embedded world: Baby You Can Drive My Car

embedded world is a large conference held every February in Nürnberg, Germany (about an hour by train or car from Munich). It occupies six halls in the conference center. Last time I came it snowed heavily. But when they were setting up the booth earlier in the week, it was 20°C/72°F. But it was not to last and it snowed anyway, although not much. But after balmy Barcelona (for Mobile World Congress), it was a shock to be in temperatures below freezing.

Last time I came to embedded world, a decade ago, it was largely low level with a lot of focus on development tools for embedded and on real-time operating systems (RTOSs). Now both of those areas have largely gone open-source. If there are themes this year they are:

• Automotive. Everyone is trying to position what they do as critical for advanced driver assistance systems (ADAS) and autonomous vehicles.
• Safety and security. These are closely inter-related since there is no safety if your systems can be hacked and the software altered, and obviously safety and security are key aspects of ADAS.
• System testing. Can you say ADAS? Or ISO 26262?

Cadence had a booth at the show and, while there are a couple of other things, the focus was on IP for automotive, especially using Tensilica processors for ADAS, fitting right in with one of the main themes of the show.

For instance, here is an example of lane detection. The picture to the left is the video input, driving along 880 past San Jose Airport. Of course, in a real vehicle this input would be provided by a camera, typically concealed behind the rear-view mirror.

The picture on the right shows the lanes detected. In this demo, nothing is done with the data, but in a real car typically the car will keep you in the lane and not let you drift out to the left or right unless you have signaled that you are going to change lanes. Otherwise, the car will adjust the steering wheel to keep the car centered in the lane. Coupled with intelligent cruise control that keeps the car a constant distance from the car in front, this functionality provides the sort of autonomous driving in normal freeway conditions that several high-end cars feature today.

Another interesting demo is the car door anti-trap protection. All cars today have this: if you close a window and it encounters an obstacle (such as a child’s head) the system immediately reverses and opens the window again. When I worked at VaST we had a demo of an electronic control unit (ECU) for controlling a seat in a car over CAN. I forget if we ever actually hooked it up to a real CAN and a real seat so that you could click the simulated buttons on the screen (that are typically on the car door) and see the seat move appropriately. But the anti-trap demo has a real car door with a real window.

In a real car, the software runs on an ECU, but on this demo it is running on a virtual system platform (VSP). The code is written for the ARM® Cortex®-A9 that will be in the “real” ECU when it is built. But the VSP runs it on a PC at speeds pretty close to the real hardware. Flynn Schwiegelshohn of the University of Bochum in the Ruhr created this example of “hardware in the loop” (HIL) in association with the Cadence Academic Network.

It all works as advertised. Flynn put his fist in the gap between the window and the top of the door and closed the window. Once it hit his fist the system backed the window off. The way it works is that the motor that closes and opens the window has a Hall sensor that sends a signal back on each rotation. If the rotation rate slows (or stops), then something is obstructing the window and the window needs to be reversed.

The advantage of this sort of software development is the fidelity. Once the code is working in the VSP environment, it should be possible to take that code (the actual binary, not a recompilation) and run it on the real Cortex-A9 based ECU. It is a lot easier to debug on a virtual platform on the software engineer's desk and so the whole development can be virtualized.

Automotive Ethernet is a rapidly growing standard, driven by the need for networks that are fast enough to transfer video and audio. Two years ago there was the first automotive Ethernet conference that was lightly attended. This year it was orders of magnitude bigger. On the booth Cadence was demonstrating audio over automotive Ethernet, not as demanding as video but still probably more than you could put on CAN, the previous primary communication standard in vehicles.

Automotive Ethernet is going to be a very quickly adopted standard in automotive. Cadence has an Automotive Ethernet Mac available as IP. See my earlier blog here.

We also had a demo of our Protium rapid prototyping system. This was showing a Tensilica core, but this time not a real one. The netlist of the core was loaded into the FPGAs in the prototyping system which, in principle, could then be used for debugging the processor or for debugging the software that was being written to run on it. In a nice mind-twist, the FPGA-emulated Tensilica core was actually running the pedestrian detection demo that could be seen elsewhere running on the actual core.

The biggest market for semiconductors today is the mobile market. That is not so much a sweet spot for embedded since smartphones are general-purpose devices that run any App you choose. The next biggest market is automotive. Although only 100M cars are manufactured per year, compared to billions of smartphones, there is a large semiconductor content in them. The value of the electronics in a car is forecast to rise to 40-50% of the overall cost. The transition to ADAS and autonomous driving is the factor driving this increase.

Next year embedded world will be a bit later in the year (maybe so it won't snow). It will be 14-16th March 2017, still in Nürnberg, of course. We'll be another year closer to autonomous vehicles.