Early Digital Computers at Bell Telephone Laboratorieslopez/articulos/00948904.pdf · development arm, and Western Electric was AT&T’s manufacturing division. Western Electric and

It is my intent to convey the flavor of comput-er development at Bell Telephone Laboratories(Bell Labs) during the 20-some years followingBell Labs’ entry into the digital computer era.In doing so, I describe various problems thatBell Labs was addressing and characteristics ofthe machines developed as solutions. Duringthis period, Bell Labs was AT&T’s research anddevelopment arm, and Western Electric wasAT&T’s manufacturing division. WesternElectric and AT&T each owned 50 percent ofBell Labs, and AT&T owned 100 percent ofWestern Electric. The combination of AT&T,Western Electric, Bell Labs, and the telephonecompanies owned or controlled by AT&T wasfrequently referred to as the Bell System.

In what follows I will not attempt to dis-

tinguish between the terms calculator andcomputer. For example, one could take theposition that the Model I was a calculator andnot a computer because of its lack of a storedprogram and other deficiencies depending onone’s exact definition of computer. Also, I willnot discuss Bell Labs’ telephone call-switchingmachines.

Model IBell Labs’ entry into the digital computer era

occurred in 1937 when a research mathemati-cian, George R. Stibitz, noticed similaritiesbetween circuit paths through electromagnet-ic relays (used by telephone companies forswitching telephone calls) and the mathemati-cal binary notation for numbers. To explorethis thought over a weekend, Stibitz used somerelays to build a binary adder. This adder wouldgive, as output, the binary digits for the sum oftwo one-digit binary numbers that were inputto the device. He called it the Model K, the Kreferring to his kitchen where he constructedit. Figure 1 shows a replica of the Model K thatStibitz, shown in Figure 2, constructed in 1991for George Keremedjiev, director of theAmerican Computer Museum. When Stibitzdemonstrated the device to his managementthe following week, they were unimpressedthat relays could perform binary arithmetic.

However, in 1938, Stibitz was asked todesign an electromechanical relay digitalcomputer that could perform arithmetical cal-culations using complex numbers. Samuel B.Williams, a telephone system design engineer,was assigned to supervise its engineering andmanufacturing. What had happened was thatStibitz’s management had become aware thatBell Labs’ computational staff was in an over-load situation and felt that a Stibitz-designedrelay machine could help alleviate the over-

22 IEEE Annals of the History of Computing 1058-6180/01/$10.00 © 2001 IEEE

Early Digital Computers at BellTelephone LaboratoriesM.M. Irvine

This article relates highlights from the digital computer developmentactivities at Bell Telephone Laboratories for roughly the period1937–1958. The history begins with a researcher using relays to builda binary adder on his home kitchen table, continues with relaycomputers designed for military use, and culminates with computersdeveloped after Bell Labs invented the transistor.

Figure 1. G.R. Stibitz’s Model K binary adder.(Courtesy of the American Computer Museum.)

load problem. At that time Bell Labs was usingpeople with commercial mechanical calcula-tors to solve problems, many involving com-plex numbers. Complex numbers have theform a + ib where a and b are real numbersand i is the square root of minus 1. Complexnumber calculations were used extensively indesigning the many electrical filters thatmade transcontinental telephony possible atthat time.

The machine Stibitz designed was initiallyknown as the Complex Number Calculator(CNC) and later as the Model I. Stibitz fre-quently referred to it as the ComplexCalculator; others at Bell Labs referred to it asthe Complex Number Computer. It becamefully operational in early 1940 and was in con-stant use for about nine years.

The Model I used approximately 460 relays.It had three teletype terminals. (A teletype is akind of typewriter device that was frequentlyused for sending messages over telephone andtelegraph lines.) It had the capability to readand record on punched paper tape. To savecomputational time, the Model I had two par-allel calculating units, which allowed themachine to simultaneously perform calcula-tions on both parts of a complex number.

The Model I was the first digital computer touse multiple terminals. However, service wason a first come, first serve basis and the Model Icould work on only the problem from a singleterminal at a time. When that problem was fin-ished, the machine immediately switched tothe next problem in the queue.

The Model I was the first digital computer todemonstrate use from a remote location. Thefirst occurrence took place at the September1940 meeting of the American MathematicalSociety at Dartmouth College in Hanover, NewHampshire. After Stibitz made his presentationexplaining the details of his new machine,audience members were invited to submitproblems to the computer via a teletype termi-nal in the room. This terminal was linked by atelegraph line to the computer, located at BellLabs in New York City. Both John Mauchly(one of the ENIAC designers) and NorbertWiener (a pioneer in cybernetics and coiner ofthe word) spent a great deal of time at the ter-minal that day testing the Model I’s capabili-ties. It was able to give answers to problems inabout a minute.

The total cost of the Model I was $20,000,which at that time was considered an extreme-ly large expenditure (Stibitz called it “an astro-nomical sum”). Because of this high cost,Stibitz was unsuccessful in convincing his man-

agement to fund more advanced features hehad designed for his machine.

Model IIWith the advent of World War II, Bell Labs

loaned Stibitz to the NDRC (National DefenseResearch Committee) to conduct studies insupport of the war effort. One of the problemshe worked on was how to develop a cost-effective method for testing designs for the M-9 analog antiaircraft gun director that Bell Labswas developing. A gun director used position-al information, usually obtained from opticalor radar sources, to compute the future posi-tion of an attacking aircraft. This informationwas then used to set the timing fuse of theshell and to point the gun to the proper aim-ing position for firing so that the shell explod-ed on or in close proximity to the aircraft. See

July–September 2001 23

Figure 2. Inventor George R. Stibitz in the early1940s. (IEEE Spectrum, Nov. 1974, © 1974 IEEE.)

Editor’s NoteThree prior articles focused on the early efforts at Bell

Laboratories:

D. La Porte and G.R. Stibitz,“Eloge: E.G. Andrews, 1898–1980,” Annalsof the History of Computing, vol. 4, no. 1, Jan. 1982, pp. 4–5.

E.G. Andrews and H.W. Bode, “Use of the Relay Digital Calculator,”Annals of the History of Computing, vol. 4, no. 1, Jan. 1982, pp. 5–13.

E.G. Andrews, “Telephone Switching and the Early Bell LaboratoriesComputers,” Annals of the History of Computing, vol. 4, no. 1, Jan.1982, pp. 13–19.

the sidebar, “The Gypsy Computer,” for addi-tional M-9 information.

During the development process, Bell Labsfound itself requiring test equipment that didnot exist. This equipment would move the input

knobs of the M-9, exactly as a person wouldmove them, in tracking an attacking aircraft andcheck to determine if the signals issued by theM-9 would keep the gun accurately aimed.

Stibitz designed a dynamic tester for the M-

24 IEEE Annals of the History of Computing

Early Digital Computers

The Gypsy ComputerThe computer used to conduct simulated

flights of the B-68 under control of the XMH-3TRADIC (transistorized airborne digital comput-er), although analog in nature, was a trail-blazing machine, nicknamed Gypsy.

“Gypsy’s” story began in 1940 when D.B.Parkinson and C.A. Lovell invented a pioneeringelectrical analog computer to be used for control-ling antiaircraft guns.1 It used shaped wire-woundpotentiometers and vacuum-tube amplifiers toperform standard arithmetic operations.

This led directly to the development of theM-9 gun director to control the US Army’sheavy antiaircraft guns. This is the gun directorfor which George Stibitz designed the Model IIthat was used for testing the M-9 during itsdevelopment. The first production M-9 wasdelivered to the Army in December 1942.Although it served in many areas, probably itsbest-known achievement was its performanceduring the month of August 1944 when 89 of91 V-1s (“buzz bombs”) launched by the enemyfrom across the channel and aimed at Londonwere shot down over the cliffs of Dover.

The success of the M-9 was such that BellLabs was asked to develop a similar device forcoastal defense guns used to protect US shoresfrom enemy ships. The new task was simpler.Where the M-9 had to deal with high-speed,maneuverable targets in 3D space, the Coast

Artillery had to deal with low-speed targets ofminimal maneuverability constrained to the sur-face of the water.

For these reasons, the M-8 gun data com-puter (as the new device was called) was con-siderably simpler than the M-9 gun director. Atthat time it was customary to use the term direc-tor if the data processing equipment was inclose proximity to the optical or radar source ofdata, and to use the term computer if the trajec-tory data came from a remote location as wasthe case with the M-8 data computer.

Quoting Fagan,1 page 158 (emphasis in thefollowing quote is mine):

After the war, the technology developed for firecontrol, using the DC signal methods of the M-9director and the M-8 computer, was applied, underthe direction of Emory Lakatos, to the design of avery useful general-purpose analog computer(GPAC), nicknamed the “Gypsy,” which wasextensively used for over ten years for the solutionof problems in many scientific and engineeringareas. This was the basis for the development of theanalog computer industry. The first Gypsy was builtlargely from unused components for M-8computers leftover from the wartime development.

Gypsy was placed into service in 1949 andproved to be so useful that another was built afew years later. The two machines could be usedseparately for small problems or connectedtogether to work on larger problems. Thesemachines were donated to the PolytechnicInstitute of Brooklyn in 1960 when their work-load was assumed by a more modern, com-mercial analog computer. Figure A showsGypsy.2,3

References1. M.D. Fagen, ed., A History of Engineering &

Science in the Bell System—National Service inWar and Peace (1925-1975), Bell TelephoneLaboratories, New York, 1979.

2. A.A. Currie, “The General Purpose AnalogComputer,” Bell Laboratories Record, New York,Mar. 1951, pp. 101-108.

3. E. Lakatos, “Problem Solving with the AnalogComputer,” Bell Laboratories Record, New York,

Figure A. Richard W. Hamming setting up aproblem on Gypsy. (A History of Eng. & Sciencein the Bell System Comm. Sciences [1925-1980],courtesy AT&T Archives.)

9 gun director. Using appropriate digital inputdata supplied by paper tape, the tester couldsimulate the aircraft’s path in a form suitablefor input for the M-9. Bell Labs was given theresponsibility by NDRC to do the detaileddesign and to build the simulator. It used thesame relay technology as the Model I and hadapproximately the same number of relays. TheModel II took its program instructions andinput data from punched paper tape. It had arepertoire of 31 instructions, and the outputwas a punched paper tape with trajectory datain a form suitable for input to the gun direc-tors. When not being used to develop targetdata, the machine helped researchers solve avariety of other problems.

The simulator was installed at Bell Labs inNew York City in September 1943. Stibitz hadnamed it the Relay Interpolator, although thatname was soon changed to the Model II. Nearthe war’s end, Bell Labs moved the simulator tothe Naval Research Laboratories in Washington,D.C., where it continued on active duty until1961.

Model IIIBecause of the Model II’s modest capabili-

ties, Stibitz was unable to include all the calcu-lations he wanted to incorporate into hissimulator. Consequently, even before theModel II became operational, Stibitz had pro-posed a more sophisticated and powerful com-puter. This new computer also addressed thecomputationally intensive task of testing anti-aircraft equipment. With its additional capa-bility, it could simulate a shell’s ballistictrajectory after the gun director had suppliedthe fusing and aiming information and afterthe gun was fired. Comparing the explodingshell’s simulated detonation position with theplane’s simulated position at the time of thedetonation gave the miss distance and thus theeffectiveness of the shot.

The Model III, shown in Figure 3, was com-monly referred to as the Ballistic Computer. Itused approximately 1,400 relays, three times asmany as the Model I and II. It was installed atCamp Davis, North Carolina, in June 1944 andin 1948 moved to Fort Bliss, Texas, where it wasupgraded with additional registers and servedan additional 10 years working on variousproblems.

Model IVBell Labs built a Model IV for US Naval

Ordnance. Much like the Model III, it was builtto perform the same kinds of tasks. Additionalcircuits were added to account for the guns

being mounted on the deck of a rolling, pitch-ing ship. It was delivered to the Naval ResearchLaboratories in Washington, D.C., in 1945 andremained in service until 1961. The Navyreferred to this machine as the Error DetectorMark 22.

Model VThe government’s need for additional com-

puter power continued. To help address thisneed, Bell Labs was awarded a contract in 1944to use this relay technology to produce twonew machines that would be markedly morepowerful than their ancestors.

These new, more powerful machines wereboth known as the Model V. Although thedesign allowed for a total of six arithmeticunits, these two machines were each equippedwith only two units. The Model V could bedivided into two independent machines towork on two different problems simultaneous-ly or combined into one machine to handlelarger problems. It was the first of the Stibitzmachines to use the floating decimal point.Each of the Model Vs had about 9,000 relays,approximately 55 pieces of teletype equipment,and weighed about 10 tons. The Model I wasbuilt for a cost of $20,000; in contrast, eachModel V cost about $500,000.

The Model V could be programmed to solvea wide variety of mathematical problems. Theseincluded solving problems in probability theo-ry, solving systems of simultaneous equations,solving both ordinary and partial differentialequations, performing integration, and simu-lating physical systems.

Stibitz and others often described the


Figure 3. The Model III computer. (“A Review of the Bell Laboratories’Digital Computer Developments,” Proc. Joint AIEE-IRE Computer Conf.,© 1952 IEEE.)

machines’ useful work by estimating how manypeople, while using desk calculators, would benecessary to perform the equivalent amount ofwork. Stibitz estimated that the Model III didthe work of 25–40 people, depending on thetype of problem being solved, and that theModel V did the work of 225 people.

The first Model V was delivered to theNational Advisory Committee for Aeronauticsat Langley Field, Virginia, in 1946; the secondone was delivered to the Ballistic ResearchLaboratories, Aberdeen, Maryland, in 1947. TheAberdeen machine was later sent to an armysite at Fort Bliss, Texas, and after a period of usewas donated to the University of Arizona.

In 1958 the first Model V was donated toTexas Technological College. While on its wayto Texas, however, the truck transporting ittipped over, critically damaging the machine.The damaged Model V served out the rest of itslife as spare parts for its twin at the Universityof Arizona.

Model VIBell Labs built one last machine of this series

for its own use. Named the Model VI, it was asimplified version of the Model V. It had onlyone arithmetic unit and about half the numberof relays as its predecessor (4,600 versus 9,000).Additional features included the capability tohandle 10-digit numbers in contrast to theModel V’s seven-digit capability.

The Model VI was installed at Bell Labs in1950 where researchers used it to solve a largevariety of Bell Labs’ research and developmentproblems. It was the first digital computer I eversaw in operation. In 1956, Bell Labs donated itto the Polytechnic Institute of Brooklyn whereit was used for research and instructional pur-poses. In March 1960, it was given to the BiharInstitute of Technology in India where, I wastold, it was India’s first digital computer.

These relay machines had high reliabilityand high accuracy, and they were able to run forlong periods of time unattended. During theirentire working life, only two errors were everreported from all three Model V and Model VImachines as a result of machine failure.

Stibitz1 has given a delightful account of hisinvolvement with these computers, andAndrews2 has published a review of them. Anexcellent account of the machines’ design anduse can be found in Williams’s book.3 For stillmore details, see the literature for specific infor-mation on the respective models.4-8

Table 1 lists some of the machines’ character-istics. Figure 4 shows the Model VI, and Figure 5shows one of the remote operating stations.

The AMA computerThe success of these early relay machines

encouraged Bell Labs to apply the same tech-nology to an area vital to AT&T—the billing oftelephone customers for calls made.



Table 1. Design characteristics of Models I to VI, adapted and modified from work by Andrews.9

Complex Number Calculator Design features Model I Model II Model III* Model IV Model V Model VINumber of built-in routines 2 0 0 0 4 200Decimal point: fixed or floating Fix Fix Fix Fix Float FloatMultiplication Yes ** Yes Yes Yes YesDivision Yes No Yes Yes Yes YesSquare root No No No No Yes YesSpecial trig features No No *** *** Yes NoSpecial log features No No No No Yes NoSelf-checking (%) No 90 100 99 100 100Number of relays 460 490 1,400 1,425 9,000 4,600Pieces of teletype equipment 4 5 7 7 55 16Digits per number 8 2 to 5 1 to 6 1 to 6 1 to 7 3,6,10

Multiplication time in seconds per five-digit number N/A N/A 1 1 0.8 0.8

Number of problem stations 3 1 1 1 3 and 4 2Arranged for unattended No No Yes Yes Yes Yes

* This column applies to the Model III before its modification in 1949. **With multiplier specified in program.***Very limited application.

In this system, a data recorder—located atthe switching machine handling the calls—captured the necessary data. The recorder usedoil-impregnated, three-inch-wide, perforatedpaper tape. There was space for 28 holes acrossthe tape allowing recording of six digits, eachrepresenting a single item. Adjacent rows wereabout one-tenth of an inch apart. Four or sixrows were required per call depending on thetype of call (local, long distance, and so on).10

One recorder would handle 100 telephonelines, and data for a particular call would beinterspersed with data from other calls. Datarecorded on the tapes included calling andcalled number, start and stop times of the call,the call type, and so forth. These tapes wereperiodically shipped, usually daily, to an auto-matic message accounting center for process-ing. At the AMA center, the tapes were readinto an elaborate perforated tape operationcalled assembly. The collection of equipmentdoing this task was known as the assembler,shown in Figure 6. It consisted of a paper tapereader, relay logic, and 10 tape perforators.

The assembly operation was a two-stage tapesorting process that resulted in all the data for aparticular call appearing on physically adjacentrows of the tape. This assembly operation also

arranged the call data into a format suitable forinput into the AMA computer. Drew gives adetailed account of this assembly operationand equipment.11

The AMA computer was a relay digital com-puter that performed the arithmetical opera-tions necessary for customer billing. Thecomputer’s output consisted of perforatedpaper tapes that were used as input to fourmore stages of paper tape sorting (to put thecall data in numerical order by phone number).The paper tapes resulting from this operationwere used as input to a printer that printed thecall data in a form readable by the billing clerkswho prepared the final bill. This somewhat


Figure 4. Frame equipment of the Model VIcomputer. (“A Review of the Bell Laboratories’Digital Computer Developments,” Proc. Joint AIEE-IRE Computer Conf., © 1952 IEEE.)

Figure 5. Remote operating station and storage input tables for theModel VI, which was originally installed at the Murray Hill, New Jersey,laboratory in 1949. (“Bell Laboratories Digital Computers,” BellLaboratories Record, Mar. 1957, courtesy AT&T Archives.)

