Kirill
Isakov
a,
Olli
Sorsa
b,
Taina
Rauhala
b,
Santeri
Saxelin
b,
Tanja
Kallio
b,
Harri
Lipsanen
a and
Christoffer
Kauppinen
*a
aDepartment of Electronics and Nanoengineering, Micronova, Aalto University, P.O. Box 13500, FI-00076, AALTO, Finland. E-mail: christoffer.kauppinen@gmail.com
bDepartment of Chemistry and Materials Science, Aalto University, P.O. Box 16100, FI-00076, AALTO, Finland
First published on 8th November 2022
With the development of microscale and standalone electronic devices the demand for microscale energy storage is increasing. Micro-supercapacitors are emerging as the candidate for microscale energy storage, especially when combined with energy harvesters. In this work, we enhance the capacitance of porous silicon (PS) supercapacitor electrodes up to 4× by adding a new high-surface-area nanoelectrode on the existing topography, thus forming a hierarchical 3D supercapacitor electrode that can be used in micro-supercapacitor applications. The nanoelectrode is based on grass-like alumina (GLA) – a recently reported conformal nanoporous coating – with two surrounding TiN films, all materials deposited by atomic layer deposition. The GLA nanoelectrode can be deposited conformally on complex topographies like here on PS, as the total thickness of the electrode structure does not exceed 200 nm allowing it to be fitted in existing electrodes. The GLA nanoelectrode increased the capacitance of the PS supercapacitors alone by up to fourfold or 4× and reduced the self-discharge to a mere 25% loss after 20 h compared to a TiN coated PS reference and state-of-the-art PS both with significantly higher losses. The GLA nanoelectrode showed remarkable stability for 10000 galvanostatic cycles with a decrease in the capacitance only by 5% and no structural changes were identified from SEM images. Microelectronics compatible processing, the conformal deposition process and the nanoscale thickness of the GLA nanoelectrode allow it to enhance 3D electrodes commonly used in micro-supercapacitors.
Currently, self-powered microsystems primarily rely on thin film batteries as their energy storage.1 However, batteries are limited in their lifetime and power output, and microbatteries are not an exception. Supercapacitors can be quickly charged or discharged, which means fast energy transport,4–6 and are able to sustain a large number of cycles, often more than a million cycles, before degradation,1 making them a promising candidate as the required alternative for batteries.
In conventional supercapacitors, the energy is stored as a thin electrochemical double layer at the electrode–electrolyte interface (capacitive process). Alternatively, energy can be stored through reduction/oxidation of the electrode surface (faradaic process). With the faradaic processes, a much higher energy density can be reached but it occurs much more slowly and drags down the efficiency of the device which is why they cannot be utilized when considering devices with an extremely high power. Currently conventional supercapacitors are too large to be integrated with microscale electronic devices, therefore micro-supercapacitors compatible with microelectronics fabrication must be used.
These micro-supercapacitors have gained attention in recent years, and some approaches utilize various 3D architectures to allow an increased energy density1 such as interdigital deep reactive ion etched silicon beams,7 carbon microrods,8–10 interdigital micromolded activated carbon,11 porous silicon,12 porous carbon nanotubes and/or graphene.13,14 When using porous silicon (PS) as the active electrode material for the micro-supercapacitor, a thin TiN coating (5–30 nm) has been shown to improve the stability, conductivity and capacitance of the PS structure.15,16 Additionally, TiN coated PS has been shown to exhibit excellent wettability with aqueous and organic electrolytes and superior electrolyte wettability compared to just PS both with aqueous and organic electrolytes.17
Atomic layer deposition (ALD) is known for the ability to conformally coat complex 3D structures.17–22 ALD has also been widely utilized in supercapacitors, mainly for the deposition of pseudocapacitive active materials and to create a protective layer on the support material and/or to enhance the wettability.17,23,24 TiN has also been used as a pseudocapacitive material25 but we did not notice any pseudocapacity in aqueous Na2SO4. Recently, grass-like alumina (GLA) a nanostructured conformal thin film based on the ALD of Al2O3 has been reported26 and used for various applications including broad band and omnidirectional antireflection coatings for glass26 and packaged black silicon photodiodes,27 superhydrophobic antireflection coatings,28 and for enhanced IR absorbers.29
In this work we demonstrate hierarchical porous silicon supercapacitors with improved characteristics such as energy density, capacitance and self-discharge by conformally coating the porous silicon substrate with a GLA nanoelectrode. The GLA nanoelectrode is made by sandwiching the GLA nanostructure between two ALD TiN films. Microstructured PS is conformally coated with this GLA nanoelectrode to achieve a hierarchical structure with higher energy density/charge storage in a coin cell supercapacitor and is suitable for silicon-based micro-supercapacitors with various geometries, including interdigitated configuration. The GLA nanoelectrode is continuous and regular even on 40:1 aspect ratio PS. Finally, we investigate the electrochemical behavior of the assembled supercapacitor cells and establish the capacitance enhancement by adding the GLA nanoelectrode, as well as investigate the cell behavior during 10000 charge–discharge cycles.
