It became clear during 1948 that unexpected technical difficulties would delay completion of the Univacs. The machine for the Office of the Air Comptroller was primarily for use by George Dantzig, noted for his work on Linear Programming applied to economic problems. He urged the Air Force to contract with NBS for the development of an interim machine which would serve multiple purposes. Furthermore, the Bureau was in need of a computer for its own scientific research. Consequently, NBS began development, in its own laboratory, of an interim machine for use by NBS, the Army and Air Force, and the Census Bureau, which wished to test automatic methods on the 1950 Census. The interim machine at NBS was completed and put into continuous operation in May 1950, a year before the first commercial machine was delivered. It was originally called the Standards Eastern Automatic Computer to distinguish it from the Standards Western Automatic Computer still under development at the NBS Institute for Numerical Analysis in Los Angeles.

In the early years of its use, SEAC was operated 24 hours per day, 7 days per week. During this time four different kinds of demand were made on the machine. It was used by mathematicians from NBS and many other institutions to do the calculations that spurred the advance of numerical analysis. It was also used by the engineering staff of the Electronic Computers Laboratory to develop new components and to expand the capabilities of SEAC. A maintenance staff kept constant watch on the 12,000 diodes and over 1000 vacuum tubes by doing diagnostic testing. Finally, with the support of other agencies, NBS developed entirely new uses for computers and tested these uses on SEAC. Despite these competing demands, SEAC achieved reliable operation 77 % of the time during its critical first three years of operation. This was both a surprise and a source of encouragement to the young field of computer development.

The mathematical applications of SEAC were well represented by the publication of the NBS Handbook of Mathematical Functions [2] (which is separately described in this volume). One major calculation carried out dealt with nuclear energy. The early engineering and maintenance developments are described in an NBS Circular published in 1955: Computer Development (SEAC and DYSEAC) at the National Bureau of Standards [1]. It consists of an introduction by S. N. Alexander, the Chief of the Electronic Computer Laboratory which developed SEAC, and reprints of eight papers by NBS staff which were previously published in various journals and computer conference proceedings. Their titles give an idea of the scope of NBS contributions to computer design.

Foreword, A. V. Astin

Introduction, S. N. Alexander

SEAC, S. Greenwald, S. N. Alexander, and Ruth C. Haueter

Dynamic circuitry techniques used in SEAC and
DYSEAC, R. D. Elbourn and R. P. Witt

DYSEAC, A. L. Leiner, S. N. Alexander, and R. P. Witt
System design of the SEAC and DYSEAC, A. L. Leiner,
W. A. Notz, J. L. Smith, and A. Weinberger
High-speed memory development at the National
Bureau of Standards, R. J. Slutz, A. W. Holt, R. P.
Witt, and D. C. Friedman

Input-output devices for NBS computers, J. L. Pike and
E. L. Ainsworth

Operational experience with SEAC, J. H. Wright, P. D. Shupe, Jr., and J. W. Cooper

SEAC-Review of three years operation, P. D. Shupe, Jr. and R. A. Kirsch

The Greenwald, Alexander, and Haueter paper describes the organization of SEAC from the block diagram level down to the circuitry level. It exhibits. pictures of the tens of thousands of hand-constructed components which today are manufactured by automatic methods. To us today, it seems even more amazing than it did in 1950 that the thing could work at all!

This circuitry using diodes, vacuum tubes, and pulse transformers is described in the Elbourn and Witt paper.

The memory of SEAC initially consisted of what we would today describe as 3072 bytes, (not kilobytes or megabytes!). This memory, in the form of acoustic pulses in mercury columns, is described in the paper by Slutz, Holt, Witt, and Friedman, as are subsequent developments of a higher speed electrostatic memory and a memory design using individual components for each bit.

Because of the very limited memory capacity of SEAC, it was important to be able to store information on external media. These media were also used for feeding data and programs to SEAC. The paper by Pike and Ainsworth describes the various magnetic and paper tape media which served this purpose. One magnetic device was an adapted office dictating machine, modified to store digital information. Each programmer had one of these wire cartridges on which the whole history of his or her programming experience would be stored. These wire cartridges still exist today, but there are no devices to read them, thereby losing to history much of the detailed early experience on the first computer!

