Supporting Information Effect of polarization reversal in ferroelectric TiN/Hf 0.5 Zr 0.5 O 2 /TiN devices on electronic conditions at interfaces studied in operando by hard X-ray photoemission spectroscopy Yury Matveyev a*, Dmitry Nergov a, Anna Chernikova a, Yury Lebedinskii a, Roman Kirtaev a, Sergei Zarubin a, Elena Suvorova a,c, Andrei Gloskovskii b and Andrei Zenkevich a a Moscow Institute of Physics and Technology, 9, Institutskiy lane, Dolgoprudny, Moscow region, 141700, Russia b Deutsches Elektronen-Synchrotron, 85 Notkestraße, Hamburg, D-22607, Germany c A.V. Shubnikov Institute of Crystallography, Leninsky pr. 59, Moscow 119333, Russia * To whom correspondence should be addressed: matveyev.ya@mipt.ru
S1. Sample growth procedure Si (100) with native 2-nm-thick SiO 2 layer was used as a substrate. TiN layer ~20 nm in thickness was grown as a bottom electrode by thermal Atomic Layer Deposition (ALD) process based on the TiCl 4 and NH 3 precursors at T=400 C in Sunale R-100 Picosun OY reactor. Further, the sample was transferred to another ALD reactor (Sunale R-100 Picosun OY) and 10- nm thick Hf 0.5 Zr 0.5 O 2 (HZO) film was deposited at T=240 C using Hf[N(CH 3 )(C 2 H 5 )] 4 (TEMAH), Zr[N(CH 3 )(C 2 H 5 )] 4 (TEMAZ) and H 2 O as precursors and N 2 as carrier and purge gas. The top TiN electrode was deposited at room temperature by magnetron sputtering from TiN target in Ar:N atmosphere (3 sccm N and 2.2 sccm Ar) at p=2 mtorr in Edwards Auto 500. Since the maximal probing depth for HAXPES analysis at the probing x-ray energy E=6 kev is ~ 20 nm, the thickness of the top TiN was chosen d ~ 12 nm to ensure both the continuous and conducting film and sufficiently high yield of photoelectrons from the buried HZO layer. In order to induce the crystallization of HZO layer in non-equilibrium ferroelectric phase, samples were subjected to rapid thermal annealing in Ar at T = 400 o C for t = 1 min. The top electrodes were further patterned by maskless optical lithography (Heidelberg Instruments μpg101) with SF 6 plasma etching (PlasmaLab System 100). The lateral size of the top electrodes was defined 300 μm in diameter in order to fit the X-ray beam size (~100x300 μm 2 ). Since the conductivity of the top ultrathin TiN electrode was rather limited, the contacts were made via d~50 nm thick Al paths and pads, deposited on e-beam evaporated SiO 2 (100 nm) layer, also patterned by maskless optical lithography. The electrical contact to the common continuous bottom TiN electrode was made by the two pads, formed at the sample corners by SF 6 plasma etching of all above layers and covered by Al. S2. Sample electrical characterization The fabricated TiN/Hf 0.5 Zr 0.5 O 2 /TiN capacitor devices were examined prior to HAXPES experiment using Cascade Summit 1100 probe station coupled with Agilent semiconductor device analyzer B1500A containing two B1530A waveform generator/fast measurement units and two source/measurement units connected via a selector. For in situ switching of polarization in HZO and simultaneous measurements of I-V characteristics during HAXPES experiments, the two-channel Agilent B2912A source-meter unit was used. In all electrical measurements, the bottom electrode was grounded while the bias voltage was applied to top electrode. In order to get rid of parasitic capacitance, the polarization current was measured at the bottom electrode. The leakage current through the devices was measured to be I ~ 0.1 pa at U = 0.1 V bias for both polarization directions (see Fig. S1).
