Pehr Anderson, Wenkai He, Yao Ma, Brian Slutz, Ken Lynch
In The Structure of Scientific Revolutions, T. S. Kuhn proposes a model to explain the general pattern of scientific revolutions. In the Kuhnian model, science enters into a stage of normal science following the establishment of a paradigm (e.g., the Newtonian Physics). A paradigm in science serves three functions: 1, to provide a metaphysical worldview; 2, to provide analytical tools such as equations, criteria of problems and solutions, and unarticulated rules of game; 3, to provide an institutional system to recruit and train new members. Normal science is thus a process of accumulative problem-solving process. However, after many anomalies accumulate the paradigm begins to lose respectability and belief. People begin to actively search for a substitute to the old paradigm. When most members of the younger generation shift to a new paradigm, a revolution in science happens. After a successful revolution, a new kind of normal science makes its progress under the leadership of the new paradigm [Kuhn].
Is the Kuhnian model of scientific revolutions applicable to engineering revolutions? We try to answer this question in this essay by studying the history of a concrete case of engineering revolution, the magnetic core memory. As a new paradigm in computer memory technology which bases upon using the hysteresis loop of magnetic materials to store binary bits digitally, it distinguishes itself from the old cathode ray storage-tubes paradigm (including the Williams storage-tubes, the Selectron storage-tubes, and the electrostatic storage-tubes) which employs the analog-state cathode ray to store binary bits in the parallel-mode electronic digital computations. From the mid-1950's to the mid-1970, the magnetic core memory is the most important memory devices in the compute memory technology. Its inventor Jay W. Forrester is also elected to Inventors' Hall of Fame.
Emerson Pugh has studied the history of magnetic core memory from the perspective of its internal technical aspects. Paul Edwards has studied the Cold War social-political environment of Forrester's Project Whirlwind in which the magnetic core memory is invented to solve the problem of storage. Kent Redmond and Thomas Smith have studied the history of Project Whirlwind based upon intensive studies on the achieves of Project Whirlwind. However, Pugh's study does not give sufficient attention to the social-political factors and their effects on the invention of the magnetic core memory. Edwards's external approach fails to reveal how the technical aspects of the magnetic core memory are affected by the social-political factors. Redmond and Smith's efforts can be described as a kind of "retrospective reconstruction" from the perspective of the successor, Forrester. They under-evaluate the degree of uncertainties in technical inventions and over-evaluate the role of inventors' technical visions. In their story of Project Whirlwind, Forrester's shift to digital computation and his shift to the magnetic core memory are attributed by Redmond and Smith to Forrester's technical visions. However, technical visions are highly personal things yet an engineering project like Project Whirlwind is involves many relevant actors such as military agencies which do not share Forrester's technical visions. Redmond and Smith's Whig-style history makes them blind to the discrepancies in their own materials (see details in Section III).
In our study, we try to integrate the internal history of the invention of the magnetic core memory with its external history. We will argue that the Cold War military demands for the real-time control systems (e.g. the SAGE Project) not only provide a rich research and development resources and a huge procure market for the magnetic core memory, but also affect directly Forrester's randomly accessible 3-D array of the magnetic core memory. Meanwhile, in order to bridge the gap between the personal character of technical visions and the social character of an engineering project, we employ the term "heterogeneous engineering" that is defined by Donald MacKenzie as the "engineering of the social as well as the physical world." [MacKenzie, p.28.]
We recognize the accumulativeness of engineering efforts in both the old cathode ray storage-tubes paradigm and the magnetic core memory paradigm, which is analogous to the accumulativeness of the Kuhnian normal science. Yet we find in our study that the Kuhnian paradigm-shift is unable to explain the complexity of the transitional period from the old paradigm to the new paradigm. In fact, Margaret Masterman has pointed out that the Kuhnian paradigm -shift is contradictory to his definition of paradigm [Masterman]. In order to explain the transitional period, we introduce the term "magnetic core culture," by which we mean a set of knowledge, skills, ideas, and devices based around using magnetic core materials to store binary bits. [we get this term from discussions on "gyroscope culture." MacKenzie, p31.] Members of the magnetic core culture include: 1, manufacturers of magnetic core materials such as the Arnold Engineering Company; 2, Forrester at MIT, Jan Rajchman at RCA, An Wang at Harvard, and Mike Haynes at IBM, etc. We will argue that Forrester's entry into the magnetic core culture in June 1949 and the military resources backing his Whirlwind project (later become part of the SAGE project) stimulate greatly the development of the already exist magnetic core culture which culminates as a paradigm in computer memory technology.
An Interview with Jay Forrester
Jay Forrester leads the "wig history" of the invention of core memory. An extremely active man, Forrester's secretary organizes his workday into 15 minute increments so he can squeeze as much time out of the day as possible. Our interview with Forrester gave us insight to his aggressive personality traits that were instrumentally responsible for the way the history of core memory has often been portrayed.
We found Forrester's character to be very charming. His stories were very entertaining, even with himself always as the hero. We are not sure if this self inflating personality trait has always been with him. Perhaps it is a product of his aloofness that both Perry Crawford and Bill Papian described; his absorption in his own work has led it into being his favorite subject.
In one story, Forrester unlocked a Spartan looking cabinet to withdraw a folder that had the phrase "TOP SECRET" stamped across the front (it had subsequently been re-stamped with the word "DECLASSIFIED"). Opening this white folder, he pulled out a folded sheet. Forrester carefully unfolded it and covered his entire desk to show an enormous grid with every box filled. He explained that with this sheet, in 1947 he presented to the military how much manpower was required to develop computers over the next 15 years. Each column equated to a year and each row was dedicated to each aspect of the military that computers would change. The way he derived the numbers was by making the assumption that the computer was going to be bigger than radar. Because his assumption was correct, and he knew how much radar had cost to develop, Forrester was surprisingly accurate in his forecast.
When asked about von Neumann and his ENIAC computer, Forrester exclaimed his admiration for the famed mathematician. But he then checked his own enthusiasm with the recollection that his Whirlwind computer was much more robust (which of course is what he was after from the beginning). He turned the conversation to the recollection of an MIT dinner he once shared with von Neumann on the left and Norbert Wiener on the right. Forrester contrasted their personalities; von Neumann being able to speak on any subject and Wiener choosing to speak on topics that only he can usually dominate. Perhaps this attack on this well respected Nobel Laureate was just casual conversation, but it also revealed his rivalry with his well tenured colleague (a senior position that Forrester was not able to attain in Wiener's department).