All PS chips were dipped into HF for 1 min to strip off the native SiO2 (Fig. 1b and c). Immediately after the oxide strip, the chips entered the ALD reactor for TiN deposition via thermal ALD (Fig. 1d). The deposition proceeded for 750 cycles at 500 °C with TiCl4 and NH3 as precursors, and N2 as a carrier gas targeting at 15 nm thickness. TiCl4 was pulsed for 100 ms with 60 sccm carrier gas flow, and NH3 was pulsed for 100 ms with 80 sccm carrier gas flow. The ALD TiN layer thicknesses were calculated from a known growth-per-cycle (GPC) GPC = 0.2 Å for this specific process, at this temperature for this tool. After this, half of the chips were used as supercapacitor plates as such (3 chips with 50 μm PS and 3 chips with 200 μm PS) as shown in Fig. 1d (referred to in the text as PS cells), and the other half was further processed with a GLA nanoelectrode (same number of chips). The next step for the ones with the GLA nanoelectrode is the deposition of Al2O3 by ALD (Fig. 1e). Thermal ALD deposition was executed for 130 cycles at 120 °C with 200 ms trimethylaluminum and 200 ms H2O pulses, N2 gas carried both precursors to the reaction chamber with a flow of 150 sccm, resulting in 11 nm (GPC = 0.85 Å) of alumina which was verified by He–Ne ellipsometry from monitor samples of flat Si. The deposited Al2O3 was transformed into GLA by immersion into de-ionized water (DIW) at 90 °C for 30 min (Fig. 1f); this process is described elsewhere in detail.26 Whole chips were immersed and then dried with a N2 gun. The final step in the GLA nanoelectrode process is the deposition of a top TiN layer (Fig. 1g). This layer was produced with the same process as the bottom TiN layer, but the target thickness was 10 nm, which corresponds to 500 ALD cycles. Al2O3 deposition was carried out in a BENEQ TFS-500 ALD system, and TiN deposition was carried out in a Picosun SUNALE R-200 Advanced ALD system. Both materials were deposited via thermal ALD processes.
Scanning electron microscope (SEM) images of the supercapacitor chips with a GLA electrode were taken at a cross-section before and after galvanostatic cycling using Zeiss Supra 40 field emission SEM. Cross-sections were obtained by cleaving the supercapacitor chips along Si crystal planes.
Fig. 4 shows the two fundamental electrochemical characterization methods for the 200 μm PS + GLA cell: cyclic voltammetry and galvanostatic cycling. In the CV curves, a rectangular shape is obtained and in the galvanostatic cycling, a triangular shape. These are the ideal shapes of an electrochemical double layer supercapacitor with the corresponding electrochemical methods which means that the cell operates with capacitive processes. With the slowest charge/discharge rates, the shapes show some non-ideal behavior due to the faradaic reaction caused by the cell components (see Fig. S1, ESI†). This is seen mainly at high cell voltages as increased faradaic current in the CV curves and as a decreasing slope in the galvanostatic charge/discharge curves. At higher rates, the faradaic reactions do not have enough time to occur to a significant degree, resulting in a mostly capacitive response of the system. Moreover, at higher rates, the self-discharge (discussed further in the end of this section) can be taken as negligible, which also affects the shape of the curves. Overall, the system works rather ideally when the rate is high enough (≥5 mV s−1 or ≥0.01 mA cm−2). Fig. S5–S20 in the ESI† contain the same data as in Fig. 4 for all the samples that were fabricated in this work.
