New Engineering Companies and the Evolution of The United States Computer Industry

Arthur L. Norberg Universityof Minnesota

The electronicdigital computerindustry emerged tentatively onto the world stagein 1946,with the foundingof two engineeringcompanies and the incorporationby severallarge firms of the new electroniccomputing ideas into their planning.Only a handfulof largefirms participated in thisnew computer phenomenon--IBM,Raytheon, Bendix, and Burroughs. The smallstartup firms includedEckert-Mauchly Computer Corporation (1946), Engineering Research Associates,Inc. (1946), and CaliforniaResearch Corporation (1950). These latterthree firms were eventuallyabsorbed into largerfirms. Eckert-Mauchly Computer Corporation and Engineering Research Associateswent into Remington-Randin 1951 and 1952, respectively,and California Research Corporationbecame National Cash Register'scomputer division in 1954. Raytheonlater confinedits computerefforts to fire control and missile systems,while Bendix,after developing several computer systems, became part of Control Data Corporationin 1962. Burroughsis now one of two main componentsof UnisysCorporation along with SperryRand, the successorof Remington-Rand,and, of course,IBM is still the principalcompany in the computerindustry. A similarstory could be told abouta numberof startups in the first half of the 1950s, but the story would be the same. The small firms with an engineeringemphasis in leadershipand capabilityprovided an importantstimulus to the early computerindustry in the United States. These businesseswere organizedat a time before a stable computerdesign was available, and they participatedin the developmentof standardschemes for designing,manufacturing, and servicingcomputers. In the receivedhistory of the early computerindustry, writers have assumedthat EckertMauchly ComputerCorporation possessed the capability and an advancedstate of knowledge,which shouldhave given them an edge. This edge was blunted by the inability to convince potential funders/purchasers.Authors have also assumed that IBM resisteddevices that wouldaffect the tabulatormarket, and when they realized it mighttake away their market share, they entered the market late with dedication and overwhelmedthe smallerfirms. EngineeringResearch Associates (ERA) is not evaluatedin this analysis,because it is seenonly as a military contractorrather than a competitorfor commercialbusiness. Many in the industryassume that

BUSINESS AND ECONOMIC HISTORY, VolumeTwenty-two, no. 1, Fall 1993. Copyright¸ 1993 by the BusinessHistory Conference. ISSN 0849-6825.

181 182

IBM waited until EMCC demonstratedthe techniquesand then enteredthe market. I believethe evidencedemonstrates a contraryposition, that is, that all thesecompanies struggled with computerideas in the early years,because theyneeded to learna greatdeal about how to implementdesigns and produce a digital computersystem. Each contributeddifferently to the process. What madeIBM andERA differentfrom EMCC was the resourcesthey coulddraw on. Within the company,IBM could supportR&D on computersbecause of their successfultabulator business. ERA had the supportof the U.S. Navy. EMCC had neither of these resources. The areasof the governmentand businessesEMCC relied on for supportwere cautiousabout when success would be achievedand held backsomewhat. In addition,EMCC had a range of problemsunconnected with machinedesign. The story goes that lef• to themselveswith the right financing,EMCC Wouldhave succeeded. If we examine these companiesmore closely using some criteria of Nelson and Winter, we see a differentand more complexhistory. Nelsonand Winter, in their studyAn EvolutionaryTheory of EconomicChange, bring out three areas of firm capabilitythat are particularlyrelevant to the historyof computing. Theseareas are the needfor clear and achievableobjectives, an adequatestate of knowledge,and choicesembedded in the talentsof the personnel. With respectto objectives,they notedthat

objectiveslike profit, market share,or growth do not serve to guide actionin the absenceof specificunderstanding as to how they are to be achieved. Unlessthis understandingis obvious, sharedby all thosewho are involvedin decisionmaking, even the deepestcommitments to a commonultimate objectivewill not serve to focus attention and coordinate action. To serve this purpose,objectives must be articulatedin sucha way that they are relevant to the decisionsat hand....choiceof operational objectivesis an importantarena for managerialdecision [3, p. 56].