In his last comments, Forrester expressed that he was unsatisfied with the fact that historians wait until the participants in history are no longer around. He accused that historians do this to be able to force their own "historical opinions" and views about the events with little opposing debate. It was an interesting point. When historical participants are dead, it makes it significantly harder them to rebuff bold assumptions and more interesting conclusions created by the historian.
We believe the reason Forrester made these comments was to break down any historian's conflicting view of the history he is a part of. It is the reason why he gives academic talks promoting himself as the hero inventor of core memory (such as his core memory symposium that was given at Northeastern University) without giving much mention to the rest of the core culture that he was late to join.
Jay Forrester's aggressive personality trait of ripping down opposition was extremely important to the success of Project Whirlwind. The trait can also be found in Stark Draper, for example, when he defeated Gamaw (Gamaw raised the "problem of the vertical", a serious problem with inertial guidance that could be equated with the problem of memory in the Whirlwind computer) (p.72, Mackenzie). Had Forrester not possessed this personality, he would not have survived the yearly "inquisitions" of the ONR, and his project would have become a failure. The commonality of this personality trait could be why a non-collaborative history is so often told.
Because Forrester is such an excellent debater, the fame of core memory falls into his possession. Forrester held the trajectory of the developing Project Whirlwind steady, fending off opponents and rolling up his sleeves to tackle reverse salients, such as the memory problem. Meeting Forrester was insightful, allowing to gain an appreciation for why this man has been able to aggregate the credit for the invention of the computer, when so many people played a part in it's development.
The Rise of Core Memory Culture
The invention of magnetic core memories is important in the history of technology because it drove the first wave of the computer revolution. The common story associated with this technology is that Jay Forrester, while leading Project Whirlwind at MIT, invented magnetic core memory in 1949. However, if a more inclusive history is studied, it will show that many pioneering researchers contributed in some way to this invention. Starting as early as 1945, the idea of using magnetic cores to store information was published, and by 1948, storing bits of information in magnetic cores was actually achieved. By taking a closer look at these early attempts at creating magnetic core memory, it will be evident why they were not were able to have the same impact as Forrester's invention. The interesting issues from the perspective of a historian is why the early attempts failed to be revolutionary, yet at the same time, how they contributed to the emergence of a magnetic core culture. This culture, to borrow MacKenzie's definition, is defined as a set of knowledge, skills, ideas, and devices based around using magnetic core materials to store binary bits. [MacKenzie, pg 31]
An inclusive history of the invention of magnetic core memory would include John Presper Eckert, An Wang, Jay Forrester, Jan Rajchman, Mike Haynes and William Papian. Eckert was an engineer that contributed heavily to the ENIAC project, the first electronic computer developed. Wang, as a researcher at the Harvard Computing Lab in 1948, was the first person to implement magnetic cores to store information. These two researchers contributed to the early attempts at creating magnetic core memory that led to the rise of a magnetic core culture. By 1950, Forrester, Rajchman, Haynes and Papian joined this culture and contributed to its ultimate success. Forrester, the leader of Project Whirlwind, adopted magnetic cores in 1949 to meet the memory needs of the project. His ideas on memory were implemented through the research efforts of Papian. Rajchman, an engineer at RCA, developed his ideas of using magnetic memory in parallel with Forrester and engaged in a long patent suit with him over the invention. Haynes, as a Ph.D. student at the University of Illinois, applied magnetic cores to implement logic and later established the magnetic core research efforts at IBM, the company that ultimately benefited the most from the invention. These researchers all joined and contributed to the magnetic core culture that was solidified through the success of Project Whirlwind.
In 1945, John Presper Eckert proposed storage of information in a 2-dimensional array of magnetic cores. Eckert was a pioneering engineer with experience designing and building computers and understood the crucial issues for memory. While working on the EDVAC, Eckert published a paper proposing to use a 2-D array of toroidal ferromagnetic cores. This paper was widely circulated in the computer research community and Eckert told a number of people about his idea on magnetic core memories, including Jay Forrester. [Pugh, p.89.] Rajchman also admitted that this idea was communicated to him through another person who had talked with Eckert. Thus, even though Eckert never implemented the idea, he significantly influenced much of the later developments in this area.
Eckert was unable to pursue his ideas on magnetic core memory further because his research focus became directed to the commercial market. In 1946, because Eckert was unable to get appointed a professorship at the Moore school, [Goldstine, p240], he formed one of the first computer companies. Along with John William Mauchly, another researcher at the Moore School of Engineering, he started the Eckert-Mauchly Computing Company (EMCC) to build computers for the commercial market. Because of the need to sustain the company, the funding devoted towards research could not nearly match those provided by the military for ENIAC or Project Whirlwind. Also, since the target was towards civilian use, the market was smaller and the tastes were for the more conservative approaches. As an example of the conservative approach of the commercial marketplace, EMCC had to agree to provide equipment to convert data stored on magnetic tape to punched cards in order to obtain a contract from the Prudential Insurance Company. Thus, given the new environment in which he was working, Eckert had to focus on the priority of building a company and could not afford to implement his pioneering ideas on memory.
In 1948, An Wang implemented the first magnetic cores to store information in a computer. Wang was a researcher at the Harvard Computing Lab under the guidance of Howard Aiken, a pioneer in the field of computers engineering. Aiken gave Wang a specific problem to solve for the Harvard Mark IV computer, which was the lab's first attempt at building a purely electronic computer. Aiken wanted Wang to devise "a way to record and read magnetically stored information without mechanical motion." [Wang pg 52] Wang figured out how to store and read information using magnetic cores and created a serial method of accessing and writing the bits of data, much like a serial delay line. Although this serial method was used in the Mark IV computer, it wasn't a revolutionary change in computer memory design. A main reason was that because it was necessary to cycle through the series to get to a particular bit, as the memory becomes large and larger, it becomes increasingly cumbersome to access the memory. Even though his specific application of core memory did not survive, the concept of using individual magnetic cores to store bits became the dominant basis of memory for the next twenty year.
