- Email: [email protected]
Band alignment at the interfaces of Al2O3 and ZrO2-based insulators with metals and Si
Journal of Non-Crystalline Solids 303 (2002) 69–77 www.elsevier.com/locate/jnoncrysol
Section 4. Band alignment of high-k dielectrics with Si: measurement and impact on electrical properties
Band alignment at the interfaces of Al2O3 and ZrO2-based insulators with metals and Si V.V. Afanas’ev a
a,*
, M. Houssa a,b, A. Stesmans a, G.J. Adriaenssens a, M.M. Heyns b
Department of Physics, University of Leuven, Celestijnenlaan 200D, 3001 Leuven, Belgium b IMEC, Kapeldreef 75, 3001 Leuven, Belgium
Abstract Internal photoemission of electrons was used to determine the band alignment in metal (Mg, Al, Ni, Cu, Au)-oxidesilicon structures with Al2 O3 - and ZrO2 -based insulators. For Al2 O3 - and ZrO2 layers grown on Si by atomic layer deposition the barrier height between the Si valence band and the oxide conduction band was found to be 3.25 and 3.1 eV, respectively. Thermal oxidation of the Si/oxide stacks results in a barrier height increase to 4 eV for both cases due to formation of a silicate interlayer. However, there is a significant sub-threshold electron emission both from silicon and metals, indicating a high density of states in the band gap of the insulators. These states largely determine the electron transport across metal oxides and may also account for a significant dipole component of the potential barrier at the metal/ZrO2 and metal/Al2 O3 interfaces. Ó 2002 Elsevier Science Ltd. All rights reserved. PACS: 73.20.At; 73.30.y; 73.40.Qv; 73.50.Pz; 73.61.Ng
1. Introduction Thin insulating layers of metal oxides with dielectric constant higher than that of SiO2 (e ¼ 3:9) are currently considered as candidate gate dielectric materials for metal–oxide–silicon (MOS) devices capable of reducing the direct tunneling of electrons [1–5]. To suppress the tunneling between metal and silicon the energy barrier (U) at the
* Corresponding author. Tel.: +32-016 327167; fax: +32-016 327987. E-mail address: [email protected] (V.V. Afanas’ev).
metal/oxide and Si/oxide interfaces must be sufficiently high. Consequently, the barrier height, together with the dielectric constant of an oxide, represent the parameters of primary importance for selection of an insulator for device application. So far, due to the lack of experimental data, the interfacial barriers were evaluated theoretically using the bulk parameters of solids involved [6,7]. However, they may not necessarily reflect the properties of thin films. Aluminum oxide (Al2 O3 ) provides a clear example of such differences: the reduction of the oxide thickness from 5.5 to 3.5 nm was reported to result in a 0.5-eV Al/Al2 O3 barrier lowering [8]. Moreover, fabrication of laterally uniform ultra-thin insulators requires a low process
0022-3093/02/$ - see front matter Ó 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 0 2 2 - 3 0 9 3 ( 0 2 ) 0 0 9 6 7 - 5
70
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
temperature to avoid reaction and/or intermixing with the Si substrate which may affect the barrier height. For instance, the Al/Al2 O3 interface barrier height increases from 1.8 to 2.2 eV for the oxides formed at room temperature [9–13] to 2.45 eV for thermal oxidation in air at 300 °C [14], 2.9 eV for thermal oxidation in O2 at 425°C [15], to reach 3.05 and 3.2 eV for the Al2 O3 layers deposited at 850 [16] and 900 °C [17], respectively. Such an uncertainty in U leads not only to large inaccuracy in the evaluation of the electron tunneling probability, but also to a largely undefined metal–silicon work function difference (/ms ) thus hampering determination of the electric field strength and the fixed charge density in the oxide from the electrical measurements. The knowledge of barrier heights is also needed to analyze the electron transport mechanism(s) across the MOS insulator. While the band-to-band tunneling is highly sensitive to the energy offset, it is also possible that the traps in the oxide will mediate the electron transport between the Si and metal, as it happens in the Si MOS structures subjected to electrical stress or irradiation [18–20]. In the latter case the most important parameter might be the energy depth of the corresponding trap which determines the trap-to-trap and trapto-band emission probabilities [21,22]. Clarification of the current transport mechanism is of vital importance for optimization of the gate dielectric technology: Would the trap-assisted injection be dominant, bulk properties of the oxide must be improved, otherwise, interface engineering will be needed to reduce the gate leakage current.
2. Experimental To ensure the lateral uniformity of the insulating layers we used the atomic-layer chemical vapor deposition (ALCVD) process which provides an atomically controlled growth mode through saturation of each adsorbed monolayer of the metal oxide atoms [23,24]. The Al2 O3 (0.5–15 nm) and ZrO2 (3–20 nm) layers were deposited on clean (HF dip prior to ALCVD) or SiO2 -covered (1 0 0)Si substrates (n- or p-type, nd , na 1 1015 cm3 ) at 300 °C using H2 O, Al(CH3 Þ3 and ZrCl4 precursors
as described elsewhere [22,25]. ALCVD grown Al2 O3 /ZrO2 stacks were also analyzed. After deposition, some samples were oxidized in dry O2 (1 atm) at temperatures in the range 500–800 °C to grow additional oxide at the interface between Si and the Al2 O3 or ZrO2 layers. Transmission electron microscopy results indicate that this treatment does not lead to intermixing of the layers [25]. MOS capacitors of 0.5 mm2 area were defined by thermal evaporation of semitransparent (15-nm thick) metal electrodes onto the oxides in high (106 Torr) vacuum from a resistively heated W boat. A set of metals (Mg, Al, Cu, Ni, Au) with substantial difference in electronegativity was used. For the sake of comparison, the same metal electrodes were also evaporated onto 4–5-nm thick thermal SiO2 layers grown on (1 0 0)Si at 800 °C. No postmetallization anneal was performed to ensure a minimal metal/oxide chemical interaction. MOS structures were characterized electrically at room temperature by dc current–voltage (I–V ), capacitance–voltage (C–V ) (102 –106 Hz) and by internal electron photoemission measurements (IPE) [26]. The I–V curves of the MOS structures were measured not only in darkness but also under excitation with photons from an Arþ ion laser (hm ¼ 2:41–2:73 eV) to reveal the photon-stimulated electron transitions [27]. From the accumulation capacitance of MOS structures with as-deposited uniform Al2 O3 and ZrO2 layers, the low-frequency relative dielectric constants were found to be 8 and 15, respectively. Upon oxidation, the dielectric constants of both the interfacial silicon oxide and the high-permittivity oxide were found to increase [25]. More details regarding electrical properties of the ALCVD oxides have been published elsewhere [22,25,28–30]. The IPE spectral curves were measured in the photon energy range hm ¼ 2–5 eV (the photon energy was always kept well below the insulator bandgap width – 5.4 eV for ALCVD ZrO2 , Ref. [22]) with spectral resolution of 2 nm. IPE yield (Y ) was defined as photocurrent normalized to the incident photon flux [26,31]. For a positive metal bias the IPE from the silicon substrate is measured, which spectral dependence is independent of the type of metal used, as reported elsewhere [30]. When the metal is biased negatively, the current across the insulator is related to
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
electrons emitted from the metal gate and exhibits the corresponding spectral threshold variations in accordance with the metal/oxide barrier height changes. To avoid the contribution of electrons simultaneously photoinjected from the Si substrate to the current from the metal, the spectral measurements on thin (d < 10 nm) insulators were performed under a bias ranging from 2 to 4 V, i.e., high enough to prevent the Si-emitted electrons from reaching the gate. For the same purpose the IPE spectra from Si were mostly measured in the MOS structures with Au electrodes [30]. The density of electrons injected during the IPE experiments was kept below 1012 cm2 to avoid substantial trapping in the oxide. The CV measurements after the IPE experiments were used to affirm negligible charging.
3. Results and discussion 3.1. IPE spectra and thresholds The IPE spectra for n-Si/Al2 O3 /Au and n-Si/ ZrO2 /Au MOS structures (metal biased positively)
Fig. 1. Cube root of the IPE yield as a function of photon energy for MOS structures with different dielectrics: (a) as-deposited 5-nm thick Al2 O3 ( ) and oxidized at 650 °C for 30 min () or at 800 ° C for 10 min (M) as compared to a 4.1-nm thick thermal SiO2 (O); (b) as-deposited 7.4 nm ZrO2 ( ) and oxidized at 650 °C for 30 min () or at 800 ° C for 10 min (M). All curves are measured under an applied electric field of 2 MV/cm in the insulating layer closest to Si. The arrows E1 and E2 indicate onsets of direct optical transition in the Si crystal. The spectral thresholds UV are indicated for different oxides. The energy width of the monochromator slit is indicated in (a); the error in the IPE yield determination is smaller than the symbol size.
