I worked on the ACS project. I worked on an operating system design but I report here on plans for the physical packaging for the machine.

The ICs for the machine were to have been laid out in a plane of about 1 meter square. The IC’s were then called “MECL II” or perhaps that was a trade name later adopted for that class of IC. The IC’s were to be arranged in a square array with bonding pads all visible on one side of the plane. They were five ICs to the inch in both directions, 25 per square inch. The IC was square with 100 bonding pads on each side. These pads were connected to signal wires with “pressure bonding” which was accomplished by using physical force to mash the wire’s end onto the pad. A device (which I never saw) would do this automatically of course.

The wires that connected these pads were of two categories depending mainly on their length. For runs of less than 1.2 inches simple insulated wire was used. Its diameter was 1 mil (1/1000 of an inch). For longer runs the wire was grounded coaxial, 2 mils thick. The main subjective difference was that you could see the 2 mil wires. You could also see the others if you were nearsighted much as you can see a spider web if the light is just right.

I saw a mockup of this technology of a few square inches, which had been wired at random to approximate the expected wiring density. While it looked like mold from a normal reading distance, it felt like a nylon carpet to the finger.

In the mockup the wiring was only about 1/2 inch thick. It was claimed that the pressure bonding machine could poke down, vibrating, thru the loose mat of wires to make a new connection with little danger of breaking an existing connection.

Maintenance

This was still the era when systems were repaired in the field. Computers of that day would fail several times a year, especially ones with large component count. Perhaps systems would be shipped with the wiring machine attached, for corrections of design flaws. It could also wield a scope probe for debugging. With the scope probe option it was also possible to consider debugging the machine remotely. The human skill to debug failures was expensive enough that the debugger would likely be remote. There were about 400,000 physical test points in a meter square plane. Sampling scopes were already necessary for that class of circuit. I suppose that they were already digital in which case their signal could be transmitted to a remote diagnostician. This plan was far from settled when the project was canceled.

More ACS lore