Obviously, the Arma Micro Computer is not a microcomputer according to modern definitions, since its processor was made from discrete components. But it's an interesting computer in many ways. First, it is an example of the aerospace computers of the 1960s, advanced systems that are now almost entirely forgotten. People think of 1960s computers as room-filling mainframes, but there was a whole separate world of cutting-edge miniaturized aerospace computers. (Taking up just 0.4 cubic feet, the Arma Micro Computer was smaller than an Apple II.) Second, the Arma Micro Computer used strange components such as transfluxors and had an unusual 22-bit serial architecture. Finally, the Arma Micro Computer evolved into a series of computers used on Navy ships and submarines, the E-2C Hawkeye airborne early warning plane, the Concorde, and even Air Force One.
The Arma Micro Computer
The Micro Computer used 22-bit words, which may seem like a strange size from the modern perspective. But there's no inherent need for a word size to be a power of 2. In particular, the Micro Computer was designed for mathematical calculations, not dealing with 8-bit characters. The word size was selected to provide enough accuracy for its navigational tasks.
Another strange aspect of the Micro Computer is that it was a serial machine, sequentially operating on one bit of a word at a time.2 This approach was often used in early machines because it substantially reduced the amount of hardware required: it only needs a 1-bit data bus and a 1-bit ALU. The downside is that a serial machine is much slower because each 22-bit word takes 22 clock cycles (plus 5 cycles of overhead). As a result, the Micro Computer executed just 36000 operations per second, despite its 1 megahertz clock speed.
The Micro Computer had a small instruction set of 19 instructions.3 It included multiply, divide, and square root, instructions that weren't implemented in early microprocessors. This illustrates how early microprocessors were a significant step backward in functionality. Moreover, the multiply, divide, and square root instructions used a separate arithmetic unit, so they could execute in parallel with other arithmetic instructions. Because the Micro Computer needed to interact with spacecraft systems, it had a focus on I/O, with 120 digital inputs or outputs, configured as needed for a particular mission.
Circuits
The Micro Computer was built from silicon transistors and diodes, using diode-transistor logic. The construction technique was somewhat unusual. The basic circuits were the flip-flop, the complementary buffer (i.e. an inverter), and the diode gate. Each basic circuit was constructed on a small wafer, .77 inches on a side.5 The photo below shows wafers for a two-transistor flip-flop and two diode gates. Each wafer had up to 16 connection tabs on the edges. These wafers are analogous to integrated circuits, but constructed from discrete components.
The wafers were mounted on printed circuit boards, with up to 22 wafers on a board. Pairs of boards were mounted back to back with polyurethane foam between the boards to form a "sandwich", which was conformally coated. The result was a module that was protected against the harsh environment of a missile or spacecraft. The computer could handle a shock of 100 g's and temperatures of 0°C to 85°C as well as 100% humidity or a vacuum.
Because the Micro Computer was a serial machine, its bits were constantly moving. For register storage such as the accumulator, it used six magnetostrictive torsional delay lines, storing a sequence of bits as physical twists that formed pulses racing through a long coil of wire.
The photo below shows the Arma Micro Computer with the case removed. If you look closely, you can see the 22 small circuit wafers mounted on each printed circuit board. The memory driver boards and delay lines are towards the back, spaced more widely than the other printed circuit boards. The cable harness underneath the boards provides the connections between boards.4
Transfluxors
One of the most unusual parts of the Micro Computer was its storage. Computers at the time typically used magnetic core memory, with each bit stored in a tiny ferrite ring, magnetized either clockwise or counterclockwise to store a 0 or 1. One drawback of standard core memory was that the process of reading a core also cleared the core, requiring data to be written back after a read.
The Micro Computer used ferrite cores, but these were "two-aperture" cores, with a larger hole and a smaller hole, as shown above. Data is written to the "major aperture" and read from the "minor aperture". Although the minor aperture switches state and is erased during a read, the major aperture retains the bit, allowing the minor aperture to be switched back to its original state. Thus, unlike regular core memory, transfluxors don't lose their data when reading.