Wang's attempts at using magnetic core memory fell short because of his background and because of the research environment. When he created the solution to the memory problem, Wang had only been working at the Harvard Computing Lab for two months. The problem that Aiken gave to him was very specific and only required an understanding of physics and electronics engineering, not of the logical architecture of digital computer. Wang didn't understand the larger issues of memory in a computer system as well as Forrester did at the time and thus didn't have the same insights as Forrester. It is possible that if Wang had pushed his invention further, he might have come to the invention that Forrester reached. Wang hindsightly regretted not pushing the idea further to explore potential extension of the core memory concept. [Wang 60] However Forrester's painstaking process of invention indicated that the step forward was not as easy as Wang has supposed. Another significant hindrance to Wang's efforts was Harvard's policy on "not pursuing research in developing technologies once they had matured to the point where they had commercial applications." [Wang 61] Thus, by late 1949, it was apparent that the Harvard Computing Lab would no longer build computers and Wang left to start his own company in 1951.
Comparatively speaking, Forrester was luckier than Wang. For it would be easier at MIT to do engineering research about the electronic digital computers, a work as important as theoretical research to the nascent digital computer yet was despised by the high-browed Haarvard as research about commercial applications. How did Forrester come to work on storage devices of binary bits?
These two early attempts were the main contributions leading to the rise of magnetic core culture. If a single event could be marked as the birth of magnetic core culture, it would be the symposium at Harvard in September 1949, at which An Wang presented his invention. This event led to Haynes's interest in magnetic cores and his Ph.D. research in using them to implement logic. Later, as a researcher at IBM, he led the company into the magnetic core culture by implementing a 2D array of magnetic cores and starting the research efforts there in this area. The symposium also allowed Forrester an opportunity to meet with representatives from companies that provided magnetic cores and initiated the dialogue into the materials science research. The main significance of this symposium was that it was the first opportunity for a collection of researchers to communicate with each other the developments of magnetic core memories. This spirit of sharing continued after the symposium and is represented with the frequent communications between Papian and Wang and between Papian and Rajchman. Having explored the birth of magnetic core culture in the early attempts and the developments of other researchers and companies that joined this culture, it is now possible to explore how Forrester arrived at his revolutionary extension of the magnetic core concept, which ultimately validated and established the magnetic core culture.
Analog to Digital
In March of 1946, Project Whirlwind was born at the Massachusetts Institute of Technology's (MIT) Servomechanisms Laboratory as a digital computer project funded by the Office of Naval Research (ONR) for the purposes of creating a general purpose flight simulator. Before Whirlwind, there had existed at the Servomechanism Lab since December of 1944 an ONR project known as the Airplane Stability and Control Analyzer (ASCA), an analog computer attempt at building such a simulator. It was the ASCA project, which metamorphosed into Project Whirlwind. From our vantage point in the 1990s, in this world of computer guided missiles and the Internet, it seems the most logical decision in the world to evolve a problem of analog control into one of digital control, and indeed, Forrester has long been praised for his foresight. However a careful examination of the technological and political cultures before and after the inception of Project Whirlwind, shows that the switch from analog to digital, while in no small part a progressive decision, was one more of expediency than of vision. Had Forrester continued attacking from the analog perspective, it was probable that his project would have been cancelled or at least drastically scaled back.
To fully explain the postwar climate in the world of automatic control it is necessary to delve somewhat into the history of control systems and of digital computers, two very separate technological trajectories until tied together in Project Whirlwind. Analog computing, simply put, is using continuous or non-quantum techniques to perform a calculation. Examples of early analog computation devices include the sundial and the slide rule [Small, 8]. When we consider the computation inherent in control techniques, we can also speak of devices such as engine governers and other feedback systems as rudimentary analog computers. Feedback theory matured quickly in the early twentieth century, especially after World War I, due in part to the creation of control devices for military technology. Automatic pilots and fire control systems are good examples of this maturation. By 1930 Vannevar Bush at MIT had invented the differential analyzer, a device combining mechanical methods with electric power sources (therefore an electromechanical analog computer) to solve 'complex differential equations' [Edwards, 46].
While analog control had a long history or at least an extended genealogy, electronic digital computing was in its infancy at the time Project Whirlwind began at MIT. Alan Turing, the famed British mathematician, led a group of British scientists and engineers who created the first electronic digital computer, the Colossus, in 1943. The Colossus project is largely unknown because of British war secrecy [ibid, 16] An American group at the University of Pennsylvania Moore School followed with the ENIAC in 1946 [ibid, 51]. Meanwhile John von Neumann, who had played a role in the ENIAC project, conceived the first 'internal stored program' digital computer, named the EDVAC, which was actually preceded into operation by later British designs [ibid, 51 and 62]. Not only was it a young discipline, but digital computing was considered to be useful only as high complexity calculators.
So we have two threads, one of analog control and one of digital calculation, being sewn together after World War II. This intertwining was not caused by Forrester, rather he helped to prove practical the theory of information being developed by Norbert Weiner and Claude Shannon, which held that calculation, communication and control were different aspects of the same entity [ibid, 66]. In fact, at least two people preceded Forrester with ideas of digital (or numerical) techniques for control systems. In the 1940s, Jan Rachman, an engineer at RCA, looked into designing an automatic digital system for the control of antiaircraft weaponry [ibid, 67]. Ironically, Jan Rachman went on to become the defendant in a lawsuit with Jay Forrester over the rights to several patents for magnetic core memory systems. Additionally, an MIT graduate student named Perry Crawford, whom Forrester knew quite well, did his master's thesis in 1942 on 'Automatic Control by Arithmetical Operations', also concentrating on the problem of a fire-control system for antiaircraft weaponry [Redmond, 27] Neither of them implemented a practical system.
Before Forrester switched to digital computing ideas, his group worked on implementing the ASCA flight simulator using analog techniques. It was the Special Devices Division (SDD) of the ONR that contracted Forrester and the Servomechanism Laboratory to build the simulator, for the purpose of streamlining the training of military pilots. Captain Luis de Florez headed up the SDD and was an alumnus of MIT. Interestingly, however, de Florez did not first ask MIT to build the ASCA. He went first to the famous Bell Telephone Laboratories (BTL), which had tremendous experience in the design of flight simulators. It was only after BTL declined the contract, citing more pressing war time concerns, that de Florez propositioned his alma mater [ibid, 2-4]. So it was, that in December 1944 Forrester and his group of engineers began work on the ASCA project. '...By May 1945 they were willing to admit that they had, in their earlier, relative ignorance, underestimated the development cost' [ibid, 20, f.13]. Cost alone was not the only factor; Forrester became increasingly frustrated with the complexity of the system and the inability of current analog techniques to cope with the problems [Edwards, 76]. The SDD group knew of the problems that Forrester was having and also knew that better people for analog computer design were out there, specifically Arthur Vance and his group at RCA [Redmond, 25, f.24]. It was in the summer of 1945 that Jay Forrester began to consider the benefits of digital computation methods. Perry Crawford especially had an effect, telling Forrester of the exciting potential of digital control. He did, however, also warn Forrester that 'the sort of problem that Forrester's group was attacking lay still too far in the future to be of any help' [ibid, 27]. By October of 1945, Crawford had gone to work for the SDD, the group paying Forrester's bills.
