SUPPLEMENTARY INFORMATION

Ferroelectric-field-effect-enhanced electroresistance in metal/ferroelectric/semiconductor tunnel junctions Zheng Wen, Chen Li, Di Wu*, Aidong Li and Naiben Ming National Laboratory of Solid State Microstructures, Nanjing University, Nanjing 210093, China Department of Materials Science and Engineering, College of Engineering and Applied Sciences, Nanjing University, Nanjing 210093, China National Center of Microstructures and Quantum Manipulation, Nanjing University, Nanjing 210093, China *Corresponding author: diwu@nju.edu.cn Present address: Ningxia Key Laboratory of Photovoltaic Materials, Ningxia University, Yinchuan 750021, China NATURE MATERIALS www.nature.com/naturematerials 1

I. Band diagram and tunneling electroresistance calculations Fig. S1. Schematic representations of charge density profiles and corresponding electric field distributions for the low (a, c) and the high (b, d) resistance states of the metal/ferroelectric/semiconductor FTJ. Fig. 1 is also shown (in light colors) to assist the understanding (top: Fig. 1a and c; bottom: Fig. 1b and d). As shown in Fig. S1 (and also in Fig. 1), a ferroelectric tunnel junction (FTJ) composed of a metal and a heavily and uniformly doped n-type semiconductor separated by an ultrathin ferroelectric barrier in thickness d is considered. A rectangular barrier of height U is assumed when the barrier is not polarized. The ferroelectric polarization P creates bound charges with a surface charge density of ±P at the two surfaces of the barrier. We assume the screening length in the metal side is 2 NATURE MATERIALS www.nature.com/naturematerials

SUPPLEMENTARY INFORMATION zero, namely, the ferroelectric bound charges at the metal/ferroelectric interface can be efficiently screened by charges in the metal side. Therefore, the barrier height at the metal/ferroelectric interface is fixed. We treat the screening in the semiconductor following the way used in MOSFET devices.[1] When the polarization points to the semiconductor, the semiconductor surface is driven into accumulation. It can be regarded as a metal. As in MOSFET devices, the density of accumulated majority carriers peaks at a finite distance, approximately 1 nm, away from the semiconductor surface.[1] We model the accumulated electrons as a charge sheet with zero thickness in a density σ A and at a distance δ away from the semiconductor surface, as shown in Fig. S1a.[1] A depolarization field develops due to the incomplete screening (Fig. S1c). The potential energy drop -qδϕ across the barrier due to the depolarization field can be obtained by solving the Poisson's equation, q( P σ A) x qδϕ( x) =, where -q is the ε ε 0 F electron charge, ε 0 and ε F the dielectric constant of the vacuum and the ferroelectric barrier, respectively. σ A can be obtained from the flat band condition as σ A = dε S dε S +δε F P, where ε S is the dielectric constant of the semiconductor. This is consistent with the results in metal/ferroelectric/metal tunnel junctions.[2] When the polarization in the barrier is switched, pointing toward the metal, the negative bound charges in the ferroelectric/semiconductor interface have to be screened by the ionized donors in the semiconductor. This makes a region, w D in width, in the semiconductor surface depleted of electrons. As pointed out previously, NATURE MATERIALS www.nature.com/naturematerials 3

