RAPID COMMUNICATIONS PHYSICAL REVIEW B VOLUME 57, NUMBER 16 15 APRIL 1998-II Al/ZnSe 100... Schottky-barrier height versus initial ZnSe surface reconstruction M. Lazzarino, G. Scarel, S. Rubini, G. Bratina, L. Sorba,* and A. Franciosi Laboratorio Nazionale TASC-INFM, Padriciano 99, I-34012 Trieste, Italy and Department of Chemical Engineering and Materials Science, University of Minnesota, Minneapolis, Minnesota 55455 C. Berthod, N. Binggeli, and A. Baldereschi Institut de Physique Applique´e, Ecole Polytechnique Fe´de´rale de Lausanne, CH-1015 Lausanne, Switzerland Received 22 September 1997 Al/ZnSe 100 Schottky barriers fabricated on c(2 2), 2 1, and 1 1 reconstructed surfaces were studied by means of photoemission spectroscopy and first-principles calculations. Relatively similar values of the Schottky barriers were found for interfaces fabricated on Zn-stabilized c(2 2) and Se-dimerized 2 1surfaces, while substantially lower values of the p-type barriers were predicted theoretically and observed experimentally for junctions grown on the Se-rich 1 1 surface. S0163-1829 98 51716-X The possibility of tuning the Schottky-barrier height has In Fig. 1 we show SRPES results for the Se 3d and Zn 3d been attracting attention since the inception of the study of core-level emission at a photon energy of 120 eV from sur- metal/semiconductor junctions.1 Recently, renewed interest faces exhibiting the three types of RHEED patterns repro- has been stimulated by the contact problems that plague ducibly obtained in sequence upon annealing. After anneal- wide-band-gap semiconductors such as ZnSe and GaN.2­7 ing at 260 °C, the Se-rich 1 1 surface gives rise to a We report experimental and theoretical studies of the relatively broad Se 3d line shape which reflects a high Schottky barrier at Al/ZnSe 100 interfaces showing that the binding-energy contribution associated with Se-Se coordina- barrier height is strongly dependent on the initial composi- tion at the surface,9 and a Zn 3d/Se 3d integrated intensity tion of the semiconductor surface. The observed experimen- ratio R 0.81 0.03. Following annealing at 330 °C, the tal trend is that while the c(2 2) and 2 1 surface termi- Se-stabilized 2 1 reconstruction gives rise to a sharper Se nations correspond to similar barrier heights, a 0.25 0.05 3d line shape, and a Zn 3d/Se 3d integrated intensity ratio decrease in the p-type barrier is found when the junction is R 0.95 0.03. This reconstruction is also observed during fabricated on the Se-rich 1 1 surface termination. The the- ZnSe MBE in Se-rich growth conditions, and is believed to oretical trend in the barrier is compellingly similar, i.e., junc- correspond to a surface terminated by a fully dimerized tions corresponding to the c(2 2) and 2 1 terminations monolayer of Se.9 For higher annealing temperatures are predicted to have relatively similar barrier heights ( 430 °C , the well-known c(2 2) reconstruction9­11 is within 0.05 eV , while those obtained for the Se-rich 1 1 seen to correspond to a single Se 3d doublet, a clearly de- surface are expected to have a lower p-type barrier, and spe- fined high binding-energy shoulder in the Zn 3d line shape, cifically lower by 0.45 0.20 eV for the measured excess Se and R 1.21 0.03. This reconstruction is also observed dur- coverage of 0.41 0.18 ML of the 1 1 relative to the 2 1 reconstruction. Our first-principles calculations for model interface configurations explain our experimental re- sults in terms of a variable Se-induced local interface dipole. ZnSe epilayers 500 nm thick were grown by molecular- beam epitaxy MBE on GaAs 100 .8 All epilayers were Cl doped (n 1 ­ 3 1018 cm 3). A thick Se cap layer was used to protect the samples during transfer in air to the photoelec- tron spectrometer. The Se cap layer was thermally desorbed in situ, and different surface reconstructions-as determined by reflection high-energy electron