None of these companiesdefined clear and achievableobjectives in 1946; they cameto them overtime. Overarchingobjectives were apparentto the participantsin computing,but operationalobjectives were not. We need to add to this criterion evaluationof the role of the customerin setting objectives. An adequate state of knowledge about computing was just developing. Indeed, it took a decadeto arrive at sufficientknowledge to developall the aspectsof constructinga stablecomputer system. Along the way, all the firms interestedin digital computingmachines struggled with pushingback the envelopearound the known into the unknown. And the accumulatedexperience of the personnelin these firms influenced the definitionof problemsinvestigated and the solutionsproposed. The stateof knowledgewas arguablyadequate in 1946, but becameso by 1950, and many groupsplayed a role in the definition. The talentsof the personnelin these groups,while impressive,needed honing. The searchfor a stablecomputer design by all involved inside and outsidethese firms is the key to understandingthis history. In the questfor 183 a stabledesign, emphases were placed on speed, storage, and reliability. Early computerdesigns were slow, had small storage, and were insufficientlyreliable to encouragefaith. Speed could be increasedin two ways: decreasing interactiontime of partsof the computer,such as reducing distances electrical signalsmust travel, and increasingthe amountof dataand instructionsstored insidethe machine. Thus, an array of researchprograms in companiesand universitiespursued techniques to increasestorage and decreasethe time to isolatea singledata element, keeping costs in mind in mostcases. This search culminatedin the developmentof magneticcore memory, which sweptthe industryin the mid-1950s. The reliabilityissue was, of course,twofold also. The need for reliability in componentsled to new developments.Industry developeda rangeof new componentssuitable for usein computers,including massproduction techniques for their manufacture. Designerscrafted more reliablecircuits for the transferof dataand instructionsinternally. In this paper, I concentrateon the two 1946 startupsin the computer business:Eckert-Mauchly Computer Corporation (EMCC) and Engineering ResearchAssociates, Inc. (ERA). EMCC beganwith a generaldesign for a computer based on the work of the principals at the University of Pennsylvania. ERA organized to design and develop data processing machineryto advancethe stateof the art beyondwartime devices. By 1951, each had a computerto market; both machineswere importantdesigns and contributedto the stabledesign that emergedin the middle of the 1950s,both contributedsubstantially to an understandingof speed,storage, and reliability for computers.In the developmentof thesefirms, nothing was routine, in the Nelson and Winter sense. After 1956, a short decade after the introduction of the ENIAC, however, use of digital computerswas spreadingand a digital computer industry was beginning to flourish. The computer stimulatedtechnical developmentsand modes of socialbehavior that made the computingenterprise into a major phenomenon.Out of this came a new world that, over the next four decades,and maybe beyond, containednew opportunitiesand new tensions. The seedsof theseopportunities and tensionswere presentin the efforts of the men and women working in this emergingenterprise in the decadebeginning in 1946.

The Companies

The storyof the foundingof the new digital computerindustry goes back to early 1945. Even duringthe ENIAC developmentperiod (1943-1945) Eckert and Mauchly were discussingthe design of future internally-controlled computersand of the prospectsfor commercialization.They slippedeasily into plansto build a commercialstored-program electronic computer, and in the springof 1946, establishedan enterpriseto do so. EMCC neverdeviated from the objectiveto build commercialmachines. A principallesson of the war had beenthe value of rapid deciphering of privatemessages of othergovernments. The Navy concludedthat more sophisticatedmachinery would be neededin a more complicatedpeacetime world. Decipheringhad been doneby a sectionthat was composedof a 184