And so it was that in November of 1945, Forrester began focusing his staff on the idea of digital computation methods. A report of a staff conference Forrester held at this time read, "Techniques for high speed electronic computation have not been worked out, and several months' work will be necessary to properly evaluate the process" [ibid, 34]. Yet by mid-January of 1946, Forrester was able to propose to the Navy with "sufficient confidence" that digital techniques be employed [ibid, 41]. Thanks in good part to Perry Crawford's cheerleading in the military and political world, the Navy was prepared to take the plunge into the digital realm [Edwards, 80]. Thus, it was in March 1946 that Project Whirlwind was born.
This of course was not the end of Forrester's worries. He and his group were a gifted bunch, but the 'Pennsylvania technique', named after the designers of ENIAC at the University of Pennsylvania, was still quite new to them. In fact, as late as June 1946, 3 months after the project had been awarded, Forrester reported that, "We are not in a position to decide what must be built by next June until we know the best principles to use as a foundation" [Redmond, 54, f.17]. One way he and his engineers learned the best principles was to attend a summer course on digital computer design at the University of Pennsylvania from July to August of 1946 [Notes on Computer Course].
It was not just inexperience that made Forrester's leap from analog to digital so contentious. The fact that this was no typical digital calculation device is important to consider. The large-scale digital projects of the day were not intended for real-time use, let alone in something so complex as an aircraft simulator. This real-time, large storage requirement mandated a reliable, fast, cheap memory that just was not available at the time. The storage device of the day was the mercury delay line, an unsuitable choice due to its sequential nature. Electrostatic storage techniques offered randomly accessible memory but at the inception of Project Whirlwind the feasibility of using electrostatic tubes as memory was not certain.
Looking back, we see that Forrester's group was the second choice for working on the ASCA problem, after Bell Laboratories, who turned the project down. When Forrester reached a large obstacle in the project, the sheer complexity of implementing the aeronautical equations by analog methods, he knew that the SDD knew of his problems and also that the SDD knew that others out there had a better practical grasp of analog computer design. The possible cancellation of the ASCA project required Forrester to look for a new direction for his project, to keep it afloat.
This new direction was digital. Forrester had learned about digital computation techniques in the ENIAC in 1945. He had also talked with Perry Crawford about digital control methodology and knew that Crawford saw a great future for such technology. The problem was that at the same time, digital control seemed a thing of the future, with little practical evidence supporting such a rapid shift to digital. Still, since Crawford was an employee of the SDD and with digital control being his pet project, it probably was not terribly hard for Forrester to convince the Navy to endorse his switch to digital computation. While it was true that his group was unskilled at digital computer design, they were bright and picked things up easily. The problem still lay in finding a storage device.
Digital Computers: for Control or Calculation
In March 1946, Forrester and his associates finally decided to adopt digital computation techniques to attack the aircraft analyzer problems. The Navy's Office of Research and Inventions (ORI) approved their proposal and named it as Project Whirlwind. [Redmond and Smith, p43.] Nonetheless, the difficulties of building a high speed, reliable digital computer for real-time control systems proved to be much greater than Forrester originally imagined. Understandably, Forrester and his associates had to focus their efforts on building a digital computer. They then needed to explain to their sponsor what would be the unique virtue of their digital computer project, especially when it was compared with other digital computer projects.
It was obvious that Forrester and his associates, young and unknown engineers, were in a inferior position compared with the leading figures in digital computer technology, such as John von Neumann of the Institute for Advanced Study (IAS), Howard Aiken of Harvard, J. P. Eckert and J. W. Mauchly of the University of Pennsylvania, and G. R. Stibitz and S. B. Williams of the Bell Telephone Laboratory.(see Fig.1) Forrester thus had no choice but to do heterogeneous engineering. To military officials, he advertised the state-of-the-art digital computer and its application perspectives in military services; to mathematician s working for his sponsor, he stressed the engineering aspects of his digital computer project, particularly the requirement of real-time control which was "not imposed on other machines under construction." [Mina S. Rees p113.] In the meantime, Forrester tried to mobilize support from MIT's leading administrators by relating Project Whirlwind to MIT's post-W.W.II agenda of regaining its leadership in digital computer technology.
Project Whirlwind was originally under the supervision of the Navy's Special Devices Center (SDC), a subordinate of ORI. SDC had a good working relationship with Forrester's group. Perry Crawford, SDC's principal scientific advisor, was not only Forrester's friend but also shared Forrester's vision of applying digital computers for real-time control. [Interview with JWF by authors, Nov.21, 1997] However, when the direct technical responsibility for Project Whirlwind was taken over by the Mathematics Branch of the newly established Office of Naval Research (ONR) in 1948, [Redmond and Smith, Project Whirlwind, p.95] Forrester had to confront with the negative reviews of his project from ONR's mathematicians. The engineer's conception of digital computer found it in conflict with the mathematician's conception.
Warren Weaver, a mathematician and head of the Naval Research Advisory Committee (NRAC) at the time, visited Project Whirlwind on Feb. 15, 1947. As a mathematician who had done much work on fire control, Weaver appreciated the importance of engineering in construction of digital computers in his visit. [Memorandum, JWF to N. McL. Sage, subj.: "Warren Weaver, visit to Laboratory, Feb.15, 1947" Feb.27, 1947] Yet after Weaver visited IAS on Feb.26, 1947, he was "greatly influenced by von Neumann" and expressed "a more favorable inclination toward theoretical research at the Institute for Advanced Study than at M.I.T." [JWF, "Computation Book No.45," Notebook Supplement 45JWF27] After inspection of Project Whirlwind in 1947, Mina Rees, then head of the ONR's Mathematics Branch, became concerned about Forrester's group's lack of the "mathematical competence needed in the design of a new type of digital computer." [Memorandum, Perry Crawford to Director, SDC, subj.: "Discussion of Project Whirlwind at ONR Conference on 28 October, 1947," Oct.29, 1947] Even after Forrester recruited professor Philip Franklin from MIT's Mathematics Department to strengthen his group's competence in mathematics, the mathematician Francis J. Murray from Columbia University, a disinterested reviewer requested by ONR, still reported on Nov.12, 1947 Project Whirlwind's "insufficient attention to the mathematical side of the program." [Francis J. Murray, Nov.12, 1947, cited in Redmond and Smith, p.81.]
Forrester felt distressed by the negative reviews of his project by mathematicians. He complained that mathematicians lacked understanding of engineering problems of applying digital computers in real-time control systems. It was true that the logical architecture of the Whirlwind machine was followed that of the IAS machine. [Richard E. Smith, "A Historical Overview of Computer Architechture," Annals of the History of Computing, Vol.10, No.4, 1989, p.280] Yet Forrester argued that between the two machines there were still significant differences on the electronic circuitry to achieve the similar logical architecture. On Oct.11, 1948, Forrester and Robert R. Everett compared the Whirlwind computer program with the IAS computer program which was favored by mathematicians.(see Fig. 2) They pointed out that these differences were resulted from a "different basic philosophy and ultimate objective." They then concluded that two programs were not in direct competition. For "I.A.S. is engaged in scientific research, the goal of which is the study of high-speed computing technique while M.I.T. is engaged in engineering development, the goal of which is to produce and use computers." [Memorandum L-6, JWF and R. R. Everett, subj.: "Comparison Between the Computer Problems at the Institute for Advance Study and the M.I.T. Servomechanisms Laboratory," Oct.11, 1948]
Forrester's position on the importance of engineering in digital computer construction proved to be more convincing to his colleagues at MIT than to mathematicians. In a letter to Weaver in the early of 1946, professor Samuel Caldwell of MIT's Electrical Engineering department and head of MIT's Center of Analysis not only criticized von Neumann's lack of appreciation of the engineering problems in building digital computers but also indicated that MIT had the "key men required for the theoretical, developmental, and engineering aspects of the problem." [Samuel Caldwell, letter to W. Weaver, cited in Karl L. Wildes and Nilo A. Lindgren, A Century of Electrical Engineering and Compute Science at MIT, 1882-1982, Cambridge: The MIT Press, 1985, pp. 232-233.] In a letter to the staff of employees of Center of Analysis in early 1947, Caldwell wrote:
"... in the field of electronic computation we entered the war among the leaders and emerged in a much less favorable position. ... resumption of our work in electronic computation development, at a greatly increased rate, stands as the largest and most important single item on our future development program."[Ibid., p.233.]
As early as in April 1946, MIT president K. T. Compton and Caldwell arranged with the Rockefeller Foundation a two year, $100,000 study of electronic digital computation. It was named as Rockefeller Electronic Computer (REC) project with an aim at "an appraisal and development of fundamental methods, both mathematical and instrumental, that would provide a well-founded basis for the subsequent design and construction of a machine." [K. Compton to W. Weaver, 26 June 1947, cited in Larry Owens, "Where Are We Going, Phil Morse? Changing Agendas and the Rhetoric Obviousness in the Transformation of Computing at MIT, 1939-1957," Annals of the History of Computing, Vol.18, No.4, 1996, p.34.] In the early June of 1947, however, when Caldwell found that the scale of the REC program was to small to compete with Project Whirlwind, he proposed to cancel REC, for the "final machine [of Whirlwind] will be able to meet both Navy and MIT needs." His proposal was approved by MIT president K. T. Compton. [Ibid., p.235]
In 1948, MIT's administrators became concerned about the ever-rising cost of Project Whirlwind and the pressure from ONR. Yet after reading Ralph Booth and J. Curry Street's favorable report of Whirlwind's engineering achievements, MIT's administrators were convinced to support Project Whirlwind. [Redmond and Smith, Project Whirlwind, p.109.] In order to justify Whirlwind's ever-rising cost, MIT president K. T. Compton requested in September 1948 a report on all possible applications of digital computers for military and civilian purposes. [K. T. Compton to N. Mcl. Sage, Sep.8, 1948] Forrester and his associates then prepared a detailed report which discussed possible application of digital computers in guided missile data reduction, high-speed computer traffic networks, cryptography, interception networks, air traffic control, industrial process control, simulation and training, logistics, etc., with proposed cost of $2 billion in fifteen years. [JWF, H. R. Boyd, R. R. Everett, H. Fahnestock, and R. A. Nelson, "Forecast for Military Systems using Electronic Digital Computers, Report L-3," Servomechanisms Laboratory, Sep.17, 1948] Meanwhile Forrester also widely advertised the vision of using digital computers in real-time control systems. (fig 3.)
Although Forrester stressed that his group was working on a prototype digital computer, the astronomical cost of building Whirlwind computer made it difficult for it to be copied. In fact, as Mina Rees pointed out later, "Whirlwind was never copied." [Mina Rees p116.] In light of the requirement of using digital computer for scientific calculation, Whirlwind computer was unnecessarily fast, therefore too expensive. No matter how grand and imaginative Forrester's vision using digital computers for real-time control would be on paper, it did not fit to the transformation from wartime research to peacetime research in post-W.W.II period. The social environment became so adverse to Project Whirlwind that in early of 1950, it was actually at the edge of being stopped. [Redmond and Smith, Project Whirlwind, p.155.]
The coming of Cold War in 1950, however, saved Forrester's project. In March 1950, George E. Valley, Jr., then head of newly established the Air Defense System Engineering Committee, decided to fund Project Whirlwind was with a hope to use it in an air defense system, later became Semi-Automatic Ground Environment (SAGE)project. The cost-crisis of Whirlwind found it solution in the SAGE project with a proposed cost of $60 billion between 1951 and 1965. [Edwards, p.97]
It is worth noting that no reviewer of Whirlwind is able to translate the cost-crisis of Whirlwind into a technological crisis, i.e., the electrostatic storage tubes are not adequate to support huge real-time control systems. Since Project Whirlwind was the only digital computer project which was imposed on the requirement of real-time control, experts of other digital computer projects simply lacked the practical, unarticulated knowledge to single out the real weak point in Forrester's project, the inadequacy of storage devices. Mathematicians of ONR, on the other hand, admitted their incompetence in evaluating engineering merits of Forrester's project, depended mainly upon vacuum-tubes experts' evaluation of storage-tubes in Project Whirlwind.
