Charge-coupled devices-A new approach to

MIS device structuresCharge-coupled devices require no diffusion for theiractive parts and can be conceptually fabricated with two maskingsteps. These new devices are finding applicationas image sensors and memory elements

W. S. Boyle, G. E. Smith Bell Telephone Laboratories, Inc.

Recent advances in materials and processing have oscillators, for example, have become important asresulted in a new class of information-handling struc- sources of power for millimeter waves. These devices hadture-the charge-coupled device. This three-layer their genesis in work performed by W. Shockley and W.structure creates and stores minority carriers, or their T. Read a decade earlier. The injection laser, first demon-absence, in potential wells near the surface of the strated in 1962, is only now becoming a device in whichsemiconductor. The minority carriers move from under the understanding of a heterojunction materials systemone electrode to a closely adjacent electrode on the has proceeded to a point at which continuous room tem-same substrate when a more negative voltage is ap- perature operation is possible.plied to the adjacent electrode. Because of their high Similar delays are apparent in the most important of alltransfer efficiency, these devices have already found semiconductor technologies: low-level signal processing.application as image sensors. In addition, there is Throughout the past decade, integrated-circuit tech-every expectation that memories made by use of the nology has been dominated by bipolar transistor struc-stored-charge concept will be less expensive and tures that comprise metal and silicon dioxide films onfaster, and will require less power than a magnetic silicon and use a process technology of diffusion andcounterpart now in use. oxide-masking. There has been little tendency to vary this

system even though the many changes in materials andprocessing have been explored experimentally.The reason for the delay between structural concept

Rarely do important technological applications follow and application is easy to find. Although it may be possi-major advances in our understanding of the fundamental ble to demonstrate the feasibility of an idea in a novelphysics and chemistry of a phenomenon or a materials device structure, its wide application depends on a com-system. Rather, it is more likcely that the improved under- plete understanding of both the physical phenomenonsstanding from fundamental studies will give rise to some involved and the materials and processes required torudimentary device structures that may not be used for fabricate the structure. For these reasons, progress inseveral years before the original idea finds widespread major device technologies tends to await the developmentapplication. In the past year, the IMPATT and transit-time of a large amount of background information. Improve-

18 IEEE spectrum JULY 1971

Authorized licensed use limited to: IEEE Publications Staff. Downloaded on October 6, 2009 at 11:59 from IEEE Xplore. Restrictions apply.

FIGURE 1. A piezoelectric surface wave transducer withmetal stripes formed by electron-beam resist techniques.

FIGURE 2. (Above right) A-Video display of excess cur-rents from defects in a silicon-diode array camera tube.B-The same display from a defect-free array.

ments then occur in quantumlike jumps. Integrated-cir-cuit technology is at the stage where in the next few yearswe are going to see just such a major change. The follow-ing areas, in which significant progress has been made,lead to this conclusion.

Semiconductor progressFirst, after more than a decade of intensive study, we

are beginning to understand interface states in thesimplest systems-such as silicon dioxide on silicon.This understanding not only encompasses the physicaldescription, but also includes the phenomenologicalknowledge of how to preserve desirable properties undersevere stress. Of particular importance are dual dielec-trics, which give the benefits of the low surface-statedensity available with silicon dioxide together with thehigh stability from less permeable materials, such assilicon nitride and aluminum oxide. Semiconductor de-vices can now be made that do not require the protectionof a hermetically sealed enclosure. This single develop- hybrid circuit.ment changes the whole nature of integrated electronics The ability to fabricate structures with a thickness ofby allowing a new freedom in interconnecting chips; no less than one micrometer is a second factor. For somelonger do the connections have to pass through vacuum time it has been possible to produce insulating films ofseals. In this way a much higher order of integrated cir- high quality in this thickness range, but only recently hascuit is possible, with complete subsystems such as digital ion implantation produced well-controlled doping pro-processors and memories taking the form of one large files with such dimensions. An improvement in precision

Boyle, Smith-Charge-coupled devices: A new approach to MIS device structures 19

Authorized licensed use limited to: IEEE Publications Staff. Downloaded on October 6, 2009 at 11:59 from IEEE Xplore. Restrictions apply.

of at least one order of magnitude over any other process slide of silicon. In the first devices we made, defects werefor both the number of impurity ions and their position a serious problem. Figure 2A shows the image of thehas been clearly demonstrated. leakage current obtained from a target made three years