71
are shown in Fig. 1(a) and (b), respectively. Because replacement of the Au gate electrode with Al has no effect on the IPE curves, the latter correspond to electron emission from the Si valence band to the conduction band of the corresponding insulator, as shown in the insert in Fig. 1(b). This is affirmed independently by observation of the IPE spectrum modulation by direct optical transitions in the Si crystal [26,32] as indicated by arrows E1 and E2 in Fig. 1(a). These transitions cause deviation of spectral curves from the ap3 parent Y ðhm UÞ dependence thus limiting the spectral interval of linear fit and causing a substantial increase of the IPE threshold determination error as compared to the energy width of the monochromator slit (0.02 eV). For as-deposited Al2 O3 the IPE spectral threshold (UV ) appears to be slightly above 3 eV, while for ZrO2 it is 0.15 eV lower. To obtain the exact barrier energies, the spectral thresholds were measured at different strengths of electric field and then extrapolated to zero field in the Schottky coordinates, as shown in Fig. 2. The resulting barriers between the top of the Si valence states and the bottom of the oxide conduction states are 3:25 0:08 eV for Al2 O3 and 3:1 0:1 eV for ZrO2 as compared to the 4:25 0:05 eV barrier at the Si/SiO2 interface, which is also shown for comparison. The Schottky plot allows us to determine the effective
Fig. 2. Schottky plot of the spectral thresholds for IPE from the Si valence band into the conduction band of SiO2 layers of different thickness (in nm): 55 ( ), 5.8 (), 4.1 (M), as compared to 5-nm thick Al2 O3 (O) and 7.4-nm thick ZrO2 (}) layers. The symbol size corresponds to the error of the threshold determination.
72
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
image-force dielectric constant ei related to the refraction index n of the insulator near the Si surface (ei n2 ). From the slopes of the lines in Fig. 2, ei values are obtained as 2:1 0:2, 3:5 0:5, and 5 1 for SiO2 , Al2 O3 , and ZrO2 , respectively. Within experimental error, these values coincide with the square of the refractive index of bulk SiO2 (n ¼ 1:456), Al2 O3 (n ¼ 1:77), and ZrO2 (n ¼ 2:2) suggesting a physically abrupt transition between silicon and the oxide [30]. Comparison of the above results with earlier data on IPE from Si into Al2 O3 layers deposited at high temperature or subjected to high-temperature anneal [16,17,33] indicates that in the as-deposited state the barrier height is much reduced (3.25 eV vs. 3.9–4.1 eV). The 2.1 eV conduction band offset resulting from our value also appears to be significantly lower than the theoretically predicted 2.8 eV value [7], thus affirming the importance of interface effects. Upon oxidation, the major IPE emission shifts to the spectral range of hm > 4 eV, which suggests formation of a silicate layer as explanation of earlier experimental data. In the Si/ ZrO2 system the barrier increases to a similar value of 4 eV starting from 3.1 eV for the as-deposited sample. Interestingly, crystallization of ZrO2 observed upon oxidation at T > 500 °C [27] seems to have no substantial effect neither on the energy of the ZrO2 conduction band edge (with respect to energy bands of the Si substrate) nor on the bandgap width of oxide itself (spectra not shown). Oxidation results in reduction of the IPE yield in the spectral range of 3 < hm < 4:5 eV, which indicates a decrease in the density of states related to the conduction band states of the high-permittivity metal oxide. Instead, the spectral features characteristic for SiO2 become pronounced as the thickness of the thermally grown interlayer exceeds 1 nm. Worth noticing here is the fact that the IPE curves of the oxidized gate stacks do not reach the shape corresponding to the IPE into pure thermal SiO2 (O in Fig. 1(a)), not even at the highest oxidation thermal budgets applied. A significant ‘tail’ is observed in the sub-threshold region, particularly pronounced in the case of Al2 O3 . More details for ZrO2 are revealed by the spectral curves measured in the sub-barrier spectral range using Arþ laser excitation, shown in Fig. 3. This photo-
Fig. 3. IPE yield as a function of photon energy for n-Si/7.4 nm ZrO2 /Au MOS structures in as-deposited state ( ) and after oxidation at 500 °C for 60 min (), 650 °C for 30 min (M), and at 800 °C for 10 min (O). The measurements are done with an applied field strength in ZrO2 layer of 3 MV/cm, metal biased positively. The error in the yield determination is smaller than the symbol size.
current caused by excitation of electrons via the oxide conduction band tail states decreases with increasing oxidation temperature, but does not disappear entirely. Apparently, the electron exchange between Si and the band tail states in ZrO2 gets impeded upon growth of a SiO2 -like interlayer. However, the latter is likely to be a silicate containing a considerable density of gap states, which provide under-barrier pathways for electrons. Examples of spectral curves of IPE from a metal into 5-nm thick Al2 O3 and ZrO2 layers are presented in Fig. 4. For Al2 O3 only data for Al and Au are shown in Fig. 4(a), which are similar to the data reported by DiMaria for IPE into high-temperature deposited aluminum oxide [17]. For ZrO2 the spectra of IPE from Mg ( ), Al (), Ni (M), Cu (O), and Au (}) are presented in Fig. 4(b). As the IPE from the same metals into thermal SiO2 of the same thickness (curves not shown) obeys the 2 Fowler law [34] Y ðhm UÞ (in agreement with literature results for thicker oxides, see Refs. [35– 37]), Fowler co-ordinates ðY 1=2 hmÞ were used to determine the IPE spectral thresholds. A cubic fit Y 1=3 hm was found to be less good; it results in a systematic 0.1 eV red shift of the IPE threshold which is comparable to the error arising from the extrapolation inaccuracy. It needs to be mentioned
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
73
Table 1 Barrier heights (UV ; U) at the interfaces between (1 0 0)Si, metals, and SiO2 , Al2 O3 and ZrO2 insulators Insulator
SiO2 (0.05 eV) Al2 O3 (0.1 eV) ZrO2 (0.1 eV)
Fig. 4. (a) Square root of the IPE yield from the metal as a function of photon energy for MOS structures with Al and Au electrodes deposited on a 5-nm thick as-deposited (open symbols) and oxidized at 800 °C for 10 min (filled symbols) Al2 O3 insulator. The strength of the electric field in the oxide is (in MV/cm) as follows: Al on as-deposited oxide: 0.2 ( ), 0.5 (), 1.0 (M), 1.5 (O), 3.0 (}); Al on oxidized Al2 O3 : 0.5 ( ), 1.0 (j), 2.0 (N), 4.0 (.); Au on as-deposited oxide: 0.5 ( ), 1.0 (), 2.0 (M), 3.0 (O), 4.0 (}), metal bias is negative. (b) Square root of the IPE yield from the metal as a function of photon energy for MOS structures with different metals deposited on a 5-nm thick as-deposited ZrO2 layer. The electric field strength in the oxide is 2 MV/cm, metal biased negatively. The lines illustrate the determination of the IPE spectral thresholds. The energy width of the monochromator slit is indicated in (b); inaccuracy of the IPE yield determination is smaller than the symbol size.
here is that the power of the yield dependence on photon energy is determined exclusively by the shape of the energy distribution of excited electrons in the emitter and by the energy dependence of the barrier surmount probability [38]. For electrons emitted from the same metal, their energy distribution must be the same and the variation of the spectral curves will reflect mostly the properties of the potential barrier. Therefore, deviation of the IPE spectra from the Fowler law in the near-threshold region (Fig. 4) indicates a lateral non-uniformity of the interfacial barrier. Two more features are revealed by the data shown: First, the thermal oxidation at a temperature as high as 800 °C does not change the metal/insulator barrier significantly as compared to the as-deposited state (cf. data for Al in Fig. 4 (a)). Second, the electric field has only a weak effect on spectral thresholds (within the indicated accuracy limit of 0.1 eV), suggesting a smaller Schottky barrier reduction as compared to the IPE from Si into Al2 O3 and ZrO2 (cf. Fig. 2). The barrier heights U are
Emitter (1 0 0)Si
Mg
Al
Ni
Cu
Au
4.25 3.25 3.1
2.50 2.6 2.6
3.15 2.9 2.7
3.70 3.5 3.2
3.85 3.6 3.25
4.10 4.1 3.5
listed in Table 1 for the studied insulator/metal pairs, and UV values are given for (1 0 0)Si/insulator interfaces. To enable a meaningful comparison between different oxides, the measured barrier heights are plotted in Fig. 5 together with literature data for SiO2 and Al2 O3 as a function of the Pauling metal electronegativity XM [39]. For thermal SiO2 , panel (a), the barriers measured on 5-nm thick oxides () perfectly follow the trend of barrier increase with the metal electronegativity reported earlier for thicker oxides (symbol from Ref. [35], [36] for W). In the case of Al2 O3 , panel (b), the literature data (symbol from [17], [33] for W), on average, lie slightly above the values determined for both as-deposited () and thermally oxidized (M) AL CVD layers, though for Mg and Au the agreement is perfect. Taking into account the spread of the data, the increase of the interfacial barrier height with the metal electronegativity appears to be the
Fig. 5. Barrier height at the metal/insulator interface as a function of the gate metal electronegativity for 5-nm thick thermal SiO2 (a), 5-nm thick ALCVD Al2 O3 (b) and ZrO2 (c). Circles indicate the literature values for thicker oxides, squares refer to the experimental results on as-fabricated oxide layers, triangles refer to the ALCVD layers oxidized in O2 at 650 °C for 30 min. Lines guide the eye; the symbol size corresponds to 0.1 eV inaccuracy in barrier height determination.
