Space applications of superconductivity: digital electronics

This is the fourth of a seven part series on the potential applications of superconductivity in space. Superconducting electronics offers a variety of remarkable properties including exceptionally high speed and low dissipation. The technology can be applied to both high speed computing and measurement. Its attributes may make it attractive for both ground-based and space applications.

Space applications of superconductivi digital electronics Richard E. Harris

Superconducting electronics offers remarkable new capabilities for both ground-based and space applications. In fact it appears that it may be possible in 10 to 20 years to construct a satellite computer having the computing power of today's most powerful ground based computers. This technology also will make possible new instruments capable of examining phenomena occurring on a picosecond (10 -12 s) time scale. Superconducting electronics offers a unique variety of properties which make these revolutionary advances seem compelling for applications requiring the highest possible circuit speed and density. In computer language the technology consists of a fast active element (the Josephson junction), an active element with remarkably low dissipation, and essentially lossless transmission lines which can be terminated by matched loads to minimize reflections and achieve the fastest response. In the language of instrumentation, superconducting electronics offers ultra-low-noise, high speed of response and immense bandwidth. These advantages, when compared with conventional electronics, are the result of some fundamental physical laws and not due to lack of development of other techno. logics. In fact superconducting electronics has achieved these advantages in the infancy of its development, while semiconductor electronics is a most mature technology. For applications requiring the ultimate in performance, there appears to be at present no known competitor to superconducting electronics, although, of course, some new technology may appear in the future. Nevertheless, how widespread the use of superconducting technology will become also depends on its economic viability, a subject which is difficult to assess at this stage. Because the technology is so new we begin by discussing fundamental considerations which appear to suggest that cryogenic technology will offer significant advantages for future digital devices. Secondly, we show how the active element in superconducting electronics (the Josephson junction) works and discuss the technology for fabricating the devices. Finally, we briefly review the characteristics of published circuits, and project the capabilities of future superconducting computers and instruments. The author is with the Electromagnetic Technology Division, National Bureau of Standards, Boulder, CO 80303, USA, Work supported by NASA under contract number A--437018 (JM). Paper received 7 July 1979.

Limitations of present digital technology Serious difficulties are being encountered in achieving sub. stantial further increases in computer speed. Because the speed of light is finite, the maximum dimension of a processor may be no bigger than say half the distance an electromagnetic signal can travel in one machine cycle time. The Cray-1 is one of the fastest available computers, having a cycle time of 12.5 ns (1 ins = lO--9s). Its processor and one million 64-bit words of memory are contained in a volume of only 1.7 m 3. Its maximum dimension, measured in terms of the time required for light to traverse it, is 5 ns. Within that space are dissipated 115 kW. It is a remarkable engineering feat to remove that heat. In order to construct a 1 ns computer, the diameter may be more than about 5 cm (2 in.), with a volume of 6.5 X 10-Sm 3 (65 cm3). To obtain this diameter it has been assumed that the velocity of propogation within micro striplines in the machine is reduced from that of free space by a factor of three. As will be shown later, it is likely that such a computer, constructed with the same technology as the Cray-l, would in fact dissipate more heat than the Cray. Thus the task of cooling such a computer would be still more difficult than with the Cray and perhaps impossible.

Fundamental principles governing digital logic devices The fundamental principles which demonstrate the advantages of superconducting electronics have been well discussed by Keyes.27 -3° Indeed, his papers are required reading for those wishing to thoroughly investigate the subject. Keyes begins by pointing out that the power dissipated by any device is described by V2/Z, where V is some average voltage across the device and Z is its impedance. He then develops the values of V and Z using the following arguments. In a computer, a signal must progress through a large number of active devices and transmission lines. If only a small amount of noise is added to the signal at each stage, the signal will rapidly become lost in the noise. Thus each stage of a computer standardizes the signal as illustrated in Fig. 1. As can be seen, variations of the input signal are moderated by the non-linear response; that is, small amounts of noise at the input are suppressed at the output. To achieve such standardization, a non-linear device must be used. To achieve a non-linear response the input voltage must be greater than

0011-2275/80/004171-11 $02.00 © 1980 IPC Business Press CRYOGENICS . APRIL 1980

171

f f

/'

f

A useful way of characterizing the dissipation in a logic device is through the product of its power dissipation and logic delay time. The result is roughly the energy dissipated in a switching process. Fig. 2 shows a plot of delay time as a function of dissipation for a variety of semiconductor and superconductor devices. The diagonal lines represent a constant Pr product. It can be seen that superconducting devices have a Pr product which is typically 1000 to 10 000 times smaller than present room temperature devices, According to (1) and (2) this Pr product should vary as the square of the operating temperatures: (300/4) 2 ~ 5600. Indeed, this is in the middle of the observed reductions.

/J ,/,1/

/// ////~e

////

Josephson logic

Input Fig. 1 Illustration of the non4inearity required of each active element in a digital system. Because a wide range of inputs give either a low- or high-level output, noise at the output is reduced. The output is said to be standardized

kT/e. If V ~ kT/e, then the motions of the charge carriers in the device would be affected little in comparison with their thermally-induced motions, and a linear response would result. For the sake of simplicity we will assume that V>~ 10 kT/e. The impedance of the device is obtained by noting that all computers or instruments must eventually communicate with the outside world at impedance levels of the order of that of free space. The latter has an impedance of Z o = 377 ~2. For the most efficient power transfer, the device impedance must not be too far from Z o. Thus, for our order of magnitude argument, we use Z o for the impedance of the device. Therefore, remembering that we have assumed that V ~> 10 kT/e, the minimum power dissipation is P~

(10 kT)2 ]Zo.

