CA2034650C

CA2034650C - Ellipsometric control of material growth

Info

Publication number: CA2034650C
Application number: CA002034650A
Authority: CA
Inventors: David Erik Aspnes; William Erik Quinn
Original assignee: Bell Communications Research Inc
Current assignee: Iconectiv LLC
Priority date: 1990-06-15
Filing date: 1991-01-21
Publication date: 1993-07-20
Anticipated expiration: 2011-01-21
Also published as: EP0533678A1; JPH05507349A; WO1991020093A1; US5091320A; EP0533678A4; CA2034650A1

Abstract

Abstract of the Disclosure A method and apparatus for controlling the growth of a multi-species film. During the film growth, an ellipsometer(40,52,54,62,64,70,72) continuously monitors the surface(24) on which the film is growing. The ellipsometer data is used to calculate(38) the effective complex dielectric constant of the thin-film/substrate structure. A sequence of such data is used in a model calculation to determine the composition of the top portion of the thin film. The measured composition is compared with the target composition and the amount supplied of one of the species is correspondingly changed(26).

Description

-1- 203~50 F;cld of thc /nl~ension The ;nvention relates generally to a method and apparatus for grOWiDg materials. In particular, the iDVention relates to an optieal techDique for controlling the thickDess and composition of thin films, such as semiconductor iilms, S grown by deposition.
Background of thc rnvention Fabrication of advanced electronic and opto-electronic integrated circuits often involves the growlh of a multi-layer structure of different compositions, for example III-V semiconductor compounds and their alloys. Sometimes the composition or 10 thic~ness of the layers is critical and must be well controlled. For instance, quantum-well semiconductor lasers may use rn~Ga~ As in a very thin active layer to emit light. Both the compositioD or alloying ratio x and the layer thickness will deterrnine the emissiOD
wavelength. The procedure of post-growth characterization of composition aDd growth rate to readjust the depositioD parameters for future depositioDs, although the standard 15 praetice iD the past (U.S. Patents 4,483,725 to Chang and 4,786,616 to Awal er al.), is becoming iDcreaSing]y unsatisfactory as the comple~ity of the structures increases (e.g., quaternary compositioDs) and the required accuracy of growth becomes more demanding.
Even once the process parameters have beeD established, uneoDtrolled, and possibly uncontrollable, variations iD the growth process may iDtroduce enough variation to put a 20 dev;ce out of specificat;on, thus reducing the device yield.
Many measurement tecbniques are amenable to in situ ebaraeter;zatioD in wbich tbe sample is measured wbile in the growth chamber at ambient pressures comparable to those during the growth process. Thereby, surface eontam;nation and ox;dation are minimized. However, if the in sir~l measurement 25 requ;res interruption of the growtb, ;t Deitber accurately characterizes an uninterrupted growth nor ;s ;t sat;sfactory for production-l;ne use.
Real-time characterization of layer depositioD, that is, in siru measurement of the growing surface during coDt;Du;ng growtb and without ;nterruption, prov;des a better uDderstaDding of tbe growth process. However, using that 30 ebaraeterizat;on for real-tlme eontrol of the growth has seldom been ;mplemented.
Aspnes et ai bave disclo6ed reai-time gro vth control in Unitcd States Patent 4,931,132 issucd June 5,1990. They relied on rcflcctance difference spectroscopy, which measures the difference in retlectcd intensities for two polarizations of iight ineident upon lhe growing 6urfaee. Tbeir teehnique effeetively eharaeterized the ~, . '' .

.

