Tailoring Al-SiO₂ interfacial work function using an organophosphonate nanolayer

Metal-oxide interfaces are playing increasingly important roles in nanoelectronics, actuators, sensors, photonics, and power conversion as devices become smaller and more sophisticated.^1–6 Hence, understanding and manipulating the electronic properties of inorganic interfaces, e.g., the effective work function Φ_eff at metal-ceramic interfaces, will facilitate nanoelectronic applications. Nanomolecular layers (NMLs) are an attractive tool to modify the multiple properties of inorganic interfaces.^7–16 For example, we have shown that introducing organosilane NMLs at metal-ceramic interfaces can alter interfacial phase formation paths,^7,8 inhibit diffusion,^9,10 and lead to multifold enhancements in interfacial fracture energy^9,11 and thermal transport.^12,13 We have also reported electrostatic effects of chlorosilane NMLs on metal-oxide-semiconductor capacitors^14,15 and shown that organophosphonate NMLs can influence the Φ_eff at Au-HfO₂ interfaces.¹⁶ 

Here, we report on the effects of chemical termini of organophosphonate NMLs on Φ_eff at Al-NML-SiO₂ interfaces using ultraviolet photoelectron spectroscopy (UPS) and capacitance-voltage (C-V) measurements (Fig. 1). Unlike our previous work¹⁶ that reported Φ_eff changes in Au-NML-HfO₂ structures, here, we use C-V measurements to capture Φ_eff at Al-SiO₂ interfaces that are susceptible to oxidation. We find that a mercaptan-terminated organophosphonate NML decreases Φ_eff by ∼0.67 eV, while a methyl-terminated NML produces no observable Φ_eff shifts, underscoring the key role of the metal-NML bond. X-ray photoelectron spectroscopy (XPS) measurements indicate that the Φ_eff shift is due to Al-O-S bond formation at the Al-NML interface. Such Φ_eff shifts arising from metal-NML bonding could be potentially leveraged to tailor the electronic properties of inorganic interfaces for applications.

The substrates used in this study were p-type Si(001) wafers doped with 1.5 × 10¹⁵ cm⁻³ boron and capped with a 55-nm-thick thermally grown SiO₂ layer. We obtained a gradation of SiO₂ layer thickness ranging from 35 nm to 55 nm, by etching the silica by gradually withdrawing wafers immersed in a ratio of 1:10 by volume of (5.8 M) HF solution, as detailed elsewhere.¹⁴ The graded SiO₂ layer allowed us to test capacitors with different effective oxide thicknesses (EOTs) on the same wafer, thereby minimizing statistical variance in work function determined from C-V measurements described later below. We note that work function determination by UPS is independent of oxide thickness, and a flat layer is desirable to minimize uncertainties due to changes in the escape angle of the photoelectrons.

The wafers were successively washed with ethanol, in a 3:1 ratio by volume of H₂SO₄:H₂O₂ piranha solution at a temperature between 70 and 90 °C for 45 min, with deionized water, and then dried under ultrahigh purity (99.999%) N₂. We deposited 12-mercaptododecyl-phosphonic acid (MDPA) and dodecyl-phosphonic acid (DDPA) NMLs on the silica surface via the tethering by aggregation and growth method.^17,18 In particular, the substrate surfaces were immersed in 1 mM ethanolic solutions of the organophosphonates in a vacuum oven (∼1 × 10⁻³Torr) set at 70 °C for 6 h. After evaporation of the ethanolic solution, the NML-coated wafers were annealed at 140 °C for 1 h without breaking vacuum, then rinsed, and sonicated in ethanol, and immediately transferred to a 9.0 × 10⁻⁷Torr base pressure thermal evaporator for Al deposition.

A 20- to 30-nm-thick Al film was evaporated onto the wafers at a deposition rate of ∼0.1 nm s⁻¹ through a stainless steel contact mask. Silver paint was used to make the back contact to the Si(001) substrate. We carried out C-V measurements on metal-NML-SiO₂-Si(100) capacitors with Ag-paint back-contacts in a light-sealed chamber using an HP 4284A LCR meter at 500 kHz. The metallized area of each capacitor was measured using an optical microscope.

