8 Impulse Switching of Ferrites J: 0 -' w 4 Li: o EMORY designs at present reflect either the applications, size, or environment of the memory. Many desired memory characteristics oppose each other in the determination of the memory design and there has been no general solution to the basic problems associated with magnetic storage. Lincoln Laboratory's concern primarily has been to improve the memory speed, and hopefully obtain improved efficiency and design freedom. It has only recently become clear that these three factors, speed, efficiency, and design 'flexibility, are closely related and in fact lead to generalized memory designs. Efforts at memory improvements have been concentrated in two areas; the magnetic elements, which is called the elemental approach, and the nonelemental sections or memory proper, called memory methods. Recent memory methods include techniques such as Load Sharing,t Set a Line,? Anticoincident Current, and Linear Selection schemes. The elemental approach includes the work on thin films, Twistors 3 and multiaperture devices. 4 M Memory Improvements In order to obtain improved speed capabilities, 1inear selection memory designs have been utilized. There are also other important advantages of linear selection that have accelerated its use. However, it was clear that memory methods alone would not provide the speed improvements desired. An elemental approach that has been named Impulse Switching has provided added memory speeds that are an order of magnitude improved over previous methods. To evaluate the technique, ferrite cores are currently being used as the magnetic element; however, impulse switching is a mode of operation not restricted to ferrites. Basically, impulse switching utilizes current drives that are controlled in both amplitude and width. By adjusting these parameters it is possible. to control how much of the core switches, and in effect, there is a partial switching of the core. In the 5O-30-mil size core, for example, there is 10 mils of material available for complete switching, but switching may be restricted to as little as 1 mil of material. This technique might be described by saying that impulse currents electrically, rather than physically reduce the size of the core to improve the speed. ~ I I ] 50- 30 CORE 1 ~ 0 R. E. McMAHON I [ I l '0 c:,~~' ~ ~ NORMAL CURRENT DRIVE 3 -' 15: 2 « -1 Impulse Switching Characteristics 0 The general characteristics obtained by impulse switching will best describe its capabilities and application. The switching constant (Sw) for impulse driven cores shows an effective decrease relative to the Sw obtained with normal drive currents. Fig. 1 shows the Sw curve for both cases. There is, in practice, a family of Sw curves representing different degrees of partial switching. The switching time ts is defined as where Hm = applied field Ho = threshold field for irreversible domain wall motion and is approximately equal to He, the coercive force. A reduction of Sw as shown in Fig. 1 provides a proportional decrease in the switching time. There is also an apparent change in coercive force He when impulse currents are applied. As the drive current widths are decreased, the current magnitude necessary to initiate switching increases proportionately. This improves the speed still further, since the switching time is inversely proportional to the applied field. The pulse characteristics are obtained by applying a write current variable in magnitude and width. An additional current called an exciter (opposite to the normal inhibit) is applied coincidently with the write current. The exciter current is set below the coercive field and is slightly wider than the write current. The coincidence of a write and exciter current cause partial switching during this interval. However, the absence of the exciter of course does not allow switching. The write current is first set at a particular value while the width is reduced to small values causing partial switching. There is a decrease in switching time for both the read and write core outputs and a decrease in core output voltage. A new value of write current is chosen and again the current width is reduced to obtain switching time and output voltage data. Curves of this sort are shown in Figs. 2 and 3. Fig. 4 shows the 4 .!. Ts Fig. 1. 9 (J.< 5 -I 10 ) Applied field versus switching time signal-to-noise ratio for these data. Additional data are taken with the read current set at a new value and again varying the write current in magnitude and width. Figs. 2, 3, and 4 are taken for a read current of 1.0 ampere. Additional curves for read currents of 0.5 ampere, 1.5 amperes, and 2 amperes will change switching time, output voltage, and signal-tonoise ratios considerably. In Fig. 4, the signal-to-noise ratio is obtained by applying a program of partial write and exciter currents in a manner similar to that used in normal core testing. It is observed that in many cases improved signal-tonoise ratios (U v1 , d vz ) are possible for the impulse mode. In Fig. 3 the reduction of read output to zero for a write current of 0.6 ampere at a width of 0.1 microsecond (usec) is an example of the apparent increase in coercive field for decreasing current width. For a write current of 0.6 ampere no switching will occur below O.l-p,sec width. At this width, however, the curves indicate switching will occur if the current is raised to 0.8 ampere. The regions below 70-mpsec (millimicroseconds) widths in Figs. 2, 3, and 4 are difficult to measure because of current rise time limitations. It is also difficult to show a composite picture of these curves, but the following generalizations can be made. 1. Variable core voltage outputs possible at almost any switching time. are 2. Switching times as low as 20-50 mJ.tsec are possible with good signal-to-noise ratio. 