hastilyassembled group ofcryptologists, mathematicians, physicists, engineers, andchess and bridge masters. The Navy's first orderof businesswas to try to keepthis group together. Foiled at keepingthe primegroup together after the war as civilianemployees t6 pursuesuch work underdirect supervision, the Navy assistedin the establishmentof a privatecompany, comprised of many of thosesame men, to performthe sameinvestigations with classified contracts. This companywas ERA locatedin St. Paul, Minnesota. ERA's primeobjective, though different than that of EMCC, wasjust assimply stated as thoseof EMCC: to build dataprocessing equipment for useby the U.S. Navy. In thesetwo cases,overall objectiveswere set through discussions betweenthe companiesand their customers; proximate decisions taken within the companyfocused on engineeringissues. What was not clear in the beginningwas how muchof a role the customerwould play in shapingthe product. Partly, this resultedfrom the attemptto bring a new, previously unavailableconcept to marketby small firms. In fact, customerinvolvement greatlyimproved the result. ERA influencedthe field in ways both similar to and differentthan EMCC. Among the similaritiesare major inventionsfor storagetechniques, commercializationin the early 1950sof a machineproduced originally for a military purpose,and as a fountainheadfor new companies.The principal differencesare the manner of operationof the company,that is, tight classificationin the earlyyears; production of a volume(published in 1950) that containedan assessmentof techniquesavailable in 1949 for designof computers that influenced developmentsworldwide; and a prudent manufacturingand marketingstrategy. Like EMCC, ERA alsohad a customerto satisfyand developed systems for the customer'suse. The companyevolved to meetcustomer needs in new ways. EMCC was on an almost continuoussearch for customersand investors.While they succeededeventually, as did ERA, in producinga good system,they had to emphasizethe developmentside over marketing and use. By contrast, IBM entered the businessmachines area in 1911 as Computer-Tabulating-RecordingCompany, a successorcompany to the TabulatingMachine Company of HermanHollerith. Between1911 and 1945, IBM haddeveloped a commandingposition in thetabulating machine business. Sincethe aim of companieslike EMCC was to replacetabulating equipment with electroniccomputers, IBM had a businessto protect. From IBM's perspective,all developmenthad to be evolutionaryfrom the business perspective,in essence,the customer'sperspective, regardless of how revolutionarythe technology.Beginning in the late 1930s,IBM developed electroniccalculators. During the war, IBM personnelassisted Harvard's Howard Aiken with the designand building of the AutomaticSequence ControlCalculator, a relay computer.After this project,IBM designedand built their own SelectiveSequence Control Calculator and, in 1946, organized an electroniclaboratory whose mission it was to investigateconcepts for electronicdevices. Proceeding in fits andstarts, this group enlarged over time and in the early 1950sdesigned two machinesfor two markets:the IBM 650 and the Defense Calculator,which became the IBM 701. Even though 185

customerswere consulted,IBM set its own objectivesthrough discussion insidethe company[1].

Developing the Knowledge Base

The presencein all these companiesof electricalengineers with experiencein advancedcircuit design,some with experienceon computer projects, of mathematicianswho addressedthe use of computersand participatedin the design of new techniquesfor problem solution on computers,and physicists with knowledgeof magneticand electric techniques for the developmentof storageand communication components for computers placedall thesecompanies at the forefrontof thefield. The EMCC grouphad an edgein thatmany of themhad participatedin designof a previousmachine. But the IBM and ERA staffsthat had contributedto wartime cryptologyand advancedelectronic devices were not far behindthe peopleof EMCC. We turn now to descriptionsof how they appliedtheir skills to the designof new computer systems.

EMCC

During late 1945 and early 1946, Eckert and Mauchly engagedin discussionswith various government groups at the National Bureau of Standards(NBS), the CensusBureau, and the Navy aboutthe possibilityof buildinga computerfor one or more of theseagencies. NBS conducteda surveyof developmentamong various projects and sought advice from George Stibitz, an early computerdeveloper at Bell Laboratories.Stibitz concluded that the circumstanceswere not readyto proceeddirectly to constructionof a computer. Further research,he thought, was essential. Among other recommendations,Stibitz suggested a phased approach to acquisition--research anddesign; building and testing of components;and construction of a machine. In June1946, NBS submitteda proposedcontract along lines suggested by Stibitz. The contractwith NBS, acceptedby EMCC in September1946, called for the companyto "supplythe necessaryqualified personneland facilities for and prepareplans, specifications,and wiring diagramsfor automaticallysequenced electronic digital computingmachine or machines suitablefor generalmathematical computations and for preparationof census reports,and constructand test such modelsof componentsas may, in the opinionof the Contractorsor ScientificOfficer, be necessaryto insurethe adequacyof theseplans and specifications" [7, Folder 117]. Eventually,this was understoodto mean the constructionand demonstrationof a delay line memorydevice and a magnetictape input/outputdevice. The task had been divided by EMCC into six parts. (1) Memory: besidesthe mercurydelay registers, the groupinvestigated other methods, such as beam deflectiontubes. (2) Electroniccircuits for arithmeticand control units:here, also, an arrayof'circuits suitable for severalmemory systems were under investigation.(3) Magnetictape materialsand readingand recording heads. (4) Key-operatedinput mechanismsand automaticoutput printers. EMCC decidedthat already designed equipment would be the bestsolution to 186