74
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
same as in the case of SiO2 and thus close to the ideal case dU=dXM ¼ 1 [40]. For both as-deposited () and oxidized (M) ZrO2 layers, panel (c), the increase of the barrier height with metal electronegativity is found to be considerably smaller (ideality factor dU=dXM 0:75Þ than for SiO2 and Al2 O3 . 3.2. Work function differences On the basis of the barrier values listed in Table 1 the metal/silicon work function difference /ms was calculated for both the n- and p-type Si samples used in the present study. The results are shown by lines in Fig. 6 and are compared with the flatband voltage (VFB ) values determined from the high-frequency C–V curves in the n- (filled symbols) and p-type (open symbols) MOS structures with Al2 O3 (a), Al2 O3 /ZrO2 (b), and ZrO2 (c) dielectrics, both as-deposited (circles) and oxidized at 650 °C for 30 min (squares) or at 800 °C for 10 min (triangles). The dotted lines on the panels indicate the /ms behavior in n-Si MOS capacitors with thermally grown SiO2 dielectric. The comparison between /ms and VFB values reveals three features: First, for the same metal, /ms values measured in the MOS structures with thermal SiO2 are considerably smaller than those in MOS structures with an ALCVD deposited insulator. Sec-
Fig. 6. Flatband voltages of n- (filled symbols) and p-type (open symbols) MOS capacitors with 5-nm thick Al2 O3 (a), stacked 1.5 nm Al2 O3 /5 nm ZrO2 (b), and 5-nm thick ZrO2 (c) insulators as a function of gate metal electronegativity. Solid and dashed lines show correspondingly the behavior of the metal–semiconductor work function difference /ms for n- and ptype Si, dotted lines indicate /ms for an n-type MOS structure with thermal SiO2 insulator. The error of VFB determination is smaller than the symbol size.
ond, the systematically observed trend VFB > /ms for the as-deposited ALCVD layers indicates the presence of a substantial density of negative charge in the metal oxides, particularly in ZrO2 -based insulators. The density of this charge is effectively reduced by oxidation of the deposited dielectric. Third, and by contrast, the Al-gated samples exhibit a systematic VFB shift to below /ms indicating a positive charge buildup in the insulator upon Al evaporation. The latter may be related, for instance, to a chemical interaction of Al with the Hcontaining fragments at the oxide surface, leading to the formation of protons [25]. The important result of this /ms study consists in revealing a large difference in the values measured in MOS structures with thermal SiO2 and with an ALCVD deposited insulator, which indicates the presence of a significant interface dipole. Growth of a thin silicon oxide at the interface between the Si substrate and the ALCVD layer has no impact neither on the Si/ALCVD oxide band alignment nor on this dipole (cf. Fig. 6) suggesting that the latter is associated with the metal/oxide interface. The latter suggestion finds support from the observed non-idealities in the IPE spectra of the metal/oxide interfaces. First, the tails of the Y hm curves refer to non-uniformity of the barrier. As may be seen from Fig. 4, they show a clear trend to increase for metals with a lower barrier (Al, Mg) as compared to Au. The latter suggests that, in addition to the conduction band tail states [30], the barrier may also be affected by a nonuniform polarization layer possibly related to negative charge arising from electrons trapped in the oxide at levels below the metal Fermi level [41]. Next, a weak dependence of the metal/oxide barrier height on the strength of electric field in the oxide suggests the presence of a strong built-in electric field which is also consistent with the formation of an interfacial polarization layer. Finally, the weaker U-electronegativity dependence for ZrO2 (cf. Fig. 5(c)) may also be related to the relative enhancement of the negative dipole for the metals with smallest barriers (Mg, Al). The latter effect may be of concern for practical applications of the ALCVD gate insulators: the threshold voltage control by using nþ /pþ -poly-Si gates or metal electrodes with different work function may
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
appear not so efficient as in the MOS devices with a SiO2 gate dielectric. 3.3. Electron tunneling vs. trap-assisted transport The determined interface barrier heights and the contact potential values allowed us to analyze in detail the I–V curves of MOS structures with high-e dielectrics. Examples of such curves for nSi/ZrO2 /Au structures (positive metal bias) are shown in Fig. 7(a) for the as-deposited samples ( ) and oxidized at various thermal budgets (500 °C for 60 min (); 650 °C for 30 min (M); 800 °C for 10 min (O)) samples as well as one annealed in high vacuum (650 °C, 30 min, N). Similar curves were observed when Au was replaced with another metal indicating that the measured current is due to electron injection from Si substrate. Apparently, oxidation of the Si/ZrO2 structure results in significant reduction of the leakage current probably related to the growth of a silicon oxide at the interface. The importance of oxygen is clearly indicated by lack of current reduction upon thermal treatment in high vacuum. To reveal the injection mechanism, the Fowler–Nordheim (FN) plots [42] of the dark (filled symbols) current and the photocurrent measured under excitation with 2.71
Fig. 7. (a) I–V curves (positive metal bias) of n-Si MOS structures with 7.4-nm ZrO2 insulator in as-deposited state ( ), after oxidation at 500 °C for 60 min (), at 650 °C for 30 min (M), or at 800 °C for 10 min (O), compared to the sample annealed in high vacuum at 650 °C for 30 min (N). (b) The Fowler–Nordheim plots of dark (filled symbols) and photocurrent (2.71 eV, 100 mW, open symbols) measured in n-Si MOS structures with 7.4-nm ZrO2 insulator in as-deposited state (circles), after oxidation at 500 °C for 60 min (squares), at 650 °C for 30 min (triangles), or at 800 °C for 10 min (inverted triangles) under positive metal bias.
75
eV photons at incident power of 100 mW (open symbols) are shown in Fig. 7(b) for both as-deposited and oxidized samples. It is clearly seen that with increasing oxidation temperature both currents decrease dramatically. Also, in the low-field range the currents do not exhibit the FN behavior typical for electron tunneling, suggesting a trapassisted electron transport. The slope of the FNtype I–V curves observed at high field values increases with oxidation temperature approaching that for the case of FN tunneling from Si into SiO2 . The latter indicates the formation of a SiO2 like interlayer as the major reason for the reduction of the leakage current in the oxidized Si/ZrO2 structures. The influence of an interlayer between the highe insulator and silicon is also seen from the FN plots of the dark and photocurrent (hm ¼ 2.71 eV, 100 mW) measured in n-type Si MOS structures with as-deposited and oxidized stacks of 1.3 nm thermal SiO2 /5 nm ZrO2 and 1.5 nm Al2 O3 /5 nm ZrO2 , shown in Fig. 8(a) and (b), respectively. The presence of a pre-grown SiO2 layer (Fig. 8(a)) significantly reduces the leakage current already in as-deposited (not oxidized) sample. After oxidation, a further drop of the trap-related currents in the low-field range is observed, but it has only a marginal influence on the high-field electron tunneling. The reduction of both the dark and
Fig. 8. The FN plots of dark (filled symbols) and photocurrent (2.71 eV, 100 mW, open symbols) measured in n-Si MOS structures with stacks of 1.3 nm SiO2 /5 nm ZrO2 (a) and 1.5 nm Al2 O3 /5 nm ZrO2 ; (b) insulators in as-deposited state (circles), after oxidation at 500 °C for 60 min (squares), at 650 °C for 30 min (triangles), or at 800 °C for 10 min (inverted triangles) under positive metal bias.