(1)

The minimum dissipation varies as T2 having a value of 1.66× 10-4 W for T - 3 0 0 K a n d 2.95 x 10-a War T---4K. A further consideration shows that dissipation will increase with increasing speed of response. When a device changes state, the voltage across it changes. To develop this voltage, energy 1/2 CV~ must be added. The capacitance C is that of the device. When the device is returned to its original state, the energy is dissipated. Thus the minimum dissipation for a device which is charged and discharged every r second is

Many of the difficulties with conventional digital logic are overcome by superconducting logic. The active device in superconducting logic, the Josephson logic gate, is formed from one or more Josephson junctions.a4 A Josephson junction is formed by two superconducting electrodes, separated by a thin insulating layer only a few tens of angstroms thick. To achieve its remarkable properties, the junction must be cooled to a low temperature (~ 4 K) so that the junction electrodes are superconducting. Junctions have been used in two configurations to form Josephson logic gates. The first is simply a single Josephson junction having insulated control wires over it (Fig. 3). The magnetic field which is produced by a current flowing through a control wire can cause the junction to change state. The second type of Josephson logic gate is called a superconducting quantum interference device, or SQUID. This type of device is also often referred to as a superconducting interferometer.39,4° It is composed of two, three, or more small Josephson junctions. A three-junction device is shown in Fig. 4. Connected in parallel, these junctions are small enough so that the magnetic field produced by the control wire does not affect them directly. Instead, the switching is induced by the field passing through the loops of superconductors connecting the junctions. This type of device is more sensitive to magnetic fields than is the single junction, and is therefore a better candidate for miniaturization. During the last few years, it has become the dominant type of device.

io' =., t-t

P = CV~/2r.

(2)

Clearly, for a given C and V, shorter response time leads to higher dissipation. Hence if a cycle time r is required, then the clear goals of technology are to reduce the voltage and the capacitance. Capacitance is largely determined by device dimensions. In turn, these are determined by the method of lithography used to produce the device. Thus, technologies do not differ too much in capacitance. Therefore whenever it becomes difficult to remove excess heat, a clear advantage will go to the technology which provides the lowest voltage and therefore lowest heat dissipation. As demonstrated above, this might be best accomplished with a low temperature technology.

172

10 -2

10 °

10 +2

10 +4

10 +6

Power dissipotion, /~W Fig. 2 Delay time as a function of power dissipation for a variety of electronic technologies. The product of delay and dissipation gives roughly the energy required for one switching operation. The diagonal lines are for equal energies. Josephson devices are seen to be as fast as any other technology, but to produce 1000 to 10 000 times less energy. The energy units (n J, p J, f J, aJ) are 10";9 10 -12.10-15f, 10-18 J, respectively ~

C R Y O G E N I C S . A P R I L 1980

Control

high speed. We note that in Fig. 2, the transferred electron device (TED) exhibits similar response time compared to a Josephson junction, but, of course, the TED has much higher dissipation.

Superconducting transmission lines Although superconduct7-" :'':'" (no~ to scale)

~ Insulotor ~-^''' . . . .

ing technology offers a remarkably fast, low-dissipation active element, there remains another very significant advantage over other technolgies: - lossless, low dispersion striplines.

"~ ~ Lower electrode

Fig. 3 Illustration of a thin f i l m , Josephson junction. T w o control lines are shown above, and insulated from the junction

M3

SiO

I0

~

.

..

~

0

12 5

~

~

~

Nb ground

Fig. 4 Quantum interference Josephson gate composed of three junctions in parallel. The labels Mn refer to different layers of metallization. Two control lines are shown on the left. They are not drawn across the top of the device to reveal junction structure. / c is a control current, and/g is the bias current for the gate. From reference 40. Present interferometers are somewhat different in detail 32,4

Advantages of superconducting technology Device speed The intrinsic response time, 7"j, of a Josephson junction is determined by the superconducting energy gap, 2A, by the relation112 = r,/2A.

In a very fast integrated circuit it is important that high speed signals be propagated throughout each chip, and from chip to chip, faithfully. Thus, waveguide-like structures are required. In thin-film technology a convenient structure is the stripline. In one usual geometry, this line is composed of a conducting ground plane, covered by an insulator, with a narrow line of conductor on top. When used properly, this structure will propagate signals with little loss or dispersion. Kautz has recently demonstrated, moreover, that such superconducting striplines are significantly superior to normal state lines at exceptionally high speeds and with substantial miniaturization. 26 In Fig. 6 we show (from reference 26) the attenuation per unit length and phase velocity of several types of striplines: Cu (295 K), Cu (4.2 K) and Nb (4.2 K). Examining first the attenuation for the various striplines, we see as expected that room temperature copper is poorest, but that cooling the copper to 4.2 K produces an improvement by an order of magnitude at low frequency. However, superconducting niobium is superior, by many orders of magnitude, to even cooled copper.

I000

(3)

P

For lead, which has an energy gap of 2.5 x 10 -3 eV, 7"j = 0.27 ps. A more serious limitation is due to the intrinsic capacitance C in a device of this sort which consists of two planar electrodes separated by an exceptionally thin insulating layer. An RNC time constant, where RN is the junction normal state resistance, determines the response time of the device. Although data on this time constant is scarce for realistic alloys from which junctions are made, RN can be estimated from information in a patent for an alloy of Pb with 6.5% In a . The capacitance C can be estimated approximately, from a variety of papers. The resulting dependence Of RNC on the current density/c and the barrier thickness is shown in Fig. 5. The figure is probably correct to half an order of magnitude. Reliable junctions have been made with critical current densities up to at least 2 x 103 Acre-~ . This corresponds to RNC/T J = 13. Thus, for superconducting lead, RNC = 3.5 ps, an extremely low value when compared to conventional electronic devices. It may be possible to fabricate reliable devices with much higher critical current densities, making it possible to achieve the intrinsic device response time. However, as is usual, other parameters will also serve to make it very difficult to achieve this performance in a circuit.

Dissipation That Josephson junctions offer dramatically lower dissipation than semiconducting devices has been discussed earlier. It should be noted here, however, that low dissipation is probably at least as significant an advantage as

CRYOGENICS.