3. The write interval switching time is approximately equal to the read switching time. R. E. McMAHON was with Lincoln Laboratory, Massachusetts Institute of Technology, Lexington, Mass., and is now with Transistor Applications, Inc., Boston, Mass. The work reported here was performed at Lincoln Laboratory, a technical center operated by Massachusetts Institute of Technology with the joint support of the United States Army, Navy, and Air Force. McMahon-Impulse Switching of Ferrites From the collection of the Computer History Museum (www.computerhistory.org) 31 .~ .4 w ;; Fig. 2 (left). Read switching time versus write width EXCITER CURRENT 250 MA @ .8p.s READ CURRENT 1 AMP @ .4 EXCITER CURRENT 250 MA @.8p.s READ CURRENT 1 AMP@.4p.s N > ~ 8 :J I- o .3 <!) z I ~ (,) <t 6 ~ 5 0:: I- ~ .2 Cfl o z a <t w o 0:: 7 4 I- ~ .1 o(,) .3 .4 CURRENT WIDTH (p.s) .5 .6 SigFig. 4 (right). nal-to-noise ratio versus write current width .1 .2 .3 WRITE CURRENT .4 .5 .6 WIDTH (p.s) General Advantages These three facts represent the advantages of impulse switching, and in fact, demonstrate that memory speed, efficiency, and design flexibility are controllable for the impulse switching mode. Since the voltage output level of the core is adjustable at any speed setting, the input power required for a particular switching speed and the input power required for a particular voltage output are effectively separated. If, for example, a memory design requires a read switching time of 0.1 J.lsec, and an output voltage level of 15 mv (millivolts) is satisfactory, then by choosing a proper current and width relation, this may be obtained. The total power required is less than would be the case for obtaining this speed by simple overdrive methods. In overdriving methods, an unnecessary output voltage level of 500 mv would result indicating waste power. The write interval switching time in impulse switching is approximately equal to the read switching time. The degree of partial switching is controlled during the write time and the function of the read current is simply to read out completely. The fast switching speed during the write time represents an improvement in power efficiency as well as speed. Previously, > , ~ UJ o :::> I- :::i WRITE 350 I AMP 300 CURRENT memory speeds were improved by overdriving during the read time while the write operation being of a coincident nature, allowed no overdrive. The read switching time could be an order of magnitude faster than the write time indicating the large power applied during the read time is wasted if the complete memory cycle time is considered. The average memory power of impulse switching is also low even at cycle times of 0.5 J.lsec because partial switching results in lower duty cycles. Calculations of peak and average power indicates that impulse switching compared with other elemental or memory methods (at comparable signal levels) requires much less power. The switching times in the region of 2(}-50 m,usec, although almost unusable with present memory circuitry, represents possible memory cycle times of 0.1 to 0.2 usec. Switching times below these values for ferrites represent the crossover point between wall motion and rotation. a Nondestructive read out conditions exist in the speed region below 20 m,usec and is worthy of further considerations. The previous discussion should also suggest added design flexibility of the memory circuitry-. Previously, driver limitations imposed design restrictions on the memory size. Using impulse switching, VALUES + \ .6 AMP .4 AMP IL. ~ 250 ~ 200 {EXCITER CURRENT} 250 MA@.SILS IL. I- 5 150 {READ CURRENT} I AMP @.4ILS ~ 100 UJ a: 50 UJ a: o u .2 WRITE 32 .3 :4 CURRENT WIDTH .5 IlLS) .6 Fig. 3 (left). Core read output voltage versus write width the memory element operation is variable and may be used to eliminate driver design problems. The variable output voltage level can also be adjusted to provide sense amplifier design freedom or overcome system noise level difficulties. Related Problems A program has been initiated to provide a better understanding of impulse switching and in particular, the mode of operation and the mechanism that are involved. Recent experiments indicate that the model of partial switching assumed is valid. The impulse switching process has been experimentally observed to initiate at the core center and expand outward. The physical amount of core material involved in the switching process is directly controlled by the current amplitude and width. Experimental observations indicate that irreversible domain wall motion is the major contribution to this mode of operation. The rate of wall motion and consequently the distance moved is proportional to the applied