Method and device for in-situ determination of the temperature of a wafer

Abstract

Method for residual oscillation-suppressed emissivity-corrected in-situ pyrometry, ECP, of wafers (1) during coating in coating systems, comprising the following steps: - Radiation of a sample radiation A0 of a pyrometer wavelength λ p and a sample radiation B0 of a calibration wavelength λ c onto the surface of a wafer during coating in a coating system; - Determining an intensity I R the sample radiation A reflected from the wafer R ; - Determining an intensity s E the thermal radiation E emitted by the wafer, - Determining an intensity I c from the wafer at a second wavelength λ c reflected sample radiation B c ; - Determine the reflection R p of the wafer at λ p from intensity I R and intensity I 0Rthe incident sample radiation A0, - Determine the current reflection R c of the wafer at λ c from intensity I c and intensity I 0C the irradiated sample radiation B0, - Determine the temperature T w of the wafer during coating with a layer using the emissivity-corrected pyrometry, ECP, method, - Determining a phase difference τ between reflection Fabry-Perot oscillations of the reflection R p , R p -FPOs, and the emission Fabry-Perot oscillations of the thermal radiation emitted by the wafer, E-FPOs, during the growth of a layer material on the wafer, - Determine an effective reflection scaling α=α T the reflection R p during the growth of the layer material on the wafer, - Determine the temperature T wof the wafer using the ECP method using the phase difference τ and the reflection scaling α.

Description

[0001] The invention relates to a method and a device for in-situ determination of the temperature of a semiconductor substrate (hereinafter also referred to as a wafer) during the coating of this / these wafer(s) with layer structures within a process chamber, in particular to a method for increasing measurement accuracy in the pyrometric in-situ temperature measurement of a wafer using emissivity-corrected pyrometry (ECP). Furthermore, aspects of the invention disclose a calibration structure for said method and a measurement specification for calibration. Description of the invention

[0002] Methods for determining the temperature of a sample or a wafer and temperature measuring devices with a pyrometer and an evaluation device connected to the pyrometer for signal transmission are generally known.

[0003] Modern process pyrometers for thin-film processes are emissivity-corrected, as is known, for example, from DE 4419476 A1 or DE 10 2022 101 806 A1. This means that, in addition to the pyrometer detection, a reflection measurement is also integrated at the same wavelength. This reflection measurement is used to determine the current emissivity of the wafer during the process, which is modified by a layer structure on the surface, and thus to appropriately correct the pyrometer signal, enabling a more precise temperature determination.

[0004] With the improvements for correcting systematic pyrometer measurement errors, disclosed, for example, in US Pat. No. 6,398,406 B1 and in WGBreiland, "Reflectance-Correcting Pyrometry in Thin Film Deposition Applications," 2003, it is also known that systematic errors in this emissivity correction, caused, for example, by additional thermal radiation within the process chamber, must be appropriately corrected. This additional thermal radiation to be corrected does not originate from the wafer, which is hot due to the process, but rather from other hot parts in the process chamber. Such a correction is disclosed, for example, in DE 102020126597 A1.

[0005] Another known disturbance variable for precise pyrometric wafer temperature measurements, not yet included in the corrections according to US 6398406 B1 but now recognized as equally important, is the variation of the pyrometer effective wavelength with the wafer temperature. This variation of the pyrometer effective wavelength with the wafer temperature is unavoidable for physical reasons and was already described in W.G.Breiland, "Reflectance-Correcting Pyrometry in Thin Film Deposition Applications," 2003. Thus, DE 44 19 476 C2 discloses a corresponding method and device for determining the layer thickness and the substrate temperature during the coating of semiconductor substrates (wafers), which suitably correct the measurement errors resulting from the variation of the pyrometer effective wavelength, at least for a selected wafer temperature.

[0006] The core of the emissivity-corrected pyrometry method is that, using a single, combined apparatus at a suitable wavelength λ P both a pyrometric measurement of the thermal radiation s(T, λ P ) as well as a reflection measurement R(λ P ) at the wafer surface. Since under certain, generally well-fulfilled conditions, the reflection R(λ P ) directly the emissivity ε(λ P ) = 1 - R(λ P ) of the wafer surface can be calculated, thus the true temperature of the wafer T w can be calculated by an algorithm which results directly from Wien's approximation of the Planck function: 1Tw=1Tcal−λpc2lns∗(1−Rcal)scal*(1−R)

[0007] Equation (1) uses exactly the notation from the already cited patent US 6398406 B1 (there equation (3)). T cal , R cal and Scal are values ​​that are determined during a recalibration of the pyrometer on the process chamber, as is usual according to US 6398406 B1, which is carried out with a wafer of known reflection R cal at a known wafer temperature T C eal = (T cal -273)°C in the temperature range of 400°C to 600°C. T C marks temperatures that are measured in degrees Celsius.

[0008] Through this procedure, the measured temperature of the wafer T w no longer distorted by layer thickness effects (e.g. Fabry-Perot interference within the layer structure), which a “simple” pyrometry measurement, which only uses the thermal radiation s(T, λ P ) of the wafer, always significantly interfere.

[0009] US 6398406 B1 shows that a total of five systematic disturbance parameters for the wafer temperature measurement can be corrected under certain conditions by just one correction parameter y. In the MOCVD vertical reactor used there, these "certain conditions" were met, so that equation (1) could be extended by the correction parameter y as follows: 1Tw=1Tcal−λpc2lnsexp∗(1−Rcal+γ)scal∗(1−R+γ)

[0010] Equation (2) uses the notation from US 6398406 B1 (equation (8)). Equation (2) was derived in US 6398406 B1 from: Rexp=αR+β and sexp=Cδ(1−R)Lb+γ˜Lb+ξRLb where the five error sources named in US 6398406 B1 (scaling error α of the reflection measurement, offset error β of the reflection measurement, scaling factor δ in the thermal emission signal, additional thermal radiation γ̃ (reaching the detector directly) and additional thermal radiation ξ (reaching the detector after a reflection from the wafer) are included in equations (3). These 5 parameters are referred to as disturbance parameters. The measured quantity s exp is accordingly distorted by these disturbance parameters and no longer corresponds exactly to the thermal radiation s(T, λ P ).

[0011] The constant C is the scaling factor of the pyrometer to be calibrated under the measuring conditions in the process chamber and L b is the exponential term of the Planck distribution in the Wien approximation. Lb=e−c2 / λpT with temperature T, wavelength λ p and the second radiation constant c2.

[0012] In US 6398406 B1 it is shown that by rearranging the equations (3,4) the 5 disturbance parameters α, β, δ, γ̃ and ξ can be transformed into two new correction parameters p and y, whereby the experimentally measured thermal radiation s exp can be described by: sexp=p(1−Rexp+γ)e−c2 / λpT

[0013] Of course, the new parameters p and y calculated from the disturbance parameters α, β, δ, γ̃ and ξ only become true correction parameters after their values ​​have been determined by appropriate calibration. Since the new parameter p was determined during the then usual recalibration of the pyrometer at the process chamber (using a wafer of known reflectance R cal at a known wafer temperature T calin the temperature range from 400°C to 600°C and under conditions in which the disturbance parameters α, β, δ, γ̃ and ξ act in the same way as in the coating process), only the correction parameter y remains in equation (2), which is to be determined by a suitable calibration procedure described in US 6398406 B1.

[0014] However, the application of equation (2) and US6398406B1 is limited to coating processes where the total thickness of the film at λ Ptransparent layers on the semiconductor wafer remains smaller than a critical limit (e.g., 3 µm). If this limit is exceeded, the measured ECP wafer temperatures according to equation (2), despite correction with the best-chosen correction parameter γ, exhibit characteristic residual oscillations that are ±90° out of phase with the Fabry-Perot reflection oscillations (reflection FPO) of the growing layers. These (usually small) ±90° out of phase temperature residual oscillations are measurement artifacts that disrupt the temperature control of the coating system during temperature control processes.

[0015] DE 102020111293 A1 therefore discloses a corresponding improved method and a device for determining the layer thickness and the substrate temperature during the coating of semiconductor substrates (wafers), which for a selected wafer temperature (by suitable adaptation of the reflection wavelength λ Rto the effective pyrometer wavelength λ P ) even for very thick, at λ P transparent layer stacks with high temperature measurement accuracy can be used.