APRIL

1980

P

.IOO

IOO

7 z Commonly achieved

1

~ l u e s as of 1 / 7 8 - -

Intrinsic device limit

1 2 Io

IO4 Jc' amp cm

IO6 2

Fig. 5 Estimated response time of Josephson tunnel junction, when limited by RNC , as a function o f critical current density, R N is the normal state resistance of the junction and C is its capacitance. The left ordinate gives RNC normalized to the intrinsic response time r j = IT/2A. The right ordinate gives the time in ps for a lead alloy junction. The solid line (with circles) is from an analytic expression derived from a variety of papers and a patent. The numbers next to the circles give the tunneling barrier thickness. The x's are measured values from several papers

173

I

1012 Hz. The reason for this difference is that the penetration of the electromagnetic field into the line and ground plane is frequency dependent in the normal metal case, but inde. pendent of frequency in the superconductor case. We see therefore that superconducting striplines should be much superior to normal state ones.

t ta

Kautz has made this result more clear by calculating the change in a pulse travelling down these three types of lines. In Fig. 7 his results are shown for a pulse having half-width at half-maximum of about 0.5 ps. While this pulse width is considerably faster than those found in conventional electronic circuits, it does illustrate the difference in performance between normal and superconducting lines. As can be readily seen, the thicker dielectric is superior in all cases, because a larger fraction of the field energy is contained within the dielectric, rather than within the conductors. However, thick dielectrics may be inconsistent with the miniaturization required in very fast machines. For the 0.2 pm dielectric thickness and 295 K or 4.2 K copper, the maximum usable line length is seen to be less than 0.1 cm; for niobium it is almost 1 cm.

E

.

"_o

)--

e-=

tj

o

O.

l/ I0-2

/

/-'7

¢' / e I0

IO

io IO Frequency, Hz

12 I0

Fig. 6 Attenuation and phase velocity of striplines. Attenuation (solid) and phase velocity (dashed) of striplines having conductors made of copper at 295 K, copper at 4.2 K, and niobium at 4.2 K. In each case the dielectric is silicon monoxide having thickness of 0.2 ~m 26

As Kautz shows, lo~agerpulses are less distorted by aU three types of striplines, but superconducting ones are always superior. For a given conductor, the usable line length is greater for longer pulses and thicker dielectrics. Thus while normal striplines are quite adequate at present, superconduct. ing ones may be required at the higest speeds and greatest miniaturization.

Perhaps more significant, however, is the phase velocity. To avoid distortion of a pulse propagating through the stripline, it is important for the phase velocity to be the same for all frequency components. For the copper lines the phase velocity changes significantly with frequency, but for the niobium lines the phase velocity is independent of frequency for frequencies up to the energy gap frequency, about . (L~95K)

Matched terminating resistors Superconducting striplines not only permit a relatively undistorted pulse to be propagated throughout a chip, or machine, but such lines can Nb (42K)

Cu (4K)

\

/

iI

./"

'm//I \,, //

i'.,. I

iI

I

~

~

I

I

I

I

/

I,\ "....\'',,....

/

O

2

4

-2

0

I 2

i 4

-?.'

•

0

\

"~ ~ ,."\~

::./J'--'~. ,7\%11

Time, t-tl, ps Fig. 7 Propagation of Gaussian pulses for striplines made of several materials, two dielectric thicknesses, and various line lengths. Pulses are for line lengths of zero (solid curve) 0.1 cm (dot~lash), 1 cm (dash), and 10 cm (dot). Curves which nearly coincide with the zero length curve or the horizontal axis have not been shown. The time origin has been shifted t o cause all curves t o appear nearly at the center of the time axis 26

174

C R Y O G E N I C S . A P R I L 1980

also be terminated by a matched load. Thus reflections from the ends of the lines are substantially reduced. Such reflections might interfere with the devices sending the signal, or at least reduce the effective rise time of the signal to that time required for the reflections to die off. Matched terminations are another significant aspect of superconducting electronic technology.

Fabrication of superconducting integrated circuits For several years published descriptions of methods for fabricating reliable Josephson junctions have been available, s,ls The technology not only includes processes for fabricating lead alloy junctions (Pb-Au-In), but also resistors (Auln2), and microstrip lines. Furthermore, special designs for packages which permit fast signals to move on and off a chip have appeared in the literature. 23 Details are becoming available in the published literature about the yield or reproducibility of the processes. It has been revealed that junctions can survive the order of 500 thermal cycles to liquid helium temperature and back to room temperature .33However, another report shows that they change resistance irreversibly when stored at room temperature. The changes amount to tens of percent per month. Nevertheless, storage in the freezing compartment of an ordinary refrigerator inhibits changes for periods of many months. Substantial improvements over these capabilities have probably been developed already but not publish. ed. Further progress should continue in the future.

Superconducting circuits already achieved A variety of logic and memory functions have already been achieved in superconducting integrated circuitry. Published results are shown in Table 1. It should be noted for all of the devices having linewidths of 25/am or more that no attempt has been made to achieve

Table 1.

maximum speed. Several results do show, however, that delay times under 50 ps are possible, when 5/am linewidths are used. It is difficult to project the rapidity with which delay times will be further reduced. Several points should be noted, however: reductions in delay time will accompany further reduction in linewidth; reductions in linewidth, as in semiconductor technology, below about 2/am can be achieved using electron beam lithography; and because superconducting circuits made using optical lithography already offer so much improvement over semiconductor circuits there may, for the time being, be little driving force for further reductions in size.

The superconducting computer Description of a hypothetical superconducting computer The most compelling reason for constructing a computer from Josephson tunneling logic is that by taking advantage of its low dissipation, more compact, higher-speed machines can be constructed. We assume that it will eventually be possible to construct a computer having a cycle time of 1 ns or less. We consider the implications of this speed on the size, dissipation and memory density required in such a machine. The arguments given are very general. To be made more specific, a particular machine would have to be designed for a particular function. In space for example, one might choose to minimize dissipation at the expense of speed, versatility, or memory capacity. On the ground, dissipation will be a less significant problem.

Cycle time To attempt to justify the assumption that a 1 ns machine can be built, we simply note that 100 ps simple logic devices have been constructed. In a typical computer a signal must propagate through 8 to 10 logic levels during one machine cycle. Thus 8 to 10 times 100 ps gives about

Superconducting circuits already achieved

Type

Response

Dissipation

time ps Single junction NDRO* NDRO* NDRO* SFQ* SFQ*

Logic Logic Logic Logic Logic Logic Logic Adder (1 bit) Multiplier (4 bit) Memory (64 bit) Shift A/D A/D

38 80 600 95 50 100 165 170 235 60 42-120 32 13 500 27000 5000 6200 16000 5000

Number junctions

Linewidth,

References

/am 0.1 #W

28 40 17 1

/aW nW /IW /~W

35 /~W x 45 20 /aW x 8 10 /aW x 4

1 3 3 1 1 1 1 3 3 3 6 3 7 45 322 8 20 16

1 2 50 2 25 3 50 25 3 5 5 5 2.5 25 25 13 25 25 8

24 2 38 25 37 9 16 20 39 1 32 7 6 19 21 17 36 31 14

N D R O -- nondestructive readout m e m o r y . S F Q -- single f l u x q u a n t u m m e m o r y , d e s t r u c t i v e r e a d o u t ; * - - t h e s e units could form the basis for a random access m e m o r y