The two papers by Wright, Shupe, and Cooper and by Shupe and Kirsch describe how SEAC was maintained in productive operation despite the fact that failure in any one of the tens of thousands of components or solder connections would render the computer inoperative, or so it was believed. Actually, after about ten years of ostensibly fault-free operation, it was discovered by Kirsch that there had been a wiring error in SEAC from the beginning of its useful life. A program written in a certain way would have failed, but the conditions for this failure had never occurred in the many years of normal operation. Computers of today are not as reliable as SEAC aspired to be, so we use "workarounds" to live with errors.

Two important lessons learned from the design of SEAC were incorporated in the design of DYSEAC in 1953. The reliability of SEAC's components led to the use of similar components in DYSEAC, and new kinds of logical organization were also incorporated. These are described in the two papers by Leiner, Alexander, and Witt and by Leiner, Notz, Smith, and Weinberger.

DYSEAC was built in movable trailers for the U.S. Army and is believed to be the first computer to use printed circuits. In the short time after its construction at NBS, before it was removed by the sponsoring organization, it was possible to demonstrate some of its important properties. It had extensive abilities for real-time interaction with outside devices. One of these was SEAC itself, which was able to interrupt computations on DYSEAC and send data into files on DYSEAC without disruption of the DYSEAC computation. This was an early example of hardware-based time-sharing. Other examples that ultimately led to industrial process control were first demonstrated on the DYSEAC.

The last of the four kinds of uses of SEAC was the development of entirely new applications for computers. The first uses of SEAC (and indeed most computers) were for mathematical calculations. Soon thereafter it was recognized that computers, as symbol manipulators, could also process alphabetical information. This led to so-called business data processing. But the ready availability of SEAC for innovation without the need for commercial motivation encouraged its use beyond these two conventional areas. Two examples of the many such innovative areas started on SEAC were Image Processing and Chemical Structure Searching.

In 1957 the first picture was fed into a computer when a rotating drum scanner was connected to SEAC [3]. This project demonstrated that it was possible to perform image processing operations on scanned pictures, using the (for that time) great processing speed of the computer. Progress in scanning technology can be seen in the difference between one of the first scans and a modern one (Fig. 2). Among the many disciplines

Fig. 2. Father and daughter-two scans separated by 40 years and storage requirements in the ratio of 1400:1.

impacted at NBS by this new field of image processing were metallurgy [4], character recognition [5], and criminology [6]. Today, this technology, first started on SEAC, allows us to do CAT scans in medicine, receive satellite images from space, scan bar codes in grocery stores, and do desktop publishing [7].

Experience with non-numerical processing techniques led to the realization that different kinds of data. structures could be manipulated with SEAC. One of these was the structural diagram used in organic chemistry. In the process of examining new patent applications, the U. S. Patent Office needed to search files of chemical compounds. Previously, chemical names had been the only means of representing compounds, but this type of search was not very satisfactory owing to inconsistent nomenclature. It was shown [8,9] with the use of SEAC that structural diagrams could be represented in digital form and searched on the computer, thereby finding records that could not have been found

using nomenclature alone. Chemical structure searching is now a multimillion dollar industry which plays a major role, for example, in design of new drugs.

The rapidity with which SEAC and its descendants influenced a wide variety of scientific, commercial, and even intellectual disciplines encouraged the early pioneers to believe that "Nothing will be restrained from them which they have imagined to do" [10]. Many people still believe that today.

Prepared by Russell A. Kirsch.

Bibliography

[1] A. V. Astin (ed.), Computer Development (SEAC and DYSEAC) at the National Bureau of Standards, Washington, DC, National Bureau of Standards Circular 551, U.S. Government Printing Office, Washington, DC (1955).

[2] Milton Abramowitz and Irene A. Stegun (eds.), Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables, National Bureau of Standards Applied

Mathematics Series 55, U. S. Government Printing Office,
Washington, DC, June 1964.

