History of Computing in France: A Brief Sketch

computer industry french research

In the second half of the 1950s, laboratories acquired commercial computers: Bull Gamma ET, Elliott 402, IBM 650 and 704, etc. 1960s

III The Computer Industry: Companies' behaviour, strategies and the component Problem

As no computer know-how was available in the public research system, the first two French computer manufacturers, Sociйtй d'йlectronique et d'automatisme and Compagnie des machines Bull, adopted risky policies -- either very innovative (SEA) or rather conservative (Bull). A third set of companies, the electrical equipment manufacturers, did not invest in computers until 1960, then chose to buy American licenses. In this chapter, I will particularly put the stress on the relationship between the French computer manufacturers and the component industry, between 1952 and 1972.

At the beginning of the 1950s the computer manufacturers bought components from specialised producers and soldered these together to form the circuits that constituted the central unit, the "processor", of the machine. Twenty years later, advances in miniaturisation and integrated circuitry had made it possible to produce a complete processor as a single component, the microprocessor. Over this period, therefore, the two professions had experienced fundamental changes not only in the techniques they used but also in their very businesses: whilst in the 1950s the development of electronics for mass-market products (radio and TV) had been the driving force behind the transistor industry, in 1972 it was the computer industry which generated the main demand for semiconductors (39% of the total production of s.c.), especially for integrated circuits (58%); and to this we should add the demand for industrial control devices and telecommunications equipment, both making more and more use of digital electronics.

How then did the French computer industry conduct itself during this revolution? To what extent did the manufacturers offer a market that would stimulate the semiconductor producers? Was there any collaboration between the two in the design of new circuits? What effect did the Plan Calcul, launched in 1967, have on their relations?

We cannot avoid these questions if we want to understand the relative decline of the electronics industry in France during the period. Here we shall attempt some answers by studying the relevant attitudes of the main computer manufacturers: IBM France, SEA, Bull, CSF, CGE, Compagnie des Compteurs, CII.

a) Sociйtй d'Electronique et d'Automatisme (SEA)

Formed in Paris in 1948, SEA -- Sociйtй d'Electronique et d'Automatisme -- was both a research consultancy and a producer of specialised electronic calculating equipment Despite the loss of most of its archives with the disappearance of CII, the history of SEA is beginning to become fairly well known, thanks to the participation of its former partners and senior staff in conferences on the history of computing, e.g. Colloques surl'Histoire de l'Informatique en France. . Its Director, F-H Raymond, an electrical engineer with a Doctorate in physics, had worked in the Navy's research laboratory and afterwards in various enterprises where he had become familiar with the then state-of-the art in electronics. SEA established its reputation and founded its business on its designs for analogue computers, various products for telecommunications and automation, and simulators for aeronautics. It installed the first digital computers in France in 1955, and seemed to be overflowing with inventiveness: it had patents in text processing, in virtual memory, in teleprocessing and in electronic telephone switching. Initially these were intended for the scientific world, with business applications considered later. Some of these projects failed because the quality of the components produced in France around 1954 was not up to the technical demands of SEA, and importing from America was made difficult by customs barriers. The situation was to improve within a few years. In 1960 SEA produced a small scientific computer", the CAB 500, which was to become its best-seller; at the same time it launched a series of transistorised business machines, the SEA 3900s. Its joining the Schneider group made possible the production and marketing of its machines on an industrial scale. In 1967, when it had a staff of 850 and its Director was participating in formulating the Plan Calcul, SEA was to all intents and purposes absorbed into CII, Compagnie Internationale pour l'Informatique.

Ferrite Cores

Raymond has written of the views he held at the beginning of the 1950s: "The primary question was, how to design memories? Various laboratory solutions had been experimented with, in France, in Great Britain and in America ... In my opinion at the time, none was suitable from an industrial point of view. When I was in the USA in 1946 I had met Jan Rajchmann, then deputy to Zworykin who was head of the RCA laboratories at Princeton, and luckily I was able to get him to give a lecture to the French “radioelectricians” on ferrite core memories (Rajchmann 1952: 479-491). His talk helped to plant the idea that these were a genuine possibility into the French engineers community. Cores had to be produced. Only Radiotechnique took up the idea, and we were able to make our first 'core planes' by buying wired and tested cores from Radiotechnique.” (Raymond 1989). Fifteen years later RTC (Radiotechnique-Coprim) were producing 250 million cores annually. The collaboration with Philips's French subsidiary was to remain a consistent feature of SEA's innovation strategy.