CRYOGENICS. APRIL 1980

175

1 ns. A second, and seemingly frivolous, argument is based on the folklore that a cryogenic device can supplant a room temperature one only if it does something otherwise impossible or improves the performance of the room temperature device by a factor of ten. We note that a cycle time of 1 ns is roughly a factor of ten better performance than the Cray-1. Given the intrinsic response time of less than I ps for a Josephson junction, it is conceivable that computers having cycle times in the range of 10 to 100 ps may one day be built. However, an improvement in speed of one to two orders of magnitude will provide so much more computing power than today's machines that pressures for faster machines may not develop until after 1 ns machines have come into substantial use. Density It is simple to show that the required density in such a machine is within the present state-of-the-art. We assume the computer is contained within a sphere of diameter 5 cm. TbA~hypothetical computer is illustrated schematically in Fig. 8. If we consider the total machine volume of 65 cm 3 to be one-tenth processor and the remainder memory we can ask for the density required in each.

Let us assume the processor has the same number of circuits as an IBM 370/168 which is 2 x 10. s Thus an average density of 31 000 circuits cm-a is required. If we further assume that the integrated circuits are on 0.25 mm thick substrates ( a common value for present silicon wafers) and that these substrates are spaced by 0.25 mm, we find 20 layers cm_ ~ In addition we assume that packaging limitations decrease the available space by a factor of 10. Thus we require only an average of 1600 circuits cm -2 with an area per circuit of 6.3 x 10_ s cm 2, excluding packaging. This is about the area of present generation logic devices. In the case of the memory, the requirements are slightly more strigent. Let us assume the memory contains 156 000 words, each having 64 bits, or a memory of 10 7 bits. Using the same wafer and spacing as before, and assuming that 90% of the area is for control circuitry, and interconnections, one requires a density of 85 000 bits crn-2 , or an area per bit of 1200/lm 2 . Since a memory device with nondestructive readout has already been reported having an area of 900/~m2,2'~7 the requirements for memory density can be assumed not to be a restriction. While the memory considered here is not large by present standards, it is fast. Significant amounts of additional memory could be added outside of the 5 cm diameter sphere. A recent US Patent, Is suggests the possibility of constructing a memory cell having an area of only about lO/am 2 . Such a device would undoubtedly be fabricated using electron-beam lithography. Allowing a factor of ten for decoding circuitry and packaging, l0 s bits could be contained within 5 cm 3 , a package 1.7 cm on a side. Heat removal could be more difficult with this exceedingly high density. These estimates are simplistic and are only intended to illustrate the feasibility of constructing such a machine. Significant packaging problems would have to be overcome to realize the very high projected density. Anacker (Computing at 4 K IEEE Spectrum (May, 1979) 26-37) has recently given a much more realistic view of a superconducting computer. His work contains a fairly detailed plan of the entire computer, including the re. markably small inter-connections required. He discusses a machine having a 4 ns cycle time and 1.3 x lO s bits of

176

main memory, in a volume of 4 x 103 cm. a He conservatively assumes a 5 gm linewidth technology which has already been dated by smaller.and faster 2.5/zm capabilities. Power consumptzon It is the low dissipation of supercon. ducting electronics which makes possible the high density computer described above. Of particular significance is that superconducting memory requires no stand-by power, the information being stored in the form of circulating super. currents. In these order-of-magnitude arguments, we shall assume that heat removal from the memory can be neglected.

Heat removal from the processor is also encouraging. We consider our hypothetical computer to be constructed using the recently published logic. A single tunnel junction, when in the voltage state, carries a current typically of order 1 mA and a voltage about equal to the gap voltage of 2.5 mV. Thus it dissipates a few/lW. SQUID logic devices having a dissipa. tion of about 1,u W have been reported. We assume for simplicity that each 'circuit' in our computer dissipates 10/~ W including the gates, power supply, and other losses. Multiplying by the 2 x 10 s circuits in this machine, and 0.5 because we assume no more than half the processor gates are operating at a given time, we arrive at a net dissipation of 1 W, exceedingly small when compared with conventional machines. The 1 W dissipation must be removed at 4 K using an approximately 300 K work input. Carnot efficiency multiplies these powers by a factor of approximately 75. Furthermore, it is thought that refrigerator inefficiency and heat leaks can be accounted for by another factor of ten. Thus the dissipation at room temperature would be about 750 W, still much smaller than conventional machines, such as the 115 000 W dissipation of the Cray-l. It therefore appears quite likely that a superconducting computer having the processing capability and memory of today's most powerful machines would dissipate sufficiently low power to be flown in space. Its extreme speed also makes it a strong candidate for replacing many ground-based computers. It is most significant that this dissipation produces a heat flux in the processor of 0.16 Wcm-2 from the surface of the circuits into the liquid helium. It would probably be possible to construct a machine having a heat flux of about 1 W cm-2 . Our conceptual machine has a lower heat flux than this upper limit. This point is extremely significant because the limiting factor in conventional machines is the removal of heat from chips or boards, not the cost of operating a refrigerator. Nevertheless, superconducting machines may require lower room temperature power input than present machines. Applications of a superconducting computer

As with most computers, the range of applications of a superconducting computer is quite broad. However, because the superconducting computer offers such a major advance in computing speed, it may make possible applications which are more than just evolutionary. One can think of this range of applications as divided into three areas, each being more removed from present applications than the preceding one. Applications of the first kind consist of the usual applications of computers, for logging data, data reduction and analysis, etc., but with substantial increases in speed. Furthermore, decreases in dissipation may make this type of computing in space more widespread.

