Friday, April 11, 2014

Happy 50th System/360! Pt.5: Anatomy of an SLT Card

IMGP4196

When IBM introduced System/360 they also announced an entirely new circuit technology called Solid Logic Technology or SLT. It was spawned out of the desire by IBM to make ever-more miniature parts to put in their computers. Integrated circuits was what everyone seemed to be looking forward to in the early 60s and IBM did a good amount of research into development of ECL logic being implemented on a single chip. This research proved successful, but it wouldn't see use until the 70s. Due to the reliability of producing small silicon transistors individually on a large substrate and cutting them out, IBM decided to use a "hybrid circuit" technology for the System/360.

IBM developed individual, minute planar silicon transistors and diodes with a glass backing that they would then affix to a 1/2 inch square ceramic chip. The ceramic chip had 12 pins on it with printed metal traces that made connections between the pins and the transistors and diodes. Resistors were constructed by depositing carbon ink between two traces and trimming them to the right value. These little ceramic chips would be coated with a silicone gel, then covered with an aluminum cap and potted underneath with rubber. This is the hybrid package. Most anyone would call it a silicon chip, but it's not an integrated circuit as we would think of today. Each chip consists of anywhere from 0 to 4 transistors or 0 to 8 diodes and a number of resistors. They are built so that the individual components can be isolated and tested.

SLT transistor diagram
A diagram of the glass-backed transistors and dual-diode devices. They are flip-chip modules resting on copper balls to make contact with the ceramic substrate.From the SLT Designer's Handbook

S360_SLT2S360_SLT4
SLT was actually quite an improvement in miniaturization. First a 'chip' with some transistors or diodes and a hand for reference. Second, I've heard that "this is a time where 1000 transistors overflow a thimble"

SLT chip bare
A bare SLT chip shows the mounted transistors and diodes as well as trimmed carbon resistors.

These chips would then be assembled on fiberglass boards to form a logic function. The amount of gates on a single card are equivalent to that of a 7400 series logic chip. It is not very dense. IBM's 360 and Early 370 Systems mentions that when the system was first announced, the low-density logic was criticized as being behind the times. It absolutely was, but when the reliability tests came back they found it was much more reliable than what was the cutting edge, what IBM had rejected initially for this exact reason.

Looking at how IBM made these cards, it's hard to see how they could fail. The fiberglass backing is fireproof and resistant to moisture. The aluminum covering the chip won't corrode. And there are multiple layers of sealant on the transistors. In order to reduce issues caused by bad spring contacts, IBM put the female connector on the card. The backplane it plugs into is just a grid of gold-plated pins. If a spring-loaded contact goes bad, it is part of the card- so replacing the card fixes the problem, better than having to diagnose and replace a broken card slot.

Also, every single contact is gold-plated. The pins on the backplane are and the Beryllium-Copper spring contacts on the card have a huge pad of gold on the end of each one:
IMGP4198

The pin spacing on all of these boards is 1/8 inch. Numbering of the socket pins is based on the backplane, so some weird pin numbers are found:
SLT 1-6 PAC

There were Double-height boards as well as double-width boards including a double-height double-width board that held 24 'chips.'

These cards were mounted on backplanes about the size of a standard sheet of printer paper. They contained gold-plated pins and a plastic mold that the cards fit into, preventing misplacement. The backplanes were multi-layered with power distribution and some signal lines, vertical paths on the back, horizontal paths on the front with the cards. Additional connections were made using wire-wrap. These backplanes would then be assembled into a "gate," possibly 4 by 5 planes in dimension. These were all connected by laminated ribbon cable that ran through channels next to the cages built around the backplanes.

slide24
A great example of the whole SLT gate in action. The backplanes, ribbon cable and card cages can all be seen inside this model 67 at Newcastle University.

IBM360_SCP_card-cage-6IBM360_SCP_flip-chip-rows-3
A close-up view of a backplane, shown alone in a diagnostics box. Note the metal bar that acts as a support and registration pin for the card.

I own a number of SLT cards that I bought off of ebay a while ago. They have been a source of inspiration for producing my own SLT cards, but also serve as some great, fun display pieces. Working with a number of different documents about SLT from Bitsavers, I have managed to figure out what different circuit types are on the cards that I have. What the cards themselves do are a little beyond what I have been able to figure out. Knowing the chip circuits and chasing the traces on the board would however tell a lot about them.