problemshere. (5) Servodevices for controllingtape. (6) Logicaldesign: the groupnoted only that specialcontrol features that mightallow morecompact and efficientproblem coding were to be devised. Eckert and Mauchly estimatedthe developmentcosts to be between $413,000 and $671,000,representing the minimumand efficientrequirements [9, Box 4a; 4, 267]. The two partnersassumed they would be able to obtain other contractsfor sale of the machine that would provide additional developmentfunds. Amongthe manycontacts made by Mauchly in late 1946 and early 1947, four standout becauseof later contracts:Army Air Forces, Northrop Aircraft Company,A. C. Nielson, Co., and the PrudentialLife InsuranceCompany. For example,toward the end of 1946, Mauchlybegan conversationswith the NielsonCompany, and on 4 January1947 he senta proposalfor the constructionof a groupof "electronicstatistical machines" to them. In early April, Mauchly was hired as a consultantto the Northrop Company. Northrop,and North AmericanAviation as a subcontractor,were workingin the Snarkmissile program of the Army Air Force. Mauchlyspent ten daysfrom 5 to 16 May at the Northropfacility in Hawthorne,California. Each day he spenttwo-hours "instructing a small group on the subjectof electronicdigital computers." The purposeof the instructionwas to informthe groupon whathad been accomplished in the field so far, and"to explorethe possibilityof applyingsimilar techniques to controlproblems encountered in the project work" [7, Folder 330]. By late June, Northrop had entered negotiationswith EMCC for a computer"to prove the feasibilityof a particular methodof navigation."This machine would be calledthe BinaryAutomatic Computer(BINAC), and in the mindsof Mauchlyand Eckert,would serve as a prototypefor the largermachine to be designedfor the NBS. At about this time, Mauchly approachedthe Prudential insurance Company.The time waspropitious, because the company'sEdmund Berkeley had only recentlycompleted an evaluationof the computerfield. By mid- December,he had talked with people at many organizationsinvolved with calculatingmachinery, and he had concludedthat the problemsof the Prudentialcould only be handledon a general-purposedigital machine. Eckert and Mauchly prepared a two-part proposalfor the Prudential on the "Applicationof High SpeedComputing Machines to CertainProblems of the PrudentialLife insuranceCompany" [7, Folder 18]. The proposalwas about equallydivided between a discussionof the problemsfaced by the Prudential thatwere susceptible to solutionby an EDVAC-typemachine and a description of the machinesystem. A detailedconsideration of the Prudentialproposal sheds light on at least four areas. (1) The proposalrevealed EMCC's progressin designof applicationsand assessmentof the machine'scapability, which supplements informationin the reportssent to NBS. (2) It provided evidenceof the considerationsthat slowedEMCC's progresson the NBS contract. (3) The proposalcontains a view of EMCC's emphasison applications.And (4) the role of the customerin helping to set objectivesis fairly explicit. While Eckertand Mauchlyevaluated the Prudential'stasks and designedways that their machinecould handle them more efficiently,and how the machinecould be modifiedto do so,they hadthat muchless time to completethe designon 187

the machine. But if EMCC had not expendedthis effort on applications,the entire processof machinedesign might have been slowedanyway due to insufficientdevelopment funds. The first part of the proposaloutlined the way the machinewould handlethree principalactivities of the company:premium billing; mortality studies;and group insurance. For each activity, EMCC describedthe machine'sapproach, the numberof peoplerequired to use the machine,the costto the companyof the basicservice, including the cost'of the machine, and the cost to convertthe punchedcard and other files to magnetictape. They emphasizedthat the cost of conversionwas a one-timecost. The machinerycosts came to $373,500,which could be reducedto abouthalf under a schemeto distributedevelopment costs. The secondpart of the proposal containeda detaileddescription of the ordercode to be usedin the machine, called Code C-2 (dated 7 May 1947). Includedwith the code plan was a general descriptionof the componentsof the machine. EMCC presented detailed design specificationsof the tape systemand how data would be recordedon the tape (with no more than 20% lossof space),the typewriter operationand speed,the contentscapability of delay-line memory, and methodsfor checking. They alsooffered some comments on the mannerof operationof the controlcircuits. While EMCC preparedtheir report to NBS in June 1947, Mauchly askedfor anotherextension of the NBS contract,to 1 August1947. Ten days later, Mauchly proposeda specificwork plan for the addedtime, which involvedtesting the reliabilityof variousreading and recordingmethods for magnetictape, doing such tests on a full scale, model tape transport mechanism,and evaluatingthe mercurydelay line storagedevices using the pulseenvelope system. Time wouldalso be devotedto designof high-speed flip-flopand gated oscillators for regeneratingsignals in mercurydelay lines [8, Box 1]. NBS agreed,and the contractwas extendedonce again, this time to 15 August 1947 [6, Exh. 6024]. So in August 1947, EMCC was bringingthe NBS machinedesign contractto a closein the hope of obtainingthe contractto build the machine for the CensusBureau; accepted a contractwith the Prudentialto designa similarmachine for them(most of the problemhere was applicationcodes and demonstrationof equipmentneeded, as we have seen);and accepteda contract from Northropto designand build a differentbut somewhatsimilar computer (this one by 15 May 1948). While all thiswork on hardwareproceeded, Mauchly and Betty Snyder tried to perfect and extend the capabilitiesof the code. Several new codes were transmittedto the Prudential--C4and C5, and of a numberof visitsby Prudentialpeople to EMCC to discussvarious ways to codeproblems and to convertpunched card data to magnetictape [7, Folder23]. Other technicaldesigns were under investigationand, in mid January 1948, EMCC filed a disclosurenotice for an electrostaticmemory system designedby HermanLukoff. This was part of Eckert'splan to investigateas manymethods for memoryas possible.The systemLukoff designedcontained a CRT, timing and deflection circuits, a regenerationunit for memory operation,and power supplies[6, Exh. 6296]. 188