Figure 6. The assembler integrated a paper tape reader,relay logic, and tape perforators for processing telephonecall data. (“The AMA Assembler,” Bell Laboratories Record,May 1952, courtesy AT&T Archives.)

cumbersome tape sorting system might seemstrange today. However, it was not until theBell System Data Processing project (moreabout that later) that Bell Labs decided to com-puterize the sorting operation.

The AMA computer, shown in Figure 7, con-sisted of four cabinets of relay equipment withlamps and control panels, a tape reader, and 14tape perforators.12

Although the just-described system mayseem overly complex and unwieldy, it workedwell. At this time the Bell System had about 25million customers making nearly 200 millioncalls per month. These calls had to be billed to

the customer, accurately and on time. At thesame time, the billing system had to be a low-cost operation to add only a negligible amountto the billable phone call (65 percent of whichwere 15 cents or less). The first AMA centeropened in Philadelphia in 1948. 13

The AMA assembler-computerAs the demand grew for additional process-

ing capability, engineers studied the just-described system intensely for possiblemodifications. These studies resulted in thedevelopment of the AMA assembler-computer,a new relay computer that performed the tasksof both the earlier tape assembler and AMAcomputer. One assembler-computer used lessfloor space and replaced about three of the tapeassemblers and AMA computers. This comput-er consisted of 11 relay bays, one tape reader,and up to 14 AMA tape perforators containedin seven cabinets. Figure 8 shows the assem-bler-computer.

Storage for the start and stop times of 100calls was accomplished by means of a memorycomposed of 1,000 dry-reed relays (a glass-enclosed, hermetically sealed, magneticallyactuated contact). An additional 2,300 wire-spring relays were used for the logic and con-trol portions of the machine.

The inventor of this machine was Amos E.Joel Jr., and the invention’s patent was thelargest one issued in the US up to that time(1960). It contained about 250 claims anddrawings.

More than 100 assembler-computers werebuilt. They served chiefly until the late 1960s,when they were largely replaced by commercialcomputers.14

The throwdown machineIn developing the call-switching machines,

Bell Labs conducted extensive simulations ofproposed designs to determine how well theywould perform under various telephone call-ing patterns and traffic loads. These simula-tions enabled the design engineers to locatebottlenecks in the system and better optimizetheir call-switching design. The input to thesesimulations was various calling patterns andtraffic loads.

To test the operation of a new switchingmachine called the No. 5 crossbar, a manual sim-ulation was run by a team using card files,ledgers, and written records. Dice would bethrown down during the simulation to makerandom decisions, which led to the followingmachine’s being called the throwdown machine.

In 1949, William Keister, Alistair Ritchie,



Figure 7. The AMA computer, showing fourcabinets of control equipment, a tape reader, andseveral of the 14 tape perforators. (“The AMAComputer,” Bell Laboratories Record, July 1952,courtesy AT&T Archives.)

Figure 8. The assembler-computer, which consisted of 11 relay bays,one tape reader, and up to 14 tape perforators. (“The AMA Assembler-Computer,” Bell Laboratories Record, Oct. 1957, courtesy AT&TArchives.)

and G.R. Frost mechanized this manual simu-lation by developing the throwdownmachine,15 shown in Figure 9. This was oftenreferred to as the KRF machine after the initialsof the developers’ last names.

This special-purpose machine could handleall the significant elements of a No. 5 crossbarinstallation, small or large. Engineers used thedata obtained from these simulations to makedesign modifications to the No. 5 crossbar,which improved traffic throughput andenabled determination of load-handling capac-ities for various configurations.

The throwdown machine was composedchiefly of telephone relays and rotary steppingswitches. It had about 800 relays and 57 switch-es—47 of the 22-position, 6-circuit type and 10of the 44-position, 3-circuit type. In addition,there were 60 cords with plugs and 509 jacks.Simulations were run with the aid of four oper-ators. By the mid-1950s, electronic computerswere used for these kinds of simulations.

Gunnery system simulatorDuring World War II, the Navy Bureau of

Ordnance asked Bell Labs to develop a numberof fire control radars to provide defense fortheir combatant ships. The air defense systemhad a search radar that scanned the skies look-ing for enemy aircraft. When an attacking air-craft was detected, the system transferredpositional data to a narrower beam trackingradar that would start tracking the object todevelop data for calculating gun orders. Withthese data, a gun director would calculate aim-ing and firing data for the gun, including fuzesettings. This system controlled dual-purposegun batteries able to fire against both surfaceand air targets.

As the war progressed, the speeds of attack-ing aircraft steadily increased, making the airdefense task more difficult by allowing less timefor a response. Ironically, an additional com-plication for the air defense arose fromimprovements to the tracking accuracy of thefire control radars. The improved tracking accu-racy was achieved by increasing the radar fre-quency, thereby decreasing the radar’s beamwidth. This narrower beam width made it moredifficult for the tracking radar—using the con-siderably less accurate search radar data on arolling, pitching ship—to acquire an attackingaircraft target.

Because of these problems, in the closingdays of the war, the Navy asked Bell Labs toundertake a system study of the entire problem,from search radar detection until the burst ofthe projectile against (or in close proximity to)

the aircraft. The postulated air threat was acoordinated attack by substantial numbers ofaircraft traveling at speeds up to 1,500 knots,allowing little time to organize the air defense.A further complication was that, under airattack, a ship would take violent evasive action,thus continually changing the effective arcs offire of its guns.

The study was carried out over a two-yearperiod, by a team composed of Walter A.MacNair, Bernard D. Holbrook, Alexis A.Lundstrom, and Walter H. MacWilliams. Theteam issued its “Naval AA” report in July 1947.The report proposed automatic track-while-scan of the skies with search radars anduse of the track coordinates to simplify theacquisition of attacking aircraft by the trackingradars. In an automatic track-while-scan sys-tem, a search radar continually scanned the skyto obtain 3D positional data on aircraft withinradar range. The data processing system wouldfirst associate new radar data with establishedstored aircraft tracks, then initiate new trackson aircraft not presently being tracked. Next, itcalculated the velocities of all the aircraft andpredicted their future positions.

In this proposal, the mechanics of assigningincoming aircraft to gun directors would besimplified by use of a display and switchingconsole. This combination was identified bythe acronym TEWA, for Target Evaluator andWeapon Assigner. Further, the study proposedthat a computer be designed to analyze the tac-tical situation and automatically generateassignments subject to manual override. Thisuse of a computer to control the air defense was


Figure 9. The throwdown machine, often called the KRF machine afterthe initials of the developers’ last names. (“A Throwdown Machine forTelephone Traffic Studies,” The Bell System Tech. J., Mar. 1953, courtesyAT&T Archives.)

termed ATEWA (Automatic Target Evaluatorand Weapon Assigner).

Thus, officers would conduct the air defenseby using sophisticated displays and switchingfacilities, with the possibility of carrying outthe defense automatically, using the ATEWAcomputer but subject to manual override. Itwas further proposed that a gunnery systemsimulator (GSS) be built to simulate the attackand the defense response in order to comparethe effectiveness of manual and automaticdefenses against the postulated attacks. TheBureau of Ordnance agreed with the recom-mendations, and Bell Labs designed and builtboth the ATEWA and the GSS.

The GSS had the capability of simulatingattacks of up to 10 aircraft against a defenseinstallation of four tracking radars and gundirectors, and four gun mounts. Target flightpaths were input by teletypewriter punchedpaper tapes. The targets would then be priori-tized, either manually or automatically, byestimated time to reach the ship. The mostthreatening targets would be assigned (manu-ally or automatically) to the tracking radars andto the gun mounts that were most favorably sit-uated to engage them.

The real-time ATEWA computer consisted ofmulticontact relays and implemented defensestrategies based on quantifying the threats ofindividual targets and the availability of track-ing radars and gun mounts. Once an incomingaircraft was assigned to a particular trackingradar and gun mount, simulated gunfire wouldstart. The simulated gunfire began after anappropriate slewing time delay, which repre-sented the time required to move the gunmount from its initial position to the positionnecessary to fire on the attacking aircraft.

The effects of gunfire on the attacking air-craft were simulated as well. A probability of killfor each shot, represented by the symbol p, wasestablished based on the aircraft’s flight pathand the distance of the target from the ship atthat instant. A single-shot kill probability com-puter calculated the probability that each indi-vidual shot would destroy the aircraft beingfired at, as a function of the aircraft’s range forthat shot. A separate computer, the ElectronicDice Thrower, determined, on a probabilisticbasis, whether each particular shot woulddestroy its attacking aircraft based on the com-puted kill probability of that shot. Data wereaccumulated on the success of that particularsimulated attack, and the attacks were replicat-ed to obtain statistical significance. No gunmount would be eligible for reassignment untildestruction of the attacking aircraft was assured.

As remarked earlier, the simulated defensecould be carried out either manually or auto-matically. Manual override of automatic deci-sions could always be made, thus permittingcomparison between manual and automaticcontrol of the defense.

The GSS consisted of about 40 racks ofequipment and was probably one of the largestaggregates of electronic equipment up to thattime.

The simulations carried out in 1950 and 1951proved to be very valuable, and, as a result, theNavy asked Bell Labs to build a system for theUSS Northampton based on the preceding work.The equipment, called the Gun Fire ControlSystem Mark 65, was installed in the ship andreceived extensive shipboard testing.

