RF-Characterization of HZO Thin Film Varactors

Abstract

A microwave characterization at UHF band of a ferroelectric hafnium zirconium oxide metal-ferroelectric-metal (MFM) capacitors for varactor applications has been performed. By using an impedance reflectivity method, a complex dielectric permittivity was obtained at frequencies up to 500 MHz. Ferroelectric Hf_0.5Zr_0.5O₂ of 10 nm thickness has demonstrated a stable permittivity switching in the whole frequency range. A constant increase of the calculated dielectric loss is observed, which is shown to be an effect of electric field distribution on highly resistive titanium nitride (TiN) thin film electrodes. The C-V characteristics of a “butterfly” shape was also extracted, where the varactors exhibited a reduction of capacitance tunability from 18.6% at 10 MHz to 15.4% at 500 MHz.

1. Introduction

CMOS compatible materials with extraordinary electrical properties (e.g., piezoelectric, ferroelectric, or multiferroic) are becoming very attractive in the upcoming technology areas, like neuromorphic computing, 5G, etc. An example of such CMOS compatible materials is ferroelectric hafnium oxide (HfO₂) discovered by Boescke et al. [1]. It is considered now as a very prospective candidate for nonvolatile memory devices [2] and neuromorphic computing [3,4,5]. It can be deposited by using atomic layer deposition (ALD), which makes it feasible to be included into large-scale industrial production. Ferroelectricity in such films occurs in an orthorhombic non-centrosymmetric polymorphic modification of HfO₂ (o-phase), with space group Pca2₁, upon the doping with Si or other elements [6], including Zr [7]. Hf_0.5Zr_0.5O₂ (HZO) shows a low temperature of formation of the o-phase (400°) [8,9], which is a good characteristic for Back-End-of-Line (BEoL) integration [10]. The ferroelectric phase is usually enhanced by capping of the HfO₂ with a titanium nitride (TiN) electrode, which increases the surface energy and stabilizes the orthorhombic phase [11].

Varactors, based on ferroelectric HZO interdigitated capacitors were first proposed by Dragoman group [12,13,14,15,16]. We have also proposed the use of CMOS compatible devices in the BEoL, based on HZO varactors in a parallel-plate MFM configuration [17,18], but due to the lack of experimental data at high frequencies, the dielectric relaxation was not considered. In fact, the results of studies of the high-frequency properties of HfO₂ are controversial. Some authors report almost no deterioration of dielectric properties at frequencies up to 5 GHz [19] or even up to 20 GHz [20], whereas others report the drop of capacitance in a range of 1–10 MHz [21,22,23,24]. In [20], complicated methods were used, which include 2-port measurements with a Vector Network Analyer (VNA) and a subsequent extraction of permittivity through a co-simulation, whereas in [21,22] the permittivity characterization via LCR-meter is performed, which normally has limitations in the MHz-range.

In this work, by using a simple reflectivity VNA technique, we were able to investigate the complex permittivity and the C-V characteristics of HZO in a frequency range of 10 MHz to 500 MHz.

2. Materials and Methods

Tetraethoxysilane (TEOS) isolating layer of $1\mu \mathrm{m}$ thickness was deposited on top of a Si wafer using chemical vapour deposition (CVD) (Figure 1a). Afterwards, a MFM stack was deposited in several steps. A bottom titanium electrode of titanium nitride (TiN) with 20 nm thickness was deposited on top of TEOS using atomic layer deposition (ALD). Hafnium oxide was deposited by ALD with HfCl₄ and ZrCl₄ precursors on top of the bottom TiN layer. The Hf:Zr cycling ratio was set to 1:1. Finally, the HZO layer was capped with a 10 nm TiN electrode, deposited by physical vapor deposition (PVD). The obtained MFM stack was annealed by rapid thermal anneal (RTA) at 400 °C for 60 s. High-resolution grazing incidence X-ray diffraction (GIXRD) (Figure 1b) was used to verify the ferroelectric phase, where it can be seen that the HZO was crystallized completely into the orthorhombic/tetragonal phase.

For the further RF characterization where a triple-contact ground-signal-ground (GSG) probe is used for S-parameter measurement, concentric Au electrodes were deposited by a lift-off lithography (Figure 2a,c). This method for measuring the permittivity of the thin films with a triple contact, also called an annular ring method, is described in [25,26,27,28]. The structure represents a series combination of two capacitors (Figure 2b). Its main advantage is that only one lithography step is needed for the fabrication, yet the main drawback is the unavailability of a proper de-embedding (due to one-port measurement) and therefore a high sensitivity to parasitics [25,28].

3. RF Characterization

High-frequency electric measurements were performed on a Keysight N5247B PNA-X vector network analyzer in the frequency range between 10 MHz and 500 MHz by applying a -15 dBm signal. The 50 µm pitch Ground-Signal-Ground Infinity probes were used to contact the ring structure. An SMU B2902A was used for the DC bias application. The DC bias was applied upon RF signal using the bias tee. Additionally, a co-simulation was performed in Ansys HFSS to analyze the distribution of the electric field between the electrodes.

3.1. Complex Permittivity

The complex permittivity was measured by utilizing a procedure described by [25,27], where two ring structures with different inner dot radii are measured (see Appendix A).

In Figure 3a the real and imaginary permittivity values in frequency domain are plotted. With an increase of bias from V = −3 V towards the coercive voltage V_c = 1 V, both values are changing, indicating a stable varactor tuning. In Figure 3b the loss tangent is plotted with frequency. Sweeping the bias voltage also shifts the loss tangent, so that the maximum value lies at V_c, which can be explained by the increase of intrinsic losses due to the distortion of the crystal lattice, as well as the extrinsic losses due to the motion of the domain walls [28]. As the frequency increases, the loss tangent at −3 V increases by an order of magnitude from 0.02 to 0.2. The data was exported to

Table 1. Comparison of the performance of the state-of-the-art hafnium oxide thin-film capacitors at different frequencies.