Most of the boards I have consist of three types of logic chips: 361493, 361494 and 361495. In order they are an and-or-inverter, a direct coupled inverter and an and-or extender. This leads me to believe that the cards I have are bus drivers and gates. They are all low-speed, so maybe they are from a peripheral device or smaller model system.
367493circuit361494circuit361495circuit
Schematics of the individual modules, in ascending order.

A look at all the boards I have:
IMGP4193IMGP4191
IMGP4188IMGP4186
IMGP4185IMGP4184
IMGP4183IMGP4181
Last, a view of one card, the one pictured at the top with its connector loose. One of the metal tabs that holds the plastic molding on the connector was missing allowing it to slide off relatively easy. I decided this would be a good board to uncap an SLT module on, as it was no longer mint. It gives a good view of some things you can't see from the documentation.
IMGP4182
It is a duplicate of the top right card in the above set of photos.

IMGP4207
With plastic flag and end connector removed, the shape of the board and its traces can be seen better. Silicone gel has been removed from the chip to better display the components.

IMGP4203
A rear view. In both of the above photos it is easy to see that the connector is a part of the board. The plastic connector only serves to guide the pins on the backplane to the pads on the connectors. But just look at all that gold!

The connector seen on these cards had an exceptional lifespan for IBM. Even the 1/8 inch spaced boards and backplane technology did. IBM was installing 4-wide boards in this style as late as the 80s. Although, by then they had been using more standard types of chips, called Monolithic System Technology, or MST. Still packaged in aluminum cans, but with more pins and one little chip of silicon. SLT also got a trip to the moon. Or at least helped get people there. A hardened type of SLT was used int the Launch Vehicle Digital Computer of the Saturn V rocket.

I have learned a lot of this information in a quest to recreate SLT cards. Initially it doesn't seem that hard. The logic can be made using surface-mount components (if you don't mind not having them in fancy little cans). Also, in theory the connector isn't that hard, pins on the board, a receptacle on the card. But it's not a standard pin-header. It's a 1/8 inch spaced connector that is specially molded to fit a particular socket. I have made CAD drawings of an approximation of the plastic molding, but that still leaves contacts to make. I had for a while toyed with making SMS cards because if you also used surface-mount components you would have enough room for capacitors and resistors to get the right response from modern transistors and also never have to drill the board because they're one-sided. But then, you need a 1/8 inch card-edge connector with wire-wrap pins. And they're hard to find...

I consider replicating connectors to be the major limiting factor in reproducing plug-compatible SLT and SMS cards.

Thursday, April 10, 2014

Happy 50th System/360! Pt. 4: Design

When I was thirteen I went to the science museum in Boston where they had a display about computers. Contained in that display was the front panel from a System/360 model 30. I was quite fascinated by the design of the thing. This is the point in time I can pinpoint as the beginning of my fascination with the System/360. The console seemed quite well laid out, with well divided sections broken up like a Mondrian painting. All the buttons, switches and knobs are clear, stand on their own and are easy to see and read. And who could resist seeing the almost iconic emergency power off button in the top right corner? Many of these systems were actually powered down for the last time using those buttons, because why not press (or pull) the big red button you never got to use? It certainly seems that IBM pulled out a lot of stops when designing the look of the main system. I have heard someone describe the consoles of these machines as conservative, partly because they just use simple lamps reading out binary digits. But one first has to take a look at the climate of the office and design looks of the 50s and 60s to see why the console and the system as a whole really just worked for what IBM was trying to do.

IBM has a wonderful web page about industrial design at IBM that contains a video showing the beginning of the Standard Modular System (SMS); the first standardized solid-state packaging system for IBM. SMS defined not just the way that logic boards were constructed, but also how they were arranged into complete systems. Cabinets were standardized along with the components and cables that made them up. But aside from the original objective of SMS, to reduce the number of parts IBM needed to produce to service computers, it also had a side effect of being one of the first systems to use industrial design concepts. IBM products leading up to this time looked like war-era cars - drab and boring - the cases on them just there to funnel the cool air over the components.

S360M20_1964
Old, bulky card processing equipment in back, System/360/20 in front.