[0016] The application of equation (2) (according to US 6398406 B1) is also limited to coating processes where the process temperatures do not differ significantly from the pyrometer calibration temperatures. This was still the case at the time of publication of US 6398406 B1, since the semiconductor structures industrially produced using MOCVD at that time were primarily III-arsenide and III-phosphide layer structures on GaAs and InP wafers. Such structures are coated at 600°C to 750°C, i.e., at temperatures only approximately 100-200K above the pyrometer calibration temperatures at that time. According to US 6398406 B1, this pyrometer calibration temperature at the process chamber (post-calibration) is between 400°C and 600°C, although the Al / Si eutectic temperature of 577°C is known to have been frequently used for this purpose.However, for the MOCVD processes that dominate industry today for the deposition of III-nitrides on silicon and SiC substrates, process temperatures in the range of 1000°C to 1300°C are typical. Therefore, the recalibration of the pyrometer channel according to EP 2365307 B1 is carried out very precisely using compact reference light sources, which can be conveniently inserted into the process chamber during maintenance and enable absolute pyrometer calibrations at reference temperatures that are only 100K to 200K below the actual process temperatures for III-nitride processes.

[0017] The disadvantage of all these known ECP methods is that for certain coating processes of multilayers on semiconductor wafers (e.g., III-nitride epitaxy on silicon and SiC substrates) and for very high accuracy requirements, they are no longer sufficient to eliminate all Fabry-Perot interference artifacts from the wafer temperature measurement with sufficient accuracy. This particularly applies to residual temperature oscillations in coating processes where particularly critical measurement conditions occur, such as a) when particularly thick layers or layer systems are applied, which also consist of layer materials which, at the pyrometer wavelength λ P are transparent (k = 0), and / or b) if Fabry-Perot oscillations (FPO) with a particularly large amplitude occur in the reflectivity measurement and the maximum of the Fabry-Perot oscillations in the reflectivity measurement is at very large values, which means that little thermal radiation can escape from the wafer, and / or c) if additional light scattering effects occur at the location of the pyrometer measurement within the semiconductor layer structure in combination with additional thermal radiation from hotter components in the reaction space (ie hotter than the temperature of the semiconductor wafer to be measured), and / or d) if actual deviations occur between the effective pyrometer wavelength λ P and the effective reflection wavelength λ R(which result from the shape of the Planck curve and the spectral width of the pyrometer wavelength filter). According to DE 102020111293 A1, such deviations can be compensated for a selected temperature, but not for an entire temperature range, and / or e) if apparent deviations occur between the FPO period width of the emission measurement (at the effective pyrometer wavelength λ P ) and the reflection measurement (at the nominally identical effective reflection wavelength λ R ), which actually result from the different aperture angles of reflection and emission measurement (effective reflection angle of incidence range and effective emission detection angle range). DE 10258713 B4 discloses in Fig. 4 an ECP pyrometer measuring head, which significantly reduces the susceptibility of the reflection measurement to wafer wobble in the process due to these different aperture angles of the reflection and emission measurement, and / or f) for small deviations between the emission FPO line shape and the reflection FPO line shape, which arise from the geometric proximity of the emission light source to the growing layer structure (for GaN / Si structures, this is the upper ~50 µm of the Si substrate at 1000°C). An extreme example of such emission light intensity resonances is described in: Fenolosa et al., "Thermal emission of silicon at near-IR frequencies mediated by Mie resonances," ACS Photonics 2019.

[0018] Fig. 1 shows a planetary reactor as an epitaxial reactor in which, for example, large silicon wafers (here: 5 wafers with a diameter of 200 mm each) are coated with very thick and complex III-nitride layer structures at high process temperatures (800°C-1200°C).

[0019] Fig. Figure 1a shows a schematic side view of such a reactor. The wafers 1 lie on satellite wafer carriers 2, each of which rotates around its satellite rotation axis 3. The main rotation of the main susceptor 4 around the central main rotation axis 5 is superimposed on this satellite rotation. The wafers are heated indirectly via the heater 6 and the heat transfer heater→main susceptor→satellite wafer carrier→wafer. The wafer temperature is measured via the pyrometer detection optics 9 through the measurement window 8, which is integrated into the reactor outer wall 7. Both the temperature at λ pThe radiation E emitted thermally by the wafer and the wafer reflection, which varies during the layer growth process, are measured. The two wavelengths λ p and λ c whose reflected light rays (A or B) are evaluated in their intensity ratio to the incident light rays (A0 or B0).

[0020] Fig. Figure 1b shows a top view of the planetary reactor. It shows a 5x200mm configuration, as used in the production of modern GaN / Si power transistors. The five wafers were labeled W1 ... W5. The dotted line shows measurement trajectory 13, along which measurements are taken with the ECP pyrometer during a full main rotation (duration, e.g., 12 s). Ideally, this measurement trajectory passes through the centers of the five wafers, i.e., the centers of rotation of the five satellite rotations. The directions of rotation of the main rotation and the satellite rotations are indicated by the arrows.

[0021] Furthermore, Fig. 1b shows three exemplary wafer zones 14a, 14b, and 14c, on which measurements are triggered at suitably programmed times during each main rotation, which can then be assigned to fixed radial positions on the wafers. In the example, on wafer 1, the wafer center 14a, the wafer edge (14c), and a position at half the wafer radius 14b are detected. On satellite W2, Fig. 1b additional areas 11 and 12 are marked, the meaning of which is explained below: 1. The spatial resolution of the measurements is determined by two parameters: a) the size of the measurement spot (circular, diameter approximately 1 mm) and b) the "wandering" of the measurement spot on the wafer during the measurement time (-200 ms) due to the superposition of the main rotation and the satellite rotation. This is represented by the black areas 11. 2. Since the main rotation and satellite rotation are not synchronized, there are larger radial segments (gray measurement area segments 12) for all wafer zones, each of which is measured "randomly." However, since no epitaxy is ideally homogeneous, this leads to additional measurement signal fluctuations and a damping of the FPO amplitudes, particularly in the outer wafer zones near the edge. In the central wafer zone, these effects of satellite rotation are smaller but still detectable.

[0022] The Fig. Figure 2 shows an example of a III-nitride layer structure produced in such a planetary reactor, which is based on a coating process that has particularly critical measurement conditions for the wafer temperature and at the same time requires a very high accuracy of the wafer temperature measurement.

[0023] Fig. Figure 2a schematically shows a III-nitride layer structure of a current GaN power transistor on silicon, consisting of a layer stack typically 2-4µm thick and composed of p-GaN cap layer 108, AlGaN barrier layer 102, AlN interface layer 103, i-GaN channel layer 107, Fe-GaN layer 106, GaN / AlGaN buffer and strain management layer 104, nucleation layer (AlN, AlGaN) 105 and the silicon substrate 101.

[0024] The Fig. Figure 2b outlines such a transistor structure after further technology steps with a metal contact 109 on the now structured p-GaN 108. Fig. Figure 2b also shows the most important functional element of such a transistor structure, the 2-dimensional electron gas 110, which is located directly at the interface between the i-GaN and the AlGaN barrier, and whose properties depend heavily on the composition and layer thickness of the AlGaN barrier layer 102. Therefore, the wafer temperature must be set extremely precisely for all wafers in the reactor shortly before reaching this interface during the coating process. These very tight wafer temperature tolerances must also be maintained across the respective wafer surfaces (wafer center to wafer edge). For the associated control processes, the residual temperature oscillations must be minimized to values ​​significantly below 0.2 K, which was previously not possible with the state of the art.

[0025] Fig. 2c shows a small section of a wafer reflection measurement at a pyrometer wavelength λ P=950nm during the growth of such a transistor structure in a planetary reactor. This data segment was chosen to Fig. 1 to make the special features of reflection measurement in such a reactor immediately visible. This wafer reflection at the pyrometer wavelength (here: λ P =950nm) must be measured with high precision for any form of emissivity correction, but has specific characteristics in the planetary reactor, which must be considered in the ECP method. Fig. 2c, a process step without growth (after the growth of the Fe-GaN layer 106 and before the growth of the i-GaN layer 107), in which the reflection should actually remain constant, was analyzed specifically. The measured values ​​of a central measurement zone 14a are compared with a measurement zone 14c at the wafer edge (W2 edge ). The reflection values ​​of the measuring zone 14a (W2 center) in the wafer center show little "noise" (signal fluctuations due to minimally varying total layer thickness on the wafer surface), whereas the measurement on measurement zone 14c (at the wafer edge) shows very clear signal fluctuations in the reflection signal. These signal fluctuations are caused by the smallest local growth rate variations during the previous growth of the underlying, very thick buffer layer 104 leading to significantly different FPO phases of reflection during the growth of the Fe-GaN layer 106. Fig. 2c, during Fe-GaN growth, the reflection maxima in the wafer center are reached approximately 50 s later than at the wafer edge. At a growth rate of approximately 0.9 nm / s, this corresponds to a layer thickness difference of 45 nm of GaN in the underlying buffer layer 104. Although the FPO phase differences within the radial measurement surface segments 12 are significantly smaller, they nevertheless generate these reflection signal fluctuations. The latter have nothing to do with the measuring device, but rather result from the remaining non-homogeneity of the epitaxy in combination with the double rotation in the planetary reactor. Particularly at the wafer edge, the stochastic position of each real single-point measurement within the radial segments 12 leads to these measurement signal fluctuations. These were shown here as an example for the reflection, but also occur in the raw temperature signal (via the emissivity fluctuation that is opposite to the reflection fluctuation).This will have to be taken into account in the algorithm of the improved, residual oscillation-free ECP method presented below.

