24k #14 — Angle of Incidence, Reflection
I wrote in September about the side view mirror being knocked off my van. A couple of days and a couple of thousand dollars later, the van was as good as new. It was a good practice run. This time, the mirror was knocked off by an aggressive hit-and-run driver passing with ten gallons of pick-up truck in a five-gallon-wide median.
I loaned the van to family for a vacation and the accident happened in another state. I’m on Amtrak now on my way to retrieve the van after the insurance adjusters have finished their work.
Many of the Amtrak cars plying the northeast corridor are ‘Amfleet I’ models. These cars are about as old as I am. Their resilient nature, rebuild after rebuild, gives me hope for my own trusty Ford Transit Connect. If the sheet metal on the Transit is a little more corrugated than before, then just that much closer to Amfleet spec.
The trip has been a great opportunity to brush up on the wiring diagrams for the van and think about the Amfleet-sized history of Ford vehicle electronics. Ford introduced their first electronic engine control (EEC) units just as the first Amfleet cars were being delivered. My first encounter was in my first car, a 1983 Ford Escort. By this time, Ford was on the 4th generation of their engine control system and had switched from Motorola to Intel as microprocessor vendor of choice.
Some early microprocessors show their calculator heritage. They know how to add and subtract and render results in binary-coded decimal. If an Intel 4004 was anything like a computerized version of a mechanical desk calculator, then the Intel 8061 built for Ford was like a computerized version of a carburetor and distributor. The job of a calculator is to accept problems, perform calculations, and display results. The job of carburetor and distributor are to make small changes in the running of a factory in which something explodes every few milliseconds. How daunting is that? Hard to say, like the computational difficulty of an optical lens, drinking straw, or a mechanical watch. Carburetors typically matched the two key reagents, oxygen and hydrocarbon, roughly by mass with mechanisms that were part straw and part watch. Substantial one-time work went into their design and into a specific application, but they were relatively simple. Distributors mechanically routed electrical spark impulses to the right cylinder at the right time and were typically synchronized to the engine through gears with a separate mechanical linkage that could vary the phase of the distributor with respect to the motor.
If the purpose of the carburetor and distributor were not to coordinate the delivery of inputs to the motor, you might have said it was the other way around — the purpose of the chemical inputs was to drive the distributor as a timepiece.
Engine management computations are complex mostly because we choose to use a computer for them. Weights and springs and venturis otherwise do a fine job 95% of the time. While Ford sold around 2 million cars in 1923 and the same number a century later, their market share collapsed from 70% to 13%. In a nutshell, that’s the problem. With a quarter of a billion cars now registered in the US, the 5% of the time when mechanical gadgets don’t work harmoniously would amount to an environmental disaster. It already did! The landmark Clean Air Act was passed in no small part to fix a nationwide smog epidemic fed by cars. The fix sounded easy — make sure the gas was burned instead of being dumped into the air behind the car. Implementing the fix was less easy. The first line of defense was a new kind of catalytic reactor put in the exhaust pipe. Extra air was injected after the engine and reacted with unburned gas in the exhaust as it flowed over a bed covered with platinum and palladium. You thought computers in the 1980s were expensive? Platinum has entered the chat. The second line of defense was accurate control of the fuel/air ratio to ensure complete combustion. Instead of just weighing the reagents, the cars would now continuously titrate the mix using an oxygen sensor to determine complete combustion. This is where the computers come in.
I don’t have good numbers at the moment for the relative price of a catalytic converter and a microprocessor in 1983, but the 40 years later, the catalytic converter for my van is almost four thousand dollars from Ford. The list price for a new engine computer is around five hundred bucks. Suddenly, computers looked almost like a bargain. Ford had ideas about making them even cheaper. 8061 systems didn’t need much RAM to run the same loop forever from ROM. A simple addressing scheme treated the first page of memory, held on-chip, as a register file similar to the 6502 zero page. Another simplifi-complication was the pin-saving memory bus. The most natural way to save pin and package with memory is to multiplex the data and address bus. Ford and Intel did approximately that, except that they sometimes skipped the address portion altogether. ROM is Read-Only Memory, see, and RAM stands for something totally else. It’s Random-Access Memory. A program running in the same loop from ROM doesn’t really have random accesses. It could run sequentially from a paper tape and rewind only at the bottom of each iteration. The cost was that the memory needed its own copy of the program counter to feed the right data and a mechanism to re-sync the counter on branches.
The 8061 had another key feature. Incoming interrupts were stored in a FIFO and tagged with a timestamp from a 16-bit counter so that the timing portion of real-time could be maintained even with a slow processor. Output pulses could be scheduled in a similar way. It was this facility that made the 8061 as much chronograph as computer.
Ford extended this line of custom processors for several years before returning to Motorola, though for PowerPC. Intel continued with the Ford-specific 8061 line for other customers and applications. The Intel i8096 processor (or MCS-96) was closely related, though with a more conventional memory interface.
How does a processor with a 240-byte register file relate to our 24k challenge? It seems as unlikely as my Ford needing two mirrors in two months. A later member of the MCS-96 family, the 87C196KC, mapped 0x4000 - 0x9FFF to an external memory. The subsequent 87C196KD mapped that extra 24k to _internal_ EPROM. The angle of incidence from a Ford mirror is equal to the angle of reflection as we bounce from processors with only 24k to processors with an extra 24k.
Speaking of the 80196 and rebuilding rail vehicles, here’s an 80196 retrofit for locomotive motor control. The Indian WAG-5 locomotive that’s retrofit is almost as old as the Amtrak Amfleet cars.
This post pairs well with “Drive” / The Cars / Heartbeat City, 1984.