C R Y O G E N I C S . A P R I L 1980

Cryogenic

/contoiner

jMemo~

Processof~ /

[ 10

T 5 Stole, cm

I 2

I I

r 0

Fig. 8 Illustration of a hypothetical superconductingcomputer, giving a suggestion of the size o f processor,memory and cryogenic container. A space-borne container would probably be designed differently

Applications of the second kind might rely on the increased computing speed and larger memories which are possible in a superconducting computer. For example, weather modelling and forecasting may become much more timely and accurate given the expanded capabilities of a superconducting computer. In another example, simulated wind tunnel testing might prove much faster and less expensive than the real alternative. Applications of the third kind are those in which a computer might perform revolutionary new tasks made possible by increased speed and memory. One might ask when considering these applications, what are the possibilities for artificial intelligence-like tasks with such a powerful computer? Can one, without waiting for theoretical developments, perform tasks by brute force? Could one send such a computer on an interplanetary voyage, trusting it to analyse data, control the vehicle, and only selectively return data to earth? Would such a computer reduce the need for a man in space? Never before has it been possible to conceive of sending such a powerful machine, which requires so little power, into space. The implications of these computers could be profound indeed! The present state.f-the-art The list of achieved superconducting circuits in Table 1 gives some suggestion of the present state of superconducting computer development. Most of the work has gone on within the IBM Corporation, at Yorktown Heights, NY, in the US, and in Zurich, Switzerland. It appears that more than 100 scientists and engineers are working to construct a prototype computer within the next five years. However, other US institutions are becoming involved with work at Bell Laboratories, Sperry-Univac, the National Bureau of Standards, and Hewlett-Packard, among others known at the present time. Ultimate achievable performance The 1 ns computer described previously probably does not represent the ultimate achievable from this technology. Given

CRYOGENICS. APRIL 1980

the 0.27 ps response time of a single junction, it is conceivable that a computer having a cycle time below 100 ps is possible. Supporting this possibility is the fact that even a 1 ns machine is within the bounds of already achievable density. However, it should be noted that while very fast integrated circuits have been developed, there exists little published work on packaging these circuits in such a way as to maintain speed in a whole assembly of the circuits. Projected development rate It is possible that a prototype superconducting computer will be demonstrated within five years. Such a projection is difficult to substantiate since much of the work is of a proprietary nature. Nevertheless, a commercial superconducting computer might become available within ten years. The technology may be applicable not only to large computers where it would make a unique contribution to computing power, but also to medium scale computers where the high speed might make simpler, less expensive, hardware possible. Because a superconducting computer generates such small dissipation, it is a natural candidate for use in space. Of course, the circuitry would have to be made spaceworthy. However, a much bigger task is the construction of a suitable lightweight, reliable refrigerator. (Refrigerators will not be a serious problem on the ground, as it is possible to achieve reliability using several, and weight is no problem.) It should also be emphasized that this new technology is more difficult than present computer technologies. The higher speed of response now requires attention, which was not required in the past, to all interconnections, each of which must be properly matched with microstrip lines at the proper impedance level. Furthermore, packaging must be done much more carefully with similar attention to interconnections. These factors may seriously interfere with the present distribution system for electronic components. That is, it may be much more difficult for one company to make a superconducting integrated circuit for sale to another company which assembles these circuits into a computer. The circuit manufacturer might necessarily produce a much larger, more integrated piece of hardware. Indeed, this may already be occurring with present electronics, with more and more component manufacturers now offering complete systems.

Superconducting instruments The requirement of very fast measurements by military devices and the potential arrival of very fast computers both indicate a need for new, fast measurement capabilities. That Josephson tunneling logic is a strong candidate for this new class of instruments is suggested by its high speed and low dissipation, by its past successes in this field (some Josephson instruments are now available commercially), and by the development of technology for making reliable thin-fihn Josephson junctions. The following is a brief review of past and present applications of Josephson devices. In the past, Josephson instruments have been characterized as containing a single Josephson junction. Typically, this junction was a point contact (like a cat's whisker diode), or possibly a tiny constriction in a thin film. The use of one junction, rather than more, was largely because photolithographic techniques had not been applied to Josephson devices.

177

The single Josephson junction was typically incorporated into a loop of superconducting material to form a SQUID.4~ A SQUID is capable of detecting very small magnetic fields, of the order of 10-14 T(10 -lo oersted). Most of the Josephson instruments now in use incorporate this device. SQUID devices are available commercially from at least five companies. When used directly to detect magnetic fields, or their gradients, SQUID devices are used for magnetic anomaly detection (such as submarines), for magnetocardiography (a non-body-contact method of measuring human heart performance), magnetoencephalography (magnetic brain waves), geophysical measurements and prospecting. Additional uses of SQUID's are for rf power and attentuation standards, ultra-low temperature thermometry, and precise current comparison. In each case the Josephson device made possible previously impossible measurements, or made a substantial improvement over previous measurements. In addition, single Josephson junctions, not used in the SQUID configuration, are used to maintain the US legal volt, and for ultrasensitive broadband or heterodyne detection in the miUimetre and far infrared region of the electromagnetic spectrum. ~° Single Josephson junctions have been used for microwave and infrared harmonic generation and mixing at frequencies up to 30 THz (10/am wavelength). The development of lithographic fabrication methods for complex (multi-junction) superconducting circuits provides new opportunities for high-speed devices. Fast instruments demand a number of characteristics from the technology from which they are formed. In many ways they are quite similar to very fast computers. One must consider what is meant by 'fast'. An instrument should be capable of characterizing events on a time scale which is shorter than that in which the event occurs. Of course that means as fast as possible, because there are always events of scientific interest which offer remarkable speed. A more practical measure of 'fast' might be a speed somewhat faster than events occur in commonly available circuitry, such as in future computers. Thus, instrument response times of the order of I ps would be most useful, since the fastest present instruments are sampling oscilloscopes with response times of 25 ps. Superconducting instruments should be capable ultimately of speeds in the 1 ps range. A more comprehensive definition of 'fast' might recognize the difference in types of measurements. For example, it might take more than a second to make an exceptionally high accuracy measurement of voltage. However, even for a measurement of this sort, the total time required can be reduced through the use of faster active devices. High speed devices are necessarily small. Recall that signals propagating at the speed of light travel only 300/Jm (0.030 cm) in 1 ps. If a variety of sensors in an instrument are to analyse a signal at close to the same time, they must be in close proximity to one another. The size scale is, of course, determined by the time scale. If a variety of elements in the circuits are located close together, their dissipation must be sufficiently small that their operating temperature does not rise substantially. The heat produced by the fastest conventional devices already prevents more than a small number from being located on the same integrated circuit chip. The need for an improved technology is thus evident. If an instrument is to analyse analogue signals, those signals must be accurately transmitted throughout the instrument. Thus, the use of lossless, properly terminated, lines is of great advantage in a fast instrument.