field and the duration of the field. Calculations of the average distance moved by domain walls agrees with experimental evidence of the amount of material switched. s Reversible wall motion occurs if either the current amplitude or width is below a critical value. The applied energy must be sufficient to move the walls beyond certain potential levels or reversible motion results. This accounts for the nonswitching conditions described earlier at current amplitudes and widths below critical values. In addition at very small current widths (below 20 m.usec), no switching will occur until the current amplitude is increased orders of magnitude above the expected or calculated value. There is no wall motion possible below a critical current WIdth. The mode of operation ob- McMahon-Impulse Switching of Perrites From the collection of the Computer History Museum (www.computerhistory.org) tained for current widths below 20 m,usec and very large current amplitude is one of rotation. The switching control and modes of operation that exists are providing an excellent study method for magnetic phenomenon. The current width control and its effect on margins has so far not been a problem. There is considerable variation possible in the current width at a particular speed setting, without a decrease in margins. The heat problem presumed to exist at these speeds has been found to be greatly reduced in the impUlse mode of operation. The power associated with the core may be expressed as, Ls 80 CORES 400 rna PNPN TRIGGER PULSE PNPN 20V @ 20 p-a Fig. 5. P-n-p-n sequential driver where V = volume of material switched j=frequen'2Y of operation Vout = core output voltage 1m = applied current ts = switching time 1 amp .2fLs For impulse switching, the volume of material and the switching time are very small compared to their values for normal high speed methods. For example, the power for an 51 50-30 core at one megacycle using standard overdrive methods is 40 milliwatts (mw) , while at the same speed, utilizing impulse switching, the power is 0.3 mw even with an output voltage of 50 mv. This low power level results in very small temperature increases. The memory circuitry required for the impulse switching has presented some problems. To obtain maximum operation, current rise times of 10-40 m.usec are necessary at current levels of 0.5 to 1.0 ampere. At present the circuit of Fig. 5 using p-n-p-n elements is used to provide this current. The circuit is restricted to repetition rates of about 30 kc, but is in use in a sequential I,OOO-word 80-bit 0.5 ,usee memory under construction. The circuit of Fig. 6 will operate above one megacycle and will be used for random access memory applications. References 1. A LOAD SHARING MATRIX SWITCH, G. Constantine, Jr. I.B.M. Journal of Research and Development, New York, N". Y., Jul. 1958, pp. 20411. 2. FERRITE APERTURED PLATE FOR RANDOM ACCESS MEMORY, A. R?jchman. Proceedings, J. Institute of Radio Engineers, New York, N. Y., vol. 45, no. 3, Mar. 1957. 3. THE TWISTOR, A. H. Bobeck. The Bell System Technical Journal, New York, K. Y., Kov. 19.57. lvlcA,fahon-Irnpulse Switt;hi:ng JL SL iSl -v Fig. 6. Random access driver 4. A HIGH-SPEED LOGIC SYSTEM USING MAGNETIC ·ELEMENTS AND CONNECTING WIRE ONLY, H •. D, Cran~. Stanford Research Institute Report, Stanford, Calif. 5. THE UTILIZATION OF DOMAIN WALL VISCOSITY IN DATA-;HANDLING DEVICES, Vernon L. Newhouse. 1957 Western Computer Conference. 6. NUCLEATION OF DOMAINS OF REVERSE MAGNETIZATION AND SWITCHING CHARACTERISTICS OF MAGNETIC MATERIALS, B. Goodenough, N. J. Menyuk. Engineering· Note E-532, Digital Computer Laboratory, Massachusetts Institute of Technology, Lexington, Mass. Discussion Lloyd Lambert (Aeronutronic Systems Inc.): If you write continuously, do you tend to walk up the loop? Mr. McMahon: Continuous partial write currents will fully switch the core; however in a linear selection memory only one write current is possible before the next read out, so that this is not a problem. In addition the improved signal-to-noise ratios indicate that this one write current and any number of successive exciter currents will not disturb the core. The use of impulse switching in coincident current applications has not been attempted although preliminary work indicates it could be done. Core matrix selection switches have .been designed using impulse switching with excellent results. . G. M. Hyde (Lincoln Laboratories): To your knowledge, is anyone working on a program for a computer to determine the optimum value of switching time and amplitUde? Mr. McMahon: No, but we intuitively know the values of currents and widths. We hope to do some work along these lines later. W. Lawrence, Jr. (International Business Machines Corporation): How critical are pulse widths at O.I-microsecond switching time? . Mr. McMahon: The width variation is of . course the control which allows us to obtain this design flexibility. It can also contribute to a reduction of margins if it varies considerably. We have found that all the circuits so far designed to produce impulse currents display very small variations in current width and no unfaXQ1,<l.ble reduction. in margins exist. r! Ferrit£:$ From the collection of the Computer History Museum (www.computerhistory.org) S3