In the review of Whirlwind by Ralph Booth (an electrical engineer) and J. Curry Street (a Harvard Physics Department's vacuum-tube expert) on Aug.26, 1948, we read that "they were enthusiastic about the potential of the storage tube and strongly urged the Navy be asked to acquaint itself with the high promise of this development."[Redmond and Smith, Project Whirlwind, p.109.]
In the spring of 1949, Dr. Harry Nyquist of the Bell Telephone Laboratory and Dr. Karl Spangenburg, head of ONR's Electronics Branch, visited Project Whirlwind in company of Mina Rees and C. V. L. Smith from ONR's Mathematics Branch. They were all favorably impressed by the storage-tube development program. [Redmond and Smith, Project Whirlwind, p.151.]
In the report dated Dec.1 1949 by the Ad Hoc Panel on Electronic Digital Computer of the Committee on Basic Physical Sciences of the Research and Development Board of the federal Department of Defense, which not only was the most critical of Whirlwind but also suggested stop Whirlwind, it still reported that "consideration be given to MIT's excellent staff on system studies and on the development of specific computing components, especially the storage tube." [Report of the Ad Hoc Panel, Dec.1, 1949, cited in Redmond and Smith, Project Whirlwind, p.153.]
Forrester, in collaboration with MIT's Radiation Laboratory, the Sylvania company, the Raytheon company, worked painstakingly to reduce access time and enhance reliability of their electrostatic storage-tubes. After Forrester's invention of "marginal checking technique", which made it possible to remove bad tube before its actual failure, Whirlwind's storage system became much more reliable, and much more costly. Yet in the June of 1949, Forrester's perceived insufficiency of the electrostatic storage-tubes to support a complex real-time control system forced him to keep on eye on other possible means of storage. From June 1949 on, Forrester and William N. Papian began to work actively on magnetic core storage devices. However, Forrester found himself in an awkward position: if he let his reviewers and critics known his thorny technical problem, he would have no chance to defend his project by advertizing its grand application-perspectives. Thus Forrester had to keep reporting every improvement on electrostatic storage-tubes to his sponsor, ONR, to show the progress of his project. As late as on March 16, 1950 did the magnetic core memory devices first appeared in reports to ONR. [Project Whirlwind, Summery Report No.24, third quarter, 1950, p.14] However, Forrester did not make this problem known to others. Even after the Air Force began to fund Project Whirlwind, in the final report of Project Charles (a preliminary project of the SAGE project) dated 1 August 1951, Forrester still claimed that "by the addition of auxiliary drums and special operations to facilitate sorting and coordinate conversion, a WHIRLWIND-type computer should be able to handle the problem [in an air defense system]." [George E. Valley, Jr., "How the SAGE Development Began," Annals of the History of Computing, Vol.7, No.3, July 1985, p.214] If Valley had known the storage probelm in Forrester's project, he would have given a second thought on his decision to fund Project Whirlwind.
Before George E. Valley, Jr., then head of newly established the Air Defense System Engineering Committee, heard about Forrester's digital computer project from Jerry Wiesner at MIT in January 1950, he had been unaware that the Forrester's project of aircraft analyzer had been transformed into a digital project.[Ibid., p.207.] After Valley visited Whirlwind, he was impressed by the high speed and reliability of Whirlwind computer and by Forrester's preliminary work on using digital compute for real-time control, especially air traffic control. So on March 6, 1950 Valley decided to fund Whirlwind and rent Whirlwind computer temporarily for one year to use it in air defense system. It was almost impossible for Valley to understand the technical crisis covered by the cost-crisis in such a short period of less than three months. Although Valley was greatly impressed the reliability of Whirlwind computer, in the summer of 1951 Valley finally realized that the "Whirlwind storage tubes ... weren't going to yield reliable, round-the-clock service." [Ibid., p.216.] Yet by now Valley had staked his reputation on Forrester's project, so he continued to pour millions into Forrester's development of magnetic core memory devices.[Ibid., pp.216-217.]
Being imposed to the harsh requirement of real-time control, Forrester had to achieve very high speed and reliability in his electrostatic storage tubes. This extremely expensive efforts incurred many critiques and triggered a cost-crisis of Project Whirlwind in 1949. Yet the attained high speed and reliability won the favor of Valley who then supported Forrester with huge funds. Yet the real nature of the cost-crisis was not publicly perceived. When the technical nature of the crisis became manifest in the summer of 1951, Forrester and his associates had been working on the magnetic core memory for almost two years. On the other hand, Valley had already staked his reputation on Forrester's project. Therefore Forrester's agenda of being a successful inventor and Valley's agenda of being a credible scientific advisor converged together to push the invention of magnetic core memory to its final success.
Now let us consider the technical crisis of the electrostatic storage-tubes behind the cost-crisis of Whirlwind.
Reverse Salient: Analog Addressing Reaches its Limits
This section establishes a relationship between the addressing strategy used to select bits within a memory array and the interconnect problem. The interconnect problem is fundamental to the "march of progress" in computer technology known as Moore's Law. The question is not how to build a smaller memory cell. The challenge is how to connect more cells together at lower cost.
As it applied to Project Whirlwind, the question of addressing was the question of how to cheaply select from the 16,384 bits required in the original Whirlwind specification. At the time of Project Whirlwind, electrostatic tubes were the dominant paradigm for developing high-speed random access memory. Rather than connect each memory element in some addressable scheme, the tube would deflect an electron beam to scan horizontally and vertically across a field of storage points. The bits were stored by charges placed on a screen. A holding beam kept the charge from dissipating by constantly scanning the screen.
Figure Addressing by beam deflection (an analog process) in a storage tube
Since the each tube could hold an unknown number of bits, the cost was essentially constant O(1) with the n bits. Storage techniques with expensive bit elements had a cost of O(n), but if the bits were cheap, then the drive electronics would swamp the equation. If the bits were addressed by row-column addressing, the cost was O(n1/2). Forrester was interested in creating row-column-height addressing to reduce the cost to O(n1/3). "The number of circuits to be controlled could be the cube root of the number of stored points." (Forrester July 27, 1949)
By converting a digital address into analog X,Y deflections, storage tubes promised a fixed-cost device, which could increase in storage density without necessarily increasing in cost. Storage tube memory was essentially attempting to avoid the interconnect problem altogether.