near surfaces of III V semiconductors, specifically, whether the uppermost layer was a group-III cation layer or a group-V anion layer Therefore, they suggested using their spectrometer to interrupt the supply of growth species in atomic layer epitaxy, in which a crystal is grown an atomic layer at a time Reflection high-energy electron diffraction 5 (RHEED) provides similar information about the near-surface region Both techniques could in principle be used by integrating over time to obtain thickness and composition information of a growing layer Atomic layer epitaxy is, however, considered too difficult and slow for production-line semiconductor fabrication Optical ellipsometry has long been recognized as providing accurate 10 characterizations of semiconductors and other materials, their surfaces, layer eompositions and thiclcnesses, and overlying o~cides Ellipsometry probes the surface to within an absorption length of the measuring radiation, a useful range for layer growth As will be discussed later, it can measure the real and imaginary effective dielectric eoDstaDts, which can be interpreted in terms of layer thiclcness and composition The 15 probing wavclength may bc changed in spectro-ellipsometry (SE), thus varyingthe absorption length and the dependence of the data on the composition of the material Hotticr ct al havc discloscd rcal-timc cllipsomctric charactcrization in a technical article entitlcd "Surfaec analysis during vapour phasc growth", appearing in Journal of Crystal Growth, volumc 48, 1980 at pagcs 644-654 In a rclatcd artielc entitled "Qualitative and 20 qusntitative asDessments of the growth of (aluminum, gallium) arsenide-gallium arsenide heterostruetures by Jn JI~U cllipsomctry" appcaring in Rcvuc Physlquo Appliquc, volume 16,1981 at pages 579-589, Laurenee ct al cmphasizcd that post-e~periment calculations wcrc rcquircd to dcrivc thc composition and thicl~ncss of grown layers Drevillon discussed In J/~U ellipsometry in a technical article entitled 2S "/n Jltu analysls of the growth of semieonduetor matcrials by phasc modulatedclllpsomctry from UV to IR" appearlng in Surfacc and Intcrfacc Analysls of Mlcroolcctronlc Matcrlal~ rroccJJlng and Growth, SPI~ Proeeedings, volume 1186, 1989 at pages 110 121 He eoneluded thc artielc with thc obscrvation that his worlc demonstrated tbe potential of ellipsometry for In Jltu proecss eontrol Roblllard has diseloscd thc usc of cllipsomctry for tbc eontrol of thln tllm growtb in U S Patcnt 4,434,025 Howcvcr, Robi11ard was eoneerned primarily wlth tho amorphlelty or erystallinlty of thc fllm, and thus wlth tbc growtb tcmperature Ho eould thus dcrlvc sutflelcnt eontrol Informatlon from a slnglc sct of cllipsometrie data takoo tor all analyzcr anglcs Hc also mcasurcd thc film tbiekncss by comparing rclative 3S Intcnsltles ot two peaks In thc sct, whleh pcalcs had bccn cquallzcd in a ealibratlon step prlor to thlekness mcasurcmcnt Roblllard's tcehnlque however lael~s eontrol ovcr the tllm eomposltlon Sl mmary of thc Invcntton Accordingly, an object of the invention is to provide real-time control of layer depositiom Another object is to provide composition and thickness control for 5 the growth of semiconductor structures having multiple layers of different compositioDs.
A furthes object is to provide such control with the accuracy available with ellipsometry.
The invention can be summarized as the method and apparatus for the real-time control of material growth by ellipsometry. While a film is being grown on 10 a substrate of different composition, aD ellipsometer measures complemeDtary ellipsometric data of the filmlsubstrate combination. A series of such data are used in one of several models to appro~imate the dielectric coDstaDts, and thus the composition and thiclcness, of the thin film having a thickness less than the optical thickDess of the dlip~ometer radiatiom For thicknesses greater than the optical thickness, only 15 composition is determined. Control circuitry compares the measured composition to the target compositioD and dynamically corrects the growth parameters for the film growth.
Brlcf Dc~crlptlon of thc Drawlng FIG. 1 is an illustration useful in understanding ellipsometry.
E~IG. 2 is a block diagram of the ellipsometric growth control 20 apparatus of the prcseDt iDventiom PIG. 3 is another illustratioD useful iD understanding ellipsometry.
E7IG. 4 i~ an illustratioD of typical raw data recorded by an ellipsometer.
FIG. S is a logic flow cbart for the ellipsometer measurement cycle.
~G. 6 is a logic flow chart for setting the photomultiplier tube voltago.
FIG. 7 is an illustratior. of the spectral energy dependence of the pseudo-dielectrlc function for two different materlals.
PIG. 8 i~ an illu~tration of the locu- of the pseudo-dielectric 30 tUDCtiOn durlng growth ot a multi-layer structure.
I7IG. 9 ls a logl¢ nOw chart for the control ol growth.
FIG. 10 i~ an IllustratioD of the locl of the pseudo-dielectrlc functioD during controlled and non-controlled growth of a thin film.
Dctallcd Dc~crlptlon The inventioD utlllzes elllpsometrlc characterlzation of a growing lilm tor the real-time coDtrol of the growth proce~s. Bctore fully dcscrlbing the iD~eDtion, tho priDclples ot clllpsometry wlll bc descrlbcd, follo ved by the dcscrlption of . .

some preliminary experiments, all useful for understanding the invention.
Ell)psom~try In PIG. 1 is illustrated some of the formalism required in discussing ellipsometry. A sample to be tested has a specular surface 10. Light falls on the surface 5 10 along an incident direction 12 at an angle ~ with respect to a normal 14 of the surface 10. The iwident light is specularly refiected from tbe surface 10 in a reflected direction 16, also at the angle ~ to the normal 14. The transmitted component is here ignored.
The light is assumed to be monochromatic. A plane of incidence is defined to include the incident and refiected directions 12 and 16, and thus to be perpendicular to the surface 10 10.
The incident light has two polarization directions characterized by comples field amplitudes E, for the electrical polarization parallel to the plane of the surface 10 (that is, perpendicular to the plane of incidence) and Ep for the electrical polarization perpendicular to EJ (that is, lying in the plane of incidence). Thus, four 15 quaDtities in the two comples amplitudes define the incident light. However, one is the instantaneous phase of the light wave and is thus generally unmeasurable and unimportant, and another is related to the total intensity of the light and is important only for the esperimental implementation of the ellipsometer. The polarization charaeterlstics of the incident light are determined by only two parameters, which are 20 eonveniently written as the intenslty-independent, comples ratio X = Ep/E, of the eomples field amplitudes, where X is referred to as the polarization state.
Similarly, the renected compla~ smplitudes are E,' perpendicular to the plane of incidence and Ep ' parallel to the plane of incidence. The comple~crenectance coefficients are then defined by 25rp = Ep'/Ep, (1) and r, = E,'/E,. (2) The eomples refleetanee ratio p ls defined as p = rpfi" (3) 30 and is often represented by the real angles ~ and 1~ defined as p - tan~ el~. ~4) In an elllpsometrle e~periment, tbe state of polarization X of the ineldent beam Is established by, tor esample, passing the beam through a linear polsrlzor. The boam is then refleeted at non-normal ineidenee from the surface, which 3S ehanga the polarlzstion ~tate from X to X'- The value Of X' ls then determined ezperlmentally by methods that wlll be deserlbed below. The ratio X'/X ls then s 203465~

calculated, leading to X = Ep Es = rp = p. (S~
X EPEJ' r5 Although ellipsometric data are usually presented in terms of the S complementary data ~ and 1~ or tan~ and COSI~, it is eonvenient for material characterization to define a comples pseudo-dielectric function having complementary real and imaginary parts <~> = <~1> + i<e2> (6) for the model of a thick fllm adjaccnt a vacuum That is, the surface 10 is assumed to be 10 the upper surface of a semi-infinite, homogeneous material having a complex dielectric eon~tant <e> and the area above the surface 10 has unity dielectric constant Thepseudo-dielectric constant and the comple~c reflectance ratio are related by e~> = sin2¢~+sin2~tan2~¦1 Pl, (7) whlch 1- used to relate the measured p to the properties of the sample through the model lS csleulatlon If the thiclcness of a homogeneously grown layer approaches the ab~orptlon Icngth of the light, ~> approachcs thc eomplcs dielectric constant ~ of the matcrlal of that laycr Ellipsomctry thu~ morc casily relatcs to material properties than doe~ reneetance differcnee spcetroseopy, whieh measurcs 1~ 12 - 1~ l2 (8) An cmbodimcnt of thc prcscnt invcntion is illustrstcd in the block dlagram of ~IG. 2 and Includcs a thin film growth systcm, a cllipsomctcr attached to the growth ~y~tcm, and a eomputcr-based eontrollcr receiving data from the ellipsometer and eontrolling thc opcratlon of thc growtb sy~tcm 2S Thln Fllm Growth SyJt~m In a demon~tratlon of tho Inventlon, the growth system was a Model V80H ~y~tem eommerelally avsllable from V(~ In~truments, Inc of Danvers, Ma~-aehu~ett~ Thl~ ~y~tem allow~ growth by organo metalllc molceulsr bcam epitaxy tOMMB~), whieh combine~ moleeular beam cpitasy wlth ehemlcal vapor dcposition As30 Illu-tratcd in PIG 2, a vaeuum ehsmber 201~ pumped by a dif(uslon pump 22 to a base prc~urc of lO-Il Torr A sub~trate 241s thermally bondcd to an unillustrstcd hcated ~ub~trate holder For growth of the AlGaA~ famlly of materials, gaseous .' ,, ,'. , ,' ' " . ' ' ' ' "' "
, . ' ~ ''',,''"' .; . ' ' ' ~, ' , ,, ... ,." ~ . . .. . : . :
. .