Similarly prepared wafers, but with only a 3 nm-thick Al film, were transferred from the evaporator to the PHI 5400 instrument with a 15-min air-exposure for UPS and XPS measurements. We carried out UPS measurements with a –9.00 V surface bias through a circular Mo mask for an improved spectral resolution at low photoelectron kinetic energies. Sample charging was countered using ∼1.2 eV beams of electrons and Ar⁺ at 20 μA. We acquired core-level spectra by XPS from NML-functionalized SiO₂ using an Al Kα probe beam at preselected surface-to-detector takeoff angles θ_sd and at analyzer pass energy of 23.5 eV. The spectral resolution of the XPS was 0.5 eV, while for UPS, the resolution was 0.15 eV. The background was subtracted using the Shirley method,¹⁹ and the 284.8 eV adventitious C 1s peak was used as an internal reference for correcting spectral shifts due to residual charging.

The NML-treated SiO₂ surfaces exhibit the P 2p signature at 133.2 eV, indicative of surface-bound phosphonate moieties in the NML [Fig. 2(a)]. MDPA-capped SiO₂ surfaces show [see Fig. 2(b)] two S 2p sub-bands corresponding to pristine mercaptan (164.5 eV) and oxidized (∼30%–35%) mercaptan (168.0 eV).²⁰ The Al-SiO₂-Si-Ag capacitors reveal two Al 2p sub-bands: a 74.8 eV aluminum oxide²¹ signature and a 72.3 eV peak corresponding to unoxidized Al. The Al-SiO₂ and Al-DDPA-SiO₂ structures with the 3-nm-thick Al, used for UPS, also show both Al peaks. However, Al-DDPA-SiO₂ structures exhibit a 24% metallic Al sub-band, in contrast to Al-MDPA-SiO₂ structures that essentially reveal only the oxidized Al peak. We attribute the absence of metallic Al at the Al-MDPA interface to Al-O-S bridges formed through the reaction of Al with the SO_x moieties formed via mercaptan oxidation prior to Al deposition. Similar Al oxidation by ester- or alcohol-terminated monolayers has previously been reported.^22,23

Variable takeoff angle XPS on NML-capped SiO₂ surfaces before Al deposition reveal a greater attenuation of the P 2p peak with decreasing θ_sd, than the C 1s or the S 2p bands. This result indicates that the phosphonic acid moieties in both MDPA and DDPA are proximal to the SiO₂ surface, anchoring the molecules to SiO₂ via P-O-Si bonds (Fig. 3). The NML thickness t_NML was determined by fitting the Si 2p core-level peak intensity to

where I and I₀ are the Si 2p peak intensities corresponding to NML-covered and bare SiO₂, respectively, and λ = 3.95 nm is the experimentally determined²⁴ Si 2p photoelectron mean free path through an alkane-chain monolayer. We find that t_MDPA=1.25 nm and t_DDPA=0.74 nm, which are significantly lower than the respective molecular lengths l_MDPA = 1.81 nm and l_DDPA = 1.66 nm, suggesting that both organophosphonates form monolayers with average tilt angle θ_MDPA=46° and θ_DDPA=64° with respect to the surface normal.

Estimates of the NML coverage N_NML from our variable take-off angle XPS data yield N_MDPA ∼3.8 ± 1.1 × 10¹⁴ cm⁻² and N_DDPA ∼2.3 ± 1.5 × 10¹⁴ cm⁻². The lower coverage of the DDPA may also explain the lower t_NML despite similar theoretical lengths l_NML of both molecules (Fig. 3). For N_NML determination, we assumed uniform surface coverage and used

where I_P2p and I_Si2p and x_P2p and x_Si2p are the integrated intensities and sensitivity factors of P 2p and Si 2p bands, respectively, ρ_SiO2 = 2.25 × 10²² cm⁻³ is the number density²⁵ of thermal SiO₂, and λ_Si = 3.8 nm is the Si photoelectron mean free path²⁶ through SiO₂.

NML-induced Φ_eff shifts (ΔΦ_eff) were determined from the flat-band voltage V_FB shifts from the C-V characteristics of Al-NML-SiO₂-Si-Ag and Al-SiO₂-Si-Ag structures, at different effective oxide thicknesses, EOT [see Fig. 4(a)]. We extracted the NML dielectric constants k_MDPA =1.4 ± 0.5 and k_DDPA=2.87 ± 3.5 by comparing the capacitances with and without NMLs at a given EOT. Figure 4(b) shows V_FB vs. EOT plots for Al-NML-SiO₂-Si and Al-SiO₂-Si capacitors. The V_FB was determined by the second derivative method²⁷ and the standard flat-band capacitance method²⁸ both of which yield V_FB values within ± 0.05 V. We then obtain Φ_eff = 3.3 ± 0.13 eV for MDPA and Φ_eff = 4.0 ± 0.17 eV for DDPA by invoking the relationship between V_FB, the effective metal work function $Φeff$⁠, the semiconductor work function $ΦS$⁠, the SiO₂ relative dielectric permittivity $εox$⁠, and the SiO₂-Si interface charge density Q_f using

