`
`225
`
`Some problems of MOStechnology
`
`J. A. Appels, H. Kalter and E. Kooi
`
`Introduction
`
`Scientists and engineers working in MOS transistor
`technology are charged with the production of MOS
`transistors and integrated circuits that possess certain
`specified characteristics, are stable in behaviour, and
`give high productionyields. The specified requirements
`determine the various steps in the production process:
`from the design geometryto the choice and techniques
`of oxidation, etching, diffusion and other processes in
`the manufacture of a MOS transistor !"!. Some of
`
`magnitude of the charge induced in the channel. A
`theoretical analysis based on detailed physical consid-
`erations has shownthat this is to be expected lI,
`It has also been found that the surface mobility is
`dependent on the crystal orientation at the surface. For
`electrons the mobility is greatest for the (100) plane of
`silicon; the surface mobility in this plane can even ap-
`proach the value of y« in the bulk. For the holes the
`
`the problems which this involves are described in this
`article; the structure and operation of the MOStran-
`sistor, which are dealt with elsewhere in this issue [?!,
`are assumedto be generally familiar to the reader.
`A typical example of a quantity that is determined
`by design geometry and technological processes is the
`transconductance of the MOStransistor. In the article
`just noted [2]
`it
`is shown that
`the transconductance
`—andhence the current that the transistor can carry
`at the maximumpermissible gate voltage—is propor-
`tional to
`
`
`
`B=
`
`eCoxwil.
`
`teanm Gl
`
`is the mobility of the charge carriers in the
`Here yz
`channel, Cox the capacitance of the gate per unit area,
`wis the width and / the length of the channel ( fig.
`/).
`The mobility « depends on the semiconductor
`matcrial of which the MOStransistor is made. For
`
`practical reasonsthis is almost invariablysilicon. One
`of these reasons is that it
`is relatively simple to apply
`effective isolating layers to silicon by oxidation. Al-
`though impurity centres or defects may be present at
`the Si/SiO»interface, the nature and concentrations of
`these impurities can nowbe controlled, and they can
`in fact be used toalter the behaviour of a MOStran-
`sistor in a desired direction. Much ofthis article will
`be concerned with the Si/SiQ»interface.
`In the bulk of the silicon the mobility 4 may be
`regarded as a constant of the material. At the surface
`the mobility is usually appreciably lower than in the
`bulk. Not only may it be affected here by the impurities
`or defects, but it
`is also found that ~ decreases with
`increasing gate voltage, and therefore depends on the
`J. A. Appels, Drs. H. Kalter and Dr. E. Kooi are with Philips
`Research Laboratories, Eindhoven
`
`|. Schematic diagram of a MOS transistor, made on a
`Fig.
`P-type silicon substrate. Two diffused zones of N* silicon con-
`stitute the source S and the drain D. Between them, isolated by
`an oxide layer Ox, is a metal control electrode, the gate G.
`If
`G is sufficiently positive, a concentration of free electrons occurs
`under the gate, forming an N-type conducting channel between
`source and drain. The length / and width wof the channel, and
`the thickness / of the oxide are the chief factors that determine
`the characteristics of the MOStransistor.
`
`mobility is greatest for the (111) orientation, but hole
`mobility is substantially Icss than electron mobility.
`To achieve the maximum carrier mobility, and hence
`the maximumtransconductance, the best choice is an
`N-channel transistor on a silicon chip whose surfaceis
`oriented in the (100) plane.
`is also necessary to
`If a high value of # is desired it
`have a high Cox (see equation 1), and for this purpose
`the oxide layer under the gate is made as thin as pos-
`sible. The minimum thickness is mainly determined by
`
`“) A description of the photo-etching and diffusion processes is
`given in: A. Schmitz, Solid circuits, Philips tech. Rev. 27,
`192-199, 1966.
`(2) J. A. van Nielen, Operation and d.c. behaviour of MOS tran-
`sistors; this issue, page 209.
`1 N. St. J. Murphy, F. Bers and 1. Flinn, Carrier mobility in
`MOStransistors; this issue, page 237.