The capacitance of each cell in the galvanostatic cycling is shown in Fig. 5. The capacitance has been quantified with the same charge/discharge currents as shown in Fig. 4 and plotted before and after 10000 galvanostatic cycles. The capacitance is normalized with the area of the electrode (1 cm2) in Fig. 5a and with the total volume of the active layers in the cell in Fig. 5b. To determine the volume, the height of the PS pillars is considered as the depth of the sample. For example, the total active volume of the 200 μm PS cell is 2 × 10 mm × 10 mm × 0.2 mm = 40 mm3. The volume normalized capacitances of the 50 μm PS and 200 μm PS are almost identical. Similarly, the values for 50 μm PS + GLA and 200 μm PS + GLA are close to each other. This shows that the GLA has been conformally synthesized inside the pores between the PS pillars and it increases the area of the electrochemical double layer. The capacitance increase of the GLA nanoelectrode is 2.5–4.0-fold which is impressive. The best performing cell, 200 μm PS + GLA shows capacitance values of 5–11 mF cm−2 which is the largest areal capacitance ever reported for silicon based aqueous microsupercapacitors corresponding to 120–270 mF cm−3, depending on the applied current density.
The increase of the PS cell capacitance with the addition of the GLA nanoelectrode can solely be attributed to the increase in surface area. This can be seen from the capacitance formula where for electrochemical double-layer capacitors εr is the relative permittivity of the electrolyte, ε0 is the vacuum permittivity, d is the effective thickness of the double-layer and A is the area of the capacitor. As the PS cell has 15 nm of TiN facing the electrolyte and the PS + GLA cell has 10 nm of TiN facing the electrolyte, the surface chemistry of these situations is the same. Therefore from the capacitance formula it can be concluded that the increase in capacitance is solely due to the increased surface area.
All the cells show a decrease of capacitance with increasing current density. With higher current, the electrochemical double layer is forced to form faster which in some cases results in a lower capacitance. In our case, the main cause of capacitance decrease is the limited diffusion rate of the electrolyte in and out of the PS pores. The capacitance decrease is higher for the samples with the GLA which is explained by the smaller volume for the electrolyte to diffuse between the pillar structure and the higher capacity which requires more electrolyte to be diffused. The 10000 galvanostatic cycles do not cause a substantial decrease in the capacitances which means that the structure of the material is stable in these conditions. The volumetric capacity of the PS cells in these measurements was ∼11 μA h cm−3 and for PS + GLA it was ∼32 μA h cm−3 at 0.1 mA cm−2. The coulombic efficiency, which is determined as the ratio of discharge and charge capacity, was >97% for all the cells when the current density was at least 0.01 mA cm−2. The volumetric energy of the PS cells in these measurements was ∼4 μW h cm−3 and for PS + GLA it was ∼12 μW h cm−3 at 0.1 mA cm−2. The energy efficiency of the cells was mainly 75–85%.
The stability of the materials was studied with galvanostatic cycling. Because the cells had major differences in capacity, the cells were cycled with a current equivalent to 10 s per cycle. This corresponds to 28 h of cycling and the current density varied in the range of 0.05–1 mA cm−2. The voltage window of the cycling was 0.0–0.8 V. The capacity decrease during galvanostatic cycling is presented in Fig. 5c. The system suffers from only minor capacity reduction as all the cells have roughly 95% or more of the initial capacity left after 10000 cycles. Generally, an increased surface area leads to a higher amount of side reactions with the electrolyte, leading to a lowered cycle life. In contrast, in the case of the GLA nanoelectrode, the cycle life is not lowered by the additional surface area provided by the GLA layer, demonstrating the excellent performance of the GLA nanoelectrode in improving the capacitance of the supercapacitor. The roughly 95% of the initial capacity left after 10000 cycles shows the viability of the GLA nanoelectrode as an approach to increase the performance of real-world supercapacitors.