178

All of these considerations are, of course, the same as those discussed earlier with respect to computers. Thus, in the fastest instruments the use of superconducting technology is compelling. I mportant elements of instruments

There are a variety of possible instruments, but those to be discussed here make use of two simple components, a current comparator and a sample-and-hold circuit.

Current comparator A Josephson gate is essentially a current comparator. It is used in its simplest form in digital circuits where it must determine only whether a signal represents a '0' or a '1'. However, it can also be used in an analogue sense as well. If the gate is biased in the zero voltage state, w i t h some current flowing through it, then the gate will switch at some well-defined current flowing through the control line over the gate. One can vary the switching point by providing a bias current through a second control line over the gate. Thus, an adjustable current comparator results. Under favourable conditions it should operate as fast as any Josephson gate - that is, ultimately in about 1 ps. Such current comparators would be important elements in many of the devices to be discussed below or they might be the funda. mental element in a fast trigger circuit. Sample-and-hold Sample-and-hold circuits are conventionally used to sample a voltage in a very short time, and maintain it (as in the form of a charge on a capacitor) until it can be characterized by a slower device. A sample-and-hold circuit may also be useful in a superconducting instrument for the same purpose. Alternately, some superconducting instruments may operate in such a short time that a sample.andhold circuit will not be necessary. One possible kind of superconducting sample-and-hold device would simply store a current in a superconducting persistent current loop. The current would be switched in and out of the loop using a Josephson junction in the loop. The product of the critical current I o of the junction and the inductance L of the loop would have to be sufficiently large, however, since the maximum flux in the loop, LIo, is quantized in units of¢o = 2.07 x 10-~s Wb m-2 , the superconducting flux quantum. Thus, the fractional resolution could not be greater than ¢o [LIo. The speed of response of such a simple circuit might be determined by the time required for the Josephson junction to go from normal state to the superconducting state where the flux would be trapped. Times as short as 1 ps might be achieved. Logic circuits One may also use logic in an instrument. A counter, for example, must perform logical operations. Such elements of an instrument might be quite similar to those in superconducting computers. A few possible Josephson instruments

Sampling device - analogue Published aa'n and unpublished reports of sampling devices suggest that such instruments have already been constructed having a time resolution of a few ps. Such sampling devices are constructed from current comparators similar to the type described above and should provide the best means for characterizing fast repetitive signals. Fast counters It should be possible to construct fast counters using superconducting technology. The speed of response would be determined by the speed of the circuitry for the least significant bit. More significant bits

CRYOGENICS. APRIL 1980

would change state more slowly and would therefore not be as important. Careful engineering of the first stage might bring the response time down to a few picoseconds, corresponding to a counting rate of the order of 500 GHz. Examples of the use of such fast counting rates might be in high accuracy clocks, or in detecting particles at very high flux, or in A/D converters.

AID converters In conventional electronics the fastest, but seldom used, analogue--to-digital converters are of the parallel type. That is, the converter contains one current (or voltage) comparator for every possible level of the input analogue signal. Thus, a 4-bit converter might have about 16 comparators. Subsequent circuitry then condenses the 16 bits of level information into a 4-bit binary word. A similar design in superconducting technology would undoubtedly provide the highest possible Speed. The sampling rate would probably be determined by the time required to store the digitized signal in preparation for the next one. Sampling rates as high as 10 to 100 GHz might be achieved. Two types of A/D converters have been tested and described in the literature: a successive approximations type, 31 using now outdated technology, and a parallel type, 14 using somewhat more current technology. Both are 4-bit converters. The latter makes use of multiple-quantum interference in a three-junction interferometer like that described earlier. Using this novel feature, it is possible to construct a fully parallel converter which requires only N interferometers, a significant advantage compared to the above design which might require 2 N interferometers. Furthermore, a design has been proposed, 22 which makes use of flux quanta moving in and out of a superconducting loop as the basis for digitizing an analogue signal. A/D conversion is subject to the limitation that one cannot directly digitize a signal which is changing faster than one bit of accuracy during the sampling time of the converter. Sample-and-hold circuits are used to maintain a signal until it can be digitized. The time during which the sample-andhold is sensitive to variations in the signal (aperature time tap ) may limit the speed with which signals can be digitized. It seems unlikely that a superconducting sample-and-hold circuit could operate in less than the intrinsic response time for a Josephson junction: i~2z~ = 0.27 ps for lead. During this period the signal may change no more than one part in 2 N for an N-bit converter. Assuming a sine wave at frequency f, one f'mds,fmax = (Tr2Ntap) -I. The maximum sampled frequency fmax is plotted in Fig. 9 as a fuction of the number of bits of accuracy. It is readily seen that a reduction in the aperture time of the sample-and-hold circuit dramatically increases fmax. Thus, achieving the minimum aperture time represents a high priority task if extremely high-speed A/D converters are to be realized. A long range goal might be to reduce tap below a picosecond. Altemately, sampling theory shows that aperture times are not required if the sampler determines the average of the signal over a precisely determined period. This approach might be used to circumvent the need tbr exceptionally short apertures. Applications of very fast A/D converters might include communications (where digital filtering or image processing is used), data acquisition (conversion of rapidly varying signals to digital form), and very broadband radars.

CRYOGENICS

. APRIL

1980

£~9

E "5

2

4

6 Number of bits

8

10

12

Fig. 9 Maximum sampling rate for A / D converter as a function o f accuracy, or number o f bits, Curves shown are for various aperture times tap of the sample-and-hold circuit

Digital instruments for real-time signal processing

Given a high-speed digital technology, it becomes possible to conceive of real-time applications of computational techniques which previously were done off-line, or which could be applied only to very slow signals. One such example involves Fourier analysis of signals. Such signals might pass through a superconducting A/D converter of the type discussed above, and then be digitally analysed using a special purpose, fast Fourier transform or other type of processor designed to perform these calculations at the highest possible speed. Such an FFT processor could be useful in a variety of instruments, including fast digital filters.