76
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
photocurrent by the Al2 O3 interlayer (Fig. 8(b)) is not as strong as by SiO2 suggesting a lower interfacial barrier for electrons at the Si/Al2 O3 than at the Si/SiO2 interface, in accordance with the IPE results. Only after oxidation at 800 °C the currents typical for the SiO2 /ZrO2 stack are reached. A behavior very different from the one discussed above with respect to oxidation is revealed by I–V curves of electron injection from Au into ZrO2 measured in p-type Si MOS structures (metal biased negatively) and shown in Fig. 9(a) as FN plot. The dark current (filled symbols) is barely affected by oxidation. The laser-induced current (open symbols) decreases in the low-field range with increasing temperature of oxidation suggesting a reduction in the density of traps available for photoionization; yet, there is little change at high field. The FN plots of the dark current measured in similar MOS structures with Al (open symbols) and Au (filled symbols) electrodes are compared in Fig. 9(b) (three curves are shown for each sample). Despite significant difference in the barrier height for electrons at the interfaces of ZrO2 with Al and Au (cf. Table 1), hardly any variation in the slope of the FN plots is observed in Fig. 9(b). The latter indicates that the emission of an electron from the
Fig. 9. (a) FN plots of dark (filled symbols) and photocurrent (2.71 eV, 100 mW, open symbols) measured in p-Si MOS structures with 5 nm ZrO2 insulators in as-deposited state (circles), after oxidation at 500 °C for 60 min (squares), at 650 °C for 30 min (triangles), or at 800 °C for 10 min (inverted triangles) under negative metal bias. (b) FN plots of dark current measured in p-Si MOS structures with 5 nm ZrO2 insulators and Al (open symbols) and Au (filled symbols) electrodes in as-deposited state (circles), after oxidation at 650 °C for 30 min (squares), or at 800 °C for 10 min (triangles) under negative metal bias. Three curves are shown for each sample.
metal into the conduction band of ZrO2 does not constitute the rate-limiting step. Rather, the insensitivity to the metal type suggests that the current is determined by emission of an electron from some deep, stable against oxidation, gap state in ZrO2 . This may account for a lower current from Al than from Au measured at nominally the same electric field strength at the metal/ZrO2 interface (cf. Fig. 9 (b)) despite the 0.8 eV lower barrier height for Al as compared to Au (Table 1). Apparently, a negatively charged polarization layer is formed at the Al/ZrO2 interface effectively reducing the electron emission probability from oxide traps. This picture is consistent with the non-ideal behavior of the IPE characteristics discussed earlier and points towards the trap-assisted tunneling as a dominant conduction mechanism in ALCVD Al2 O3 and ZrO2 layers.
4. Conclusions IPE analysis of interfaces of (1 0 0)Si and various metals with thin thermally grown SiO2 and ALCVD Al2 O3 and ZrO2 insulators provided directly the electron barrier energies between the filled states of the metals and Si and the conduction band of the respective insulators. The asgrown Al2 O3 and ZrO2 layers show a barrier with Si by 1 eV lower than SiO2 . In metal/Al2 O3 contacts, the barrier height increases with electronegativity of the metal in the nearly ideal way dU=dXM 1, while for as-deposited ZrO2 insulators a less ideal behavior is observed with dU= dXM 0:75. The obtained barrier values remain unchanged down to the oxide layer thickness of 3 nm for Al2 O3 and ZrO2 , and 1 nm for SiO2 [30]. The metal/silicon work function differences in the MOS structures determined on the basis of the measured barrier heights appear to be different from those for the SiO2 insulator. This is ascribed to the formation of a negative dipole layer at the interfaces of metals with ALCVD oxides indicating the presence of a substantial density of electron traps. The traps present in ALCVD insulators determine the electron transport and their influence can be efficiently reduced by introduction of a SiO2 interlayer. The latter suggests that the observed
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
insulating properties of the high-e dielectric stacks have not yet reached their fundamental limit determined by band offsets at the interfaces and can thus be improved further.
References [1] R.A.B. Devine, L. Vallier, J.-L. Autran, P. Paillet, J.-L. Leray, Appl. Phys. Lett. 68 (1996) 1775. [2] O. Nakagawara, Y. Toyota, M. Kobayashi, Y. Yoshino, Y. Katayama, H. Tabata, T. Kawai, J. Appl. Phys. 80 (1996) 388. [3] G.B. Alers, R.M. Fleming, Y.H. Wang, B. Dennis, A. Pinczuk, G. Redinbo, R. Urdahl, E. Ong, Z. Hasan, Appl. Phys. Lett. 72 (1998) 1308. [4] G.D. Wilk, R.M. Wallace, J.M. Anthony, J. Appl. Phys. 87 (2000) 484. [5] M. Copel, M. Gribelyuk, E. Gusev, Appl. Phys. Lett. 76 (2000) 436. [6] J. Robertson, C.W. Chen, Appl. Phys. Lett. 74 (1999) 1168. [7] J. Robertson, J. Vac. Sci. Technol. B 18 (2000) 1785. [8] J. Kadlec, K.H. Gundlach, Solid State Commun. 16 (1975) 621. [9] A.I. Braunstein, M. Braunstein, G.S. Picus, C.A. Mead, Phys. Rev. Lett. 14 (1965) 219. [10] A.I. Braunstein, M. Braunstein, G.S. Picus, Phys. Rev. Lett. 15 (1965) 956. [11] F.L. Schuermeyer, J.A. Crawford, Appl. Phys. Lett. 9 (1966) 317. [12] O.L. Nelson, D.E. Anderson, J. Appl. Phys. 37 (1966) 77. [13] A.M. Goodman, J. Appl. Phys. 41 (1970) 2176. [14] W. Ludwig, B. Korneffel, Phys. Status Solidi 24 (1967) K137. [15] K.W. Shepard, J. Appl. Phys. 36 (1965) 796. [16] N. Szydlo, P. Poirier, J. Appl. Phys. 42 (1971) 4880. [17] D.J. DiMaria, J. Appl. Phys. 45 (1974) 5454. [18] D.J. DiMaria, E. Cartier, J. Appl. Phys. 78 (1995) 3883. [19] A. Scarpa, A. Paccagnella, F. Montera, G. Gibaudo, G. Pananakakis, G. Ghidini, P.G. Fuochi, IEEE. Trans. Nucl. Sci. NS-44 (1997) 1818. [20] V.V. Afanas’ev, A. Stesmans, J. Electrochem. Soc. 146 (1999) 3409.
77
[21] C. Svensson, I. Lundsrom, J. Appl. Phys. 44 (1973) 4657. [22] M. Houssa, M. Tuominen, M. Naili, V.V. Afanas’ev, A. Stesmans, S. Haukka, M.M. Heyns, J. Appl. Phys. 87 (2000) 8615. [23] M. Ritala, M. Leskela, Appl. Surf. Sci. 75 (1994) 333. [24] M. Ritala, K. Kukli, A. Rahtu, P.I. Raisanen, M. Leskela, T. Sajavaara, J. Keininen, Science 288 (2000) 319. [25] M. Houssa, M. Naili, C. Zhao, H. Bender, M.M. Heyns, A. Stesmans, Semicond. Sci. Technol. 16 (2001) 31. [26] V.K. Adamchuk, V.V. Afanas’ev, Progr. Surf. Sci. 41 (1992) 111. [27] V.V. Afanas’ev, A. Stesmans, Phys. Rev. Lett. 78 (1997) 2437–2440. [28] M. Houssa, A. Stesmans, M. Naili, M.M. Heyns, Appl. Phys. Lett. 77 (2000) 1381. [29] M. Houssa, V.V. Afanas’ev, A. Stesmans, M.M. Heyns, Appl. Phys. Lett. 77 (2000) 1885. [30] V.V. Afanas’ev, M. Houssa, A. Stesmans, M.M. Heyns, Appl. Phys. Lett. 78 (2001) 3073. [31] V.V. Afanas’ev, A. Stesmans, M.O. Andersson, Phys. Rev. B 54 (1996) 10820. [32] T.H. Di Stefano, J.E. Lewis, J. Vac. Sci. Technol. 11 (1974) 1020. [33] R. Ludeke, M.T. Cuberes, E. Cartier, Appl. Phys. Lett. 76 (2000) 2886; R. Ludeke, M.T. Cuberes, E. Cartier, J. Vac. Sci. Technol. B 18 (2000) 2153. [34] R.H. Fowler, Phys. Rev. 38 (1931) 45. [35] B.E. Deal, E.H. Snow, C.A. Mead, J. Phys. Chem. Solids 27 (1966) 1873. [36] R.J. Powell, R.C. Beairsto, Solid State Electron. 16 (1973) 265. [37] P.M. Solomon, D.J. DiMaria, J. Appl. Phys. 52 (1981) 5867. [38] R.J. Powell, J. Appl. Phys. 41 (1970) 2424. [39] L. Pauling, The Nature of the Chemical Bond, 3rd Ed., Cornell University, Ithaca, NY, 1960. [40] S. Kurtin, T.C. McGill, C.A. Mead, Phys. Rev. Lett. 22 (1969) 1433. [41] C.J. Wang, T.H. DiStefano, CRC Crit. Rev. Solid State Sci. 5 (1975) 327. [42] R.H. Fowler, L. Nordheim, Proc. Roy. Soc. A 119 (1928) 173.
