system life-cycle cost, and stay ahead of competition ...

With this question catalogue (and associated costs), there is the point where the music could stop also for large OEMs. The challenge will not come only from the application side in an integrated system (applications can be a solvable challenge despite complexity, via additional sensors and sensor functions, higher computing power, and more advanced algorithms) - the probability is high that the robustness challenges and bottlenecks will emerge at the integrated embedded platform level, which only hosts all those functions. This is both the gluing and the isolation point for different hosted functions, and a significant cost factor throughout the system lifecycle.

#integrated hashtag

#software hashtag

#platform hashtag

#automotive hashtag

#zonal hashtag

#future hashtag

#architecture hashtag

#embedded hashtag

#autonomous