The resulting system is called non-destructive readout (NDRO), compared to the destructive readout (DRO) of regular core memory.6 The Micro Computer used non-destructive readout memory to ensure that the program memory remained uncorrupted. In contrast, if a program is stored in regular core memory, each instruction must be written back as it is executed, creating the possibility that a transient could corrupt the software. By using transfluxors, this possibility of error is eliminated. (In either case, core memory has the convenient property that data is preserved when power is removed, since data is stored magnetically. With modern semiconductor memory, you lose data when the power goes off.)
The photo below shows a compact transfluxor-based storage module used in the Micro Computer, holding 512 words. In total, the computer could hold up to 7808 words of program memory and 256 words of data memory. It appears that transfluxors didn't live up to their promise, since most computers used regular core memory until semiconductor memory took over in the early 1970s.
Arma's history and the path to the Micro Computer
The Arma Engineering Company was founded in 1918 and built advanced military equipment.7 Its first product was a searchlight for the Navy, followed by a gyroscopic compass and analog computers for naval gun targeting. In 1939, Arma produced the Torpedo Data Computer, a remarkable electromechanical analog computer. US submarines used this computer to track target ships and automatically aim torpedos. The Torpedo Data Computer performed complex trigonometric calculations and integration to account for the motion of the target ship and the submarine. While the Torpedo Data Computer performed well, the Navy's Mark 14 torpedo had many problems—running too deep, exploding too soon, or failing to explode—making torpedoes often ineffectual even with a perfect hit.
Arma underwent major corporate changes due to World War II. Before the war, the German-owned Bosch Company built vehicle starters and aircraft magnetos in the United States. When the US entered World War II in 1941, the government was concerned that a German-controlled company was manufacturing key military hardware so the Office of Alien Property Custodian took over the Bosch plant. In 1948, the banking group that controlled Arma bought Bosch from the Office of the Alien Property Custodian, merging them into the American Bosch Arma Corporation (AMBAC).8 (Arma had earlier received the rights to gyrocompass technology from the German Anschutz company, seized by the Navy after World War I, so Arma benefitted twice from wartime government seizures.)
In the mid-1950s, Arma moved into digital computers, building an inertial guidance computer for the Atlas nuclear missile program. America's first ICBM was the Atlas missile, which became operational in 1959. The first Atlas missiles used radio guidance from the launch site to direct the missile. Since radio signals could be jammed by the enemy, this wasn't a robust solution.
The solution to missile guidance was an inertial navigation system. By using sensitive gyroscopes and accelerometers, a missile could continuously track its position and velocity without any external input, making it unjammable. A key developer of this system was Arma's Wen Tsing Chow, one of the driving forces behind digital aviation computers. He faced extreme skepticism in the 1950s for the idea of putting a computer in a missile. One general mocked him, asking "Where are you going to put the five Harvard professors you'll need to keep it running?" But computerized navigation was successful and in 1961, the Atlas missile was updated to use the Arma inertial guidance computer. It was said to be the first production airborne digital computer.9 Wen Tsing Chow also invented the programmable read-only memory (PROM), allowing missile targeting information to be programmed into a computer outside the factory.
The photo below shows the Atlas ICBM's guidance system. The Arma W-107A computer is at the top and the gyroscopes are in the middle. This computer was an 18-bit serial machine running at 143.36 kHz. It ran a hard-wired program that integrated the accelerometer information and solved equations for the crossrange error function, range error function, and gravity, making these computations every half second.10 The computer weighed 240 pounds and consumed 1000 watts. The computer contained about 36,000 components: discrete transistors, diodes, resistors, and capacitors mounted on 9.5" × 6.5" printed-circuit boards. On the ground, the computer was air-cooled to 55 °F, but there was no cooling after launch as the computer only operated for five minutes of powered flight and wouldn't overheat during that time.