Section 4. Band alignment of high-k dielectrics with Si: measurement and impact on electrical properties
Band alignment at the interfaces of Al2O3 and ZrO2-based insulators with metals and Si V.V. Afanas’ev a
a,*
, M. Houssa a,b, A. Stesmans a, G.J. Adriaenssens a, M.M. Heyns b
Department of Physics, University of Leuven, Celestijnenlaan 200D, 3001 Leuven, Belgium b IMEC, Kapeldreef 75, 3001 Leuven, Belgium
Abstract Internal photoemission of electrons was used to determine the band alignment in metal (Mg, Al, Ni, Cu, Au)-oxidesilicon structures with Al2 O3 - and ZrO2 -based insulators. For Al2 O3 - and ZrO2 layers grown on Si by atomic layer deposition the barrier height between the Si valence band and the oxide conduction band was found to be 3.25 and 3.1 eV, respectively. Thermal oxidation of the Si/oxide stacks results in a barrier height increase to 4 eV for both cases due to formation of a silicate interlayer. However, there is a significant sub-threshold electron emission both from silicon and metals, indicating a high density of states in the band gap of the insulators. These states largely determine the electron transport across metal oxides and may also account for a significant dipole component of the potential barrier at the metal/ZrO2 and metal/Al2 O3 interfaces. Ó 2002 Elsevier Science Ltd. All rights reserved. PACS: 73.20.At; 73.30.y; 73.40.Qv; 73.50.Pz; 73.61.Ng
1. Introduction Thin insulating layers of metal oxides with dielectric constant higher than that of SiO2 (e ¼ 3:9) are currently considered as candidate gate dielectric materials for metal–oxide–silicon (MOS) devices capable of reducing the direct tunneling of electrons [1–5]. To suppress the tunneling between metal and silicon the energy barrier (U) at the
* Corresponding author. Tel.: +32-016 327167; fax: +32-016 327987. E-mail address: [email protected] (V.V. Afanas’ev).
metal/oxide and Si/oxide interfaces must be sufficiently high. Consequently, the barrier height, together with the dielectric constant of an oxide, represent the parameters of primary importance for selection of an insulator for device application. So far, due to the lack of experimental data, the interfacial barriers were evaluated theoretically using the bulk parameters of solids involved [6,7]. However, they may not necessarily reflect the properties of thin films. Aluminum oxide (Al2 O3 ) provides a clear example of such differences: the reduction of the oxide thickness from 5.5 to 3.5 nm was reported to result in a 0.5-eV Al/Al2 O3 barrier lowering [8]. Moreover, fabrication of laterally uniform ultra-thin insulators requires a low process
0022-3093/02/$ - see front matter Ó 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 0 2 2 - 3 0 9 3 ( 0 2 ) 0 0 9 6 7 - 5
70
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
temperature to avoid reaction and/or intermixing with the Si substrate which may affect the barrier height. For instance, the Al/Al2 O3 interface barrier height increases from 1.8 to 2.2 eV for the oxides formed at room temperature [9–13] to 2.45 eV for thermal oxidation in air at 300 °C [14], 2.9 eV for thermal oxidation in O2 at 425°C [15], to reach 3.05 and 3.2 eV for the Al2 O3 layers deposited at 850 [16] and 900 °C [17], respectively. Such an uncertainty in U leads not only to large inaccuracy in the evaluation of the electron tunneling probability, but also to a largely undefined metal–silicon work function difference (/ms ) thus hampering determination of the electric field strength and the fixed charge density in the oxide from the electrical measurements. The knowledge of barrier heights is also needed to analyze the electron transport mechanism(s) across the MOS insulator. While the band-to-band tunneling is highly sensitive to the energy offset, it is also possible that the traps in the oxide will mediate the electron transport between the Si and metal, as it happens in the Si MOS structures subjected to electrical stress or irradiation [18–20]. In the latter case the most important parameter might be the energy depth of the corresponding trap which determines the trap-to-trap and trapto-band emission probabilities [21,22]. Clarification of the current transport mechanism is of vital importance for optimization of the gate dielectric technology: Would the trap-assisted injection be dominant, bulk properties of the oxide must be improved, otherwise, interface engineering will be needed to reduce the gate leakage current.
2. Experimental To ensure the lateral uniformity of the insulating layers we used the atomic-layer chemical vapor deposition (ALCVD) process which provides an atomically controlled growth mode through saturation of each adsorbed monolayer of the metal oxide atoms [23,24]. The Al2 O3 (0.5–15 nm) and ZrO2 (3–20 nm) layers were deposited on clean (HF dip prior to ALCVD) or SiO2 -covered (1 0 0)Si substrates (n- or p-type, nd , na 1 1015 cm3 ) at 300 °C using H2 O, Al(CH3 Þ3 and ZrCl4 precursors
as described elsewhere [22,25]. ALCVD grown Al2 O3 /ZrO2 stacks were also analyzed. After deposition, some samples were oxidized in dry O2 (1 atm) at temperatures in the range 500–800 °C to grow additional oxide at the interface between Si and the Al2 O3 or ZrO2 layers. Transmission electron microscopy results indicate that this treatment does not lead to intermixing of the layers [25]. MOS capacitors of 0.5 mm2 area were defined by thermal evaporation of semitransparent (15-nm thick) metal electrodes onto the oxides in high (106 Torr) vacuum from a resistively heated W boat. A set of metals (Mg, Al, Cu, Ni, Au) with substantial difference in electronegativity was used. For the sake of comparison, the same metal electrodes were also evaporated onto 4–5-nm thick thermal SiO2 layers grown on (1 0 0)Si at 800 °C. No postmetallization anneal was performed to ensure a minimal metal/oxide chemical interaction. MOS structures were characterized electrically at room temperature by dc current–voltage (I–V ), capacitance–voltage (C–V ) (102 –106 Hz) and by internal electron photoemission measurements (IPE) [26]. The I–V curves of the MOS structures were measured not only in darkness but also under excitation with photons from an Arþ ion laser (hm ¼ 2:41–2:73 eV) to reveal the photon-stimulated electron transitions [27]. From the accumulation capacitance of MOS structures with as-deposited uniform Al2 O3 and ZrO2 layers, the low-frequency relative dielectric constants were found to be 8 and 15, respectively. Upon oxidation, the dielectric constants of both the interfacial silicon oxide and the high-permittivity oxide were found to increase [25]. More details regarding electrical properties of the ALCVD oxides have been published elsewhere [22,25,28–30]. The IPE spectral curves were measured in the photon energy range hm ¼ 2–5 eV (the photon energy was always kept well below the insulator bandgap width – 5.4 eV for ALCVD ZrO2 , Ref. [22]) with spectral resolution of 2 nm. IPE yield (Y ) was defined as photocurrent normalized to the incident photon flux [26,31]. For a positive metal bias the IPE from the silicon substrate is measured, which spectral dependence is independent of the type of metal used, as reported elsewhere [30]. When the metal is biased negatively, the current across the insulator is related to
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
electrons emitted from the metal gate and exhibits the corresponding spectral threshold variations in accordance with the metal/oxide barrier height changes. To avoid the contribution of electrons simultaneously photoinjected from the Si substrate to the current from the metal, the spectral measurements on thin (d < 10 nm) insulators were performed under a bias ranging from 2 to 4 V, i.e., high enough to prevent the Si-emitted electrons from reaching the gate. For the same purpose the IPE spectra from Si were mostly measured in the MOS structures with Au electrodes [30]. The density of electrons injected during the IPE experiments was kept below 1012 cm2 to avoid substantial trapping in the oxide. The CV measurements after the IPE experiments were used to affirm negligible charging.
3. Results and discussion 3.1. IPE spectra and thresholds The IPE spectra for n-Si/Al2 O3 /Au and n-Si/ ZrO2 /Au MOS structures (metal biased positively)
Fig. 1. Cube root of the IPE yield as a function of photon energy for MOS structures with different dielectrics: (a) as-deposited 5-nm thick Al2 O3 ( ) and oxidized at 650 °C for 30 min () or at 800 ° C for 10 min (M) as compared to a 4.1-nm thick thermal SiO2 (O); (b) as-deposited 7.4 nm ZrO2 ( ) and oxidized at 650 °C for 30 min () or at 800 ° C for 10 min (M). All curves are measured under an applied electric field of 2 MV/cm in the insulating layer closest to Si. The arrows E1 and E2 indicate onsets of direct optical transition in the Si crystal. The spectral thresholds UV are indicated for different oxides. The energy width of the monochromator slit is indicated in (a); the error in the IPE yield determination is smaller than the symbol size.
