A Survey of Magnetic and Other Solid-State Devices for the Manipulation of Information

210 IRE TRANSACTIONS ON CIRCUIT THEORY September

_ +(l + Ar,B~lSe’B’CErC)I, 1 ~52’ = e( V,, - I,R,)I,. ’

(27) evC7.G 1 -I- h-BBrIc/akB (23)

If I, is expressed in terms of Ic = Pm/V,,, where P, Differentiate (27) with respect to temperature,

is the maximum dissipation, (23) becomes dl,- 1 dT - e(V,, - 2I,R,) ’ cw

1 -= 40 + ArB~~I.4+sj~,J . ep, 1 + Ar,B’Pm/ffCRVCE (24) Combine (28) and (I@, and substitute Ic in terms of

Rearranging p, and R,

4RP, _ 1 4

1 e L dhdk + IlrBBj

--?- (& - l)]. (25) pm = h In +s ‘i,bOV,,n,, 41 k 4P,,Rc/z’

Taking the natural logarithm of (25)

For Case 3) due to the presence of external resistance The author wishes to express his thanks for the in- in series with the collector, R,, there is a voltage drop valuable assistance given by J. Gibson and Dr. W. iLII. IcRc between the power supply, V,,, and the collector. Webster of the Radio Corporation of America, where The heat conduction equation should be modified as most of this work was done.

A Survey of Magnetic and Other Solid-State for the Manipulation of Information*

JAN A. RAJCHMANt

Devices

HE MANIPULATION of digital information by automatic machines such as high-speed computers, information or data-handling machines, process

control mechanisms, etc., may consist of a great variety of tasks, some of which may involve very complex logic. From the point of view of the physical realization of the automaton, these logical tasks amount to a transformation, according to a given set of more or less complicated but definite rules, of a set of “on-off” signals into another set. Such a transformation can always be accomplished by a system of interconnected physical elements which have the following attributes: nonlinearity, fast response, gain, and storage.

Any physical quantity necessarily varies continuously (neglecting quantum effects). To obtain a digital signal there must be a qualitative difference between various ranges of this quantity. To discriminate between these ranges there must be thresholds below which some effect

* Manuscript received by the PGCT, May 29, 1957. This paper was presented at a tutorial session of the 1957 IRE-AIEE University of Pennsylvania Transistor and Solicf:State Circuits Conference, at the invitation of the Conference Program Committee.

t RCA Labs., Princeton, N.J.

does not occur and above which it does occur. The simplest case is that of a single threshold producing a binary element with two distinct ranges of values. A relay is a typical example of such a device, the ‘value of the con- ductance through the contacts being almost zero or very high depending on whether the. current through the solenoid is smaller or greater than some threshold. The subdivision of the range of variation of the continuous variable into more than two ranges by several thresholds, requires, in general, artifices which must be very complicated to insure clear discrimination. In practice, the exponential combinations of a very few simple binary devices, which are equivalent in range to more complicated multithreshold devices, are almost always preferable. The notion of a definite threshold is an idealization of the behavior of physical effects in which some variable varies nonlinearly with respect to some other variable. For example, a vacuum tube can be considered as a binary device because it is conducting or nonconducting, depending on whether the grid voltage is below or above cutoff. In this application, the range of continuous control of the grid, which should be ideally zero, is straddled to obtain on-off operation. As sharp nonlinearities as possible are, of course, the most desirable.

1957 Rajchman: Solid-State Devices for the-Manipulation of Information 211

Fast response of the manipulating element is of primary DIODES AND TRAI\TSISTORS

interest, because the more important application of The semiconductor. diode is really the first solid-state automata involve very long sequences of operations. To d evice to have made an important imprint in computers. obtain the solution in a practical time, each elementary It was in use before the transistor was conceived. It has a OperatiOnmuSt be performed in a shorttime.It is COlllmOn- definite threshold Separating a region of practically no place to say that the faster the machine, the more versatile conductivity from a region of high conductivity, and it it is and the more ambitious the problems it can solve.

In addition to devices possessing nonlinearity, there is capable of operating in high-speed circuits. However, it has no gain or storage properties (except in one unusual

must also be devices capable of storing signals. The mode of operation). Logical AND and OR gates made of incoming signals which occur over a period of time must d’ d be stored to be ‘available at the time dictated by the logic

10 es are classical and can be combined to make decoders, encoders, and generators of any specific logical switching

of the operation which is to be performed on them. functions. Tens of thousands of diodes were used in some Storage is also necessary to enable the machine to perform early large computers in conjunction with tubes which various sequences of operations in which the result of provided the gain and storage properties lacking in one operation is the basis of the next. Storage of bivalued d. d 10 es. These were hybrid semiconductor-vacuum tube signals can be obtained either by the natural states of computers mentioned here for historical interest. some materials, such as the remanent states of a magnetic The transistor makes possible all solid-state computers. core, or else by a feedback artifice which creates two c onsidered as a computer element, the transistor can be stable conditions, as for example, in a tube or transistor flip-flop.

viewed as having: 1) a very sharp threshold-an input voltage of f 0.1 volt turns the transistor from an on

Another characteristic required from our candidate is condition of about 100 Q impedance to an off condition gain, i.e., the requirement that the output signal of the device have a greater energy than the input. This is

of 100 kB; 2) a fast operation-the switchover can occur in about one microsecond in many types of transistors,

necessary in order to overcome the inevitable losses in the and in a time measured in millimicroseconds in newer coupling circuits when a given element drives another. types; 3) a reasonably large gain-it is easy to drive 4 or For universality of logic circuit design, a gain large 5 t ransistors by a single transistor, and; 4) storage proper- enough to enable one element to drive several others, ties through the artifice of a flip-flop connection. at least three and preferably five others, is necessary. The existence of the threshold permits the use of

The vacuum tube was the first device to have all the t ransistors in logical circuits. A typical circuit is illustrated requirements: threshold, gain, high-speed, and storage (by proper artifice). It was the basis for the birth of

on Fig. 1 (a). This circuit, similar to a diode logic circuit

high-speed computers. For many years it seemed that the control of the motion of electrons in vacuum was the “1

G ~ .

“7 only tool available for fast switching and storing devices. Y : Various ingenious counting, computing, gating, and OUTPUT

storing tubes were developed. Less than a decade ago, the advent of the transistor opened a new era, the era of e3 solid-state devices. As is well-known, the new device is OUTPUT

capable of performing all the functions of the tube as well or better. It requires no heater power. It is very ’ small. It promises to be inexpensive and to have a long ?I % e3 %

life. Perhaps equally as important as the specific advant- _ - = ages of the transistor, was the realization that electron optics was no longer the sole possible tool and that the e, whole realm of solid-state phenomena should be looked into. On the heels of transistors came the modern de- (4 (bJ velopment of magnetics, which is second only to the Fig. l-Transistor AND and OR gates. (a) OR or AND gate. transistor in its spectacularity. Active research was (b) AND gate.

initiated in many new fields such as ferroelectricity, photoconductivity, electroluminescence, and cryogenics.

