Reflectance-difference spectroscopy of mixed arsenic-rich phases of gallium arsenide „001…M. J. Begarney,1L. Li,2C. H. Li,1D. C. Law,1Q. Fu,1and R. F. Hicks1,*1Chemical Engineering Department, University of California, Los Angeles, California 900952Department of Physics and Laboratory for Surface Study, University of Wisconsin, Milwaukee, Wisconsin 53201共Received 12 May 2000兲The relationship between the reflectance difference spectra and the atomic structure of arsenic-rich recon-structions of GaAs共001兲 has been investigated. Scanning tunneling micrographs reveal that a rougheningprocess occurs as the surface structure changes with decreasing arsenic coverage from 1.75 to 0.75 monolayers共ML兲. At 1.65 ML As, small pits, one bilayer in depth and having the same c(4⫻ 4) reconstruction as the toplayer, form in the terraces. At the same time, gallium atoms are liberated to the surface, disrupting the c(4⫻ 4) ordering. At about 1.4 ML As, (2⫻ 4) domains nucleate and grow on top of the c(4⫻ 4). Furtherdesorption of arsenic causes the underlying layer to gradually decompose into a metastable (2⫻ n) phase (n⫽ 2, 3, or 4兲, and finally into the (2⫻ 4). In the reflectance difference spectra, negative peaks at 2.25 and 2.8eV correlate with the c(4⫻ 4)-type arsenic dimers. However, the intensity of the latter feature strongly de-pends on the presence of adsorbates, such as alkyl groups and gallium adatoms. By contrast, the intensity of thepositive peak at 2.9 eV is directly proportional to the density of (2⫻ 4)-type dimers.I. INTRODUCTIONReflectance difference spectroscopy 共RDS兲 is an effectivein situ probe of the surface reconstructions of compoundsemiconductors during growth by molecular-beam epitaxy,chemical beam epitaxy, and metalorganic vapor-phaseepitaxy.1–4The technique determines the relative differencein the near-normal reflectance of light polarized along thetwo principle axes of the surface, and since the bulk crystalsare isotropic, the spectra quantify the optical anisotropy ofthe first few atomic layers. The reconstructions of galliumarsenide 共001兲 have been widely studied using thistechnique.5–7While the line shapes of the c(4⫻ 4), (2⫻ 4)/c(2⫻ 8), and (4⫻2)/c(8⫻ 2) surfaces have been cor-related with electron diffraction patterns 关reflection high-energy electron diffraction and low-energy electrondiffraction兴,1the physical origin of the reflectance anisotropyhas not been conclusively determined. Uncertainty remainsas to whether the reflectance anisotropy is a result of bulk-surface transitions,8–12or by transitions among the molecularorbitals of the surface dimers.13,14Early efforts to account for the reflectance anisotropy in-volved the calculation of the surface dielectric function ofsimplified GaAs surfaces.13,14In these studies, (2⫻ 1) and(1⫻ 2) dimerized surfaces were used to approximate thearsenic-rich (2⫻ 4) and gallium-rich (4⫻ 2) reconstructions.The authors identified transitions within the dimer structuresas the source of the RDS spectra, and obtained rough quali-tative agreement with the experimental data. Based on morerecent, first-principles calculations, other researchers haveconcluded that transitions between bulk valence states andunoccupied surface states are primarily responsible for thereflectance anisotropy.9–12While these results more closelymatch the RDS data, discrepancies in the energies and mag-nitudes of spectral features remain. The best agreement be-tween theory and experiment has been achieved throughtight-binding calculations of the dimer structures.8The influ-ence of both discrete dimer structures and surface-modifiedbulk wave functions were cited as contributing factors to theobserved anisotropy.In this paper, we report on the reflectance difference spec-tra of a series of gallium arsenide 共001兲 reconstructions atarsenic coverages ranging from 1.75 to 0.75 monolayers共ML兲. The main structural features on these surfaces are ar-senic dimers, which are bonded to either a sublayer of As orGa atoms. These two kinds of arsenic dimers are referred toas ‘‘c(4⫻ 4)-type’’ and ‘‘共2⫻4兲-type,’’ respectively. To aidin the interpretation of the spectra, a direct comparison hasbeen made between the RDS line shapes and the atomicstructures as seen by scanning tunneling microscopy 共STM兲.We have found that surface roughening occurs as a result ofthe nucleation and growth of the (2⫻ 4) phase on top of thec(4⫻ 4) phase. This process is accompanied by gallium outdiffusion into the c(4⫻ 4) layer. Furthermore, we have dis-covered that the reflectance anisotropy is affected not onlyby the types of arsenic dimers, but also by the presence oftransitional structures, such as a new (2⫻ n) phase, and byadsorbates, including alkyl species and gallium adatoms.II. EXPERIMENTAL METHODSGallium arsenide films, approximately 1␮m in thickness,were grown on nominally flat GaAs 共001兲 substrates in ahorizontal flow metalorganic vapor-phase epitaxy 共MOVPE兲reactor. The substrates were doped n-type with 1⫻ 1017Si atoms/cm3. The wafer temperature during growthwas 550⫾25 °C, and the organometallic reagents, triisobu-tylgallium 共TIBGa兲 and tertiarybutylarsine 共TBAs兲, wereused at concentrations of 5 and 50 ppm, respectively.Palladium-diffused hydrogen was the carrier gas, and the to-tal reactor pressure was 20 torr. The wafers were cooledimmediately following growth with the TBAs and H2sup-plies maintained until room temperature was reached. Thisensured that the surface would have the maximum arseniccoverage possible. The samples were transferred to an ultra-high vacuum cluster tool via a turbo-pumped interface cham-PHYSICAL REVIEW B 15 SEPTEMBER 2000-IIVOLUME 62, NUMBER 12PRB 620163-1829/2000/62共12兲/8092共6兲/$15.00 8092 ©2000 The American Physical Societyber. Details of this system have been described previously.15A commercially available ISA/J-Y Nisel reflectance dif-ference spectrometer was used for our measurements of theGaAs surfaces. The spectral range varied from 1.5 to 5.2 eV,with increments of 0.025 eV and integration times of 1000ms. Three scans were recorded consecutively and then aver-aged to minimize the noise. Real-space images of the recon-structions were obtained using a Park Autoprobe/VP scan-ning tunneling microscope. Tunneling was out of filled stateswith a sample bias of ⫺3.0 to ⫺4.0 V and with a tunnelingcurrent of 1 nA.To