diffraction RHEED - were obtained by varying the annealing conditions. Al over- layers 2­3 nm thick were evaporated in situ on ZnSe sub- strates kept at room temperature, with thickness determined using a quartz thickness monitor. Surfaces and interfaces were examined after quenching to room temperature by monochromatic x-ray photoemission spectroscopy XPS us- ing Al K radiation 1486.6 eV , an overall energy resolu- FIG. 1. Synchrotron radiation photoemission SRPES results tion electron plus photons of 0.8 eV, and an effective for the Se 3d and Zn 3d core-level emission from the Se-rich 1 photoelectron escape depth of 1.5 nm, or by soft-x-ray 1 bottom-most spectra , Se-dimerized 2 1 midsection , and synchrotron radiation photoemission spectroscopy SRPES Zn-stabilized c(2 2) topmost spectra reconstructions of the at the Synchrotron Radiation Center of the University of ZnSe 100 surface. The corresponding values of the Zn 3d/Se 3d Wisconsin­Madison, with an energy resolution of 0.2 eV. integrated intensity ratio R are 0.81, 0.95, and 1.21, respectively. 0163-1829/98/57 16 /9431 4 /$15.00 57 R9431 © 1998 The American Physical Society RAPID COMMUNICATIONS R9432 M. LAZZARINO et al. 57 ing ZnSe MBE in Zn-rich growth conditions, and is believed to correspond to a surface terminated by half a monolayer of Zn atoms on a complete ML of Se, i.e., to an ordered array of Zn vacancies within the outermost layer of Zn atoms.9­11 The shoulder on the low kinetic-energy side of the main bulk-related Zn 3d feature, in particular, is believed to be associated with such Zn surface atoms.9 Recent total-energy calculations have shown the c(2 2) and 2 1 reconstructions to be the lowest energy con- figurations among those examined for Zn-rich and Se-rich surfaces, respectively.12 In fact, the Se-rich 1 1 reconstruc- tion is not observed during ZnSe MBE, and has only been reported during desorption of a Se cap layer. Kahn and co- workers reported a Se-rich 1 1 reconstruction for an an- nealing temperature of 200 °C, and tentatively associated this reconstruction with the presence of 2­3 ML of excess Se on the surface.9 Lopinsky et al. studied the same surface and reported strong similarities between the corresponding elec- tronic states and those observed upon deposition of a single amorphous monolayer of Se onto an unreconstructed, Se- terminated ZnSe surface.13 The R values in Fig. 1 were used to model the surface composition under three main simplifying assumptions: i a single photoelectron escape depth -independent of the sur- FIG. 2. X-ray photoemission spectroscopy results for the Al 2p left face termination-exists for the bulk and surface regions; ii and Zn 3d center core-level emission, together with the corresponding the photoemission intensity can be expressed as the sum of position of the Fermi level EF in the gap for Al/ZnSe 100 junctions fabri- the emission intensities from discrete 100 atomic planes; cated by depositing 3-nm-thick Al overlayers on c(2 2), 2 1, and 1 1 iii the emission from a given plane at a depth d from the surface reconstructions three topmost sections, top to bottom, respectively . For comparison we also show bottom-most spectra SRPES results for a surface is attenuated by e d/ . The measured R values are junction fabricated on the Se-rich 1 1 reconstruction. The low binding- then found to be consistent with an excess Se coverage x of energy 3d doublet dashed line reflects Zn atoms within the metallic over- about half a monolayer for the 1 1 surface used in this layer. The high binding-energy doublet dot-dashed line derives from the study as compared to the 2 1 surface14 semiconductor substrate. The position of the bulk Zn 3d emission is used to Prior to metal deposition, the positions of the Zn 3d and derive the position of the valence-band maximum EV and the conduction- band minimum E Se 3d centroids relative to the valence-band maximum E C relative