Furthermore, new techniques in pattern generation ago with the best technology we had at that time. Eachhave decreased lateral dimensions for masking opera- one of the white spots is a single diode with excess leakagetions. Figure 1, for example, shows a device fabricated current. A normal leakage current is about 10- " ampere.by A. Broers et al. I that consists of a metal pattern laid The main source of the difficulty has been traced to clus-down on a piezoelectric material to form a transducer. ters of imperfections introduced at the time that high-The individual fingers are 0.17 um wide and are spacedon 0.5-Mum centers. He uses a fine electron beam to exposea specially developed photoresist material, poly-methylmethacrylate, and then proceeds in a conventional wayto employ this material as a mask to etch the patterns. A InsulatorThis kind of fine geometric control makes it possible tofabricate devices with improved high-frequency per-formance and to make integrated circuits with higherpacking density. It also presents the opportunity of fabri- Depietioncating structures with lateral dimensions that are com- ' Vparable to depletion-layer widths and oxide thickness, Ila factor of importance in at least one device family.A third area of progress is in the preservation of the t

homogeneity of the semiconductor material throLrghout EFthe fabrication of device structures. All high-temperature Bprocess steps--such as diffusion and oxide growth- Bdegrade the crystalline perfection of starting material. n-typesemiconductorOnly recently, however, did the full extent of this damagebecome evident or even important. The problem wasbrought very forcefully to our attention recently when weattempted to fabricate silicon-diode-array camera Depletiontubes-arrays of 500 000 individual diodes on a single u edge V

\+_A -r o, Distance

I-a. °.Ka. It.. I-.I_1 FIGURE 5. Plots of electron energy as a function of dis-Input Output tance through an MIS structure. In (A), t = 0, no charge is

Metal gates stored at the surface. Charge is stored at the surface inB,_/7(B),t= X.

Input o 2 1 12 3 14 _5161 O OutputFIGURE 6. A theoretical plot of depletion width (x,), sur-face potential (o,), and capacitance (C) of an MIS structureGate p-diffused as a function of charge at the interface (Q'). Q is the maxi-mum amount of charge that can be stored under condi-

FIGURE 3. A-Circuit diagram for an IGFET bucket tionsdescribed in the text.brigade. B-Cross section of the integrated circuit form.

FIGURE 4. Cutaway of a charge-coupled device. Thedashed line represents both the edge of the depletion re- 0.8 4.0gion and the potential distribution. Voltages are typical.

v1=-5V V2-1OV V1=-5V0.6 3.0

04 2.0

|id 02 10

0 0.2 0.4 0.6 0.8 1.0

Depletion condition Q'/Q

20 IELr spectrun jui.,y 1971

Authorized licensed use limited to: IEEE Publications Staff. Downloaded on October 6, 2009 at 11:59 from IEEE Xplore. Restrictions apply.

temperature oxidation takes place. Subsequent develop- ceptually simple form of functional charge-storage devicement efforts have led to nearly complete elimination of that cannot be constructed from discrete components.such defects (Fig. 2B). The device stores a minority-carrier charge in potentialThese advances have come about from trying to im- wells created at the surface of a semiconductor and trans-

prove an existing technology. Where these possibilities ports the charge along the surface by moving the po-will lead is uncertain; however, to be more specific, we tential wells. In its simplest form the structure that ac-shall discuss a class of information-handling devices that complishes this consists of an array of closely spacedmake use of most of these advances. The common factor metal electrodes that overlay an insulator deposited onin this device family is that information is represented by a uniformly doped semiconductor substrate. Figure 4stored charge. Several such structures have been reported shows an n-type bulk (minority carriers are holes) devicein the past year. One class uses stored charge in an x-y in a situation wherein a sufficiently large negative biasarray of elements for a random-access memory.2 There potential has been applied to all the electrodes to produceare many commercially available insulated-gate field- inversion, and the center electrode has a slightly largereffect transistor (IGFET) dynamic memories in which applied potential. There is no inversion layer, only de-advantage is taken of the large off-resistance of an IGFET pletion, because a negligible number of minority carriersto store charge on a gate capacitance. are present. This is not a steady-state situation; never-Another element, reported by P. T. Panousis,3 con- theless, the surface will remain depleted for times of the

sists of a metal oxide semiconductor (MOS) capacitor order of seconds before thermally generated minoritythat is charged and interrogated through a bipolar carriers accumulate because the present silicon art pro-switch. A second class of devices includes shift registers, duces material with low densities of surface and bulk-in which charge is passed from element to element in a generation centers. If minority carriers are introducedlinear array.'4 An integrated-circuit version has been through any one of several means to be described shortly,reported by F. L. J. Sangster.' '7 It consists of either bi- they will collect at the surface in the potential minimumpolar or IGFET transistors that are connected in series, defined by the excess potential on the central electrode.with capacitors connecting base to collector or source The potential-energy diagram for the metal-insulator-to drain. Figure 3 illustrates this circuit for IGFETS. All semiconductor (MIS) structure is shown in Fig. 5. Theof these devices are conventional circuit elements con- Fermi energy of the metal appears on the left; on thenected by wires, even though they can be put in inte- right the band edges of the insulator and semiconductorgrated-circuit form. are shown as functions of distance perpendicular to the