`
`Page 1 of 13
`
`TSMCExhibit 1005
`
`
`
`226
`
`PHILIPS TECHNICAL REVIEW
`
`VOLUME 31
`
`the breakdownfield-strength (about 103 V/um); prac-
`tical values frequently lie between 0.05 and 0.25 um.
`The dimensionsofthe silicon chips set an upperlimit
`to the width w of the channel, and of course the chance
`of a defect increases with increasing w. A width of a
`few millimetres is fairly easy to achieve, and special
`techniques can be applied to give a channel with a
`width of a few centimetres [4],
`The length / of the channel cannot be made very
`small without running the risk of “punch-through”,i.e.
`a flow of current between source and drain outside the
`channel. The length / is usually a few microns, but
`special methods can be usedto bring it down to about
`1 micron. A very short channel is particularly impor-
`tant in MOStransistors for the UHF band (5).
`In addition to the transconductance 8, the parasitic
`capacitances play an important part in fast transistors.
`The mostdetrimental oneis usually the feedback capac-
`itance between drain and gate [6]. This capacitance
`depends on the amountof overlap between drain and
`gate: it can be reduced by bringingthe gate into accur-
`ate register with the channel region. Various useful
`methods that we have developed for this will be
`discussed in this article.
`The speedofintegrated circuits made with MOStran-
`sistors is mainly limited by the parasitic capacitance
`between wiring and substrate. MOS transistors are
`therefore made with thick oxide layers underthe wiring
`but with thin oxide layers at the active regions. This
`approachalso tends to prevent the formation of para-
`sitic MOStransistors; these can be formed when a
`voltage applied to a conductor induces a conducting
`channel in the substrate underneath the conductor. In
`the transition from the thick oxide to the thin oxide
`there has to be a step in the metallization; this has
`often proved to be a weak spot. We havetherefore
`developed a process in which the thicker oxide is
`embedded deeperin the silicon substrate, so that any
`steps above the surface are smaller. This is known as
`the LOCOSprocess (local oxidation of silicon), and
`will also be described in this article.
`First of all we shall take a closer look atthesili-
`con/silicon-dioxide interface. The surface defects pres-
`ent there and the contact potential of the gate metal
`and the substrate doping all have an important effect
`on the threshold voltage, i.e. the minimumgatevoltage
`needed to form a channel 2). In fact these defects can
`have a much greater
`influence than the contact
`potential and substrate doping. They can change the
`threshold voltage by tens of volts, whereas the changes
`due to differences in contact potential between dissim-
`ilar metals and the variation of substrate doping that
`occurs in practice amount to only a few volts. The
`presence of mobile ions can result in a slow change
`
`in the threshold voltage. Control of the threshold
`voltage and making surethatit is stable are the main
`factors that decide which technology should be followed.
`
`Thesilicon/silicon-dioxide interface
`
`treatment given here of the sili-
`The theoretical
`con/silicon-dioxide interface makes no pretence at
`being complete, but is a simple model that is neverthe-
`less capable of explaining many experimental results,
`and one that has also been found useful for qualita-
`tively predicting the behaviour of the Si/SiO2 system
`from the processing conditions that were used whenit
`was made. In this model we distinguish between defects
`of two kinds:
`a) Surface states — states that can exchange charge
`with the silicon, and which can be described in
`physical terms as quantum states with an energy
`level between the valence and conduction band;
`b) Oxide charge — fixed positive charges (ionized
`donors) near the interface and presumably in the
`oxide.
`Weshall now consider both types of defect in turn.
`
`e
`
`e
`
`e
`
`Si: Si
`
`3
`
`~ Si: Sit Sit
`
`~- Si2T0: Si?
`
`6
`
`O
`O
`ee
`e
`ar)
`Si Ss3t Sit
`
`St Sits:
`
`Fig. 2. a) Crystal lattice of silicon. At the surface of the crystal
`(top of the figure) each atom has an unpaired electron. 6) Where
`the surface of the silicon crystal is covered with silicon dioxide,
`the lattices of the two substances do not exactly match. As a
`result silicon bonds remain unsaturated in places.