[0026] The Fig. 1a (side view of the reactor) are used for the highly accurate wafer temperature measurement and the elements of the ECP measuring system (9, E, A, A0, B, B0) Fig. The double rotation of the 5 wafers shown in Figure 1b (top view of the wafer plane in the reactor) serves to ensure the required high homogeneity of all coating parameters in order to achieve very high process yields for such coating processes.

[0027] The extremely high requirements for wafer temperature accuracy and homogeneity in the coating of the AlGaN barrier 102 result from the fact that this layer is used in the final semiconductor components manufactured from it ( Fig. Figure 2b shows an example of such a power transistor structure, which is responsible for the most important performance parameters. Since both the composition of this AlGaN barrier (typically: 10%...30% Al with 0.1% accuracy) and its layer thickness (15nm-30nm with 0.5nm accuracy) depend heavily on the wafer temperature during the process, DE 102020100481 A1 and DE 102019104433 A1 employ a wafer temperature fine control, which requires a correspondingly high real-time measurement accuracy for the wafer temperature measurement. The current measurement accuracy after a γ-correction is approximately ±0.5K and is determined by the residual oscillations in the temperature measurement signal that remain even after the γ-correction.Furthermore, as shown below, the requirements for applying a γ-correction according to US 6398406 B1 are not met in the planetary reactor, and even scaling the gamma parameter with the measured reflectance (DE 102020126597 A1) only partially corrects these weaknesses and leads to errors in the absolute temperature value. The γ-correction, as the state of the art, is therefore no longer sufficient. The reasons for its limitations in minimizing the residual temperature oscillations in measurements in the planetary reactor are explained below.

[0028] According to US 6398406 B1, some causes of residual oscillations in the pyrometric measurement signal can only be corrected by a correction parameter γ according to the state of the art (Eq. (2)) if the conditions used in US 6398406 B1 for deriving Eq. (2) are met, i.e.: 1. when the sources of additional, non-wafer IR radiation in the reactor are at the same temperature as the wafers. This results from equations (3,4), since in the exponential term L b the same temperature T is used for the thermal radiation from the wafer and from the sources of additional radiation; 2. if all five of the disturbance variables named in US 6398406 B1 remain constant over time during the process (i.e. concerning the 5 parameters scaling error α of the reflection measurement, offset error β of the reflection measurement, scaling factor δ in the thermal emission signal, the additional thermal radiation γ̃ reaching the detector directly and finally the additional thermal radiation ξ reaching the detector only after a reflection from the wafer); 3. if only ideal, homogeneous, non-optically scattering layer materials are used within the sample and the substrate material is suitable for light of the pyrometer wavelength λ Pis not transparent; and finally 4. When the usual recalibration of the pyrometer is performed in the process chamber using a wafer of known temperature and known reflectance, so that the additional thermal radiation γ̃ and ξ also contribute during this recalibration. If other recalibration methods are used (e.g., EP2365307B1), minimizing the residual temperature oscillations using the correction parameter γ can even be counterproductive, as it distorts the measured absolute temperature.

[0029] Typically, none of these four prerequisites is met for today's particularly important groups of processes, such as liquid-phase oxidation deposition (MOCVD) in so-called planetary reactors for the epitaxial growth of group III nitrides (GaN, AlGaN, InGaAIN) on foreign substrates such as silicon and SiC. These processes are particularly important for the production of power electronic transistors with very high power and high switching frequencies.

[0030] Even if equation (2) were applicable, a complete elimination of residual temperature oscillations would still not be possible because equation (2) does not account for the often crucial error sources of the critical measurement conditions d), e), and f) (as described above). Therefore, practice shows that despite the best possible adaptation of equation (2) to the conditions in planetary reactors (DE 102019104433 A1) and despite the best possible minimization of additional sources of interference radiation in the reactor (DE 102020101066 A1), residual ECP oscillations with amplitudes of approximately ±0.5 K remain in relevant processes, which can disrupt precise temperature control in important epitaxy processes. Examples of such important temperature control processes can be found in DE 102020100481 A1 and DE 102019104433 A1.

[0031] The invention is therefore based on the object of overcoming or at least mitigating these problems of the prior art and providing a method for increasing measurement accuracy in pyrometric in-situ temperature measurement. In particular, residual oscillations in the pyrometric measurement signal resulting from phase differences between the emission FPOs and the reflection FPOs are to be corrected in real time.

[0032] According to the invention, this object is achieved by an improved method for residual oscillation-suppressed emissivity-corrected in-situ pyrometry according to claim 1. Preferred embodiments of the invention are contained in the subclaims.

[0033] The inventive method for residual oscillation-suppressed, emissivity-corrected in-situ pyrometry of wafers (in coating processes) dispenses with the reduction of the five error sources (scaling error α of the reflection measurement, offset error β of the reflection measurement, scaling factor δ in the thermal emission signal, thermal direct additional radiation γ̃, and thermal indirect, wafer-reflection-dependent additional radiation ξ) to only a single correction parameter y. This so-called γ-correction, which is used in US 6398406 B1 and DE 102020126597 A1, is - as already explained above - no longer successfully applicable to the current conditions in modern epitaxial reactors.

[0034] Alternatively, two new correction parameters are used: an FPO phase correction τ and a re-scaling factor α T the reflection measurement. The parameter α Twas chosen because this generally well-known reflection scaling is often necessary (the reflection measurement conditions change during the process due to thermal effects on the reactor or due to the unavoidable strain-induced warping of large substrate wafers) and because α T the effect of all other remaining disturbances can also be completely compensated (in combination with improved pyrometer recalibration at the reactor).

[0035] The completely new FPO phase correction parameter τ, however, is introduced according to the invention because all previous variants of the γ-correction remained ineffective against the frequently occurring and technically never avoidable small phase shifts between the reflection FPO and the emission FPO for all temperatures simultaneously.

[0036] This FPO phase difference τ is preferably measured in s (second). Both small differences in the wavelengths λ p and λ R as well as FPO phase-shifting effects due to slight differences in the effective measurement angle of the R (reflection) and E (emissivity) measurement of the thermal radiation are taken into account in this correction parameter τ.

[0037] Accordingly, the new algorithms according to the invention for the complete minimization of the temperature residual oscillations are preferably based on the following reformulation of the previous equation (5): sexp(t)=pT(1−αTR(t+τ))e−c2 / λpTw

[0038] Similar to p and γ in Eq.(5), p T and α T functionally depends on all disturbance parameters α, β, δ, γ̃ and ξ, where the functional relationships p derived from Eqs. (3) and (6) T =p T (α, β, δ, γ̃ ,ξ) and α T =α T(a, β, δ, γ̃ ,ξ) differ from those for p and γ.

[0039] Equation (6) states that during the ECP measurement during layer growth under constant growth conditions, the FPOs of the measured emission radiation s exp (t) may have a (small) phase difference τ with respect to the FPOs of the simultaneously measured reflection R(t), which may be caused e.g. by very small differences between the effective pyrometer wavelength λ P and the effective reflection wavelength λ R or resulting from the above-mentioned critical measurement conditions d), e), and f). This phase difference τ is largely responsible for residual oscillation components in the measured wafer temperature T w , which exhibit a ±90° phase shift to the reflection FPOs. For an ideal ECP measurement according to equation (1), τ would be 0.

[0040] The scaling error α Tin equation (6), however, stands for all non-idealities in the ECP measurement, which lead to inconsistencies between the amplitudes of the reflection FPOs in R(t) and the amplitudes of the emission FPOs in s exp (t). Causes for such inconsistencies can be both actual errors in the reflection scaling and effects of the additional thermal scattered light in the reactor (compare the critical measurement condition c above and the perturbation parameters γ̃ and ξ already introduced in US 6398406 B1).