- 6 - 20346s0 triethylaluminum (Al (C2H5)3), triethyl gallium (Ga (CzH5)3), and arsine (AsN3) are supplied through respective metering valves 26, 28, and 30 to ports 32 and 34 and a cracking unit 36 The cracking UDit 36 cracks the AsH3 into As2 and H2 Thereby, controlled beams of Al (C2Hs~3, Ga (C2Hs)3, and As2 are irradiated upon the substrate S 24 The alkyls Al(C2Hs)3 and Ga(C2Hs)3 crack on the substrate surface to provide Al and Ga. For the AlGaAs system, an overpressure of As is maintained so that the growth is controlled by the amouDts of Ga and Al.
The OMMBE growth system is controlled by a computer system 38, with lceyboard and display The purchased growth system included a first PDP 1153 10 computer and interfaces for the growth system A second computer, a Digital Equipment Corporation computer, model 1173, was added for the ellipsometer The Ga and As metering valves are controlled by the first computer However, the illustrated control liDe to the Al metering valve 26 was reconnected to the second computer for purposes of feedbaclc growth control EllipJomctcr Op~cs As discussed above, an ellipsometer establishes light in a definite state of polarization and analyzes the state of polarization that results after the light has been reflected at non-normal incidence from a specular surface Many different eonfigurations have been devised to perform these functions, as summarized in, e g, the 20 boolc Elllp~omc~ry and Polarlzcd Ll8ht by R M A Azzam and N M Bashara (North-Holland, Amsterdam, 1977) The ellipsometer part of PIG 2 is a rotating-polarizervariant of the rotating-analyzer ellipsometer previously described by D E Aspnes e~ al.
ID a tcehoieal artiele entitled "High preeision seanning ellipsometer," appearing in Applied Optlc~, volume 14, 1975 at pages 220-228 and further elaborated bg them in anotber 25 tocbnicai artlcle entitled "Methods for drift stabilization and photomultiplier linearization for photometrie cllipsometers and polarimeters," appearing in Rcvlcw of Sclen~lf~c InJtrl~mcntJ, ~lolume 49, 197B, at page~ 291-297.
In PIG 2, a light source 40 is a 75 W short are Xe lamp contained iD an unillu~trahd howing Thc light bcam trom thc lamp 40 is focused by a concave 30 mtrror 42 onto an iris 44, wbieh is tbc tbird of four iriscs 46, 48, 44, and 50 spatially dcfiDing tbo ineldcnt bcam 12 and tbc bcam 16 reflcctcd off tbc substratc 24 Tbe first t~vo iriro~ 46 and 48 arc in thc ineidcnt bcam 12; tbc last two iriseo 44 and 50 arc in the rcnoetcd bcam 16 Thc lamp bousing and othcr ~hielding surrounding tbe incidcr~t beam 12 minimlzo tbormal~ (air eonvcetion currcnts tbat cauu tbc lamp imagc to ~himmcr and 3S thcrcforc introduec unwantcd noi~c) A rotatable quartz Rochon prism 52 acts as the polarizer and splits the incident beam 12 into aD undeviated primary beam arld an orthogonally polarized 2 defiected secondary beam, both beams being nearly linearly polarized The unwanted secondary beam is blocked by the second iris 48 The polarizer 52 is carried in a hollow 5 shaft 54 of a motos assembly 55 that rotates the prism 52 at a mechanical rotational frequency of 58 Hz A shutter 56 is closed for part of a measurement cycle to establish a zero baseline for the detection electronics to be described below and also to establish stray light leve1s The incident beam 12 enters the vacuum chamber 20 through a 10 window 58, strikes the substrate 24, and is reflected therefrom as the reflected beam 16, which passes out of the vacuum chamber 20 through another window 60 Both the windows 58 and 60 are fabricated according to the plans disclosed by Studna et al. in a technical artic1e entitled "Low-retardance fused-quartz window for real-time optical applications in ultrahigh vacuum," appesring in Jol~rnal of Vacuum Science and 15 Tcchnology A, ~olume 7, 1989 at pages 3291-3294 Thus, they introduce minimum optical distortion aDd can be locally baked to remove arsenic Both the iDcident and refiecte~ beams 12 and 16 form an angle ~ of about 70 with the normai of the substrate A fiecond quartz Rochon prism 62 acts as the reflected beam 20 analyzer and splits the reflected beam 16 into orthogonal nearly linearly polarized compoDents in the same manner as the polarizer 52 The unwanted secondary beam isbloelced by the fourth iri~ 50 The analyzer 62 is mounted in a computer-controlled Inde~ing head 64 to precisely control its azimuth angle A with respect to the plane of iDCideDCC OD the ~ubstrate 24, both fot signal aDalysis and to establish the precise 25 azimuth of the plane of ineidence as will be described below A focusing mirror 66 and a foldiDg mirror 68 direct the primary beam into the entrance sllt of a monochromator 70, which di~perse~ the wavelengths and establishes which of the wavelengths shall emerge rom tbe e~lt slit aDd be detected by a photomultiplier 72 For wavelengths longer than about 370 Dm, an optleal fllter 74 is plaeed into the primary refiected beam 16 to block 30 ~ecood order light from passing through the monoehromator 70 Elllp~omc~cr Elcctronlc~
The elllp~ometer of PIG 2 is automatieally eontrolled by the ~econd eomputer In the eomputer system 38. The eleetronies interfacing the computer to the ellipsometer will DOW be diseussed An optieal eDeoder 1~ attached to the shaft 54 containing the polarlzer 52 snd provide~ two pulsed ~Ignal~ to the eomputer system 38 The first signal l~ a pulse that repeats at 120 equal intervals over a single rotation of the shaft 54, that is, ,, , . : . :

