Complex oxide-based ferroelectric tunnel junctions (FTJs) show excellent nonvolatile memory characteristics promising for emerging technology. However, integration of these epitaxially grown FTJs electrically with a silicon substrate remains challenging due to their incompatible lattice structures and poor electronic interfaces resulting from the direct synthesis techniques. Here, we present an epitaxial SrRuO3/PbZr0.2Ti0.8O3/SrRuO3 FTJ integrated electrically with a doped silicon substrate after a layer transfer process. The tunnel currents of the FTJ on silicon show a large tunneling electroresistance (∼1 × 105%) effect, which is explained by a numerical FTJ model incorporating pinned dipoles at the interfaces. This proof of concept of the integration of functional oxide heterostructures with silicon opens a pathway to beyond-CMOS computing devices using unconventional materials.
Few atomic layers of complex oxides when grown epitaxially on a single crystalline substrate can show various functional properties promising for emerging technologies.1–5 For example, they may possess ferroelectricity, ferromagnetism, multiferroic, or high-temperature superconductivity,1–3,5,6 and their electronic properties may vary from metallic-to-semiconducting-to-insulating states.7,8 Integration of this wide spectrum of functional properties can lead to a new generation of computational devices.5
Among various functional oxide-based ferroelectric (FE) and multiferroic devices, the FE tunnel junction (FTJ) had been extensively investigated as it can function as a non-volatile memory with non-destructive readout and ultra-low energy consumption.9–11 A schematic of the SrRuO3(SRO)/PbZr0.2Ti0.8O3(PZT)/SrRuO3(SRO) FTJ on a conducting P++-Si substrate is shown in Figs. 1(a), and the corresponding energy band diagram of the FTJ structure is shown in Fig. 1(b). The modulation of the tunnel barrier height (Φb↓,Φb↑) and effective tunnel barrier width (W↓ + tPZT, W↑ + tPZT) of the FTJ can be achieved by flipping the polarization states inside the FE oxide (PZT) from the down (P↓) to up (P↑) state (or vice versa). Along with this polarization flipping, different charge screening capacities at top and bottom electrodes,12 different effective masses at two different polarization states,13–15 and the presence of pinned dipoles in dead layers near metal-FE oxide interfaces16,17 can result in a large tunneling electroresistance (TER).
Despite significant progress in understanding and performance improvement of the functional oxide-based FTJ,9 the lack of silicon compatibility of complex oxide materials18 impedes the practical integration of the FTJ with current Silicon-Complementary Metal Oxide Semiconductor (Si-CMOS) technology. Although the recent discovery of Si-compatible, binary oxide FE materials such as doped hafnium dioxide (HfO2)19,20 has opened a new route to this end, the performance of the HfO2-based FTJ such as its tunneling electroresistance (TER) is still much lower compared to the giant TER observed in traditional functional oxide-based FTJs.9 Moreover, the large coercive field of HfO2 is a key bottleneck for low power operation. As such, many research groups still focus on synthesizing complex oxide-based FTJs on Si by direct growth techniques such as pulsed laser deposition (PLD). However, there is a serious bottleneck of integration of these FTJs with Si as high temperature growth of a single crystalline complex oxide film using PLD unavoidably results in intermixing of materials and formation of a damaged interface with Si. This acts as a detrimental buffer layer electrically decoupling the virtual Si substrate from the FTJ.5,21–23 Hence, actual electrical contact of the FTJ with the Si substrate could not be achieved before. As the Si-CMOS will remain the major technology in the near future, the integration of a complex oxide FTJ with Si is extremely important for technology leap.
Recently, we have demonstrated a layer transfer technique for integration of single layer epitaxial complex oxides with Si. In this technique, complex oxide films are grown on lattice matched complex oxide substrates and then transferred on to a target substrate.24 In this Letter, we demonstrate that a functional heterostructure (SRO/PZT/SRO) can also be transferred and integrated with a Si substrate. This technique can eliminate the deleterious effects of high temperature PLD growth of heterostructures on Si, prevent formation of any damaged layer at the oxide/Si interface, substantially reduce the process complexity such as complex etching of thin bottom electrode,25 and ensure the direct electrical contact of the bottom electrode SRO of the FTJ with the conducting Si substrate. We demonstrate that such a transferred SRO/PZT/SRO FTJ exhibits a large TER effect controlled by ferroelectric polarization.