`
`Page 2 of 13
`
`
`
`1970, No. 7/8/9
`
`MOS TECHNOLOGY
`
`227
`
`Surface states
`bonds mayact not only as electron donorsor traps for
`holes, but also as traps for electrons, since trapping an
`If the crystal lattice terminates abruptly at the sur-
`face ofasilicon chip, then a large numberof unsaturat-
`electron changesa silicon atom with an unpaired elec-
`ed silicon bonds are to be expected,i.e. each atom in
`tron into an atom witheight electronsin its outer shell.
`the outside layer of silicon atoms should have an un-
`This is the inert-gas configuration:
`paired electron (jig. 2a). Since there are about 1015 Si
`:Si: +e“ =:Si: wee (4
`atomsper cm? at the surface, one would expect about
`the same numberof unsaturated bonds on a “clean”
`surface. If the silicon is oxidized, as it is in the case
`underconsideration, then the number of unsaturated
`bonds is of course lower, but it is not equal to zero
`because there will probably not be an exactfit between
`the Si and SiOe networks (fig. 25). We shall now. con-
`sider what electrical effects can result from the un-
`saturated silicon bonds.
`It is very probablethatit will take less energy to raise
`an unpaired electron into the conduction band than to
`raise a paired valence electron; in other words, the un-
`
`Theeffect may also be seen as the giving-up of a hole:
`
`‘Si: :Si: fet, wees (5)
`
`and we may then conclude that the relevant energy
`level must lie in the forbidden band.
`From a wide variety of measurements 7) it has been
`found that energy levels do in fact occur in the for-
`bidden band, and that broadly two groups may be
`distinguished: a group near
`the conduction band
`
`C5555OR QOGK0
`
`D
`LUMUP0G00:0:4,00 0-0:
`Met
`O55HSC
`
`%
`
`SX
`
`
`Ey xx
`A EABER BED SSS?
`
`VES HIIHG
`SO OO 0000
`
`
`Meeatetatatetatatetctetetctetetateteteteteten
`
`
`
`Fig. 3. The unpaired electron of a silicon atom with an unsaturated bond has an energy Ess
`which lies in the forbidden band between valence band (energy Ey) and conduction band
`(energy E;). The atom mayoccur as a donor; by giving up an electron (ontheleft) or taking
`up a hole (on the right) it then acquires a positive charge. If there is a high electron con-
`centration the atom may also occur as an acceptor and acquire a negative charge.
`
`paired electron possesses an energy level thatlies in the
`forbidden band. A silicon atom to which such anelec-
`tron is bound may give up this electron or take up a
`hole, but in both cases the atom itself becomes posi-
`tively charged (fig. 3):
`
`+
`‘Si: Si: te, 2... Q)
`:
`+
`et+ :Si:ae Si: 2.
`
`2 ew.
`
`. 8)
`
`— these are probably acceptor levels — and a group
`near the valence band — probably donorlevels. De-
`pending on the voltages applied in the measurements,
`there is a tendencyfor electrons or holes to concentrate
`at the Si/SiOe interface; if there is a high electron con-
`centration the defects act mainly as acceptorlevels, but
`at a high hole concentration mainly as donor levels.
`On the same sample the number of acceptor levels
`found in one measurementis invariably almost equal
`
`
`If we are dealing, for example, with P-typesilicon,
`then there are many holes and the equilibria (2) and (3)
`shift to the right. If moreover the energy gap Ess— Ey
`is small, a number of holes from the silicon may be
`trapped, and therefore the hole conduction near the
`surface of the crystal is not so good as in the bulk of
`the material.
`It is also conceivable ‘that the unsaturated silicon
`
`(4) R. D. Josephy, MOStransistors for power amplification in
`the HF band; this issue, page 251.
`{51 R. J. Nienhuis, A MOS tetrode for the UHF band witha
`channel 1.5 um long; this issue, page 259.
`.
`(6] Pp, A. H. Hart and F. M. Klaassen, The MOStransistor as
`a small-signal amplifier; this issue, page 216.
`(7] E. Kooi, The surface properties of oxidized silicon, Thesis,
`Eindhoven 1967.