[0041] The scaling factor p Tin equation (6) is calibrated separately, for example, according to the method described in EP 2365307 B1 or using suitable, novel calibration wafers that can also be used at very high temperatures (600°C–1000°C). Such a novel calibration wafer is presented below as an example as part of the method described here. Whether calibration with a standardized reference light source according to EP 2365307 B1 is still useful depends on the relative proportion of the disturbance parameters α, β, δ, γ̃, and ξ that are still effective in the specific epitaxial reactor. It was found that the additional thermal radiation γ̃ (reaching the detector directly) and the additional thermal radiation ξ (reaching the detector after reflection at the wafer) are generally no longer relevant in planetary reactors, since these stray light effects are reduced by the size of the wafers (200mm diameter), the flat geometry of the planetary reactors ( Fig. 1a) and possibly through additional design measures, such as those according to DE 102020101066 A1, generally no longer occur. However, for older reactors, γ̃ and ξ may still need to be considered and corrected.

[0042] With equation (6) and p T =1 (after separate pyrometer recalibration) equation (1) can be reformulated into 1Tw(t)=Acal−λpc2lnsexp(t)(1−αTRexp(t+τ)) where Acal=1Tcal+λpc2lnscal(1−Rcal) The pyrometer recalibration parameter known from US6398406B1 is used to correct the factory calibration of the pyrometer according to the conditions in the process chamber. A cal is preferably determined using a calibration wafer in the reactor, the temperature of which T ca and its reflection R calare exactly known at the time of this in-situ calibration, whereby this recalibration must be carried out under the same reactor conditions as the actual epitaxy process.

[0043] The method according to the invention for residual oscillation-suppressed emissivity-corrected in-situ pyrometry of wafers in coating systems preferably has two aspects: 1. In-situ pyrometer recalibration is performed using a calibration wafer at high process temperatures (comparable to the process conditions of the actual film growth). This procedure ensures that, in older reactors, any interference from additional thermal scattered light is also included in this calibration. 2. The well-known emissivity-corrected pyrometry with γ-correction during layer growth is replaced by a novel emissivity-corrected pyrometry with α-τ-correction, which enables the complete minimization of both residual oscillations that are in phase (0° phase difference) or in anti-phase (180° phase difference) to the reflection FPOs, as well as residual oscillations that are ±90° out of phase with the reflection FPOs.

[0044] According to one aspect of the present invention, a method for residual oscillation-suppressed emissivity-corrected in-situ pyrometry, ECP, of wafers (1) in coating systems is disclosed, comprising the following steps (cf. Fig. 3): - Determining a thermal radiation (E) emitted by the wafer (1) at a pyrometer wavelength λ p , preferably within a bandpass Δλ p . - irradiating a sample radiation (A0) onto a surface layer of a wafer (1), wherein the spectral intensity distribution I0(λ) of the sample radiation A0 is preferably within the bandpass range Δλ P is adjustable so that for at least one temperature in the relevant temperature range for the effective reflection wavelength λ R applies: λ P =λ R ; - Determining a total intensity of the sample radiation (A) reflected by the wafer (1) to calculate the reflection R p from the intensities of A and A0. - Determine the temperature (T w ) of the wafer (1) by means of the ECP method, characterized by - Determining a phase difference τ between reflection Fabry-Perot oscillations of the reflected sample radiation A and the emission Fabry-Perot oscillations of the thermal radiation E, where - the temperature (T w) of the wafer (1) is determined by means of the ECP method using the phase difference τ.

[0045] Preferably, a reflection scaling factor α T used for reflection measurement and the temperature (T w ) of the wafer (1) by means of the ECP method additionally using the scaling factor α T certainly.

[0046] The optimal value for α T results from the complete disappearance of the in-phase or anti-phase (0° or 180°) residual oscillation components in the resulting ECP-corrected thermal emission intensity s ECP (t)=S exp (t) / [1-α T R exp (t)].

[0047] The optimal value for τ results from the complete disappearance of the residual oscillation components phase-shifted by ±90° in s ECP (t).

[0048] Since each FPO residual oscillation can be completely described by the classical ECP method by its in-phase / antiphase (0° or 180°) phase-shifted residual oscillation component and its ±90° phase-shifted residual oscillation component, the result of this α-τ-ECP method is a temperature T w of the wafer, which is completely free of FPO residual oscillations - regardless of the cause of these residual oscillations in the classic ECP process. To ensure this temperature T w of the wafer is identical to the true temperature of the wafer, the pyrometer recalibration parameter A cal preferably be determined by using a calibration wafer of known reflectance and known temperature under process conditions that best resemble the process conditions of the sample during layer growth.

[0049] Preferably, a Virtual Interface Fit (VI-Fit) method is used to calculate the measured reflection Rexp (t), to determine the reactor and process-related fluctuations of the reflection signal and then to calculate a phase-shifted reflection R exp (t+τ) to which the reflection fluctuations are transferred.

[0050] This VI-Fit provides four parameters: the growth rate r and the three following auxiliary quantities: the amplitude of the reflection at the virtual interface, the phase of the reflection at the virtual interface and, if necessary, a scaling parameter α R to improve reflection calibration. At the same time, this VI-Fit provides a noise-free fit curve R fit (t) to the measured, FPO-dominated reflection R exp(t). This fit curve can be phase-shifted by adding a virtual intermediate layer of thickness Δ between the virtual interface and the simulated growing layer, where n and k of this virtual intermediate layer are chosen to be identical to the growing layer. The layer thickness of Δ is typically between -1 nm and +1 nm. This results in a fit curve R that can be phase-shifted by Δ. p fit (t+τ).

[0051] However, the procedure for determining the optimal phase difference τ still requires an important intermediate step: smallest growth inhomogeneities in the reactor and statistically fluctuating measuring locations on the wafer (due to wafer rotation) can lead to real, small fluctuations in R exp (t) and in s exp (t), which are in antiphase and thus in the T resulting from the ECP w (t) are no longer effective. The procedure described so far now eliminates in R fit(t) these real, small inhomogeneity fluctuations, so that when using R fit (t) for the ECP method in the resulting ECP-corrected thermal emission intensity s ECP (t) the antiphase real fluctuations from s exp (t). Therefore, after the VI-Fit, the reflection R p (t) the time course of these inhomogeneity fluctuations δR(t)=R exp (t)-R fit (t) and then again to the phase-shifted fit curve R fit (t+τ) are imposed before R exp (t+τ)= R fit (t+τ)+δR(t) is used for ECP correction.

[0052] Preferably, the phase difference τ is assigned to a virtual additional layer thickness Δ, so that τ can be automatically adjusted for different consecutive process steps (with different growth rates r of the layers) by τ=Δ / r.

[0053] The optimal value for Δ results from the complete disappearance of the ±90° phase-shifted residual oscillation components in the resulting ECP-corrected thermal emission intensity S ECP .

[0054] Preferably, the temperature (T w ) of the wafer using the ECP method with full compensation of the following disturbance parameters: scaling error α of the reflection measurement, offset error β of the reflection measurement, scaling factor δ in the thermal emission signal, additional thermal radiation γ̃ , ​​additional thermal radiation (after reflection at the wafer) ξ.

[0055] Preferably, the phase difference τ is given by the following equation (6*): sexp(t)=pT(1−αTRexp(t+τ))e−c2 / λpTw defined, where s exp (t) the measured thermal radiation of the wafer, which depends on time t, α T a scaling parameter of the reflection measurement, p Ta calibration wafer attached to the process chamber as part of A cal scaling factor of the pyrometer to be determined and R exp (t+τ) is the measured reflection of the wafer after phase shift τ.

[0056] Preferably, the temperature is determined during vapor-phase epitaxy. Preferably, the temperature is determined in a planetary reactor during the epitaxial growth of group III nitrides on foreign substrates.

[0057] Preferably, the temperature to be determined (T w ) of the wafer is defined by the following equations (7*) and (8*) 1Tw(t)=Acal−λpc2lnsexp(t)(1−αTR(t+τ)) Acal=1Tcal+λpc2lnscal(1−Rcal) where A cal is determined by means of a calibration wafer in the reactor, the temperature of which T cal and its reflection R cal are exactly known at the time of this in-situ calibration.

[0058] Preferably, the ECP measuring system has at least two reflection wavelengths: λ p (pyrometer wavelength of the sample radiation A) and λ c . (Calibration wavelength of sample radiation B).

[0059] Preferably, the calibration wavelength λ c . in the short-wavelength spectral range, where the layer materials of the layer stack are still optically transparent. For GaN / Si processes, the spectral range between 500 nm and 700 nm is preferred.

[0060] Preferably, the layer structure of the calibration wafer is selected so that it remains (as far as possible) similar or the same as the production wafers produced in the epitaxial process under ECP temperature monitoring.

[0061] Preferably, the layer structure of the calibration wafer is selected such that the reflection of the sample radiation B has a characteristic minimum or maximum in the temperature range just below the process temperature range (for GaN / Si epitaxy: between 800°C and 1000°C).