when discussing tunneling through the space charge region, it is helpful to assume a complete depletion throughout the space charge region.[3] Therefore, as shown in Fig. S1b, the screening charge density per unit area σ D can be simplified as qn D w D, where N D is the doping concentration in the semiconductor. The parabolic potential energy barrier, obtained by solving the Poisson s equation, extends into the semiconductor as qδϕ(x) = q2 N D 2ε 0 ε S (x d w D ) 2. In the ferroelectric barrier, the potential energy increases by qδϕ(x) = q(p σ D )x ε 0 ε F. The screening charge density and the width of 2dε S space charge region can be solved as σ D = P 2 dε + w ε S D F and w D S ε S ε F P = ε d + d 1+ 2. Using ε s =200 for the semiconductor and ε F =100 for ε ε qn ε d F F D S the ultrathin ferroelectric film, in a ferroelectric tunnel junction with barrier thickness d=3 nm and P=10 μc/cm 2, w D can be estimated to be 4.6 nm for a 10 20 cm -3 doping concentration. The potential energy increase at the semiconductor surface (x=d) q 2 N D w 2 D /2ε 0 ε S increases with increasing polarization in the barrier. When the potential increase exceeds the difference between the Fermi level and the conduction band minimum E F -E C, the tunneling electrons have to experience an additional barrier. A decrease of tunneling transmittance is expected. Following the treatment of Zhulavlev et al.,[2] the conductance per area is calculated using 2 2 G 2q d k// = T ( E, k// ) 2 F, where the transmission coefficient T(E F, A h (2π ) k // ) is evaluated by solving the Schrödinger equation numerically. The rectangular 4 4 NATURE MATERIALS www.nature.com/naturematerials

SUPPLEMENTARY INFORMATION potential barrier U is assumed to be 0.5 ev, as used in previous reports.[2] The Fermi level of the degenerate semiconductor is set to be 100 mev above the conduction band minimum, reasonable for heavily Nb doped SrTiO 3.[4] ε S is set as 200, close to the value of SrTiO 3 at room temperature.[5] ε F is set as 100, reasonable for ferroelectric ultrathin films, such as BaTiO 3 and PbTiO 3.[6-10] The screening length in accumulation δ is assumed to be 0.5 nm when N D is 5 10 19 cm -3 and changes reciprocally with the doping concentration N D in the semiconductor. This latter assumption is arbitrary. However, this does not produce a significant difference because the ON/OFF ratio is mainly determined by the resistance of the OFF state where the semiconductor is depleted. The calculated A/G values and ON/OFF ratios are plotted in Fig. S2 as functions of polarization in the ferroelectric barrier. As the polarization increases, the space charge region develops and corresponding band bending occurs in the semiconductor surface. When the polarization exceeds a threshold, the conduction band minimum is bended above the Fermi level and an extra barrier is formed. The resistance of the OFF state increases abruptly due to the reduction of tunneling transmittance. However, the resistance of the ON state only exhibits a slight decrease with increasing polarization due to the decrease of the average barrier height in the ferroelectric. Moreover, since the width of the space charge region is a function of doping concentration, the TER can be effectively modulated by the doping concentration in the semiconductor, as shown in Fig. S2. Larger ON/OFF ratio is observed for lower doping concentration because wider depleted region is required to 5 NATURE MATERIALS www.nature.com/naturematerials 5

screen the ferroelectric bound charges. Fig. S2. a, Resistivity area product of the ON and OFF states as functions of polarization for semiconductors of various doping concentrations, b, corresponding ON/OFF current ratio as functions of polarization. II. Structure and polarization in BTO/Nb:STO heterostructures Fig. S3. a, Reflective high energy electron diffraction intensity oscillations during epitaxial growth of the ultrathin BTO on Nb:STO substrate, b, Reciprocal space mapping around the (013) Bragg reflection of BTO/Nb:STO heterostructures. As shown in Fig. S3a, clear intensity oscillations indicate a layer-by-layer growth of BTO thin film on single-crystalline Nb:STO substrates. The thickness of 6 NATURE MATERIALS www.nature.com/naturematerials

SUPPLEMENTARY INFORMATION BTO layer is 7 u.c. by counting the number of reflective high energy electron diffraction intensity oscillations. The ultrathin BTO film is fully strained on Nb:STO substrates as demonstrated by 2-dimensional X-ray diffraction intensity mapping around the (013) reciprocal spot, shown in Fig. S3b. Fig. S4. PFM phase image of tip-written ferroelectric domains collected after 2 hours. Ferroelectric domains, as shown in Fig. 2c, are written on BTO/Nb:STO. Then PFM phase images are taken repetitively within 2 hours. As shown in Fig. S4, the same domain structure with clear phase contrast can still be observed after 2 hours. III. Device structure of Pt/BTO/Nb:STO FTJs Fig. S5. Schematic device structure of the Pt/BTO/Nb:STO FTJ. NATURE MATERIALS www.nature.com/naturematerials 7