to the Fermi level EF at the interface right-most V section . were determined using a least-squares fit and a linear ex- trapolation of the leading edge of the valence band. Upon centroid was found 9.13 0.03 eV below EV . After metal deposition of 2­3 nm of Al, new measurements of the core- deposition, the Zn 3d line shape shows two contributions. A level positions and the known position of the spectrometer low binding-energy doublet develops during the early stages Fermi level EF were used to infer the p-type Schottky barrier of interface formation, and reflects Zn atoms displaced from p EF EV and the n-type barrier n EC EF , with the semiconductor and segregated in the metallic overlayer. EC EV 2.70 eV at room temperature. The procedure is This reacted component is visible as a shoulder in the XPS illustrated in Fig. 2. Prior to metal deposition, the centroid of results, and is emphasized in SRPES due to the higher sur- the Zn 3d core doublet was found 9.16 0.04 eV below EV face sensitivity. The high binding-energy doublet derives for all surfaces. Upon metal deposition, the Al 2p core-level from the ZnSe substrate, and can be used to obtain n emission appears at the position expected for elemental me- 0.78 0.04 eV, in good agreement with the XPS result. tallic Al for all interfaces examined. The position of the Zn Previous determinations of the n-type Schottky barrier for 3d centroid varies instead for the three interfaces, and can be Al/ZnSe junctions focused on the c(2 2) reconstruction used to infer the position of EV relative to EF . and yielded values of 0.55 0.10 eV Ref. 15 and 0.58 From the results in Fig. 2 we determined n 0.55 0.10 eV Ref. 10 , consistent with those reported here for 0.06 eV ( p 2.15 0.06), n 0.59 0.06 ( p 2.11 the same reconstruction. The 0.24 eV increase in n de- 0.06), and n 0.79 0.06 eV ( p 1.91 0.06) for in- crease in p for interfaces fabricated on the Se-rich 1 1 terfaces fabricated on c(2 2), 2 1, and 1 1 reconstruc- reconstruction in Fig. 2 is among the highest barrier changes tions, respectively. The quoted experimental uncertainty ap- versus semiconductor reconstruction which have been re- plies to each individual value of the barrier, but the ported to date.1 To investigate the microscopic mechanisms uncertainty on the barrier height variation in the series, that that may account for such a large change in the Schottky depends only on the determination of core-level shifts, is barrier, we performed first-principles calculations of the band comparatively smaller typically 0.03 eV . Interface reac- alignment for a series of model interface configurations. tions do not affect the Schottky-barrier determination in Fig. As in recent studies of Al/GaAs 100 junctions,6 the cal- 2. This is demonstrated by the SRPES results for a junction culations were performed within the local-density approxi- fabricated on a 1 1 surface in the bottom-most section of mation to density-functional theory DFT , using the pseudo- Fig. 2. Prior to metal deposition, the position of the Zn 3d potential plane-wave method.16 To model the isolated Al/ RAPID COMMUNICATIONS 57 Al/ZnSe 100 SCHOTTKY-BARRIER HEIGHT VERSUS . . . R9433 For convenience the Schottky barrier was decomposed as p Ep V. The band-structure term Ep is the differ- ence between the Fermi energy of the metal and the valence- band edge of the semiconductor, EV , in the bulk, each mea- sured relative to the average electrostatic potential of the corresponding crystal.6 This term does not depend on the interface, and was determined from standard bulk band- structure calculations for ZnSe and Al. The second term V is the electrostatic potential-energy lineup across the inter- face, and contains all interface-specific features. This is the only term in p which may change in the presence of inter- facial perturbations. The potential energy lineup V was derived via Poisson's equation from the