Although the equipment did an excellent jobin controlling the gunnery defense, antiaircraftguns gave way to surface-to-air missiles. BellLabs modified the defensive system to controlsurface-to-air missiles, with one or two missilelaunchers per ship, keeping the track-while-scanelements. Over a period of some 20 years,weapon director systems for air defense wereinstalled in 74 US combatant ships and in 20ships of seven foreign navies.16 The systems firstused Terrier and later Tartar missiles.

Walter MacWilliams, in a 7 November 1999letter to me, relates that an important bit ofserendipity occurred during these simulationruns:

The Simulator contained a great many vacuumtubes and electronic components and of coursemany component failures occurred, causing timelost in locating and replacing the failed parts. Webecame very sensitive to downtime resulting fromcomponent failures! As a result, the design of theequipment that was to be installed in ships wassubjected to a painstaking process of scrutiny.Electronic components were used at well belowtheir rated voltages and power dissipations to min-imize the chance of failure. In addition, accountwas taken of over-voltages that could occur fromcomponent failures, so that in many cases a circuitwould not be disabled by even a single (unlikely)component failure. As a result, the equipment thatwas installed aboard ships had an unparalleledrecord of reliability, leading to demonstrative affec-tion on the part of fire control personnel.

The transistor gating matrixThe GSS contained, as one of its compo-

nents, the first use anywhere of transistors toperform a circuit function other than one char-acterizing its electrical properties. It came aboutin this way: MacWilliams, who had proposed



the design and construction of the GSS, alsodesigned some of its components, one of whichwas a matrix of gates that channeled pulses rep-resenting gunnery projectiles to circuits thatcomputed the appropriate single-shot killprobability, as a function of the range to theaircraft being fired on. This matrix was calledthe gun-to-p-computer switching matrix.

MacWilliams had designed a perfectly satis-factory gun-to-p-computer matrix circuit usingtriode vacuum tubes. However, Bell Labs hadannounced the invention of the transistor, andseveral engineers obtained experimental tran-sistors to explore their applicability for variousapplications. Jean H. Felker, a colleague ofMacWilliams, was one of these recipients.MacWilliams persuaded Felker to get him someof the preproduction units so that he couldexplore whether transistors would be useful for his application. Using these samples,MacWilliams designed and built a 4 × 10 tran-sistor gate matrix (one transistor per gate) thatworked well. He then incorporated the designinto the GSS itself, where it performed satisfac-torily over the life of the simulator. Figure 10shows the transistor matrix.17

I have related this first application of thetransistor here because of its association withthe gunnery system simulator. However, fromhere on, the transistor holds center stage withrespect to Bell Labs’ computer development.

The beginning of the transistor eraWith Bell Labs’ invention of the transistor

in 1947, Bell Labs’ management recognized thetransistor’s immense importance and devotedextensive laboratory effort to understanding allfacets of transistor physics and technology. Asthat understanding grew, Bell Labs made it apolicy to share this information with other sci-entists and engineers. Many symposia and con-ferences were held for scientists and engineersfrom the military, universities, and business.

Bell Labs summarized its development knowl-edge in November 1951 by publishing 5,500copies of The Transistor, a 792-page book. It wasaffectionately called the Gray book because of itscover’s color and, interestingly, now commandsup to $625 on the secondhand book market. In1958, a three-volume, 1,778-page book—Transistor Technology18—was published, elaborat-ing on the work accomplished since the Graybook’s publication. Its cover was also gray.

Much of this technology was developed ina small department led by Jack A. Morton,staffed by scientists and engineers from manydisciplines. The department’s responsibility wasto convert the fledgling transistor into a con-

sistent, manufacturable device that was under-stood at the engineering level and applicable toreal problems. By mid-1950, several explorato-ry analog and digital applications were beingpursued under a contract that the military’sJoint Services sponsored and the US ArmySignal Corps administered. Three of the digitalprojects were a six-digit angular positionencoder, a four-digit reversible binary counter,and a serial adder with a pair of shift registers.

A shift register can receive a set of digits intime sequence or in parallel, hold the digitsindefinitely, and move the digits in response toa signal. A serial adder produces a sum and car-ries digits when numbers to be added are fedinto it. James Harris designed this five-digit shiftregister and two-word serial adder.19,20 Figure 11(next page) shows Harris holding the adder-register, which is now in the AT&T Archives. Ibelieve it to be the oldest surviving piece ofsolid-state computing equipment in existence.

Most of the early computing componentsand computers built at Bell Labs were scrapped.I sadly admit that I signed off for scrapping sev-eral of the computers used in the first ballisticmissile defense systems developed by Bell Labs.The adder-register is a rare exception to this fate.

After completing the adder-register in 1950and it having served its purpose, Harris offered itto the Signal Corps, which had supported thiswork. Neither the Corps nor Bell Labs was inter-ested in keeping it, but Harris was too proud ofthis machine he had constructed to put it on thescrap pile, so he took it home and stored it in hisattic. There it remained until 1983 when heagain offered to donate it to Bell Labs and this


Figure 10. The transistor gating matrix, withRobert C. Winans at left and Walter H.MacWilliams at right. (Courtesy W.H.MacWilliams.)

time they accepted. Bell Labs in turn loaned it tothe Smithsonian Institution where it was to goon public display in an exhibit the Smithsonianwas designing. After some months, it becameapparent that the proposed exhibit would nottake place, so the device was returned and placedin the AT&T Archives.

The device and its builder were reunited inBozeman, Montana, on 4 September 1999, whenHarris was presented with the 1999 George R.

Stibitz Computer Pioneer Award in a ceremonysponsored by the American Computer Museumand Computer Science Department of MontanaState University. AT&T Archives very graciouslyloaned the adder-register to the AmericanComputer Museum for the occasion.

Other engineers within Bell Labs alsoobtained sample transistors and experimentedto determine what could be accomplished withthis new, exciting device. Jean H. Felker, (whohad given MacWilliams the transistors used inthe GSS), received a new point-contact transis-tor that had been fabricated by Bell Labs’ tran-sistor development group. He first tried to builda linear amplifier with it but could not preventit from oscillating. Finally, he decided that if hecouldn’t prevent it from oscillating, he wouldsee if he could use it in that mode.

Felker then built a blocking oscillator that,when triggered, caused a simple pulse to occur.On examining the pulse output, he observedthat it was very fast (for that time), rising andfalling in a few hundredths of a microsecondand requiring a very small amount of power.

Bell Labs’ lore relates that Felker thenbecame notorious for roaming the halls andvarious offices, showing people the device andasking if they could possibly use it in theirwork. Unlike the usual development scenarios,he had a solution and was looking for a prob-lem. The end result of his search was the birthof TRADIC (transistorized airborne digital com-puter) when he realized that his circuit couldserve as a regenerative amplifier as part of ahigh-speed clocked circuit in a fast, lightweight,solid-state digital computer using low power.

Felker21 proceeded to show how this regen-erative amplifier could be used as the basis fora set of building blocks in a serial computeroperating at a megahertz rate. These buildingblocks included OR, AND, INHIBIT, DELAY, andMEMORY (bit storage cell).

Felker built a multiplier to demonstrate thebuilding blocks. Figure 12 shows his multiplierand its control unit. The multiplier contained38 amplifier packages and could multiply two16-digit binary numbers in 272 microseconds.The control unit on the left side of Figure 12contained 35 amplifier packages and developedthe numbers handled by the multiplier. It alsoprovided test facilities. Each transistor packagewas mounted on a separate plug, and all pack-ages were interchangeable. The dc power forthe entire apparatus was obtained from a powersupply that used semiconductor rectifiers. Partof this work was sponsored by the Navy Bureauof Ordnance.

After listening to Felker’s presentations of



Figure 12. Jean H. Felker’s multiplier and control unit. (Proc. IRE, Nov.1952, p. 1596, © 1952 IEEE.)

Figure 11. James R. Harris with the adder-registerhe built in 1950. (AT&T Archives photo, 1983.)

the new technology, the US Air Force agreedthat a transistorized digital computer would bea major improvement over the analog controlunits in their bombing and navigational sys-tems. Felker emphasized four points in his pre-sentations: solid-state digital computers wouldhave lower power consumption, require lesscubic feet of space, weigh less, and be more reli-able than the technology then in use by themilitary. In one meeting, Felker dramatized thelast point by declaring, “We will give you acomputer that is as reliable as a hammer!”20

Accordingly, in 1951, Felker led a group thatbegan work on TRADIC, with Air Force spon-sorship, to examine the feasibility of construct-ing a transistorized airborne digital computer.

At the same time that the TRADIC work wasstarted, Bell Labs found a second applicationfor transistor digital computers. In a briefdigression, I will first describe the second appli-cation before returning to the TRADIC story.

The prediction computerThis second application was a transistorized

digital computer to be used in a Navy track-while-scan shipboard radar system. In 1951,the Navy authorized Bell Labs to develop thiscomputer, known as the prediction computer.The system requirement was to simultaneous-ly track 50 aircraft. In doing so, the data pro-cessing system first associated new radar datawith established stored aircraft tracks and ini-tiated new tracks on aircraft not presentlybeing tracked. The system then calculated thevelocities of all the aircraft and predicted theirfuture positions.