IV. Effect of Nb doping concentration in Nb:STO Fig. S6. Room temperature I-V curves in the ON and the OFF states of Pt/BTO/Nb:STO with various Nb concentrations. a, 1.0wt% Nb, b, 0.7wt% Nb, and c, 0.1wt% Nb. The insets show rescaled OFF state I-V curves for each Nb concentration. Fig. S6 shows I-V curves in the ON and the OFF state of Pt/BTO/Nb:STO tunnel junctions with various Nb doping concentrations. Clear resistance contrast is observed in all samples. As expected from the resistance switching scenario proposed in Fig. 1 and the calculation presented in Fig. S2, the TER ON/OFF ratio increases with decreasing Nb concentration in the substrates (Fig. 3c), owing to the wider space charge region required to screen the ferroelectric bound charges in the depleted state. 8 NATURE MATERIALS www.nature.com/naturematerials

SUPPLEMENTARY INFORMATION V. PFM characterizations through Pt top electrodes PFM characteristics of Pt/BTO/Nb:STO are measured through Pt top electrodes (about 30 nm in thickness) using conductive diamond-coated silicon cantilevers (CDT-NCHR, NanoWorld). Hysteresis loops were collected in the DART mode with a triangle pulse applied on the Pt electrodes while the Nb:STO bottom electrodes were grounded. Phase images were recorded in single-frequency PFM mode over a 300 300 nm 2 area. Fig. S7. PFM phase and amplitude hysteresis loops of the Pt/BTO/Nb:STO FTJ. Fig. S7 shows PFM hysteresis loops of the Pt/BTO/Nb:STO FTJ measured on the top electrode. The coercive voltages of the device are about +1.5 and -2.0 V indicated by the minima of the amplitude loop. PFM loops measured from a top electrode represent mechanical responses from domains polarized by a uniform electric field. The coercive voltages are slightly different from the values determined with loops measured on a bare surface of ferroelectric thin films, which reflect contributions from domains polarized by an inhomogeneous electric field. NATURE MATERIALS www.nature.com/naturematerials 9

Fig. S8. Snapshot PFM phase images a-d with corresponding I-V curves e-h of a Pt/BTO/Nb:STO FTJ recorded after various writing pulses from the ON to the OFF states. The white and purple contrasts represent downward and upward domains, respectively. Snapshot PFM phase images from the ON to the OFF states are shown in Fig. S8 with corresponding I-V curves. Both the PFM data (Fig. S8a-d) and the I-V data (Fig. S8e-h) are collected from a same Pt top electrode. Please note that our Cypher AFM has a 20 na compliance for I-V measurement. After a +4.0 V write pulse, the polarization under the electrode is predominantly switched downward (in bright contrast) and the device is set to the low resistance ON state. The current reaches the 20 na compliance at very small read voltage. It is observed, in phase images collected after a -1.5 V and a -2.8 V write pulse, that the volume of upward domains (in purple contrast) increases with increasing write amplitude. Corresponding I-V curves show the resistance increases accordingly with domain switching under the electrode. After a -4.0 V write pulse, the polarization is switched predominantly upward and the device is set to the OFF state as evidenced by the I-V curve. These experiments on top electrodes provide a convincing correlation among the write amplitude, the domain 10 NATURE MATERIALS www.nature.com/naturematerials

SUPPLEMENTARY INFORMATION structure (and the effective polarization) under the electrode and the resistance of the device, which is in excellent agreement with the resistance switching mechanism we proposed. VI. The importance of the ferroelectric barrier Fig. S9. I-V curves recorded after applying ±4.0 V write pulses for Pt/Nb:STO a and Pt/SrTiO 3 (7 u.c.)/nb:sto junctions b. The insets in a and b show schematically the two junctions, respectively. For comparison, a 7 u.c. thick SrTiO 3 was deposited on 0.7wt% Nb:STO with the same growth parameters of BTO. Pt top electrodes were deposited on the SrTiO 3 /Nb:STO heterostructure and a bare Nb:STO substrate. Fig. S9 shows I-V curves of the Pt/Nb:STO and Pt/SrTiO 3 (7 u.c.)/nb:sto junctions measured after applying ±4.0 V write pulses. There is almost no resistance switching observed. This indicates unambiguously the importance of the ferroelectric barrier in the proposed tunneling resistance switching, i.e. a nonvolatile modulation on the semiconductor surface through ferroelectric field effect,[11,12] which greatly modulates the NATURE MATERIALS www.nature.com/naturematerials 11

tunneling resistance. VII. Existence of polarization modulated space charge region in Nb:STO Fig. S10. a, Capacitance of Pt/BTO/Nb:STO in the ON and OFF states as functions of frequency measured at 150 K. b, Corresponding I-V curves measured at 150 K. Chang et al. proposed a metal/ferroelectric/semiconductor tunneling structure in the 1970s.[13] Assuming there exist high density interfacial states at the ferroelectric/semiconductor interface, the TER comes from a ferroelectric modulation of Schottky barrier height due to charge/discharge of the interfacial states. The mechanism proposed in the present work is significantly different from the previous one. Unlike ferroelectric/semiconductor combinations such as evaporated SbSi on Si and chemically formed KH 2 AsO 4 on GaAs, proposed by Chang and Esaki,[13] the high quality epitaxial deposition of ultrathin ferroelectrics on lattice matched oxide semiconductors greatly suppresses the interfacial states.[11] The TER then originates mainly from the ferroelectric modulation on the space charge region in the semiconductors,[11,12] which the electrons have to tunnel through. The existence of the space charge region on the OFF state can be checked by 12 NATURE MATERIALS www.nature.com/naturematerials

SUPPLEMENTARY INFORMATION capacitance measurements. The measured capacitance C can be represented by 1 C 1 1 = +, where C BTO and C SC are capacitances of the BTO layer and the C BTO C SC space charge region, respectively. The capacitances of Pt/BTO/Nb:STO in the ON and OFF states as functions of frequency from 10 5 to 10 6 Hz are plotted in Fig. S10a. The data were collected at 150 K to minimize the effect of tunneling current on the impedance measurement. As shown, the capacitance of the OFF state is indeed smaller than that of the ON state. At 10 6 Hz, the measured capacitances are 82.7 and 62.4 pf for the ON and OFF states, respectively. The dielectric constant of BTO layer, estimated from the ON state capacitance, is about 40, consistent with values reported for ultrathin compressively-strained BTO at low temperatures.[14,15] The dielectric constant of Nb:STO depends strongly on the temperature, the doping concentration and the bias.[16-18] It increases with decreasing temperature, but decreases dramatically with increasing doping concentration and bias. The dielectric constant at 150 K of pure STO at zero bias is about 800.[14] Since the space charge region is strongly biased, it is reasonable to assume an average dielectric constant of about 200~300 for the space charge region in this heavily doped Nb:STO.[17] The width of the space charge region can then be estimated to be 5~8 nm. This indicates strong modulation of the overall barrier width by the polarization switching associated with the ferroelectric field effect. Therefore, the current can be effectively shut off in the OFF state. Fig. S10b shows the I-V curves of Pt/BTO/Nb:STO FTJ measured at 150 K. The ON/OFF ratio is about 20,000, which is comparable to the value at room temperature. NATURE MATERIALS www.nature.com/naturematerials 13