Current, ma Current, pa 0.4 0.2 0.0-0.2 Up Down -0.4-0.2-0.15-0.1-0.05 0 0.05 0.1 0.15 0.2 0.25 Bias, V Fig. S1: DC I-V characteristics of the TiN/Hf 0.5 Zr 0.5 O 2 /Si devices measured for both polarization directions The ferroelectric response of the capacitor devices under investigation was studied by the technique called PUND (Positive Up Negative Down). Typical voltage pulse train and current signals are shown in Fig. S2. An input pulse train consists of two pairs of unipolar (positive and negative) trapezoidal pulses. We use the τ = 1.5 ms pulses with τ = 0.5 ms pulse edges. Such long pulses for in operando experiments were selected in order to minimize the distortion of the pulse shape while passing along the electrical line in and outside the UHV chamber. During the first pulse in the pair of identical unipolar pulses the current I f is the sum of the current I s associated with the switching of polarization state plus the leakage and charge injection currents, while the current during the second pulse has non-polarization switching contribution and originates only from the leakage current and the charge injection. In this way, the polarization switching current I p can be found as a difference between the integral current signals generated by the first and second unipolar pulses. Polarization switching current + leakage current Leakage current only 1 6 0 0 Voltage, V -1 First pulse Second pulse Third pulse Forth pulse 0 0.25 0.5 0.75 1 Time, ms Fig. S2: A typical ex situ PUND voltage pulse train used for the pulsed switching testing and associated transient currents measured in the TiN/Hf 0.5 Zr 0.5 O 2 /Si stacks. -6
Intensity, a.u. It is worth to mention that there is the possibility that some of slow discharging traps contribute to the current only during the first voltage pulse, since during the second one they are already saturated. Such charging can be estimated by the difference between the polarization extracted from the current at the top and bottom electrode, and, according to our data, this effect accounts for less than 1% of the polarization current. In situ PUND measurements confirmed that the polarization is maintained in the sample during acquisition of HAXPES spectra. S3. HAXPES spectra analysis In order to reconstruct the potential profile from the obtained data, we utilized the following methodology. Firstly, the line shape of Hf3d 5/2 peak obtained from the nonferroelectric (not annealed) TiN/HZO/TiN sample was determined (see Fig. S3). The peak shape was simulated as a convolution of Gauss and Lorentz lines with the FWHM Guass = 1.4 ev and FWHM Lorentz = 1.4 ev. Gauss = 1.4 ev Lorentz = 0.8 ev 1667 1665 1663 1661 1659 Binding energy, ev Fig. S3: Hf3d 5/2 core level spectra taken from non-ferroelectric TiN/HZO/TiN sample and the extracted line shape parameters. Then, the Hf3d 5/2 spectrum taken from FE-HZO sample, was simulated as a sum of several components, representing virtual sub-layers in HZO, each component having its own binding energy shift proportional to the potential at the particular depth, and the intensity defined according to the Beer Lambert law (see Fig. S4). The obtained spectra were fitted with such simulated line.
Relative intensity, a.u. with potential no potential d hν e - 0 d I I e d = 10 nm λ = 4.6 nm (E K ~4 KeV) Binding energy, ev U Fig. S4: The schematic illustration of the core level spectra broadening arising from the potential distribution in dielectric layer. In order to regularize the fit, we assumed the linear potential drop across FE layer and performed the simultaneous fit of the two spectra taken at the short- and open-circuited conditions (see Fig. S5). In an open-circuit case, the additional drop of the potential corresponding to the shift of Ti2p peak position was added to the potential profile across the HZO layer. Fig. S5: The illustration of the fitting results for the polarization pointing up. The fit error was calculated by addition of the artificial noise to the original spectra and fitting of the distorted spectra. The accuracy was estimated as a 3σ in obtained distribution of potentials values at the top and bottom interfaces after 100 fitting procedures. S4. Laboratory XPS analysis of TiN/Hf 0.5 Zr 0.5 O 2 interfaces The core-level spectra were acquired with ThetaProbe XPS spectrometer (ThermoFisher) with AlKa X-ray source at E =1486.6 ev coupled with ALD reactor. All the obtained spectra
were processed with UNIFIT 2014 software package. During the fitting, we utilized the Shirley background, while the peak shape was taken as the convolution of Gauss and Lorentz lines. In order to perform the most accurate comparison of the TiN chemical state at two interfaces, only the intensity of individual components were allowed to vary, all the rest parameters (background shape, peak shape and position, doublet separation and ration between component) were fixed for all spectra. Since the TiN layers were grown in a separate ALD reactor, the Ti2p core level spectra from the bulk of bottom electrode were taken upon removal of the surface oxidized layer by Ar + ion etching (Fig. S6). The spectra consist of three components the metallic TiN, TiON and TiO 2, indicating the partial oxidation of TiN during ALD growth procedure. Ti2p TiO 2 TiON TiN Hf4f top TiN bulk top TiN interface bottom TiN interface bottom TiN bulk 468 464 460 456 452 Binding energy, ev 22 20 18 16 14 Binding energy, ev Fig. S6: The Ti2p core level spectra taken from the top and bottom electrodes, along with the corresponding Hf4f spectra, of TiN/Hf 0.5 Zr 0.5 O 2 /TiN sample under investigation. Table S1. The relative intensity of the components, corresponding to the TiN, TiON and TiO 2, in Ti2p sore level spectra, taken from top and bottom electrode of TiN/Hf 0.5 Zr 0.5 O 2 /TiN sample. Area of individual components, % Electrode Area TiN TiON TiO 2 top bulk 74 15 11 interface 70 17 13 bottom interface 54 25 21 bulk 84 8 8