Early in the project, Forrester knew about electron storage tube research at various labs around the country from the Panel on Electron Tubes, held on November 19, 1946. Literature from and mentioned storage tube projects at NYU, NOL, NRL, ESL, GE, and Raytheon.
Forrester met with people from Raytheon on February 24, 1947 and two days later evaluated Edgerton glow discharge tubes. The glow tubes stored a single bit. Later that year, Forrester proposed a three dimensional array to read tubes with a minimum of electronics. The glow tubes proved to be an unacceptable storage element while work on electrostatic storage initially looked promising.
After unsuccessful attempts to improve the addressing circuitry and bit isolation in the storage matrix of electrostatic tubes, it must have appeared that a compromise solution was necessary. Core memory was a compromise between the fixed-cost addressing expected from electrostatic tubes and individually wiring each bit.
Forresters notebook features a drawing of core memory dated June 15, 1949 with "3 mutually perpendicular conductors with plane of coil 45º to each" This orientation was extremely awkward to assemble in large arrays since the wires must be threaded in row, column and height. Later, on October 9, 1949, a revised orientation was drawn that depicts a stack of row-column addressed cores, that maintains O(n1/3) cost by combining the stacks in a row-column-height addressed cube.
[Figure Illustrations of core addressing from Forrester's notebook.]
Forrester quietly researched the possibilities of core memory in 1949, performing experiments and testing the hysteresis performance of various materials. His notebooks show a continuing interest in applying row-column-height addressing to achieve a maximal density (see fig 2). On June 15th, his notebook features a drawing of core memory with "3 mutually perpendicular conductors with plan of coil 45º to each." This is not the final form, for we see in October 9, 1950 a "Consideration of 3 dimensional ferroelectric storage array" which has stacks of planar memories, connected at the edges. The electronics perform row-column addressing to simultaneously access a vertical line of bits, giving a multi-bit memory. Jay Forresters vision of row-column-height addressing to achieve a high-density memory 1 bit-wide was not used, however, that idea was gradually refined to take advantage of the need for parallelism. Rather than building sixteen devices with O(n1/3) each, one could use one stack with O(n1/2) addressing and have sixteen bits accessed in parallel, essentially costing O(n1/3).
The actual technique that was implemented for accessing core memory was to use the X and Y terms to select a vertical line of cores, which were accessed in parallel. In this way, 2-D planes of cores could be assembled conveniently, the addressing circuitry cost was O(n1/3), while the drive electronics cost was O(n1/2). This solution to the interconnect problem would be repeated again and again. For DRAM, tiny capacitors are arranged in a grid and share the same row and column addresses, and a single bit is read out of each grid.
"In an ideal storage system, it should be possible to arrange elementary storage cells in a compact three-dimensional array; storage elements inside the volume could be selected by suitably controlling three space coordinates along the edges of the array. A scheme of this type was first described by Jay W. Forrester in a Project memorandum, M-70, dated April 29, 1947, but the suggested medium -- a glow discharge gas tube -- was unsatisfactory. A suitable medium -- small ferromagnetic cores with rectangular hysteresis loops -- now shows good promise, and research on the problem is well under way." [Summary Report Q3 1950]
This was the first mention of core memory in the project reports. New materials had been found that promised 1/2-microsecond access times, much faster than the 10 microseconds of the fastest electrostatic storage tube. The potential for a massive speed improvement enabled Forrester to public report his core research. Internally there was an acknowledged need to find something to storage tubes whose development was blocked. Externally, progress was expressed in each Summary Report with new problems detected, identified, and corrected. However, the resolution of these manufacturing problems did not result in fast or reliable performance.
Our historical studies of the invention of magnetic core memory, show that the Kuhnian model of paradigm-shift is not complete when applied to engineering revolutions. When Forrester and his associates shifted to digital computation techniques in early of 1946, there was no well-established paradigm in digital computation technology at the time. It was not clear whether the logical architecture of digital computer should be serial or parallel. [Paul Edwards mistakenly takes the serial mode proposed by von Neumann in his "First Draft of a Report on the EDVAC" on June 30, 1945 as the mode of "the von Neumann machine." [Edwards p378 footnote 20]. In fact, the logical architecture of "the von Neumann machine" is the parallel one discussed in "Preliminary Discussion" on June 28, 1946 written by von Neumann et al. [Goldstine pp.253-256.] When Forrester began to work on the magnetic core memory device in June 1949, the magnetic core memory is just a dim technical idea rather than a well-established paradigm. In Walter G. Vincenti's terminology, the magnetic core memory from 1949 to its being mass-produced in 196o's can be called as a new variation.
Since a nascent technical variation does not possess a solid technical foundation to justify itself, it thus needs heterogeneous engineering which mobilizes both technical factors and non-technical factors (social, economical, political, etc.) to shield it from critics. Efforts to explain technical invention from the perspective of inventors' visions are unsatisfactory. For visions are highly personal things which cannot be shared widely in society while engineering activities are highly socialized process in which it is impossible to expect every relevant actors to share with inventors' visions. In light of this, the idea of heterogeneous engineering would provide a solution to the dichotomy between the personal characteristics of technical visions and the social characteristics of engineering.
To further our understanding of the growing process of a technical variation, we introduce the idea of "magnetic core culture" by which we hope to explain the transitional process from the paradigm of storage-tubes to the magnetic core memory paradigm. Members in the magnetic core culture share many common things: similar technical visions, the same magnetic materials, and commonly interested symposiums. From the point of view of technical contents, they communicate with each other frequently thereby facilitating the elaboration of dimly perceived technical idea. From the social point of view, they bring their different research agendas into the magnetic core culture. Thus different social needs penetrate through different research agendas of different members in the magnetic core culture into the specific ways of using magnetic core materials to store binary bits. Moreover, because the huge expenditure is needed to develop magnetic core memory devices, the huge procurement of computers in 1950's and 1960's is therefore vital to the expansion of the magnetic core culture which finally emerges as a new technological paradigm in the computer memory industry in 1960's. In this manner, the internal technical factors and the external social factors do intertwine together seamlessly in the formation and development of the magnetic core culture.
