Field cycling behavior and breakdown mechanism of ferroelectric Al_0.78Sc_0.22N films

Artificial intelligence (AI), deep learning, and machine learning have been important topics in many areas. ^1–4) All these technologies need to operate at high speed with low power consumption. Computation in memory (CiM) is now important hardware to make a multiply-accumulation process with very fast speed with reduced power consumption thanks to its massive parallel operations. ^5–10) The usage of nonvolatile memories in CiM is an effective way to reduce power consumption while improving the speed of the operation. Among various nonvolatile memories, including magnetic, resistive, and ferroelectric random-access-memories (RAMs), ferroelectric memory provides energy-efficient edge computing thanks to its electric field operation. ^6–10) A CiM device with ferroelectric field-effective transistors (FETs) was demonstrated in 2018. ⁶⁾

The performance of ferroelectric FETs is dominated by the characteristic of the ferroelectric materials. Ferroelectric HfO₂ gate dielectrics have been used as their ferroelectricity remains even in the thickness range below 10 nm. ¹¹⁾ Moreover, thanks to its compatibility with the conventional CMOS fabrication process, a 3D NAND-type ferroelectric FET array and nonvolatile memory applications have been demonstrated. ^12–14) However, there is still some issue for ferroelectric HfO₂ to be solved.

One of the issues is the switching cycle. A typical endurance cycle of ferroelectric HfO₂ capacitors before dielectric breakdown is around 10⁹ cycles or less. ^15,16) The switching cycles can be improved to some extent by the capping process or other element incorporation, and the high endurance of 10¹¹ cycles has been successfully demonstrated. It is reported that oxygen vacancies are formed during the field cycling, and once the chain of oxygen vacancies occurs, a local current path will be formed between the electrodes. ^17–19) Once the conductance of the current path becomes high, the excess Joule heat at the path will trigger the breakdown of the device.

Recently, a new ferroelectric material AlScN was found in 2019. ²⁰⁾ A large remnant polarization (P_r) of about 100 μC cm⁻² with a box-like profile in the hysteresis curve is attractive for memory devices. The origin of the large polarization comes from the wurtzite structure of the AlScN film. The switching mechanism is explained by the movement of the nitrogen atom, changing the surface between metal and nitrogen faces. The ferroelectric and electrical properties can be adjusted by controlling the content of Sc from 10% to 40%, giving flexibility to memory design. ^21–24) Although thickness scaling showed degradation in the thin region, the ferroelectricity is kept for a thickness less than 10 nm. ^25–27) Furthermore, ferroelectric AlScN films with the room-temperature process are attractive for embedded CiM applications. ²⁸⁾ Although many works are focused on composition or thickness-dependent ferroelectric properties, there are few works on endurance, including field cycling change and dielectric breakdown. ²⁹⁾ In this paper, we measured the field cycling change, especially the change at the near-interface region, of 50 nm thick Al_0.78Sc0_.22N films and proposed a model for the dielectric breakdown.

A 50 nm thick ferroelectric Al_0.78Sc_0.22N film with top and bottom TiN layers were deposited in situ on an n⁺Si substrate in the same chamber by reactive sputtering at a substrate temperature of 400 °C. The top electrode was patterned by wet etching to form electrodes. The detailed process and film characteristics are reported in Refs. 20,25,28. The P_r and the coercive field (E_c) were >100 μC cm⁻² and 5.3 MV cm⁻¹, respectively. The breakdown field was about 8.1 MV cm⁻¹. A columnar structure is confirmed by transmission electron microscope (TEM) and X-ray rocking curve. An electrode size of 56 × 56 μm² was used for measurements.

Sinusoidal waves with a voltage amplitude (V_amp) of 28 V, corresponding to an electric field of 5.6 MV cm⁻¹, at 2 kHz were applied to the top electrode. The voltage application was ended by using a positive voltage to the top electrode, resulting in upward polarization of the devices. Note that the Al_0.78Sc_0.22N layer is already polarized in an upward direction after deposition. The numbers of switching cycles were set to 10²–10⁵ cycles. The field cycling test or the switching endurance of the ferroelectric AlScN capacitors is evaluated by capacitance–voltage (C–V), positive-up negative-down (PUND) measurement with a pulse height (V_pulse) of either 26 or 28 V, with a pulse width of 8 μs. The near-interface property change on the number of switching cycles is monitored by leakage current analyses by applying a positive voltage to the top electrode.