VIII. Resistance switching of Pt/BTO/Nb:STO up to 10 4 cycles Fig. S11. a Bipolar resistance switching by cycling with ±2.2 V pulses. b Resistance switching between 9000 and 9100 cycles for clarity. If ±2.2 V pulses are applied, an ON/OFF ratio about 40 can be maintained to more than 10,000 cycles, as shown in Fig. S11. Reference 1. Sze, S. M. & Ng, K. K., Physics of Semiconductor Devices, 3 rd Edition, John Wiley & Sons, (2007). 2. Zhuravlev, M. Y., Sabirianov, R. F., Jaswal, S. S. & Tsymbal, E. Y. Giant electroresistance in ferroelectric tunnel junctions. Phys. Rev. Lett. 94, 246802 (2005). 3. Conley, J. W., Duke, C. B., Mahan, G. D. & Tiemann, J. J. Electron tunneling in metal-semiconductor barriers. Phys. Rev. 150, 466 (1966). 4. Takizawa, M., Maekawa, K., Wadati, H., Yoshida, T., & Fujimori, A. Angle-resolved photoemission study of Nb-doped SrTiO 3. Phys. Rev. B 79, 14 NATURE MATERIALS www.nature.com/naturematerials

SUPPLEMENTARY INFORMATION 113103 (2009). 5. Van der Berg, R. A., Blom, P. W. M., Cillessen, J. F. M. & Wolf, R. M. Field dependent permittivity in metal-semiconducting SrTiO 3 Schottky diodes. Appl. Phys. Lett. 66, 697 (1995). 6. Lee, S. J., Kang, K. Y. & Han, S. K. Low-frequency dielectric relaxation of BaTiO 3 thin-film capacitors. Appl. Phys. Lett. 75, 1784 (1999). 7. Kim, Y. S. et al. Ferroelectric properties of SrRuO 3 /BaTiO 3 /SrRuO 3 ultrathin film capacitors free from passive layers. Appl. Phys. Lett. 88, 072909 (2006). 8. Li, Z., Grimsditch, M., Xu, X. & Chan, S. K. The elastic, piezoelectric and dielectric constants of tetragonal PbTiO 3 single crystals. Ferroelectrics 141, 313 (1993). 9. Tabata, H., Murata, O., Kawai, T., Kawai, S. & Okuyama, M. c-axis preferred orientation of laser ablated epitaxial PbTiO 3 films and their electrical properties. Appl. Phys. Lett. 64, 428 (1994). 10. Fu, D., Ogawa, T., Suzuki, H. & Ishikawa, K. Thickness dependence of stress in lead titanate thin films deposited on Pt-coated Si. Appl. Phys. Lett. 77, 1532 (2000). 11. Mathews, S., Ramesh, R., Venkatesan, T. & Benedetto, J. Ferroelectric field effect transistor based on epitaxial perovskite heterostructures. Science 276, 238 (1997). 12. Miller, S. L. & McWhorter, P. J. Physics of the ferroelectric nonvolatile memory field effect transistor. J. Appl. Phys. 72, 5999 (1992). 13. Chang, L. L. & Esaki, L. Nonvolatile Schottky diode with barrier height NATURE MATERIALS www.nature.com/naturematerials 15

controlled by ferroelectric polarization. IBM Tech. Discl. Bull. 14, 1250 (1971). 14. Yano, Y. et al. Epitaxial growth and dielectric properties of BaTiO 3 films on Pt electrodes by reactive evaporation. J. Appl. Phys. 76, 7833 (1994). 15. Trithaveesak, O., Schubert, J. & Buchal, Ch. Ferroelectric properties of epitaxial BaTiO 3 thin films and heterostructures on different substrates. J. Appl. Phys. 98, 114101 (2005). 16. Saifi, M. A. & Cross, L. E. Dielectric properties of strontium titanate at low temperature. Phys. Rev. B 2, 677 (1970). 17. Suzuki, S. et al. Fabrication and characterization of Ba 1-x K x BiO 3 /Nb-doped SrTiO 3 all-oxide-type Schottky junctions. J. Appl. Phys. 81, 6830 (1997). 18. Li, J., Ohashi, N., Okushi, H., & Haneda, H. Temperature dependence of carrier transport and resistance switching in Pt/SrTi 1 x Nb x O 3 Schottky junctions. Phys. Rev B 83, 125317 (2011). 16 NATURE MATERIALS www.nature.com/naturematerials