self-consistent supercell charge density us- ing the macroscopic average technique.6 The supercell cal- culations were performed with a plane-wave kinetic-energy cutoff of 20 Ry. The other computational details are as in Ref. 6. In Fig. 3 we show the macroscopic average of the potential energy V and the potential lineup across the relaxed junctions. Since an increase in V corresponds to an identi- cal increase in EC EF decrease in EF EV , the calcula- tions predict a 0.05 eV increase in n decrease in p in going from the relaxed configuration A to the relaxed con- FIG. 3. Right: Starting interface configurations employed in the super- figuration B, and a further increase of n decrease in p cell calculations. Configuration A involves Al atoms positioned at the Zn by 0.56 eV in going from the relaxed configuration B to the vacancy sites of the c(2 2) surface, below the Al fcc lattice rotated 45° relaxed configuration C. The direction and order of magni- about the 100 axis relative to ZnSe to satisfy the epitaxial relation. Con- tude of the predicted shifts are consistent with those ob- figuration B involves a ZnSe surface terminated by a full Se monolayer below the fcc metal. Configuration C involves a ZnSe surface terminated by served experimentally see Fig. 2 , suggesting that although a 50% Se­50% Al atomic layer on top of a full Se monolayer. Left: Mac- the model configurations employed may not describe the de- roscopic average of the electrostatic potential energy V and potential energy tail of the actual atomic reconstructions,18 they do capture lineup V across the relaxed junctions. Relaxation is graphically illustrated the basic electrostatic trend as a function of interface com- at the bottom for each atomic plane. Double atomic symbols denote in- position. equivalent relaxation at different sites. Since an increase in V corresponds We emphasize that atomic relaxation at the Al/ZnSe 100 to an identical decrease in EF EV , the calculations predict a 0.05 eV decrease in interfaces is substantial, and has an important effect of the p in going from the relaxed configuration A to the relaxed configuration B, and a further decrease by 0.56 eV in going from the relaxed order of 0.5­1 eV on the Schottky-barrier height, especially configuration B to the relaxed configuration C. The calculated values of p for the Se-rich configuration C. In this case, from the initial are also shown. configuration C we found that the Se­Al ZnSe 001 interfaces, we used supercells consisting of 13 0.5Se0.5 interplanar spacing at the interface increased by 40% after convergence, layers of ZnSe and 7 layers of Al. In view of the good lattice and became comparable to the Al matching between ZnSe and GaAs, many of the structural 0.5Zn0.5­Al and Se­Al in- terplanar distances of configurations A and B, respectively. considerations put forth in Refs. 6 also apply to the present case. In particular, the Al 100 direction was made parallel This large relaxation in configuration C reflects the increased to the ZnSe 100 growth axis, and the Al fcc lattice was metallic character of the bonds between the Se and the rotated 45° about the 100 axis relative to ZnSe cubic lattice Al0.5Se0.5 atomic layer. in order to satisfy epitaxial relations. The Al overlayer was To calculate p , the LDA band-structure term EV and tetragonally elongated 4% following macroscopic elasticity therefore Ep should be corrected to take into account theory, and the local atomic structure at the interface was many-body and relativistic effects. As a ground-state prop- fully relaxed. erty, V is instead accurately determined within DFT. The The starting interface configurations17 prior to atomic re- spin-orbit correction on Ep was derived from experiment, laxation are schematically illustrated on the right-hand side and amounts to 0.15 eV. The many-body corrections to the of Fig. 3. We selected simple configurations corresponding