`M. V. Whelan, Influence of charge interactions on capaci-
`tance versus voltage curves in MOSstructures, Philips Res.
`Repts. 20, 562-577, 1965; Electrical behaviour of defects ata
`thermally oxidized silicon surface, Thesis, Eindhoven 1970.
`
`Page 3 of 13
`
`
`
`228
`
`PHILIPS TECHNICAL REVIEW
`
`VOLUME 31
`
`to the numberof donorlevels found in another meas-
`urement;
`this lends plausibility to our assumption
`that the same trapping centres are involved in both
`cases.
`.
`
`‘ The assumption that the centres are related to un-
`saturated silicon bonds explains why the number of
`surface states depends on the crystal orientation of the
`silicon surface. If this is a (111) plane, then there are
`usually 3 to 5 times as many surface states as on a
`(100) plane. This suggests that the oxide networkfits
`better on a (100) crystal plane than on a (111) plane.
`Othercrystalorientations give various numbersofsur-
`face states that lie between those ofthe (100) and (111)
`planes.
`The way in which the surface states can affect the
`characteristics of a MOStransistor will be demon-
`strated by means of a numberof experimentaltransis-
`tors on a P-type silicon substrate, i.e. with an N-type
`channel. This channel would have to be induced by
`applying a positive voltage to the gate. Since the effect
`of this is a decrease in the concentration of holes near
`the silicon surface and an increase in the electron con-
`centration, the equilibria (2) and (3) shift to the left
`and the equilibria (4) and (5) to the right. This means
`that the donor states tend to becomeneutral(if they
`were not neutral already) and the acceptorstates nega-
`tive. The build-up of negative charge in the surface
`states means that the mobile charge entering the bulk
`ofthe silicon is less than the total induced charge. Con-
`sequently the threshold voltage, required for inversion,
`is higher than expected, and on increasing the gate
`voltage the subsequent increase in the inversion charge
`(and hence in the current through the transistor) is
`lower, and the transconductanceis therefore affected.
`Theeffect of the surfacestates is illustrated in fig. 4,
`which shows the Ja-Vgs curves for the experimental
`MOStransistors that all have the same dimensions but
`were annealed in different gas atmospheres after
`forming thé gate oxide in an extremely dry atmosphere
`at about 1100 °C. During the anneal, the temperature
`was kept
`low (450 °C) compared with the normal
`growth temperature of SiOz on Si (1000°C or
`higher), so that the processing steps could cause no
`difference in oxide thickness. They did, however, give
`rise to differences in the numbers of surface states,
`as may be shown from the threshold voltages and
`transconductances.
`In fact a hydrogen atmosphere
`and water vapour in an atmosphere of wet nitrogen
`even lead to negativethreshold voltages and thus ap-
`pear to remove the surface states for the most part.
`Water vapour in an oxygen atmosphere has consider-
`ably less effect. This suggests that a reduction of water
`to hydrogen plays an important part in the process.
`This hypothesis seems to be confirmed by the experience
`
`Page 4 of 13
`
`that the treatment in wet nitrogen is most effective
`when the chip is heated to a high temperature in an
`inert gas immediatély after the silicon is oxidized. This
`treatment
`reduces the oxygen content of the SiOz
`through the influence of the silicon beneath it.
`The simplest explanation for the disappearance ofthe
`surface states is a chemical reaction of hydrogen with
`the centres involved, i.e. the formation of SiH groups
`in our model. This explanation has been confirmed by
`infra-red absorption measurements[8], With the aid of
`a sensitive method of measurement it has been shown
`that
`the SiOz almost
`invariably contains a certain
`number of SiH groups, whose concentration is par-
`ticularly high when the oxidized surface is subjected to
`operations which reduce the numberof surfacestates.
`Often very little water vapour is sufficient to reduce
`the numberof surface states; a heat treatment in an
`inert gas (e.g. nitrogen or helium) which is not extreme+
`ly dry (containing a few ppm of water) maybeeffec-
`tive. It is also found that treatmentin a fairly dry en-
`vironment mayalso be highly effective if there is a base-
`metal electrode (e.g. of aluminium) on thesilicon sur-
`face. Here again the surface states underthe electrode
`disappear uponheating. It is assumed thatin this case
`a reaction of the metal with traces of water produces
`sufficient hydrogen.