, ' '. . , ', :~
,:

at a frequency of 6960 Hz At each of these pulses an analog-to-digital converter card input of the computer system 38 reads the output voltage of the amplifier of thephotomultiplier 72 and the computer system 38 stores the intensity data for further digital processing The second signal from the optical encoder on the shaft 54 is a level 5 shift that occurs at the first and sixty-first pulses of the first signal and is used to synchronize the Fourier analysis procedure to be discussed below to the azimuth angle of the shaft 54 The first signal pulse is received by the "read input of the AID input card in the computer system 38 The level shift of the second signal is received by one input bit of a 16-bit binary input/output card in the computer system 38 The motor 55 driving the shaft 54 is a hysteresis synchronous capacitancc motor driven by a power amplifier incorporating a precision oscillator for establishing the rotation frequency of the polarizer 52 The lamp 40 is powered by a dedicated DC power supply including a pulse source to initiste thc arc The shuthr 56 is opcrated by a driver and is opened and closed according to one output bit of the 16 bit inputloutput card The computer system 38 further provides control signals in the form of pulses from the 16-bit input/output card to stepper motors driving the indexing head 64 and the wavelength selector of the monochromator 70 Each stepper motor 20 receives two binary lines for bidirectional rotation The output current of the photomultiplier 72 is direcdy proportional to the instantancous detected intensity of the reflected beam 16 at the selected wavelength several of which data are used to determine the polarization state of the reflected light The photomultiplier current is converted to a voltage by an amplifier 25 This smplifler provides low-pass filtering with a time constant of appro%imately three triggering cycles (500 p s) to eliminate high-frequency noise The amplifier also receives aDd mi~es with the photomultiplier current a small triangular current wavcform and a DC
ot(set current The triangular current waveform is generated by clipping and integrating a small traction of the AC output of the power supply for the motor 55 of the shaft 54 30 The trlangular waveform averages out inaccuracies in the step increments of the AID
converter card The DC offset current is used to ensure that the signal supplied to the A/D converter Is positive so as to tall wlthin ih 0-10 V range The output of thephotomultiplier amplifier i- supplied to the ndatan input ot the A/D input card in the computer system 38 3S Thc gain of thc photomultiplier 72 is controllcd by the computer systom 38 through tho photomultiplicr power supply which proportionately controls the voltago supplled to the photomultiplier tube 72 A D/A (dlgital to analog) output of the 20~4650 computer system 38 supplies an analog voltage directly to the power supply to control its output voltsge.
Another D/A output of the second computer in the computer system 38 has been adapted to control the meterinB valve 26.
The method of determining the polarization state of the reflected radiation will be esplained with reference to FIG. 3 and using the example of anellipsometer with a stationary polarizer and rotating analyzer, which is fully equivalent to the ellipsometer of ~71G. 2 having a rotating polarizer 52 and stationary analyzer 62. If an incident beam is assumed linearly electrically polarized in in some direction, usually 10 30 with respe.,t to the plane of incidence but here illustrated vertically, upon reflection from a specular surface it becomes in general elliptically polarized, as indicated schematically by an cllip9c 80. That is, the instantaneous reflected electric vector E has a direction and magnitude tracing the ellipse 80 at the optical frequency. The ellipse is characterized by a major asis a and minor axis b and by an azimuth angle ~ of the major 15 asis a. Because absolute intensity is not material in ellipsometry, the polarization state is determined by the azimuth angle ~ and the minorlmajor asis ratio b/a. As a polarization analyzer passes only that component of the polarization ellipse 80 along the analyzer's priDcipal asis, a rotating analyzer (or polarizer) can therefore be used to sample the ellipse B0 at various points around it, thereby establishing its geometric properties.
20 Speclflcally, the photomultiplier 72 measures an intensity proportional to IE 12 at a point on tbe elllpse 80 selected by the angle P of the rotating analyzer (polarizer). The elllpsometer therefore measures an angularly dependent intensity, such as that plotted in PIG. 4. Such a sinusoidally varying intensity I is fit within the accuracy of the mea~urement to the eguation 1 = cO + clco8(2P) + c2~in(2P). (9) Ellipsometric data are measured over the full 27r range of angle P and the data are Pourier tram~formed to yield the DC, sine, and cosine coefficients cO, cl, and c2, which are related to the ~ and ~ representation of the comp1es reflectance ratio p of Equation (4) by 30tan~ I ~ tan~ (10) Co Cl md co~

whcre ~ Ir the anglo ot the priwlpal asls of the tised aDalyzer wlth respect to the plane ot incidenco ot FIG. 1.

, ., ~ , ,,'.~.', ,'.

' .