The Physics & Chemistry Department: Betting on Hybrid Circuits

In 1955 SEA set up a department for research in physics and chemistry, in order to monitor the evolution of components and to master the basic technologies needed in the design and manufacture of computers. Ten years later this had a staff of 40, organised into five groups -- hardware, devices/systems, electronics/ memories, circuits/ technology, workshop/ models.

Between 1961 and 1966 this Department made 17 studies under contracts with various military departments and with the DGRST (Dйlйgation Gйnйrale а la Recherche Scientifique et Technique); these led to 83 patent applications in the field of computer components. One of the main production tools that were developed was a machine for cutting silicon wafers by means of a power-driven wire; licences were sold in America, Great Britain and Japan. "The computer today depends on the solid state" wrote Raymond in 1966. "Neglecting research brings the certain risk of being destroyed by the competition; handing the research to other companies, even if allies, risks being served badly and too late. Depending on licences or foreign patents means being always 2 to 5 years behind in getting a machine on to the market" (Raymond 1967: ch. 5 p. 1) Keeping up with the technology gives freedom to decide on what hardware to produce, without waiting for the competitors' product announcements; and this favours a policy of effective dialogue between the computer manufacturer and the components industry. SEA could consent to sub-contract production and even the marketing of its machines, but not research.

In early 1964 the chiefs of the Physics & Chemistry Department went to America to study the state-of-the-art in solid-state physics, visiting about ten industrial and university laboratories The French team visited Philco (Lansdale and Blue Bell), IBM (Thomas J. Watson Research Center), RCA, Bell Labs, Johns Hopkins University, Stanford Research Institute, Caltech, Fairchild, Motorola and others.. They were disappointed by what they saw of monolithic integrated circuits. "Whilst, technologically, the integrated circuit is in the line of semiconductors, it must be a compromise for the simple reason that it is made up of different devices in the same material. This compromise must be to the detriment of performance. Further, the fabrication technology is so complex that the yield [the number of "good", i.e. usable, circuits, as a percentage of the total number produced] is less than 3%. Thus with its poor performance and high cost the integrated circuit does not meet the needs of the computer industry."

This judgement, which seems astonishing to us now, was based on notes made on the spot by the two specialists. Originally, integrated circuits were indeed too expensive to be used in computers manufactured for the commercial market, and their use was limited to military, space and nuclear applications. It was not until the end of the 1960s that the yield for TTL circuits reached about 30%, when the average price fell from $ 50 to $ 2.3.

On the other hand, the two Frenchmen found confirmation among their American hosts of the view taken by SEA five years previously, that the right fabrication technology was that of depositing thin layers on an insulating substrate. Passive elements -- conductors, resistors, capacitors -- could already be made in this way, the problem was how to add active elements.

In the mid-1960s SEA incorporated pilot hybrid circuits in pre-prototype computers for testing, and prepared for industrial production with Radiotechnique (the good collaboration of the two companies was partly based on the fact that Dreyfus-Alain, at SEA, and Brunet, head of s.c. development at RTC, were both alumni from the Paris Ecole de Physique-Chimie). This represented a change in the nature of the company from an assembler of components to a designer, a change stimulated by the evolution of semiconductors. In 1965 Raymond proposed to the Defence administration a computer project, "CAB 15000", intended to result in a compatible range comprising at least three processors, with the expectation of producing about a thousand units altogether (Raymond 1965). In the event, only one example was produced, using hybrid circuits operating at 20 MHz, developed by SEA in collaboration with Radiotechnique. Receiving no Government support, the project was soon cancelled.