This paper deals with selected salient aspects of these with the obvious .advantage of power amplification, is an various developments as they pertain to the manipulation OR or an AND circuit depending on the polarity of the of information in digital form, and in some instances in bias. Another arrangement of transistors for an AND analog form. Semiconductor diodes and transistors, circuit, illustrated in Fig. l(b), consists in using a simple covered by many other papers of this conference*, are series connection. Such a connection is satisfactory with treated briefly. There is a more detailed account of transistors because of their low resistance when on, and magnet& followed by a few remarks on the other, the fact that there are no complications attendant to relatively newer fields. heaters as in the case of series vacuum-tube circuits.


A very important property of the transistor is the ,possibility of controlling either electron or “hole” currents. In a vacuum tube circuit, a typical problem is that of charging and discharging a condenser. There is always a capacitive load due to stray capacitance, even if the main load is not capacitive. In very fast operating circuits it is usually the limiting factor. The conventional way to solve the problem is to charge the condenser by the electrons from the vacuum tube and discharge it by ,an electron flow through an anode resistance as shown in Fig. 2(a). The greater the rate of charging, the larger

I I I I I I f-

---

f T

\ I k? --- ; 1. 5

Fig. Z-Tube circuits for charging and discharging a condenser.

the tube current and the larger the resistance which acts as a shunt during charging. The rate of discharging, on the other hand, when the tube is off, is greater when the resistance which is the sole discharge path is smaller. Therefore, the value of resistance must be a compromise between the optimum for the charging and discharging cycles. To avoid this conflict a second tube can be used, as shown in Fig. 2(b). One tube is used to bring electrons to the condenser through its anode and the other to with- draw the electrons through its cathode. Such vacuum tube arrangements are very awkward since they require the series connection of relatively high-anode voltages and offer problems of cathode-heater capacitance and breakdown. The example illustrates clearly that a linear element such as a resistance is a detriment in digital circuits, which should include only nonlinear elements, and these should be capable of conducting currents of negative as well as positive charge carriers. _’

The conditions for the transistor are much more favorable because a p-n-p transistor can be connected with an n-p-n transistor as shown in Fig. 3(b). To charge the condenser (load), the bases of both transistors are made positive thereby rendering transistor n-p-n conductive and transistor p-n-p nonconductive. This allows electron current from source E, to charge the condenser. To discharge the condensers, the bases are made negative, reversing the roles of the two transistors. This allows hole current from the Ez source to discharge the condenser. This “complementary symmetry circuit” is quite simple and does not have the shortcomings of the vacuum-tube circuit. The possibility of controlling “hole” or positive particle current as well as electron current, is the main distinguishing feature. Of course, it is possible to use a single transistor to charge and discharge the condenser, as in conventional vacuum tube circuits [see Fig. 3(a)].

e I

LOAD LOAD

.- ZE’ 1

(b) Fig. 3-Transistor complementary symmetry circuit. (a) Single

transistor circuit. (b) Complementary symmetry circuit.

Two remarks can be made concerning the use of transistors in high-speed circuits. 1) When a square voltage is applied to the base of the transistor, the current through it rises and decays exponentially. Therefore, in order to render the current response more rectangular, it is customary to use a speed-up condenser in an RC network in the base circuit as shown in Fig. 3(a) and+3cb). This tends to sharpen the rise and the fall. 2) When the transistor is saturated, minority carrier storage limits severely the speed of response because the storage time is relatively long. Care must be taken to avoid saturation. This can be accomplished either by limiting the current by some external artifice, or by the use of clamping diodes.

MAGNETICS

In the last decade, magnetic cores made of material having a square hysteresis loop, have become important manipulators of digital information. New materials, devices, circuits, and systems are still being discovered and invented at an unabated rate. The new field of magnetics is an art, growing too fast to have become a science.

The fact that ferromagnetic materials possess hysteresis gives them inherent storage capabilities, since there can be several remanent states after the removal of energizing magnetomotive forces. If we consider a given hysteresis loop, the core can be magnetized clockwise or counterclockwise. Actually, if we consider a whole family of hysteresis loops, there are an infinite number of remanent states. Only two are usually used in circuits by forcing the core to describe a given loop.

The existence of remanence makes the magnetic core a natural element for storage since the core remains in the remanent state indefinitely without any. holding power. In order for the core to be useful in switching, it must have a threshold. It is primarily for this reason that materials with square hysteresis loops have been sought. Also this is required to render the two states more distinct. Historically, the first materials with square loops were obtained by using very thin alloy tapes that were heat treated in a very exacting way. Such tape cores are still

1957 Rajchman: Solid-State Devices for

the basis of many devices. However, the synthesis of ferrospinels or “ferrites” with square hysteresis loops greatly broadened the usefulness of ferromagnetic devices. These materials allow the making of cores with great uniformity at much less cost.

This paper describes the philosophy underlying some of the commonly used circuits and devices using square- hysteresis-loop cores. Magnetic circuits will be treated in greater detail than transistor circuits, as the author has done more work in that field and believes it to be less known to the members of this conference.

The storage arraiys of cores with operation based on current coincidence 1-S are, perhaps, of the most spec- tacular core circuits. If we consider an array of cores arranged in rows and columns with two windings on each core, said windings being connected in series along rows and columns, it is possible to energize one row and one column with a current of such an amplitude that only the core receiving both the column and row excitation has enough magnetomotive force to switch over, whereas all cores subjected to only a row or a column excitation have insufficient excitation to switch over (Fig. 4). This

Fig. 4-Array of cores.

two-to-one selection system is widely used. It is interesting to consider that the current-coincident system can be extended to three-to-one selection and in fact to selections with greater ratios. For example, if all the cores of the array were subjected to a minus one-half current, where one unit of current is the value nominally required to switch over a core, the application of plus one-half currents on the row and column would switch no core, and further- more no core in the array would have an excitation greater than a half. If many planes had identical row and column excitations of plus one-half, only the ones in which the minus one-half excitation on the cores was absent would switch over. Consequently, with a two-to-one

1 J. W. Forrester, “Digital information storage in three dimen- sions using magnetic cores,” J. Appl. Phys., vol. 22, pp. 44-48; January, 1951.

2 J. A. Rajchman, “Static magnetic matrix memory and switch- in circuits,” RCA Rev., vol. 13, pp. 183-201; June, 1952.

3 J. A. Rajchman, “A myriabit magnetic core matrix memory,” PROC. IRE, vol. 41, pp. 1407-1421; October, 1953.

9

the Manipulation of Information

I”

213

;

Fig. 5-Three-to-one selection and generalization.

discrimination a three-dimensional selection can be obtained (Fig. 5). This possibility results from the fact that both sides of the hysteresis loops are used.

Such use can also be exploited to, get a higher discrimination in selecting a single core in a plane. For example, if an inhibiting current of minus one-third were applied to the whole plane and row and column selecting currents of plus two-thirds were used, then the selected core would get precisely one unit of drive while all other cores would get plus or minus one-third units of drive. One could utilize the threshold of the hysteresis loop by using more than two or three selecting signals. In fact, one could generalize to n signals (‘see table of Figure 5). When this is done the ratio of currents of the nonselected to the selected element approaches unity as n increases and consequently a more definite threshold or a so-called “squarer” hysteresis loop is required.’ Conversely if the hysteresis loop is extremely poor, one can use redundant switching. For example, not only can one select the row and column, but also the diagonal which goes through a given point4 (Fig. 6). The excitation of the diagonal is not logically necessary, but improves the ratio of currents in the selected and nonselected cores. In the gamut of these possibilities, the systems using a two-to-one or three-to-one selection are the only ones that are widely used.