`
`25mMA
`
`20
`[N,b-H,0)
`
`He]
`
`S
`
`O,+H,,0) 0
`
`-30
`
`0
`
`50
`
`Fig. 4. The Ju-Vgs characteristics of a number of geometrically
`identical MOStransistors which have been processedin different
`gas atmospheres after the chips had been oxidized.
`
`
`
`1970, No. 7/8/9
`
`MOS TECHNOLOGY
`
`229
`
`the oxide layer. The sodium atom breaks the bond
`Wemaytherefore conclude that the structure of the
`between an oxygen andasilicon atom,anditself forms
`interface is generally very dependent on the crystal
`a bond with the oxygen atom. Asa result, one ofthe
`orientation ofthe silicon, on the method of growing
`valence electrons of the silicon loses its bond, and as
`the oxide and on the subsequent
`treatments. Many
`this electron is easily released, a positively charged
`experiments can be explained on the assumption that
`centre is formed.
`the silicon bonds are or are not saturated with hydro-
`the interface may conceivably be
`The sodium at
`gen. We can becertain, however,that this does not give
`replaced by otheralkali metals and even by hydrogen.
`a complete description of the interface. A more exact
`This may perhaps explain why, even underfairly clean
`theory would have to take into account, for example.
`conditions, oxidation in steam gives rise to more oxide
`the occurrence of SiOH groups and particularly the
`charge than oxidation in dry oxygen. On the other
`influence of other
`impurities (whether deliberately
`hand, it has also been observed that heating in hydro-
`introduced or not) on theinterface structure. We shall
`return to this in the next section.
`
`
`
`'
`4;
`y
`(4r
`
`/f
`
`°
`
`7
`
`19 3
`10~cm
`Na
`
`108
`
`
`
`
`
`
`LOOWUGD
`
`
`
`
`
`
`
`Fig. 5. Distribution of the concentration of Na atoms in the
`oxide as a function ofthe distance x from the silicon; the hatched
`urea indicates the scatter of the measuring results.
`
`a O.
`
`0
`O
`Si +0. NapSi +0
`O
`° 0.
`
`O
`
`O
`
`2 025102 Ng, Si20
`
`0
`
`0
`
`Fig. 6. a) The location of a sodium atom in SiOe. The sodium
`atom breaks the bond between a silicon and an oxygen atom,
`and as a result one of the valence electrons of the silicon loses
`its bond;
`this electron is easily released and leaves behind a
`positive charge. b) A hydrogen atom can introduce an SiH group
`in SiOz. In this group the hydrogen atom forms a homopolar
`bond with thesilicon and there is no longer an unpaired electron.
`
`(8] These measurements were carried out by Dr. K. H. Beckmann
`and T. Tempelmann of the Philips Hamburg laboratories:
`see K. H. Beckmann and N. J. Harrick, J. Electrochem.
`Soc. 118, 614-619, 1971 (No. 4).
`
`Positive charge at the oxidejsilicon interface
`Anyone assumingthat all the difficulties are resolved
`by a Suitable after-treatment that reduces the number
`of surface states to a negligible value will be surprised
`by the result that, although the /a-Vgs curve has ap-
`proximately the theoretically expected shape after such
`a treatment, the threshold voltage often has a value less
`positive (or more negative) than was expected.
`Indeed, the N-channel MOStransistors offig. 4 have
`a negative threshold voltage after treatment in hydro-
`gen or wet nitrogen; in other words, they already have
`an inversion channel when the gate voltage is zero.
`This effect cannot be explained in terms of the surface
`states, since they have the very effect of opposing the
`inversion,
`Wemust therefore assume that there are other cen-
`tres present in addition to the ones we have mentioned.
`It is usually supposed that the effect is caused by the
`presence of positively charged centres in the oxide
`immediately adjacent to the silicon surface, although
`these are difficult to distinguish experimentally from
`ionized donor centres in the silicon near the surface.