This was followed by a project to design a family of monolithic integrated circuits: despite their initial scepticism, the authors of the 1964 report had concluded: "... However, this is not sufficient reason for us to dismiss all considerations of integrated circuits and their variants from our thoughts." By 1966 the yield of the fabrication process had passed 10% (Kilby 1966). SEA planned to use integrated circuits in the Axe 2 range of computers which would take over from the CAB 1500 about 1970. In 1965 Radiotechnique launched its DTL circuits for numerical applications, producing a million in 1967 and ceasing to lose money in this venture in 1969. However the Plan Calcul, launched in 1966, simply ditched SEA's projects.

b) Bull: A Client For the Semiconductor Industry

The Compagnie des machines Bull was incorporated in Paris at the beginning of the 1930s to exploit the patents for punched card machines taken out ten years previously by a Norwegian engineer, Frederik Rosing Bull. It was owned and managed by a group of industrialists, including in particular George Vieillard, a Polytechnician of the 1914 vintage who was Managing Director until 1962, and by members of the Calliиs family, related to Michelin and owning the Aussedat paper mills that supplied the cards required for Bull's "accounting and statistical machines". Starting with a workforce of 50 in 1931, the company won for itself a respectable fraction of the French market against the competition from well-established multinationals like IBM and Remington-Rand. Since 1948, Bull had been engaged in an ambitious international expansion (Mounier-Kuhn 1989, 1990).

Also in 1948 the management decided to set up an electronics laboratory to give them the knowledge to beef up their machines and so make these better able to face the competition, in particular that from IBM's new products. The founder of this laboratory, Bruno Leclerc, left with his colleague Henri Feissel -- both of whom had worked on radar projects at LMT, a French subsidiary of ITT -- for a visit to the USA. On their return they designed and built the Gamma 3 calculator, marketed in 1952 (Leclerc 1990a, 1990b). In their initial forecasts the company had hoped for 70 sales at 920,000 F each; in fact, in 10 years they sold more than 1200. The success of the Gamma 3 was one of the main factors in the spectacular expansion of Bull during the following decade In 1960 Compagnie des Machines Bull was one of the world's leading manufacturers of data processing machinery. It had a base of over 4,000 installations, of which a third were exported; 14,000 employees in France, ten factories, and a global turnover of 201 MF, which had multiplied by 10 over the past ten years. Its profits began to fall in 1962 and in 1963 it showed a loss of 85 MF, with no hope of recovery in the short term. A series of political and industrial manoeuvres ended in 1964 with its takeover by General Electric..

Gamma 3 was an excellent machine, superior to the equivalent IBM product (IBM 604) both in its logical design and in its technology -- logic circuits based on germanium diodes, exhaustively tested and much more reliable than vacuum tubes. But like the 604, the Gamma 3 was an electronic calculator, not a computer; the program was not stored in the machine but was wired on a plug board similar to those in a manual telephone exchange -- the standard way of "programming" punched card machines. It was not until 1956 that Bull added to the machine a magnetic drum, to give what was their first computer, the "Gamma Extension Tambour". The commercial sucess of the Gamma systems made Bull one of the major European users of germanium diodes.

Bull started to build transistorised machines in 1957. These included the Gamma 60, a powerful machine using 38,000 transistors, first delivered in 1960 (only 19 produced); Gamma 322, successor to Gamma 3, in 1961; Gamma 30, made under licence from RCA; and Gamma 10, a small commercial machine of which 1850 were sold. Up to 1970 all Bull and Bull-GE machines were built from discrete components, the technology of which made great progress in both performance and cost reduction during the decade, paralleling the microcircuit "revolution".

Until the end of the 1950s Bull did not produce electronic components, instead buying its vacuum tubes from CSF and RCA and its germanium diodes from Radiotechnique.

The components remained a competitive handicap to Bull, even the passive components. In 1956, the syndicate of the French electronics industry stressed that prices of professional equipment were 35% above the international rates. In 1961 Bull's President complained that the European manufacturers were penalised by components prices: for example, mica capacitors cost from 1 to 4 F, against 0.4 F. in the USA. No company would embark on automated production unless assured of very large orders, and Bull could not commit any orders unless the quality of the product was guaranteed. A high level engineer from the Ministry of Industry, Sueur, after a visit to Bull in 1963, recorded that automated production of components was "a factual impossibility [...], the industrialists will not risk making the investments needed [...] and the government is rather helpless Callies а Blancard, dйlйguй ministйriel pour l'Armйe de l'Air, 23 fйvrier 1961 (AN 771521/52). ."