71
are shown in Fig. 1(a) and (b), respectively. Because replacement of the Au gate electrode with Al has no effect on the IPE curves, the latter correspond to electron emission from the Si valence band to the conduction band of the corresponding insulator, as shown in the insert in Fig. 1(b). This is affirmed independently by observation of the IPE spectrum modulation by direct optical transitions in the Si crystal [26,32] as indicated by arrows E1 and E2 in Fig. 1(a). These transitions cause deviation of spectral curves from the ap3 parent Y ðhm UÞ dependence thus limiting the spectral interval of linear fit and causing a substantial increase of the IPE threshold determination error as compared to the energy width of the monochromator slit (0.02 eV). For as-deposited Al2 O3 the IPE spectral threshold (UV ) appears to be slightly above 3 eV, while for ZrO2 it is 0.15 eV lower. To obtain the exact barrier energies, the spectral thresholds were measured at different strengths of electric field and then extrapolated to zero field in the Schottky coordinates, as shown in Fig. 2. The resulting barriers between the top of the Si valence states and the bottom of the oxide conduction states are 3:25 0:08 eV for Al2 O3 and 3:1 0:1 eV for ZrO2 as compared to the 4:25 0:05 eV barrier at the Si/SiO2 interface, which is also shown for comparison. The Schottky plot allows us to determine the effective
Fig. 2. Schottky plot of the spectral thresholds for IPE from the Si valence band into the conduction band of SiO2 layers of different thickness (in nm): 55 ( ), 5.8 (), 4.1 (M), as compared to 5-nm thick Al2 O3 (O) and 7.4-nm thick ZrO2 (}) layers. The symbol size corresponds to the error of the threshold determination.
72
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
image-force dielectric constant ei related to the refraction index n of the insulator near the Si surface (ei n2 ). From the slopes of the lines in Fig. 2, ei values are obtained as 2:1 0:2, 3:5 0:5, and 5 1 for SiO2 , Al2 O3 , and ZrO2 , respectively. Within experimental error, these values coincide with the square of the refractive index of bulk SiO2 (n ¼ 1:456), Al2 O3 (n ¼ 1:77), and ZrO2 (n ¼ 2:2) suggesting a physically abrupt transition between silicon and the oxide [30]. Comparison of the above results with earlier data on IPE from Si into Al2 O3 layers deposited at high temperature or subjected to high-temperature anneal [16,17,33] indicates that in the as-deposited state the barrier height is much reduced (3.25 eV vs. 3.9–4.1 eV). The 2.1 eV conduction band offset resulting from our value also appears to be significantly lower than the theoretically predicted 2.8 eV value [7], thus affirming the importance of interface effects. Upon oxidation, the major IPE emission shifts to the spectral range of hm > 4 eV, which suggests formation of a silicate layer as explanation of earlier experimental data. In the Si/ ZrO2 system the barrier increases to a similar value of 4 eV starting from 3.1 eV for the as-deposited sample. Interestingly, crystallization of ZrO2 observed upon oxidation at T > 500 °C [27] seems to have no substantial effect neither on the energy of the ZrO2 conduction band edge (with respect to energy bands of the Si substrate) nor on the bandgap width of oxide itself (spectra not shown). Oxidation results in reduction of the IPE yield in the spectral range of 3 < hm < 4:5 eV, which indicates a decrease in the density of states related to the conduction band states of the high-permittivity metal oxide. Instead, the spectral features characteristic for SiO2 become pronounced as the thickness of the thermally grown interlayer exceeds 1 nm. Worth noticing here is the fact that the IPE curves of the oxidized gate stacks do not reach the shape corresponding to the IPE into pure thermal SiO2 (O in Fig. 1(a)), not even at the highest oxidation thermal budgets applied. A significant ‘tail’ is observed in the sub-threshold region, particularly pronounced in the case of Al2 O3 . More details for ZrO2 are revealed by the spectral curves measured in the sub-barrier spectral range using Arþ laser excitation, shown in Fig. 3. This photo-
Fig. 3. IPE yield as a function of photon energy for n-Si/7.4 nm ZrO2 /Au MOS structures in as-deposited state ( ) and after oxidation at 500 °C for 60 min (), 650 °C for 30 min (M), and at 800 °C for 10 min (O). The measurements are done with an applied field strength in ZrO2 layer of 3 MV/cm, metal biased positively. The error in the yield determination is smaller than the symbol size.
current caused by excitation of electrons via the oxide conduction band tail states decreases with increasing oxidation temperature, but does not disappear entirely. Apparently, the electron exchange between Si and the band tail states in ZrO2 gets impeded upon growth of a SiO2 -like interlayer. However, the latter is likely to be a silicate containing a considerable density of gap states, which provide under-barrier pathways for electrons. Examples of spectral curves of IPE from a metal into 5-nm thick Al2 O3 and ZrO2 layers are presented in Fig. 4. For Al2 O3 only data for Al and Au are shown in Fig. 4(a), which are similar to the data reported by DiMaria for IPE into high-temperature deposited aluminum oxide [17]. For ZrO2 the spectra of IPE from Mg ( ), Al (), Ni (M), Cu (O), and Au (}) are presented in Fig. 4(b). As the IPE from the same metals into thermal SiO2 of the same thickness (curves not shown) obeys the 2 Fowler law [34] Y ðhm UÞ (in agreement with literature results for thicker oxides, see Refs. [35– 37]), Fowler co-ordinates ðY 1=2 hmÞ were used to determine the IPE spectral thresholds. A cubic fit Y 1=3 hm was found to be less good; it results in a systematic 0.1 eV red shift of the IPE threshold which is comparable to the error arising from the extrapolation inaccuracy. It needs to be mentioned
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
73
Table 1 Barrier heights (UV ; U) at the interfaces between (1 0 0)Si, metals, and SiO2 , Al2 O3 and ZrO2 insulators Insulator
SiO2 (0.05 eV) Al2 O3 (0.1 eV) ZrO2 (0.1 eV)
Fig. 4. (a) Square root of the IPE yield from the metal as a function of photon energy for MOS structures with Al and Au electrodes deposited on a 5-nm thick as-deposited (open symbols) and oxidized at 800 °C for 10 min (filled symbols) Al2 O3 insulator. The strength of the electric field in the oxide is (in MV/cm) as follows: Al on as-deposited oxide: 0.2 ( ), 0.5 (), 1.0 (M), 1.5 (O), 3.0 (}); Al on oxidized Al2 O3 : 0.5 ( ), 1.0 (j), 2.0 (N), 4.0 (.); Au on as-deposited oxide: 0.5 ( ), 1.0 (), 2.0 (M), 3.0 (O), 4.0 (}), metal bias is negative. (b) Square root of the IPE yield from the metal as a function of photon energy for MOS structures with different metals deposited on a 5-nm thick as-deposited ZrO2 layer. The electric field strength in the oxide is 2 MV/cm, metal biased negatively. The lines illustrate the determination of the IPE spectral thresholds. The energy width of the monochromator slit is indicated in (b); inaccuracy of the IPE yield determination is smaller than the symbol size.
here is that the power of the yield dependence on photon energy is determined exclusively by the shape of the energy distribution of excited electrons in the emitter and by the energy dependence of the barrier surmount probability [38]. For electrons emitted from the same metal, their energy distribution must be the same and the variation of the spectral curves will reflect mostly the properties of the potential barrier. Therefore, deviation of the IPE spectra from the Fowler law in the near-threshold region (Fig. 4) indicates a lateral non-uniformity of the interfacial barrier. Two more features are revealed by the data shown: First, the thermal oxidation at a temperature as high as 800 °C does not change the metal/insulator barrier significantly as compared to the as-deposited state (cf. data for Al in Fig. 4 (a)). Second, the electric field has only a weak effect on spectral thresholds (within the indicated accuracy limit of 0.1 eV), suggesting a smaller Schottky barrier reduction as compared to the IPE from Si into Al2 O3 and ZrO2 (cf. Fig. 2). The barrier heights U are
Emitter (1 0 0)Si
Mg
Al
Ni
Cu
Au
4.25 3.25 3.1
2.50 2.6 2.6
3.15 2.9 2.7
3.70 3.5 3.2
3.85 3.6 3.25
4.10 4.1 3.5
listed in Table 1 for the studied insulator/metal pairs, and UV values are given for (1 0 0)Si/insulator interfaces. To enable a meaningful comparison between different oxides, the measured barrier heights are plotted in Fig. 5 together with literature data for SiO2 and Al2 O3 as a function of the Pauling metal electronegativity XM [39]. For thermal SiO2 , panel (a), the barriers measured on 5-nm thick oxides () perfectly follow the trend of barrier increase with the metal electronegativity reported earlier for thicker oxides (symbol from Ref. [35], [36] for W). In the case of Al2 O3 , panel (b), the literature data (symbol from [17], [33] for W), on average, lie slightly above the values determined for both as-deposited () and thermally oxidized (M) AL CVD layers, though for Mg and Au the agreement is perfect. Taking into account the spread of the data, the increase of the interfacial barrier height with the metal electronegativity appears to be the
Fig. 5. Barrier height at the metal/insulator interface as a function of the gate metal electronegativity for 5-nm thick thermal SiO2 (a), 5-nm thick ALCVD Al2 O3 (b) and ZrO2 (c). Circles indicate the literature values for thicker oxides, squares refer to the experimental results on as-fabricated oxide layers, triangles refer to the ALCVD layers oxidized in O2 at 650 °C for 30 min. Lines guide the eye; the symbol size corresponds to 0.1 eV inaccuracy in barrier height determination.