. lo- Z034650 The data acquisition for the ellipsometer is controlled by a program written in Fortran and Macro languages which is stored in and executed by the second computer in the computer system 38 The fiow diagram for the data acquisition subroutine is shown in FIG 5 In the first step 86, the shutter 56 is closed in order to S measurc the background contribution Step 88 accumulates a number NR0 of data cycles of intensity data from the photomultiplier 72 Each data cycle corresponds to half a revo1ution of the polarizer 52 or to 60 intensity measurements The beginning of step 88 is synchronized with thc level shift from the rotary encoder on the shaft 54, to which the azimuth angle of the analyzer 62 was fi~ed by the head 64 in a precalibration step Each 10 of the 60xNR~ intensity data is synchronized by the 6960 Hz pulse train from the rotary cncodcr After thc accumulation of the data, they are checl~ed to see that no data have becn droppcd In the rare case of dropped data, the step 88 is repeated The data sccumulated in step 88 are Fourier analyzed in step 90 to yield the DC, cosine and sine baclcground coefficients, here represented by VDC0, VCOS0, and VSIN0, which are then 15 ~tored See the Appl~cd Optlc5 artic1e of Aspnes et a1 for the Fourier analysis procedure ID stcp 92, thc shuffcr 56 is opcned for the measurement In step 94, a numbcr NR of data cyclc- of intcnsity data are accumulated The timings in step 94 fo11Ow the ~ame timing~ as in thc baclcground stcp 88 The values of NR0 and NR are cho~en by the operator and are typically 10 aDd 100, respcctivcly, reprcscnting 20 re~pectlvely 5 and 50 mechanical rotations of thc polarizer 52 In step 96, Fourier analy~l~ produco~ the DC, cosinc and sinc mcasurcd cocfficient~, VDCI, VCOSl, and VSINI. ID ~tcp 98, the bacl~ground coemcient~ arc ~ubtracted from the measured coeUicicnh to yicld thc correctcd cocfficient~ VDC, VCOS, and VSIN, corresponding to co~ cl , nd c2 The control logic for thc subroutinc that sets the voltagc of the photomultipllcr tubc 72 is shown in 171G 6 Upon eDtry, a repeat counter is reset in step 102 ID the tir~t ~tcp 104 of thc loop, thc rcpcat countcr 1~ incremented by one In step 106, tho olllp~ometric data VDC, VC09, and VSIN aro talcen This step 106 corresponds to ~hp~ 92, 94 and 96 In PIG 5, although typically only 4 data cycles are accumulated 30 (NR ~ 4) In to~t 108, it 1- detormlnod whothor thc DC cocfficicnt VDC is within prodetermined limlt~, tor e%amplc, of (4 7 ' 0.05) V ~o a~ to cnsure that cvcry part of the pbolomultlpller amplitlor output voltago wlll lle wlthln tho 0-10 V range of the A/D
coDvortor tor any po~lble valuo ot p It tbo voltago 1~ ~vlthin thc llmlts, thc subroutinc is o%ltod; It not, In h~t 110 It 1- dchrmlncd whcthcr thc prc~ct llmit of thc rcpcat counter 35 ha~ boon oxceodod It w, In ~top 112 an orror mouago 1~ prlnted and the oubroutiDc is o%ltod; It not, coDtrol ~tay~ wlthln tho loop In ~tep 114, a new control voltage Is calculatcd tor thc powcr ~upply of thc photomultlplicr tubc This calculation is based on . . . : -2~)346SO
. 11 -the measured VDC and the change required to bring it to 4 7 V In step 116, the new eontrol voltage is issued to the photomultiplier tube power supply through the D/A
converter output of ehe second computer of the computer system 38 Thereafter, the loop repeats S Preliminary Expcriments Before the remainder of the apparatus of the invention is described, two preliminary esperiments will DOW be discussed Much technological interest centers on GaAs and its ailoys with AlAs. Therefore, we used the spectrometer and growth chamber of FIG 2 to measureI0 the spectral dependences of the pseudo-dielectric function <e> of GaAs and Alo 36Ga0 64As No growth control was performed in these preliminary experiments Because there measurements were performed upon optically thick films, <e> ~ ~ The speetra1 dependonces of ~el> and <e2> illustrated in FIG 7 show that 2 6 eV is afavorable energy at whieh to monitor the growth of the alloys of AIxGal_xAs because of 15 the strong dependence of <e> on the alloying percentage x arising from the El and El+~l transitions at this energy The choice of 2 6 eV allows a measured value of e2 for an optleally thielc film to be eonverted to a eomposition x, as will be described below A1though data for only one temperature are shown, the dielectric function at this energy is also dependent, although less strongly, on growth temperature Accordingly, the spectrometer was set to an energy of 2 6 eV for real-time measurements of the pseudo dielectrie function <e> in another preliminary e~periment A GaA~ substrate was loaded into the OMMBE~ station Layers of AIxGal_xA~ were sequentially