SAE Convergence: Improving automotive microcontroller performance
by Eric Bryant on Oct 20th 2006 at 1:35PM 
Without a doubt, modern expectations for automotive performance, safety, and comfort are pushing "intelligence" throughout the vehicle, with microcontrollers showing up in a variety of devices that draw together chassis control, driver assistance, and risk management. At the SAE Convergence electronics show, Patrick Leteinturier from Infineon talked about the trends that we might expect for these tiny semiconductor brains.
Moore's Law may dictate growth in computational power of about an order of magnitude (10x) every five years or so, but that's not a reasonable assumption for automotive devices, according to Leteinturier. The reasons are numerous - the harsh automotive environment, customer expectations for longevity (20 years), and the heat generated by increased clock speeds (where as a PC CPU might dissipate 60W, something in the neighborhood of 1W is considered acceptable for automotive microcontrollers). Expect instead something more along a doubling of power every five years, with current engine management devices going from 250 million instruction cycles per second nowadays to about 2 billion/second by 2020. Memory will grow at a similar rate from 2MB to 16MB over the same period of time. Process geometry shrinks will also occur, but also not at the same rate of other electronic devices. The problem here once again is reliability, but another factor is the heavy use of analog circuits in the automotive environment - such devices can't be shrunk in the same manner as digital circuits.
Adding to reliability will be multiple redundant processors, although how these will be packaged is yet to be determined. Techniques such as built-in self test (BIST) and built-in self repair (BISR) will improve memory integrity (ever more important as storage space grows), and new memory storage technologies such as FeRAM might replace Flash in the next decade.
This might not be the most exciting innovation, but it's the sort of steady improvement on the component level that enables technology that's a bit more obvious in its impact on our driving experience.
[Source: Infineon]
Get a WordPress.com Blog
Reader Comments (Page 1 of 1)
James Noble 2:32PM (10/20/2006)
This may be innacurate. As computers in vehicles do more they're useage of RAM and need for speed (computationally) will increase. The various levels of redundancy will be one of other techniques in software engineering that will be used. Having a computer park a vehicle or drive it down the road can take more than 2 megabytes of ram (these applications will definitely utilize redundancy).
Reply
TJ 10:42PM (10/20/2006)
Sure a standard PC may use 60W, but a laptop has nearly the same power and uses much less power. In terms of raw performance two megs and 250MHz today is pathetic. Even going up to 16megs and 2Ghz by 2020 is pretty weak by today's standards.
For comparison consider that the portable Sony PSP game machine has a pair of 333Mhz MIPS R4200s with a 160Mhz gfx chip and 32megs of memory today (on market since early 2005). While a car has the luxury of a relatively un-jarring life time of cruising on flat roads with the odd speed bump hit too hard the PSP must be designed to withstand abuse of clumsy kids.
When I sit in my carburated 20 year old Jeep with my AM/FM radio, cassette playing walkman, CB radio, little black day timer, and film camera, and book for the kids to read I've got a total of 0 MIPs of computing power.
When I sit in my new cutting edge Mercedes with my satellite radio, 60gig video iPod, cell phone, blackberry, 10 mega pixel camera, and Gameboy for the kids I've got lots of computing power made to withstand abuse. Each common simple battery operated device is more powerful than my car.
I think the real reason cars are so primitive is engineers haven't found anything useful to do with all that computing power.
Reply
lincmercguy 1:44AM (10/21/2006)
TJ,
You cannot compart microcontroller systems to PC's. That PSP or your IPOD will not last 20 years, and will not see the temperature extremes that the systems in your car will. Try putting that PSP in the freezer for three hours, then heat it up to 150 deg F. for an hour. Then repeat that over and over for about a week and see if it works. Most of the processing capability is for the graphics. Also, your IPOD doesn't have that much processing power.
If you packed that much computing power into every microcontroller system in your car, the car would cost $20k more and crash at random.
A microcontroller is an entire system on a chip, including program ROM, system RAM, timers, UARTs, parallel interfaces, external address I/O on a chip about the size of a quarter. The actual die is much, much smaller than the ship. So you have everything on a motherboard in a chip that consumes about 20mA.
Most microcontrollers today have 16kB of program ROM, 256 bytes of system RAM and run at 5-30MHz. They also only cost about $5-$10/chip. Specialized applications are available if more of any part is needed. However, increased speed means more heat and more electrical power.
Most automotive systems don't really need any more performance, they need adaptability. Your regular radio, cruise control, air bags, windshield wipers, etc wouldn't be any better with processing power. What they need is communications with the other systems to have more features like speed controlled radio volume, adaptive cruise control, rain sensitive wipers, speed sensitive airbags, etc.
Jaguar tried to pack in more processing power into the Lincoln LS. It lead to complete system crashes at random, the cars would just quit working until the battery was disconnected.
Some systems are going to need more processing power. PCM's get more and more complicated, especially in hybrids. Systems that self-park will require a bit of processing power to interpret the images used to position the car.
Moore's law is just an estimite, and it applies to the die sizes and the circuit speed. That doesn't relate directly to processing power. Just because the technology exists to make mroe powerful microcontrollers, doesn't mean it will happen.
The redunancy technology isn't in software engineering. The redunancy in microcontrollers applies at the electronic level. For instance, if a processor register stops functioning, the chip would move to a backup redundant register. This is done to prevent module failure, which is becoming a bigger and bigger issue.
As for my unique perspective, I worked my way through college as a Ford mechanic, and am now an embedded systems/broadcast engineer.
Reply