Two separate computers were to performthese two tasks, but for several reasons only thecomputer to calculate velocities and predic-tions was built. First, there was a shortage oftransistors to construct two machines; next,there was a shortage of engineering personnel;finally, the velocity and prediction task wasdeemed the more demanding task. The predic-tion computer performed coordinate conver-sion, velocity computations, data smoothing,and prediction of future positions. These func-tions required a program of about 100 steps.

The prediction computer was a serial, syn-chronous machine that operated at a 3-MHz bitrate. It used diode-resistor logic and high-speedpulse amplifiers. A sequential access memorywas provided by two 318-word quartz ultra-sonic delay lines. A third 318-word delay linehandled buffering.

Five programs altogether, at a maximum sizeof 168 steps each, were wired into the machine.The programs included two diagnostics for test

purposes, one simulation program for generat-ing test data, and two programs for the track-while-scan application. Addition time was 4microseconds; multiplications required 48microseconds.

The development work was directed byArthur W. Horton Jr. This computer was com-pleted, extensively tested, and demonstrated in1957. Figure 13 shows an oblique view of theprediction computer. Table 2 details themachine’s characteristics.16,22,23


Figure 13. Oblique view of the predictioncomputer. (“A Special-Purpose Solid-StateComputer Using Sequential Access Memory,” Proc.Western Joint AIEE-IRE Computer Conf., © 1959 IEEE.)

Table 2 . The prediction computer was a special-purpose, digitalcomputer that used diode-resistor logic and high-speed pulseregenerative transistor amplifiers.

Feature SpecificationWord length 12 binary bitsNumber of transistors Approximately 1,000 (point contact)Number of diodes Approximately 12,000 (germanium) Internal memory 636 words; two ultrasonic delay linesBuffer memory 318 words, one ultrasonic delay lineProgram control 168-step magnetic-core stepping switchAddition time 4 microsecondsMultiplication time 48 microsecondsOutput Cathode ray tube (aircraft tracks), and

pen recorder (for test purposes)

Although this machine did not go into pro-duction, the knowledge gained during its designand construction was applied to the Nike ZeusAcquisition Radar Data Processing System (atrack-while-scan system). That system used spe-cial-purpose reporter-sorter and track-initiatorcomputers. A general-purpose computer han-dled the remaining track-while-scan functionssuch as data smoothing and tracking, a satellitetest, impact point prediction, and handover oftargets to a precision target-tracking radar.

The TRADIC projectThe other project begun in 1951, with more

long-lasting and far-reaching impact, wasTRADIC. This Air Force-sponsored develop-ment program called for four versions of BellLabs’ solid-state computers.

TRADIC Phase One was developed toexplore the feasibility, in the laboratory, ofusing transistors in a digital computer thatcould be used to solve aircraft bombing andnavigation problems.

Flyable TRADIC was used to establish thefeasibility of using an airborne solid-state com-puter as the control element of a bombing andnavigation system.

Leprechaun was a second-generation labora-tory research transistor digital computerdesigned to explore the capabilities of new solid-state devices for airborne computers. It was alsoused to demonstrate that a stored-program con-trol machine could be built using transistors andto conduct programming and logical designresearch for digital computers to be used in real-time control loops in weapon systems.

XMH-3 TRADIC was to be a solid-state com-puter meeting military environmental specifi-cations as the control unit for a bombing andnavigational system to be installed in a to-be-developed Air Force plane, the B-68.

Each of these four computers has beenreferred to by different names, as listed in Table3, and that has led to some confusion.

The naming confusion extends to theacronym TRADIC as well. Its origin is transis-torized airborne digital computer; frequently,transistorized is changed to transistor.

Many Bell Labs personnel, includingMorton, for example, detested the word tran-sistorized. To them it implied that transistorswere merely substituted for vacuum tubes inolder circuit designs. Of course, the solid-statemachines were completely new designs.

The TRADIC Phase One computerThe TRADIC Phase One computer was built

to demonstrate successful laboratory operationof a high-speed, general-purpose, solid-statedigital computer together with input-output(I/O) equipment representative of its use as thecontrol unit of an aircraft bombing and navi-gation system.24,25 General purpose here meansthe computer can add, subtract, multiply, anddivide numbers and move them among itsinternal units and I/O equipment, all undercontrol of any program that has been preparedfor and inserted into it. The programs, whichcontrolled the machine, were introduced via aremovable plugboard. Table 4 gives a summaryof some of its characteristics.

Felker led the development team and Harriswas the supervisor reporting to Felker. TheTRADIC Phase One computer was completedin January 1954. It was the first large transistordigital computer. Also, it was probably the



Table 3. TRADIC computers and the names they have been known by.

Computer Also known asTRADIC Phase One computer Feasibility Model TRADIC

TRADICFlyable TRADIC TRADIC Phase Two

TRADIC Flyable ModelFlyable Model TRADICFlyable Research Model TRADICTRADIC

Leprechaun Second TRADIC Feasibility ComputerXMH-3 TRADIC Control Unit for XMH-3 System

Flyable TRADICXMH-3 Computer

Table 4. The TRADIC Phase One Computer. Germanium point-contact diodes were used for logic operations and other circuitfunctions. Point contact transistors were used in circuits to reshapepulses distorted by the logic networks.

Component DescriptionNumber of transistors 684 Bell Labs Type 1734 Type A cartridge

transistorsNumber of diodes 10,358 germanium point-contact diodesWord size 16 bits-serial. Each word was stored in a

separate electrical delay line 16 microseconds long.

Number storage 16 addressable electrical delay lines. Constants were input using toggle switches.

Addition or subtraction time 16 microsecondsMultiplication or division time Less than 300 microsecondsDigital-to-analog converters Two that converted numbers to voltagesClock 30 watts at 1 megacycle supplied by an

electron tube because no transistors were available which could supply this much power at this frequency.

largest equipment unit in terms of the numberof transistors used—684—up to that time.

One of the major decisions made early inthe design stage was to use solid-state diodes forthe logic circuits. Bell Labs already had a strongbackground in diode logic. Beginning in thelate 1930s, Bell Labs had conducted an inten-sive review of various technologies that werepotentially useful for future telephone callswitching systems, and the solid-state diodewas one of those studied. This investigation ledto new and useful diode configurations for per-forming logical operations. Two of the most-used diode logic circuits, the AND gate and theOR gate, were invented by H.T. Holden26 andArthur W. Horton.27

Given these achievements, diode logic cir-cuitry was well understood and very reliable in1951 when TRADIC Phase One was beingdesigned. The use of this proven technologylimited the risk of failure in this new machine.Bell Labs decided that a follow-on computer(Leprechaun) would be developed to exploretransistor logic as a main objective.

TRADIC Phase One had a 1-MHz clock usingtransistors for pulse shaping and amplification,and it used 10,358 germanium point-contactdiodes. About one-third of the diodes wereused for the logic circuits. The remainder hadcircuit functions such as isolation, pulse shap-ing, and clamping plus some functions pecu-liar to the use of point-contact transistors.

The computer performed additions and sub-tractions in 16 microseconds; multiplicationand division, in less than 300 microseconds.Number storage was accomplished with 16addressable electrical delay lines.

The 30 watts of power for the 1-MHz clockwas supplied by a vacuum tube supply becauseno transistors were available that could supplythis much power at this frequency. Logic cir-cuits used 60 watts of dc power provided byprecisely regulated, all solid-state power sup-plies. The only adjustable elements in thecomputer were in the power and clock supplies,and the computer operators used them toadjust the dc voltage and clock phase to stan-dard values.

Programs for the computer were set up onremovable plugboards. Each plugboard couldhandle a program of 64 machine steps, togeth-er with one subroutine, recallable as needed,also of up to 64 steps.

For two years—May 1954 to May 1956—TRADIC Phase One ran 24 hours a day in a reli-ability study. For some results from the study,see the “Highlights from TRADIC SummaryReport” sidebar (next page).

One of the hallmarks of the early Bell Labs’transistor digital computers was the extensiveuse of margin checking, which let the comput-er operators determine the machine’s health. Toperform margin checking, the operators wouldvary the power supply voltage to the computerfrom nominal during operation. If the machinecontinued to operate normally, all was well.However, if the computer started making errors,that was a warning that one or more compo-nents had deteriorated and that perhaps themachine would fail if preventive action werenot taken to locate and replace the weak com-ponent. Margin checking proved highly effec-tive in early identification of weak components.

Harris introduced margin checking into theTRADIC design. In his earlier work on tele-graph equipment as an engineer with the C&PTelephone Co. of Virginia, he had observedthat traditional margin checking was a verypowerful maintenance tool that greatlyenhanced system reliability (1999, privateemail communication). Figure 14 shows Harriswith the TRADIC Phase One.

The Flyable TRADIC computerThe Flyable TRADIC computer was devel-

oped as a flying testbed to establish the feasi-bility of using a transistorized digital computeras a real-time control element in a bombingand navigation system.

In the early 1950s, Bell Labs developed theBombing and Navigation System, Optical andRadar, Type K-5, and Western Electric pro-


Figure 14. James R. Harris at the control unit ofthe TRADIC Phase One. (“TRADIC: The FirstPhase,” Bell Laboratories Record, Sept. 1958,courtesy AT&T Archives.)

duced 65 of these systems for the Air Force’s B-66B aircraft. The control units were vacuumtube analog computers. The Flyable TRADICcomputer was to be flight-tested with the radarportion of this system, with Flyable TRADICreplacing the vacuum tube analog computer asthe control unit.