`The amountofoxide charge, like the numberof surface
`states described above, is connected with the interface
`-structure. Again, with identical processing, the oxidized
`(100) planeis foundto give the lowest oxide charge, and
`the (111) plane the highest. Impurities have an impor-
`tant effect, particularly sodium. It has been clearly
`demonstrated [7] that the presence of sodium during
`oxidation can have a marked effect on the amount of
`charge, although the crystal orientation still remains
`important. It has been shown by neutron-activation
`analysis that the sodium hasa distribution in the oxide
`like thatillustrated infig. 5. Most of the sodium can be
`seento lie in the top layer of the oxide, but there is also
`an accumulation at the interface with the silicon. The
`position of sodium in the oxide structure may perhaps
`best be represented as in fig. 6a. This structure may
`be regarded as a somewhat reduced oxide structure,
`which is also to be expected on the silicon side of
`
`Page 5 of 13
`
`
`
`230
`
`PHILIPS TECHNICAL REVIEW
`
`VOLUME 31
`
`gen or in water vapour — particularly at Jess elevated
`temperature (600 °C) — may cause the charge to de-
`crease. Here again, the formation of SiH may be ex-
`pected,resulting for examplein the structureillustrated
`in fig. 66. The hydrogen atom now forms a homopolar
`bondwith the silicon atom, and this no longer has an
`unpaired electron. |
`Fixed negative charge has sometimes been found. It
`can becaused byat least one impurity — gold, which is
`often present in small quantities. Gold is also readily
`maderadioactive by neutron activation and its presence
`demonstrated in this way.
`Thepresence of sodium and of other impurities may
`have a variety of causes. The impurities may come from
`the chemicals used or from the quartz glass tubes in
`which the oxidations are carried out. Another impor-
`tant source may be dust; if this settles on hot quartz
`tubes, sodium ions mayeasily enter the tube by diffu-
`sion and thus mix with the oxidizing gas. To achieve
`good process control it is therefore important to use
`pure chemicals and to protect the quartz tube from
`dust. Where extremely clean oxides are required, water-
`cooled quartz tubes maybe used, and thesilicon chip
`may then be heated by induction heating.
`
`Forthe control of the oxide charge, cleanliness is not the only
`important consideration, and indeed it may not always be neces-
`sary; what is particularly important, as in the case of the surface
`states, is the gas atmosphere. An oxidizing atmosphere increases
`the oxide charge, especially when the temperatureis relatively
`low (the effect is shown for example in fig. 4, where heating at
`450 °C in oxygen does not in fact give a transistor of the de-
`pletion type — because there are so many surfacestates opposing
`inversion — but it does clearly alter the threshold voltage in the
`negative direction, by 15 volts). This effect of oxygen is not yet
`sufficiently understood. It is undoubtedly related to the oxidation
`mechanism: perhaps the, oxygen at the upper surface attracts
`electrons which are then generated by structural change of the
`oxide/silicon interface. It is also conceivable that traces of im-
`purities again play an important part: the transport of hydrogen
`from theinterface towards the supplied oxygen is a likely possi-
`bility, which could cause the structure in fig. 66 to change for
`example to that in fig. 6a. In any case the significance for the
`technologistis that to obtain a low oxide charge hewill have to
`end the oxidation in one way or another by tempering ina non-
`oxidizing gas.
`
`Depending on the process used, the oxide charge is
`found to have a value ranging from less than 10!9 to
`more than 1013 unit charges per cm2. At an oxide
`thickness of say 0.2 wm, this means a change in the
`threshold voltage with respect to the theoretical value
`ranging from less than 0.1 V to more than 100 V.
`Process control has now advanced to a stage where
`variation of the oxide charge with a tolerance of 101°
`charges per cm? is quite feasible. One ofthe results of
`the presenceof the positive oxide charge wasthat it was
`
`originally very difficult to make N-channel MOStran-
`sitors that did not already have an inversion channelat
`zero gate voltage in the absence ofsurface states. This
`is the main reason why most MOScircuits have been
`(and still are) made with transistors of the P-channel
`type: here of course the presence of oxide charge only
`meansthat the threshold voltage is rather more nega-
`tive, since P-channel transitors are always of the en-
`hancement type.