One of the drawbacks of supercapacitors is their relatively high self-discharge rate.30 This means that supercapacitors cannot be used for long term energy storage because the energy efficiency decreases as the device experiences self-discharge. We tested the self-discharge rate of the studied cells by charging the device up to 80% of the maximum voltage (0.64 V) and applying that voltage for 0.5 h in order to stabilize the system and to eliminate the effects of overpotential arising during the galvanostatic charging. Then the circuit was opened, and the voltage was monitored for 20 h (Fig. 5d). The cell voltage of the best performing material, 200 μm PS + GLA, drops from 0.64 V to 0.48 V in 20 h corresponding to a 25% loss. In comparison, in the state-of-the-art devices by Grigoras et al.17 the voltage dropped from 1.2 V to 0.4 V in 20 h, a 67% loss for their device. The low self-discharge achieved by our GLA nanoelectrode devices is thus clearly favorable compared to state-of-the-art microsupercapacitors. The voltage drop (in this work) is clearly dependent on the total capacity of the cell since the cells with the highest capacity (200 μm PS) experience less voltage discharge than the 50 μm PS cells. Most remarkably, the GLA further decreases the rate of self-discharge. In terms of the self-discharge, it would be beneficial to use longer pillars, so that the charge could not escape that fast.
The electrochemical impedance spectra of the cells are presented in Fig. 6. The ideal spectrum of a double layer supercapacitor is a straight vertical line at Z′ equaling the ohmic resistance of the cell. All the spectra show a small semicircular feature at high frequencies and then rise up in an almost right angle relative to the Z′-axis. The ohmic resistance of the cell can be considered as the intercept with the Z′-axis and the area-specific ohmic resistance is similar for both PS cells (∼5 Ω cm2) and for both PS + GLA cells (∼3 Ω cm2). This means that the GLA nanoelectrode coating increases the conductivity of the material. This is probably due to the thicker TiN (roughly 15 nm below and 10 nm above the GLA) which improves the in-plane conductivity. The PS electrode has only 15 nm TiN on the side facing the electrolyte. PS pores coated with conductive layers have previously resulted in higher pore surface conductivity which results in a higher utilizable capacitance.31 For the spacer cell (see the discussion in the ESI†) the area-specific ohmic resistance in 0.5 M Na2SO4 was 1.2 Ω cm2 (Fig. S3, ESI†) which means that the silicon substrate causes a significant resistance in the system. However, for a more optimized system, the thickness of the substrate can be lower which causes less resistance. In addition, wafers with much lower resistivities for example 0.005 Ω cm are commercially available compared to the values of 30–50 Ω cm used in this work. These optimization steps would even in a conservative estimate bring down the silicon resistance by a factor of a thousand. Moreover, if the materials are used in a chip supercapacitor, the system can be designed so that there is no need for the current to travel through the whole thickness of the silicon chip.
Fig. 6 Electrochemical impedance spectra of the supercapacitor cells before (left) and after the galvanostatic cycling (right) at zero charge open circuit voltage. |
The width of the high frequency feature is clearly higher for the 200 μm PS than for the 50 μm one. This suggests that the feature is caused by diffusion as for the longer pillars the electrolyte has to diffuse a longer distance. The effective area-specific resistance increase of the diffusion is ∼0.2 Ω cm2 for the 50 μm and ∼0.6 Ω cm2 for the 200 μm one. Based on this, the total performance of the cell is not that highly dependent on the diffusion and even longer PS pillars could be used with this geometry. The 10000 galvanostatic cycles do not have an effect on the EIS either, demonstrating the stability of the material in the studied conditions and indicating real-world robustness of the GLA nanoelectrode.
TiN-GLA-TiN likely contains transitionary film at the interface of different materials, such as AlOxNy and TiOxNy. This is not particularly problematic, as the TiN and Al2O3 interface is utilized in some MOSFET structures,32 where TiN plays the role of the gate material, and Al2O3 is the gate oxide. The diffusion of oxygen from alumina into TiN is reported to be about 1 nm33 or even indetectable.34 The top TiN layer in the electrode structure presented in this work has a thickness of about 10 nm, which ensures that the electrolyte-facing surface of the electrode comprises only TiN, and the oxygen diffusion phenomena inside the electrode do not alter the system from the perspective of its application in supercapacitors. Another important characteristic for the electrode is its conductivity; it was demonstrated that the addition of GLA and TiN layers on top of TiN coated porous silicon actually improves the conductivity in the supercapacitors (see the impendence spectra presented in Fig. 6).
