VOL. 52, 1964 ENGINEERING: J. A. RAJCHMAN 363 fected cells, which produce high frequency of lysogeny, show a pattern of DNA synthesis similar to that of c+-infected complexes. The functions of the cl and c2 loci are interpreted as controlling DNA synthesis during infection leading to ly- sogeny. The cl locus represses phage DNA synthesis in the first few minutes of the infection, and the c2 locus maintains the repressed state after the recovery of cellular replication. The assistance of Mrs. Margaret Dunson in the performance of these experiments is greatly appreciated. * This work was supported by U.S. Public Health Service grant GM-09252-02. t Postdoctoral research fellow of the U.S. Public Health Service. I Lederberg, E., and J. Lederberg, Genetics, 38, 51 (1953). 2 Lennox, E. S., Virology, 1, 190 (1955). 3Jacob, F., Virology, 1, 207 (1955). 4 Levine, M., Virology, 3, 22 (1957). 6 Levine, M., and R. Curtiss, Genetics, 46, 1573 (1961). 6 Smith, H. O., unpublished results. 7Kellenberger, E., K. G. Lark, and A. Bolle, these PROCEEDINGS, 48, 1860 (1962). 8 Sechaud, J., Arch. Sci., 13, 427 (1960). 9 Luria, S. E., P. K. Fraser, J. M. Adams, and J. W. Burrous, in Exchange of Genetic Material: Mechanisms and Consequences, Cold Spring Harbor Symposia on Quantitative Biology, vol. 23 (1958), p. 71. 10Kaiser, A. D., Virology, 3, 42 (1957). 1 Bertani, L. E., Virology, 12, 553 (1960). 12 Egan, J. B., and B. W. Holloway, Austral. J. Exptl. Biol. Med. Sci., 39, 9 (1961). 13 Kaplan, R. W., U. Winkler, and H. Wolf-Ellmauer, Nature, 186, 330 (1960). 14 Kaiser, A. D., and F. Jacob, Virology, 4, 509 (1957). "5Jacob, F., and A. Campbell, Compt. Rend. Acad. Sci., 248, 3219 (1959). 16 Zinder, N. D., Virology, 5, 291 (1958). 17 Levine, M., and H. 0. Smith, unpublished results. 18 Jacob, F., and J. Monod, J. Mol. Biol., 3, 318 (1961). INTEGRATED MAGNETIC AND SUPERCONDUCTIVE MEMORIES BY JAN A. RAJCHMAN RCA LABORATORIES, PRINCETON, NEW JERSEY Communicated by V. K. Zworykin and read before the Academy, April 29, 1964 1. Introduction.-Memory in computers: Electronic techniques allow the execution of millions and even billions of arithmetic or other elementary logic functions in minutes or at worst in hours. This great speed is significant only for long sequences of computations, and sequences that are strictly specified by detailed instructions. These instructions must be available at speeds compatible with those of the computation, and it must be possible to select automatically among alter- native instructions according to the results of computation in progress. The intermediate numbers occurring in a long numerical computation and the instructions forming the program for the execution of the problem must be stored. This is the role of the "memory." It must be possible to dispose of incoming in- formation, or to "write-in" the "memory," at any desired location, or "address," Downloaded by guest on October 1, 2021 364 ENGINEERING: J. A. RAJCHMAN PROC. N. A. S. and similarly to extract information, or "read," from any desired address. Such selective addressing must be possible not only in a short time but, preferably, in a time independent of the selected address or any previously selected addresses. In other words, it must be possible to access the memory at "random." The realization of high-speed random-access memories made possible the universal computers which are such an important reality today. The first truly digital high- speed random-access memory was the selectron,I an electrostatic storage tube using a matrix of selectable electron beams striking discretely made storing elements. 1010 PROBABLE REGION OF 226 CRYOELECTRIC 109_X / MEMORIES ~-A ______M 224 -222 - PROBABLE REGION 08 INTEGRATED MAGNETICOF_ 220 ____--- MEMORIES WITH Q 7 \9 __ \ASINTEGRATED CIRCUITS cr ~~~~~~~~~~~~~~~~~218 STATE OF ART _ EXPERIMENTATE63 AeM.NeAL 216 W FERRITE (63) 106 I16384 (I) co - ~~~~~~~~~~~~~~4,096 FO5 MICROFEG1 RITE 1aa024dca SOME ~~~~~~~~PROTOTYPES -256 4 -64 memories 6eedvlpd2~13 ____LAT WAFER~S(62)Ths1emreonitofary2o4nivdal- IOO1.S IOMIS FLus lO0ns l.ns FIG. 1.-Storage capacity and cycle time of various memories. Core memory: M\ore than a decade ago, the first random-access magnetic memories were developed.2', I These memories consist of arrays of individually molded ferrite cores in the form of tiny rings which are linked by two sets of parallel and other wires. The material used is not only ferromagnetic but, due to con- siderable development work, its hysteresis loop is almost perfectly rectangular. As a result, it is possible to bring any core to one or the other of its remanent states of magnetization. The state of remanence stores naturally, and without standby power, the information as to which direction the core was last magnetized. This memory property is a gift of nature. Furthermore, because of the hysteresis loop rectangularity, the core not only performs the function of storage, but participates in the selecting function inherent in addressing, in that the