[3] R. A. Kirsch, L. Cahn, L. C. Ray, and G. H. Urban, Experiments
in processing pictorial information with a digital computer,
Proceedings of the Eastern Joint Computer Conference, Dec.
9-13, 1957, Institute of Radio Engineers, New York (1958).
[4] G. A. Moore and L. L. Wyman, Quantitative Metallography with
a Digital Computer, Application to a Nb-Sn Superconducting
Wire, J. Res. Natl. Bur. Stand. 67A, 127-147 (1963).
[5] Mary Elizabeth Stevens, Automatic Character Recognition, A
State-of-the-Art Report, NBS Technical Note 112, National
Bureau of Standards, Washington, DC, May 1961.

[6] J. H. Wegstein, A Semi-Automated Single Fingerprint Identification System, NBS Technical Note 481, National Bureau of Standards, Washington, DC (1969).

[7] Russell A. Kirsch, SEAC and the Start of Image Processing at the National Bureau of Standards, IEEE Ann. Hist. Comput. 20 (2), 7-13 (1998).

[8] Louis C. Ray and Russell A. Kirsch, Finding Chemical Records by Digital Computers, Science 126, 814-819 (1957). [9] Herbert R. Koller, Ethel C. Marden, and Harold Pfeffer, The Haystaq System: Past, Present, and Future, in Proceedings of the International Conference on Scientific Information, Washington, DC, Nov.16-21, 1958, National Academy of Sciences, Washington, DC (1959) pp. 1143-1179. [10] Genesis, 11:6.

Francis L. Hermach's paper [1] launched the field of ac-dc thermal transfer metrology, which forms the basis for ac voltage and current measurement and calibration throughout the world. It laid the foundation for the techniques of ac-dc transfer and provided the first theoretical basis for the thermal transfer structures used in all national measurement institutes (NMIs, i.e., counterpart organizations to NIST) today. Hermach was the first to realize the very large improvement in capability that is possible when electrothermic elements are used as ac-dc transfer devices instead of relying on absolute instruments as had been the practice previously. The impact of the paper was, therefore, nothing less than the creation of an area of electrical metrology that continues to provide the national and working standards on which the world's NMIS base their ac voltage and current calibrations.

Although the paper contains construction details. and experimentally determined characteristics for new instrumentation developed at NBS, it also has over five pages of very detailed electrical, thermal, and thermoelectric modeling for the critical elements in the newly proposed thermal transfer standards. It contains the very first solution to the steady state temperature distribution in an ac-dc thermal transfer instrument and includes effects of Peltier and Thomson heating and low frequency error due to failure to average the applied signal.

This ground-breaking publication is the most cited work in the ac-dc thermal transfer field. It has been and continues to be cited by scientists and engineers in NMIs all over the world and is regularly mentioned for providing the basis of new calibration standards. Virtually every major NMI has a copy in its technical library. The paper is still disseminated routinely to metrologists who require a solid foundation in the field of thermal transfer

measurements.

Hermach's paper made a major contribution by proposing and describing the use of electrothermic instruments as transfer devices, as well as clearly delineating the major physics elements limiting their performance, thus creating a whole new area of calibration standards.

AC voltages and currents in the frequency range from low audio to hundreds of megahertz are measured most accurately by comparison to de standards using ac-de thermal transfer instruments. AC-DC thermal transfer structures were first applied in the audio frequency range and later at radio frequencies [2] for difference measurements of voltage, current, and power. Hermach and the staff of the NBS Electricity Division produced important developments including the first description of coaxial transfer standards and the first transmissionline analysis of such structures [3].

In general, the rate of transformation of energy from electrical to thermal form in thermal converters is proportional to the root-mean-square (rms) values of current and voltage. The heater temperature is a function of the square of the heater current even if the constants in the defining equation that describes the underlying physics vary with temperature or time. Since the response of thermal converters is calibrated on direct current at the time of use, ac-dc transfers are possible with little decrease in accuracy from drift or external temperature influences.