band structure of ZnSe have been evaluated in Ref. 19. As to ideal continuations of the semiconductor bulk while taking the LDA band-gap values in our calculations and in Ref. 19 into account the initial composition of the starting surface. are different, due to the different pseudopotentials employed, For Al/ZnSe fabricated on the c(2 2) surface we posi- we used the valence band-edge correction of Ref. 19 and tioned Al atoms at the Zn vacancy sites of the outermost scaled it by the ratio of the difference between the GW band semiconductor layer configuration A in Fig. 3 . For Al/ZnSe gap19 and the LDA band gaps in the two calculations. The fabricated on the 2 1 surface we terminated the semicon- resulting estimate for the many-body correction on Ep was ductor with a full layer of Se atoms at the ideal bulk posi- 0.50 eV. Using these two corrections, we obtain p tions configuration B . For Al/ZnSe fabricated on the 1 2.00, 1.95, and 1.39 eV for the relaxed configurations A, 1 surface, we used a virtual crystal approach to terminate B, and C, respectively. We emphasize that while the varia- the semiconductor with a 50% Se-50% Al atomic layer con- tion of the calculated Schottky barrier with interface compo- figuration C . sition is independent on the magnitude of the self-energy RAPID COMMUNICATIONS R9434 M. LAZZARINO et al. 57 roughly proportional to the Se excess coverage x on the Se- terminated semiconductor surface for 0 x 1. A 0.95 eV decrease in p is obtained for x 1 see Fig. 4 , and a 0.56 eV decrease is calculated for x 0.5 Fig. 3 . Therefore a 0.45 0.20 eV decrease in the barrier is expected for x 0.41 0.18 Ref. 14 . Recently, Chen et al. reported a 0.25-eV reduction in the p-type barrier for Au/ZnSe 100 junctions by introducing a 2­3 ML Se interlayer between the metal and the semicon- ductor, and proposed an electronegativity-based interpreta- tion of the barrier reduction.15 The similar 0.25-eV barrier reduction in the presence of vastly different electronegativity variations X ( X 0.94 for Al-Se versus 0.01 for Au-Se in Pauling's scale , and the expected saturation of the dipole in Fig. 4 for x 1, call into question the general applicability of a simple electronegativity-based approach. Our results sug- FIG. 4. Difference in the charge distribution top and average potential gest that lattice relaxation plays an important role in deter- energy bottom calculated between an interface terminated with two full Se mining the Schottky barrier in these systems and should be monolayers, and a single Se monolayer. The latter corresponds to the re- laxed configuration B, while the former corresponds to the limiting case of taken into account to improve upon electronegativity-based a type-C configuration for x 1. estimates of the interface dipole. correction, the absolute value of the barrier depends on the In summary, experiment points to an important effect of magnitude of such a correction, which was not calculated in the initial ZnSe surface composition on the Al/ZnSe 100 our work, but simply rescaled from that of Ref. 19. There is Schottky barrier. In particular, the presence of an excess Se therefore a substantial uncertainty ( 0.2 eV in the absolute coverage x at the 0.5 ML level on top of a fully Se- values of the theoretical barrier. terminated ZnSe surface gives rise to a local interface dipole The mechanism responsible for the large reduction in p that lowers substantially the p-type barrier. Theory shows for interfaces fabricated on the 1 1 surface is illustrated in that the dominant related charge transfer occurs from the first Fig. 4, where we plot the difference in the electronic charge Al monolayer to the excess Se atoms, and that the large distribution and in the electrostatic potential calculated be- relaxation of the Se-Se interatomic distance has a major role tween an interface terminated with two full Se monolayers in determining the actual