surface. On the far right the Fermi level in the n-typeCharge-coupled devices semiconductor is shown. The bias voltage VI is the ap-A third class, charge-coupled devices (CCD), has been plied potential difference between the semiconductor and

reported by the writers," and the principle has been con- metal. Figure 5A shows the potential distribution in thefirmed experimentally.9 These devices represent a con- absence of collected minority carriers; Fig. SB illustrates

the potential distribution when a saturated number ofminority carriers has been collected. In the saturationcondition shown in Fig. 5B, distribution of minority car-riers is such that their diffusion current away from theFIGURE 7. Cutaway of a charge-coupled device in (A) the

storage condition and (B) the transfer condition. surface Is exactly balanced by their dnft toward the sur-face. If less than the saturation value is accumulated, theA V1 = -5V V2 = 1OV V1 = -5V net flow is toward the surface, but if the saturation value

_ is exceeded, there is a net flow into the undepleted bulk,

Silicon FIGURE 8. A three-phase charge-coupled device.A 11 t 16 7 9l__

.. J.. r--- I. . --.B VI -5V V2=-1OV V3=-15V B -VI

-V3

6~~~~~~~~~~~~~~~~~~~~~~~~~L L

_Cs~~~ ~~~~~~~~~~~~~~~~~~x'---' ---I_ J L---

-V2_ - V 2 , E l , | l ,|1~~~~~- V

Boyle, Smith---Charge-coupled devices: A niew approach to MIS device structures 21

Authorized licensed use limited to: IEEE Publications Staff. Downloaded on October 6, 2009 at 11:59 from IEEE Xplore. Restrictions apply.

where the minority carriers recombine, just as in a for-ward-biased p-n diode. With the collection of minoritycarriers in the depletion well, the depletion width de-creases as shown in the potential-level diagram; there-fore, the differential capacity increases and, as shown, thesurface potential decreases. Note that the minority-carrier accumulation held at the surface cannot changewith rapid changes in electrode-to-bulk voltage. There-fore, the capacitance phenomenon consists only of mobilecharge removal or addition at the depletion edge as theelectrode bias is increased or decreased.

In Fig. 6, these three quantities-depletion width Xd,differential capacity C, and surface potential X,-areplotted as a function of the ratio of accumulated minoritycarriers to the saturation value. The plot here is for theparticular values of doping density Nd = , X 1014 cm38,gate bias V, = 10 volts, oxide thickness x0 = 2000 A.This results in an oxide capacitance of CO = 1.7 X 10-8F/cm2. The Debye length, which is a measure of thesharpness of the depletion edge, is 5 X 10-6 cm, whichis much less than the depletion width. For convenience,0, and C are plotted as normalized values to the gatebias and oxide capacity, respectively. As minority carriers

-(V0 + V)AV -(V'- V)

VO

FIGURE 11. A four-phase charge-coupled device that usesV0_ B silicon-gate technology.-V0

FIGURE 12. Input schemes for charge-coupled devices.L--1 r--Ij r--i r--i t--i i .

LTaJ J L J L J A

-(Vo- V) C Surface-('~~~~~~~- V) C~~~~~ avalanche-("Vo V)

I1-- 1_ _,

ii._ _ 2i'._ _8 i _ B vFIGURE 9. A two-phase charge-coupled device.

p-n junctionand gate

FIGURE 10. Basic overlapping metallization structure. Theoxide under plates 1 and 3 would be made thicker andplates land 2,3, and 4 would be connected to make the de-vice shown in Fig. 9.