To achieve this objective, the Bell Labs engi-

neers reconfigured the TRADIC Phase Onemachine to withstand the aircraft’s environ-mental conditions, including vibration, shock,and humidity, for example. Design changes tothe TRADIC I/O circuits were necessary so thatthe TRADIC computer could interface correctlywith the radar portion of the K-5 system. Otherchanges assured reliable operation in the



The following three sections—Abstract, Project Goals,and History—are reproduced verbatim from the TRADIC,Computer Research Program, Summary Engineering Report32. This report was issued 1 June 1957, ContractAF33(600)-21536 for the US Air Force, Air MaterielCommand, prepared by Bell Telephone Laboratories onbehalf of Western Electric Company.

AbstractThis report summarizes the TRADIC Computer Research

Program for the development of a solid-state technologyfor airborne weapon-control computers. Among the majoritems covered are: direct-coupled transistor logic circuitry,transistor-driven magnetic-core memory, time encodingand decoding equipment and the philosophy of their use,and the system design of the digital weapon-control com-puter. Evaluations of these and competing techniques areincluded where appropriate.

The summary proper is accompanied by two auxiliaryVolumes, of which Supplement I is a compilation of allTRADIC work on direct-coupled transistor logic circuitry.Supplement II is in the nature of a quarterly report, andpresents the details of work completed during the twelfthquarter of the TRADIC computer research project.

Project GoalsThe object of the computer research program has been

to develop a digital-computer technology which exploitsthe most recent advances in the solid-state art and is tai-lored to meet current airborne weapon-control problems.Achievement of this objective required work in severalareas including weapon-system studies, programming,logical design, input-output systems, large-scale memorysystems, digital building blocks, and device work, e.g.,work on transistors and magnetic cores.

History of the ProjectThe potential importance of the transistor in the design

of military equipment was unmistakably clear at an earlystage of transistor development. Programs for evaluationof the feasibility of the use of transistors for military appli-cations were inevitable. The prospect of relatively complexsystems operating at low power and occupying less vol-ume and weighing less than corresponding vacuum-tubesystems was attractive enough to arouse the interests ofmany people, particularly those concerned with airborne

systems. TRADIC (coined from the words transistor air-borne digital computer) was set up as one of the first proj-ects to explore the application of transistors to a particularweapons problem.

Before transistors entered the picture, computers forbombing and navigation systems had almost universallyused analog techniques. An early study concluded, how-ever, that with available point-contact transistors an ana-log computer offered fewer dividends than a digitalcomputer to perform this function. As a result, the first aimof the TRADIC project was to determine the feasibility ofthe use of point-contact transistors in a digital computerof the type required in the K-5 bombing and navigationsystem which was under development at the BellTelephone Laboratories.

TRADIC was set up to include two phases. The goal ofphase one was to develop a set of digital building blockssuitable for the construction of a complete computer andto build a small computer from these building blocks capa-ble of solving part of the bombing and navigation prob-lems. In addition to the demonstration of feasibility, it wasexpected that this computer would provide valuable infor-mation on the use of a relatively large numbers of transis-tors, particularly with respect to their reliability.

The building blocks developed in phase one used semi-conductor diodes to perform the logic. Transistor syn-chronous pulse amplifiers were used to provide gain andpulse regeneration. The circuits were designed to operateat a pulse repetition rate of 1 megacycle. The transistorswere germanium point-contact switching transistors whichwere the developmental predecessors of the Type 2N67.The phase one computer was constructed using 684 tran-sistors and 10,614 [sic] diodes. It was first demonstratedin January 1954, and is believed to have been the first tran-sistorized digital computer ever built.

The phase one computer, although limited in size, wascapable of solving the simple bombing problem and didindeed demonstrate that these techniques were suitablefor a bombing and navigation computer. In addition, thephase one computer has provided some interesting lifetestdata. In 17,103 hours of operation, 8 out of 684 transis-tors were replaced and 9 out of 10,614 diodes werereplaced, giving failure rates of 0.07 percent and 0.005percent per thousand hours, respectively. Of the eighttransistors replaced, five were detected by marginal check-ing before they caused a failure of the computer.

Highlights from TRADIC Summary Report

harsher environment. For example,the plugboards for entering programsinto the TRADIC Phase One comput-er would be unreliable in aircraftoperation and also too complex withthe longer program. Consequently,George G. Smith developed a newstorage mechanism for programs byusing a Mylar sheet with punchedholes, a system reminiscent ofpunched-card storage.

Two Flyable TRADIC machineswere constructed, one for flight-testing purposes aboard a loaned AirForce EC-131B cargo plane and theother for program development anddebugging in the laboratory. Theoperator’s position is shown in Figure15. Figure 16 shows the EC-131Bcargo plane with the equipmentinstalled. Table 5 gives some of thecharacteristics of Flyable TRADIC.

Louis C. Brown, who did much ofthe Flyable TRADIC’s programmingand flight testing, has written anAnnals article giving a detailed account of thesystem and of the issues encountered during itsdevelopment.28

LeprechaunLeprechaun, a second-generation TRADIC

computer, enabled Bell Labs engineers to fur-ther explore the capabilities of solid-statedevices for use in airborne computers. Muchtime was devoted to programming and logicaldesign research for computers to be used in

military real-time control loops.Initially, Felker led the development team and

Harris, a supervisor then, reported to him. Later,Felker was reassigned to another Bell Labs com-puter project and was replaced by Jack A. Baird.Six months later, Harris was transferred to thesame project as Felker, and shortly after that, JackA. Githens was appointed Harris’ replacement.

Leprechaun does not fit any discernible pat-tern, compared to the naming origins of otherTRADIC computers, and the reason behind it


Figure 15. Flyable TRADIC’s operator’s position. (TRADICFlyable Research Model Program Summary Eng. Report, US AirForce, 1958.)

Figure 16. Flyable TRADIC installed in the EC-131B aircraft.(From AT&T Archives press release.)

Table 5. The Flyable TRADIC computer was a synchronous, binary, serialcomputer with parallel multiplication.

Feature SpecificationNumber of transistors 2269 (WECO 2N67)

416 (WECO GA-52996)Number of diodes 10,740Word size 16 serial bitsNumber of storage slots (variable) 52 serial delay-line storage slotsNumber of storage slots (constant) 80 fixed constant-storage slotsProgram store Mylar sheet with punched holesAddition or subtraction time 16 microsecondsMultiplication time 64 microsecondsDivision time 304 microsecondsProgram solution time 50 millisecondsInput/output 8 shaft-angle inputs

6 incremental inputs (slew and track)16 manual inputs (hand set and track)28 outputs (meters, servos, radar)

Clock 4 phase, 1.036 MHzWeight 1,193 pounds (excluding air conditioning)Size 39.5 cubic feetPower 450 watts

bears telling. Because this new computer wouldhave the ability to run much longer programsthan the TRADIC Phase One, the Bell Labsdevelopment group started referring to it as theLong Program computer. One day, while in theshower, Robert A. Kudlich—who was responsi-ble for Leprechaun’s system and logicaldesign—starting musing on how to make anacronym from those three words. He startedwith LONPROCOM. As he repeated that nameit sounded to him like Leprechaun (1999, pri-vate phone-call communication). Later, the restof the group felt it was appropriate and thename stuck. Githens29 has related why he feltthe name particularly suited this machine:

For when one turns to Webster, as everyone mustto check the spelling, one finds that it is a veryappropriate name. Leprechaun is from the MiddleIrish lu, meaning “little” and from corpan, diminu-tive of corp; corp is from the Latin corpus meaning“body.” Little body—that is just what we were afterwhen we started designing this machine. Further,

the dictionary reveals that inIrish folklore a leprechaun isa little fairy generally con-ceived as a tricky old man,who if caught may reveal thehiding place of treasure. Afterabout a year of struggle, wehave caught our Leprechaun,although we are still seekingthe treasure.

A coincident-currentmagnetic memory servedas storage. The logic func-tions were implemented bydirect-coupled transistorlogic circuitry (DCTL).

Although Philco invented this circuitry, Harrisnamed it. Leprechaun was a binary, parallel,asynchronous machine. The operation codeprovided all the common arithmetic, logical,and transfer operations—a total of 28 opera-tions. Table 6 summarizes Leprechaun’s maincharacteristics. A more detailed description ofLeprechaun has been given by Jack Githens29

and later by Jack Baird.30 An exhaustive, 812-page review of Leprechaun’s details was pre-pared and edited by Githens and Gilmartin.31

Figure 17 shows Leprechaun at its Bell Labs’Whippany, New Jersey, location. Leprechaunbecame operational in 1956 and, after exten-sive use at Bell Labs, was delivered to the AirForce. It was installed at the Wright-PattersonAir Development Center at Dayton, Ohio, in1959, where it was used in evaluating Air Forceweapons systems.

Harris has recently given an assessment of thehistorical significance of this machine (privatecommunication, letter, 27 September 1999):

The Leprechaun computer was a very significantadvance in three major areas.

First, this machine was the earliest solid-statecomputer to use true stored-program control,sometimes called von Neumann architecture. Toput this a little differently, Leprechaun was thefirst solid-state computer to employ software!

