TROS: IBM mainframes stored microcode in transformers | Hacker News Reader

TROS: IBM mainframes stored microcode in transformers

righto.com

189 pointsmagnat36 comments

Just a few years later, HP did 16 layer PCBs instead of the mylar tape and achieved ROM that was very dense and fast for its time[1]; though apparently yields were quite poor.

1: http://www.hp9825.com/html/the_hp_9100_rom.html

So while we are in awe, there is some group of engineers that are going "hey grandkid, let me tell you about Mylar memory. They pushed the envelope back in the day. It would be nice if some of these articles would have names so we could go "wow Bob, that was some amazing engineering".

What’s Mylar memory? Is it magnetic core memory?

The article starts with

"This unusual storage mechanism used a stack of Mylar sheets to hold 15,360 bits, equivalent to 1920 bytes."

Yes I know it even has a picture. My question is what is it?

It's a sheet of Mylar that has had conductive traces deposited on it, forming wires. This means it can be bent to go around curves without the use of additional connectors.

The interesting thing to me about the TROS sheets is that they were initially manufactured to be the same [0] but were programmed by punching out (cutting) the traces in places to zig-zag the current path either through a transformer loop or around it. Presumably there was a machine that positioned a punch to the correct position, based on coordinates from an input punch-card deck, before descending and cutting through the traces.

[0] Note that there are two designs - one with the inputs on the left and one with the inputs on the right, to line up on the connector block to the diodes. Although it's possible it's the same design, just flipped before the punching step.

Strangely enough, there are three tape designs that have the conductors offset slightly. The stack of Mylar tapes cycles through the three designs. The purpose is to prevent wires in neighboring layers from being parallel, which would cause stray capacitance.

Got it. When I saw the squares it reminded me of magnetic core memory. I thought the squares were the punches but I can see it in this image[1].

[1] http://static.righto.com/images/ibm-360-tros/tros-diagram.jp...

Just read the article in full. Mylar is even contrasted to core memory.

No.

From the article, it appears that this memory resembles paper tape, in that holes are punched or not punched, leading to a physical kind of "rom".

I hadn't heard of this before.

Made possible in part by Richard Feynman's development of electroplated plastic.

If you want names, transformer read-only storage was originally invented by T. L. Dimond in 1951 and used by Bell Labs. At IBM Hursley Laboratory, Antony Proudman was responsible for memory development including TROS.

Oh, the Dimond ring translator.[1] What a hack.

As I've pointed out before, the early history of computing was a struggle to get enough memory. Any kind of memory. IBM had an electronic multiplier in test before WWII. Eckert, at Columbia University, built some incredible semi-programmable kludges out of IBM tabulator equipment. (He went on to be one of the designers of the UNIVAC I.) But memory remained insanely expensive, more than a million dollars a megabyte, into the 1970s, 30 years after the first all-electronic arithmetic unit.

[1] http://etler.com/docs/Crossbar/articles/30-AMATranslator.pdf

Confusingly, there were two unrelated Eckerts in early computing. Wallace Eckert was an astronomer at Columbia who figured out crazy ways to use punch card accounting machines to do scientific computation, and wrote the book "Punched Card Methods in Scientific Computation".

Presper Eckert, who was unrelated to Wallace Eckert, is the better-known co-designer of the groundbreaking ENIAC computer (along with Mauchly). They then designed the UNIVAC I, basically the first successful commercial computer.

Oops, my bad. Thought those two were the same.

I've heard about program store sheets in the Bell System early electronic computers, but I'm not sure how they compare to this. Never have found a good explanation of how they worked, just a few of my older colleagues described a stack of "memory sheets" they'd swap when it was time to change the computer's programming.

The #1 ESS telephone switch's processor used two types of memory. Ferrite sheets were used for read/write memory to store temporary data, in a manner similar to magnetic core memory, while twistor memory was used for rewritable but mostly read-only storage, to store the system programs.

https://en.wikipedia.org/wiki/Twistor_memory

https://en.wikipedia.org/wiki/Number_One_Electronic_Switchin...

My stepdad worked for AT&T back in the 70’s/80’s. He took us to his office one night and we saw an old ESS switch that had magnetic core memory. I just started programming at the time but remember it creeping me out for some reason.

We also listened in on a random phone call and called a small restaurant in France to listen to the menu of the day.

Huh. Kind of similar to rope memory in function but packaged completely differently.

The article actually mentions that and compares the two: Rope memory has a higher storage density (important for a spacecraft) while mylar TRO is easier to change the programming (hole punch a stack of sheets vs threading a lot of wire through cores)

TFA also notes that the roles of the wires in the two types of memory is reversed

Old tech is cool to me because so much of it was human scale.

One can just see the parts. Love it.

As impressive as frightening.

What is the frightening part to you?

Sincere question. I am just curious for perspective. I like old tech and am always interested in others general take on it.

The amount of ingenuity and complexity and also a bit of "wildness" factor

Reminds me of pre digital music. Same things. Humans who knew their shit and who presented raw skill, acumen.

Yeah. I get that.

What if we are somehow forced to return? Some disaster or other.

Will we still be those same quality humans?

There's a bit of that in the pre digital eras. Things were much more complex and strange. Before doing anything with electric systems you had to study sophisticated mathematics. Now you can just get along with some arithmetics.

Can anyone find the speed of this memory?

TROS had a cycle time of 625 nanoseconds with an access time of 240 nanoseconds including address decoding and delays through the sense amplifier.

My Apple 2 has 100ns RAM. I am pretty sure slightly faster machines used 80ns.

This old memory speed seems fairly respectable for the time.

GPIB, a bus system from the mid 60s, requires a response time of <100 ns to a particular signal. That's one of the reasons why it's still practically impossible to implement in software only 50 years later.

And we hit that on tech of the era too.

Damn cool.

We had discrete logic in the late 60s that achieved toggle rates of a couple hundred MHz (ECL), it was just kinda expensive and needed loads of power. It was (and afaik still is) however widely used in aerospace/military applications (high standing currents and low impedance transmission lines probably make for excellent radiation and interference immunity) and used to be common in high end high frequency test equipment (e.g. counter/timers, synthesizers). I've heard ECL was also very popular for implementing cryptographic devices due to its low noise / switching transients.

That's extremely impressive given the year and the underlying technology.

More than meets the eye

I knew it! OS/360 was a decepticon plot all along!

Reply on news.ycombinator.com