Metal Oxide

1 \ 2 3 4 C

_J~i~¶-r, lLightl

22 IEEE spectrum JULY 1971

Authorized licensed use limited to: IEEE Publications Staff. Downloaded on October 6, 2009 at 11:59 from IEEE Xplore. Restrictions apply.

accumulate, the differential capacity increases by a factor potentials on 2, 5, and 8 are increased so that the chargesof 3, and the surface potential by a factor of 8. These in- move over one position. In Fig. 8C, electrodes 2, 5, andcreases provide a large signal for determining the pres- 8 have become the storage sites. In this way, coded in-ence or absence of charge. The physical position of this formation can be shifted along the linear array. A varietycharge in the storage condition is indicated by the plus of geometries other than this three-phase structure utilizesigns in Fig. 7A. the same basic concept. One (Fig. 9) is a two-phase systemThe interesting feature of this structure is that it is in which the potential on adjacent electrodes alternates

possible to place the metal electrodes close enough to- between Vo - V and V0 + V. The electrodes have stepsgether and to apply voltage differences of sufficient mag- along their length; that is, the oxide thickness is steppednitude to obtain the potential distribution shown in Fig. so that a different potential appears beneath each in-7B. In this case, the bias potential V3 exceeds V2, which dividual electrode.'9'21 The shifting operation is analo-causes carriers to be transferred from one electrode to gous to that obtained with the triplet geometry pre-the next. Subsequently, the potentials on the electrodes viously described.can be readjusted so that the quiescent storage site is Figure 10 shows a basic structure that implements thelocated at the third electrode. Figure 8 shows just such a two-phase device by connecting the electrodes eithersequence, which effects transfer along a linear array. The directly, capacitively, or with an external dc bias. The useelectrodes are connected in groups of three and operated of overlapping electrodes also reduces the separationwith a three-phase voltage supply to give direction to the between plates from a photolithography tolerance to antransfer operation. The left column shows the potentials insulator thickness. Alternatively, the device can be usedapplied to each electrode. The response V, is sufficient to with a four-phase drive. Figure 11 shows such a deviceprovide a depletion region in the semiconductor. Al- in which silicon-gate technology is employed.though V2 is larger than V, and produces the storage site At the start of an array of shifting electrodes the chargepreviously described, V3is still larger and effects the trans- can be generated by any one of a number of means.fer. In Fig. 8A, charges are stored under electrodes I and Some of these are shown in Fig. 12. In Fig. 12A, charge is7; no charge is stored under electrode 4. In Fig. 8B, the generated through the application of a large enough pulse

to produce avalanche breakdown in the semiconductor.Figure 12B illustrates a means of injection that uses theleft electrode to produce an inverted region adjacent to a

Injection into substrate Ap-n diode formed in the bulk. The p region is wired to the

L1 I1LI1+vFIGURE 15. Photograph of the fabricated device of Fig. 14.

p-n junction or Schottky barrier I(reverse biased) -

Floating gate -

-vs~~~~~~~~~~~~vFIGURE 13. Output schemes for charge-coupled devices.

FIGURE 14. Longitudinal cross section of a charge-coupled8-bit shift register.Input Input Transfer pads output Outputdiode gate gate diode

Boyle, Smith-Charge-coupled devices: A new approach to MIS device structures 23

Authorized licensed use limited to: IEEE Publications Staff. Downloaded on October 6, 2009 at 11:59 from IEEE Xplore. Restrictions apply.

-v1 -V2 V1 < v2 bulk and becomes a source of minority carriers to supportA the inversion. A minority-carrier current flows along the

i________________________________ surface to the right and delivers a minority-carrier ac-

l + + _ cumulation to the region under the next electrode, whicht-) - - - - - - - -is biased to produce a potential well. This is similar to<__________J gating charge from source to drain in an iiFFr. Figure

L| L 12C shows that the minority carriers can be generatedoptically and collected under the electrodes. In this lattermode of operation the device operates as an imaging

B device and has the built-in features of image storage andt = 0 B scanning.

If{ 5 \Figure 13 illustrates methods of detection. In Fig. 13A,

c the substrate connects to ground through a resistor.v I + + + + When charge is transferred to the last electrode, a positive

voltage causes holes to be injected into the substrate,and a current passes through the ouput resistor. In Fig.

C 13B, a diode at the end of the line is reverse-biased to ao < t C voltage - V. which is more negative than any of the stir-

/+ \ \ face potentials used for transfer. When charge is trans-c ferred to the diode position, it produces an output currentv + + + current !, in the external circuit. In Fig. 13C, the fact thatFIGURE 16. The transfer of charge from one plate to the capacitance of the MOS structure changes with chargeanother. C and V refer to the conduction and valence is used in a capacitive division circuit, where the voltagebands. V. could be connected, for example, to the gate of an

IGFEr.A cross section of one of the device structures studied'1

is shown in Fig. 14. The transfer pads are arranged for athree-phase operation, P,,, P.,, P::,, -, P;,8 with every

FIGURE 17. The fraction of charge remaining under a plate third pad connected together as shown in Fig. 10, whereas a function of normalized time. The abscissa is plated as/T, which spreads the region where T < 1, and results in the three-phase drive was illustrated. Injection is con-straight parallel lines over the diffusion portion. trolled by the potential on the input gate, which can in-