Work on BINAC beganin earnestin the late springof 1948. Delivery had beenspecified in the contractas 15 May 1948. Towardthe end of May, A1Auerbach reported that the designsof the computer,memory, power supply and controlbox were each90 to 100% complete. Only the computerand the memorydrafting had beendone and most of the materialfor constructionof theseunits had beenacquired. Each was abouthalf constructed.Nothing else had been done on the other units besidesdesign [6, Exh. 6470]. Lukoff reportedthat mostof the engineerswere assignedto the BINAC projectat this time. Lou Wilson, AI Auerbach,Jim Weiner and Bob Shawwere the major designersof the "CPU." Brad Sheppardand Lukoff were responsiblefor the tape-loaderinput device. And PresEckert was everywhere helping with design decisions[2, p. 77]. The machinewas delivereda year late in mid 1949, and work resumedon the UNIVAC. By this time, negotiationwith customerswas complete;all design characteristicshad been approvedby the several customersand componentshad been testedsatisfactorily. But satisfactory componentsand a completelyfunctioning system proved to be two different things. Many moredesign changes were made to the UNIVAC systembefore it was deliveredin 1951, but the machineperformed well for the Census Bureauand for the othercustomers who receivedsystems in the 1950s.

ERA

In its first proposalto the Navy, ERA focusedon threeareas for investigation. First,ERA wantedto makean assessmentof the type andnature of problems arisingin the Navy for solutionon data processingmachines. This type of investigationwould generate information about the similaritiesand differences amongproblems with an eye towardreformulating them to fit the machines "recentlydeveloped" and understandingthe requiredaccuracy and amount of storage.The problemsnot solvableby knownmachines could be usedfor a secondarea of investigation:decisions about "the direction of developmentof the computingmachine art." The third areawould involve ERA in the design and constructionof the variouscomponents needed for new solutions[2, p. 2]. The proposalwent on to discusswhat ERA saw as the first problemneeding attention:storage. "The storageproblem is oneof the mostdifficult in building computingmachines" [2, p. 5]. ERA receivedits first contractfor suchwork in February1946. The contractdid not coincidewith the areasof the proposal,but involveda series of projects.Since it tookalmost five monthsfor personnelto be releasedfrom the service,it was only in summerthat the work underthis contractwas well underway[9]. With the beginningof work on this and othercontracts from privatefirms, ERA was a broaderbased company than EMCC. Functionally, ERA developedfour productareas, classified as areasA, B, E, andN, two of which evolved from the above contractwith the Navy. The A group (8 projects)addressed aviation problems, such as a groundspeed recorder and a parachutelanding shock reducer, mostly contracts for varioussections of the U.S. Air Corps. The B projects(4 in numberat this time) includedseveral for Navy and Army agencies. Anotherset of airline projects(E, 11 in all) were soughtby the presidentJohn E. Parkerto stimulatecash flow. These 189