Development of new components became one of the main concerns of Bull's R&D during the 1960s; it absorbed between 10% and 20% of the R&D budget The Rapport Schulze & Bigard gives the following provisional research budget for 1964:

General advanced studies 6 MF

Scientific and process-control calculators 7.6

Business-machine devlopment 14

Enhancement and improvement of existing
business-machine hardware 4.3

Miscellaneous 2.1

Total 34 MF. This was accompanied by a great increase in the company's interest in the academic science community, of which the "Bullists" had taken little notice up to then. The management began to court the DGRST and the minister responsible for research, with the aim of attracting Government support. However, Bull received little support for work concerning hardware development -- partly, it seems, for political reasons: the Callies, owners of Bull, were outspoken antigaullists; moreover, this manufacturer of accounting machines had a mediocre technical reputation in the military milieux.

In the mid-1960s the semiconductors, still germanium, came mainly from Cosem-CSF and Radiotechnique, where Bull sponsored special production lines. Later General Electric, who controlled Bull, sought to persuade them to buy from Sesco, a joint subsidiary of Thomson and GE; but Bull's engineers expressed a preference for Texas and Fairchild, on grounds of both performance and price. From 1967 Bull-GE research teams designed exclusively machines based on integrated circuits, but the combination of their own shaky financial position and GE's scrimp and save policy prevented them from producing any such machines before the 1970s.

c) CSF and CGE: Adopting American Technologies

The CAE (Compagnie Europйenne d'Automatisme Йlectronique) was formed on 1 July 1960 as a joint venture by CSF, Intertechnique and the Californian corporation Thompson-Ramo-Woolridge. Intertechnique had won in 1958 a contract with EDF for control of nuclear reactors by tendering the small RW-300 computer, which it would build under licence from Ramo-Woolridge (RW). TRW offered technology in the form of licences for RW machines.

The main partner, CSF, seemed to hold all the cards needed for success in informatics. It was one of the leading French companies in electronics with a staff of 15,000, including 1400 engineers, and a turnover of 600 MF; backed financially by Banque de Paris et des Pays-Bas, it had the resources to invest massively in a new technology.

Exceptionally for France at that time, CSF had a strong tradition of research; it was the only big company headed by a "Normalien" An graduate of the pure science-oriented Ecole Normale Supйrieure, in Paris., Maurice Ponte, who had a doctorate in physics. Its research facilities were divided between two centres: electronic and particle physics in the Warnecke and Simon laboratories at Corbeville, with 800 staff including about 100 engineers and research scientists, which alone made about 30 MF by selling its research in France and elsewhere; and physical chemistry at Puteaux, with 250 staff including 50 engineers, where one of the three departments worked on semiconductors under the direction of Claude Dugas, a former assistant of Yves Rocard's “Visite au Centre de physique йlectronique et corpusculaire de la CSF”, L'Onde йlectrique n° 400, July 1960. The Semiconductor department was working mainly on silicon power transistors, tunnel diodes, field-effect devices and photoelectric cells. Between 1963 and 1965 the laboratory obtained three contracts with DGRST, as part of a general programme for electronics R&D. 22 doctorate theses had emanated from this laboratory between 1933 and 1960, for research directed by professorial friends of Maurice Ponte (de Broglie, Grivet and Rocard). Ponte came close to sharing the Nobel Prize with G.P. Thomson in 1927: it was only a delay/scarcely-explicable hold-up in the transmission of his paper that prevented this.. Applied research was done in most of the company's subsidiaries.

CSF produced electronic hardware for the professional market, mainly for the military; its industrial capability was well suited to producing the computers of the period, small series in runs of a few dozens. The group was also the biggest producer of components in France: Vacuum tubes, ferrite cores, diodes and transistors. In principle, this vertical integration ensured control of costs, control of quality of supplies and technical command in the designing of systems. From this point of view CAE represented the opening of a new outlet for CSF component.

With the RW-300, CAE imposed itself as the leading provider of control equipment for nuclear power stations, and added new RW machines to its catalogue. Later it developed a machine based on the American technology but with an improved logic, the CAE 510; this proved a success as a real-time scientific computer. At the same time, to provide the missile guidance systems for France's nuclear weapons, it built under licence the RW-130 which had already been used in the USA for various military applications; from this it derived upgraded computers by adding silicon transistors to TRW's germanium components.

