Originally Posted by Chris Scott
In the A320, the mother of civil digital FBW, the designers took a deliberate decision to go for well-proven COMMERCIAL chips, which were already being used in large numbers in a wide range of applications. For the SECs (spoiler-elevator computers), they chose the 80186. ...... Airbus and Sextant Avionique had the option of developing what they called "mathematically correct" micro-processors.........
If the missions of the SECs and ELACs (elevator-aileron computers) are not much more demanding in 2008 than in 1988, and replacement chips are still available, why indeed go for a more complex chip?
The 8086 was a 16-bit architecture in which all processing and all data channels were 8-bit. It means you had to do everything twice: once for the lower half and once for the upper half of the 16 bits. The 80186 had full 16-bit data paths and processing.
The option to develop a "mathematically correct" chip was there only in theory. To find out what happened when the Brits tried to do that, Google on "Viper RSRE". You'll see Brian Randell's submissions to Risks about what happened commercially with the Viper chip at
http://catless.ncl.ac.uk/Risks/10.15.html#subj4
http://catless.ncl.ac.uk/Risks/11.73.html#subj1 and you can follow the story by searching through earlier Risks issues.
The points are that (1) you have to get it right; (2) other people have to believe you have gotten it right; (3) you then have to fabricate and sell quite a few of them. Fabricating a special chip to put in a few thousand Airbi was not then a commercially viable proposition, and a risky proposition even if the entire development costs were written off (read: government subsidised). Things are somewhat different now.
There is, however, a problem with supply. It costs a lot of resources to keep chips in production over decades, and manufacturers don't do it, with processor capabilities doubling every 18 months (Moore's "Law") and SW writers making full use of it.
Parts of FAA ATC ran until recently on mainframe computers that were many decades old. The manufacturers had stopped supporting them; the FAA had to build its own fabrication labs, and then it had a hard time keeping the boffins around who knew how to fabricate and repair the processing elements after retirement age. It became a crisis on which the NTSB produced a special report.
I know one major ATC en-route system designed in the 90's (weren't they all?) based on banks of commercially-available desktop computers. For a number of years, the service provider has been stockpiling replacements by buying up all stocks of these machines wherever they surface in the world (ebay and so on). This business of keeping processors around for the projected lifetimes of aviation systems is not a trivial problem.
PBL