74
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
same as in the case of SiO2 and thus close to the ideal case dU=dXM ¼ 1 [40]. For both as-deposited () and oxidized (M) ZrO2 layers, panel (c), the increase of the barrier height with metal electronegativity is found to be considerably smaller (ideality factor dU=dXM 0:75Þ than for SiO2 and Al2 O3 . 3.2. Work function differences On the basis of the barrier values listed in Table 1 the metal/silicon work function difference /ms was calculated for both the n- and p-type Si samples used in the present study. The results are shown by lines in Fig. 6 and are compared with the flatband voltage (VFB ) values determined from the high-frequency C–V curves in the n- (filled symbols) and p-type (open symbols) MOS structures with Al2 O3 (a), Al2 O3 /ZrO2 (b), and ZrO2 (c) dielectrics, both as-deposited (circles) and oxidized at 650 °C for 30 min (squares) or at 800 °C for 10 min (triangles). The dotted lines on the panels indicate the /ms behavior in n-Si MOS capacitors with thermally grown SiO2 dielectric. The comparison between /ms and VFB values reveals three features: First, for the same metal, /ms values measured in the MOS structures with thermal SiO2 are considerably smaller than those in MOS structures with an ALCVD deposited insulator. Sec-
Fig. 6. Flatband voltages of n- (filled symbols) and p-type (open symbols) MOS capacitors with 5-nm thick Al2 O3 (a), stacked 1.5 nm Al2 O3 /5 nm ZrO2 (b), and 5-nm thick ZrO2 (c) insulators as a function of gate metal electronegativity. Solid and dashed lines show correspondingly the behavior of the metal–semiconductor work function difference /ms for n- and ptype Si, dotted lines indicate /ms for an n-type MOS structure with thermal SiO2 insulator. The error of VFB determination is smaller than the symbol size.
ond, the systematically observed trend VFB > /ms for the as-deposited ALCVD layers indicates the presence of a substantial density of negative charge in the metal oxides, particularly in ZrO2 -based insulators. The density of this charge is effectively reduced by oxidation of the deposited dielectric. Third, and by contrast, the Al-gated samples exhibit a systematic VFB shift to below /ms indicating a positive charge buildup in the insulator upon Al evaporation. The latter may be related, for instance, to a chemical interaction of Al with the Hcontaining fragments at the oxide surface, leading to the formation of protons [25]. The important result of this /ms study consists in revealing a large difference in the values measured in MOS structures with thermal SiO2 and with an ALCVD deposited insulator, which indicates the presence of a significant interface dipole. Growth of a thin silicon oxide at the interface between the Si substrate and the ALCVD layer has no impact neither on the Si/ALCVD oxide band alignment nor on this dipole (cf. Fig. 6) suggesting that the latter is associated with the metal/oxide interface. The latter suggestion finds support from the observed non-idealities in the IPE spectra of the metal/oxide interfaces. First, the tails of the Y hm curves refer to non-uniformity of the barrier. As may be seen from Fig. 4, they show a clear trend to increase for metals with a lower barrier (Al, Mg) as compared to Au. The latter suggests that, in addition to the conduction band tail states [30], the barrier may also be affected by a nonuniform polarization layer possibly related to negative charge arising from electrons trapped in the oxide at levels below the metal Fermi level [41]. Next, a weak dependence of the metal/oxide barrier height on the strength of electric field in the oxide suggests the presence of a strong built-in electric field which is also consistent with the formation of an interfacial polarization layer. Finally, the weaker U-electronegativity dependence for ZrO2 (cf. Fig. 5(c)) may also be related to the relative enhancement of the negative dipole for the metals with smallest barriers (Mg, Al). The latter effect may be of concern for practical applications of the ALCVD gate insulators: the threshold voltage control by using nþ /pþ -poly-Si gates or metal electrodes with different work function may
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
appear not so efficient as in the MOS devices with a SiO2 gate dielectric. 3.3. Electron tunneling vs. trap-assisted transport The determined interface barrier heights and the contact potential values allowed us to analyze in detail the I–V curves of MOS structures with high-e dielectrics. Examples of such curves for nSi/ZrO2 /Au structures (positive metal bias) are shown in Fig. 7(a) for the as-deposited samples ( ) and oxidized at various thermal budgets (500 °C for 60 min (); 650 °C for 30 min (M); 800 °C for 10 min (O)) samples as well as one annealed in high vacuum (650 °C, 30 min, N). Similar curves were observed when Au was replaced with another metal indicating that the measured current is due to electron injection from Si substrate. Apparently, oxidation of the Si/ZrO2 structure results in significant reduction of the leakage current probably related to the growth of a silicon oxide at the interface. The importance of oxygen is clearly indicated by lack of current reduction upon thermal treatment in high vacuum. To reveal the injection mechanism, the Fowler–Nordheim (FN) plots [42] of the dark (filled symbols) current and the photocurrent measured under excitation with 2.71
Fig. 7. (a) I–V curves (positive metal bias) of n-Si MOS structures with 7.4-nm ZrO2 insulator in as-deposited state ( ), after oxidation at 500 °C for 60 min (), at 650 °C for 30 min (M), or at 800 °C for 10 min (O), compared to the sample annealed in high vacuum at 650 °C for 30 min (N). (b) The Fowler–Nordheim plots of dark (filled symbols) and photocurrent (2.71 eV, 100 mW, open symbols) measured in n-Si MOS structures with 7.4-nm ZrO2 insulator in as-deposited state (circles), after oxidation at 500 °C for 60 min (squares), at 650 °C for 30 min (triangles), or at 800 °C for 10 min (inverted triangles) under positive metal bias.
75
eV photons at incident power of 100 mW (open symbols) are shown in Fig. 7(b) for both as-deposited and oxidized samples. It is clearly seen that with increasing oxidation temperature both currents decrease dramatically. Also, in the low-field range the currents do not exhibit the FN behavior typical for electron tunneling, suggesting a trapassisted electron transport. The slope of the FNtype I–V curves observed at high field values increases with oxidation temperature approaching that for the case of FN tunneling from Si into SiO2 . The latter indicates the formation of a SiO2 like interlayer as the major reason for the reduction of the leakage current in the oxidized Si/ZrO2 structures. The influence of an interlayer between the highe insulator and silicon is also seen from the FN plots of the dark and photocurrent (hm ¼ 2.71 eV, 100 mW) measured in n-type Si MOS structures with as-deposited and oxidized stacks of 1.3 nm thermal SiO2 /5 nm ZrO2 and 1.5 nm Al2 O3 /5 nm ZrO2 , shown in Fig. 8(a) and (b), respectively. The presence of a pre-grown SiO2 layer (Fig. 8(a)) significantly reduces the leakage current already in as-deposited (not oxidized) sample. After oxidation, a further drop of the trap-related currents in the low-field range is observed, but it has only a marginal influence on the high-field electron tunneling. The reduction of both the dark and
Fig. 8. The FN plots of dark (filled symbols) and photocurrent (2.71 eV, 100 mW, open symbols) measured in n-Si MOS structures with stacks of 1.3 nm SiO2 /5 nm ZrO2 (a) and 1.5 nm Al2 O3 /5 nm ZrO2 ; (b) insulators in as-deposited state (circles), after oxidation at 500 °C for 60 min (squares), at 650 °C for 30 min (triangles), or at 800 °C for 10 min (inverted triangles) under positive metal bias.