grown at 600C with x = 0 07, 0 19 and 0 36 Each of the layers were growD to a thicltneos of about 275 nm thielc The pseudo-dieleetric fuDction 25 was measured duriDg the deposition When the real and imaginary parts <el> and~e2~ of the pseudo dieleetrie funetion are plotted against each other, there results the series o~ spirals of ~7IG 8 The spirals begin at the bullc value of e of the optically thick substrate layer and converge to a different value eorresponding to the bullc value of e of the optlcally thiek grown ~ilm Similar loei are obtained when the ellipsometric data are 30 ropresented by ~ and 1~, as has been reported in the previously eited artieles by Hottier et al and Lsurenee et al ~ int, it is observed that the end-point values, that i~, e, can be approslmated by an analytle e~pre-sion ~ - ~, + (9 22x - 24 14x2) + I ( - 16 94x - 0 33x2), (14) 35 where x 1~ the alloying fraetlon and e, ls the 2 6 eV dieleetric function of the GaAJ
substrate This equation Is represented by the dashed llne in FIG 8 Sueh an equation CaD be usod to relate the measured value ~ to the eompositioD x The drift, resembling a , ~, ' pig tail, in x by 0 02 at the end of the second spiral was caused by a swelling of the seal in the triethylaluminum valve 26 The spiral loci of <e> can be e~plained in terms of the Fresnel reflectance e~cpressions of the so-called three-phase (substrate 1, overlayer 2, ambient 3) 5 model where the boundaries between the substrate, overlayer, and ambient phases are mathcmatically sharp If rl2 and r23 are the reflectances at the substrate-overlayer and overlayer-ambient interfaces, respectively, then as described, e g, by Azzam andBashara, the reflectance rl3 of the three-phase system is given by Z rl2 rl3 + 1 (15) 10 wherc z = c2i~d (16) and c~ = (eO- sin2~)1n~ ~17) and where ~ is the angle of incidence, ~ is the wave vector of the probing light, ~0 is the 15 complac dlc1ectric constant for the bullc ovcrlayer material, c is the speed of light, and i~ thc frcqucDcy of thc probing light I~quations (15) through (17) are valid for both s aDd p polarized light as long as thc appropriate e~prcssions for rJ and rp are substituted for r/~, rc~pcctlvcly, scc Azzam and Bashara Spccifically, r, and rp are defined in terms of thc dlclcctrlc functlons e" eO, and ea = 1 of thc substratc, overlayer, and ambient, 20 rc~pcctlvcly If the dlelectrlc functions of thc substrate and ovcrlaycr arc nearly thc ~amc, as Is thc casc hcrc, scc Flg 7, thcn rl2 ~ 0 and Equation (15) can bc expanded to tlr~t ordcr in rlz Thc rc~ultlng c~prc~sion again to first order in rl2 Is rl3 = r23 + (r12--r23)Z- (18) 25 It the ovcrlaycr 1~ nona~istcnt (d = 0, Z = 1), then rl3 = rl2, while if the overlayer is optlcally th{ck (d # ~, Z ~ 0) thcn rl3 = r23- BCtWCCD thcsc extreme values, rl3follow~ an exponcntlsl ~plral wlth the perlod and ratc of convergence dctcrmlned by the rcal and Imaglnary parts of ~ as dcflncd In Eq (17) A~ thl~ functlonal torm appllcs to both r, and rp, It also applles to 30 thc ratlo p - rp/r, and tbcrcfore al~o to <c> ~or larger r~2 such that thc first-order e~pan~ion 1~ not a good approxlmatlon, hlghcr-ordcr corrcctlon tcrms distort the simple e~ponentlal ~plral form but thc overall ~hapc rcmaln~ qualltatlvely the samc, Control Thcory .. ., : . ...
., The pseudo-diolectric function or other ellipsometric forms of data can be advantageously used to monitor both the composition eO(d~ and thickness d of growing layers. The use of the pseudo-dielectric function for control presents some difficulties however. Neither the pseudo-dielectric function <e> nor the complex5 reflectance ratio from which it is obtained provide direct information about the dielectric function eO of the layer being deposited. This information must be derived from these data.
Several methods of achie~ring this goal can be used, such as directly fitting the e~act Presnel equations to the data without approximation using eO and d as 10 adjustable parameters, or calculating eO itself by a suitable construction. The former method has the ad~rantage of generality but requires substantial computing power. In a demonstratioD of the invention, we used a second method based on the appro%imation of Equation (18), which is applicable to the G~AsMl,~Gal_~As system and is much simpler to implement.
In terms of the pseudo-dielectric function, we write Equation ~18) as <-(d)> = eO + (e,--eO)cl2~d. (19) Let us consider the two values <e(d+~d/2)> and <e-~d/2)>, where ~d is a thickness increment that we may select corresponding, e.g., to the difference between two or more 20 successive olllpsometer roadings. If ~ dl << 1, then we have appro~imately <e(d)> = <e(d + I~d /2~> + <e(d--~d /2~>