The arrays of individual cores have become the classical solution for random-access memory chiefly because they have proven to be much more reliable than previously used systems. (Fig. 7) Memories with hundreds of thousands of bits are common in commercially made machines.

4R. C. Minnick and R. L. Ashenhurst, “Multiple coincidence zay;t5;torage systems,” J. Appl. P&s., vol. 26, pp. 575-579;

2 *

214 Lieptember

Fig. B-Redundant selection by the addition of diagonal windings.

While very much larger memories have been built, it seems that the technique of individual cores becomes uneconom- ical when extended to the storage of millions of bits. Con- siderable research is in progress to substitute other magnetic elements for the individual cores. Patches of thin films evaporated on a suitable substrate are being widely investigated.5 Recently, we have developed a

ferrite memory plate molded with an array of apertures in which the magnetization around each aperture stores one bit of information.6 The ferrite being an insulator, it is possible to print a winding directly on the plate and thus eliminate some of the manual threading (Fig. 8 and Fig. 9, opposite). We believe that memory capacities of millions of bits become practical through these techniques which greatly simplify the making of the memory. It is interesting to note that there is essentially no interference between the magnetizations around adjacent apertures because, for a given current through an aperture, the magnetizing force diminishes gradually with radial distance and, at a well-defined distance, (chosen to be less than half the width of the leg between adjacent-apertures) becomes smaller than the threshold of switchover.

Magnetic switches can be made to select one or several outputs out-of-many by using a number of windings, each linking certain cores in series in such a way that the energization of selected groups of windings will produce an algebraic sum of magnetomotive forces which exceeds the threshold of switchover on the selected core only.2

6 A. V. Pohm and S. Rubens, “A compact coincident-current memory,” Eastern Joint Computer-Conference, New York, N. Y.; December, 1956.

6 J. A. Rajchman, “Ferrite apertured plate for random access memory,” PROC. IRE, vol. 45, pp. 325-334; March, 1957.

Fig. 7-Typical core array.

There is a considerable difference between such switches and the coincident current memory, which is also combinatorial, in \that the nonselected cores in the switch can support arbitrarily large magnetomotive forces tending to drive them in the direction of their existing saturation, whereas this was not possible in the memory arrays because the state of remanence is unknown by definition.

Let us consider the example of a switch that might be called a combinatorial decoder switch. This is an example of a switch with eight outputs and three inputs whose purpose it is to select one output for every possible combination of inputs (Fig. 10). The input signals are in pairs, each one linking half the cores. The first input links the cores by juxtaposed halves, the second by interlaced quarters, and the third by interlaced eighths. The direction of winding is such that a current sent in one or the other branch of each input pair tends to magnetize the cores further into saturation. Consequently the applications of the inputs has no effect per se. If during the presence of the inputs, all cores are energized in a direction tending to reverse their magnetization, only the core which .is not inhibited by the input currents will have a net reversing magnetomotive force and, therefore, will be the only .one to switch over. After the core has been switched, it can be restored-by energizing a winding which links all cores. Decoding switches of this type have been used successfully as input switches for magnetic core memories.

Rajchman: Solid-State Devices *for the Manipulation of Information

Fig. g--Ferrite-memory apertured plate with printed winding.

Fig. R-Stack of memory plates.

The usefulness of the switch can be broadened by using output windings that link certain ones of the cores according to any desired code. When the selecte’d core switches it produces an output signal only in those particular outputs which link it (Fig. 11). This type of switch can be considered to be a decoder-encoder device, since it takes the binary code and decodes it and then encodes it in a new set of bi-valued signals. This amounts to code conversion. Any desired switching function can be obtained with such a switch. However, in general, this would be extremely extravagant in the number of cores required since there must be a core for every possible combination of inputs. For example, if there mere 20 inputs, a million

OUTPUT CORES

Fig. IO-Tube-driven decoder switch.

III

110

101

100

011

010

001

000

> OUTPUT

II

10

IO

01 -

IO

01

01

00

\I n INPUTS

RESTORE -/

m OUTPUTS

Fig. 11-Decoder-encoder or universal switch.

cores would be required. Consequently, the mere mech- anization of a truth table by such a decoder-encoder switch is not practical, in general, for circuits performing math- ematical and logical operations. It is necessary to take into account the specific logical relations between the input and the output in order to be able to reduce the number of cores to a practical value. This, in general, requires the switching of cores in succession and involves the transfer of information from core to core.

The classical example, in which the switching of one core causes the switching of another, is the so-called magnetic shift register.’ A row of cores are coupled to one another by means of circuits between windings which include diodes. In addition to the coupling windings, there are two energizing or “advance” windings, each linking every other core. The purpose of the shift register is to “register” a pattern of information and to “advance” this pattern along the register in response to the advance pulses (Fig. 12). To understand the operation, let us assume first that all cores are in a given normal state of remanence except one which is in the other. Let the energization of the advance windings be of such a direction as to tend to bring all energized cores to the normal

7 A. Wang and W. D. Woo, “Static magnetic storage and delay line,” J. Appl. Phys., vol. 21, pp. 49-54; January, 1950.


P-N

P--N N--P

Fig. la-Shift register.

state. When the advance winding linking the core in the ‘abnormal state is energized, this core will switch over, but all the other cores linked by the same winding will be merely driven further into saturation. The switching of the selected core induces a voltage in its output winding which produces a current in the coupling circuit tending to bring the next core of the row to the abnormal state. With the proper choice of the number of turns, to account for flux conservation and losses in the circuit, the next core will, indeed, switch over. In doing so this second core induces in turn a voltage in its output winding. This voltage would tend to produce current in the output circuit, and thereby impede the switchover, were it not for the diode which is so connected as to prevent this current flow. It turns out that with this choice of polarity of connection of the diodes in the row of cores, the switchover of the first core induces a voltage in its input winding of such direction as to produce a current in the input side which is not blocked by the diode. To prevent these currents from impeding the turn over of the selected core various artifices are used, as for example, a resistance and a shunting diode.

An arbitrary pattern of information rather than a single bit can be advanced through the register. Of course, this arbitrary pattern is stored at any one instant in half of the cores of the row, the cores of the other half being all in the normal state N. Since this type of register requires one core storing one bit of information and another ready to receive it, it is called a “two-core-per- bit” register.

The example of the shift register embodies the essential problem encountered in logical magnetic circuits: namely that the transfer of information from one core to another in a chain requires that there be isolation from the preceding and following links in the chain while the transfer is being made. To obtain this isolation diodes are used in most circuits. Alternatively some other device with nonlinear relation of current vs voltage can be used. A variation of the classical shift register is one in which the information is transferred from a core to a condenser

and then from condenser to a core.’ In this (‘one-core- per-bit” register only a single advance winding is required.

The example of the shift register shows that the addition of diodes to magnetic cores permits transfer of information from core to core indefinitely. This implies that there is gain. The gain results from the fact that the switching of a core to one polarity produces no load current because of the blocking effect of the diode and requires, therefore, only the amount of energy necessary to make up the losses in switching the core, whereas the switching of the core to the other polarity does produce a load current and resultant transfer of energy from the source to the load. The logical input step “primes” the core and determines, thereby, that an output is produced during the “drive” of the core. An unprimed core produces no output during drive. Amplification by means of a passive hysteretic element, is seen to be obtained with an intrinsic time delay between input and output. In conventional magnetic amplifiers, this is translated into the requirement that the frequency of the signal be much lower than that of the energizing carrier.