[0062] The accuracy of the method is better, the smaller λ c (without leaving the transparency spectral range of the layers). Furthermore, it is advantageous if the layer structure on the calibration wafer is very similar to the layer structure of the wafers in the actual coating process.

[0063] According to a further aspect of the present invention, a method for controlling the temperature T w a wafer (1) in a planetary reactor during gas phase epitaxy for growing semiconductor layers of group III nitrides, wherein the temperature T wof the wafer is determined by means of the ECP method according to one of the preceding claims. Preferably, the determined temperature T w of the wafer is compared with a reference value and the temperature T w of the wafer is adjusted taking this comparison into account, e.g. by means of the heating system 6 or via the rotation speed of the satellite plate 2.

[0064] Preferably, the temperature T w of the wafer is determined before a control time segment in a learning time segment, from which the correction parameters α and τ or Δ result through a corresponding 2-parameter optimization towards the complete disappearance of the T w (t)-residual oscillations.

[0065] Preferably, a best-fit curve is additionally fitted to the reflection FPOs measured in the learning time segment in R exp (t) is determined.

[0066] For temperature control in the immediately following control time segment, preferably each individual R exp -measurement point receives an α-Δ correction as part of a single-point correction process. The control time segment is therefore also called the autocorrection time segment. This is preferably done by fitting the fitted curve R fit (t) and the phase-shifted fit curve R fit (t+τ) from the learning time segment into the immediately following autocorrection time segment. In the autocorrection time segment, a transfer of the current inhomogeneity-related R exp (t) fluctuations on the extrapolated phase-shifted curve R fit (t+τ) before it is used for ECP correction. This also corrects for possible minor drifts in the mean growth rate during the process.

[0067] Preferred method variant: if the growth rates in the learning time segment and in the autocorrection (AK) time segment differ significantly (e.g. due to different process conditions in different process steps), the method can still be applied in a modified form: The new growth rate r AK is determined by extended VI-Fit in a correspondingly extended layer model (maintaining α and Δ, but also r and the virtual interface parameters), so that extrapolated curves R are obtained for the subsequent control measurement in the autocorrection time segment. fit (t) and R fit (t+τ). This allows the ECP method to be applied instantaneously to each individual point without any further fitting procedure.

[0068] According to a further aspect of the present invention, a device for residual oscillation-suppressed emissivity-corrected in-situ pyrometry ECP of wafers (1) in a coating system is disclosed, comprising: a light source (47) for irradiating sample radiation (A0 and B0) onto the surface of a wafer (1); a spectral filter unit (45) with a bandpass characteristic for filtering the sample radiations (A and B) reflected by the wafer and a thermal radiation (E) emanating from the wafer, a detector unit 40 for measuring the total intensity of the radiations A, B and E, a control unit (50) configured to convert the spectral intensity distribution I0(λ) of the sample radiation A0 to the effective effective wavelength λ R =λ p for reflection measurement R p and an evaluation unit (60) arranged to determine the temperature T from the measured intensity of the radiations A and E wof the wafer by means of the method according to one of the preceding claims and to enable the pyrometer recalibration by means of a calibration wafer by evaluating the radiation B.

[0069] The present invention further discloses a computer program product comprising instructions which, when executed by a computer, cause the computer to carry out the steps of the method according to claim 1 and the dependent claims.

[0070] The present invention further discloses a computer-readable (storage) medium comprising instructions which, when executed by a computer, cause the computer to carry out the steps of the method according to claim 1 and the dependent claims.

[0071] The invention is explained below in exemplary embodiments with reference to the accompanying drawings. They show: Fig. 1 a schematic representation of the geometric and measuring conditions in a planetary reactor for the MOVCD (metal-organic vapor deposition) of layer structures made of semiconductor materials (a: side view; b: top view); Fig. 2a-c a layer stack of a GaN / AlGaN power transistor on silicon wafer (a: layer stack as grown in epitaxy; b: exemplary schematic transistor geometry after etching of the layer system; c) Example of measurement data, which during a coating process for a layer stack according to a) show that the special configuration on the planetary reactor (see Fig. 1) leads to additional fluctuations being imposed on the measurement signal. Fig. 3 an apparatus and method for emissivity-corrected wafer temperature measurement: a: functional sketch of the apparatus and b: block diagram of the algorithm and its synchronization to the coating process Fig. Figure 4 shows pyrometrically determined temperatures during the growth of a relatively thick GaN layer under constant growth conditions. Accordingly, they contain a) the result according to the prior art (γ-correction); b) the result according to the invention after α-τ correction; and c) the result of an α-τ correction, in which, for testing purposes, the transfer of reflection fluctuations to the α-τ-corrected reflection curve was omitted. Fig. Figure 5 shows an embodiment of the method, in which the α-τ correction parameters are first determined in a learning time segment and then applied to individual measurement points in a real-time correction process in an autocorrection time segment. a) shows the measured data and b) explains the virtual interface (VI) model used. Fig. 6 shows an embodiment of the method according to the invention, in which (unlike in Fig. 5) the α-τ correction extends over three process steps, each with a different growth rate of the GaN layers. Fig. 7 a calibration wafer for in-situ calibration of the pyrometer. Shown in a) is a calibration wafer structure that is almost identical to a real GaN / Si power transistor and in b) the in-situ measurement data (temperature, R(λ p ) and R(λ C )), where the R(λ C =633nm) curve shows a characteristic, temperature-specific minimum at T cal =943°C. In c) the exact determination of this T cal -value is shown, using the first derivative of a polynomial fit to these measured data.

[0072] The Fig. 1 has already been explained in detail above. Therefore, it should only be added here that the side view Fig. Figure 1a illustrates that these reactors are generally very flat, as the vertical distance between the wafer support (2, satellite plate) and the ceiling plate 10 is much smaller than the horizontal reactor diameter. This makes it almost impossible, at least in the wafer center, for thermal scattered light from sources outside the wafer (e.g., from the main susceptor 4) to reach the detection optics of the ECP pyrometer 9. However, the basic principle of heat transport in this reactor (heat transfer heater 6 → main susceptor 4 → satellite wafer carrier 2 → wafer 1) results in the large surfaces of the main susceptor 4 being at higher temperatures than the wafer surface—and thus becoming sources of additional thermal scattered light in the reactor.Therefore, in newer reactors (e.g. according to DE 102020101066 A1) main susceptor cover plates are used, so that the disturbing detection of thermal radiation from the main susceptor should be greatly reduced even for the wafer surfaces near the wafer edges (i.e. close to the main susceptor).

[0073] In the top view of the Fig. Figure 1b clearly shows that the measurement trajectory 32 of the ECP pyrometer is positioned by the measurement window position in the reactor such that it passes through the wafer centers of all five wafers. At the same time, it is clear that this trajectory sweeps over different sub-areas of the outer wafer regions with each new main rotation—due to the satellite rotation not being synchronized. This can, as already described above, lead to small fluctuations in the R during the entire coating process. p-measurement signal, which, however, usually remain small due to the high coating homogeneity in such reactors. These small R p Fluctuations are masked by the larger FPO, but are relevant for the ECP method presented here. These fluctuations can, depending on the ratio of the main rotation to the satellite rotation, also take the form of beat signatures in R p For ECP measurement under ideal conditions (without phase shift between the R exp (t) and s exp (t) FPOs) this is not relevant for the ECP temperature measurement, since the emission s exp (t) and the reflection R exp (t) react in antiphase to the layer thickness fluctuations, thus leaving the ECP-corrected temperature signal free of these fluctuations and beat signatures. However, for the α-τ correction presented here, these fluctuations or beat signatures must necessarily be aligned with the phase-shifted R exp(t,τ) values ​​are transferred (see also Fig. 4c).

[0074] Another special feature of the planetary reactor should be mentioned here. Due to the size and design principle of the reactor, the main rotation is relatively slow. At a typical main rotation frequency of 5 rpm, a measurement value for a given position on a wafer (e.g., the wafer center of wafer W1) is only recorded every 12 seconds. This poses a challenge for the wafer temperature fine control, which is usually achieved via the flow of the rotating gases of the satellite plates 2 (DE 102020100481 A1 and in DE 102019104433 A1). Therefore, the satellite rotation must be adjusted based on the temperature measurement at a single measuring point. However, a single measuring point does not have an "FPO phase," which is corrected by τ in the method presented here. This explains the effort required for the algorithm for the real-time phase shift of the Rp -FPOs was necessary.