Second, Leprechaun was the first computer touse junction transistors. It was known from thevery early days of the transistor that the junc-tion type had the potential to be much superiorto the point-contact type. Why? The point-contact device was inherently difficult to under-stand and analyze and duplicate. In contrast, thejunction device, when it became available, lentitself to analysis and prediction at all levels, thatis, device design, device manufacture, and cir-cuit application.



Table 6. The Leprechaun computer.

Feature SpecificationNumber of transistors 5,500 (Raytheon CK761, GE 2N137, Philco Surface Barrier

SB100, Western Electric GA52830)Memory 18,000 coincident-current magnetic coresMemory capacity 1,024 18-digit words including parity check bit and sign bitAddition or subtraction time 48 microsecondsMultiplication time 350 microsecondsDivision time 430 microsecondsPrimary input Ferranti photoelectric reader and paper tapePrimary output Teletype high-speed punchOptional output Electric typewriterSize 15 cubic feet, not including I/O equipmentPower 250 watts

Figure 17. F.W. Hodde (right) feeds a problem into Leprechaun while N.J.Powell prepares the machine for operation. (“Military Applications,” BellLaboratories Record, June 1958, courtesy AT&T Archives.)

Finally, Leprechaun was a direct precursor ofthe integrated logic circuit that is now the engineof world economy. The machine used as a build-ing block a DCTL (Direct Coupled TransistorLogic) logic plate containing only transistors andresistors, a scheme perfectly adapted to be real-ized later in the form of an integrated circuit. ADCTL logic plate is shown in Figure [18]. Theplate was built in universal form, without wiring,and then was wired to realize a specific logicdesign. In this sense the DCTL plate foreshad-owed the silicon master slice.

DCTL has been described by Harris32 anddiscussed in depth (21 papers by 13 authors) inSupplement 1 (A Handbook of Direct-CoupledTransistor Logic Circuitry).31

TRADIC Summary Engineering ReportAs the research phase of the TRADIC devel-

opment program reached completion, aSummary Engineering Report33 was issued thatreviewed the project history, outlining whathad been learned and giving suggestions forfuture research.

To convey the flavor of the development ofthese three machines and to give the developers’thoughts in their own words, the “Highlightsfrom TRADIC Summary Report” sidebar repro-duces the Abstract, Project Goals, and Historysections of that report written in 1957.

The XMH-3 TRADIC Although ultimately never brought to com-

pletion, XMH-3 TRADIC was a solid-state com-puter to be built to military specifications andused as the control unit of the XMH-3Bombing and Navigation System. This systemwas to be installed in a two-person, to-be-developed high-speed tactical bomber (B-68)that would be effective within a radius of 1,000nautical miles at all hours and in all weather.The Air Force selected two aircraft companies,Douglas and Martin, for the design competi-tion for the B-68 development contract. Theairframe, with all of its equipment and payload,was known as the 302A Weapon System.

The XMH-3 Bombing and NavigationSystem was quite sophisticated for its time. Itfeatured an inertial data generator (for inertialguidance), a bombing and navigation radar set,a three-beam radar terrain avoidance system,and a photographic unit.

The mission profile called for the plane tofly at high altitudes over friendly terrain (toconserve fuel) and, upon entering hostile terri-tory, descend to a low altitude (100 to 200 feet)to avoid radar detection and radar-directedweapons’ fire.

On reaching the target, the aircraft wouldexecute one of several bomb release maneuversas shown in Figure 19. The entire flight wasunder the XMH-3 TRADIC computer’s control.

These were to be highly briefed missions,and a photographic unit would take picturesalong the way, compare those pictures withbriefing photos, and the navigator made anynecessary position corrections to the system.

Simulated flights of the B-68 under the con-trol of the XMH-3 TRADIC computer were con-ducted on an analog computer (see the


Figure 18. Direct coupled transistor logic (DCTL)plate (top view). (TRADIC Computer ResearchProgram Supplement I, A Handbook of Direct-CoupledTransistor Logic Circuitry, US Air Force, 1957.)

Figure 19. 302A Weapon System weapon-delivery modes. (XHM-3System Third Interim Eng. Report, US Air Force, 1956.)

“Gypsy” sidebar) by Bell Labs’ MathematicsResearch Department. A favorite exercise wasto fly low over the English Channel toward theWhite Cliffs of Dover, which gave the terrainavoidance component a good workout.

The XMH-3 TRADIC design relied heavily onthe experience gained during development ofTRADIC Phase One, Flyable TRADIC, andLeprechaun. Major changes had to be made tomeet the size and weight requirements as well asthe military specifications covering, for example,humidity, temperature, and vibration. The com-puter’s major characteristics are listed in Table 7.

The Air Force, in April 1956, delivered a C-131B to Bell Labs to be used in initial flight test-ing of the XMH-3 Bombing and NavigationSystem. In the fall of that year, the Air Forcecancelled the 302A Weapon System project forbudgetary reasons. Additional information onthe XMH-3 TRADIC can be found in the ThirdInterim Engineering Report.34

The digital technology developed on thisproject and the projects discussed earlier wereused as the basis for a series of new digital com-

puters, developed in partnership with Univac.These new computers were used on the NikeZeus, Nike X, Sentinel, and Safeguard inter-continental ballistic missile defense systems.

The Bell System data processingcomputer

When transferred from the Leprechaun proj-ect, Felker was appointed systems engineeringdirector of a project charged with developing acomputer for the Bell System telephone com-panies’ use in business operations. This com-puter, initially envisioned as a replacement forthe assembler-computer, would use magnetictape instead of perforated paper tape to capturethe call data. Later, under Felker’s leadership,the charter was broadened to include some ofthe telephone companies’ business operations.

The organization to carry out the develop-ment work was established in 1956, and thecomputer under development was called theBell System Data Processing (BSDP). Initially, aseparate computer was to be used for the four-stage tape-sorting operation discussed earlier.However, as the work progressed it becameclear that the sorting operation could beaccomplished in the BSDP and a separate com-puter would be unnecessary. Personnel weredrawn from the Bell System telephone compa-nies to aid in establishing the requirements forthe new computer. Many of the Bell Labs engi-neers who had experience with the earlier com-puters were reassigned to the project, and thisis the project to which Harris was transferredfrom the Leprechaun project.

The requirements were established and themachine’s major components’ design wasunder way when, in 1958, Bell Labs manage-ment made the decision to cancel the project.A survey of soon-to-be-available commercialcomputers convinced management that thetelephone companies’ needs could best be metby using commercially available machines.

A footnote to this history is that shortlyafter the cancellation of the BSDP computer (ithad never had another name), Felker was trans-ferred from Bell Labs to AT&T and later servedas operations vice president of New Jersey Bellbefore he retired to Spain to paint.

Bell Labs then established an organization,reporting to a vice president, called BusinessInformation Systems-Programs (BISP) for thepurpose of developing software for commer-cially available computers for the telephonecompanies’ business operations. Once again,engineers from the telephone companies wereassigned to Bell Labs to aid in establishing sys-tem requirements, and Felker was coaxed out



Table 7. The XMH-3 TRADIC computer’s proposed characteristics.Five computers were to be built: one laboratory breadboard model,one flight test model, and three deliverable models for the B-68.

Feature SpecificationProgram storage Punched-card storage as used in Flyable

TRADICFlight-plan storage Ferrite-core memory made nondestructible

by the placement of permanent magnets near each core that should show a “1”. This type of memory was referred to as a Shachell Store. Stored in this memory would be up to four flight paths in a flight plan, using a maximum of 30 fix-and-aim points. Each fix or aim point required 161 bits.

Variable memory storage Ferrite-core memory, minimum capacity of 256 18-bit words

Logic circuitry DCTL as used in Leprechaun, although where Leprechaun used about 5,500 transistors, this would be 10,000 transistors.

Computing cycle 0.25 second or less. Operation would be asynchronous between start and completion.

Operation and address size 17-bit word (5 bits specify operations, 11 bits specify addresses and constants, and 1 bit indicates a program step that the computer cannot modify)

Weight 273 poundsSize 10,560 cubic inches of spacePower 160 watts

of retirement to be the VP heading up the BISPdevelopment effort.

EpilogueThe military supported much of Bells Labs’

computer research and development work.Although many other organizations alsoreceived strong support from the same source,Bell Labs was in a somewhat unique position.At this time, AT&T was a highly regulated com-pany and had no legal right to develop com-puters for other than its own internal orgovernment use. This fact shaped computerdevelopment at Bell Labs during the time peri-od being discussed.

The Bell System was an appreciable marketfor computers, as evidenced by the approxi-mately 100 assembler-computer machinesdelivered to the telephone companies. So it wasnatural that, when Bell Labs determined thetelephone companies’ computing demandswere sufficiently great to warrant developmentof a suitable follow-on computer, the BSDPproject was established.

In my opinion, the decisive moment for thefuture of computer research and development atBell Labs occurred in 1958 when managementdecided that suitable commercial computerswould be available in the near future and can-celed the BSDP project. Bell Labs further empha-sized its withdrawal from the field by forming apartnership with Univac for developing com-puters for ballistic missile defense systems.

However, the various telephone call switch-ing machines that Bell Labs designed andWestern Electric manufactured are truly special-purpose computers optimized for call switch-ing. These switching machines were producedby the hundreds. Several other small more gen-eral computers were also built.