100 ~~~~~~~~~~~~~~~~ducean inversion layer that connects the first plate to asource of minority carriers at the input diode. A symmetri-

6011 11 S | _ cal arrangement is used for detection of the output signal

with a reverse-biased diode.40 ll a IW Figure 15 illustrates this structure. In the central sec-

tion, where the transfer pads are located, there is a thin(1200-A) oxide and beyond that region a thick (5000-A)

20 ~~~~~~~~~~~~~~~oxide,which prevents depletion in regions other thanbeneath the transfer electrodes. All of the second elec-trodes in transfer triplets are connected together by means

10 | 1|1 | __ of a diffused crossunder. In this early device, the plates10 S 11 1|11 1 are 250 pm by 50 pm and are separated by 3-Am gaps.

Of particular interest are the mechanisms that inhibitP 86 ¢l 8 i8 the complete transfer of charge from one plate to the

CL ~~~~~~~~~~~~~~~~next.Our early work indicates that this is not a simpleoh 4 ~~~~~~~~~~~~~~problemand depends on at least two distinct factors.

The first involves the dynamics of the free carriers thatundergo motion through both the miechanisms of dif-

129 ~~~~~~~~~~~~~~~~fusionand drift. The second involves trapping effects atinterface states. Our preliminary data indicate that, withthe material system and geometries we have used, boththese effects are present but are not sufficiently large to

0.81 * norypreclude useful applications. We have, for example, made0.25 1 4 9 16 2s 36 49 tobservations on structures where the total amount of

charge left behind at each transfer is less than 0.1 percent0.4 ~~~~~~~~~~~~~~~~~atshifting times of 2 pAs.

A qualitative feeling for the magnitude of these twoefrects may be obtained by first considering the dynamics

0.2 of minority carriers. Figuire 16A illustrates the distribuitionof carriers the instant after bias has been increased on thetransfer electrode of length L. Figure 16B shows the

0.10.51 4 9 16 2 6 4

corresponding potential distribution. In Fig. 16C, the po-0.251 4 ormaize ti6e 256t4 tential distribution has changed because a few of theNormalized time, r - ,t minority carriers have moved out through the action of

24 liEE spectrum JULY 1971

Authorized licensed use limited to: IEEE Publications Staff. Downloaded on October 6, 2009 at 11:59 from IEEE Xplore. Restrictions apply.

1.0 E _ e * e l 111 ll lli | I 7 _ _ _ diffusion alone. At this point in time, however, the re-arrangement of potential is such as to provide a drifting

0.8 |11 i | l |l | l l | 1E11 field for some of the remaining carriers over part of theirC ~~~~~~~~~~~~~~~path.This effect, as well as the enhanced transfer eflii-

ciency resulting from fringing fields that exist in finite

.i 0E geometries, are discussed qualitatively in Ref. 8.... Mathematically, this means solving the transport equa-

* . *..~~~~~~. ~~.~~ ton in the presence of a potential distribution that is ac ..~~~~~~~~~~~~function of the free carrier density. Approximate solu-

tions to these coupled equiations show that the drift com-

0. 2 4 6 8 10 12 N ponent is much larger than the diffusion component ex-Number of triplets cept for the last fraction of charge to be transferred.' 11,2,20

FIGURE 18.Transfer efficiency as a function of the number Aplot'2 of the amount of charge remaining at a site as aof bits for 150 kHz and 1 MHz. function of time after a square-wave pulse is applied to an

adjacent electrode is shown in Fig. 17. The normalizationFIGURE 19. Transfer efficiency as a function of the number time is m,, = L2/1u,,,, where L is the length of a plate, ,u isof bits for various operating voltages. the mobility, and ii,, = I volt. The quantity V, = Sq,, is a

1.0 ~~~~~~~~~~~~~measureof the initial quantity of charge stored. There isa rapid drop initially, which is faster for a large amount ofcharge (storage voltage = V, = 10 volts) than for the

0.8 s~~~~~~~~~~maller amount (V., = 2 volts). When only a small2< amount of charge remiains, the slopes change as both be-

.~come diffusion-limited. These calculations essentially~~ ~ agree with the experimentally observed charge-transfer

6 2N efficiencies..um. rThe upper limit of the amount of charge lost to surface

.Rtatescan be estimated as follows: The maximum n omberof charges stored Ns is the oxide capacity times the gatevoltage divided by the electron charge. For typical valuesthis is 2 X 1012 charges/cM2. If the surface is kept under

0.4 i~~~~~~~~~~~~~nversion at all times, only those states that have a time~~~~~~~ ~~~~~~constant that corresponds to the time allowed for shifting

.2 ~~~~~~~~~~~~~~willbe effective in reducing transfer efficiency. With a~~ ___ ~~~~~~~ density of states of 10ll Cm- 2. eVW and if a band kTwide

,.. ~~~has the right time constant, the maximum fraction lost!V ~ ~will be2 X 10- .