76
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
photocurrent by the Al2 O3 interlayer (Fig. 8(b)) is not as strong as by SiO2 suggesting a lower interfacial barrier for electrons at the Si/Al2 O3 than at the Si/SiO2 interface, in accordance with the IPE results. Only after oxidation at 800 °C the currents typical for the SiO2 /ZrO2 stack are reached. A behavior very different from the one discussed above with respect to oxidation is revealed by I–V curves of electron injection from Au into ZrO2 measured in p-type Si MOS structures (metal biased negatively) and shown in Fig. 9(a) as FN plot. The dark current (filled symbols) is barely affected by oxidation. The laser-induced current (open symbols) decreases in the low-field range with increasing temperature of oxidation suggesting a reduction in the density of traps available for photoionization; yet, there is little change at high field. The FN plots of the dark current measured in similar MOS structures with Al (open symbols) and Au (filled symbols) electrodes are compared in Fig. 9(b) (three curves are shown for each sample). Despite significant difference in the barrier height for electrons at the interfaces of ZrO2 with Al and Au (cf. Table 1), hardly any variation in the slope of the FN plots is observed in Fig. 9(b). The latter indicates that the emission of an electron from the
Fig. 9. (a) FN plots of dark (filled symbols) and photocurrent (2.71 eV, 100 mW, open symbols) measured in p-Si MOS structures with 5 nm ZrO2 insulators in as-deposited state (circles), after oxidation at 500 °C for 60 min (squares), at 650 °C for 30 min (triangles), or at 800 °C for 10 min (inverted triangles) under negative metal bias. (b) FN plots of dark current measured in p-Si MOS structures with 5 nm ZrO2 insulators and Al (open symbols) and Au (filled symbols) electrodes in as-deposited state (circles), after oxidation at 650 °C for 30 min (squares), or at 800 °C for 10 min (triangles) under negative metal bias. Three curves are shown for each sample.
metal into the conduction band of ZrO2 does not constitute the rate-limiting step. Rather, the insensitivity to the metal type suggests that the current is determined by emission of an electron from some deep, stable against oxidation, gap state in ZrO2 . This may account for a lower current from Al than from Au measured at nominally the same electric field strength at the metal/ZrO2 interface (cf. Fig. 9 (b)) despite the 0.8 eV lower barrier height for Al as compared to Au (Table 1). Apparently, a negatively charged polarization layer is formed at the Al/ZrO2 interface effectively reducing the electron emission probability from oxide traps. This picture is consistent with the non-ideal behavior of the IPE characteristics discussed earlier and points towards the trap-assisted tunneling as a dominant conduction mechanism in ALCVD Al2 O3 and ZrO2 layers.
4. Conclusions IPE analysis of interfaces of (1 0 0)Si and various metals with thin thermally grown SiO2 and ALCVD Al2 O3 and ZrO2 insulators provided directly the electron barrier energies between the filled states of the metals and Si and the conduction band of the respective insulators. The asgrown Al2 O3 and ZrO2 layers show a barrier with Si by 1 eV lower than SiO2 . In metal/Al2 O3 contacts, the barrier height increases with electronegativity of the metal in the nearly ideal way dU=dXM 1, while for as-deposited ZrO2 insulators a less ideal behavior is observed with dU= dXM 0:75. The obtained barrier values remain unchanged down to the oxide layer thickness of 3 nm for Al2 O3 and ZrO2 , and 1 nm for SiO2 [30]. The metal/silicon work function differences in the MOS structures determined on the basis of the measured barrier heights appear to be different from those for the SiO2 insulator. This is ascribed to the formation of a negative dipole layer at the interfaces of metals with ALCVD oxides indicating the presence of a substantial density of electron traps. The traps present in ALCVD insulators determine the electron transport and their influence can be efficiently reduced by introduction of a SiO2 interlayer. The latter suggests that the observed
V.V. Afanas’ev et al. / Journal of Non-Crystalline Solids 303 (2002) 69–77
insulating properties of the high-e dielectric stacks have not yet reached their fundamental limit determined by band offsets at the interfaces and can thus be improved further.
References [1] R.A.B. Devine, L. Vallier, J.-L. Autran, P. Paillet, J.-L. Leray, Appl. Phys. Lett. 68 (1996) 1775. [2] O. Nakagawara, Y. Toyota, M. Kobayashi, Y. Yoshino, Y. Katayama, H. Tabata, T. Kawai, J. Appl. Phys. 80 (1996) 388. [3] G.B. Alers, R.M. Fleming, Y.H. Wang, B. Dennis, A. Pinczuk, G. Redinbo, R. Urdahl, E. Ong, Z. Hasan, Appl. Phys. Lett. 72 (1998) 1308. [4] G.D. Wilk, R.M. Wallace, J.M. Anthony, J. Appl. Phys. 87 (2000) 484. [5] M. Copel, M. Gribelyuk, E. Gusev, Appl. Phys. Lett. 76 (2000) 436. [6] J. Robertson, C.W. Chen, Appl. Phys. Lett. 74 (1999) 1168. [7] J. Robertson, J. Vac. Sci. Technol. B 18 (2000) 1785. [8] J. Kadlec, K.H. Gundlach, Solid State Commun. 16 (1975) 621. [9] A.I. Braunstein, M. Braunstein, G.S. Picus, C.A. Mead, Phys. Rev. Lett. 14 (1965) 219. [10] A.I. Braunstein, M. Braunstein, G.S. Picus, Phys. Rev. Lett. 15 (1965) 956. [11] F.L. Schuermeyer, J.A. Crawford, Appl. Phys. Lett. 9 (1966) 317. [12] O.L. Nelson, D.E. Anderson, J. Appl. Phys. 37 (1966) 77. [13] A.M. Goodman, J. Appl. Phys. 41 (1970) 2176. [14] W. Ludwig, B. Korneffel, Phys. Status Solidi 24 (1967) K137. [15] K.W. Shepard, J. Appl. Phys. 36 (1965) 796. [16] N. Szydlo, P. Poirier, J. Appl. Phys. 42 (1971) 4880. [17] D.J. DiMaria, J. Appl. Phys. 45 (1974) 5454. [18] D.J. DiMaria, E. Cartier, J. Appl. Phys. 78 (1995) 3883. [19] A. Scarpa, A. Paccagnella, F. Montera, G. Gibaudo, G. Pananakakis, G. Ghidini, P.G. Fuochi, IEEE. Trans. Nucl. Sci. NS-44 (1997) 1818. [20] V.V. Afanas’ev, A. Stesmans, J. Electrochem. Soc. 146 (1999) 3409.
77
[21] C. Svensson, I. Lundsrom, J. Appl. Phys. 44 (1973) 4657. [22] M. Houssa, M. Tuominen, M. Naili, V.V. Afanas’ev, A. Stesmans, S. Haukka, M.M. Heyns, J. Appl. Phys. 87 (2000) 8615. [23] M. Ritala, M. Leskela, Appl. Surf. Sci. 75 (1994) 333. [24] M. Ritala, K. Kukli, A. Rahtu, P.I. Raisanen, M. Leskela, T. Sajavaara, J. Keininen, Science 288 (2000) 319. [25] M. Houssa, M. Naili, C. Zhao, H. Bender, M.M. Heyns, A. Stesmans, Semicond. Sci. Technol. 16 (2001) 31. [26] V.K. Adamchuk, V.V. Afanas’ev, Progr. Surf. Sci. 41 (1992) 111. [27] V.V. Afanas’ev, A. Stesmans, Phys. Rev. Lett. 78 (1997) 2437–2440. [28] M. Houssa, A. Stesmans, M. Naili, M.M. Heyns, Appl. Phys. Lett. 77 (2000) 1381. [29] M. Houssa, V.V. Afanas’ev, A. Stesmans, M.M. Heyns, Appl. Phys. Lett. 77 (2000) 1885. [30] V.V. Afanas’ev, M. Houssa, A. Stesmans, M.M. Heyns, Appl. Phys. Lett. 78 (2001) 3073. [31] V.V. Afanas’ev, A. Stesmans, M.O. Andersson, Phys. Rev. B 54 (1996) 10820. [32] T.H. Di Stefano, J.E. Lewis, J. Vac. Sci. Technol. 11 (1974) 1020. [33] R. Ludeke, M.T. Cuberes, E. Cartier, Appl. Phys. Lett. 76 (2000) 2886; R. Ludeke, M.T. Cuberes, E. Cartier, J. Vac. Sci. Technol. B 18 (2000) 2153. [34] R.H. Fowler, Phys. Rev. 38 (1931) 45. [35] B.E. Deal, E.H. Snow, C.A. Mead, J. Phys. Chem. Solids 27 (1966) 1873. [36] R.J. Powell, R.C. Beairsto, Solid State Electron. 16 (1973) 265. [37] P.M. Solomon, D.J. DiMaria, J. Appl. Phys. 52 (1981) 5867. [38] R.J. Powell, J. Appl. Phys. 41 (1970) 2424. [39] L. Pauling, The Nature of the Chemical Bond, 3rd Ed., Cornell University, Ithaca, NY, 1960. [40] S. Kurtin, T.C. McGill, C.A. Mead, Phys. Rev. Lett. 22 (1969) 1433. [41] C.J. Wang, T.H. DiStefano, CRC Crit. Rev. Solid State Sci. 5 (1975) 327. [42] R.H. Fowler, L. Nordheim, Proc. Roy. Soc. A 119 (1928) 173.