The Atlas wasn't originally designed for a computerized guidance system so there wasn't room inside the missile for the computer. To get around this, a large pod was stuck on the side of the missile to hold the computer and gyroscopes, as indicated in the photo below. This doesn't look aerodynamic, but I guess it worked.
The Atlas guidance computer (left, below) consisted of three aluminum sections called "decks". The top deck held two replaceable target constant units, each providing 54 navigation constants that specified a target. The constants were stored in a stack of printed circuit boards 16" × 8" × 1.5", covered in over a thousand diodes, Wen Tsing Chow's PROM memory. A target was programmed into the stack by a rack of equipment that would selectively burn out diodes, changing the corresponding bit to a 1. (This is why programming a PROM is referred to as "burning the PROM".11) The diode matrix was later replaced with a transfluxor memory array, which had the advantage that it could be reprogrammed as necessary. The top deck also had connectors for the accelerometer inputs, the outputs, and connections for ground support equipment. The bottom deck had power connectors for 28 volts DC and 115V 400 Hz 3-phase AC. In the bottom deck, quartz delay lines were used for storage, representing bits as acoustic waves. Twelve circuit cards, each with a faceted quartz block four inches in diameter, provided a total of 32 words of storage.
Arma considered the Micro Computer the third generation of its airborne computers. The first generation was the Atlas guidance computer, constructed from germanium transistors and diodes (in the pre-silicon era). The second-generation computer moved to silicon transistors and diodes. The third-generation computers still used discrete components, but mounted on the small square wafers. The third generation also had a general-purpose architecture and programmable transfluxor memory instead of a hard-wired program.
After the Micro Computer
Arma continued to develop computers, improving the Arma Micro Computer. The Micro C computer (1965) was developed for Navy ships and submarines. Much like the original Micro, the Micro C used transfluxor storage, but increased the clock frequency to 972 kHz. The computer was much larger: 3.87 cubic feet and 150 pounds. This description states that "the machine is an outgrowth of the ARMA product line of micro computers and is logically and electrically similar to micro-computers designed for missile environments."
In mid-1966, Arma introduced the Micro D computer, built from TTL integrated circuits. Like the original Micro, this computer was serial, but the Micro D had a word length of 18 bits and ran at 1.5 MHz. It weighed 5.25 pounds and was very compact, just 0.09 ft3. Instead of transfluxors, the Micro D used regular magnetic core memory, 4K to 31K words.
The widely-used Litton LTN-51 inertial navigation system was built around the Arma Micro-D computer.12 This navigation system was designed for commercial aircraft, but was also used for military applications, ships, and NASA aircraft. Aircraft from early Concordes to Air Force One used the LTN-51 for navigation. The photo below shows a navigation unit with the Arma Micro-D computer in the lower left and the gyroscope unit on the right.
In early 1968, the Arma Portable Micro D was introduced, a 14-pound battery-powered computer also called the Celestial Data Processor. This handheld computer was designed for navigation in crewed earth orbital flight, determining orbital parameters from stadimeter and sextant measurements performed by astronauts. As far as I can tell, this computer never made it beyond the prototype stage.
Conclusions
The Arma Micro Computer is just one of the dozens of compact aerospace computers of the 1960s, a category that is mostly forgotten and ignored. Another example is the Delco MAGIC I (1961), said to be the "first complete airborne computer to have its logic functions mechanized exclusively with integrated circuits". IBM's 4 Pi series started in 1966 and was used in many systems from the F-15 to the Space Shuttle. By 1968, denser MOS/LSI chips were used in general-purpose aerospace computers such as the Rockwell MOS GP and the Texas Instruments Model 2502 LSI Computer. 13
Arma also illustrates that a company can be on the cutting edge of technology for decades and then suddenly go out of business and be forgotten. After some struggles, Arma was acquired by United Technologies in 1978 for $210 million and was then shut down in 1982. (The German Bosch corporation remains, now a large multinational known for products such as dishwashers, auto parts, and power tools.) Looking at a list of aerospace computers shows many innovative but vanished companies: Univac, Burroughs, Sperry (now all Unisys), AC Electronics (now part of Raytheon), Autonetics (acquired by Boeing), RCA (bought by GE), and TRW (acquired by Northrop Grumman).