rangedfrom broadcastconsulting to plansfor an airlinereservation program, many of whichwere in the originalprospectus for ERA. And therewere 13 N projectsfor the Navy with suchcolorful namesas Celery, Alcatraz, and Orion [9]. In laterhistories, ERA becamenoted as a resultof developments in two of theseproject areas: B and N, and we will focuson aspectsof them here. Another contract to ERA from the Office of Naval Research also took effect in August1946 [10]. This contractcontained three objectives: (1) a surveyof the computingfield similarto the oneproposed by ERA, including an analysisof problemsto be solved,the componentsand equipmentneeded in the solution,and a formulation,as explicitas possible, of the requirements for new componentsor techniquesneeded for the solutionof navalproblems; (2) Researchlooking toward the developmentof thesenew componentsand techniques,such as the availability of multiplexingtechniques for use in storageand transmissionof data;and (3) the furnishingof consultingservices to the Office of Naval Researchon questionsconcerning the developmentand applicationof computingequipment and techniques [9]. This contractcalled for ERA to examineeverything other projects were trying to accomplishand to designa storagedevice with multiplexedinput and output. It wasfrom this clausein the contractthat ERA's emphasison magnetic storage was to emerge. In a few months,from sometimein mid-1945 to March 1946, the ERA people,Howard Engstrom, Charles Tompkins, Lawrence Steinhardt, and John Howard,in particular,had analyzed this new field andreached the level of the Eckert-Mauchlyand the Institutefor AdvancedStudy yon Neumann/Goldstine groupsin an appreciationof the natureof the problem. Eachof thesegroups went in separateways, of course.But this proposalshows that the ERA team wasjust asthorough as the othertwo in identifyingthe natureof the problem and carvingout a promisingdirection. Next ERA had to work with their customerson designspecifications and neededfunctional parts, and, through testingof components,learn the art of computersystem construction. By June1946, ERA hadorganized several projects related to the search for a better data storagesystem. One of thesewas to analyzethe use of photographicfilm as a potentialtape source. Another was to examinesolid- state delay lines as a storagemedium. Yet anotherfocused on the use of magneticmedia as a storagesource, the primaryarea of personneltalent. In the beginning,these projectswent under the codenamesOrion or N-1011. Eventually,ERA emphasizedonly magneticmedia, even thoughunder the Navy's urgingthey continuedto investigateother possible storage media. The machine to incorporatethese ideas was codenamedGoldberg. Goldbergwas analogousto Binac in the role it playedin ERA's path to the 1101 system. (IBM had its own analogueto Goldberg/Binac:the Tape ProcessingMachine (1949-1951).) Goldbergwas designedto analyzedata of a "teletypenature" at a rate of 20,000 pulsesper second.There were to be six functionalparts to the Goldbergmachine. (1) An IBM tape readerwould be used for punchedtape input and outputto the machine. (2) Two magnetic drumswere to be constructed,one for internaland one for externalmemory. The internalmemory drum of 31 inchesdiameter was to have 22 or more tracksand store 5000 magnetized"spots" around its circumference.The Navy 190

specifiedthat the data on this drum shouldbe analyzedat a rate of 20,000 itemsper second. Three magnets would be employedfor reading,erasing, and rewriting. The calculationalpart of the machinewas composedof translatorsand counters.(3) Goldbergwas to havethree types of translatorsfor converting data from one systeminto binarycode. The complexityof thesetranslators went from two-unittranslators with two inputsand four outputs,to six-unit with six inputs and 64 outputs,to a 36x36 matrix. (4) There were 36 electronicpulse counters, each with a maximumcapacity of 9999, controlled by a gatetube in the inputsignal circuit. (5) Four controlsignals for timing, starting, stopping,and canceling,and it was necessaryto provide control circuitsfor these. In addition,the plan specifieda set of controlcircuits for the functioningof the counters,reading schemes, and printingresults--a total of nine circuitsin all. And (6) a mechanicalprinter was to be suppliedhaving a capacityof 38 five-digit numbers[5]. During May, the final designsfor the readingand re-writingamplifier and the electronicmemory units were completed,and productionprototypes were under constructionby the end of the month. But other circuit design changeswere being undertaken,especially in the carry-overchassis. The continuousread, erase and rewrite featurewas successfullytested for durations of up to 90 minutes. "No deteriorationof the patternwas observed."Other circuitswere in variousstages of design. A comprehensivestudy of control circuitsand equipmentbegan. Much work still remainedto be doneover the next sevenmonths--the projected completion date of dram construction[5]. The sortingout and educationprocesses that occurredfrom mid-1946 to mid-1947, whenGoldberg was delivered,allowed ERA to learnthe field of computing, accumulatethe engineeringskills to pursue research and developmenton an organizedand sustainedbasis, and to carve out for themselvesan area of computingmachine development that would have an importantinfluence on the field. Thu•, the critical years in ERA's growth were 1947 and 1948. In theseyears they changedfrom an electronicproject shopto a computerdesign company. In October1946, Tompkinswrote a memorandumto William Norris aboutfive probablesources of computingmachine business. Besides ONR and NBS, he cited the United StatesArmy, PrudentialLife Insurance,and the University of Chicago [10, Box 7]. In spite of the awardingof the ONR contractin August 1946, containingas it did requirementsfor a survey, servicesand bread-board models, ERA was still tryingto obtaina contractfor a larger computingmachine project. During 1947 underthe ERA contractwith ONR to surveycomputer projectsin the United States,Tompkins, Cohen, and othersvisited a number of installationsto discussthe work of thoseprojects and other subjectsof mutual interest. For this work, in 1947 they visitedMIT (Whirlwind), Brown University (the solid-statedelay line work of Arenberg),the University of Pennsylvania(EDVAC), Harvard (Mark II and III), and knew about developmentsat the IAS. ThroughNBS theyobtained the instructioncode for Univac [9]. From theseconsiderations came ERA's computerdesign project labeled B-3001. 191

