 |
Figure 1 Schematic of normal incidence reflectance experiment and screen print of ADVISOR analysis software. |
W. G. Breiland, L. A. Bruskas, A. A. Allerman, T. W. Hargett, J. F. Klem, and M. M.
Bridges
wgbreil@sandia.gov
Dept 1126 MS 0601
Sandia National Laboratories,
Albuquerque, NM 87185-0601
Epitaxial growth of compound semiconductors is accomplished with Metal Organic Chemical Vapor Deposition (MOCVD) or Molecular Beam Epitaxy (MBE). In both these methods, many process variables must be precisely controlled to achieve the proper combination of layer thickness, chemical composition, lattice matching, and doping levels required by modern compound semiconductor devices. We have found that a simple and robust in situ monitor, normal incidence reflectance, has proven to be an indispensable tool for achieving stringent device requirements for epitaxial growth. Although this probe has been recently been used in real-time control of layer thickness, it has also been extremely useful as a pre-growth calibration tool. These pre-growth calibration techniques may be implemented easily with commercial deposition systems that are incapable of real-time control.
The schematic on the left of Fig. 1 shows the reflectance experiment. Light is launched at the wafer and detected at normal incidence using a beam splitter arrangement. Interference of light between interfaces of the thin films cause oscillations in the reflectance signal as the films are grown. The monitor requires only one window, polarization is not a factor, and the effects of wafer wobble can be minimized. We have found that light from a 5 W tungsten-halogen lamp provides an extremely stable (1 part in 10,000) source. A 10 nm bandwidth interference filter is used to isolate a single wavelength. The detector is typically a silicon PIN photodiode using DC signal averaging.
The reflectance monitor may be used in either of two modes. It can be used as a real-time in situ fault detector during every growth run. An experienced grower can easily determine that some problem has occurred during the run by glancing at the reflectance signal waveform and comparing it to a previous successful run. The real-time nature of the monitor allows one to pinpoint the source of the fault (eg. a gallium source going dry on the fourth layer). It is also possible to use the monitor as an absolute reflectance monitor to extract quantitative information about growth rates and chemical composition. This latter use of the probe has proven to be a very powerful tool for improving device yield and minimizing calibration runs.
The image on the right of Fig. 1 shows the interference data from a pre-growth calibration run. These data are analyzed using the ADVISOR (Analysis of Deposition Using Virtual Interfaces and Spectroscopic Optical Reflectance) method [1]. Growth rates for each bubbler source and for several doping conditions are determined to better than 1% accuracy. These numbers are fed to a model-based automated recipe generator that is used by the MOCVD tool to grow a device structure.
|
Figure 1 Schematic of normal incidence reflectance experiment and screen print of ADVISOR analysis software. |
Results of the ADVISOR approach are shown in Fig 2. The left plot shows the reproducibility of the cavity wavelength values for 770 and 845 nm vertical cavity surface emitting laser (VCSEL) structures [2]. These devices are very complex, consisting of nearly 700 layers with composition grading in the distributed Bragg reflector (DBR) mirrors, quantum wells, and a single-wavelength cavity that must be the correct thickness. These devices can be grown with very high yield by performing a single 1 h calibration run, such as is shown in Fig. 1. This calibration is typically only performed for a week’s worth of device fabrication. Note that a 770 laser structure was grown as a special run in the middle of a batch of 845 nm lasers. The model-based recipe generation allows one to be agile in device fabrication by using the same set of pre-growth calibration values to generate a variety of device structures.
The right graph in Fig. 2 shows the results of real-time process control during the MBE growth of a complicated reflectance modulator structure [3]. ADVISOR analysis was done on early layers to modify the deposition rates. Squares represent cavity wavelengths that would occur without correction. Circles represent the measured cavity wavelength with ADVISOR control. Note the large increase in device yield that was achieved with the in situ monitor.
 |
Figure 2 Measures of device structure accuracy resulting from ADVISOR analysis and process control. |
In our most recent application of normal incidence reflectance, real-time ADVISOR/MBE control was used to determine the thickness of most layers of the first reported room temperature CW 1.3 micron VCSEL. [4].
Normal incidence reflectance may also be used to determine surface temperatures during growth. A single- wavelength pyrometer will exhibit large (up to 100 °C) artifacts in temperature if it is not corrected for the change in emissivity that occurs during thin film growth. For smooth, flat, opaque surfaces, the emissivity is simply (1 - reflectance). An instrument that simultaneously records thermal emission and normal incidence reflectance may thus be used to perform emissivity-correcting pyrometry during thin film growth. Fig 3 shows a plot of reflectance at 550 nm (left scale, smooth data) superimposed on the 900 nm emissivity-corrected surface temperature (right scale, noisy data) [5]. Subtle temperature effects may be observed during the growth. For example, a pause was required between the GaAsP and GaAs layers. The figure reveals that the reason for this pause was to allow the system to recover to the 530 °C growth temperature. The figure also illustrates the high signal-to-noise that may be achieved with reflectance waveforms. The layer labeled “GaAsSb” is a 77 Å quantum well between 120 Å GaAs barriers.
 |
Figure 3 Reflectance (left axis) and surface temperature (right axis) during growth.. |
Reflectance has also been used to monitor plasma etching on patterned wafers. Unlike many silicon-based etching processes, selective etch recipes are rare. It is thus useful to watch the time-reversed interference pattern evolve during an etch process in order to stop the etch at the correct interface or within the correct layer. Complications due to photoresist etching may be dealt with using digital filtering methods.
1. W.G. Breiland and K.P. Killeen, J. Appl. Phys. 78 (1995) 6726.
2. W. G. Breiland, H. Q. Hou, H. C. Chui, B. E. Hammons, J. Cryst. Growth, 174 (1997) 564
3. W. G. Breiland, H. Q. Hou, B. E. Hammons, and J. F. Klem, XXVIII SOTAPOCS Symposium, Electrochem. Soc. 1998.
4. K. D. Choquette, J. F. Klem, A. J. Fischer, O. Blum, A. A. Allerman, I. J. Fritz, S. R. Kurtz, W. G. Breiland, R. Sieg, K. M. Geib, J. W. Scott, and R. L. Naone, Electron. Lett. to be published.
5. A. A. Allerman, T. W. Hargett, and W. G. Breiland, unpublished, (2000).