While those designs worked, they looked out of place due to modern designers like Marcel Breuer, Charles and Ray Eames, and Eero Saarinen. Offices and office furniture were becoming the fashionable designs themselves, rather than merely sporting fashionable designs. Simplicity and new materials were cool, and filigree was out. Mondrian themes, mimicking Piet Mondrian's artwork, were popular. With SMS IBM made the change and released their best selling computer, the IBM 1401. The 1401 had the double-threat of having a good, simplistic design and redesigned peripherals, including a super fast and good quality printer, the 1403. Due to the amount of effort required to design the processors, many of the peripherals present on the 1401 were offered with slight modifications as part of the System/360.

The 1401 fit in with these modern offices. Companies who were proud of and wanted to show off their new computers no longer had to be embarrassed that their computer looked outdated, even if it wasn't.

IBM1401_729Tape_1407Console_COLORToronto-KingSt-Datacenter-1
Photos from Ed Thelen's 1401 site

So when one looks at the console of the System/360 and says that it looks "conservative" they're right in a technical sense, but they're wrong in the design sense. It had to look professional and fit in with the environment of a large (or small) business, but also look a bit flashy so as to not look out of place.

360genrl2_color

The cabinets of the system itself, bright colors on the doors, black on the sides and tops, look like Eames shelves.

S360_M75

The decision to not make a separate front panel, but hang it on the side of the cabinet, had the positive side-effect of requiring an almost free-floating desk with a built-in console typewriter to be used. This setup makes the front of the computer look quite light and open. The use of bright colors makes them more inviting and eye-catching, but even in black-and-white the computers have amazingly sharp lines and light shades. There are so many great photographs of these computers because they just look so good to begin with.

360mod40
A 360/40 just begging for you to IPL it.

IBM's biggest rival at the time, at least in supercomputing, was Control Data Corporation who had Seymour Cray as their main architect. CDC machines and later Cray Research computers had some quite interesting designs to them, frequently looking like they came out of a sci-fi film. They sold well, not as much as IBM machines, but that was mostly because they were expensive supercomputers. When asked about why the machines had such designs, Cray remarked that if someone is paying a good amount of money for a computer, the computer should look like it is worth it. IBM may not have been able to unseat CDC in computing power, but they did understand this concept - that shows well in the System/360.

us__en_us__ibm100__system_360__360_family_circle__800x638



us__en_us__ibm100__system_360__watson_at_360__620x350
It was the man pictured, Tom Watson Jr. who ushered in the era of industrial design at IBM when he became CEO in 1956. Here he is pictured with a model 20.
From the IBM 100 site

Wednesday, April 9, 2014

Happy 50th System/360! Pt.3 Isn't it Fabulous?

tumblr_mk8e4hm1nx1r49q4co1_1280

This is a bit of a silly post about all the fabulous "action shots" that were produced with these computers as a primer for a discussion about the design of the 360 series. A lot of these were generated, by IBM and also bored operators. These are some of my favorites.

360mod65a
"I just need a little program, like this high."

ibm360_07

image 55
These three are ready to get things done! (it helps if the computer is on though)

Tuesday, April 8, 2014

Happy 50th, System/360! Pt.2: The Family

ibm360-1
An early promotional photo showing, across the bottom, the then models 30, 40, 50, 60 and 70 processors

For the second part of my anniversary series for the System/360, I want to introduce the members of the computing family; the processors that made up the backbone of the series. In previous years, these would have been completely separate devices or even series, but with System 360 all of these processors had a single unifying theme, or architecture. I'll start off with those that were announced in 1964 and finish with the models that were introduced later.



A quick note about IBM part numbers in this era:
Most every processor at this point had some kind of 4-digit number to designate it, such as 7090, 1401, 1620. Upgrades would be in a similar range; 7094, 1440. And lastly, peripherals would be interspersed according to what line they were offered with; 1403, 1402, 7330. With a whole new series like 360, the "series code" was 2000 and the processors were the base-most elements; model 30=2030 and so on. Other peripherals were given different different prefixes, such as 24XX contained all the tape drives, 25XX was card handling, etc. It was a good way of keeping track of things, but they didn't always stick to it anyway. Just something to keep in mind when reading further.