[0075] Fig. Figure 2 shows the layer stack of a GaN / AlGaN power transistor on a silicon wafer: a) the layer stack as grown in epitaxy; b) exemplary schematic transistor geometry after etching the layer system. Fig. Figure 2c explains how the interaction of the very thick, IR-transparent layers with the peculiarities of the planetary reactor leads to small fluctuation values ​​being superimposed on the ECP temperatures when there is no transfer of the R p (t)-measurement value fluctuations on the R p (t+τ) curves.

[0076] The typical layer sequence of a GaN / AlGaN power transistor is Fig. 2a: The nucleation layer 105 on the silicon substrate 101 serves to maintain the crystal quality of the entire layer structure, as it mitigates the effects of the different lattice constants of the silicon substrate and the group III nitrides. In the GaN / AlGaN buffer layer 104, relaxations are initiated by special compositions and layer thicknesses of the sublayers contained therein. These relaxations minimize crystal defects and adjust the strain in the layer structure at growth temperatures (approximately 1000°C) such that the different thermal expansion coefficients of the substrate and layer materials result in the wafer being almost flat again after cooling, thus being suitable for subsequent processes (e.g., lithography). The Fe-GaN layer 106 is very high-resistance due to compensation doping in order to improve the breakdown voltages of the transistors.The i-GaN channel layer 107 is undoped and is grown at lower growth rates than the Fe-GaN layer in order to produce it with as few defects as possible. This is important because the 2-dimensional electron gas (2DEG) 110 forms at its upper interface with the AlGaN barrier layer 102. This 2DEG, with its extremely high charge carrier mobility, is intended to enable the high currents of the power transistors. This 2DEG is generated by the polarization effects at the i-GaN / AlGaN interface. It is therefore important that the AlGaN barrier layer 102 be grown at an extremely precise wafer temperature, as otherwise its aluminum content and layer thickness would vary, which in turn would affect the reproducible characteristics of the 2DEG. This is the real driving force for ever-increasing demands on the measurement accuracy of in-situ temperature measurement in such processes.This is relatively uncritical for the remaining two layers in the stack: the AlN intermediate layer 102 is only a few monolayers thick and is intended to ensure that the unavoidable statistical compositional fluctuations in the AlGaN barrier layer do not lead to potential inhomogeneities at the i-GaN / AlGaN interface, which in turn would reduce the mobility of the 2DEG. A p-GaN layer 108 is required as gate material to interrupt the 2DEG locally beneath the gate electrode—thus, the transistor is in the desired off-state without applying a voltage to the metal contact 109.

[0077] Fig. Figure 2c utilizes a special feature of this transistor structure to visualize the signal fluctuations in the planetary reactor, which otherwise remain invisible under the doping of the FPOs. Since the Fe-GaN layer 106 and the i-GaN channel layer 107 have very different doping levels and growth rates, a growth pause is inserted between them during the MOCVD process to adjust the process parameters accordingly. During this process pause, the FPOs also "rest" in the in-situ reflection signal, since the growth rate is zero. This was used, as explained above, to detect the different R p -to display measurement signal fluctuations in the central wafer zone 14a and in a wafer edge zone 14c.

[0078] The Fig. 3 explains the apparatus and method for emissivity-corrected wafer temperature measurement.

[0079] The Fig. Figure 3a shows a schematic representation of an embodiment of an apparatus according to the invention for carrying out a method according to the invention for residual oscillation-free, emissivity-corrected in-situ pyrometry on a wafer 1 during a coating process. This arrangement completely solves, in particular, the problem of temperature residual oscillations for total optical thicknesses n d (average refractive index × total layer thickness) in (preferably) layer stacks with total layer thicknesses < 20 µm.

[0080] The device comprises • a light source 47 for irradiating the sample radiations A0 and B0 onto the surface of the wafer 1, wherein the spectral intensity distribution I0(λ) of the sample radiation A0 is within the bandpass range Δλ P is adjustable so that for at least one temperature in the relevant temperature range for the effective reflection wavelength λ R applies: λ P =λ R ; • a spectral filter unit (45) with a bandpass characteristic for filtering the sample radiation (A and B) reflected from the wafer 1 and a thermal radiation (E) emanating from the wafer 1 • a detector unit 40 for measuring the intensities of the filtered radiations (A, B and E) • a control unit (50) configured to measure both the reflection R p adjust the spectral intensity distribution I0(λ) of the sample radiation (A0) so that λ R =λ p for at least one temperature in the temperature range relevant for the MOCVD process, as well as to separate and synchronize the reflection measurement and the emission measurement via the light modulator 46, as well as to synchronize all measurements with the process steps of the MOCVD system and to forward the resulting measurement results to them. • and an evaluation unit (60) designed to determine the temperature (T w ) of the wafer by means of the method according to one of the preceding claims, as well as the temperature calibration of the pyrometer based on the reflection signatures R c a calibration sample.

[0081] Fig. Figure 3b schematically shows the sequence of the method according to the invention. Reference numeral 31 represents the epitaxy process.

[0082] Reference numeral 32 represents the process sub-step with increased accuracy requirements for the temperature measurement (e.g. for a wafer temperature fine control of the satellites 2 under otherwise absolutely constant epitaxial conditions) - in the example process for a III-nitride power transistor, this is usually the Fe-GaN and / or the i-GaN growth step.

[0083] Reference numeral 32a represents a learning time segment within process sub-step 32 for determining the parameters α T and τ to minimize the TT residual oscillations; reference numeral 32b represents an autocorrection time segment within the process sub-step 32 with application of the parameters α T and τ based on an R generated in 32a p (t)-forward extrapolation. The temperature values ​​determined in real time at individual measurement points during this autocorrection time segment are suitable for use in satellite temperature fine-tuning control tasks.

[0084] While in the learning time segment the parameters α T and τ are determined by evaluating a longer measurement period (containing several Fabry-Perot oscillations), these previously determined parameters α T and τ in the auto-correction time segment 32b in cyclic repetition 38 applied to the most recent individual measurement point.

[0085] Reference numeral 33 represents, as a sub-segment of 32a, a virtual interface fit (WG Breiland and KP Killeen; Journal of Applied Physics 78, 6726 (1995)) to the measured R p -FPOs, where for the reflection measurement R exp (t)=R p (t) of the growing GaN layer, a fit curve is generated, the growth rate r is determined, and the fluctuation deviations of all measurement points are determined. This virtual-interface fit (VI-Fit) utilizes the known optical properties n and k of the growing layer (here, GaN at approximately 1000°C) and replaces the properties of all underlying, previously grown layers with just two auxiliary (fit) parameters of a virtual interface.

[0086] Reference numeral 34 represents the determination of α T and τ using the 2-parameter residual oscillation minimization in the ECP wafer temperature T w (t). Used for this iterative determination of α Tand τ, the fit curve R previously obtained in step 33 fit (t). The τ-phase shift in R fit (t+τ) is realized by an additional, extremely thin (-1 nm ... 1 nm) GaN layer in the VI model. The thickness Δ (in nm) and the growth rate r (in nm / s) yield τ= Δ / r. The goal of this numerical parameter optimization for α T and τ is a linear, residual oscillation-free temperature profile resulting from the application of Eq.(7). In its term α T R exp (t + τ) inserted: R exp (t+τ) = R fit (t+τ) + δR(t), where the second summand δR(t)=R exp (t)-R fit (t) represents the transfer of reflection fluctuations from the original reflection measurement to the FPO phase-shifted measurement. For the α T and τ parameter optimization with the aim of minimal FPO residual oscillations in the temperature signal T w(t), common, well-known numerical minimization algorithms are used. The result is an α τ R exp (t + τ) curve and residual oscillation-free temperature measuring points T w (t) in the entire time segment 2a.

[0087] Reference numeral 35 represents an extrapolation of both R fit (t), ie, the VI-Fit curve, as well as the α T R fit (t + τ) curve, ie, the result of the (α,τ) optimization in the future time domain 32b.

[0088] Reference numeral 36 represents for each further measuring point in 32b the determination of the difference δR(t)={R exp (t)-R fit (t)}.

[0089] Reference numeral 37 represents the auto-correcting ECP (without further fitting process) by transferring this difference δR(t)={R exp (t)-R fit (t)} to the also extrapolated α T R fit (t + τ) curve, whereby α T R exp (t+τ) = α TR fit (t+τ) + α T {R exp (t)-R fit (t) using Eq.(7) the α-τ-corrected measured value T w (t). Each resulting temperature measurement point T w (t) is free of residual temperature (TT) oscillations and thus suitable for fine temperature control.

[0090] The necessity of the seemingly complicated additional effort to transfer the local measured value fluctuations δR to the extrapolated R fit (t+τ) curve results from the Fig. 1 described fluctuations in the R and E measured values. Only through this combination of extrapolated R fit (t+τ) curve with the δR(t) correction, a phase correction of individual R exp (t) measurement points are possible, which then ultimately completely minimizes the residual temperature oscillations according to equation (7).