`oF
`
`Determination of oxide charge and surface states by meas-
`urement of the MOScapacitance
`
`Information about the nature.and number of the surfacestates
`can be obtained bya.c. circuit measurements that determine how
`the effective capacitance of the capacitor formed by gate, oxide
`layer and substrate depends on the applied d.c. voltage. A number
`of measurements have been made on MOSconfigurations spe-
`cially designed for the purpose, with the metal contact on the
`oxide much greater than the gate of a transistor but small with
`respect to the dimensionsof the silicon wafer, which wasentirely
`covered on the otherside by a metal substrate contact. These con-
`figurations might be referred to as MOScapacitors ( fig. 7). The
`
`
`
`
`the ter-
`Fig. 7. MOS capacitor. The capacitance measured at
`minals is that of the series arrangement of Csiog and Csi. The
`magnitude of Cs; depends on the applied d.c. voltage.
`
`Fig. 8. Variation of the capacitance C of a MOScapacitor (on
`P-type silicon) with the applied d.c. voltage Vgs in the theoretical
`case where there are no surface states and no oxide charge. In
`this case the energy bands at Vgs = 0 are not bent and the capac-
`itance measuredis the flat-band capacitance Cr. The MOS capac-
`itor has a capacitance equal to Csiog When thesilicon directly
`beneath the oxide is a good conductor. This can be demonstrated
`quite clearly with low-frequency a.c. voltages (curve LF), but at
`higher frequencies the agreement is not so good (HF).
`
`Page 6 of 13
`
`
`
`1970, No. 7/8/9
`
`MOS TECHNOLOGY
`
`231
`
`to about Csio,, but at higher frequencies C remains small. This
`capacitance C measuredat a particular frequency may be regarded
`is because the supply and removalof charge carriers in the inverted
`as the resultant of the series configuration of a capacitance Csio,
`layer of an MOS capacitor cannottake place fast enough at such
`across the oxide layer and a capacitance Csi, which is related to
`high frequencies. In MOS transistors this effect is not so pro-
`the fact that the charge on the lower “plate” of the capacitor has
`nounced,
`the inverted layer here being connected with source
`the form of a space-charge cloud in the silicon. We may write:
`and drain diffusion.
`boot
`Two typical examples of the results of capacitance measure-
`C
` Csiog
`Csi
`ments can beseen in fig. 9a and b. These measurements were not
`made on MOScapacitors but on MOS transistorsspecially made
`If Csidg >>Csi, then C x Csi; if Csi > Csidg then C Csiog-
`for the purpose, since it was also required to measure the drain
`In these equations Csio, is a constant, but Cs; depends on the
`current Jag. This is also included in the figures. The transistors
`thickness of the depletion layer, that is to say on the applied
`on which the measurements represented in fig. 9a and fig. 9b were
`voltage and on the doping concentration of the silicon.
`carried out had the same shape and dimensions and were made
`If there is no oxide charge and there are no surface states, we
`on P-type substrates having the same conductivity (50 Qcm).
`may expect the relation between C and Vgs in a MOSconfigura-
`There wasonlyaslight difference in the production process: after
`tion on a P-type substrate to be represented by curves like those
`oxidation (16 hours at 1200 °C in oxygen) and a phosphorus
`in fig. 8. This figure also applies to N-type material, provided
`diffusion (4 hours at 1150 °C in dry nitrogen) the transistor of
`the positive and negative Vgs scales are interchanged.
`
`28pF
`
`28pFy 26
`
`c
`
`24
`
`22
`-30
`
`~-20
`
`10
`qy sat
`
`05
`
`!
`!