~ eO + (,-J - eO) C 2 1 1~ d (20) and /~<e(d)~ = ~(d + Ad /2)> - <e(d--l~d /2)>
~ 2 ~ d ~ 0) c21~d . (21) The ~/arlable d indleate~ an average over a range l~d eentered about d.
We ean now eombine Equations (20) and (21) to eliminate the taetor ~,- eO)C21~d, and solve the result for the desired overlayer dlelectrie function eO.
'rhe result ls eO _ < ~(d) ~ ~ 2~ d (22) Beeau~e ~e(d)~ and ~<e(d)~ are measured values, we ean obtain the desired qUaDtity ~0 trom measurement it v~e malte appropriate e~timate~ tor Ad and ~. In praetiee, 10%
aeeuraey i~ sutfielent tor the~e e~timate~.

..
... .
. ' ' .. ', ...... . .
.

.

Z0346~;0 The e~cpression is actually much more general than the Fresnel derivation would indicate because the substrate-overlayer reflectance rl2 is really the ratio of the refiected to incident electric fields at the boundary 1--2 As this ratio is definable even in the absence of a boundary, we can suppose the existence of a running S virtual boundary Np~,-8d deep, where NPD is the number of points that we can average to obtain better estimates of <e(d)> and /~<e(d)> and ~d is the thickness increment of the film per point Thus, Equation (22) yields the dielectric response of the outermost region of the overlayer of thicl~ne99 NpU ~d = ~d-Aecording to Equation (14), the imaginary part of the dielectric 10 eonstant Im{~O(d)} varies almost linearly with x A target imaginary dielectric constante"2 can be derived from Equation (14) for a target alloying percentage x, Then, the difference in imaginary parts is proportional to the difference in alloying percentages ~x = x(d)--x" that is, Im{eO(d)}-e,,2 = al /~x (23) 15 where aI = 16 94 from Equation (14) Combining Equations (22) and (23) produces the eontrol equation al ~x = Im{ <~(d)>- 2~ d [<~(d+~d)>-<e(d)>l}-el~2~ (24) The only remaining undetermined quantity in Equation (24) is ~ d.
Thl~ quantiq ean be ealibrated by separate measurement or ealculated from flow-rate 20 data to the ~uMcient aeeuracy of ~0%
With all psrsmeten thus determined, Equstion (24) esn be used as the eootrol equatioo lo whieh ~x i~ redueed to zero That is, the sign aod magnitude of l~x aro u~ed to feed baclt eootrol ~igoal~ to the growth ~y~tem such that the magnitude of I~X ir redueed Ooe difflculty in sehieving fioe eootrol with the pre~ent Implementation i~ that the ~e(d)> ~ignal i~ relatively noi~y on the time scale of ~d In order to remove the noise and avoid in~tabilitie~, eO(d) i~ obtained from a running average ot Np" of p~eudo dieleetrie data Io partieular, we least -quares fit Np" equally ~paeed polnt~ Iying on a ~tralght line to the Is~t Np" poiot~ of dats plotted in the complex 30 ~e~ plane Then <e(d)> and (~e(d+/~d)>-~e(d)>)/~d are equal to the midpolnt ~nd ~lope, re~peetively, ot thl~ Iine Furthermore, to avold lo~tsbilitie~ io the ~ystem aod to reduce the etteet ot ool~e, the valve voltsge Vv i~ eorreeted ooly ioerementslly for a given value of Thu~, for every point tbe now vslvo voltsgo Vv' i- rolated to tho old valvo voltago Vv :, , ~ . , by Vv~ = Vv+CI-~x t2~) where the control constant Cl is chosen to increment Vv by a selected fraction of the total error on each point, typically 1-2% The sign of Cl is chosen to reduce ~x to zero The S size of C~ for immediate 100% correction is an experimentally known parameter of the growth system and the material family Thus, compositional errors are incrementally reduced to zero over a period of time determined by the operator The above procedure allows other forms of averaging to be used in the system, e g, on <~(d)> itself, as long as thc time constants of the other forms of 10 averaging are short compared to that of Equation (25~ We have found it to be advantageous to aqerage fi(d)> over Np,~ points to better follow local fluctuations in composition, as long as the correction fraction Cl ~x < 1/Np,J
Other types of control may be used with different combinations of proportional, integral, der;vative and time-filtered feedback, for example Growth Control Log~c The control of the growth was performed by a program stored and e~ccuted in the second computer of the computer system 38 The logic flow chart for the control program is shown in Fig 9 This program was written for an acample of theIDventloD In whlch ll lGaA~ i8 grown on a GaAs substrate Only the Al source is 20 dynamically controlled by the ellipsometer and the second computer The Ga and As sources are controlled by the first computer of thc computer system 3~ and are turned on to grow additlonal GaAt before the illustrated program is e~ecuted Thc procedurc bcgiDs in step 122 by loadiDg the control parameters trom thc l~eyboard of the computer ~y~tem 38 The parameters include ~-,2. the target 25 Imaglnary part of the dielectric function calculated from Equation (14) for a given target ailoylDg perccDtage x and the estimatcd ~tartlDg voltage setting Vv or VVO for the trlethylalumlnum valve 26 callbrated for x. The target dielectric constant ~,, 2 is calculated ~rom Equatlon (14) based on x However, the triethylaluminum valve 26 is Inltially closed with an is~ued value of VVO = O V In step 124, the monochromator is 30 run to the measùremeDt energy, here selected to be 2 6 eV
Thc loop begins by tran~errlng control in step 126 to the photomultlpller ~oltago settlng subroutlne, dewrlbed In ~IG 6, for checlclng thephotomultlpller tube voltage a~d ~or adju~tlng It l~ Decc~sary Intenslty data are then acqulrod in step 128 wltb a call to the proccdure o~ 17IG S In step 130, the pseudo 3S dlelectrlc con-taot- ~c~ 2~ are calculated accordlng to Equations (9), (10), (11), (4), and (7). ID step 132, the calculated pseudo dielectrlc constaDt- value~ are stored and dlsplayed, preterably plotted agalDst each other on an ~ y plottcr Thls completes tho . .
, : ,," ~' . ,,, :, .

.

