**Recent Advances in Functionalized Mesoporous Silica Frameworks for E**ffi**cient Desulfurization of Fuels**

#### **Shruti Mendiratta \* and Ahmed Atef Ahmed Ali \***

Department of Chemical and Petroleum Engineering, University of Calgary, Calgary, AB T2N1N4, Canada

**\*** Correspondence: shruti.mendiratta@ucalgary.ca (S.M.); ahmed.atef.dr@gmail.com or

Received: 29 April 2020; Accepted: 3 June 2020; Published: 5 June 2020

**Abstract:** Considerable health and climate benefits arising from the use of low-sulfur fuels has propelled the research on desulfurization of fossil fuels. Ideal fuels are urgently needed and are expected to be ultra-low in sulfur (10–15 ppm), with no greater than 50 ppm sulfur content. Although several sulfur removal techniques are available in refineries and petrochemical units, their high operational costs, complex operational needs, low efficiencies, and higher environmental risks render them unviable and challenging to implement. In recent years, mesoporous silica-based materials have emerged as promising desulfurizing agents, owing to their high porosity, high surface area, and easier functionalization compared to conventional materials. In this review, we report on recent progress in the synthesis and chemistry of new functionalized mesoporous silica materials aiming to lower the sulfur content of fuels. Additionally, we discuss the role of special active sites in these sorbent materials and investigate the formulations capable of encapsulating and trapping the sulfur-based molecules, which are challenging to remove due to their complexity, for example the species present in JP-8 jet fuels.

**Keywords:** mesoporous silica; nanomaterials; desulfurization; fuel; JP-8

#### **1. Introduction**

According to the World Health Organization (WHO), exposure to air pollution is the topmost environmental health risk factor, having led to one in eight global deaths in 2016 [1]. Reducing the greenhouse gas emissions is listed as one of the sustainable goals of the United Nations which directed their climate change mitigation efforts to apply stringent regulations, promote intelligent transport systems and innovative construction techniques, and invest in cleaner energy from fossil fuels through greenhouse gas capture, storage, and sequestration [2]. As per the 2016 Climate and Clean Air Coalition (CCAC) report, fine particles (PM2.5) present in the air, arising as a consequence of high sulfur content in fuels, are among the major contributors to air pollution. The presence of sulfur in fuels is linked directly to increased air pollution through direct emissions of harmful sulfur oxide (SOx) compounds in the atmosphere, resulting in acid rain, as well as indirectly by deactivating/poisoning the catalysts used in the refining systems and decreasing the efficiencies of emission control devices [3]. For example, black carbon released from the combustion of diesel can be captured through filters; however, these filters perform well only in the presence of low sulfur concentrations. Therefore, desulfurization of fuels becomes imperative for the preservation of environment, protection of health, and for increasing the lifetime of components playing essential roles in power generation systems.

High chemical affinity of sulfur atoms for metallic cations have made metal-oxide loaded solid sorbents to play vital role in removing sulfur content at mid, as well as high, temperatures. Hydrodesulfurization (HDS) is one of the conventional methods to desulfurize fuel in refineries where organic sulfur is converted to H2S employing transition metal catalysts loaded on alumina substrates.

ahmed.atef1@ucalgary.ca (A.A.A.A.)

Low cost transition metal oxides of Mn, Zn, Fe, Co, and Mo act as active components of these catalysts due to their high thermal stability [4–9]. However, energy demanding operating conditions involving high temperatures and pressures, low octane/cetane values, and selectivity towards simple refractory sulfur compounds in comparison to more complex aromatic species makes this process less efficient and makes room for new alternatives. Although few companies, like Akzo Nobel, ExxonMobil, and Nippon Ketjen, have come forward with a new trimetallic HDS catalyst (Mo-W-Ni, NEBULA) capable of producing diesel oil with a sulfur content as low as 10 ppm sulfur content, this process is still very energy intensive and requires hydrogen. Much effort is now being directed towards low temperature purification systems with high adsorption capacities for sulfur by exploring the efficiency of adsorbents, such as zeolites, metal–organic frameworks (MOFs), carbonaceous materials, ionic liquids (IL), and silica (Scheme 1). However, the efficiency of conventional adsorptive desulfurization methods can only be enhanced in the presence of better adsorbents or by using a combination of sulfur removal platforms. 

**Scheme 1.** Various porous materials reported for desulfurization of fuels.

The discovery of mesoporous molecular sieves (MCM-41) possessing hexagonal porous channels was a hallmark in materials science, not only in terms of opening avenues of new materials, such as mesoporous silica nanoparticles, mesoporous carbon nanosheets, and MOFs, for catalysis, gas adsorption, and drug delivery but also towards creation of new zeolites and their derivatives [10–20]. Structural marvels of mesoporous silica (MS) have also motivated researchers in recent years to investigate their performance in desulfurization of fuels, and they are being investigated as supports for metal oxides, owing to their high porosities, large surface areas, structural flexibility, and easy functionalization. High pore volumes and confined pore characteristics of one dimensional MCM-41 (with a 2–4 nm pore diameter) and two dimensional Santa Barbara Amorphous (SBA)-15 (with a 4.5–30 nm pore diameter) and their derivatives make them ideal representatives for adsorbing common aromatic S-based molecules present in fuels, such as benzothiophene (BT), dibenzothiophene (DBT), or

4,6-dimethyldibenzothiophene (4,6-DMDBT). To the best of our knowledge, this review article is the first to focus on the design and performance of mesoporous silica frameworks and the corresponding nanoparticles for desulfurization of fuels.