Conferencesfor this B-3001 projectstarted toward the endof February 1947, apparentlybetween Tompkins and ONR personnel. By 10 March Tompkinshad met with Arnold Cohenand GeorgeHardenbergh to discuss "thepreparation of the list of componentsfor computingmachines...and to start additionalresearches concerning magnetic storage of pulses." Three groups would do the pulse work, led by Coombs, Rubens, and Gutterman, respectively. ERA developeda twelve-pointprogram concerned with head,media, circuitand system problems. Further work wasto be doneon headshape and size,lamination for improvingthe efficiencyof high-frequencyresponse, the sizeof gapsbetween the headsand the magneticsurface, and the materialsout of which the headwas made. Possiblenew researchon mediacomponents wouldbe balancedwith researchon placementof thesecomponents on drams. By the beginningof September1947, in all its essentialfeatures the work on the magneticstorage system was completeand ERA had achievedthe objectivesset out in the contract[5]. With the completionof this project,John Howard recommended that "ERA hold immediatelya large-scalesymposium for all personnelwho might have ideas as to possibleapplications" of the storagedevice. Seventeen membersof the engineeringstaff attendedthe meeting. The ideas for use offeredby othersat the meetingwere not very muchadvanced over the first descriptionof productscirculated at the time of ERA's founding. They had to do with masscommunication problems, as a consulterfor wage rates in connectionwith punchedcard machines,store credit informationfor quick reference,use in storingdata for mapsand weather,and rapid recordingof experimentaldata. One of the more intriguingsuggestions at the meetingby Howard was the couplingof the dram with an IBM machine[9]. Howardreceived so many suggestionsat the meetingthat he was able to assemblethem into fourteencategories in a twelve-pagememorandum to Ralph Meader preparedon 8 March [9]. The categoriesseem to be rankedin orderof importance.Seven ideas were groupedin a categoryon "sequence control applications." These ranged from automaticcontrol of looms for patternweaving and of machinetools to automaticcontrol of guidedmissiles and airplanes.This latteridea was alreadypart of the analysisby Tompkins of the needsof the variousmissile development groups of theNavy aspart of the componentssurvey. In the "dynamicinventory applications category," we find airlinereservations systems, large-scale inventory problems, the census, and automaticaccounting systems for banks. Over the next two yearsmany of theseideas were presented by ERA to corporationsand the government with limited success. ERA's aim to build a computingsystem had beenachieved, however. By thistime, ERA washeavily involved in completingthe components survey for ONR, deeplyimmersed in a designof a majorcomputing system for the Navy, the systemto be known as the ERA Atlas, and in its commercial version,the ERA 1101. They were alsonegotiating with NBS for designof yet anothercomputer based on the needsof the U.S. Air Controlleftsoffice. Thesetasks precluded shifting personnel .to an investmentventure of the firm at the expenseof assuredcontracts. 192

With Atlas, ERA went throughthe samebuilding and testing steps as EMCC. After considerableinteraction with the Navy, ERA deliveredthe Atlas in December1950. With someminor modificationsand permission from the Navy, ERA offeredthis machinefor salethereafter as the 1101.