;~03~650 measurement part of the control cycle The program continues with the evaluation of the correction to the control ~loltage VV0 In order to establish a baseline, the illustrated program allows by the operation of test 134 for Nl points to be taken before Al GaAs growth begins A
S typical value of Nl is 10 Thc feedbaclc coDtrol, to be described later, takes an average of Np~s data points in order to determine smoother va1ues of eO In FIG 9, Np,~ is represented by N2 Therefore, as determined in test 136, the ne~ct (Npl3- 1) points of pseudo-dielectric data are stored in locations 2 to Np~s in a data buffer i~ step 138 A typical 10 vaiue of Npt~ is 31 In step 140, the current value voltage VV0 is output to the valve 26 The first c~ecution of step 140 begins the growth of thc AlGaAs film Prior to the accumulation of Np~s data points, thc cstimated startiDg value of VV0 is output oDce (Np"- 1) poiDts of pseudo-dielectric data have been loaded {nto thc buffer, control of branch 136 passes to step 142 In step 142, the buffer is down 15 shiftcd by onc to accommodate the most current data which in step 144 are loaded into thc last buffer location The oldcst data point is eliminated in the process In stcp 146, the most recent Npl" data values are used to calculate the imaglnary diclectrlc eonstant e2 O(d) of thc film according to Equation (22) and the Icast-squares procedure described abovc Thc ratio of thc averagc spacing between the 20 indivldual Np" points of thc rcal and imaginary parts of ~e>, mu1tiplied by thc indicated constants, providc thc dcrivativc portion of Equation (22). The dielectric constant portlon of the eonotant ~, defincd In Equatlon (17), is appro~imated by either the most recent valuc of eO(d ) or thc targct valuc e, Thc fit is intcrpolated to thc avcragc value of d tor thc Np" poinb to provlde <e(d )> . Thcreby, eO(d) i~ calculated according to 2S Equatlon (22) from the last Np" data poinb In step 148, the corrected valve voltage Vv' or VV0 is calculated Irom Equations (24) and (25) The corrccted valve control voltagc VV0 is then issued to the valve controllor 26 in ~tep 140 via the D/A output of the second computer and the cycle repeat~. The period ol the cycle i~ about 1 seeond While no growth termination was esplicitly provided in the e~ample program of PIG 9, growth could be terminated based on the standard mass flow ealibration A procedure i~ de~cribed below to use the invcntion to additionally measure the tblcl~ne~ ot a deposited lilm, which vould be u~ed as the termination condition An e~ample ot the effectiveness ot the present control schemc for 3S malDtalDlng compo~itioD ot a thlclt AI~Cat~ J ~amplc ovcr an c~tcndcd pcriod of time is Illu~trated In the p-eudo dleleetrle loei ~hown In FIG 10 ~n Al0 20Ga0 80A~ film vas grown on a GaA~ ~ub~trate wlth thc OMMBE gro-vth systcm and cllipsomchr of FIG 2 ~ ' '','` ' ~ ,, ,: ' , : ' -The pseudo-dielectric functions were recorded during growth First, the growth was controlled according to the program of FIG 9 The locus of the resulting pseudo-dielectric constaDts is represented by the solid line (some of the noise has been smoothed out of FIG 10) The spiral locus indicates uniform composition of the film grown under S feedbaclc control Immediately thereafter, in a comparison osperiment another AlO ~OGaO 80As film was grown on a fresh GaAs substrate In the comparison esperiment, the valve control signal VV0 was fised at the value established at the end of the controlled esperiment That is, the second film was grown with no real-time feedbacl~ control The resulting locus is shown by the dashed line in FIG 10 The 10 estreme variations in the comparison trajectory shows that the OMMBE growth system was not performing adequately Upon further data review, the valve control voltage VV0 was found to be undergoing large escursions during the controlled experiment It therefore appears that even during the controlled esperiment the OMMBE system was malfunctioning but the feedbaclc control nonetheless produced a uniform film of the 15 desired composition Othcr Embodimcnts The invention can be applied to a production environment in which a pseudo-dielectric locus has been established as the standard for a production run of multi-layer structures There is no requirement that any layer be grown optically thic~
20 Any deviation from the standard loeus results in the eontrolled ehange of the metering devlee The dlreetlon of deviation determines the direetion of ehange, whieh reflects the new focal polnt of the spiral determined by the new composition For instance, in the ~plral trajeetory of AlGaAJ on GaAs, an outward, or leftward as illustrated in FIG 10, devlatlon In the upper Initlal part of the spiral indieates eseess Al while an inward or 25 rightward devlation indieates escess Ga Such a standard locus could also be used to control layer thiclcDes~ by termioatlng growth or changing stoichiometry when a specific polnt on the standard locus has been reaehed The invention ean use the full three-phase model of Equation (15) or models Ineorporatlng addltlonal phases Purther, these equations can be integrated 30 over esperlmentally derived values of ~d to provide the thiclcness termination eondition The Inventlon is not limited to OMMBE growth of semieonductor thln fllms Although the Inventlon 1- partleularly advantageous for semiconductorbotero~trueturer In wbleb tbere are small varlations ot a eomples dieleetrie eonstnnt, the loventlon ean be applled to non semleonduetor fllms a~ Iong as the film and the substrate 35 havo dlttoront dloleetrle eonstants The Inventlon ean also be u~ed for dlfferent growth metbods, tor esample, vapor ~ourcos using mas~ flow or pressure eontrol, e-beam evaporator~ and thermal evnporators Por effeetive dynamie eontrol, the amount of ~, '' ~ ....
.;, .. ;: - : ,,, material emitted by one of the sources should be changeable on a short time scale. The invention can advantageously be used in a relatively high-pressure environment, such as chemical vapor deposition, that precludes the use of electron spectroscopy. Although the embodiment discussed here uses only a single wavelength, one can envision the use of S multiple wavelengths selected to optimize independently sensitivities to thickness and compositions.
The invention thus provides for the precise control of material composition and thickness in the growth of thin films. The ellipsometric equipment is relatively ine~pensive and imposes few constraints on the growth procedure.
10 Nevertheless, it offers a precision and reproducibility not previously available.

Claims

1. A controlled growth apparatus, comprising:
a chamber capable of holding a substrate onto which a film is grown;
a plurality of sources within said chamber capable of simultaneously irradiating a surface of said substrate with a plurality of respective species to form said film;
at least one metering device capable of controlling an amount of a respective one of said species irradiated upon said surface;
an ellipsometer operable during an operation of said sources, illuminating said surface with a beam of light, and providing ellipsometric data from light reflected by said surface from said beam of light; and a controller generating a control signal from a sequence of a plurality of said ellipsometric data, said control signal controlling said at least one metering device during said operation of said sources.

2. A controlled growth apparatus as recited in Claim 1, wherein said controller generates said control signal based upon a target composition of said film.

3. A controlled growth apparatus as recited in Claim 2, wherein said control signals dynamically controls relative amounts of said species irradiated by said sources.

4. A controlled growth apparatus as recited in Claim 2, wherein there are at least three of said sources, whereby said target composition is at least a ternary composition.

5. A controlled growth apparatus as recited in Claim 4, wherein said growth apparatus comprises chemical vapor deposition.

6. A controlled growth apparatus as recited in Claim 1, wherein said control signal causes said sources to dynamically vary a composition of said film

7. A controlled growth apparatus as recited in Claim 1, wherein two of complementary ellipsometric data completely characterize a polarization state of said reflected light.

8. A controlled growth apparatus as recited in Claim 1, wherein said ellipsometric data are in a form of pairs of complementary ellipsometric data.

9. A method of controllably growing a film, comprising the steps of:
simultaneously irradiating a surface with a plurality of species, whereby a film is grown on said surface;

illuminating said surface with an incident beam of light during said irradiating step, whereby light is reflected from said surface in a reflected beam of light;
deriving ellipsometric data from said incident and reflected beams;
and dynamically controlling said irradiating step in response to said ellipsometric data.

10. A method as recited in Claim 9, wherein said irradiating step irradiates said surface with at least three species.

11. A method as recited in Claim 10, wherein said controlling step controls a positive rate at which said irradiating step irradiates said surface with at least one of said species.

12. A method as recited in Claim 11, wherein said controlling step controls said irradiating step additionally according to a target composition of said film.

13. A method as recited in Claim 12, wherein said controlling step controls said irradiating step to vary a composition of said film.

14. A method as recited in Claim 11, wherein during said irradiating step said film is grown as a semiconductor film.

15. A method as recited in Claim 9, wherein said deriving step includes determining a polarization state of said reflected beam.

16. A method as recited in Claim 15, wherein said determining step uses at least a three-phase model, said film comprising an intermediate one of said phases.

17. A method as recited in Claim 9, wherein said deriving step drives said ellipsometric data in a form of pairs of complementary ellipsometric data and said controlling step controls said irradiating step in response to a sequence of said pairs.