At the beginning of the process, a 20 nm La0.7Sr0.3MnO3 (LSMO) sacrificial layer was grown epitaxially on a single crystalline STO(001) substrate using the PLD technique followed by growing the SRO(20 nm)/PZT(6 u.c.∼2.3 nm)/SRO(20 nm) FTJ, which ensures the minimum lattice mismatch and good quality of the epitaxial films.24 Figure 1(c) shows the atomic force microscopy image of a transferred control PZT sample with an ∼2.4 nm thickness, grown following the same deposition parameters as the trilayer. The grown FTJ structure was symmetrical, as both top and bottom SRO have identical surface termination SrO-TiO2 with the ferroelectric oxide PZT film. Such symmetrical electrodes were chosen to eliminate any work function difference induced built-in-electric field inside the FTJ. The in situ growth of both SRO electrodes with PZT is essential to avoid extrinsic deleterious effects at SRO/PZT interfaces, resulting in spurious TER effects.26 In subsequent steps, LSMO was wet etched and the SRO/PZT/SRO FTJ was separated from LSMO and transferred on the doped P++-Si (1020/cm3) using a polymethyl methacrylate (PMMA)-based transfer stamp. Prior to the transfer of the SRO/PZT/SRO FTJ, the P++-Si substrate was annealed in hydrogen (H2) gas for passivation of the Si surface, which ensured low resistance Si/SRO contact formation with minimum interface trap states.
Figure 1(d) shows the switching current response of an FTJ (with top SRO grounded) under time-varying electric fields. A negative pulse was used to set the polarization first. After that, as the voltage is swept up, current through the FE layer increases. It starts decreasing before the supply voltage reaches its peak. This is a typical current response of the FE material originating from the switching of dipole orientation. Similar characteristics are also observed at negative pulses. The ferroelectricity of the transferred SRO/PZT/SRO/Si FTJ was further verified using a technique commonly known as the positive-up-negative-down (PUND) technique.25,27,28 Figure 2(a) shows the measured switchable polarization (Psw) of a transferred 5 μm × 5μm FTJ with respect to the pulse amplitudes (Vpulse). The pulses were applied on the top SRO when the P++-Si substrate was grounded. For the PUND measurement, initially, the FTJ was completely turned into a high (low) resistance state by applying a preset voltage pulse and, subsequently, two positive (negative) voltage pulses of 1 ms were applied. The pulses were applied with varying amplitudes (Vpulse) to switch the FTJ into a low (high) resistance state. The small magnitude of switchable polarization in a low voltage range (−1 to +1 V) in Fig. 2(a) implies that only a small fraction of ferroelectric domain nucleation and domain growth can take place.27 The magnitude starts rising sharply in the region −1 to −3 V, demonstrating an increased number of ferroelectric domains with reversed polarization. However, in the region +1 to +3 V, the increase rate of switchable polarization (Psw) with respect to increasing pulse amplitudes (Vpulse) is substantially lower compared to the region of −1 to −3 V. This implies that a non-switchable pinned dipole pointing to the top SRO may be present at the bottom SRO/PZT interface, inhibiting the ferroelectric domain growth and domain nucleation. The magnitude of the switchable polarization (Psw) saturates at −65 μC/cm2 and +38 μC/cm2 beyond −2.8 V and +3.2 V, respectively, which implies the complete ferroelectric domain switching and polarization reversal.
Although the SRO/PZT/SRO FTJ is symmetrical in structure, the magnitudes of switchable polarizations (Psw) are not equal. The saturation polarization (Ps), which can be approximated as half of the switchable polarizations (Psw/2), is Ps− = −32.5 μC/cm2 at −2.8 V and Ps+ = +19 μC/cm2 at +3.2 V. The smaller magnitude of saturation polarization (Ps+) at +Vpulse implies the presence of a large non-switchable pinned dipole near the bottom SRO/PZT interface pointing to the top SRO, which reduces the actual saturation polarization (Ps+) from +32.5 μC/cm2 to +19 μC/cm2. During the transient switching measurement, the oppositely oriented pinned dipole at the bottom SRO/PZT interface can relax a portion of the switchable polarization inside ferroelectric PZT, which increases the subsequent non-switching transient current component, resulting in a smaller saturation polarization (Ps+ = +19 μC/cm2).28,29 However, at −Vpulse, the direction of the pinned dipole at the bottom SRO/PZT interface is the same, as −32.5 μC/cm2 does not add any relaxed switching current component of PZT during the non-switching current measurement.
Figure 2(b) shows the current–voltage (I–V) characteristics of the FTJ when bias voltages (Vbias) were applied on the conducting P++-Si substrate keeping the top SRO grounded. The tunnel current ratio in the low-to-high resistance state is ∼1 × 103 at +0.5 V. The resistance hysteresis loop with respect to the switching bias voltages (Vbias) was also measured at + 0.5V [Fig. 2(c)]. The FTJ was initially at a low resistance state with Rlow = 106 Ω at Vbias = −5 V. It switched from a low-to-high resistance state with Rhigh = 109 Ω at +4 V and a high-to-low resistance state with Rlow = 106 Ω at −5 V. The larger voltage required to switch the FTJ into the low resistance state is also consistent with having a pinned dipole at the bottom SRO/PZT interface, directed toward the top SRO mentioned in the PUND measurement.28 The magnitude of the saturation polarization in low and high resistance states is Ps+ = +19 μC/cm2 and Ps− = −32.5 μC/cm2, respectively. The FE polarization in PZT verified by the PUND measurement as well as the observed resistance switching hysteresis loop of Fig. 2(c) confirms that the polarization controlled resistive switching is the dominant transport mechanism in the transferred SRO/PZT/SRO FTJ.15 The resistance area product of the transferred FTJ on Si is 0.025 GΩ μm2 and 25 GΩ μm2 in low and high resistance states, respectively, which is also consistent with the previous report on the PZT-based FTJ.15 However, the large TER such as ∼1 × 105% in a symmetrical SRO/PZT/SRO FTJ integrated with a conducting Si substrate is remarkable.