- The transfer mechanism has been studied" in an opti-cal injection experiment whereby a light source, whichmoves along the array, provides a reproducible amount ofinjection at any one of the electrodes in a sequence.

0.2 ~ .5Figures 18 and 19 show some data that have been ob-tained in this way. In Fig. 18, the total efficiency plottedas an ordinate is the ratio of the amount of charge re-covered at the collecting diode when the light beam is

S positioned at one of the electrodes divided by the chargeobtained when the beam is positioned at the last elec-trode. The abscissa is the number of triplets along whichthe charge has passed. The voltages used are the same at

Number of tripletsFIGURE 20. A 96-bit, three-phase CCD. The enlarged por-tion shows the output end of the device.

Boyle, Smith-Charge-coupled devices: A new approach to MIS device structutres 25

Authorized licensed use limited to: IEEE Publications Staff. Downloaded on October 6, 2009 at 11:59 from IEEE Xplore. Restrictions apply.

_~ 1x tacted by another deep diffusion through the epitaxial7 jAm

layer. In total, five separate diffusion steps are required,7,ffim each with its accompanying oxide masking and photo-*7t lithography steps. After all the desired impurity profiles

have been developed, two more masking steps are re-p- quired to make ohmic contact to the transistor and

delineate the metalization pattern. In the past few years,some simplification of bipolar structures has taken place.FIGURE 21. A cross section of a junction-isolated planar For example, by the use of much thinner epitaxial layers,it is possible to arrange for the n+ collector diffusion toperform the dual function ofproviding isolation and mak-ing contact to the collector. Even further simplificationhas been reported recently by B. T. Murphy et al." Thefield-effect devices have from their earliest days been

both frequencies and the percentage, in parentheses, is the conceptually easier to fabricate-a standard insulated-transfer efficiency per electrode. We do not have enough gate field-effect transistor requires only one diffusion fordata as yet to map out in detail the frequency dependence the establishment of the source and drain regions.of the transfer process; however, the early data and cal- Charge-coupled devices require no such diffusion forculations indicate that the transfer efficiency falls off their active parts and can be conceptually fabricated withrapidly with frequency changes. The transfer phenomenon two masking steps-one for the metal pattern, another tois even more complicated for the transfer efficiency as a define the regions where thick and thin insulating film isfunction of resting and transfer potential. Figure 19 required. In practice, IGFEr-type structures are used forshows that the transfer falls off very rapidly as the resting input-output and regeneration at the periphery of largepotential is reduced from 3 volts to 1 volt. There is a arrays. Since these elements constitute only a small frac-further complication in the variation of the transfer with tion of the area of a chip, overall yields should be highthe transfer potential V, (with a constant bias potential). because of the basic simplicity ofthe CCD elements.An unexplained maximum occurs for a transfer potentialof 10 volts. Some of the complexity may arise from the New applications for semiconductor devicesrelatively large separation between electrodes, which re- By and large, the major application for all semicon-sults in poorly defined potentials in the regions of free ox- ductor devices has been in the time-honored role of anide surface lying between the electrodes. Experiments un- active device that can provide gain or act as a switch. In-der way use structures similar to Fig. 11 and will give deed, it is surprising and somewhat disheartening to findmuch better control of the potential distribution over all that all integrated circuits are still recognizable in termsregions through which the carriers move. In this way, some of discrete building blocks of resistors, capacitors, andof the anomalous results may be resolved. These better- transistors. At last the situation is beginning to change.defined conditions will provide a new tool to explore This is now the case in the search for elements that willexperimentally and directly new facets of interface states. perform the memory function. The change is also of par-Not only should we be able to determine surface-state ticular importance in some large digital machines, suchdensities very close to the band edge, but we will also as electronic switching systems, where there is at least anbe able to examine the capture and emission cross sections order-of-magnitude more memory elements than logicas a function of the electric field parallel to the surface. gates or switches. In the past, a bistable flip-flop circuitAdvantage has been taken of these high transfer effi- has been used to form a memory cell without the use of