Finally, the Micro Computer illustrates that terms such as "microcomputer" are not objective categories but are social constructs. At first, it seems obvious that the Arma Micro Computer is not a real microcomputer. If you consider a microcomputer to be a computer built around a microprocessor, that's true. (Although "microprocessor" is also not as clear as you might think.) But a microcomputer can also be defined as "A small computer that includes one or more input/output units and sufficient memory to execute instructions" (according to the IBM Dictionary of Computing, 1994)14 and the Arma Micro Computer meets that definition. The "microcomputer" is a shifting concept, changing from the 1960s to the 1990s to today.
For more, follow me on Twitter @kenshirriff or RSS for updates. I'm also on Mastodon as @[email protected]. Thanks to Daniel Plotnick for providing a great deal of information and photos. Thanks to John Hartman for obtaining an obscure conference proceedings for me.
Notes and references
-
I should mention the danger of "firsts" from a historical perspective. Historian Michael Williams advised "not to use the word 'first'" and said, "If you add enough adjectives to a description you can always claim your own favorite." (See ENIAC in Action, p7.)
The first usage of "micro-computer" that I could find is from 1956. In Isaac Asimov's short story "The Dying Night", he mentions a "micro-computer" in passing: "In recent years, it [the handheld scanner] had become the hallmark of the scientist, much as the stethoscope was that of the physician and the micro-computer that of the statistician."
Another interesting example of a "micro-computer" is the Texas Instruments Semiconductor Network Computer. This palm-sized computer is often considered the first integrated-circuit computer. It was an 11-bit serial computer running at 100 kHz, built out of RS flip-flops, NOR gates, and logic drivers. The 1961 article below described this computer as a "micro-computer", although this was a one-off use of the term, not the computer's name. This brochure describes the Semiconductor Network Computer in more detail and Semiconductor Networks are described in detail in this article. Unlike modern ICs, these integrated circuits used flying wires for internal connections rather than a deposited metal layer, making their design a dead end.
The Texas Instruments Semiconductor Network Computer. From Computers and Automation, Dec. 1961. -
Most of the information on the Arma Micro Computer in this article is from "The Arma Micro Computer for Space Applications", by E. Keonjian and J. Marx, Spaceborne Computing Engineering Conference, 1962, pages 103-116. ↩
-
The Arma Micro Computer's instruction set consisted of 19 22-bit instructions, shown below.
Instruction set of the Arma Micro Computer. Figure from "The Arma Micro Computer for Space Applications". -
This block diagram shows the structure of the Micro Computer. The accumulator register (AC) is used for all data transfers as well as addition and subtraction. The multiply-divide register is used for multiplication, division, and square roots. The product register (PR), quotient register (QR), and square root register (SR) are used by the corresponding instructions. The data buffer register (S) holds data moving in or out of storage; it is shown with two 11-bit parts.