amplitude of the addressing currents can be chosen so that the selected core switches while the other cores in the selected lines remain completely unaffected. In this way electronic switches are required only for the lines of the array and not for every core, so that their number, Downloaded by guest on October 1, 2021 VOL. 52, 1964 ENGINEERING: J. A. RAJCHMAN 365 although large, is still tolerable even when these switches are vacuum tubes. With the later advent of the transistor, the classical solution to random-access memory became the transistor-driven and sensed arrays of ferrite cores. Since its advent, the core memory has been gradually improved4' and has reached a range of capacity and speeds illustrated by the hatched area on Figure 1. Need for more storage capacity: Despite great progress, the memory is still the main limit in the further broadening of processing information by machines. The demands on memory result simply from the fact that all information as to what to do, how to do it, and what data to do it on, flows through the memory. Obviously, the storage capacity and speed of access of the memory determines the capability of the whole processing machine. Among specific examples of problems demanding more storage capacity are: solution of differential equations, aerodynamics, particle accelerators, plasma physics, etc.; problems of large matrix inversions and linear equation systems; combinatorial and symbol manipulative problems; problems involving masses of data such as weather prediction, statistical analysis, pattern recognition, etc. An indication of the demand for larger storage capacities can be seen in the growing use of electromechanical storage systems such as rotating drums, disks, and moving magnetic cards. These systems are a compromise dictated by neces- sity. They attain capacities of millions to billions of bits at the cost of access times measured in seconds, and the necessity for maintenance characteristic of high-speed complex mechanical devices. Clearly then, electronic random-access memories of capacities as large as hundred millions or billions of bits and with access times measured in microseconds are extremely desirable. How can they be attained? 2. Integration Techniques.-Are further gradual improvements in the ferrite core memory likely to lead to memories with billions of bits? Probably not, for the following reasons. The cores are molded and tested individually on automatic machines. Highly performing and uniform cores are thus obtained. The cores are threaded on wires by semiautomatic methods to form plane arrays. The arrays are interconnected by manual operations. Finally, the whole magnetic structure is connected to its associated transistor circuitry. This circuitry is made up of individually con- structed, mounted, and connected transistors and other circuit elements. The entire memory is therefore an example of highly evolved techniques for handling individually made components, or in modern parlance, highly evolved "non- integrated techniques." Incidentally, there are usually one hundred to two hundred as many magnetic cores as semiconductors, but the cost of the magnetic and semiconductor subassemblies are about equal. The degree of automation required for a billion-bit memory can be appreciated from the simple arithmetic fact that a billion seconds constitute thirty years. Therefore, a production rate of one-per-second is two or three orders of magnitude too slow to be practical. Molding and testing machines have attained rates of ten cores per second; still, many such machines must be used merely to provide the cores fast enough. Threading of the cores is a much slower operation still. It is clear that enormous effort in automation is necessary to obtain assemblies with billions of cores by such inherently brute force methods. Downloaded by guest on October 1, 2021 366 ENGINEERING: J. A. RAJCHMAAN PROC. N. A. S. What is needed is a method for making all the magnetic storing elements and all the linking windings by some batch fabrication which in a single step can produce the whole magnetic assembly. In short, an integrated technique is needed. Many different approaches to integrating the magnetic structure were undertaken. An early development was that of the apertured ferrite plate6 molded with an array of holes and covered with a metallic coating so patterned as to pro- vide one winding linking all holes. In the more recent microferrite memory,7 the same printing of winding through the apertures permits the use of extremely small cores. In FIG. 2.-Laminated ferrite structure, 16 X 16 bits (aspirin the FLEA memory' utilizing tablet for size comparison). metallic sheets of permalloy etched with an array of holes and covered with an insulator, all the necessary windings are provided by a printed technique. Fast switching and the possibility of integrated-batch fabrication of thin films of permalloy only a few thousand angstroms thick'0 13 gave incentive to substantial research in recent years.