360genrl_colormod30_Sys360
The Model 30

The IBM 2030 in IBM series code, or model 30, was the initial baby of the system. It was the lowest performance machine and looked quite unlike many of the other computers with its bank of knobs for data entry and backlit lucite indicator panel. It is my favorite, partly because of these features. This model was aimed more at the commercial industries for payroll and related tasks. All the computers in the series are 32 bit, but the model 30 has only an 8-bit path to memory, requiring 4 memory cycles to read or write a word to memory. One of the methods that IBM used to keep the lower-cost models cheap was to implement microcode, a read only storage that interprets the instructions to internal signalling, rather than using dedicated circuitry. This was slower, but also cheaper and smaller. The model 30 was the only model to use what was called Card Capacitor Read Only Storage, or CCROS. CCROS was unique because the data contained in it was encoded in Mylar punched cards that had metal patterns printed on them. This allowed the computer to be easily modified if so desired, but was a massive headache to get working. People who switched from an earlier 1401 system often complained that the System/360 was more unreliable than the 1401, most likely due to early issues with this device.


T373925A360genrl2_color
The Model 40

The IBM 2040 was the next step up, with a faster microcode and a wider, 16-bit path to memory, but with an 8-bit arithmetic unit. It was also the first model in the series to use Transformer Read Only Storage, or TROS, for microcode. TROS was similar to "rope" memory in that bits were encoded by whether or not a drive wire passed through a sensing ring. It was faster and more reliable than CCROS. The Model 40 also incorporated some innovations on its front panel that were present on higher-end models. It used individual, front mounted lamps and had a unique way of multiplexing bays of lamps for busses and having the individual bits labelled properly. A knob on the side of the panel would rotate a bank of labels behind a small window that would also operate a switch, changing the readout on the lamps. This model was the most popular, showing up in many universities and businesses.


S360_M50IBMSystem360_Mod50
The Model 50

The IBM 2050 was the first in a series of computers to replace what was considered to be the scientific processing line at IBM, the 7000 series. Unlike the model 30 and 40 it had a 32-bit arithmetic unit and a microcode based on a balanced capacitor store, similar but more reliable than CCROS.


360mod65System_360
The Model 60 (actually the Model 65)

The IBM 2065 was a real workhorse. It used the same microcode as the model 50, but had a 60 bit wide arithmetic unit with an auxiliary 8 bit unit for floating point operations. It also stored its registers in flip flops rather than some variant of core memory, which made it blazing fast (at least comparatively). It also used a different, faster type of main memory.

By this time, the computers are actually getting quite large. The model 30 and 40 pretty much had the same size case (about two refrigerators worth) But as can be seen from the image at the top of the page, the higher models started having compartments jutting out in all directions to house the increasing circuitry (in production the model 40 did not have the rear compartment it has in the image). The cabinets were even taller than those of the first two models. And to think that the computers still needed an additional module or two to talk to peripherals... no wonder people wanted to free up floor space when they were decommissioned.


S360_M75us__en_us__ibm100__apollo__rtcc__800x820
The Model 70 (actually the Model 75)

The IBM 2075 is actually quite an important computer. It was the only computer of the original line-up to not use microcode, but faster hard-wired logic. Other than that factor, it is essentially a beefed-up model 65. What makes this an important computer is that it had a higher addressable memory space and was essentially the supercomputer of the series at the time it was released. But overall, it is important because it was a cluster of five of these computers at NASA's Real Time Computing Center that ran all of the Apollo missions. This is my second favorite computer of the series.


Later Releases

S360_M20_colormodel22S360_1401
Models 20, 22 and 25

These models were later add-ons to the low end to replace unit-record (card processing) equipment. The model 20 was designed in Europe and was not entirely compatible with the larger systems. The models 22 and 25 were small systems that resembled the model 30 and (I think) used TROS for microcode.


S360_M85_ColorS360Mod85_Console
The Model 85

The model 85 was a very unique computer. It did not have a front panel like the other models of the series. Not only was the panel sitting on its own as a desk, but it had a video screen that went along with it. There are very few lights on the front panel, because it was accompanied by a twin microfilm module that provided document reference and also data readout, using the microfilm as a mask and labels. It was hardwired and used a fast thin-film memory for a lower portion of memory. It wasn't very popular and not many were made. It's hard to even find good pictures of it.


The Model 44

The model 44 was essentially a model 40, but with hardwired logic and circuitry for real-time applications. It was intended for scientific purposes but was not very popular.


ibmsuper
Models 91, 95 and 195

These models were the last of the 360 series. They were the highest performance, with special high-speed memories and all hardwired logic. The 95 was a 91 with thin film memory and only two were ever made, both for NASA. The model 195 was almost a System/370 and a model made with 370 circuitry was available as part of that line.