It can be shown generally, that logical circuits for producing any desired switching functions can be made up of configurations of AND and OR gates, and NOT circuits. The properties of isolation between gates and of gain between input and output which were required for the shift register are necessary and sufficient for these more generalized circuits. For this reason, the example of the shift register is fundamental.

Another type of magnetic circuit depends on current steering.“l’ This circuit is of interest because it requires only a few electronic drivers and the amplitude of the output current is largely independent of the properties of the magnetic cores, and depends only on the regulation of the electronic drivers. Consider a circuit having many parallel branches, each including a winding on a core in series with a diode (Fig. 13). Let all the cores be set in one direction, the normal direction, except one which is set in the abnormal direction. Let us now have a winding which links all the cores and which is connected in series _ with all the parallel branches. A current flowing through this winding, tending to bring all the cores to the normal state, will switch over the abnormal core and induce thereby a voltage on its output winding of such chosen polarity as to tend to make conducting the corresponding diode, and consequently tending to cut off .a11 the other diodes. The result of these tendencies will be to cause the current to flow entirely through the selected branch. The flow of the branch current through the selected core will tend to oppose the effect of the driving current, so that in order to have an excess of magnetomotive force necessary

8 R. D. Kodis, S. Ruhman, and W. D. Woo, “Magnetic shift register using one core per bit,” 1953 IRE CONVENTION RECORD, part 7, pp. 38-42.

9 M. Karnaugh, “Pulse-switching circuits using magnetic cores,” PROC. IRE, vol. 43, pp. 570-584; May, 1955.

10 J. A. Rajchman, and H. D. Crane, “Current steering in magnetic circuits,” IRE TRANS., vol. EC-6, pp. 21-30; March, 1957.

Rajchman: Solid-State Devices for the Manipulation of Information 217 1957

---t----J DRIVE

. Fig. la-current steering principle.

for switchover, there must be more turns on the switching winding than there are on the branch winding. The setting of the core that selects the conductive branch produces no branch current in itself since the back-to- back diodes block any possible current flow. This setting is akin to priming the circuit. The desired branch current has an amplitude which is the amplitude of the drive current source since the branch current is the drive current and is independent of the core and diode properties.

The usefulness of the principle of current steering can be illustrated by means of the decoder switch mentioned earlier. The three pairs of selecting windings that were tube driven in the previous example are now driven by current steering (Fig. 14). Each tube is replaced by a core and diode. The parallel pairs of selecting windings are connected in series. The inputs to the switch set one core in each pair differently from the other. The drive brings all the input cores to a standard direction of magnetization, thereby switching one core in Jeach pair. This causes the drive current to be steered through one -branch of every pair. Consequently, all output cores but oneSwill be subject to inhibiting currents and the selected one will be subject to a reversing current, just as was the case in the tube driven decoder. That particular core will switch over and produce an output in its output winding. These outputs do not have the benefit of current steering in this particular example. Current steering was used merely for selection, but not for the outputs. However, it is possible to connect a diode in series with the series-connected parallel selecting branches, so as to use current steering both for selection and for producing the output.

Another example of a current-steering circuit is one in which new settings are produced by the steered current itself. It is “two-phase stepping register” which could also be thought of as a current steering commutator (Fig. 15). The purpose of the circuit is to deliver a given current in succession to a number of loads. The steering cores are

-STEERING CORES

Fig. 14-Current steering decoder switch.

t + DRIVE”1 DRIVE*2

Fig. 15-Current steering commutator.

divided into two groups. A first drive current, steered by a given core (k) of one group flows through a corresponding load and switches a core (k + 1) of the second group. In turn, a second drive current, steered by the previously set core (k + I), flows through its corresponding load and sets a new core (k + 2) in the first group and so on. The circuit is similar to the conventional magnetic shift register (used as a ring counter) with the important difference that the advance current itself is steered through the successive loads. It is possible to design the commutator so that most of the energy of the driving sources appears in the load and only a small part is dis- sipated in the steering cores and diodes. Experimental current-steered commutator switches have been operated with 80 per cent of the drive energy appearing in the loads. Currents of one to two amperes have been steered to successive loads, in time intervals of one to a few micro- seconds.

The combinatorial switches, magnetic shift registers,

IRE TRANXACTIONX ON CIRCUIT THEORY September

and commutator switches described thus far were all driven from current sources. A current source is a device furnishing a given current without regard to the voltage this current will produce in the load, hence can be considered as a generator with infinite internal impedance. The current-steering principle epitomizes the current- driven type circuits. Magnetic switches can also be driven by voltage sources, that is, .by generators which produce a given voltage without regard to the current produced in the loads and which can therefore be considered to have zero internal impedance. ’ Some of the well-known voltage-driven magnetic

computer circuits depend on the use of the core as a synchronous magnetic amp1ifier.l’ Consider the core of Fig. 16 with an input and an output winding and with

Fig. 16-Principle of voltage-driven circuit (Ramey).

diodes in each circuit as shown. Power is provided to both windings, i.e., ac sine waves of the same frequency and phase. Square waves or any other symmetrical waves can be used also. The two voltages Es, and EL, have at any instant the relative polarities shown and have amplitudes proportional to the number of turns of their respective windings. The operation of the amplifier is in two steps: reset and gating. During the half-cycle with polarities as indicated in the figure, current can flow only in the input or reset circuit because of the direction of the rectifiers. This causes the magnetic core to proceed from saturation (or reset state) an amount dependent on E, and E,‘,. If E, is zero the core is switched over by EL,, if it is equal to EL, there will be no voltage across the cores since E, and EL, will cancel so that the core will not switch over. When E, is zero and the core switches over, there is no current in the output winding even though the direction of the diode would permit it because the voltage E,,, which is greater or at least equal to the voltage induced on the output winding, keeps the diode cutoff.

The second or gating step is in the next half-cycle. The voltages are the reverse of those shown in the figure. If the core was not switched. over in the preceding step (E, = EL,), then the voltage E,,, which tends to reset the core, appears almost entirely on the load while the load current simply brings the core further into saturation. If the core was switched over in the preceding step (E. = 0) then the voltage E,, will appear almost entirely across the output winding with only a small voltage drop across the load, due to the magnetizing current. It is seen, therefore, that if there is no input (E, = 0) there is practically no output (EL = 0), and if there is input (E, = EL,) there is

I1 R. A. Ramey, “Magnetic amplifier circuits and applications,” AIEE Conference Paper, 1953.

an output (E, = E,,). In the conversion of the intelligence signal from the input to the output several quantities can change. 1) The voltage E,, can be greater than EL,, 2) there can be more power in the output than the input because the input needs to provide only enough power to magnetize the core while the output can provide far in excess of that amount, and 3) the output occurs one-half cycle after the input.