Conclusion

From this brief history,we learn first that the startingpoints for the three companieswas the same: they all emphasizedstorage systems developmentfirst. For ERA, the storagesystem was an end in itself, but it developedinto a projectfor a completemachine--the Atlas andthe 1101. For both EMCC and IBM, storagesystems were startingpoints in a longer programto producea computer. None of the companies,indeed no project anywhere,had a sufficientknowledge of how to build functioningcomputer storageelements or any otherpart of the machineto ensuresuccess. Other authorshave called attentionto the transferof informationamong all the projects. This played a particularly importantpart in ERA's activities. Informationtransfer was also significant to IBM, but lessso to EMCC, though they did benefit from attendanceat the variousconferences held in the late 1940s. Second,the strong interactionbetween designersand users in the determinationof machinespecifications and functioningability hasnot been examinedgiving equal weight to both setsof people. Users demandshave beentreated in the literatureas reactionary,whereas designers ideas have been portrayedas revolutionary. In fact, the roles of NBS, the CensusBureau, and the PrudentialInsurance Company in EMCC's designand planningwere criticalto bringinga functionalmachine to market. EMCC's contractswere designedby them to convergeand at the sametime meet customerdemand. The view in 1946 that such a device could be constructed from standard radio partsdid not preparethe designersor the anxioususers for the large number of obstaclesthe designersneeded to overcomebefore an operatingmachine becameavailable. Many designchanges occurred along the way, and a numberof componentsand test instrumentsneeded to be newly designedand built for properperformance in computersystems. The U.S. Navy played a different role for ERA. Navy contracts provideda systematiclearning and R&D periodfor ERA, duringwhich they could advancefrom smallerto largerscale system contracts. While IBM's start may be consideredslow from some perspectives,the time frame for bringing a machineto market is about the same as for EMCC and ERA. IBM's objectiveswere quitedifferent than those of EMCC, eventhough they were somewhat similar in their end--to obtain commercial customers for the new machinery. IBM's approachto the developmentproblem was quite differentthan EMCC's, andin somerespects different than ERA's. In at least oneinstance, IBM's R&D programconverged with ERA's andIBM contracted with ERA to pursuedesign of magneticdrams for use in IBM machinery. This contractheld out the possibilityof significantcommercial business for ERA, and might have spelled the difference between independenceand absorption,except that IBM decidedto go it alone and finance their own 193

development. This decisionwas reachedafter much discussionwithin the companyabout customer needs and market potential. Third, eachcompany experienced about a five-yeartime from startof design to delivery of a computersystem. EMCC began first in 1946 and deliveredthe BINAC in 1949. The BINAC, whichincorporated many features prominentin the laterUNIVAC, wasdelivered to NorthropAviation in mid- 1949. UNIVAC 1 was deliveredto the CensusBureau in early 1951. ERA received a contractto designand build the Atlas in 1947 and delivered the machineto the National SecurityAgency in December1950. Its commercial counterpart,the ERA 1101, was first deliveredin 1953. Developmentat IBM on the 701 and the 650 tookjust aboutas long. The IBM 650 project,begun in 1948, assumedsustained activity in 1952 and produceda machinein 1954. The DefenseCalculator appeared in 1951, and its commercial version the IBM 701 reached the market in December 1952. The IBM systemsdesigns were more complete in that they considered manufacturingrequirements and customerservice needs. The routineof the IBM tabulatorbusiness was transferredin so far as possibleto the makingof computer systems. EMCC and ERA developedbetter processesfor manufacturingand servicinglater, as evidencedby the successionof modified models of the UNIVAC and the ERA 1101.

References

1. CharlesJ. Bashe,Lyle R. Johnson,John H. Palmer,and EmersonW. Pugh,1BM's Early Computers(Cambridge, MA, 1986). 2. HermanLukoff, From Dits to Bits:/1 PersonalHistory of the ElectronicComputer (Portland, OR, 1979). 3. RichardR. Nelson and SidneyG. Winter,An EvolutionaryTheory of EconomicChange (Cambridge,MA, 1982). 4. Nancy Stem, From Eniac to Untvac(Bedford, MA, 1981). 5. Arnold A. CohenPapers, Charles Babbage Institute, University of Minnesota. 6. HoneywellCollection, Charles Babbage Institute, University of Minnesota. 7. JohnW. MauchlyPapers, Van Pelt Library,University of Pennsylvania. 8. National Bureau of Standards,Record Group 167-76-67,National Archivesand Records Agency. 9. SperryCorporation Records, Hagley Musuem and Library. 10. YutterCollection, Hagley Musuem and Library.