A Customer for Texas Instruments

CAE grew rapidly, its turnover increasing by a factor of ten between 1963 and 1966, partly as a result of it's taking over the informatics activities of another company, CGE (now Alcatel). The association between CSF and CGE led CAE to take out a licence from another American company, SDS (Scientific Data Systems), whose products CGE had begun to import.

CAE “frenchified” American calculators, meaning that it built these, partly with elements made in France: ferrite cores, magnetic reading heads and various other parts. Strangely, the active components -- transistors, and integrated circuits especially -- came mostly, not from Cosem, but from Texas Instruments, who in 1961 had opened its factory at Villeneuve-Loubet. According to the former technical director of CAE, Cosem's silicon transistors were too unreliable and the development took too long, doubtless because CSF had not acquired the same command of silicon technology as had its American competitors. "CSF's policy was to be a second source; they were not used to undertake any development unless requested and underwritten by the Administration Conversation (26-09-1992) with Robert Chambolle, former Technical Director of CAE.."

Five years later a government study implicitly condemned CSF's policy of simply following behind the leaders. "Lacking big enough military orders or broad enough markets developed early enough, the French industry, taken as a whole, was unable to establish a firm financial base, for it usually started to manufacture this or that family of products just when the selling price had fallen to a very low level. Consequently it can not meet from within its own resources the high cost of the micro-electronics studies on which its future depends. [ ... ] Without the means, and perhaps also due to a lack of judgement, CSF-Cosem was late in developing the products [ICs]that should now be taking over from their predecessors [Ge discrete components]."

The short term view taken earlier by its management in favour of germanium saddled the company with a certain amount of technical backwardness. According to the then head of the CSF St.Egrиve laboratory, the most difficult technological revolution the component manufacturers had to contend with was not the germanium transistor, which did not require a production organisation essentially different from the vacuum tubes', but silicon, an especially difficult maaterial to handle. In addition, St.Egrиve, in the Alps, was 600 km from the Paris CSF laboratories and communication between the two was slow and arduous; whilst the key to success in micro-electronics lies in improving the fabrication processes, and this requires close collaboration between research and production. Further, CSF was not accustomed to mass production. Its priorities were for custom products made in small numbers for the Army and for nuclear research, which offered few possibilities of economy of scale -- a tentative diversification into radio and TV for the mass market failed; as for discrete components, its publicity boasted "Cosem, the main European producer of diodes". It left the gate open for silicon components, through which the new American companies poured -- Texas Instruments, Fairchild and Motorola: France was Motorola's main customer in Europe in 1966, with a turnover of $ 2 M.

D) The Plan Calcul: National Ambitions and American Technology

The Plan Calcul, one of the most ambitious technological programmes of the Fifth Republic, aimed at establishing an informatics industry that would guarantee France independence from the American manufacturers.

The government's policy was to shape a "national champion", a company, preferably big (if necessary, formed by "inducing" several companies to merge) which the State would support through R&D grants and preferential purchases. The "Calcul" and "Composants" plans meant, primarily, concentrating the industry so as to reduce internal competition and generate a "critical mass" effect. The weakness of such a course is evident: the "national champion" is formed, not as the product of market pressures -- which would have left a number of competing firms -- but in response to criteria imposed by the State in the interests of its own needs, and to existing alliances between this or that branch of the government service and this or that firm A service devoted to “managing” and encouraging mergers (“Bureau des fusions et regroupements d'entreprises”) was created within the Ministиre de l'Industrie by a 6 October 1967 decree !. It will therefore orient its structure and its strategies to the demands of the State rather than to the needs of the market, and will find it very difficult to change its behaviour and ethos so as to remain competitive.

The main industrial protagonists in the Plan Calcul were CGE and Thomson-CSF. Their main objective was neither the closing of the gap in component production (Cosem's doctrine was "controlled delay" and CGE had sold its components subsidiary to Philips-Radiotechnique) nor the development of a computer industry (in an official report in 1966 they told the government that they feared this would provoke reprisals from IBM), but the control of government funding. As the government submitted the distribution of its funding to the condition that the industry “concentrated” its activities, CGE, CSF and Thomson began a series of mergers so as to regroup their specialised subsidiaries. The resultant companies were anointed "national champions". This enabled them to both drain off most of the public money and eliminate any rival firms who might try to compete for subventions. Thus in the computer field, SEA was absorbed; in components, Radiotechnique was marginalised; and the new entrants, Silec and later Efcis, were doomed.

