16

Early on, even fairly small ROM's were quite useful. For example, you could fit a minimal upper-case only font for a terminal into something like 256 bytes. And that's a scale that a human could lay out fairly easily. Fabbing even a small mask ROM has a fairly significant initial cost, and takes quite a bit of time.

But printed circuit boards were a pretty mature technology by the time mask ROM was emerging as a viable option. And if you only needed a grid of something like 64x32 traces, that would not require a huge board. Assuming traces of 0.1" (like pins of DIP sockets and breadboards), a 256 byte "ROM PCB" would only be a bit bigger than 6". Several kilobytes could fit pretty easily in a small backplane.

So, did anybody ever just take the design of a mask ROM and print it directly as traces on a multilayer PCB, rather than in silicon?

edit to add: To be clear, I am not asking about something like a diode array where you have a circuit board full of components, with something like 1 or more components per bit. I am asking about a PCB design where the layout of the traces is the ROM. Such that for example each "stored bit" can be a trace that routes to a power pin vs disconnected or a ground, with no additional per-bit components other than the connectivity of the traces in the PCB. Obviously, I understand you can build discrete logic circuits that work as memory.

Improve this question
7
  • 5
    6" (15 cm for the literate) was in the early 70s still hard to make and rather expensive. Beside that, yes, of course have ROMs been build from single diodes. Any kind. One I remember was a key row/column to ASCII encoder ROM build from a long stretch of diodes.
    – Raffzahn
    Commented Jun 6, 2023 at 18:01
  • 6
    Wikipedia has an entire article about diode arrays used this way.
    – Michael Kohne
    Commented Jun 6, 2023 at 18:44
  • 4
    The traces are easy, but you'll need a lot of diodes.
    – Wayne Conrad
    Commented Jun 6, 2023 at 21:34
  • 6
    It is a bit unclear what exactly you are asking about. Simply printing a pattern on a PCB will not give very many bits of information, you need some kind a coding/decoding logic. The simplest as far as i know is one diode per "active" bit, either per one or per zero (your choice of logic level). In addition the PC would need decoding logic adressing the "bytes" on the card.
    – ghellquist
    Commented Jun 7, 2023 at 10:43
  • It's important to realize that mask ROM on integrated circuits does have a transistor per each bit stored.
    – jpa
    Commented Jun 7, 2023 at 16:32

4 Answers 4

Reset to default
25

The HP9100A had a ROM which used inductive coupling between adjacent layers on a 16-layer PCB to store data: http://www.hp9825.com/html/the_hp_9100_rom.html. This approach apparently reached a density of 1000bits per square inch.

Improve this answer
3
  • 4
    God that's a beautiful hack.
    – Sneftel
    Commented Jun 7, 2023 at 9:43
  • 1
    Nice, this is probably the closest to what is being asked that we'll get. The way how no current can be induced back into a disconnected row line neatly solves the ghosting problem that simply connected traces would get.
    – jpa
    Commented Jun 7, 2023 at 16:29
  • I wonder if it would be practical to make a sandwich with two driving layers, a ferrite layer, and an interchangeable sensing layer, in such a way that hitting a row wire each with 2/3 of the current needed to flip the ferrite, and then hitting the row wire with 1.5 times the current needed to flip it back, would make it possible to observe the orientation of the sensing wire over the row/column intersection.
    – supercat
    Commented Jun 8, 2023 at 17:51
20

IBM used punched mylar circuits as ROMs.

0's and 1's were encoded by cutting some traces, circuits were probed through small inductors.

http://www.righto.com/2019/11/tros-how-ibm-mainframes-stored.html

Mylar foil Assembly

Improve this answer
5
  • 1
    True, but they are related to core memory, not PCB nor diode based ROM.
    – Raffzahn
    Commented Jun 6, 2023 at 20:01
  • 1
    @Raffzahn the question doesn't ask specifically about diode-based ROM, and these are PCBs.
    – Criticizing Israel not allowed
    Commented Jun 9, 2023 at 12:45
  • @user253751 calling them PCB is quite a stretch. Also, they are not building a ROM on their own but part of a stack with many additional components. Using only a single as the OP implies would be not a great idea.
    – Raffzahn
    Commented Jun 9, 2023 at 21:13
  • 1
    @Raffzahn True, they're not much of a "board", but that's just because, with no components on it, there's no advantage to it being rigid. You could make it a board and use traces for the inductor grid, but there wouldn't be much of an advantage to doing that since you lose the easy ability to swap the cheap sheets and save very little on construction costs.
    – Perkins
    Commented Jun 9, 2023 at 23:21
  • @Raffzahn Another term for a PCB is "printed wiring board" - which criteria don't these meet?
    – Criticizing Israel not allowed
    Commented Jun 10, 2023 at 6:44
18

DEC had a bootstrap ROM module for PDP-11's with 32-words of memory: BM792 Unibus Module. Each bit position had a diode. If the diode was snipped out it flipped that bit from 1 to 0. Experienced users just snipped one lead and lifted the diode slightly. A drop of solder could be used to reset the bit.

I think there's one in the basement.

Improve this answer
1
  • 2
    I remember being annoyed that the 11/20 at the National Scientific Balloon Facility didn't have this. I toggled in the boot program enough times to have it memorized ツ
    – John Doty
    Commented Jun 7, 2023 at 16:56
8

While I am sure it has been done, keep in mind that concurrent with the beginning of the microprocessor era was also the beginning of the EPROM era. According to the linked Wikipedia article, the Intel 1702 was announced in 1971, and held 2,048 bits (256x8). There are also apparently one-time-programmable EPROMs that were regular EPROMs except no quartz window so no easy way to erase them.

So I would suggest that large (e.g., several hundred bits or more) semiconductor ROMs from discrete components were probably only a practical choice from sometime in the 1960s until the early 1970s. After that, EPROMs were the way to go.

Improve this answer
1
  • 2
    Especially in their prior from of PROMs. Also, the technology of mask programmed ROMs, that is standard cell ROMs where only the last metallisation layer defined content, were already a thing short before (E)PROM and widely used to do semi custom ROM.
    – Raffzahn
    Commented Jun 6, 2023 at 20:00

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .