Introduction

Conventional silicon doping is increasingly impeded due to the spatial dimensions approached by nanotechnology. Several fundamental physical principles counteract the substitutional incorporation of dopant impurities (e.g. B, P, or As) on Si lattice sites and their ionization to become electronically active donors of majority charge carriers. The formation energy for the substitutional dopant integration1,2 as well as the ionization energy3,4,5,6 increase strongly with decreasing Si dimensions. Moreover, conventional doping of Si nanoscale devices faces severe technological challenges: diffusion-related, steep radial gradients in the doping profile of Si nanowires7,8,9; the inadvertent but inevitable diffusion of source/drain dopants into field-effect transistor (FET)-channels10; surface segregation and inactivation of dopants11; and statistical fluctuations by random numbers/positions of dopants in Si nanovolumes12,13,14. These problems render conventional Si doping unsuitable for future nanoelectronics.

As an alternative to classical impurity doping, modulation doping of III-V semiconductors was introduced in the late 1970s15. Homostructure modulation doping was proposed for Si and Ge nanowires using conventional dopants in a Si shell around nanowires, allowing for separation of majority charge carriers from their parent donor impurities16,17. However, this approach does not solve issues of dopant inter-diffusion, deficient dopant ionization or statistical distributions of dopant number and position. An impressive work-around, referring back to junctionless FETs (Lilienfeld, 1925), was demonstrated by Colinge et al.18. This approach solves the problem of highly abrupt p-n junctions, though ultimately scaled devices again suffer from nanoscale-Si doping obstacles. Hypervalent doping of free-standing Si quantum dots was demonstrated using chemical surface engineering19. However, so far this approach is not at all CMOS compatible. Molecular monolayer doping20,21 was developed to achieve ultra-shallow p-n junctions via RTA-diffusion. Since this concept is based on classical impurity dopants, its applicability in low-dimensional systems is subject to the same constraints outlined above.

Here, we demonstrate a novel approach: modulation doping of Si by aluminium (Al)-induced acceptor states in silicon dioxide (SiO2). Using computer-aided materials design, we predict that Al atoms in SiO2 generate acceptor states 0.8 eV below the Si valence band edge. These states capture electrons from Si over a distance of several nanometres, providing holes to Si as majority charge carriers. We confirm this concept experimentally via capacitance-voltage (CV) and deep level transient spectroscopy (DLTS) on SiO2:Al/Si-based MOS capacitors. This doping technique is robust against out-diffusion, quantum confinement, dielectric confinement, and self-purification since it relocates doping from confined Si nanovolumes to adjacent bulk-like SiO2.

Results

Density functional theory simulations

Acceptor modulation doping of SiO2 was modelled via real-space density functional theory (DFT). We embedded a Si10 nanocrystal22 in three monolayers (MLs) of SiO2, presenting the ultimate theoretical test of the doping concept because Si nanocrystal, SiO2 and modulation acceptor (Al) form one approximant. Figure 1a and b show the highest occupied (HO) molecular orbital (MO) and the lowest unoccupied (LU) MO associated with Al. High probability densities occur within the Si10 nanocrystal and at the Al acceptor, although both are separated by 3 ML of perfect SiO2. This finding shows that the Si10 nanocrystal can still be positively ionized despite significant quantum confinement-induced bandgap widening. Figure 1c shows the electronic density of states (DOS) of the Si10 nanocrystals in pure SiO2 vs. SiO2:Al, along with the HOMO and LUMO of a 1.9 nm Si nanocrystal, fully terminated with hydroxyl (OH) groups, featuring SiO2 embedding23. The Al acceptor state (β-LUMO) exists 0.26 eV below the HOMO of the 1.9 nm Si nanocrystal, clearly showing that Si nanocrystals of d ≥ 1.9 nm can obtain positive majority charge carriers (holes) from Al acceptors in SiO2. Transferring the situation to bulk Si, the Al acceptor state is located ~0.8 eV below the valence band edge. The principle of direct acceptor modulation doping is shown in the electronic band structures in Fig. 1d for Si nanocrystals and in Fig. 1e for bulk Si.