The launching of the Plan Calcul was expressed by the creation of a “Dйlйgation а l'informatique” at Government level, two research centres, IRIA and LETI and an industrial complex consisting of CII (computers), SPERAC (peripherals) and SESCOSEM (components). These creations were accompanied by large investments, 7,500 MF (about $ 1,500 M) between 1967 and 1975.

CII (Compagnie internationale pour l'informatique) was formed in December 1966 by the merger of SEA and CAE, enforced in a very hostile climate. The two companies had been competitors, and on bad terms with one another for a long time. They approached the Plan Calcul with conflicting strategies: SEA wanted to strike out on a new line, investing mainly in research so as to "skip a generation" and have new, very competitive products to offer in the early 1970s, whereas the CAE team took a shorter-term view and proposed building SDS computers under licence. Given the relative strengths of the industrial, and above all political, forces involved, the actual outcome of the creation of CII was that SEA, the more innovative company, was absorbed into CAE, who imposed their own strategy and products: thus the first products of the national champion were American computers built under licence. Only in 1969 did CII begin to manufacture its home-brewed computer range, “IRIS”.

For components, the State agreed with the industry a "5-year Micro-electronics Convention" Plan Composants was announced by the Minister for Industry in April 1967, just at the moment when CSF published its deficit. It was to provide 100 MF over 5 years, mainly to support production of integrated circuits. This was followed by a second convention in 1972 which doubled the support to 200 MF over 5 years, equivalent to over 900 MF in 1991.. This, the first of a long series of plans concerning components, guaranteed the French producers a certain amount of government finance, officially to support research aimed at reducing the estimated 3-year technical lead of the Americans. All this was, in fact, inscribed in the context of the big industrial manњuvres of the period: Thomson absorbed CSF and the semiconductor activities of the two groups were combined in the merging of SESCO and COSEM to form SESCOSEM Thomson and CSF had reached agreement in the field of semiconductors in the summer of 1966. An Italian subsidiary, Mistral, which supplied the consumer market, was added to the combination resulting from their merger, which then had a turnover of 135 MF, with one third of the French semiconductor market and 10% of the European. Thus Thomson-CSF was for a time the second European producer, after Philips., which received 20 MF between 1969 and 1973 for support of its R&D. To what extent was the support for research diverted into cover for losses? An unsigned note in the DGRST papers, dated 13 February 1968, complains: "Nothing of this has been achieved. The Components Plan is simply a rescue operation, for saving Company X when its losses -- due to bad management -- have become catastrophic Quoted by Zysman (1982: 173). SESCO and COSEM were both in deficit after the mid-1960s. Merging the two was far from forming the "critical mass" that was the official aim of the merger. In 1969 SESCOSEM's research department had a budget of 30 MF and employed 90 engineers, less than a quarter of the resources committed to component research by Texas in the USA, whom France sought to overtake.". The evolution of Cosem's turnover showed a peak of discrete Ge components in 1966, followed by a rapid decline; combined with weaknesses in the sales force and above all with technical backwardness in Si products, this resulted in a general fall in Cosem's production in 1967. Financial results fell to zero in 1966, to - 3 MF the next year. Cosem never became profitable again.

Yet CSF-COSEM did carry out research on integrated circuits. In 1964 it had produced an experimental TTL circuit under contract with the Air Force, and in 1968 a 16-bit MOS memory under contract with DGRST at the Corbeville laboratory. But the step from laboratory to industrial scale was always long and difficult. The DTL and TTL circuits were still in the development stage at COSEM in 1967, and it was not until 1968 that the first production was presented at the Paris component fair. The forecast then was that the output of TTL circuits would reach 100,000 per month in 1969; now, Texas had been producing the same "chips" -- their Series 54 and 74, of which COSEM's were clones -- at their plant at Villeneuve-Loubet since 1966. In 1969 the American company reduced the price of its MSI circuits by 25%, forcing its competitors either to follow suit or to throw in the towel. It did the same again in 1971, just when the recovery attempt at Sescosem was beginning to bear fruit; Sescosem then began to consider giving up production of TTL circuits.