Block diagram of the Arma Micro Computer. Figure from "The Arma Micro Computer for Space Applications".For control logic, the location counter (L) is the 13-bit program counter. For a subroutine call, the current address can be stored in the recall register (RR), which acts as a link register to hold the return address. (The RR is not shown on the diagram because it is held in memory.) Instruction decoding uses the instruction register (I), with the next instruction in the instruction buffer (B). The operand register (P) contains the 13-bit address from an instruction, while the remaining register (R) is used for I/O addressing. ↩
-
Arma's original plan was to mount circuits on ceramic wafers. Resistors would be printed onto the wafer and wiring silk-screened. (This is similar to IBM's SLT modules (1964), although IBM mounted diode and transistors as bare dies rather than components.) However, the Micro Computer ended up using epoxy-glass wafers with small, but discrete components: standard TO-46 transistors, "fly-speck" diodes, and 1/10 watt resistors. I don't see much advantage to these wafers over mounting the components directly on the printed-circuit board; maybe standardization is the benefit. ↩
-
The Micro Computer used an unusual mechanism to select a word to read or write. Most computers used a grid of selection wires; by energizing an X and a Y wire at the same time, the corresponding core was selected. The key idea of this "coincident-current" approach is that each wire has half the current necessary to flip a core, so the core with the energized X and Y wires will have enough current to flip. This puts tight constraints on the current level, since too much current will flip all the cores along the wire, but not enough current will not flip the selected core. What makes this difficult is that the properties of a core change with temperature, so either the cores need to be temperature-stabilized or the current needs to be adjusted based on the temperature.
The Micro Computer instead used a separate wire for each word, so as long as the current is large enough, the cores will flip. This approach avoids the issues with temperature sensitivity, an important concern for a computer that needs to handle the large temperature swings of a spacecraft, not an air-conditioned data center. Unfortunately, it requires much more wiring. Specifically, the large advantage of the coincident-current approach is that an N×N grid of wires lets you select N2 words. With the Micro Computer approach, N wires only select N words, so the scalability is much worse.
For more on Arma's memory systems, see patents: Memory Device, 3048828 and Multiaperture Core Memory Matrix, 3289181. ↩
-
The capitalization of Arma vs. ARMA is inconsistent. It often appears in all-caps, but both forms are used, sometimes in the same article. "Arma" is not an acronym; the name came from the names of its founders: Arthur Davis and David Mahood (source: Between Human and Machine, p54). I suspect a 1960s corporate branding effort was responsible for the use of all-caps. ↩
-
For more on the corporate history of Arma, see IRE Pulse, March 1958, p9-10. Details of corporate politics and what went wrong are here. More information on the financial ups and downs of Arma is in "Charles Perelle's Spacemanship", Fortune, January 1959, an article that focused on Charles Perelle, the president of American Bosch Arma. ↩
-
Wikipedia says that Arma's guidance computer was "the first production airborne digital computer". However, the Hughes Digitair (1958) has also been called "the first airborne digital computer in actual production." Another source says the Arma computer was the "first all-solid-state, high-reliability, space-borne digital computer." The TRADIC (Transistorized Airborne Digital Computer) (1954) was earlier, but was a prototype system, not a production system. In turn, the TRADIC is said by some to be the first fully transistorized computer, but that depends on exactly how you interpret "fully".
This is another example of how the "first" depends on the specific adjectives used. ↩
-
The information on the Arma W-107A computer is from "Atlas Inertial Guidance System: As I Remember It" by Principal Engineer John Heiderstadt. ↩
-
Chow Wen Tsing's PROM patent discusses the term "burning", explaining that it refers to burning out the diodes electrically. To widen the patent, he clarifies that "The term 'blowing out' or 'burning out' further includes any process which, by means less drastic than actual destruction of the non-linear elements, effects a change of the circuit impedance to a level which makes the particular circuit inoperative." This description prevented someone from trying to get around the patent by stating that nothing was really burning. ↩
-
Details on the LTN-51 navigation system and its uses are in this document. ↩
-
For more information on early aerospace computers, see State-of-the-art of Aerospace Digital Computers (1967), updated as Trends in Aerospace Digital Computer Design (1969). Also see the 1970 Survey of Military CPUs. Efficient partitioning for the batch-fabricated fourth generation computer (1968) discusses how "The computer industry is on the verge of an upheaval" from new hardware including LSI and fast ROMs, and describes various LSI aerospace computers. ↩
-
The "IBM Dictionary of Computing" (1994) has two definitions of "microcomputer": "(1) A digital computer whose processing unit consists of one or more microprocessors, and includes storage and input/output facilities. (2) A small computer that includes one or more input/output units and sufficient memory to execute instructions; for example a personal computer. The essential components of a microcomputer are often contained within a single enclosure." The latter definition was from an ISO/IEC draft standard for terminology so it is somewhat "official". ↩
10 comments:
Great article, wow I never knew there could be a micro that early. Is there an emulator?