[0091] The Fig. Figure 4 explains the process using the example of the growth of a relatively thick GaN layer under constant growth conditions and corresponds to a learning time segment 32a. Fig. 4a the result according to the state of the art (γ-correction), the Fig. 4b the inventive result after α-τ correction and the Fig. 4c) the result of an α-τ correction, in which the transfer of the reflection fluctuations to the α-τ-corrected reflection curve was omitted for testing purposes.

[0092] It is clear from Fig. 4a and Fig. 4b: while the γ-correction can only reduce the amplitude of the residual temperature oscillations to about ±0.5K, the α-τ-correction provides a virtually completely residual oscillation-free temperature profile (deviations <±0.1K). However, the remaining minimal temperature fluctuations may well be real variations in the wafer temperature (for example, due to small fluctuations in the gas flow of the satellite rotation). Fig. Figure 4c serves more as a didactic and illustrative example: if the δR(t) fluctuation transfer step is omitted from the α-τ correction, residual oscillation-free but "noisy" temperature curves are obtained. The reason is that the fluctuations remain in the raw emission signal, but have been eliminated in the reflection signal by the fitting process. This also further demonstrates that the fluctuations are not caused by noise in the measuring device, but rather—as described above—by the peculiarities of the planetary reactor.

[0093] Fig. Figure 5 provides an example of the complete procedure in a) and outlines the VI model used in b). As shown in a), the α-τ correction parameters are first determined in a learning time segment. These are then applied to individual measurement points in the autocorrection time segment using real-time correction methods. The same data set was used for this as in Fig. 4, except that the learning time segment 32a was reduced to the first part of the data set, so that the method could be tested in the autocorrection time segment 32b in the second part of the data set. In b), it is outlined how, for this VI fit, the optical properties of all underlying layers are combined in a virtual interface and how, by introducing an additional virtual layer 111 of thickness Δ directly at the virtual interface (VI), a phase shift τ is made effective. The result demonstrates that a forward extrapolation of the α-τ correction parameters determined in the learning time segment also leads to the complete minimization of the temperature residual oscillations in the autocorrection time segment. Furthermore, it was found that the method for fluctuation transfer to the extrapolated R fit (t+τ) curve is also suitable to fully compensate for smaller fluctuations and drifts in the growth rate.

[0094] Fig. Figure 6 shows an embodiment of the method according to the patent, in which (unlike in Fig. 5) the α-τ correction extends over three process steps, each with a different growth rate of the GaN layers. Process step 1: the growth of Fe-GaN, process step 2: a growth pause for adjusting the process parameters, and process step 3: the growth of i-GaN. The temperature control is to take place in process step 3, which, however, is too short (only about 1.5 FPOs) to implement the process as in the example of the Fig. 5 is only applicable within this process step. However, it is clear from the reflection measurement (λ p =950nm): The optical properties of Fe-GaN and i-GaN are identical, as they exhibit identical FPO amplitudes. The growth rates r Fe-GaN and r i-GaN are different, because the FPO period lengths are different (r Fe-GaN > r i-GaN). In addition, the growth rate during the growth pause is not zero, but slightly negative (some Fe-GaN is etched back, the FPO runs slightly backward).

[0095] In order to apply the procedure, two learning time segments were used: In learning time segment 1, VI-Fit is used to determine: r Fe-GaN , and, as auxiliary quantities, the two parameters of the virtual interface. By α-τ optimization to minimize the TT residual oscillations in this Fe-GaN step, we obtain: α T and Δ=r Fe-GaN / τ. In learning time segment 2, only the two growth rates are determined: r pause (etching rate) and r i- GaN This is achieved by simply extending the model to grow two additional GaN layers on the previously determined, unchanged virtual interface structure of the Fe-GaN layer. τ i-GaN is then calculated from τ i-GaN =r i-GaN / Δ. The R resulting from this multilayer model (VI / Δ / Fe-GaN / Pause / i-GaN) fit (t)- and α T R fit (t + τ) - curves are again calculated using r i-GaN and τ i-GaN extrapolated into the subsequent autocorrection time segment.

[0096] In the auto-correction time segment, the δR(t) values ​​are again calculated using the extrapolated R fit (t) to which extrapolated and R fit (t + τ) curve and then using Eq.(7) the α-τ-corrected T w (t) measurements are calculated.

[0097] The resulting temperature differences between the Fe-GaN and i-GaN process steps are real. The temperature fluctuations during the growth pause are also real (the process parameters are changed there according to the epitaxy recipe). Nevertheless, the resulting temperature in the auto-correction time segment is very stable (fluctuation <0.1 K). This example shows how important it is to assign an effective layer thickness difference Δ (in nm) to the phase shift τ (in seconds). Only in this way can the GaN growth rate r be determined for sub-steps with different GaN growth rates. i a consistent phase shift τ i = Δ / r iThe resulting temperature fluctuations, especially during the growth pause, are real, as the change in process conditions influences the wafer temperature. Thus, while the process conditions differ for Fe-GaN and i-GaN, the effective optical properties n and k of Fe-GaN and i-GaN are indistinguishable in the spectral range used here.

[0098] Fig. Figure 7 shows a calibration wafer for in-situ calibration of the pyrometer. A) shows a calibration wafer structure that is almost identical to a real GaN / Si power transistor, and b) shows the in-situ measurement data (temperature, R(λ p ) and R(λ C )), where the R(λ C =633nm) curve shows a characteristic, temperature-specific minimum at T cal =943°C. In c) the exact determination of this T cal -value is shown, using the first derivative of a polynomial fit to these measured data.

[0099] The Fig. The calibration wafer structure shown in Figure 7a is almost identical to a real GaN / Si power transistor ( Fig. 2a) and can therefore be manufactured by the user. The top layer of this calibration wafer is an AlGaN barrier layer 102. Its aluminum content ensures the thermal stability of the layer stack even at high temperatures, so that such wafers can be exposed to multiple temperature ramps between 800°C and 1050°C in various MOCVD reactors for calibration purposes without any noticeable changes in the layer structure.

[0100] Fig. Figure 7b shows that with a suitable choice of the layer thickness of the i-GaN channel layer 113, the reflection (here at λ c=633nm) passes through a characteristic minimum during the cooling process in the temperature range between 800°C and 1000°C, which is suitable for pyrometer calibration. In a temperature ramp between 800°C and 1000°C, this R(λ c ) minimum always at T cal=943°C, as it depends only on the layer thicknesses in the structure and on the thermal change in the refractive indices of the layer materials involved. The sensitivity of the method is based in particular on the large total thickness of the layer stack, preferably more than 1 micrometer, more preferably more than 2 micrometers, even more preferably more than 3 micrometers, and even more preferably more than 5 micrometers. The wavelength position of this last FPO minimum is specifically adjusted by the total optical layer thickness n(T)*d (n is the effective, average refractive index and d is the total thickness of the layer stack). The minimal change in the total stack thickness d due to thermal expansion can be neglected here due to the significantly greater dependence of the refractive index n(T) on temperature T. The very large total thickness d of the layer stack thus acts as an amplification factor for the effects caused by n(T). In Fig.5b shows that when the structure cools down, the last reflection minimum, which was previously passed “forward” (during AlGaN layer growth), is passed “backward” again, since n(T) becomes smaller with decreasing temperature and accordingly the optical layer thickness n(T)*d becomes smaller again.

[0101] For the purpose of absolute temperature calibration, during a 800°C→1050°C temperature ramp with the constant temperature gradient dT / dt Tw(t)=800°C+(dT / dt)*t checked at which temperature T w of the pyrometer this R(λ c ) minimum.