It is of interest to comment on these three points. 1) The voltage E,, can be greater than EL, simply because the core acts as a transformer. If the output of one stage is to drive another it is advantageous to make E,, greater than EL, in order to have some excess voltage to allow voltage losses in the coupling loop. 2) There is more power in the output than the input because an artifice is used to remove the load while the input switches the core and to insert it back in the circuit when the output switches the core back. The artifice here consists of biasing off the output diode, connected to be otherwise conducting. In current driven circuits discussed previously the artifice consisted in connecting the diodes so as to connect or disconnect the load depending on whether it was the gating or control cycle. The input can be thought of as “priming” the core so that output can be obtained during a subsequent ‘(drive” period. 3) In magnetic amplifiers, there is always a delay between the output and input signals, of the order of a few cycles of the high- frequency power supply, required to allow for buildup. Here the delay is only half a cycle because the signal is synchronous with the power supply frequency. The delay, undesirable in conventional magnetic amplifiers, is exploited to advantage in computer applications.

A commutator switch, or a magnetic delay line” as originally called, can be made by connecting many such one-core-magnetic amplifiers in a row, as shown on Fig. 17. Here explicit use is made of the discrete time delay in

_ STAGE I STAGE 2. STAGE 3. STAGE 4. STAGE N

Fig. 17-Voltage-driven commutator switch (Ramey).

each stage as well as of the intrinsic gain which prevents the degradation of the signal from stage to stage. The operation of this counter can be described in the following manner. A pulse appearing during a reset half-cycle on the input of stage 1 is reproduced on the output of stage 1 on the second half-cycle; this provides an input signal pulse for stage 2 and so the pulse is reproduced on the output of stage 2 on the third half-cycle, etc. When a core is being switched over, the preceding and following stages must be decoupled, as was explained in connection with current driven commutators. The decoupling is obtained by forcefully biasing off the output diode when the input is allowed to conduct and vice versa.

Voltage driven circuits have been investigated quite

1957 Rajchman: Solid-State Devices for the Manipulation of Information 219

thoroughly and an all-magnetic computer was built recent1y.l’ It is made of unit blocks each including basically a voltage-driven single-core magnetic amplifier and a number of diodes. Special metal-tape-wound cores are used. The bobbins are stainless steel rather than ceramic. Some of the diodes provide the necessary .elements required for obtaining gain and isolation from other units, while others are used for logic switching. Only a very few different standard units suffice, with proper interconnection, to produce almost any desired arithmetic or logic switching function.

It is interesting to note that the current from the voltage source has many parallel paths: the selected branch and all the unselected branches. The cores in all the unselected paths switch over. This requires for each the magnetizing switchover current. In the selected path the current is much greater. If we consider the energy lost in the disturbed cores we find that it is (n - 1) 1.4,. It is interesting to compare this to the power wasted in disturbed cores in the current driven commutators. There, half the cores are disturbed elastically by a flux & resulting from the current I required to switch over the selected core. Therefore the energy loss is l/2 (n - 1) I+,. Because C& is much less than &, typically 30 to 50 times in good materials, the current-driven commutator wastes much less power in disturbed cores than does the voltage driven one.

In the magnetic logic circuits described above, the power originates from a central source, all elements of the circuit being passive and dissipating power. They require only a very few driving tubes by centralizing all sources of gain and power and, therefore, minimizing the expense and possible trouble. This is in keeping with the tube-age pretransistor philosophy, that the complexity of a circuit is measured by the number of tubes, the tubes being the most expensive and the least reliable element of the circuit. With the advent of the transistor this philosophy may no longer be valid, as it becomes practical to con- template many local elements with real gain interspersed between passive elements.

An example of such a circuit is a magnetic shift register with transistor coupling between cores as shown in Fig. 18. Actuation of one of the. advance windings, e.g., A, causes the abnormal core (or cores) to switch from p to n, as in the previous example of a core-diode shift register. This induces a voltage on the winding connected to the base of the corresponding interstage transistor and thereby makes the transistor conducting. The relatively large collector current of the transistor switches over the next core from n to p and thereby effectively causes a shift of information. It is possible to improve the operation by a feedback coupling winding (shown dotted on the figure), on the core being reversed by the advance winding, by means of which the collector current of the transistor not only tends to switch over the next core, but also helps

UT. H. Bonn, “The megacycle ferractor,” 1956 Western Joint Computer Conference, San Francisco, Calif.; February, 1956.

in turning over the core being actuated by the advance winding. This produces a “snap-action” and results in a greatly redused value of required advance current, which now needs only to “trigger” the core.

The transistor-coupled magnetic shift register is an example of the combined use of magnetic cores and transistors. The cores are used primarily for storage and the transistors primarily for gain. This type of circuit may well turn out to be, in the present state of the art, the best engineering compromise for a great variety of applications.

“-II----l11

f

Fig. 18-Transistor-coupled shift register.

In comparing the relative merits of transistors and magnetics in storage and switching circuits, no absolute dogmatic stand should be taken. It seems fairly obvious that storage of very large numbers of bits is best done using magnetic devices as the intrinsic storage medium, where the price and reliability of each element are of prime importance. On the other hand, a logical or arithmetic circuit by itself is best realized using transistors, which allow great versatility of gating and have gain at every local gate. Between these two extremes lie a large number of applications for which either transistors or magnetics can be used. For many of these a combination of the two techniques may well turn out to be the best.

THETRANSFLUXOR

The only magnetic device mentioned so far has been a simple toroidal or ring-shaped core. It turns out that the switching and storage capabilities are greatly broadened by the use of cores with several apertures arranged in various geometrical configurations. A device was developed which operates by the control of flux between the legs of the magnetic circuit formed by the apertures and hence has been called the “transfluxor.“13,14 The simplest type of transfluxor comprises a core with two apertures..

To understand the principle of this transfluxor, let .us consider a core with two apertures of unequal diameter which form three legs, numbered 1, 2, 3 in Fig. 19. Assume

I3 J. A. Rajchman and A. W. Lo, “The transfluxor-a magnetic gate with stored variable setting,” RCA Rev., vol. 16, pp. 303-311; June, 1955.

I4 J. A. Rajchman and A. W. Lo, “The transfluxor,” PROC. IRE, vol. 44, pp. 321-332; March, 1956.


the more distant leg 3, remain saturated in the original direction. After such a setting, the alternating current on winding W,, at its first proper phase, will bring back the whole of leg 2 to its original direction of saturation and thereby cause flux in leg 3 to decrease by the amount that flux. in leg 2 is increasing, which is precisely equal to that initially “set” by the setting pulse. At the next phase, leg 3 will be saturated to its original direction and the amount of set flux will be transferred back to leg 2. On successive phases the amount of flux set-in will be transferred back and forth between legs 2 and 3. This will cause voltages to be induced in the output winding W, which are thus seen to be determined by the amplitude of a single setting pulse. Even though the transfluxor utilizes material which is always saturated completely in one or the other direction, continuous or “analog” control is possible because the position of the limit between the regions of opposite saturations can vary continuously. By saturating the whole width of leg 2 “on-off” or “digital” operation can be obtained also.

UNBLOCKED PARTIALLY The operation of the transfluxor can be compared to SET

Fig. 19-Principle of the transfluxor. the action of a circus juggler who is ready to throw, back- and-forth between his hands, a ball of any particular size once it is set into one of his hands. His monotonous motions are in vain with a zero-size ball (no ball) and remain imperturbable whatever the size (below some maximum) of the ball given to him.

that at first a current pulse is sent through winding W,, and is intense enough to saturate legs 2 and 3, but not necessarily leg 1, which is deliberately made to have a larger cross sectional area. The two legs (2 and 3) will remain saturated.after the termination of the pulse, since remanent and saturated inductions are almost equal in square loop materials. Consider now the effect of an alternating current in winding W,, which produces an alternating magnetomotive force along a path surround- ing the smaller aperture, as shown within the dashed line in Fig. 19. When this magnetomotive force has a clockwise sense, it attempts to produce an increase of flux in leg 3 and a decrease in leg 2. But no increase of flux is possible in leg 3 because it is saturated. Con- sequently, there can be no flux flow at all, since magnetic flux flow is necessarily in closed paths. Similarly, during the opposite sense of the ac, the magnetomotive force has a counterclockwise sense and attempts to produce an increase in flux in leg 2, which is again impossible since that leg is saturated. Consequently, there is no flux flow at all and no voltage induced on the output winding W,,. The transfluxor is “blocked.”

A more technica description considers the amount of flux which can be reversed back-and-forth around the small aperture, or “exchanged” between legs 2 and 3. This is best characterized by a conventional hysteresis loop relating the instantaneous flux to the instantaneous magnetizing force (on leg 3) producing it. For every setting, there is a different amount of exchangeable flux and a different loop. A typical family of loops, as observed on an oscilloscope, is shown in the photograph of Fig. 20.

Fig. 20-Hysteresis loop of transfluxor output for different settings.

The Ioops vary in size from the nearly horizontal line obtained in the blocked condition to the Iargest loop obtained at maximum setting. It is apparent that the transfluxor operates as if the output -magnetic circuit consisted of a conventional one-apertured core with the essential property that the effective cross sectional area of the core can be adjusted by a single set pulse to any desired value from practically zero to a maximum value equal to the physical cross sectional area of the core.

. Assume now that a “setting” pulse of opposite polarity

from that of the first “blocking” pulse and of smaller but prescribed amplitude is applied through winding W,. The resulting magnetizing force around the larger aperture diminishes gradually with radial distance. Because a critical value of this force is required, in square loop materials to produce flux reversal, such reversal is possible up to a critical radius only. The amplitude of the setting pulse determines this radius and can be prescribed so that any desired portion of the width of leg 2 reverses its direction of saturation while the other portion, as well as

The amount of flux set by the single setting-current pulse depends on the radius of the circle within which the magnetizing force is greater than the critical value required for flux reversal, and outside of which it is smaller. The radius of this imaginary circle is proportional to the

Fig. al--Setting characteristic of transfluxor.

amplitude of the setting pulse. Therefore, below a definite threshold of setting current there is no flux set because the circle is within the large aperture. The amount of flux set increases almost .linearly as the circle sweeps the width of the intermediary leg 2. For larger values, flux is set in leg 3 as well as in leg 2, so that the amount of inter- changeable flux between legs 2 and 3, decreases as the device begins to be blocked in the opposite direction. This explains the typical shape of the plot of the useful reversible output flux vs the amplitude of the single setting pulse, Fig. 21.

The output magnetic circuit of the transfluxor is similar to a conventional transformer. It is to be expected, therefore, that power can be transmitted with high eEi- ciency from the primary to the secondary wasting only a small part for magnetizing the core. This is possible, but requires an artifice to prevent malfunction. A transfluxor originally blocked could be spuriously unblocked by flux reversing around the outer legs 1 and 3 in a sense opposite to original blocking, thereby by-passing’the blocking effect of the intermediary leg 2. Therefore, the energizing magnetomotive force of this sense, must be kept below a value capable of such reversal but large enough to produce the flux reversal around the small aperture, via legs 2 and 3, required when the transfluxor is properly set. In the opposite sense, the magnetomotive force cannot spuriously unblock, since it tends to bring the blocking leg 3 further into saturation. It can, therefore, be arbitrarily large and provide not only the required reversing magnetizing force around the small aperture, but also substantial power to a load in the secondary. With such unsymmetric energization, conveniently obtained with pulses rather than sine waves, efficiencies of 80 per cent are typical. There is a “priming” pulse of limited amplitude followed by a “driving” pulse. The drive pulse can be very short, e.g., a tenth of a microsecond, and ‘produce. by rapid flux reversal reasonably high voltages from small cores. Of course, sine wave energization, biased or unbiased, can be used also. The

Rajchman: Xolid-State Devices for the Manipulation of Information 221

pulse repetition rate or frequency of the sine wave is limited by power dissipation in the core to about a megacycle for typical transfluxors.

The first applications of the transfluxor exploited the “nondestructive” readout properties of the new device. Readout from any magnetic storage device must necessarily be dynamic, since induced voltages result from change of flux. In a conventional memory core the stored information is ascertained by changing it, i.e., by destroy- ing it.. However, in the transfluxor this need not be the case because the flux in the larger leg 1 is not altered by the “interrogation” pulses, retains at all time the stored information, and yet its value determines whether or not flux in legs 2 and 3 will be interchanged as a result of interrogation. A random access-memory system using an array of transfluxors (Fig. 22), is the outstanding example

WRITE ADDRESS LINES \

L I < READ ADDRESS LINES

Fig. 22-Array of transfluxors.

of the use of the nondestructive readout capability of the transfluxor. The two stored states are the blocked and unblocked states. Current coincidence can be utilized for addressing the selected transfluxor both for write-in and readout. For writing, pulses are applied simultaneously to one row and one column winding linking leg 1. The additive effect of these pulses on the selected transfluxor produces a setting. Because these pulses are kept below the threshold of setting, they have no effect on any other transfluxors. The direction of the writing pulses determines whether the transfluxor is set to the blocked or unblocked condition. Readout is obtained by applying pulses of the proper amplitude to the selected row and column windings linking leg 3. A readout is obtained by a pair of oppositely directed pulses on each selecting line. As a result, fluxes in legs 2 and 3 reverse back and forth and return to their initial state if the transfluxor is unblocked. If the transfluxor is blocked, there are no such reversals. The flux reversal, if any, is detected as an induced voltage on a common winding linking leg 3 of all transfluxors. The nondestructive readout eliminates the

222 IRE TRANSACTIONS

rewrite circuits required in conventional core memories. Further, it makes possible the simultaneous write-in and readout in different addresses.

The transfluxor can be compared to a latching relay: a pulse will set it to either close or open a circuit until a new pulse sets it differently. A crossbar electromechanical switch is an example of an array of latching relays so arranged that any ones of a number of row channels can be connected to any ones of a number of column channels depending on which of the relays in the array are closed. The array of transfluxors of Fig. 22 can be utilized for the same purpose. Pulse or amplitude modulated signals applied to the column windings linking leg 3 will be transmitted to those row windings linking leg 3 which are coupled through unblocked transfluxors, but not those coupled through a blocked transfluxor. The setting of the transfluxors can be obtained by using the column and row windings linking leg 1. Transfluxors permit much faster operation than latching relays and are much more reliable, as there are no contacts to wear out. This application of the transfluxor typifies its general use ‘as an on-off device: to transmit on-off or continuously modulated signals. The on and off settings are established by a single pulse and require no holding power to be maintained.

Another application of the transfluxor is for a channel commutator. A device for opening in succession one channel among many can be built using a row of transfluxors. The large aperture is considered as though it were a simple core and is connected exactly as the cores of a shift register.. The current-steering type of connections are particularly suitable. After the termination of an advance pulse one of the transfluxors is set while all others are blocked. Consequently, modulated signals can be transmitted through that channel only and will con- tinue to be transmitted until a new advance pulse selects the next channel.

The analog storage property of the transfluxor, or the capability to control according to a stored level in a continuous range, can be used to control a resonant circuit. For example, the inductance of the circuit can be connected in series with a winding on leg 3 of the transfluxor. For small signals the control is obtained by virtue of the fact that the effective permeability is different for different remanent states, and any desired remanent state can be obtained with proper setting. For large signals, the inductive component, depending on the flux excursion, and the resistive component, depending on the resulting hysteresis losses, have different values for different settings and thereby exercise control of the resonant circuit.

The transfluxor can also be used for current steering, instead of using a core-diode combination.” Transfluxor steering is achieved by the use of a transfluxor winding linking leg 3 connected in series with a branch load in each of a number of parallel branches. This is illustrated in Fig. 23 for the case of two branches. The steering action can be considered to be due to the difference in effective impedances of a blocked and set transfluxor, so

ON CIRCUIT THEORY September

SET SET

SET TRANSFLUXOR BLOCKED TRANSFLUXOR T, T2

Fig. 23-Transfluxor steering.

that by setting the two transfluxors to different states a preferential current flow is obtained. However, diodes are not required since the setting of the transfluxors causes no voltages in the output windings. The operation of the transfluxor steerer can be understood by following the detail of current drives and resulting flux changes. When the two transfluxors are blocked, as in Fig. 23, the DRIVE current will not be steered, but rather be divided between the two branches, equally if the loads are equal. If one transfluxor is set, e.g., T, in Fig. 23, the DRIVE current I, divides into a relatively smaller current I, and larger current I,. This division results from certain flux changes in the transfluxors.

.

To understand the relation between I, and I, during DRIVE, assume first that both branch loads are zero. Then the transfluxor output windings are directly in parallel and any voltages generated by the changing of flux in T, and T, must be equal. Consequently the total flux changed in each branch at every instant must also be equal as they occur at the same time. Since T, and Tz operate on different B-H loops, transfluxor T, being set and T, blocked, it takes different magnitudes of current to produce equal flux changes. For reasonably square loop magnetic material the blocked transfluxor T, requires a large current Iz to produce the same flux change that a smaller current 1, produces in the set transfluxor T,. The ratio of currents I, and I,, defined as the discrimination, in a real circuit with equal loads will start at a high value and tend asymptotically to unity as the currents I, and I, become equal. This results from the fact that eventually there are no longer any. flux changes and the currents divide according to the load impedances. The discrimination is low for large loads since the voltages

.

1957 Rajchman: Solid-State Devices for the Manipulation of Information 223

across the transfluxor windings are a small part of the total branch voltages. Consequently, the loads must be reasonably small to obtain good discrimination.

Transfluxor steering of current through one branch of a parallel pair can be repeated in many pairs connected in series and is directly applicable to driving a decoding switch. The example of a decoder with n = 3 inputs and 2” = 8 outputs used to illustrate the tube and core-diode decoders is used again in Fig. 24 to illustrate the transfluxor steering decoder. After one transfluxor in each pair is set by the inputs, the DRIVE current is steered through the branch of each pair including the blocked transfluxor and thereby causes a net reversing mmf on only the single selected output core.

SET

Fig. 24-Transfluxor steered decoder switch.

Transfluxors have been made with more than two aperturesI The many possible modes of flux transfer between the various legs of the magnetic circuit were exploited to produce interesting storing and switching functions beyond the capabilities of the simplest two-hole transfluxor. For example, a device was made to respond to a sequence of pulses, ABC, but not to the occurrance .of these pulses in any other order such as BAC or BCA. -Another example is a straight AND gate, in which all of .a number of signals must have occurred for the device -to be set.

A third aperture added to the two of the transfluxor *described earlier in detail, can eliminate any possibility of oversetting. The four legs I, 2, 3, and 4 of the trans- Auxor illustrated in Fig. 25 are of equal cross section. The .amount of flux that can be set is limited by the width of leg 1 to precisely the amount which leg 3 can accept. ‘Therefore, when the setting pulse on leg 1 becomes larger than required to transfer flux to the first-filled closer leg 3, no further flux is available for transfer to leg 4. In this -four-legged transfluxor leg 2 is a dummy which remains .always saturated in the same direction and provides the necessary return path to satisfy continuity of flux flow.

Many other multiaperture configurations of transfluxors -have been made and various switching and storing func- :tions were obtained thereby. Still a good field of research

BLOCK PfI1”E

DRIVE

(4

AFTER BLOCK AFTER SET

(b) Cc) Fig. 25-Elimination of oversetting with an additional aperture.

is offered by the study and exploration of the different flux distributions in cores made of square loop material and having complex geometrical configurations.

OTHER SOLID-STATE DEVICES

There are many other solid-state phenomena which are potential candidates for exploitation in storage and switching circuits. The semiconductor properties of diodes and transistors on one hand, and the magnetic properties of ferrites and metals on the other, are the most important effects used today and have been treate’d in some’detail in this paper. Some of the other contenders which have been proposed and on which some exploratory research has been done are worth mentioning briefly.

The phenomenon of ferroelectricity has been studied in its application to switching and storing devices.” This phenomenon is analogous to ferromagnetism in that the electric polarization in certain materials is related to the electric field in a manner similar to that in which the magnetic induction is related to the magnetic field. There is also hysteresis as is the case with the ferromagnetic materials. When a varying electric field is applied to a ferroelectric condenser, the integrated current that flows through the condenser can be thought of as being a charge trapped in the condenser and this charge is related to the electric field in the manner of the familiar hysteresis loop. There are ferroelectric materials with almost rectangular hysteresis loops. When a condenser made of such a material is driven further into saturation, practically no current flows through it and it appears to have a small capacitance. If, on the other hand, the condenser is driven in the opposite direction, a large amount of charge will flow to produce the reversal and the condenser will appear momen-

15 J. R. Anderson,. “Ferroelectric storage elements for digital computers and switchmg systems,” Elec. Eng., vol. 71, pp. 916-922; October, 1952.

224 IRE TRANSACTIONS ON CIRCUIT THEORY Septewbber

tarily to have a large capacity. The best-known material exhibiting ferroelectricity is barium titanate. Recently many other materials have been discovered, and some are most promising (GASH, Glycine Sulfate, Colominite, Thiourea).

One of the proposed applications of ferroelectricity is for very compact memory arrays. In one proposa1,15 a crystal of barium titanate is, coated with parallel conductive lines on one side, and with a set of parallel conductive lines on the other side at right angles to the first. The application of voltages on one column and one row conductor produces’ a voltage difference on the element at the intersection which is twice that which appears on the other elements of the selected lines. Consequently, if the hysteresis loop is reasonably rectangular, a voltage- coincident system can be used which is similar to the current-coincident magnetic system. Because no third electrode is available on the condenser, the readout cannot be obtained from an ad hoc electrode and one must resort for readout to the artifice of measuring the intensity of the currents which flow in the selecting lines when selecting voltages are being applied. Successful units with 256 elements have been made in a ‘size less than a square centimeter. Because of the lack of versatility caused by having only two electrodes, and difficulties connected with the stability of the material, large memory systems using ferroelectrics do not appear as practical as their magnetic counterparts at the present time.

The relatively small energy required for switchover makes ferroelectric elements attractive in some types of logical switching. Experimental ferroelectric shift registers” and ferroelectric counters have been described recently. In these circuits there are isolation problems between stages similar to those encountered in magnetic circuits which require the use of diodes. Actually both Zener and conventional diodes are used in the ferroelectric registers. The speed of the ferroelectric register is very much less than its magnetic counterpart, but it requires much less driving power.

Another interesting application of the ferroelectric condensers is an attempt to produce the same storing and gating function as the transfluxor, in a device which may be called a “transcharger”. Three cells are used to simulate the operation of a three-legged, two-apertured transfluxor (Fig. 26). The device can be thought to work as follows. When two cells connected in series are both polarized in the same direction, an ac voltage applied to the combination will reverse the polarization back- and-forth and the two cells will behave as though they were a single thicker cell. If, however, the polariza- tions of the two elements are opposite, then one or the other cell will behave as an infinite or very large impedance so that neither one will reverse its polarization. The third cell in the circuit can be used in order to set the two first

I6 J. R. Anderson, “A new type of ferroelectric shift register,” IRE TRANS., vol. EC-5, pp. 184-191; December, 1956.

BLl%lYNG SETTING

BLOCKING AND &C SETTING PULSE

Fig 26-Transcharger: a ferroelectric analog of the transfluxor.

cells into the parallel or antiparallel condition in such a way that after the setting it becomes an infinite impedance into which no charge is being deviated. Successful devices of this type have been built.

It is interesting to note that while the analogy between the ferromagnetic and ferroelectric devices is almost complete, there are two rather fundamental differences. 1) In a magnetic circuit of a core of arbitrary configura- tion, the algebraic integral of all the flux traversing a plane that intersects the core is precisely zero. For example, the sum of the fluxes of the three legs of the transfluxor is always zero. On the other hand, in a number of ferroelectric condensers connected without sources or sinks of charge, the total of the charge remains constant, but does, not need to be zero, and can have any arbitrary value set into the system initially. 2) In ferroelectric devices there is no possibility of transformer action by means of which the charge transferred from cell to cell could be multiplied by some artifice akin to windings turn-ratio used in magnetics.

At the present time, ferroelectric devices do not have the importance of ferromagnetic devices, but in some special applications such as a low-power-drain shift register they may find an important place. It is also possible that combination of ferroelectric and ferromagnetic devices will be very fruitful.

Other phenomena which have been proposed for new devices are electro-optical effects.17 Some solids, called photoconductors, change their conductivity upon illumination. A great deal of work has been done recently to obtain efficient photoconductors. Cadmium sulfide powders are among the outstanding of these materials. There are other solid materials, electroluminescent phosphors, which produce light when an electric field is applied to them. Electroluminescence was discovered about a decade ago and since has been the subject of extensive investigations.

17 E. E. Lbebner, “Opt0-electronic devices and networks,” PROC. IRE, vol. 43, pp. 1897-1906; December, 1955.

1957 Rajchman: Xolid-State Devices for the Manipulation of Iiaformation 225

Reasonably efficient screens have been obtained with powders of zinc sulfide imbedded in suitable plastics. Consider a photoconducting cell in series with an electroluminescent cell, and a fixed driving ac voltage applied to the combination. This constitutes a light amplifier since both the input and output signals are light signals. The gain, or ratio of output-to-input light intensity, can be made comparatively high with the proper cell materials and construction, and with sufficiently high voltage and frequency of the driving source.

Under certain conditions, when the output light is made to feed back to the input, two stable states are obtained: one in which there is practically no light and the other in which the light is relatively intense. To understand this, consider the variation of the gain as a function of the input for an amplifier without feedback. At low light inputs, the photoresistance is high and most of the voltage appears across it leaving a very small voltage on the electroluminescent cell. Because the light output of the cell increases very rapidly with applied voltage, at low voltages there is relatively very little light. As a consequence, the gain of the amplifier is very low. As the light input increases, so does the gain until any further increase in illumination of the photoconductors no longer produces any significant increase in the voltage on the electroluminescent material, it having reached almost the supply voltage. When this is the case, the output remains constant while the input keeps on increasing, and the gain drops. It is quite clear, that if feedback is allowed, the device will settle at one of the two possible light levels for which the gain is unity.

A number of complex logical functions can be accomplished by various electrical connections between such feedback cells. Not only are logical manipulations possible electrically, but also optically, by allowing certain photo- cqnductors to be exposed to certain electroluminescent cells and be masked from certain other cells. Shift registers and selecting combinatorial switches have been proposed.

At the present time, the important limitation on electro- optical devices: is the speed of operation, which is limited by the response of the photoconductor. The time constant of efficient photoconductors is of the order of tens of milliseconds. The attractive feature of the optical link is the absolute electrical isolation that can be achieved between the different elements which are coupled merely by light.

Another approach comes from the field of cryogenics. A superconductive computer element-the cryotron- was recently described.l* The basic idea is based on the fact that a superconductor becomes a normal conductor in the presence of a magnetic field. The transition between the conducting and superconducting state of the material depends on the value of the magnetic field. The current required to generate a magnetic field intense enough to

18 D. A. Buck, “The cryotron-a superconductive computer component,” PROC. IRE, vol. 44, p. 482; April, 1956.

control transition can be easily obtained through the superconductor itself. The cryotron can in fact act as an amplifier. In typical experimental cryotrons a tantalum wire is used as the gating superconductor. A single layer of niobium wire is wound on the tantalum. At the temperature of the liquid helium in which the cryotron is immersed the niobium remains superconductive for all values of magnetic field while the tantalum is gated from the superconductive to the conductive state. By connecting the gating element of one cryotron to the coil of another and vice-versa, a flip-flop is obtained. All the logical switching and storing functions have been shown to be realizable with cryotrons. The commercially developed helium liquefiers make the once formidable task of obtaining temperatures near absolute zero relatively easy. At present the limitation of cryotrons is chiefly in their speed of operation, there being an inductive-resistive time constant of the order of milliseconds. However, the possibilities of reducing this time constant are promising. It is likely that the cryotron will become an important computer element.

CONCLUSION

This review of the salient examples of magnetic storage and switching elements shows what great versatility is offered by the new art of magnetics. Competing solutions offered by diodes and transistors in the equally exciting new field of semiconductors bring up the obvious question of relative merits.

In general, it is conceded that storage is better accomplished with magnetics since it is an inherent property of the materials used and requires no artifice, and switching is better accomplished with semiconductor devices which have inherent gain, and also, usually sharper thresholds of switching. Switching by magnetics generally requires more power than by semiconductor devices. In spite of this, magnetic switching is preferable when the number of elements of the switch is very large. This is particularly true for the address switches of magnetic memories. It is quite feasible to make magnetic switches with tens or even hundreds of thousands of outputs, whereas switches of similar size made with present day semiconductors would be prohibitively expensive and not as reliable. Magnetic switching is preferable also in many special cases as, for example, when an absolute degree of reliability is essential or when the desired switching function is intimately related to a desired storing function as in the transfluxor.

It is likely that the transistor and magnetic techniques will complement each other and will be dominant in the near future over other techniques. A breakthrough in the newer solid-state contenders, such as ferroelectrics, photoconductors, and electroluminescent materials, or magnetorestive superconductive devices, may cause a new imprint in the already explosively growing field of solid-state manipulation of information.

Documents

A Survey of Magnetic and Other Solid-State Devices for the Manipulation of Information