Flip on the TV, and a car commercial is bound to pop up shortly touting one of two technological aspects. One is center stack integration of smartphone-style applications. The other is advanced driver assistance systems (ADAS) featuring cameras, radar, and other sensors helping cars do things like stay in lanes, recognize road signs, jostle drowsy drivers, see in blind spots, and brake to avert collisions when necessary.
Historically, many semiconductor firms shied away from automotive applications, with onerous environmental requirements and high stakes liability concerns. The allure of increasing content, especially higher-value functions such as ADAS, the “connected car,” and vehicle-to-vehicle (V2V) integration, is drawing participants back in.
Electronics content in autos has been headed up for years, starting with ECUs for powertrain and vehicle dynamics control, with a CAN network interconnecting several independent units. That was followed for integration of point-function microcontrollers with individual responsibility for a sensor or actuator, again with CAN connections. Infotainment burst on the scene, with faster interconnects like FlexRay and MOST, delivering features such as mapping and audio/video content but with limited integration to critical vehicle functions.
Advances in fabrication, test and verification continue to close gaps in device reliability, even at extended temperatures. Achieving safety has meant partitioning functions, keeping interactions between verified subsystems at a minimum. As cars become more connected, and more and more software defines the experience, that option is diminishing.
The motivation for automotive manufacturers to integrate capability in SoCs is high; the cost of certifying and maintaining tens or even hundreds of subsystems in a car multiplied across models and years is out of control. Features coming into vogue look well suited for smartphone SoCs, and in fact many of the suppliers are leveraging embedded expertise but varying degrees of success in mobile – names like Freescale, Intel, NVIDIA, Renesas, and TI among them, along with Altera and Xilinx coming from the FPGA direction.
What’s wrong with mobile SoC architectures for automotive? Nothing, if you don’t mind occasional, inexplicable burps in service. Before hitting the road, improvements in architectural integrity related to functional safety are needed.
A couple nights ago on my usually dependable dual core Android phone, I had just used MLB At Bat to watch video of the Giancarlo Stanton unscheduled tooth removal, and pulled up the IMDB app to type in a name from an old movie I was watching, and … the display went dark. Pressing the power button elicited a couple beeps but no visual response. It came back after about three minutes of listening to my trade-in threats.
That kind of SoC burp in an ADAS system is unacceptable.
Did I forget to mention autonomous cars? I love control and assistance tech, but I’m a lesser-known skeptic of autonomous transportation in general. I’ve proposed the real test for any self-driving vehicle is the elementary school nearest you at 7:30am. While an autonomous car may not careen out of control, it will very likely stop in its tracks when confused by its surroundings. Spending $40M or more on a military drone for open skies is one thing; building a driverless car to venture out on open roads for $40K or less is quite another.
Never say never, but for the time being, I’m right. Suffering from recurring nightmares of sudden acceleration, executives of a major auto manufacturer walked into a crowded room in Ypsilanti last week and pronounced: “Toyota will not be developing a driverless car.” They are all in on ADAS, and are committed to collision-prevention technology in all models sold in the US in 2017.
ADAS is rapidly heading for every car, and that means we need safer SoCs designed for the role – not just variants lifted from smartphones. Arteris FlexNoC network-on-chip is ready to deliver four critical items needed for the cause.
Multicore redundancy: Borrowing a long-standing practice from avionics, achieving ISO 26262 compliance at higher ASIL levels means getting past single point faults. For resilience, dual core lock step is required, and a NoC can ease implementation.
End-to-end data protection: Resilience doesn’t stop at the processor. With ECC running throughout the NoC to cores and peripherals, both core-side and packet transport protection is available. An AMBA interconnect only supports ECC to RAM, leaving much unprotected.
Partitioning: That firewall capability we discussed last month in the Altera Arria 10 MPSoC with FlexNoC inside? It isn’t just for network traffic; it also provides a way to partition safety-critical functions from non-critical functions, on the same SoC.
Quality of service: Traffic in the NoC can be prioritized and shaped, ensuring critical elements are able to maintain service in all conditions. This includes managing duplicate NIUs, and checks for consistency and equality SoC wide.
Of note is Mobileye, a leader in vision-based ADAS SoCs and systems, switching from another technology to Arteris FlexNoC for their latest device. The EyeQ3 features four MIPS 32 1004K cores, along with four proprietary vector microcode processors for accelerated vision functions.
Altera, Freescale, Renesas, and TI are also deeply integrating FlexNoC into SoCs targeting ADAS – and there is more to come. We’ve gotten word Arteris is planning a major automotive announcement for ARM TechCon next month … stay tuned.
Being able to integrate more functions into a properly designed SoC has enormous benefits in capability and lifecycle cost for auto manufacturers and subsystem vendors like TRW. With ADAS breaking through and becoming a must-have feature soon to appear in every new car, designers grappling with safety-critical issues should be looking at network-on-chip for help.
[*=1]New details on Altera network-on-FPGA
[*=1]10 years, 100,000 miles, or <1 DPM
[*=1]Seen at DAC! Self-Driving Cars –Victory Lap or Pile-Up?
[*=1]Google Robot Cars are Coming!