Figure 1: DFT results on SiO2 modulation doping with Al acceptors.
figure 1

(a,b) Si10 nanocrystal (cyan) in three ML SiO2 (O is red, Si grey, H white) with Al atom (yellow) replacing Si in outermost SiO2 shell, showing β-HOMO (a) and β-LUMO (b) as iso-density plots of 4 × 10−4 e/cubic Bohr radius. (c) Electronic DOS (blue, red) of that approximant (energy scale refers to vacuum level Evac). DOS for pure SiO2 embedding (cyan, orange) is shown for comparison along with the HOMO and LUMO of a 1.9 nm Si nanocrystal fully terminated with OH groups22. The two possible MO spin orientations are noted by α and β due to unpaired electron configuration caused by the acceptor (doublet). (d) Band structures showing the principle of direct modulation doping for a Si nanocrystal (NC) in SiO2 and (e) for Si bulk terminated with SiO2:Al layer. Occupation of electronic states are described by chemical potential μ and Fermi level EF, respectively.

Full size image

Electrical characterisation of modulation doped silicon

Figure 2a shows a schematic of the Si/SiO2:Al MOS capacitor, fabricated via plasma enhanced chemical vapour deposition (PECVD) of SiO2 on n-type Si wafers combined with thermal atomic layer deposition (ALD) of 1 or 2 Al-O MLs. After deposition, the structures were annealed for 30 sec at 1000 °C in Ar ambient; subsequently, thermally evaporated aluminium contacts were structured. As evident from Fig. 1e, successful modulation doping of Si will manifest itself as a negative fixed charge Qfix in the SiO2:Al film. Using 1 MHz high-frequency capacitance-voltage (CV), we measured a change in the fixed charge of ΔQfix = −2.3 × 1012 cm−2 for 1 ALD Al-O ML in SiO2 (Al1) compared with pure SiO2 reference samples (Al0) (Fig. 2b). The flatband voltage VFB was + 0.9 V and + 1.9 V for samples Al1 and Al2, respectively, whereas for the reference sample Al0, the flatband condition was achieved for −0.2 V. By using n-type Si with a donor density of 3.5 × 1015 cm−3 and Al as the gate metal, we keep the work function difference between the gate and n-Si negligible (−0.004 eV). High-frequency CV does not allow for a contribution of Si/SiO2 interface states to the CV signal24, which can therefore be ruled out as having an effect on the CV curve. High energy resolution deep level transient spectroscopy (HERA-DLTS) was carried out to further characterize the Al-induced acceptor state. Figure 2d plots the transient capacitance over the reverse bias VR at 102 K. Clearly, sample Al1 shows a dominant signal at VR = −3.5 V, whereas no transient capacitance is detected for the reference sample. The peak originates from electrons escaping from the Al-state into the Si wafer via direct tunnelling (scattering, trap-assisted tunnelling and hopping are suppressed at low temperatures). The transient time constant TW = 31 ms was kept comparatively short at the cost of signal intensity to further eliminate any trap-assisted tunnelling that is less dynamic. Using the Poisson equation, the energetic position of the Al-acceptor state in the SiO2 bandgap was calculated to be 0.5 eV below the Si valence band edge. This energy value underlines the accuracy of DFT calculations, yielding 0.8 eV. The full with half maximum (FWHM) of VR was ΔVR = 0.66 V, cf. Fig. 2c. The FWHM translates into a vertical-spatial distribution of Al atoms of ca. 0.27 nm, accounting for RTO interface roughness and binding of Al to the RTO, as well as binding to the CVD capping oxide.

Figure 2: Electronic characterization of SiO2 modulation doping with Al acceptors using Al/SiO2/Al-O/SiO2/Si MOS structures.
figure 2

(a) Sample structure showing Al modulation acceptors charged from Si substrate by electron tunnelling. (b) CV curves of reference sample Al0 (no Al-O), Al1 (1 ML Al-O) and Al2 (2 ML Al-O) measured at T = 300 K; ΔVfb and ΔQfix due to negatively ionized Al shown by coloured arrows. (c) Band structure scheme of charge transient measurements for electron release [scheme I] and for electron capture [scheme II] by Al in DLTS. (d) Electron release of Al in SiO2 with pulse time tp = 205 μs, transient time TW = 31 ms and pulse voltage Vp = +0.5 V as function of reference voltage VR measured at T = 102 K to freeze out inelastic scattering and trap-assisted processes for maximum energy resolution by direct electron tunnelling into Si. (e) Electron capture with transient time TW = 3.63 s, pulse voltage Vp = −4 V and reference voltage VR = 0 V as function of pulse time tp measured at T = 502 K to activate all transport paths (hopping, direct and trap assisted tunnelling) for maximum occupation probability of Al in SiO2. Arrows show sub-peaks in accord with Al-O MLs. Capacitance per area scale changes from pF/cm2 in (d) to nF/cm2 in (e).

Full size image

This finding confirms our h-DFT results, where Al in SiO2 provides an acceptor state even for small Si NCs and corroborates that Al undergoes virtually no diffusion during the activation anneal, in accordance with its diminutive diffusion coefficient25. In Fig. 2e, we demonstrate the reverse pulse scheme to sample electron capture into Al acceptors and plot transient capacitance over the pulse time tP at 502 K. The modulation doped sample Al1 showed one broad peak at tP = 2.0 s, with a maximum transient signal of 146 nF/cm2 or 9.11 × 1011 cm−2 transferred charge, accounting for ca. 40% of all negative Al acceptors. No electron capture signal was detected for the reference sample. Time constants TW and tp were chosen to be large compared to low-temperature electron escape in order to account for the maximum density of charged Al acceptors. The substantially higher two-peak signal of sample Al2 shows that we control the density of active modulation acceptors in SiO2, which was confirmed by HF-CV results.

Field effect passivation and improved hole tunnelling by Al-induced acceptor states

First results on carrier lifetime measurements carried out by quasi-steady-state photoconductance (QSSPC)26 are shown in Fig. 3a using mechanical grade double side polished 525 μm thick Cz-Si wafers with an ultrathin rapid-thermal oxide. Samples were not passivated in hydrogen (H2) or forming gas (FG). The spectra show a minority (hole) carrier lifetime τhole = 1 ms at an excess majority carrier density corresponding to 1 sun illumination (Δp = 1015 cm−3), presenting an increase over the reference sample with undoped annealed CVD-SiO2 by a factor of 100. The corresponding emitter saturation current density27 is j0 = 1.05 × 10−13 A/cm2. This low value obtained on preliminary samples shows the potential of SiO2 modulation doping for high efficiency silicon solar cells. With a high negative Qfix we can approximate the surface recombination velocity by28 Seff = dwafer/2τhole, yielding 26 cm/s. In Fig. 3b current-voltage measurements on 3 nm thick RTO/SiO2 stacks on p-type Si wafers are shown. Current densities of ca. 40 mA/cm2 which correspond to maximum power point (MPP) operation of high efficiency solar cells incur a potential drop of ca. 0.23 V for the reference sample while the SiO2 doped with 1 ML Al reduces this potential drop by 74% to 0.06 V. While a high negative Qfix accelerates holes to the tunnelling barrier, the Al-induced acceptor states do not have to be charged for increasing the transition probability through the tunnelling barrier. Such states contribute to improved hole tunnelling already in their neutral form due to diminishing the tunnelling barrier from the valence band offset from Si to SiO2 of ca. 4.5 eV29 to merely 0.5 eV by introducing a lateral defect band. The improved lifetimes and tunnelling currents obtained for preliminary, non-optimized Cz-wafers indicate the promising potential of SiO2:Al for passivating hole selective tunnelling contacts in high efficiency silicon solar cells.

Figure 3: Effect of SiO2:Al modulation doping on effective minority carrier (hole) lifetime and hole tunnelling current density.
figure 3

(a) Double-side polished 1.6 Ωcm phosphorus-doped 525 μm Cz-Si wafers with RTO/Al-O/SiO2 stacks on both sides (sample Al1) show more than 2 orders of magnitude higher lifetimes compared to the RTO/SiO2 reference sample (Al0). The effective minority carrier lifetime of Al1 at an excess minority carrier concentration corresponding to 1 sun illumination (Δp = 1015 cm−3) is τhole = 1 ms. (b) Hole tunnelling current density under accumulation bias on boron-doped Cz-Si wafers with RTO/Al-O/SiO2 stacks (3 nm total thickness). The Al-O monolayer in sample Al1 enables 1 order of magnitude higher hole current densities at small bias as compared to sample Al0 (RTO/SiO2 reference sample).

Full size image

Discussion

Heterostructure modulation doping of Si using intentionally designed impurities in SiO2 presents a true paradigm shift by “outsourcing” dopants to the surrounding matrix. This approach can acceptor-dope Si MOS-FETs by incorporating Al atoms in the insulating SiO2 trench or into the base and coating for fin-FETs, which circumvents all mentioned nanoscale doping problems. As shown exemplarily in Fig. 4a,c, the source and drain areas of an intrinsic Si fin-FET become p-type conductive by incorporating Al-atoms in the buried oxide (BOX). The absence of dopants within the active Si volume eliminates dopant impurity scattering, resulting in lower heat generation, higher carrier mobilities and consequently lower operating voltages. These features are very beneficial for advancing ultra-large scale integration (ULSI) and ultra-low power applications. Moreover, modulation doping avoids the so-called dopant fingerprint of nanoscale MOS-devices, i.e. statistical performance fluctuations due to variations in the exact number and distribution of dopants4,12,14. Coulomb repulsion between the charged Al-induced acceptor states self-regulates the amount of holes generated and prevents that all such states capture an electron. In fact, approximately only one in a hundred acceptor states (1 Al-O monolayer) can be charged and hole-doping is easily controlled via the areal coverage of the Si nanovolume with SiO2:Al.

Figure 4: Application examples for Si modulation doping.
figure 4

(a,c) Undoped Si fin-FETs can be provided with holes as majority carriers from Al acceptors in the buried oxide layer forming the base of the fin (a), eliminating any size limit due to conventional doping. The band diagram of such fin-FET (c) shows its working principle as hole depletion (p self-conducting) transistor. (b,d) HIT Si solar cells can be equipped with massively enhanced hole contacts where SiO2 is the optimum choice in terms of chemical bond interface passivation (b). Moreover, SiO2 provides a much increased minority (electron) barrier while accelerating holes through a low tunnelling barrier thanks to negatively charged acceptors located about 0.5 eV below the valence band of c-Si. The band diagram of such a hole contact shows its working principle to provide much increased conversion efficiencies of HIT solar cells (d).

Full size image

Apart from microelectronics, Si-modulation doping can enhance passivating tunnelling contacts in heterojunction with intrinsic thin layer (HIT) solar cells30, cf. Fig. 4b,d. Very recently, the conventional HIT-cell concept was complemented by the DASH cell concept (dopant-free asymmetric heterocontacts)31. Bullock et al. propose alkali metal fluorides such as LiFx on intrinsic a-Si:H as electron selective heterocontacts31 and MoOx/a-Si:H(i) as hole selective heterocontacts32 and achieve impressive conversion efficiencies. In a different approach, electron-selective contacts were realized by ALD-TiO2 on ultrathin tunnel-SiO233.

We propose that the Si modulation doping approach represents a competitive technology to improve efficiencies even further: Modulation acceptors in ultrathin SiO2 significantly improve hole tunnelling transport via a strongly decreased tunnelling barrier (0.5 eV vs. 4.5 eV in undoped SiO2), cf. Fig. 3b. Additionally, the acceptor states provide a negative drift field for repelling electrons as the minority carrier type, complemented by the unparalleled chemical Si surface passivation quality of SiO2. A massive increase in minority carrier lifetime and ensuing ultra-low emitter saturation current density as compared to unpassivated samples is evident from Fig. 3. Currently used amorphous Si surface layers are far from featuring such properties. Interestingly, a recent theoretical screening study investigated several elements concerning their defect state energies in SiO2 with respect to Si, in order to improve field-effect passivation and transport for majorities34. However, the energetic position derived for Al in that study differs vastly from our theoretical DFT and experimental DLTS values.

Finally, we note that one or two Al-O monolayers in SiO2 do not constitute an Al2O3 phase. Field effect passivation by negative Qfix were also demonstrated with Al2O3 films prepared by pyrolysis35, sputtering36, PECVD37, and ALD38,39,40,41. These stacks often require annealings in the range of 400 to 800 °C to build up the negative Qfix, depending on the deposition technique. Remarkably, it was shown that 1 to 2 nm silicon oxide inevitably grow during deposition and annealing and that the negative charges are majorly located at the SiO2/Al2O3 interface38,39. We hypothesize that Al2O3 undergoes an interface reaction with underlying Si most likely during the anneal to form an ultrathin layer of SiO2, thereby creating O-deficient Al2O3 (Al2Ox, x < 3) which contains O vacancies. This situation is in striking analogy to F-deficient aluminium fluoride (AlFx, x < 3) as mentioned below42,43. In one work39 the origin of the negative Qfix is ascribed to excess oxygen (O) within AlOx (i.e. x > 3). Since O is very reactive and either a radical (monatomic form) or a gas (O2) and the respective samples require an activation anneal for 15 min at 425 °C39, it is inconceivable how excess O could exist in such AlOx layers. As a result, we have reasonable doubt to consider excess O as the origin of the negative Qfix. On the other hand, average values of Qfix = −4 × 1012 cm−2 and midgap Si/SiO2 interface trap densities of 3 × 1010 cm−2 measured by mercury-probe CV (Hg-CV) wafer maps have been routinely obtained for F-deficient AlF3 (AlFx) on Si wafers with ultrathin SiO242. Light soaking increases the average value of this negative Qfix to −4.85 × 1012 cm−2 43. DFT calculations show that F vacancies are acceptor type defects which can localize electrons originating from Si and tunnelling though SiO242,43,44,45, yielding a negative Qfix. We note here that O was not present in AlFx nor did the ultrathin SiO2 layer on the Si wafer contain excess O43. These findings clearly support the picture that O vacancies provide the acceptor type defect states in AlOx manifesting the negative Qfix. In contrast, modulation acceptor states in SiO2:Al owe their existence to one O dangling bond next to a single Al atom which substitutes Si in SiO2, see Fig. 1a–c. Although Al2O3 thin films provide field effect passivation in analogy to Al-doped SiO2 on Si, the insulating nature of Al2O3 in compound with the 2 nm SiO2 layer formed during the anneal prevents improved hole tunnelling as demonstrated here for 1 Al-O monolayer in SiO2, cf. Fig. 3b. Doping SiO2 with Al also eliminates oxidizing interface reactions whereby sample processing becomes more controllable which can be a technological advantage. The performance of Al2O3 was also investigated in the limit of 1 and 4 plasma ALD-cycles and corresponding thicknesses of 1 and 5 Å40,41. The thickness of 1 Å corresponds to one thermal ALD cycle as used by us to dope SiO2 with 1 ML Al-O. It turns out that these ultrathin films deposited directly on Si create significantly less negative Qfix, though a superposition with the high positive fixed charge density in the SiNx capping cannot be ruled out.

