Why is your car less reliable than it used to be?
Probably the most reliable vehicle I have had was a 1987 Toyota pick-up with a 22R 4-cylinder engine. There was not much in the way of electronics in it – just a transistorized ignition and a radio. The rest was all mechanical or electromechanical. Not much went wrong with it. All I replaced was an igniter and a set of front hubs. I traded it in at about 160,000 miles (260,000km). It was simple enough I could do most anything in the way of basic maintenance myself.
By comparison, many of today's vehicles are loaded with electronics: "power this" and "electronic that." Vehicles even have electronics subsystems in the rear view mirror. Each piece of electronics itself is fairly reliable. Let's say that each stands a 0.1% chance of failure each year after leaving the dealer's lot. The real issue is that each function may have an MCU and associated bus as well as a host of discrete parts. The number of MCUs, FPGAs, and even ASICs, can be mind-boggling alone. What used to be a dashboard with a few mechanical gauges is now host to scores of complex semiconductors. The powertrain uses more. The accessories even more.
Let's say this is a mid-complexity car with around 200 complex semiconductor-based boards. This means there is a one-in-five chance that something will fail in the vehicle each year! If left alone, after 15 years (the length I kept my Toyota), there could be three failed items. On vehicles I've purchased following the Toyota I have had this very thing happen – and not just with minor systems, but with things that affect the engine and brakes.
Devices like FPGAs present an interesting dilemma. When used appropriately by a design team, with proper supporting components, they can significantly reduce the complexity required for a function, thereby improving reliability and safety. Alternatively, when used to add frivolous complexity to a product, they can become a drag on reliability and safety.
If one looks at the issue more closely, one can see some areas regarding modern MCUs and FPGAs that need very close attention. Some areas that the hardware team has control over include the component selection and the board construction and layout. I do not know how many high-reliability designs I have seen where the MCU/FPGA RTL/HDL used both clock edges, but the designer did not account for jitter and duty cycle in the oscillator or the quality of the PLL power supply.
Another area is the reliability of the components. To save cost, industrial/commercial parts will often be used. In many cases, no derating is performed, and a full parametric test will not be undertaken. AEC-Q100 grade parts (both semiconductor and discrete) have additional manufacturing tests associated with them to help screen for potential issues due to aging caused by heat and thermal cycling. Also, many vendors package the AEC-Q100 grade parts in a more robust grade of material. Better testing can help determine if there are issues like a damaged bus transceiver garbling part of the data over a certain temperature range. Board or subsystem performance should be checked during vibration and thermal cycling as well. The bottom line is that the cost delta for these parts is not that much compared to the cost of the mechanical components the devices serve to control, monitor, and protect.
What are your experiences with cars related to this topic?
This blog originally appeared on All Programmable Planet