AcknowledgmentsIt is with a great deal of pleasure that I acknowl-edge all the help I have received from Louis C.Brown and Walter H. MacWilliams. They havebeen very generous with their time in discussingtheir activities on the computer systems dis-cussed here. They also shared with me criticaldocumentation including photographs andcopies of their papers. And it was a lot of funreliving events with them of those long-ago days.

I owe special thanks to J.R. Harris whom Ipestered constantly during the preparation ofthis article. Jim always responded promptly tomy questions, whether they were about dates,people, events, or technology, with thoughtful,well-researched answers.

Figures A, 5 to 9, 14, and 17 are the property

of AT&T Archives and are reprinted with per-mission of AT&T.

I would also like to express my appreciationto the Annals editors for their patience with mewhile they transformed this article into pub-lishable form.

References and notes1. G.R. Stibitz, “The Relay Computers at Bell Labs,”

Datamation, vol. 13, nos. 4 and 5, Apr. and May1967, pp. 35-49.

2. E.G. Andrews, “A Review of the Bell Laboratories’Digital Computer Developments,” Proc. JointAIEE-IRE Computer Conf., Dec. 1951, Feb. 1952.(Reprinted in Bell Telephone System MonographNo. 1972.)

3. M.R. Williams, A History of Computing Technology,2nd ed., Chapter 6, IEEE Computer Soc. Press,Los Alamitos, Calif., 1997, pp. 221-234.

4. O. Cesaroo, “The Relay Interpolator,” Bell Labora-tories Record, vol. 24, no. 12, Dec. 1946, pp. 457-460; Relay Interpolator, final tech. report to theNational Defense Research Council, Bell TelephoneLaboratories, New York, 31 Oct. 1945; The RelayInterpolator, report no. R-3177, Naval ResearchLaboratory, Washington, D.C., 25 Sept. 1947.

5. J. Juley, “The Ballistic Computer,” Bell Laborato-ries Record, New York, vol. 25, no. 1, Jan. 1947,pp. 5-9.

6. Computer Mark 22, Model 0, report 178-45, Unit-ed States Navy, Washington, D.C., 12 Dec. 1945.

7. S.B. Williams, “A Relay Computer for GeneralApplication,” Bell Laboratories Record, New York,vol. 25, no. 2, Feb. 1947, p. 49; S.B. Williams, “ABell Telephone Laboratories Relay Computing Sys-tem,” Annals, Harvard Computation Laboratory,Cambridge, Mass., vol. 16, 1947, p. 41; F.L. Alt, “ABell Telephone Laboratories Computing Machine,”Mathematical Tables and Other Aids to Computation,National Research Council, Washington, D.C., vol.3, no. 21, part 1, Jan. 1948, pp. 1-13, and part 2,no. 22, April 1948, pp. 69-89; Additional picturesof Aberdeen installation, Bell Laboratories Record,New York, vol. 26, no. 5, 1948, p. 208; H.W. Bodeand E.G. Andrews, “Use of the Relay Digital Com-puter,” Electrical Eng., New York, vol. 69, no. 2,Feb. 1950, pp.158-163.

8. E.G. Andrews, “The Bell Computer Model VI,”Annals, Harvard Computation Laboratory, Cam-bridge, Mass., vol. 26, 1951, p. 20; E.G.Andrews, “The Bell Computer Model VI,” Electri-cal Engineering, New York, vol. 68, no. 9, Sept.1949, pp. 751-756.

9. E.G. Andrews, “Bell Laboratories Digital Comput-ers,” Bell Laboratories Record, New York, Mar.1957, p. 82.

10. H.D. Cahill, “Recording on AMA Tape in Central


Offices,” Bell Laboratories Record, New York, Dec.1951, pp. 565-570.

11. G.G. Drew, “The AMA Assembler,” Bell Laborato-ries Record, New York, May 1952, pp. 227-232.

12. A.E. Hague, “The AMA Computer,” Bell Laborato-ries Record, New York, July 1952, pp. 289-294.

13. J. Meszar, “Basic Features of the AMA Center,”Bell Laboratories Record, New York, Feb. 1952, pp. 70-74.

14. T.C. Rehm, “The AMA Assembler-Computer,” BellLaboratories Record, New York, Oct. 1957, pp.423-427.

15. G.R. Frost, W. Keister, and A.E. Ritchie, “A Throw-down Machine for Telephone Traffic Studies,”The Bell System Tech. J., vol. 32, no. 2, Mar. 1953,pp. 292-359.

16. M.D. Fagen, ed., A History of Engineering &Science in the Bell System—National Service in Warand Peace (1925-1975), Bell Telephone Laborato-ries, New York, 1978.

17. W.H. MacWilliams, “A Transistor Gating Matrixfor a Simulated Warfare Computer,” Bell Labora-tories Record, New York, Mar. 1957, pp. 94-99.

18. Transistor Technology, H.E. Bridgers, J.H. Scaff,and J.N. Shive, eds., vols. 1, 2, and 3, VanNostrand, New York, 1958.

19. J.R. Harris, “A Transistor Shift Register and SerialAdder,” Proc. Inst. of Radio Engineers (IRE), NewYork, Nov. 1952, pp. 1597-1602.

20. J.R. Harris, “The Earliest Solid-State Digital Com-puters,” IEEE Annals of the History of Computing,vol. 21, no. 4, 1999, pp. 49-54.

21. J.H. Felker, “Regenerative Amplifier for DigitalComputer Applications,” Proc. IRE, Nov. 1952,pp. 1584-1596.

22. J.A. Baird, “Military Applications,” Bell Laborato-ries Record, June 1958, pp. 221-225.

23. W.A. Cornell, “A Special-Purpose Solid-StateComputer Using Sequential Access Memory,”Proc. Western Joint AIEE-IRE Computer Conf., Mar.1959, pp.74-79.

24. J.H. Felker, “Performance of TRADIC TransistorDigital Computer,” Proc. Eastern Joint AIEE-IREComputer Conf., ”Performance of TRADIC Transis-tor Digital Computer,” Computers and Automation,vol. 4, no. 1, Jan. 1955, p. 16 (abstract).

25. J.R. Harris, “TRADIC: The First Phase,” Bell Labora-tories Record, vol. 36, Sept. 1958, pp. 330-334.

26. W.H.T. Holden, US Patent No. 2,299,898; Patentand Trademark Office, Washington, D.C., filed 16Oct. 1941; issued 27 Oct. 1942.

27. A.W. Horton, US Patent No. 2,244,700; Patentand Trademark Office, Washington, D.C., filed 21Sept. 1939; issued 10 June 1941.

28. L.C. Brown, “Flyable TRADIC—The First AirborneTransistorized Digital Computer,” IEEE Annals of

the History of Computing, vol. 21, no. 4, Oct.-Dec.1999, pp. 55-61.

29. J.A. Githens, “The TRADIC LeprechaunComputer,” Proc. Eastern Joint AIEE-IRE ComputerConf., 1956, pp. 29-33.

30. J.A. Baird, “Leprechaun Computer,” Bell Laborato-ries Record, New York, Feb. 1960, pp. 58-63.

31. “TRADIC Computer Research (Second Stage),”Summary Eng. Report 1, Dec. 1958, AD-208621,J.A. Githens and M. J. Gilmartin, eds.; TRADICComputer Research (Second Stage),” SummaryEng. Report, Supplement: Leprechaun, A Solid-StateDigital Computer, 1 Dec. 1958, ContractAF33(600)-34977, U.S. Air Force, Air MaterielCommand, AD-208622, J.A. Githens and M. J.Gilmartin, eds. Also, Tradic Computer ResearchProgram Supplement I, A Handbook of Direct-Coupled Transistor Logic Circuitry, 1 June 1957,Reporting Period 1 Feb. 1954 to 31 Dec. 1956,Contract AF33(600)-21536, US Air Force, AirMateriel Command.

32. J.R. Harris, “Direct Coupled Transistor Logic Cir-cuitry,” IRE Trans. Electronic Computers, vol. EC-7,no. 1, Mar. 1958.

33. J.A. Baird et al., “TRADIC Computer Research Pro-gram,” Summary Eng. Report, US Air Force, AirMateriel Command, Washington, D.C., 1 June1957.

34. Bell Telephone Laboratories, XHM-3 System ThirdInterim Engineering Report, Air MaterielCommand, USAF, Washington, D.C., 15 Nov.1956, AD-115 302.

M.M. Irvine retired from Bell Labs in 1984, which hejoined in 1955 after teaching physics at LehighUniversity, following service in World War II as aradio officer for the merchant marine. His responsi-bilities at Bell Labs focused on software developmentfor various ballistic missile defense systems and, later,for the Bell System operating companies. He servedon the editorial board of the Bell Laboratories Recordfrom 1978 to 1983. Irvine received a BS in engineer-ing physics from Montana State College and an MSand a PhD in physics from Lehigh University. He is aSenior Life member of the IEEE, a member of theAmerican Physical Society, and of the AmericanAssociation of Physics Teachers.

Contact M.M. Irvine at 511 South Willson Ave.,Bozeman, MT 59715; email [email protected].

For further information on this or any other com-puting topic, please visit our Digital Library athttp://computer.org/publications/dlib.



Documents

Early Digital Computers at Bell Telephone Laboratorieslopez/articulos/00948904.pdf · development arm, and Western Electric was AT&T’s manufacturing division. Western Electric and