Double sweep C–V curves of the capacitors after different numbers of switching cycles from 0 to −28 V, in the direction to induce the polarization reversal of the AlScN film, are shown in Fig. 1. The devices with fresh and 10² switching cycles showed little difference, showing a peak capacitance at −22 V with a sudden drop at further negative voltage application, suggesting the ferroelectric switching. The crossover capacitance between the forward and backward sweeping was at −9.4 V, indicating the presence of imprint effect in the capacitor. With larger numbers of switching cycles, another capacitance peak start to appear at −16 V with a gradual capacitance decrease for the first one at −22 V. The transition in the voltage at peak capacitance did not complete with a switching cycle up to 10⁵ times, indicating that high voltage switching components remain in the film. The increase in the capacitance of the device with 10⁵ switching cycles at −22 V is due to leakage current as the dissipation factor (tan δ) of the measurements was more than 4. The increase in the leakage current will be explained later. The capacitance crossover shifted to a positive direction from −9.4 V to −3.4 V, but since the high switching voltages remain, the shift is not due to negative charge trapping in the film. Note that the voltage at the peak capacitance in the positive voltage region showed a slight reduction with the number of cycles. This fact also supports the change in the switching voltage in the negative voltage region. It is interesting to note that the large change in the voltage at the peak capacitances is only observed in the negative voltage sweep. Although the physical reason is still unclear, there should be a physical asymmetry in the film.

Zoom In Zoom Out Reset image size

The P_r of the capacitors on the number of switching cycles, measured at different V_pulse is shown in Fig. 2. A slight wake-up effect is observed in the first 10² cycles for V_pulse of 28 V, changing the P_r from the initial 89 to 95 μC cm⁻². Further switching cycles more than 10⁴ times resulted in a slight reduction in P_r, indicating the fatigue effect. For the P_r obtained by V_pulse of 26 V, the initial value was comparable to that of 28 V, thus the applied voltage was sufficiently high for the switching voltage. However, a severe fatigue effect was observed after 10⁴ switching cycles without presenting the wake-up effect. As the difference in P_r between the two voltage pulses represents the presence of components with a switching voltage of higher than 26 V, the increase in the gap infers some portion of the ferroelectric domains, especially with low switching voltage components, are pinned by the switching cycles, presumably due to electron trapping.

Zoom In Zoom Out Reset image size

The near-interface region change upon ferroelectric switching can be deduced from leakage current analysis. It is reported that a high concentration of V_N already exists in the film at an as-deposited state; thus, downward band bending increases the leakage current. It is also reported that the concentration of V_N at the near-interface region increases by the first ferroelectric switching. ³⁰⁾ We adopt this method to capture the physical image of the near-interface region change by field cycling. The Schottky plot of the leakage current (J) of devices with different switching cycles at fresh and after 10³ and 10⁵ switching cycles measured at 40 °C are shown in Fig. 3. All the leakage current followed the Schottky emission model with slightly different parameters. This finding indicates that electrons are not conducting locally but rather uniform from the electrode. The leakage current firstly increases with 10³ cycles and then reduces with 10⁵ switching cycles.

Zoom In Zoom Out Reset image size

The Schottky emission model with a tunneling barrier at the Al_0.78Sc_0.22N/metal interface can be expressed as follows,

$$\begin{eqnarray}&&\mathrm{ln}\left(\displaystyle \frac{J}{{T}^{2}}\right)=\sqrt{\displaystyle \frac{q}{4\pi {\varepsilon }_{0}{\varepsilon }_{{\rm{i}}}}}\displaystyle \frac{q}{kT}\sqrt{E}+\,\mathrm{ln}\left({A}^{* }{T}_{{\rm{t}}}\right)-\displaystyle \frac{q{\phi }_{\mathrm{Bn},\mathrm{app}}}{kT},\end{eqnarray} \tag{ 1 }$$

where E, T, q, k, ε₀, ε_i, A^*, and ϕ_Bn,app are the electric field, absolute temperature, elementary charge, Boltzmann constant, vacuum permittivity, optical dielectric constant, effective Richardson constant, and apparent Schottky barrier height to electron, respectively. By introducing a rectangular-shape tunneling barrier, the tunneling probability (T_t), can be expressed as

$$\begin{eqnarray}&&{\rm{ln}}\left({T}_{{\rm{t}}}\right)=-2d\sqrt{\displaystyle \frac{2{m}^{\ast }q\left({\phi }_{{\rm{Bn}}}-{\phi }_{{\rm{Bn}},{\rm{app}}}\right)}{{\hslash }^{2}}},\end{eqnarray} \tag{ 2 }$$

where m*, ħ, ϕ_Bn, d is the effective electron mass, reduced Planks constant, Schottky barrier height for an electron at Al_0.78Sc_0.22N/metal interface, and tunneling distance, respectively. ³¹⁾ The physical origin of the tunneling barrier is explained by the downward band bending at the interface, presumably due to the presence of charged V_N.