Ref.	Material	Tanneal	d (nm)	fmin (MHz)	$\mathsf{ϵ}$’(fmin)	tan$\mathsf{\delta }$ (fmin)	fmax (MHz)	$\mathsf{ϵ}$’(fmax)	tan$\mathsf{\delta }$ (fmax)
[19]	HfO2	500 °C	10	${10}^{-3}$	16	0.013	5000	15	0.013
[20]	HfO2	-	10	90	19	0.01	20,000	21	0.07
[22]	La2Hf2O7	900 °C	18	2 × ${10}^{-5}$	15	0.01	2	14	0.2
[23]	Ce0.63Hf0.37O2	900 °C	56–98	${10}^{-4}$	32	-	1	25	-
[24]	HfO2	400 °C	104	2 × ${10}^{-5}$	18.8	0.01	10	1.88	150
[29]	HZO	500 °C	30	${10}^{-3}$	17.5	-	1	15	-
[15]	HZO	-	6	1000	9.8	0.09	14,000	7	0.02
Our work	HZO	400 °C	10	10	24.6	0.027	500	23.8	0.2

for comparison with other state-of-the-art HfO₂ capacitors, whose characteristics were measured at high frequency.

At first glance it might seem that a dielectric relaxation is taking place. However when implementing an Ansys HFSS simulation with the permittivity set constant over the entire frequency range and loss tangent set to 0 (Simulation 1), it is found that the simulation results fit the experimental results much better (Figure 4a) than when using the data obtained from experiment (Simulation 2). Also, when implementing the permittivity extraction used in Appendix A towards the results of Simulation 1, the same effect of reduction in permittivity is observed Figure 4b, which fits well with the experiment, while Simulation 2 does not. This means that the observed strong dielectric relaxation is an artifact in this frequency range upon application of this method, and an increase in the dielectric loss is due to other parasitic effects. This was also explained in [25], where the increase in the loss tangent was attributed to the contact resistance and to the distribution of electric fields on the lower thin film electrode. Since a highly conductive Pt electrode was used in [25], the effect of field distribution played a minor role, but in our case a moderately metallic 20 nm TiN electrode is used, so an effect of field distribution causes larger distortions in the permittivity values. Indeed, a simulation has demonstrated a high decrease in the homogeneity of field distribution on the bottom electrode Figure 4c,d.

3.2. C-V Characteristics and Tunability

Since the method, as described in Appendix A, involves the measurement of two devices, it introduces a significant variation in the extracted parameters, and thus becomes ineffective for a smooth C-V characteristics. Therefore, to extract the capacitance of a single device, a simpler approach was used (see Appendix B) to extract both capacitance and loss tangent. Since the extracted permittivity obtained by this method is less accurate, as the parasitic sheet resistance is not compensated, it was coined an effective permittivity.

In Figure 5a, a C-V characteristics is plotted at different frequency values. In the plot, C-V curves have a “butterfly” shape, which is a standard indication of a ferroelectric behavior [7,29], thus confirming the measurement authenticity. The peaks are located at coercive voltage V_c = 1 V, which is not changing with frequency. Upon increase in frequency, the loops shift to lower permittivity values, repeating the trend, described in Section 3.1. In [29], a similar drop in the C-V curves with increasing frequency is observed, although it occurs in a much lower frequency range than in our case.

Based on the C-V curves, the corresponding tunability was calculated using the relation:

$$\begin{array}{c}\hfill \tau =\frac{{\epsilon }_{max}^{\prime }\left(V\right)-{\epsilon }_{min}^{\prime }\left(V\right)}{{\epsilon }_{max}^{\prime }\left(V\right)}\end{array}$$

(1)

where ${\epsilon }_{max}^{\prime }$ and ${\epsilon }_{min}^{\prime }$ are maximum and minimum values of the real parts of complex permittivity, respectively.

In Figure 5b the corresponding Q-V characteristics are plotted at different frequency values in logarithmic scale for a better peak visualization. The Q-V minima lie on the coercive bias points, which was explained in Section 3.1 for tan $\delta$. Upon increasing frequency, the Q-factor drops due to an increase in tan $\delta$.

In Figure 5c, tunability and minimum Q-factor are both plotted with respect to the frequency. The tunability experiences a reduction from 18.6% at 10 MHz towards 15.4% at 500 MHz and the minimum Q-factor decreases from 22.7 to 1.45, respectively. This can also be attributed, at least in part, to an effect of the inhomogeneous field distribution close to the TiN electrode.

4. Conclusions

In this work we have investigated the RF properties of 1:1 Hf:Zr doped ferroelectric HfO₂ thin-film varactor, implementing the annular ring reflectivity method. In the frequency range of 10 MHz to 500 MHz, HZO shows a slight decrease in effective permittivity and an increase in effective loss tangent. The co-simulation has demonstrated an inhomogeneous distribution of the electric field on the bottom thin film TiN electrode, which is contributing to the reduction of varactor performance. Upon the bias sweep, varactors exhibit a stable shift of permittivity. For the extracted C-V-characteristics we are observing the typical butterfly shape, indicating a perfect match with low-frequency C-V measurements. The extracted tunability shows a slight decrease, which can at least partially be attributed to the inhomogeneous field distribution on the bottom electrode.