"Each basic circuit was constructed on a small wafer, .77 inches on a side."
0.77 inches seems like an odd choice, until you convert to metric. In metric it's 1.9558cm, so close it suggests the dimensions were intended to be 2cm (which would be 0.787 inches). I'm starting to notice quite a few cases where the US has regressed on metric since the optimistic days of the 1960s. Consider NASA, it's not so widely known that the AGC worked in metric and only converted to imperial units for the display for the astronauts. Or, for example, there's an interesting BBC Horizon TV show from the late 1970s where they interview the Voyager designers who freely convey the dimensions of the spacecraft in metres and talk about the distances in millions of km. Somewhere along the line NASA has backtracked into only talking in US customary units, even to an international audience.
Another great example is the Sony microfloppy disk itself, which is in fact 9cm and not 3.5 inches. It's got the same number of syllables, so there's no inherent linguistic advantage in the imperial version. I wrote a recent blog on that: (the title is a bit tongue-in-cheek): https://oneweekwonder.blogspot.com/2023/12/the-humble-microfloppy-disk-vehicle-of.html
Was the Atlas guidance computer a stored-program machine, or was it, like the Polaris system, a digital differential analyzer?
> I don't see much advantage to these wafers over mounting the components directly on the printed-circuit board
This might have been a yield thing? Especially so early in the silicon era, and such high requirements to meet (eg. G-loads).
With standard modules, you can turn them out by the thousand and then burn them in, shake them up, and whatever other abuse testing you need to handle. This requires much less variety in tooling than burning in many different whole boards, and the cost of scrapping a bad module is much less than scrapping a whole PCB. Similar to the known good die programs today, I'd guess?
Hi, great article (as usual)! However, there seems to be a typo in Note 6, which says "too much current will flip all the cores along the wire, but not enough current will not flip the selected current" - Shouldn't "selected current" be "selected core"?
"But there's no inherent need for a word size to be a power of 2. " - did you mean multiple of 8? Because 22 bits is inherently a power of 2.
@David Galloway: Ken Shirriff is correct. Although values represented by a 22-bit word are a power of two, the word size is 22, which is not a power of 2.
2, 4, 8, 16, 32, 64, 128, 256 are powers of 2 (2^1, 2^2, 2^3, 2^4, 2^5, 2^6, 2^7, 2^8 respectively). 22 is divsible by 2 but it's not power of 2.
Maybe one of the first microcomputer can be the Programma 101?
https://en.wikipedia.org/wiki/Programma_101
Really interesting article! But from looking at the logic diagram, and the instruction set, it's not clear to me how it handled input and output. You mention it had 120 I/O pins, but I don't see much discussion on how they were handled and used.
I would assume they were memory mapped to an address range? But there doesn't seem to be a load/store command, though if I squint hard enough, maybe the Store and Transfer commands do this? And what is the "Halt and Transfer" command doing? Maybe it ends reading the current value (one bit at a time) and moves onto the next command? A way to shortcut calculations when you get zero early enough and want to move on quickly to the next set of work?
Hi John S,
"But there doesn't seem to be a load/store command" It uses STC (Store and Clear) and ADD, which is like a pdp-8. If the accumulator is 0, which it will be after an STC, then ADD functions as a LOAD. On a pdp-8 they're called DCA (Dump and Clear Acc) and TAD (Two's-complement ADd).
Post a Comment