Table 1 shows a list of silicon based aqueous microsupercapacitor electrodes/devices with their relevant properties. The capacitance of the 200 μm PS cell (∼2 mF cm−2) is already comparable to other PS electrodes reported in the literature. With different silicon structure geometries, capacitances of 1–5 mF cm−2 have been reported.15–17,35,36,38 However, the GLA nanoelectrode boosts the capacitance of the PS electrodes to the next level (5–11 mF cm−2 or 2.5–4× increase), and no aqueous silicon microsupercapacitor reported as high a capacitance. Higher capacitance values (10–300 mF cm−2) can be achieved with porous carbon structures, such as carbon nanotubes, graphene and etched activated carbon and even higher ones (100–2000 mF cm−2) with pseudocapacitive materials utilizing faradaic processes such as oxides (e.g. RuO2, MnO2) or conductive polymers (e.g. polyaniline, polypyrrole).39 However, the pseudocapacitive materials lack in durability and rate capability resulting in lower maximum power and energy efficiency,40,41 and porous carbon electrodes often require additives such as binders or conductive additives which decrease the energy density.30 Unlike carbon-based supercapacitors, their silicon-based counterparts can be scaled down to accommodate microscale devices. Using a silicon substrate also allows production of supercapacitors on the same chips in parallel with other electronics.
Electrode material | Electrode capacitance (mF cm−2) | Electrode thickness (μm) | Electrolyte | Voltage window (V) | Device energy (μW h cm−2) | Device power (mW cm−2) | Ref. |
---|---|---|---|---|---|---|---|
PS + GLA | 5–11 | 200 | 0.5 M Na2SO4 | 0.8 | 0.48 | 0.34 at 1 mA cm−2 | This work |
PS | 1.7–3 | 200 | 0.5 M Na2SO4 | 0.8 | 0.16 | 0.34 at 1 mA cm−2 | This work |
TiN in a PS matrix | 3.1 | 6 | 1 M NaCl | 1.2 | 0.6 | 48 | 17 |
SiC coated Si nanowires | 1.7 | 32 | 1 M KCl | 1.0 | 0.24 | — | 35 |
In2O3 coated PS | 1.36 | 0.8 | 1 M Na2SO4 | 1.2 | 0.21 | 0.12 | 36 |
Graphitized C on poly-SiC | 0.743 | 0.5 | 1 M H2SO4 | 0.7 | 0.026 | 1.50 | 37 |
The PS electrodes are often very stable when coated with TiN and capacitance deterioration of 5–20% has been reported over thousands of cycles.15,17 Therefore, with the 5% capacitance decrease over 10000 cycles reported here, the stability of 200 μm PS + GLA is at least at the expected level. Since the energy of a supercapacitor device is more dependent on the voltage window than on the capacitance of the electrode materials, the aqueous system used in this work with the 0.8 V upper limit does not yield a high energy density (0.48 μW h cm−2, 12 μW h cm−3 for the 200 μm PS + GLA). Although, in comparison to the highest energy density aqueous systems on silicon these are reasonable results as a specific energy of up to 0.6 μW h cm−2 has been reported17,35–37 (see Table 1). The power densities obtained in both 200 μm cells, PS and PS + GLA, are equal to 0.34 mW cm−2 at a current density of 1 mA cm−2. This value is notably lower than for some other reported aqueous systems, even though the corresponding capacitances of those system were lower (see Table 1).
We chose aqueous 0.5 M Na2SO4 as the electrolyte because it is common in fundamental supercapacitor experiments. We have also tested organic electrolyte (1 M TEABF4 in propylene carbonate), but the cell did not provide the expected double layer capacitor-like behavior (Fig. S4, ESI†). This experiment was performed without the GLA which means that the issue relates to either the microstructure of the PS or the TiN layer, not the addition of GLA. It should be noted that with organic electrolyte the system is expected to have a lower capacitance and rate capability because of the larger ion size and slower diffusion. Due to the operation mode of ALD, the synthesis can be performed on a variety of different substrate geometries. The geometry was not optimized in this study. Therefore, a different geometry could enable the use of organic electrolyte and optimization of the substrate could result in even higher capacitance. Furthermore, other conductive materials instead of the TiN could be used, e.g. metals deposited by ALD. Nevertheless, the experiments performed in the aqueous electrolyte already show that the addition of GLA increases the capacitance of the porous silicon by a remarkable 2.5–4.0 times and is therefore highly beneficial.
PS | Porous silicon |
ALD | Atomic layer deposition |
GLA | Grass-like alumina |
SEM | Scanning electrode microscope |
CV | Cyclic voltammetry |
EIS | Electrochemical impedance spectra |
Footnote |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d2ya00177b |
This journal is © The Royal Society of Chemistry 2022 |