The chemical state of the bottom Hf 0.5 Zr 0.5 O 2 /TiN interface was studied upon in situ ALD growth of 2-nm-thick Hf 0.5 Zr 0.5 O 2 on top of TiN (Fig. S6). In case of bulk TiN, Ti2p spectrum consists of three components. The relative intensity of metallic TiN component, which is lower compared to the bulk value, is evident of an additional oxidation of TiN at the interface (see Table S1). We believe that the oxidation of TiN surface occurs during the exposure of TiN to the atmospheric air prior to ALD growth of Hf 0.5 Zr 0.5 O 2 layer on top. In order to quantify the thickness of oxide layer at the interface we assume the layer-by-layer growth model, and effectively subtract the spectra taken from unoxidized TiN from the oxidized ones, thus obtaining the spectrum of the surface TiON/TiO 2 layer. Then, taking the effective attenuation length for E k ~1 KeV in TiO 2, TiON and TiN: λ TiO2 2.1 nm; λ TiON 2.1 nm; λ TiN 1.9 nm [ 1 ], we estimate the thicknesses of TiO 2 + TiON layer d ~0.5 nm. XPS spectra of the top TiN electrode in the TiN/HZO/TiN capacitor under investigation were obtained upon removal of the surface Ti oxide by Ar + ion sputtering. The relative intensity of components in the spectrum indicates that TiN layer grown by magnetron sputtering is more oxidized compared to the ALD-grown TiN layer. The chemical structure of the top TiN/Hf 0.5 Zr 0.5 O 2 interface was studied by sputtering of top TiN layer until Hf4f core level line started to emerge (Fig. S6), which should correspond to the residual thickness of TiN ~4.5 nm ( ~2.5 λ TiN ). Careful analysis of the spectrum shape reveals a slight decrease of the metallic component (see Table S1), which points at the possible oxidation at the interface. Since the oxidized TiON layer lies under the bulk TiN, its thicknesses can be estimated by the following formula: d ox = λ ox ln ( 1 I oxλ m ( 1 e d m λm ) I m λ ox e d m λm where λ ox 2.1 nm, λ m 1.9 nm the effective attenuation length of TiON and TiN, I ox, I m the intensity of TiON and TiN layer, respectively, d m the thicknesses of overlaying TiN layer. Taking the residual thickness of TiN after etching d m ~ 4.5 nm, we obtain d ox ~ 1.5 nm. S5. Transmission electron microscopy and X-ray energy dispersive spectrometry analysis of TiN/Hf 0.5 Zr 0.5 O 2 /TiN structures The high-resolution TEM analysis was performed on the additional sample with twice thicker top TiN electrode (d~20 nm), which has been grown with the same procedure as the one ) 1 Cumpson P.J., Seah M.J. Elastic scattering corrections in AES and XPS. Estimating attenuation lengths and conditions required for their valid use in overlayer/substrate experiments // Surf. Interf. Anal., Vol. 25, No. 6, 1997. pp. 430-446.
subjected to HAXPES measurements. The sample cross-section was prepared with mechanical polishing followed by Ar+ ion milling (with E=5 kev reduced to 0.5 kev for final polishing at 2 to the sample surface) at room temperature. In Fig. S7 the high-resolution cross-sectional image of TiN/Hf 0.5 Zr 0.5 O 2 /TiN/Si sample is presented. Epoxy Top TiN Hf 0.5 Zr 0.5 O 2 Bottom TiN Si Fig. S7: The high-resolution cross-sectional image of TiN/Hf 0.5 Zr 0.5 O 2 /TiN/Si sample. Chemical microanalysis and mapping of samples was performed using an X-ray energy dispersive spectrometry (EDS) system (Esprit/Quantax Bruker) in scanning TEM bright and high angle annular dark field (BF and HAADF STEM) modes in a FEI Tecnai Osiris microscope (200 kv X-FEG field emission gun, X-ray detector (Super-X) with 4 x 30 mm 2 windowless SDD diodes and 0.9sr collection angle at 22 take-off angle). Quantitative analysis of Zr and Hf was done using L (2.042 kev) and M (1.645 kev) lines in order to minimize the overlapping of peaks Hf and Si at around 1.7 kev and Hf and Cu around 8 kev. For quantitative microanalysis, X-ray energy dispersive spectrometry (EDS) in STEM mode was performed on a TiN/HZO/TiN sample. It reveals a higher content of O at the interfaces between HZO and TiN (magnetron sputtered and ALD grown) layers (as well as the oxide at the silicon surface). The HAADF image (Fig. S8a) corresponds to the area taken for elemental mapping. The HZO/TiN interfaces are quite rough within 3-5 nm. The original (net counts or composition) X-ray EDS maps of Ti, Yr, Hf and O would not be conclusive because of the apparent phase intermixing on the interfaces projection on the maps plane. Therefore, to
improve interfacial O visibility, the average bulk O content from the TiN and HZO layers was subtracted from the total O content and the net count map of Fig. S7b with its associated profile in Fig. 5 both show the residual O with a significant O enrichment at the HZO/TiN top interface, and importantly, a stronger one at the HZO/TiN top interface as compared to the bottom one. The outermost surface layer of TiN also exhibits a higher O content that is due to the surface oxidation. Fig. S8: STEM HAADF image (a), and elemental map (b) of Hf, Zr and O of TiN/HZO/TiN sample