Agile digitalfiltering Digital filtering is an extremely useful concept, applicable to areas such as communications and noise reduction. Filtering can be done with rapid change in frequency, lineshape, etc., possibly controlled by the incoming signal. One application of this type of filter might be in the real-time interpretation of codec_ signals.

Multi-channel spectrum analyser Superconducting digital technology should enable the construction of an analyser which has higher speed than those using conventional technology. Such analysers are already under development, using conventional technology, for the search for extraterrestrial intelligence (SETI). Such a device could be directly coupled with a processor to examine the data in real time, thus reducing the quantity of data which must be stored or transmitted. Summary and conclusions

Superconducting electronics is a newly emerging technology of considerable promise. The motivation for the develop-

179

ment of this technology is rooted in some very general and fundamental considerations. The Josephson switch is extremely fast and can potentially achieve switching speeds of 1 ps. Its most attractive attribute is its small dissipation. The dissipation is sufficiently small to allow extremely close packaging of active elements and thus allow dramatic reduction of intracircuit communication delays. Furthermore, the use of superconducting striplines for this communication provides the opportunity for unparalleled freedom from dispersion and attenuation of signals. Broad interest in the technology is expanding, but the prime driving force behind the technology is a major commercial programme to develop a superconducting computer. With some knowledge of the characteristics of the simple logic elements, it is relatively easy to show that order of magnitude (and greater) advances can be made in both speed and capacity. A necessary result of increased speed is reduced size, and overall dissipation drops by many orders of magnitude below that of conventional computers. All of these changes (increased speed and capacity, decreased size and dissipation) are in a direction favourable to space applications, The development of such a computer may be 10 to 20 years in the future, but it is not too early to consider the opport. unities, perhaps revohitionay ones, which may appear. Several applications can be suggested now. The processing and compression of complex raw data from space experiments would provide a means of conserving the limited power available for information transmission. For example, the mapping of the heavens in any frequency region involves tremendous amounts of data. A certain amount of data filtering can be accomplished without a large computer, but a smart satellite could provide much more intelligent selection of data. A powerful computer in space could offer a means to improved weather modeling and forecasting on a global scale. The combination of the acquisition and processing of weather data could simplify and speed up the modeling process, and thus produce more timely and accurate forecasts. These applications are founded on established concepts, but it is also exciting to consider more exotic ideas. Never before have we been faced with the potential for such a large computing capability in space. Could such computers assume intelligence-like roles? If so, would the need for some difficult manned space missions be reduced? There are difficult questions which have certainly been considered before, but reconsideration might be appropriate in light of the new opportunities which are now developing. Other types of fast instrumentation are made possible by this new technology. Ultra.fast A/D converters, digital filters, counters and spectrum analysers are just a few of the devices which should be considered. The applications of such instruments are myriad. For example, improved A/D conversion could improve signal acquisition in the search for extraterrestrial intelligence (SETI). Coupled with a superconducting computer, such A/D converters could perform real-time processing of microwave signals in a very broad frequency band. This type of system would essentially be an ultra-broadband digital receiver. All of these applications are highly speculative. We assume that primarily commercial interests will develop the computer, but that is by no means certain. The peripheral development of other fast instruments (both digital and analogue) will probably accompany the computer effort. Undoubtedly these developments will be important to the space programmes of the world.

180

References 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

24 25 26 27 28 29

Baechtold, W., Forster, Th., Heuberger, W., Mohr, Tit. O. Complementary Josephsonjunction circuit: a fast flip-flop and logic gate Elec Lett 11 (1975) 203-204 Broom,R.F., Jutzi, W., Mohr, Th. O. A 1.4 mil2 memory cell with Josephson junctions [EEE Trans Mag MAG-11 (1975) 755-758 Eldrigd_e,J.M., Matisoo, J. Fabrication of variable current density Josephsonjunctions, US Patent No. 3 816 173 (issued 11 June 1974) Geppert,L.M., Greiner, J.H., He~eli, DJ., Kleppner, S. Damped three junction interferometers for latching logicIEEE Trans Mag MAG 15 (1979) 412-415 Gheewala,T.R. Josephson logic circuits based on nonlinear current injection in interferometer divecesA ppl Phys Lett 33 (1978) 781-783 Gheewaia,T.R. Delay measurement of experimental 2.5 t~m J osephson current injection logicAppl Phys Lett 34 (1979) 670-672 Gheewala,T.R. A 30 picosecondJosephson current injection logic familyIEEEJ Solid-State Circuits SC-14 (1979) 787-793 Greiner,J.H., Basavaiah, S., Ames, I. Fabrication of experimental Josephson tunnelling circuitJ Vac Sci Technol 11 (1974) 81-84 Guetet,P., Mol~, Th. O., Wolf, P. Single flux quantum memory cells 1EEE Trans Mag MAG-13 (1977) 25 -55 Hamilton,C.A. Space applications of superconductivity: microwaveand infrared detectors Cryogenics 20 (1980) Hamilton,C.A., Lloyd, F.L., Petetson, R.L., Andrews, J.R. A superconductingsampler for Josephson logic circuits ApplPhys Lett 35 (1979) 718-719 Harris,R.E. Intrinsic response time of a Josephson tunnel junctionPhys Rev B 13 (1976) 3818-3821 Harris,R.E., Hamilton, C.A. Fast Superconducting Instruments. Future trends in superconductive electronics, Charlottesville, 1978. American Institute of Physics, New York (1978) 448-458 Harris,R.E., Hamilton, C.A., Lloyd, F.L. Multiple-quantum interference superconductinganalog-to-digital converterAppl Phys Lett 35 (1979) 720-721 Havemann,R.H., Hamilton, C.A., Harts, R.E. Photolithographic fabrication of lead alloy Josephsonjunctions 3 Vae Sci Technol 15 (1978) 392-395 Henkels,W.H. An elementary logic circuit employingsuperconducting Josephson tunnelling gates IEEE Trans Mag MAG-11 (1978) 860-863 Henkels,W.H. An experimental 64-bit decoded Josephson random access memory 1EEE J Solid-State Circuits SC-13 (1978) 591-600 Henkeis,W.H. Ultralow-power, micro-miniaturizedJosephson devices having high inductance, US Patent No. 4 028 714 (issued 7 June 1977) Herell,D.J. A Josephson tunnelling logicadder 1EEE Trans Mag MAG-10 (1974) 864-867 Hertell,D.J. Femtojoule Josephson tunnelling logic gates 1EEE J Solid State Circuits SC-9 (1974) 277-282 Hertell,D.J. An experimental multiplier circuit based on superconducting Josephson devices1EEE J Solid State Circuits SC-10 (1975) 360-368 Hurrell,J.P., Silver, A.H. SQUIDDigital Electronics, Future Trends in Superconductive Electronics, Charlottesville, 1978 American Institute of Physics, New York (1978) 437-447 Jones,H.C., Herrell, D.J., Yao, Y.L. Inductancemeasurement of superconducting chip-to-packageconnectors suitable for Josephson LSI technology1EEE Trans Mag MAG-15 (1979) 432-434 Jutzi, W., Moltt, Th.O., Gasser, M., Gschwind, H.P. Josephson junctions with 1 pm dimensions and with picosecond switching times Elee Lett 8 (1973) 589-591 Jutzi, W. An inductively coupled memory cell for NDROwith two Josephsonjunctions Cryogenics 16 (1976) 81-88 Kautz,R.L. Miniaturization of normal-state and superconducting striplines NBS JRes 84 (1979) 247-259 Keyes,R.W. Physicalproblems and limit6 in computer logic IEEE Spectrum (May, 1969) 36-45 Keyes,ILW., Harris, E.P., Konnetth, K.L. The role of low temperatures in the operation of logic circuitry, Proc IEEE 58 (1970) 1914-1932 Keyes,R.W. Physical limits in digital electronics Proc IEEE 63 (1975) 740-767