ciencies in applying the charge-coupled-device concept to magnetic elements. Today, most of the fast memoriesimage sensors'4 (Fig. 12). Figure 20 illustrates a three- made from semiconductors make use of this principle.phase 96-bit linear array. Charge is transferred from right A much more natural approach is to make use of chargeto left in this photograph, with input-output diodes lo- storage. Some promising results have been obtained incated at intervals of 32 bits. This device has 10-,um plates MIS structures by using a dual dielectric and storingwith 3-,um spacings between them and has been run at 2 charge at the interface between the two dielectrics or in aMHz with better than 98 percent efficiency. floating metal gate sandwiched between two insulators.16,17The device structure that uses the charge-coupled Unfortunately, successful operation depends on being

process is significant because it makes use of some of the able to pass current in a controlled manner through one ofbasic mechanisms in semiconductors, and it may also sug- the insulators. Three more direct approaches were men-gest some new experimental tools for basic measurement. tioned earlier. There is every expectation that memoriesHowever, it also typifies the direction that the tech- made by the use of the stored-charge principle will be lessnology is taking at this time. It does this in a number of expensive and faster, and will require less power forways. operation than any of the magnetic memories that are

now in existence. It should also be possible, of course, toTrend to simpler structures make use of the shifting principles that are inherent in theFor the past ten years the majority of integrated cir- charge-coupled devices to devise new ways of interrogat-

cuits have used a basic structure that has evolved directly ing memories other than the coordinate address approachfrom planar discrete-device technology. Figure 21 shows that is used now.a cross section of a junction-isolated planar transistor.A deep p-type diffusion is formed at the periphery of the New approaches to signal processingn-p-n transistor at the center of the diagram. A buried As measured by the fundamental physical limits, allcollector, shown by the n+ region at the center, is con- present digital machines are extremely wasteful of power.

26 EEE spectrum JULY 1971

Authorized licensed use limited to: IEEE Publications Staff. Downloaded on October 6, 2009 at 11:59 from IEEE Xplore. Restrictions apply.

A figure of merit for a logic gate is equal to power times 11. Engeler, W. E., Tiemann, J. J., and Baertsch, R. D., "Surfaceswitching delay. This quantity has the units of energy charge transport in silicon," AppL. Phys. Lett., vol. 17, p. 469, 1970.

energ. 12. Strain, R. J., and Schryer, N. L., "A nonlinear diffusionrequired to process one bit of information. In practice, analysis of charge-coupled device transfer," Bell. Syst. Tech. J.,the lowest delay product of the best semiconductor de- July/Aug. 1971.vices lies in the range of 10-12 joule. R. Landauer's has 13. Amelio, G. F., private communication.shown by phenomenological reasoning that no more 14. Bertram, W. J., "Application of the charge-coupled deviceconcept to solid-state image sensors," 1971 IEEE Internat'l Cony.than a few times kT should be required for the switching Dig., pp. 250-251.function. Because kT is approximately 10-20 joule, there 15. Senhouse, L. S., Kushler, D. L., and Murphy, B. T., "Baseare eight orders of magnitude between today's per- diffusion isolated transistors for low power integrated circuits,"

IEEE Trans. Electron Devices, vol. ED-18, pp. 355-358, Juneformance and what is theoretically possible. Nowhere 1971.else, at least in the field of electronics, is there such ap- 16. Kahng, D., and Sze, S. M., "A floating gate and its applicationparent inefficiency! One does not have to search far for to memory devices," Bell Syst. Tech. J., vol. 46, p. 1283, 1967.the reasons for such poor practical results. Because we are 17. Frohmann-Bentchkowsky, D., "A fully-decoded 2048-bit

junctions to obtaincelectrically-programmable MOS-ROM," ISSCC Dig. Tech.using junctons to obta nonlinearities-in other words, Papers, pp. 80-81, 1971.switching action-the potential variations must be at 18. Landauer, R. W., IBMJ. Res. Develop., vol. 5, p. 183, 1961.least several kT/q volts or a fraction of a volt. The small- 19. Kosonocky, W. F., and Carnes, J. G., "Charge-coupled digitalest structures that can be fabricated have dimensions that circuits," ISSCC Dig. Tech. Papers, pp. 162-163, 1971.produce a capacitance of 10- 12 farad. At each switching 20. Kim, C. K., "Carrier transport in charge-coupled devices,"

Dig. Tech. Papers, pp. 158-159, 1971.operation, CV2 of energy must be dissipated and this is 21. Tompsett, M. F., "A simple charge regenerator for use withapproximately equal to 10- 12 joule. charge-coupled and bucket-brigade shift registers and the design ofOne system for performing logic that circumvents these functional logic arrays," ISSCC Dig. Tech. Papers, pp. 160-161,

practical limits for junction devices is the one that uses 1971.