and

Figure 5 shows the SiO₂ work function shift $ΔΦSiO2$⁠, determined from UPS measurements of Al-NML-SiO₂ and NML-capped SiO₂ structures. We note that $ΔΦSiO2=−ΔΦeff$ because UPS measures electron ejection from SiO₂ whereas C-V measurements measure electron injection into SiO₂. Since the Fermi level E_F of SiO₂ is within the bandgap, we calculated $ΔΦSiO2$ from the difference in E_SEO, the secondary electron onset edge. In particular, we used

and

where $hν$ is the He¹ photon energy. Before taking the difference in E_SEO, we aligned the UPS spectra with the highest occupied molecular orbital energy E_HOMO [Fig. 5(b)] to eliminate any residual charging effects.

Both UPS and C-V measurements yield similar NML-induced changes in Φ_eff [Fig. 5(c)]: ΔΦ_eff ∼0 for DDPA and ΔΦ_eff = –0.67 eV for MDPA. Such agreement despite different degrees of Al oxidation in the patterned capacitors with thick metal films used for C-V measurements, and blanket structures with thin metal films used for UPS, suggests that the Al-NML interfaces in both structures are chemically similar. At Al-NML-SiO₂ interfaces, we can express ΔΦ_eff as

where N_NML is the molecular coverage, e is the electronic charge, $δinterface$ is the interfacial dipole moment due to a molecule in the NML, $and ε0$ is the dielectric permittivity of vacuum. By inputting measured ΔΦ_eff, the N_NML and $kNML$ for MDPA and DDPA, we find that $δinterface$= 0.6 ± 0.1 D for MDPA, while $δinterface$= 0 for the DDPA.

The total interface dipole $δinterface$ can be modeled as a sum of the dipoles due to Al-NML (⁠$δAl−NML$⁠) and NML-SiO₂ (⁠$δNML−SiO2)$ bonds, and the body of the molecule $δbody$⁠, as

The UPS and C-V measurements of Al-NML-SiO₂ include all three contributions, while UPS measurements of NML-capped SiO₂ surfaces do not include $δAl−NML$ [Fig. 5(a)]. We can thus extract $δAl−NML$ from the difference between UPS measured $ΔΦSiO2$ of the Al-NML-SiO₂ and NML-SiO₂. Figure 5(c) shows that $δAl−MDPA$ = 0.9 ± 0.3 D, while $δAl−DDPA$ = 0.2 ± 0.6 D.

Our result showing a 50% higher $δAl−MDPA$ than $δinterface$ implies that the net $δbody+δNML−SiO2$ dipole is oriented opposite to the Al-MDPA dipole $δAl−MDPA$ [see Fig. 5(c) inset]. The higher $δAl−MDPA$ is attributable to Al bonding with oxidized mercaptan via Al-O-S bridges, which facilitate electron injection into the SiO₂. While the $δinterface$ = 0 for DDPA, the constituent $δbody+δNML−SiO2$ dipoles are nonzero, as measured from $ΔΦSiO2$ of DDPA-SiO₂ interface without Al. Thus, the weak Al-CH₃ interactions at Al-DDPA-SiO₂ interfaces are equal but opposite to the small $δbody+δNML−SiO2$ contributions. Our results therefore indicate that one can develop a detailed model of how NMLs influence Φ_eff by measuring structures at various stages of metal-oxide-semiconductor capacitor fabrication.

In summary, we have shown that the Φ_eff of Al-NML-SiO₂ interfaces can be tuned through the terminal moiety of the organophosphonates by altering the Al-NML bonding. MDPA modification decreases Φ_eff by 0.67 eV due to Al-O-S bridges at the Al-NML interface, while DDPA produces no detectable change in Φ_eff, due to weak Al-CH₃ interactions. While the effect of the Al-NML bond dipole on Φ_eff is dominant in MDPA, in DDPA, the cumulative NML body and NML-SiO₂ dipoles are equal but opposite to the Al-NML bond dipole. These results indicate that controlling metal-NML bonding can serve as a means to reap large ΔΦ_eff that span more than half the Si band gap. Further fine-tuning of Φ_eff can be achieved by manipulating the NML length, coverage and morphology, as shown earlier.^29,30 Our results should provide greater insight into tailoring electronic properties at metal-ceramic interfaces.

We gratefully acknowledge the National Science Foundation under Grant ECCS 1002282/301.