In 1937 mathematician and researcher George Stibitz "liberated some relays from a scrap pile at Bell Telephone Laboratories" where he worked and took them home to experiment with a personal project. Over the weekend he assembled what is generally believed to be the first binary adder circuit. The project was built in classic tinkerer style. Along with two of the relays, he fashioned two input switches out of metal cut from a tobacco tin and used two flashlight bulbs as outputs. The entire circuit was mounted to a wooden board.
Over the next decade, propelled by Stibitz' innovations in digital design, Bell Laboratories would be a significant force in the development of computer technology, producing a series of increasingly capable computers driven by electromagnetic relays.
Bell Laboratories in the 1920s and 1930s was an incubator for the advancement of switching technology. Under the combined ownership of the Western Electric Company and the American Telegraph and Telephone Company (AT&T), Bell Laboratories provided research and development for the United States' maturing nationwide telecommunications network. The engineers who designed and maintained the system operated within a private world, working with technologies that few outside the Lab had use for, and speaking a language of their own.
To handle the routing of calls, the network used what was called the panel machine switching system. The concepts behind the panel system were largely unique to the telephone network. The system relied heavily on relays which were configured as complex digital logic circuits, although the term "digital logic" had yet to be coined.
A number of the techniques they developed map directly to the basic requirements of an electronic computer. The system had the ability to store and retrieve numbers using banks of relays. The panel system also freely converted among numbering systems in order to more efficiently handle data. A number entered into the system in the human-friendly base 10 (decimal), might be converted into base 2 (binary), base 4 or base 5 during operation, only to be converted back to decimal again to be displayed to human operators.
Through the lens of history, it can seem surprising that with so many pieces of the puzzle already in place, Bell Laboratories did not move earlier to designing and building computers. In the pragmatic, business driven environment, it appears that no one found the idea interesting enough to pursue. The prime and perhaps only true mission was to sustain and expand the telephone system, and a complex, expensive calculator didn't seem to have a place in that vision. As computing pioneer Ernest Andrews would recount years later:
"In brief, then, many of the more important computer operations and concepts were, in the 1920s and 1930s, already incorporated into the panel switch system, and, as the value of the new circuits was recognized, into other systems as well. It might be asked, therefore, why automatic digital computers were not developed simultaneously. In the period from 1925, for example, this writer heard many of his colleagues say with firm conviction that relay circuit techniques could be used to carry out arithmetical operations commonly performed with desk-type adding machines. The reasons are the traditional ones of economics and failure to see a compelling need. But more importantly, no one had taken the initiative to make a thorough study of all aspects of such a project."
It was against this backdrop that when Stibitz first brought his binary adder into the lab to demonstrate to his colleagues, they were "more amused than impressed with some visions of a binary computer industry." Nonetheless, he continued to explore the concepts of binary, electronic computation as a side project and prepared the plans for a complete and functional machine.
A barrier to getting a project funded for the building of a relay calculator was the fact that the mechanical adding machines of the day worked pretty well. Bell Labs needed quite a bit of number crunching as part of their R&D efforts, but these needs were addressed by a room full of workers who would complete the needed calculations using standard mechanical adders. (The mathematics staff are somewhat jarringly referred to as "the girls" in contemporary accounts.) The prospect of spending many thousands of dollars to build a machine consisting of hundreds of relays and other components was a tough sell when the current system appeared to be efficiently meeting the needs of the engineers.
As it happened however, there was one particular requirement that was giving the team more trouble than usual. Engineers working on filter and amplifier circuits were performing calculations on complex numbers: two-part numbers made up of both the amplitude and phase of a signal. A characteristic of these complex number calculations was that there were significantly more intermediate totals that needed to be written down and re-entered when using the adding machines. Stibitz was approached by the chief of the mathematics department regarding the problem of complex numbers, and it was quickly apparent that his design concept for a calculator was well suited to tackle the equations. With an actual solution to an actual problem identified, leadership at Bell Labs finally agreed to fund development of the first relay calculator.
The project was placed under the leadership of a veteran Bell Labs engineer, Samuel Williams, and the two men set to work refining Stibitz' concepts for the special purpose of complex number calculations. To simplify the conversion between decimal and binary, Stibitz developed a system known and used today as binary coded decimal (BCD) in which each decimal digit is represented by four binary bits. The machine used two calculator units running in parallel which provided for the simultaneous processing of the two components of the complex number. The calculator used standard and familiar telephony equipment. Workstations were made by combining two 10-key push button keyboards and a commercially available teletypewriter was used for output. The system would ultimately have three of the keyboard work stations running with a control mechanism that ensured "first come, first served" operation.
The calculator was completed in October 1939. In January 1940, test and debugging was finished and the calculator was put to work at a Bell Laboratories facility in New York City. In September of that year, the machine, then called the Complex Calculator, was the subject of a somewhat famous demonstration for the American Mathematical Society held at Dartmouth College. For the demonstration a remote link was established. Problems were entered into a terminal at Dartmouth, transmitted to the New York office where the solution was calculated, and then transmitted back to Dartmouth where the answer printed on a teletype. Attendees were even invited to submit their own problems to be solved.
The Complex Calculator would come to be known as Model I. It operated productively at Bell Labs until 1949.
When the United States entered World War II, Stibitz was loaned to the National Defense Research Committee (NRDC) where he consulted on a number of control projects including anti-aircraft and other ballistics computation while continuing to work closely with Bell Labs. During this time, he continued to develop and enhance the machines while adapting them to other uses.
With Model II, which came online in September 1943, the use of paper tape was introduced. The tape system featured five-hole codes for a machine language level programming of the device with such commands as "transfer from register B to adder, "read data" and "punch."
Stibitz described the operation of Model II years later in a 1967 Datamation article:
"It was exciting and a bit weird to watch this interpolator go about its work sans human boss: days, nights, Sundays and holidays. This was a year before Mark I was formally demonstrated, and the use of teletype tapes and readers, under the control of an impersonal back of relays, was new. At the time it seemed to us we had a highly intelligent machine - the first programmable computer. It could call for the next program step from one tape and the next data from another at exactly the right instant, and detect any extra holes worn in the tape by repeated runs."
Stibitz and Bell Labs continued developing the relay computer through the war years and beyond. The last Stibitz machine was the Model V (there were actually two Model Vs produced) that was delivered to the Ballistics Research Laboratory at Aberdeen, Maryland in 1947. The Model V had more the 9,000 relays, used paper tape to store both program and data, and had two calculator units that could operate either independently on separate problems or be combined for more intensive calculations. Bell Labs would produce one final relay computer, the Model VI, in 1950 after Stibitz had left the organization and begun independent consulting.
Stibitz, George R. as told to Mrs. Evelyn Loveday, "The Relay Computers at Bell Labs," Datamation 13, no. 4 (April 1967)0: 35-44, and no. 5 (May 1967): 48-53
Andrews, E.G., Telephone Switching and the Early Bell Laboratories Computers., Bell System Technical Journal, 42: 2. March 1963 pp 341-353.
Chapuis, Robert J, & Joel, A.E. Jr., 100 Years of Telephone Switching, Part 2., IOS Press, 2003
Ifrah, Georges (2001). The Universal History of Computing: From the Abacus to the Quantum Computer. John Wiley & Sons. ISBN 0-471-39671-0.