To verify the electronic transport characteristics of the transferred SRO/PZT/SRO FTJ on Si and to evaluate the large TER effect in a symmetrical junction, we performed device simulation using the FTJ model reported earlier.30 Different saturation polarizations in low and high resistance states resulting from pinned dipoles measured in PUND are included in the numerical simulation.16,17 The details of the simulations model with electronic band parameters can be found in the supplementary material. As the FTJ is symmetrical in structure, two pinned dipoles are oppositely oriented to each other inside two ferroelectric dead layers of diL and diR with a thickness of ∼2 Å each near the top and bottom SRO/PZT interfaces, respectively. However, as the magnitude of saturation polarization in the low resistance state is lower (Ps+ = +19 μC/cm2), the magnitude of the pinned dipole near the bottom SRO/PZT interface (PiR) is larger than that of the pinned dipole near the top SRO/PZT interface (PiL). The presence of pinned interfacial dipoles with unequal polarizations directed toward the ultrathin ferroelectric oxide film between two symmetrical SRO has been verified before by the first-principles calculation.31 The oppositely oriented pinned dipoles (PiL, PiR) can produce an interface domain wall having strong bonding with the interface atoms, resulting in a larger PiR with net polarization pointing to the top SRO.31 The switchable polarization of the ferroelectric inside PZT was used the same as saturation polarization Ps+ = +19 μC/cm2 and Ps− = −32.5 μC/cm2 in low and high resistance states, respectively, as observed from the PUND measurement. The energy band of the SRO/PZT/SRO FTJ with pinned dipoles (PiL = +1 μC/cm2, PiR = −13.5 μC/cm2) and dielectric constants (εiL = 9ε0, εPZT = 90ε0, and εiR = 45ε0) is shown in Fig. 3(a) with applied bias Vbias = +0.5V at bottom SRO. The calculated tunnel current in low (ILow) and high (IHigh) resistance states at T = 300 K is shown in Figs. 3(b) and 3(c). The simulated I–V resulting from direct quantum mechanical tunneling through the ferroelectric oxide PZT has a good match with measured tunnel currents in low and high resistance states, verifying the large TER (∼1 × 105%) stemming from the FE tunnel electroresistance effect. The presence of pinned dipoles along with different switchable polarizations modulates the tunnel barrier height of the SRO/PZT/SRO FTJ from the high (ΦSRO = 1.85 eV) to low (ΦSRO = 1.7 eV) resistance state16,17,32 [Fig. 3(a)]. In addition, the polarization dependent complex band structure of the SRO/PZT/SRO FTJ results in two different effective masses of PZT (mPZT = 0.5 m0, 0.9 m0), which also results in different tunneling currents at low and high resistance states.13–15 The combination of the modulated tunnel barrier height of PZT and the effective mass variation in high (low) resistance states results in a large TER of ∼1 × 105% observed in the symmetrical FTJ13–17,32 with identical surface termination (SrO-TiO2) and screening length of SRO (lSRO). The asymmetrical I–V in Figs. 3(b) and 3(c) also supports the pinned dipole present toward the top SRO/PZT interface in low and high resistance states. Theoretical prediction of such pinned dipole assisted large TER had been reported in earlier reports.16,17 Our experimental observation and numerical simulation verify such large TER in a symmetrical FTJ when directly integrated with a conducting Si substrate.
In conclusion, we demonstrate large TER (∼1 × 105%) in the transferred single-crystalline SRO/PZT/SRO symmetrical FTJ when integrated electrically on the conducting Si substrate. Charge transport simulations based on quantum mechanical tunneling and polarization measurements unambiguously indicate that the two distinct resistance states originate from ferroelectric control of the devices. This work opens a possibility of integrating many other single crystal complex oxide heterostructures such as magnetic tunnel junctions and tricolor superlattices with silicon.
See the supplementary material for details about simulation of the FTJ current.
Scanning probe microscopy and electronic transport studies carried out at the Argonne National Laboratory were supported by the U.S. Department of Energy, Office of Basic Energy Sciences, Materials Sciences and Engineering Division. The use of the Center for Nanoscale Materials was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357. Materials growth carried out at the University of California Berkeley was supported by the Office of Naval Research under Contract No. N00014‐14‐1‐0654.