value of the dipole moment at the and a single Se monolayer. The latter corresponds to the interface. relaxed configuration B, while the former corresponds to the The work in Trieste was supported in part by INFM under limiting case of a type-C configuration for x 1. Charge the TUSBAR Advanced Research Project. The work in Min- transfer from both the metal and the Se-terminated semicon- neapolis was supported in part by the National Science Foun- ductor to the excess Se atoms at the interface is clearly vis- dation under Grant No. DMR-9116436. The work in Lau- ible, and the asymmetry in the charge transfer gives rise to a sanne was supported by the Swiss National Science well-defined dipole-field across the interface. The dipole- Foundation under Grant No. 20-47065.96. Useful discus- induced change in the electrostatic potential lineup is sions with A. Kahn are gratefully acknowledged. *Also at Istituto ICMAT del CNR, Montelibretti, Roma, Italy. 10 M. Vos et al., Phys. Rev. B 39, 10 744 1989 . 1 See, for example, L. J. Brillson, in Handbook on Semiconductors, edited 11 H. H. Farrell et al., J. Vac. Sci. Technol. B 8, 884 1990 . by P. T. Landsberg Elsevier, Amsterdam, 1992 , Vol. 1, p. 281, and R. 12 C. H. Park and D. J. Chadi, Phys. Rev. B 49, 16 467 1994 . T. Tung, in Contacts to Semiconductors, edited by L. J. Brillson Noyes 13 G. P. Lopinsky et al., Surf. Sci. 355, L355 1996 . Publications, Park Ridge, 1993 , p. 176, and references therein. 14 Under the above assumptions, and using the atomistic models employed in 2 See, for example, II-VI Blue/Green Laser Diodes, edited by Robert L. Ref. 9 for the c(2 2) and 2 1 reconstructions, given the Zn 3d emis- Gunshor and Arto V. Nurmikko Proc. SPIE, 2346, 1 1991 . sion intensity from a single 100 plane of Zn atoms I0Zn, the Se 3d 3 S. N. Mohammed and H. Morkoc¸, Prog. Quantum Electron. 20, 361 emission intensity from a single 100 plane of Se atoms I0Se, using 1 1996 . ML 1.417 Å as the interplanar spacing in bulk ZnSe, one finds 4 0 0 0 0 0 0 A. Franciosi and C. G. Van de Walle, Surf. Sci. Rep. 25, 1 1996 . Rc(2 2) IZn/ISe, R2 1 (IZe/ISe)k, and R1 1 (IZe/ISe)* k 1 x(1 5 M. Cantile et al., Appl. Phys. Lett. 64, 988 1994 ; L. Sorba, S. Yildirim, k) / 1 kx(1 k) , where k e 1.417/ and x is the excess Se cov- M. Lazzarino, A. Franciosi, D. Chiola, and F. Beltram, ibid. 69, 1927 erage in ML for the 1 1 surface relative to the 2 1 surface. Using 1996 . the experimental R values for the three surfaces at a photon energy of 6 120 eV, we find x 0.41 0.18 ML and 5.9 1.0 Å. C. Berthod et al., Europhys. Lett. 36, 76 1996 ; C. Berthod, J. Bardi, N. 15 W. Chen et al., J. Vac. Sci. Technol. B 12, 2639 1994 . Binggeli, and A. Baldereschi, J. Vac. Sci. Technol. B 14, 3000 1996 . 16 We used the pseudopotentials nonlocal form by R. Stumpf, X. Gonze, 7 S. M. Sze, Physics of Semiconductor Devices Wiley, New York, 1981 , p. and M. Scheffler, Fritz-Haber-Institut Research Report No. 1, 1991 un- 304 published . A nonlinear core correction was used for Zn. 8 M. Lazzarino et al., Appl. Phys. Lett. 68, 370 1996 , and references 17 We used the theoretical lattice parameters aZnSe Al 0 5.46 Å and a 4.15 Å, therein. and the metal-semiconductor spacing at the junction was taken as the 9 A detailed analysis of the evolution of the Se 3d and Zn 3d line shape as average of the interlayer spacings in ZnSe and Al. a function of Se desorption has already been presented by W. Chen, A. 18 Atomic interdiffusion, Al-Zn exchange reactions, and Zn segregation Kahn, P. Soukiassan, P. S. Mangat, J. Gaines, C. Ponzoni, and D. Olego, within the Al overlayer, for example, have all been reported for Al/ZnSe Phys. Rev. B 49, 10 790 1994 . The purpose of Fig. 1 is to calibrate the junctions, Refs. 1, 8, 10, and 15 and are largely neglected here. composition of the actual surfaces employed in the present study. 19 O. Zakharov et al., Phys. Rev. B 50, 10 780 1994 .