In summary, we developed in theory and experiment a heterostructure Si-modulation doping method based on aluminium-induced acceptor states in SiO2. Providing holes as majority charge carriers with such a fundamental principle represents a paradigm shift in silicon science and technology. It allows for a different strategy to define and control majority charge carriers in nanoelectronic devices and allows for passivating hole-selective contacts in high-efficiency solar cells. More generally, our modulation doping concept is transferable to other group-IV semiconductors, such as diamond or germanium, if energetically suitable combinations of impurity elements and dielectric matrices are found.

Methods

Density functional theory simulations and Poisson solver

Approximants were structurally optimized for the maximum integral over all bond energies defining the most stable configuration, using the Hartree-Fock (HF) method with 3-21G molecular orbital basis set (MO-BS)22,46 for structural optimizations and the B3LYP hybrid DF47,48 with 6–31G(d) MO-BS22,49 for electronic structure calculations with the Gaussian09 software suite50. RMS and peak force convergence limits were 8 meV/Å and 12 meV/Å, respectively. Ultrafine integration grids and tight convergence criteria were applied to the self-consistent field routine. During all calculations, no symmetry constraints were applied to the MOs. Further accuracy evaluations can be found elsewhere22,23. A one-dimensional (1-D) Poisson solver for MIS structures was coded in MatLab following Nicollian and Bruce24.

Fabrication and characterization of SiO2:Al samples

PECVD SiO2 tunnel oxides (1.5 nm) were deposited51 onto wet-chemically cleaned 100 mm Czochralski P-doped Si wafers (1.6 Ωcm) in a modified Oxford Instruments PlasmaLab–FlexAL cluster. Subsequently, the wafers were transferred under vacuum into the thermal ALD chamber for deposition of Al-O monolayers via TMA (trimethylaluminium) and H2O at 200 °C (sample Al1: 1 ALD cycle, Al2: 2 ALD cycles, Al0: H2O pulse only). Finally, the capping SiO2 layer (10 nm) was deposited in the PECVD chamber. The lifetime wafers were rapid-thermally oxidized (RTO, 2.5 nm) and then transferred into the ALD-PECVD cluster for Al-O monolayer and capping oxide deposition. All wafers were RTP-annealed (1000 °C, 30 s, Ar atmosphere). Electrical top- and substrate contacts (Al) were thermally evaporated and lithographically structured to fabricate MOS capacitors. The hole tunnelling samples were fabricated similarly on B-doped Cz-Si wafers (2.4 Ωcm) with 1.5 nm RTO and capping oxide to enable direct tunnelling and thermally evaporated Pd contacts. All silicon oxide thicknesses were measured by ellipsometry. DLTS and HF-CV were measured with a PhysTech FT1030 High energy resolution analysis (HERA) setup using a JANIS VPF 800 Cryostat. I-V was measured with an Agilent B1500A. Minority (hole) carrier lifetime was measured as average value over the entire wafer with QSSPC using a lifetime tester from Sinton Consulting.

Additional Information

How to cite this article: König, D. et al. Modulation Doping of Silicon using Aluminium-induced Acceptor States in Silicon Dioxide. Sci. Rep. 7, 46703; doi: 10.1038/srep46703 (2017).

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.