`
`[t V
`
`,
`
`10
`
`0
`20V
`
`0
`-10
`—> i,
`a
`
`
`
`Fig. 9. Variation of the gate capacitance C with the d.c. voltage Vgs in an N-channel MOS
`transistor. a) After oxidation, the transistor was processed in wet nitrogen. From thecal-
`culated flat-band capacitance Cy it follows that the flat-band voltage Vr is —12 V. The thresh-
`old voltage Vin of this transistor is —6 V, as appears from the J sat-Vgs characteristic.
`6) Here the transistor has not undergone treatment in wet nitrogen, and consequently sur-
`face states are present at the SiQe/Si interface. The flat-band voltage Vr is —30 V, the thresh-
`old voltage Vin is +80 V.
`
`The nature of the curves may be explained in qualitative terms
`as follows. In the case of a P-type substrate a negative charge
`on the measuring electrode of the MOS capacitor would increase
`the hole concentration at the surface. Charge variations caused
`by the a.c. signal used for measurement then occur so close to
`the interface that Cs: is relatively large and C is approximately
`equal to Csio,. If the negative voltage on the electrode is allowed
`to approach zero, then these charge variations gradually occur
`less close to the interface and Csi becomes smaller. As a result
`the measured capacitanceis also lower. If Vgs is raised to positive
`values, this process continues until the threshold voltage is reached
`and inversion takes place at the surface.
`The values of C found when Vs is raised still further depend
`on the frequency at which the measurement is made. At low
`frequencies (below about 100 Hz)C increases with rising Vgs up
`
`fig. 9a was subjected to heat treatment at 450 °C for a further
`30 minutes in wet nitrogen before the electrode metal was de-
`posited.
`The oxide layer was 1.2 um thick in both transistors, and for
`the substrate conductivity of 50 Qcm the threshold voltage Vin
`should be +6 V whenall other effects are neglected. As can be
`seen in fig. 9a, the threshold voltage is —6 V, i.e. 12 V lower.
`If the sum of the oxide charge and the charge present in the sur-
`face states at the threshold voltage is put at Ni elementary charges
`e per cm, we can calculate Nt from the expression:
`
`Ne = —CoxAl Vin,
`
`where A Vin is the change in Vin caused by the positive charge,
`i.e. —12 V in the present case. We then arrive at Ny = 2x 1011
`positive charges per cm?.
`
`Page7 of13
`
`
`Page 7 of 13
`
`
`
`232
`
`PHILIPS TECHNICAL REVIEW
`
`VOLUME 31
`
`From the shape of the curves offig. 9a it can be shown that
`near the interface. It is known that electric conduction
`the positive charge in this transistor must be situated almost
`in vitreous materials, such as SiOoe,
`is often due to
`entirely inthe oxide, and that the charge in the surfacestatesis
`alkali ions. Turning again to fig. 5, we see that there is
`negligible in this case. Now let us return for a moment to fig. 8.
`a relatively large amount of sodium present in the ox-
`This shows the variation of C with Ves for the theoretical case
`in which there are no surface states and no oxide charges. In this
`ide, particularly in the top layer. If all this sodium were
`case, for a gute voltage of Vgs s= 0 the energy bands in the energy-
`to be driven towards the oxide/silicon interface when
`band diagram of the MOStransistor are not curved !2), and C
`a positive voltage was applied, this would giverise to
`has a value Cr that can becalculated from the oxide thickness
`a concentration of about 10'% positive elementary
`and the substrate doping. [f oxide charges do exist, they cause
`charges per cm? (at an oxide thickness of 0.2 um this
`band curvature at zero gate voltage, and a negative gate voltage
`Vr; is then required to removethe band curvature and obtain the
`would mean a change of100 V in the threshold voltage).
`flat-band capacitance Cy, If we calculate this for a given transistor
`Such large changes are not generally found, however,
`we can use the measured C-Vgs curve to find the magnitude of
`and also the absence ofanysignificant instability when
`V; for that transistor. The voltage Vr is a measure of the oxide
`the gate voltage is negative indicates that most of the
`charge without the charge in the surface states, becauseit is
`sodium ions present in the oxide do not take part in
`measured whenthere is as yet no question of inversion and the
`the conduction. It is probable that
`reaction between
`associated electron trapping. In fig. 9a, Vr = —12 V, whichis
`therefore the