The ε_i values were extracted from the slope of the Schottky plot and are shown in Fig. 4. The initial value of 2.8 for fresh devices showed a gradual increase up to 3.5 after 10³ switching cycles. Note that the obtained ε_i values are the dielectric constant at high frequency, which is the square of the refractive index, as the transition of electrons through the interface is shorter than the dielectric response. Therefore, by the dielectric relaxation of the ionic polarization, the ε_i values are generally 1/4–1/10 of the static dielectric constant. ³²⁾ The increase in ε_i can be understood from the creation of V_N near the interface. On the other hand, further switching cycles results in the reduction in ε_i down to 2.1, less than the value of the fresh device. As the switching is performed at a high electric field of 5.6 MV cm⁻¹, the atomic rearrangement, possibly including the compositional change, at the near-interface of the Al_0.78Sc_0.22N layer can likely happen.

Zoom In Zoom Out Reset image size

By taking the intercept in the Schottky plot measured at a different temperature, as shown in Fig. 5, the ϕ_Bn,app can be obtained. The ϕ_Bn,app was found to decrease continuously with the switching cycle from 0.59 to 0.27 eV after 10⁵ switching cycles.

Zoom In Zoom Out Reset image size

Although A^* of Al_1−x Sc_x N films is not reported, we can assume an A* for Al_0.78Sc_0.22N films ranges from 10 to 50 A cm⁻² K⁻². Therefore, in the calculation, an A* of 50 A cm⁻² K⁻², ^33,34) which corresponds to an m^* of 0.42m₀, where m₀ is the mass of electrons, is used. We also adopted the ϕ_Bn to be 2.3 eV, ²²⁾ which was extracted by X-ray photoelectron spectroscopy. The extracted ϕ_Bn,app, A*T_t, and d are summarized in Fig. 6. The error bars represent the extraction accuracy from the data taken from Fig. 5. The ϕ_Bn,app, and A*T_t continuously decrease with the number of cycles. As the leakage current is the balance between the ϕ_Bn,app, and A*T_t as is expressed in Eq. (1), the increase and decrease behavior observed in Fig. 3 with switching cycles can be understood. The d, on the other hand, continuously increases with the number of switching from 4 to 6 nm, indicating that the downward bending becomes much steeper with switching. This finding indicates that the near-interface as deep as 6 nm can be modified by ferroelectric switching of Al_0.78Sc_0.22N films.

Zoom In Zoom Out Reset image size

Based on the obtained results, the picture of the physical phenomena in the bulk and at the near-interface region induced by the switching cycle test is illustrated in Fig. 7. A large number of V_N already exists in the film at as-deposited conditions. With hundreds of switching cycles, V_N is created both in the bulk and at the near-interface region. The bulk V_N creation contributes to facilitating the displacement of N atoms, creating ferroelectric domains with a low E_c. Those formed in the near-interface region bend down the conduction band to increase the tunneling probability of electrons from the electrode. With further switching cycles, electron trapping might take place in the bulk Al_0.78Sc_0.22N to pin the ferroelectric domains, especially from low E_c components, reducing the P_r. In the near-interface region, continuous creation of V_N results in further reducing the ϕ_Bn,app. The near-interface regions extend as deep as 6 nm with 10⁵ switching cycles. Once the leakage current becomes sufficiently high, the capacitor brake down by excessive Joule heat. We should note that the breakdown mechanism is different from other PZT-based or HfO₂-based ferroelectric materials, where conductive filaments formation to bridge the electrodes trigger the breakdown.

Zoom In Zoom Out Reset image size

The field cycling of ferroelectric Al_0.78Sc_0.22N capacitors revealed the reduction in the switching voltage, possibly due to the formation of VN to facilitate the movement of atoms. Also, fatigue effect was observed after switching for 10³ cycles, presumably due to electron trapping to pin the ferroelectric domains. The apparent Schottky barrier height was continuously reduced with switching by downward bending the energy band to tunnel the electrons. The breakdown of Al_0.78Sc_0.22N films is triggered by the Joule heat due to excessive reduced barrier height. The mechanism is in contrast to conventional ferroelectric materials, where the breakdown is triggered by local conductive filaments.

This work is partially supported by the Murata Science Foundation. This work was financially supported by the "Center for the Semiconductor Technology Research" from The Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education (MOE) in Taiwan. Also supported in part by the Ministry of Science and Technology, Taiwan, under Grant MOST 109-2634-F-009-027.