[0102] T w is done with the inventory calibration A cal measured 1Tw(t)=Acal−λpc2lns(t)(1−Rcal) where for R cal the reflection value R(k p ) of the calibration layer structure at T cal , i.e. at the minimum of the temperature FPO of the R(λ c ) reflection. If this R(λc )--Minimum e.g. at 951°C, i.e. at T w =(951+273)K=1224K, then with T cal =(942+273)K=1215K a correction value for A cal as follows: AcalNEW=Acal+[1Tcal−1Tw] List of reference symbols 1 wafer = semiconductor substrate with layer system on the surface 2 satellite plates with wafer holders, rotating in the gas stream 3 Rotation axis of the satellite dish for its satellite rotation 4 Main susceptor, mechanically rotating and holding the satellite plates 5 Rotation axis of the main susceptor for its main rotation 6 Heating (RF coils), heating the main susceptor and indirectly also satellite dishes 7 Housing of the MOCVD reactor 8 optical windows for measuring wafer temperature and wafer reflection 9 Detection optics of the ECP pyrometer 10 Ceiling (upper cover of the reaction chamber for the gaseous starting materials) E0 thermally generated radiation in the wafer, within the λ p Exit depth E thermally emitted radiation of the wafer A0 incident light, for measuring the reflection at the pyrometer wavelength λ P A reflected light of wavelength λ P B0 incident light, for measuring the reflection at the calibration wavelength λ c B reflected light of wavelength λ c 11 effective measuring zone measuring area of ​​a single measurement (single main rotation) 12 wafer sub-areas, statistically assigned to a measuring zone (many main rotations) 13 Measurement trajectory of the ECP pyrometer, generated by the main rotation 14 measuring zones on the measuring trajectory 13 of the ECP pyrometer, satellite-specific W iNumbering of the wafers or satellite plates in the reactor 101 Semiconductor substrate (here: silicon) 102 AlGaN barrier layer 103 AIN intermediate layer (few atomic monolayers) 104 very thick GaN / AlGaN buffer layer 105 AIN or AlGaN nucleation layer 106 Fe-GaN layer, high-resistance 106a Fe-GaN layer during growth, in the Virtual Interface (VI) model 107 i-GaN channel layer, undoped, particularly defect-free 108 GaN cap layer or p-GaN contact layer 109 metal contact of the transistor gate 110 2DEG - 2-dimensional electron gas, switchable under the transistor gate 111 virtual intermediate layer in the VI model to enable phase matching τ 31 Epitaxy process 32 sub-step of the epitaxy process that requires precision temperature measurement 32a Learning time segment (in 32) for determining the parameters α T and τ 32b Autocorrection time segment (in 32) with application of parameters α T and τ 33 Reflection FPO fit using the virtual interface method, R-fluctuation measurement 34 α T and τ determination by iterative minimization of the temperature residual oscillations 35 Forward extrapolation of the results from 33+34 36 for each measured value: R-fluctuation transfer to the R(α T ,τ)-forward extrapolation 37 ECP with α-τ real-time correction for each individual measured value 40 detector unit 44 beam splitters 45 optical filter unit 46 Light modulator, synchronized by the control unit 50 47 Light source for reflection measurements 50 Control unit (synchronization of measurement processes, including the epitaxy process) 60 Evaluation unit (generating the result data from the raw data)

Claims

[1] Method for residual oscillation-suppressed emissivity-corrected in-situ pyrometry, ECP, of wafers (1) during coating in coating systems, comprising the following steps: - Radiation of a sample radiation A0 of a pyrometer wavelength λ p and a sample radiation B0 of a calibration wavelength λ c onto the surface of a wafer during coating in a coating system; - Determining an intensity I R the sample radiation A reflected from the wafer R ; - Determining an intensity s E the thermal radiation E emitted by the wafer, - Determining an intensity I c from the wafer at a second wavelength λ c reflected sample radiation B c ; - Determine the reflection R p of the wafer at λ p from intensity I R and intensity I 0Rthe incident sample radiation A0, - Determine the current reflection R c of the wafer at λ c from intensity I c and intensity I 0C the irradiated sample radiation B0, - Determine the temperature T w of the wafer during coating with a layer using the emissivity-corrected pyrometry, ECP, method, - Determining a phase difference τ between reflection Fabry-Perot oscillations of the reflection R p , R p -FPOs, and the emission Fabry-Perot oscillations of the thermal radiation emitted by the wafer, E-FPOs, during the growth of a layer material on the wafer, - Determine an effective reflection scaling α=α T the reflection R p during the growth of the layer material on the wafer, - Determine the temperature T wof the wafer using the ECP method using the phase difference τ and the reflection scaling α. [2] Method according to claim 1, wherein the temperature T w of the wafer from a residual oscillation-free ECP-corrected thermal emission intensity s ECP is calculated. [3] Method according to one of the preceding claims, wherein the phase difference τ is obtained from the minimization of the values ​​of the R p -FPOs by ±90° phase-shifted residual oscillation components in the resulting ECP-corrected thermal emission intensity s ECP is determined. [4] Method according to one of the preceding claims, wherein the scaling parameter α is obtained from the minimization of the values ​​of the R p -FPOs in-phase or anti-phase phase-shifted residual oscillation components in the resulting ECP-corrected thermal emission intensity s ECP is determined. [5] Method according to claim 1, wherein the spectral intensity distribution I0(λ) of the sample radiation A0 is controllable such that the center wavelength λ R eff within a used optical filter bandpass λ=λ p ±Δλ p at least one temperature is identical to the center of gravity wavelength λ p eff the pyrometry measurement. [6] Method according to one of the preceding claims, wherein the determination of the temperature T w of the wafer using the ECP method is independent of at least one of the following errors or disturbance parameters of the measurement: - Error in scaling factor α R the reflection measurement - Offset error β of the reflection measurement - Scaling factor δ in the thermal emission signal - additional thermal radiation γ̃ - additional thermal radiation ξ. [7] Method according to one of the preceding claims, wherein the phase difference τ and the reflection scaling α are defined by the following equation (*): s(t)=p(1−αR(t+τ))e−c2 / λpTw where s(t) is the time t-dependent thermal radiation of the wafer, α is a scaling error of the reflection measurement, p is a scaling factor of the pyrometer to be calibrated using a suitable calibration wafer, and R is the sample radiation reflected from the wafer. Other parameters in (*) are the wafer temperature T w , the wavelength λ p and the second radiation constant c2. [8] Method according to one of the preceding claims, wherein the determination of the temperature is carried out by means of phase difference τ and reflection scaling α during the performance of a gas phase epitaxy under constant process conditions. [9] Method according to one of the preceding claims, wherein the determination of the optimal reflection scaling α is carried out by using α*R p (t) for the ECP correction from s(t) to s ECP (t). [10] Method according to one of the preceding claims, wherein the determination of the optimal phase difference τ is carried out by means of a virtual interface (VI) fit to the reflection R p (t) during coating with a known material with also known nk(T) properties. [11] Method according to one of the preceding claims, wherein the determination of the temperature T w of the wafer in a planetary reactor during the epitaxial growth of group III nitrides on foreign substrates. [12] Method according to one of the preceding claims, wherein the temperature Tw of the wafer (20, 22) to be determined is defined by the following equations (**) and (***) 1Tw(t)=Acal−λpc2lns(t)(1−αR(t+τ)) Acal=1Tcal+λpc2lnscal(1−Rcal) where A cal before applying the ECP method using a calibration wafer in the reactor, the temperature of which is T cal and its reflection R cal are exactly known at the time of this in-situ calibration. [13] Method according to claim 12, wherein the layer thickness of the calibration wafer is selected such that the reflection R c at the second wavelength λ c has a characteristic minimum or maximum in the temperature range between 800°C and 1000°C. [14] Method according to claim 13, wherein the characteristic minimum or maximum a) is passed through during a T-ramp in this T-range and b) can be assigned to exactly one wafer temperature of the calibration wafer, since the associated temperature of this wafer in this minimum or maximum depends only on the optical properties of the layers. [15] Method for controlling the temperature T w a wafer (20, 22) in a planetary reactor during gas phase epitaxy for growing semiconductor layers of group III nitrides, wherein the temperature T w of the wafer is determined by means of the method according to one of the preceding claims. [16] Method according to claim 15, wherein the temperature T w of the wafer is already determined before a control time segment in a learning time segment, from which both the correction parameters α and τ or Δ are determined. [17] Method according to claim 16, wherein in addition a best-fit curve is fitted to the reflection FPOs measured in the learning time segment in R p (t) is determined. [18] Device for residual oscillation-suppressed emissivity-corrected in-situ pyrometry, ECP, of wafers (20, 22) in a coating system, comprising: - a means for irradiating a sample radiation A0 of a pyrometer wavelength λ p and a sample radiation B0 of a calibration wavelength λ c onto a surface layer of a wafer (1); - an optical filter unit (45) each having a bandpass characteristic for filtering the sample radiation A and B reflected from the wafer (1) and the thermal radiation E emanating from the wafer (1), - a detector unit (40) designed to: Determining an intensity I R the sample radiation A reflected from the wafer (1); Determining an intensity s E the thermal radiation E emitted by the wafer, Determining an intensity I c from the wafer (1) at a second wavelength λ c reflected sample radiation B; - an evaluation unit (60) arranged to determine the temperature T from the determined intensities wof the wafer by means of the method according to one of the preceding claims. [19] Apparatus according to claim 18, further comprising a control unit (50) configured to adjust the spectral intensity distribution I0(λ) of the sample radiation A0 to set the effective effective wavelength.

Images