pracntical limis forjunictio devies2i2theone hat ses22. Bobeck, A. H., and Scovil, H. E. D., "Magnetic bubbles,"

magnetic bubbles for logic.22 Sci. Am., vol. 224, pp. 78-90, June 1971.The energy dissipated in the magnetic material per

switched bit may be as small as 4 X 10- 14 joule. Similarly,a highly idealized charge-coupled device in which charge Reprints of this article (No. X71-071) are available tois transported by a moving sine-wave potential well gives readers. Please use the order form on page 9, whichthe same loss per bit (at 10 MHz) on the semiconductor gives information and prices.slice. Both calculations neglect power-supply loss.These charge-coupled devices had their origin in the

search for semiconductors with device mechanisms similarto those obtainable from magnetic domains. Whether they Willard S. Boyle (F) iswill be able to remove the eight orders of magnitude be- executive director of thetween practice and theoretical limits is not yet clear. One nentsmDicnuctor Compo-

thing, however, is evident: Much is still left to be done in Telephone Laboratories,getting a better understanding of the fundamental physics Murray Hill, N.J. A nativeand chemistry of semiconductor structures. This is an of Canada, he receivedextremely fruitful field for the development of devices of the B.Sc., M.Sc., andPh.D. degrees from Mc-practical importance. Gill University in 1947,

1948, and 1950 respec-REFERENCES tively. Dr. Boyle joined

Bell Laboratories in 19531. Broers, A. N., Lean, E. G., and Hatzakis, M., "1.75 GHz and was later appointed head of a department thatacoustic-surface-wave transducer fabricated by an electron beam," did some of the early work on solid-state lasers. InAppl. Phys. Lett., vol. 15, p. 98, 1969. 1962 he was named director of space science and ex-2. Wood, J., and Ball, R. G., "The use of insulated-gate field-effect ploratory studies at Belcomm, Inc. He returned totransistors in digital storage systems," ISSCC Dig. Tech. Papers, tlorator ies in 1964 and a e re sentpp. 82-83, 1965. the Laboratories in 1964 and assumed his present3. Panousis, P. T., "A TRIM bipolar charge storage memory," position in 1969. Dr. Boyle isthe authorof a numberofpresented at International Electron Devices Meeting, Washington, published articles, holds 11 patents, and is a memberD.C., October 28-39, 1970. of APS.4. Janssen, J. M. L., "Discontinuous low-frequency delay line withcontinuously variable delay," Nature, vol. 149, pp. 148-149, Jan. George E. Smith (SM) is26, 1952. head of the Interface5. Hannan, W. J., Schanne, J. F., and Woywood, D. J., "Auto- Device Department atmatic correction oftiming errors in magnetic tape recorders,"lEEE Bell Telephone Labora-Trans. Military Electronics, vol. MIL-9, pp. 246-254, July/Oct. tories, Murray Hill, N.J.1965. He received the B.A. in6. Sangster, F. L. J., "Integrated MOS and bipolar analog delay physics from the Uni-lines using bucket-brigade capacitor storage," ISSCC Dig. Tech. versity of PennsylvaniaPapers, pp. 74-75, 1970. in 1955 and the M.S. and7. Sangster, F. L. J., and Teer, K., "Bucket-brigade electronics- Ph.D. in physics fromNew possibilities for delay, time-axis conversion, and scanning," the University of ChicagoIEEEJ. Solid-State Circuits, vol. SC-4, pp. 131-136, June 1969. in 1956 and 1959 respec-8. Boyle, W. S., and Smith, G. E., "Charge-coupled semiconductor tively. He joined the lab-devices," Bell Syst. Tech. J., vol. 49, pp. 487-493, 1970. oratories in 1959 and initially studied the electrical9. Amelio, G. F., Tompsett, M. F., and Smith, G. E., "Experi- properties and band structure of semimetals. Hismental verification of the charge-coupled device concept," Bell. primary interests are in the areas of new semicon-Syst. Tech. J. vol. 49, p. 593, 1970. ductor devices and the device physics of semicon-10. Tompsett, M. F., Amelio,G. F., and Smith, G. E., "Charge- ductor-insulator interfaces.coupled 8-bit shift register," AppI. Phys. Lett., vol. 17, p. 111,1970.

Boyle, Smith-Charge-coupled devices: A new approach to MIS device structures 27

Authorized licensed use limited to: IEEE Publications Staff. Downloaded on October 6, 2009 at 11:59 from IEEE Xplore. Restrictions apply.