CRYOGENICS. APRIL 1980

30 31

32 33 34 35 36 37 38 39 40 41

Keyes, R.W. Physical limits in semiconductor electronics Science 195 (1977) 1230--1235 Klein, M. Analog-to-digital converter using Josephson junctions, 1977 IEEE Int SoSd-State Circuits Conference Digest of Technical Papers 20 IEEE Cat. No. 77CH 1172-6 ISSCC (1977) 202-203 Klein, M., Herrell, DJ. Sub-100 ps experimental Josephson interferometer logic gates 1EEE J Solid-State Circuits SC-I 3 (1978) 577-583 Lahiri, S.K., Basavanah, D. Lead-alloy Josephson tunnelling gates with improved stability on thermal cycling JAppl Phys 49 (1978) 2880-2884 Matisoo, J. The tunnelling cryotron - a superconductive logic element based on electron tunnelling, Proc IEEE 55 (1967) 172-180 Van Duzer, T. Private Communication Yao, Y.L., Herrell, DJ. An experimental Josephson junction slfift register element, Int. Electron DeVices Meeting, Tech Dig (1974) 145-148 Zappe, H.H. A single flux quantum Josephson junction memory cellAppl Phys Letters 25 (1974) 424-426 Zappe, H.IL A subnanosecond Josephson tunnelling memory cell with non-destructive readout 1EEEJ Solid State Circuits SC-10 (1975) 12-19 Zappe, H.H. Quantum interference Josephson logic devices ApplPhys Letters 27 (1975) 432-434 Zappe, H.H. Josephson quantum-interference computer devices 1EEE Trans Mag MAG-13 (1977) 41-47 Zimmerman, J.E. Space applications of superconductivity: Low frequency superconducting sensors Cryogenics 20 (1980) 3

For Sale AR60HPgRIFICAIOHPLAIT 30 TON-DAY

MAJOR EQUIPMENT

Cleveland, Ohio INSTALLED

-

1--Clark CMA 6 compressor, 400 HP, synchronous 440V. 5--Cylinders: 19¼"x8",30 psi; 14½"x8", 48 psi; 5¾"x 8'" 600 psi; 2¼"x8'" 3000 psi; 3¾"x8'" 1300 psi. 1--H 2 recycle compressor I 1--N 2 blower 2--Deoxo 1'4"x,3'9" high I 1--Deoxo 1'10"x4'10" high 3--Coolers 20'8 ' long 13--Separators 2--Absorbers 12--Filters 1--Mol Sieve: blowers, exchangers, skid mounted refrigeration refrigeration. 1--Finch refrigeration compressor, 60 HP. 2--Cryogenic storage tanks, stainless steel, 2000 gallons.

HOWE.BAKERHYDROGEN.CARB MONOXIDEPLANT Produces 500,000 SCFD of 50% hydrogen & 50% carbon monoxide. Can be modified to produce up to 500,000 SCFD of pure hydrogen. Complete pkgd unit, skid mounted, easily put together. Like new conditionready for immediate delivery. Write or Call for an A p p o i n t m e n t !

IlllUBlll J

Division Areco, Incorporated

I

9301 W. Bernice St., Schiller Park, IL 60176 U.S.A. I 312-678-1500-- TX 28-9454 - Cable AARONEQCO I I

only the Finest... CUSTOMERS

APPLICATIONS

Oregon State University Bell Telephone Laboratories Wright-Patterson Air Force Base Stanford University Sandia Laboratory Argonne National Laboratory Cornell University Fermi National Accelerator Laboratory Naval Research Laboratory Princeton Applied Research Lawrence Livermore Laboratory International Business Machines Jsnis Research Company Brookhaven National Laboratory University of California Pennsylvania State University University of Illinois SUNY-Buffalo Massachusetts Institute of Technology Xerox Corporation Oak Ridge National Laboratory Max Plsnck Institute Kyoto University

M6ssbauer Effect Nuclear Magnetic Resonance Nuclear Demagnetization Magneto-Optical Superconductor Research Plasma Physics Ion Resonance Magnetoresistance Effects Faraday Rotation Gyro Cyclotron Superconducting Motor Bubble Chamber Energy Storage and Conversion Nuclear Beam Spin Flip Beam Focussing

PRODUCTS Superconducting Magnets Complete Magnet Systems Vapor Cooled Current Leads Liquid Helium Level Meters Cryogenic Accessories

...Among Many Others

a lv[-RICAN

S E N D FOR FREE B R O C H U R E

IMI GNETICS, INC.

P.O. Box R, Oak Ridge Industrial Park Oak Ridge, TN 37830/Telephone 615-482-1066

CRYOGENICS , APRIL 1980

181