Traditional thermal converters contain wire heaters or thin metal heater structures. The temperature of the heater is typically monitored with one or more thermocouples, also made of wire or thin metal film. The best-performing primary standards usually contain many thermocouples in an arrangement that minimizes ac-dc difference by reducing both heater temperature and thermal gradients. Current research at NIST includes two areas directed at new thermal converters suitable for both primary and working standards.

Multi-junction thermal converters (MJTCs) are used in very high-accuracy ac-dc difference metrology because they have very small ac-dc differences, follow the rms law of excitation, and produce high output emfs. MJTCS traditionally have been fabricated from wire heater resistors and thermocouples. The project to develop thin-film MJTCs (FMJTCs) involves the use of micro-machining of silicon and photo-lithography on thin films to produce high-performance thermal transfer standards. Multilayer FMJTCs have been designed, fabricated, and tested at NIST by J. R. Kinard, D. B. Novotny, and D. X. Huang, and new improved converters are under development [4].

The basic elements of the devices are a thin-film heater on a thin dielectric membrane, a silicon frame surrounding and supporting the structure, and thin-film thermocouples positioned with hot junctions near the heater and cold junctions over the silicon. Carefully selected materials in new thermal designs are required, along with very accurate dimensioning of the heater and thermocouples. The heater and thermocouples are sputter deposited and patterned with photolithography. Contributions to ac-dc difference from the Thomson effect and other effects are further reduced by the appropriate choice of heater alloy.

Integrated micropotentiometers are thermal transfer devices that contain FMJTCs and thin-film output resistors fabricated as an integrated structure on the same silicon chip. The figure shows an integrated micropotentiometer including the FMJTC structure. New versions of the FMJTCS and integrated micropotentiometers are under development that include new membrane materials and vacuum packaging, with the help of novel etching techniques such as front and back surface etching.

At audio frequency, thermal and thermoelectric effects ultimately limit the measurement uncertainty in conventional room-temperature thermal converters. Heater powers as high as a few tens of milliwatts and temperature differences as high as 100 K are common in some thermal converters. To reduce these effects and to achieve very high temperature sensitivity, a novel sensor employing a superconducting resistive-transition edge thermometer is being developed at NIST by C. D. Reintsema, E. N. Grossman, J. A. Koch, J. R. Kinard, and T. E. Lipe [5,6]. Since the new converter operates at temperatures below 10 K and is mounted on a platform with precise temperature control and very small temperature gradients, the thermal and thermoelectric errors are potentially quite small. Because of the very high temperature sensitivity of the superconducting transition, this converter also offers the possibility of direct thermal transfer measurements at very low signal levels.

This transfer standard consists of a signal heater, trim heater, and temperature sensor all mounted on a temperature-stabilized platform. The sensor resistance is measured by an ac resistance bridge, and the temperature of the assembly is held constant by the closed loop application of power to the trim heater. A NbTa thinfilm meander line is used as the thermal sensor, and it is thermally biased to operate within its superconducting-resistive transition region. The signal heater in the prototype device is a 7 thin-film meander line and the trim heater is a 450 PdAu thin-film meander line, both adjacent to the detector on the silicon substrate. To ensure temperature stability, the entire converter assembly is mounted on a second platform controlled at a slightly lower temperature. This intermediate stage is thermally isolated, and controlled by a second ac resistance bridge using another transition edge sensor and heater.

Using this new cryogenic converter, measurements have been made at signal power levels of a microwatt, which is around 1000 times lower than is possible with room-temperature converters. Characterization using a fast-reversed-dc source has shown that the thermoelectric errors are presently in the 1 V/V to 2 μV/V range. These early results are encouraging, but considerable improvement both in the resistance bridge performance and in the input transmission line will be necessary for this new device to be a candidate for consideration as a primary standard.

Prepared by J. R. Kinard, Jr., N. B. Belecki, and J. F. Mayo-Wells, based on excerpts from the paper The Ampere and Electrical Units [7], authored by members of the Electricity Division.