Edwards, Paul, The Closed World. The MIT Press, 1996.
Forrester, Jay, JWF Computation Notebooks from Magnetic Core Memory Box 4, AC 337. MIT Archives.
Goldstine, Herman H. The Computer from Pascal to von Neumann. Princeton University Press, 1972.
Kuhn, T. S., The Structure of Scientific Revolutions The University of Chicago Press, 1970.
MacKenzie, Donald Inventing Accuracy. The MIT Press, 1993.
Masterman, Margaret, "The Nature of Paradigm," in I. Lakatos and A. Musgrave eds., Criticism and the Growth of Knowledge. Cambridge University Press, 1970.
Pugh, Emerson. Memories that Shaped an Industry. The MIT Press. 1984.
Project Whirlwind Summary Report Nos. XX-28. Digital Computer Laboratory, Massachusetts Institute of Technology 1949-1951.
Project Whirlwind Summary Report Nos. 29-40. Digital Computer Laboratory, Massachusetts Institute of Technology 1952-1954.
Redmond, Kent and Smith, Thomas, PROJECT WHIRLWIND: A Case History in Contemporary Technology. MITRE Corporation, November 1975.
Rees, Mina S., "The computing Program of the Office of Naval Research, 1946-1953," Annals of the history of computing, Vol.4, No.2, April 1982.
Small, James S. General-Purpose Electronic Analog Computing 1945-1965.
Small, James S. IEEE Annals of the History of Computing, Vol. 15, Number 2. IEEE Press, 1993.
Wang, Dr. An. Lessons, An Autobiography. Wang Institute of Graduate Studies. 1986.
Appendix: A. Timeline
November 19, 1946
Panel on Electron Tubes includes storage tube projects at NYU, NOL, NRL, ESL, GE, and Raytheon.
February 24, 1947
Forrester met with people on the Raytheon storage tube project.
February 26, 1947
Forrester evaluated the Edgerton glow discharge tube.
January 3, 1948
Initial tests on an ST-32 demonstrate storage in a 3x3 e-tube.
Lifetime tests are just beginning. Access time ranges from 10 to 25 ms.
June 15, 1949
Forresters notebook features drawing of core memory and many notes.
"3 mutually perpendicular conductors with plan of coil 45º to each"
August 19, 1949
Forrester arranges for materials with Bejamin Faulk of Arnold Engineering Co for magnetic materials for core memory.
Q3, 1949 Summary Report
A total of 20 storage tubes have been made. Only 13 pass.
The memory circuitry will be redesigned to talk to 16x16 tubes, not 32x32.
October 9, 1949
Forrester illustrates a stack of row-column addressed cores, that maintains O(n1/3) cost by combining the stacks in a row-column-height addressed cube.
"Consideration of 3 dimensional Ferroelectric storage array"
October 12, 1949
ONR de-classified Project Whirlwind.
Q1, 1950 Summary Report
Two tubes have run without error on a pattern cycling test for several hours.
Sixteen (16x16) storage tubes run simultaneously in whirlwind, first attempt to use such a system as an integral part of a complete computer.
Q2, 1950 Summary Report
Still no 32x32 electrostatic tubes, although 16x16 yield has improved.
November 15, 1950
Forrester itemizes many problems with electrostatic storage in his notebook.
Spot diameter, current density, dimensional control, uniformity, and focussing
Q3, 1950 Summary Report
First mention of core memory in Summary Reports.
Fig.1 The list of the lecturers of the Moore School Course on electronic digital computers: 1946
Aiken, Howard H. Harvard University
Burks, Arthur W. Institute for Advanced Study
Chu, J. Chuan Moore School
Crawford, Perry O. ORI, U.S. Navy
Curtis, John H. National Bureau of Standards
Eckert, J. P. Jr. Electronic Control Company
Goldstine, H. H. Institute for Advanced Study
Hartree, D. R. University of Manchester
Lehmer, D. H. University of California, Berkeley
Mauchly, J. W. Electronic Control Company
Mooers, C. N. Naval Ordnance Laboratory
Rademacher, Hans University of Pennsylvania
Rajchman, Jan RCA
Sharpless, T. K. Moore School
Sheppard, C. B. Moore School
Stibitz, George independent consultant
Travis, I. R. Moore School
von Neumann, John Institute for Advanced Study
Williams, S. B. consultant, Moore School
Source: Martin Campbell-Kelly and Michael R. Williams eds., The Moore School Lectures (Cambridge: The MIT Press, 1985), p.xv.
Fig.2 Comparison of Whirlwind Project to the IAS program
goals real-time control scientific calculation
speed very high only the tenth of Whirlwind's
reliability 50% reliable operation time 50% reliable operation time
as a minimum as excellent and fortunate
systems interested in coordinating new insufficient attention to
components into systems coordination of components
components 30%-40% Lab. efforts to the dependence on available
development of components
operational design standards are those difficult to be extended to
common to radar and military operational packaging
flexibility great flexibility for poor flexibility for different
different control systems control systems
reports accurate and complete mainly reports on arithmetic
technical reports methods for the solution of
training some 30% of the staff is on no substantial contribution
rotating training appointment to the training of new persons
methodology systematic style "step by step" serial style
basic philosophy engineering R&D aiming scientific experiment aiming for a
for a prototype of a digital breadboard of a digital computer
Source: Memorandum L-6, JWF and R. R. Everett, subj.: "Comparison Between the Computer Problems at the Institute for Advance Study and the MIT Servomechanisms Laboratory," Oct.11, 1948.
Fig.3 Visitors to Whirlwind with interest in real-time control
North American Aviation guided missile control
General Electric ground-based control for
Watson Laboratory (AMC) air-traffic control
Air Navigation Development Board air-traffic control
Naval Electronics Laboratory the construction of
San Diego damage control and
Cambridge Field Station computer early-warning
Coles Signal Laboratory the use of computers
Guided Missile Research guided missile control
Bureau of Ordnance guided missile control
Source: JWF, Memorandum L-7, subj.: "Visitors To Project Whirlwind," Nov. 2, 1948.
Fig.4 Comparison of the cost of Whirlwind to other digital computer projects:
The Raytheon "Hurricane" $460,000
The IAS computer $650,000
The Harvard Mark III $695,000
Source: Redmond and Smith, Project Whirlwind, p.166.