Westinghouse Flight Computer

[Qbus]

Picked up a couple Westinghouse prototype computers that were part of the first-generation systems used on the F-16, No microprocessor chip, all discreet components. They have around eight cards in them. Two memory cards that are 4K by 8, a termination card, what appears to be a serial card, a dual A to D card and what I think are five cards that make up the processor. Two identical cards labeled CU that have several old Harris 0512-5B 64 X 8 PROMS and a lot of 9300 series 4 bit shift registers and two more identical cards labeled AU that have four AM934151 chips along with a bunch of additional 9300 4 bit shift registers and 9338 8 bit “deluxe” registers. Don’t know for certain yet but thing eight of them may make up the main registers for the processor. There is also a fifth card labeled “priority” but not looked too much at that yet. Interestingly I have not found any evidence of a clock yet? There is a crystal on the I/O card but think that’s just for serial port.

Have done a little with primitive systems before, but this may be the oldest system so far. Somewhat surprised by the use of static RAM and not core to retain the program, program code must have been stored somewhere external and loaded somehow or just hard coded into the structure of the processor.

Did quick inventory of all the devices on the AU and CU cards and there is nothing like a 1802 or CD4004 to be found unless its lurking on the priority card so for now I am sticking with the idea of discrete components.

What I need now is some information of how discreet processors were built back in the days before the microprocessor. Any help in finding sites with that sort of information would be really helpful.

 

[Chuck(G)]

all discreet components

Well, sure, because it's a high-security item. All discussions must be discreet.

As far as architecture not using MPU circuits, just discrete MSI, examine any early 1970s/late 1960s minicomputer. Plenty of information on those on bitsavers.

 

[durgadas311]

I spent a lot of time dissecting Wang programmable calculators (and related machines) from the 1969-1974 era. The "CPU" is generally built using multiplexers/demultiplexers, register arrays, and some sort of ALU. Some models also used serial data paths, to reduce contact counts (and simplify the ALU). These were driven by microcode stored in a "ROM" - although not conventional semiconductor ROM in the early models. They used a "wire weave" ROM, where each wire represented a word and it was woven in and out of "transformers" that pick up the pulses sent through the wire and thus "read" the code. I'm sure other methods existed, too. So, you may have to look closer to find out what it was.

The microcode could be fairly wide (the Wangs used 42/43 bits in each microcode word). It really depends on the hardware. Minimally, you need enough bits to form two source, one destination, and ALU op code. Plus some sort of execution control (branching, etc). The Wang microcode was very simple for execution, each instruction contained the address of the next instruction to execute (there was no (auto-incrementing) PC) with some bits that were dependent on various status bits (conditional branching). It required very precise placement of instructions, though, and I imagine whatever they used to "compile" it was fairly sophisticated. Anyway, things won't look anything like modern computers.

Regarding core memory vs. semiconductor - from what I've heard, core memory is temperamental and even though it does retain contents when powered off you can't guarantee it won't be corrupted during power off/on (unless you go to great lengths to ensure clean power cycles). It also is probably not tolerant of the various forces (and vibrations) exerted on a fighter jet. Although, we can't be certain whether these prototypes actually bear any resemblance to the actual hardware delivered on the first F-16s.

 

[Qbus]

Always had fun with this stuff, it’s like being in a room in the dark with a blindfold. So far I know a couple things. First is that it’s an eight bit data bus with memory built out to 8K, the CU cards have seven Harris 0512 64 X 8 PROMS with all seven sharing the same address bus but the outputs are individual lines not related to the other devices. The CE lines are all tied together and tied to VCC+

I am assuming that the CU cards are instruction decoders and instructions are all on the common data bus, bits 0 to 5 are decoded and some function carried out depending on that word. That address bus common to all the devices is only six bits wide, maybe bits six and seven are the only two data bits? Or maybe six or seven are used to tell the system if it’s an instruction or data word like in the NOVA world where one bit is used to tell the system what’s going on?

The AU card has other weird stuff going on, there is one group of eight DM9338 8 bit multiple port registers all tied to the same clock, along with two sets of four DM 9338 registers with each group having its own shared clock that appears independent of the others.

Figure this must be the register part of the processor but why two separate four bit registers and only one eight bit? Maybe there was an intention to split one register so it can be used in different ways?

And there are four AM934151G 24 pin things, is this the ALU? Have not been able to determine what they are yet.

And on top of everything else now I got to worry about Homonyms, Sorry but have always had an affliction where part of my brain does not work correctly.

 

[Chuck(G)]

Could be a 74181 - see PDF p 114 here U6N934151X.
The AMD 9xxx series is a licensed version of the original Fairchild.

 

[Qbus]

Yes! I now think the mysterious AM934151G 24 pin things are the Fairchild 9341 ALU, the original Fairchild device that’s equal to a 74181. Will need to do some checks to confirm like seeing if carry moves to the next stage and things like that but that’s what I am thinking right now. Found a 1971 Fairchild product catalog that real useful, provides direct access to what all those nine thousand family of devices are.

 

[Chuck(G)]

It's not unusual to see the 74182 carry look-ahead generator along with the 181, so you may want to check that one out.

 

[Eudimorphodon]

The F16 had several computers, which component do you think you have? Here's a StackExchange thread about the M362F-2 fire control computer that sounds like it might be vaguely similar to what you're looking at? The notable difference there is the FCC did supposedly have core memory, but if this is a bench-test unit maybe they cooked up a more compact replacement?

 

[durgadas311]

... the CU cards have seven Harris 0512 64 X 8 PROMS with all seven sharing the same address bus but the outputs are individual lines not related to the other devices. The CE lines are all tied together and tied to VCC+

This does sound like microcode, with the instruction width of (maximum) 56 bits. 64 words is not very much program, but we don't know what this device is tasked with doing. I was amazed at how few instructions it took to accomplish things on the Wang machines.

...
The AU card has other weird stuff going on, there is one group of eight DM9338 8 bit multiple port registers all tied to the same clock, along with two sets of four DM 9338 registers with each group having its own shared clock that appears independent of the others.
...

There was a Wang design that used the same register chips, I think. Those are handy because you can simultaneously read two different registers and write a third, making them ideal for connecting to an ALU (Rf=Ra+Rb) and performing an operation in a single cycle. The number of 9338 chips indicates the width (bits) of the register array (although it is also possible they expanded vertically to go beyond 8 registers).

 

[Qbus]

Couple pictures, think this was used in development of the AN/AGP-66 Radar and although pure speculation think it may have intrepid IFF data and worked with a character generator to display that data on the radar screen. That along with cursor display and the like. Have seen Data General Nova and Rolm systems that do the same thing on ground radar. It has some of the same connectors as the Rolm 1602 but its half the size. Don’t know how much further I am going to peruse this project because identifying things like microcode and the like can be hairy. Suppose if the data and address busses are brought out can conceivably build a head with direct access to the busses and things like reset and maybe step but also have a bunch of other stuff going on. Also picked up a ton of other AGP-66 parts including the production computer that’s way different looking and think a lot of that stuff may end up for sale at Hamvention in a couple months. Just that these little white prototype systems are cool boxes and have lots of other weird aero space hardware laying around so at least one of these boxes may end up being a conversation item.