In this review, we discuss the synthesis, functionalization, and desulfurization performance of various MS frameworks. We focus our attention on the promising adsorption techniques using MS frameworks containing photocatalytic, redox, magnetic, ionic liquid, and carbonaceous active sites. A brief account of the synthetic strategies for preparing a MS catalytic host framework is given, along with commercial, as well as developing, desulfurization strategies. In addition, we discuss the mechanisms for removal of sulfur rich species wherever possible and make efforts to understand the role of special active sites in such sorbents. Finally, we investigate the formulations capable of encapsulating and trapping the sulfur-based molecules that are more complex and difficult to remove, such as the ones present in JP-8 and JP-5 jet fuels. In our opinion, this review will be beneficial for a wide audience working in the fields of energy production and environmental remediation.

#### **2. Strategies to Synthesize Mesoporous Silica Catalytic Host Materials**

High surface area of MS makes it a very promising material for the adsorptive desulfurization and hydrothermal methods. The meso-templates method is one of the conventional methods for MS catalyst preparation. A typical MS synthesis involves the self-organization of surfactant species, followed by their treatment with silica precursors. Condensation of the precursors around surfactants and their consecutive removal via reflux or calcination generates perfect MS frameworks. However, the hydrothermal method using various cationic, anionic, and neutral surfactants makes this process expensive and non-ecofriendly. Fortunately, these shortcomings can be overcome by incorporating new green and economical synthetic strategies using renewable templates. In recent years, sustainable development methods involve new templating strategies using nanocellululose, triblock copolymers, and task-specific ionic liquids as templates, instead of conventional surfactants. Some methods mentioned in this section even bypass the traditional templating strategies by using molecular imprinting techniques and single or dual phase hydrolytic conditions.

#### *2.1. Templating Strategies*

#### 2.1.1. Nanocellulose Template

One of the most abundant biomaterial present in nature is cellulose, which is a low cost, reusable, and biodegradable material. Upon acid hydrolysis, this material can be transformed to form nanocellulose fibers with lengths ranging from 50 nm to a micron [21]. Nanocellulose has the potential to become an environmentally benign templating alternative. It can offer benefits of both soft and hard templates, owing to its unique structural and chemical features, including high aspect ratio and low density, as well as presence of hydrophilic surface hydroxyl groups [22,23]. Recently, nanocellulose was used as a template by Shen et al. for the development of tungsten-impregnated mesoporous silica catalysts [24]. The resulting disordered worm-like mesoporous catalysts had highly dispersed tungsten species and high surface areas in the range of 344–535 m<sup>2</sup> /g. It was observed that the pore diameter (2.3 nm to 10 nm) and pore volume (0.32 cm<sup>3</sup> /g to 0.95 cm<sup>3</sup> /g) increased significantly when the dosage of nanocellulose templates was increased in the precursor solution.

#### 2.1.2. Triblock Copolymer (P123) as a Template

Several types of triblock copolymers are used in the synthesis of porous materials as the pore-directing agents. One widely used symmetric triblock copolymer with favorable characteristics is Pluronic P123, which comprises poly(ethylene oxide)–poly(propylene oxide)–poly(ethylene oxide) repeating units. P123 exhibits high hydrothermal stability, as well as amphiphilic properties, that can form stable micelles [25]. Generally, P123 can be used to form 2-D hexagonal mesoporous silica structures using acidic conditions and room temperature [26]. However, alteration of the synthesis

conditions, such as the addition of sodium dodecyl sulfate (SDS) [27], n-butanol [28], or NaI [29], can result in the formation of cubic *Ia*3*d* mesoporous silica structures. In 2020, Tagaya et. al. successfully synthesized novel biocompatible slit-shaped mesoporous silica/hydroxyapatite hybrid particles using P123 and Cetyl trimethylammonium bromide (CTAB) for biological applications. [30]. It was found that the use of P123 as a template was beneficial in increasing the pore size of the mesoporous structure of synthesized particles. Moreover, the use of P123 imparted dispersion stability for the particles at the monodisperse state.

#### 2.1.3. Task-Specific Ionic Liquid Strategy

Task-specific ionic liquids (TSIL) have gained significant attention in recent years for their role in synthesizing nanoparticles and chiral moieties with tailorable characteristics (physical, chemical, and biological) [31]. A series of mesoporous silicas (which were further loaded with zirconium derived nanoparticles) was successfully synthesized from hydrolysis of the precursor material tetraethoxysilane (TEOS) in acidic medium in the presence of the ionic liquid 1-hexadecyl-3-methylimdazolium bromide, which acted as a template material for the mesoporous silicas. The synthesized mesoporous silica structures showed high capacity to be impregnated with the zirconium derived nanoparticles [32]. Similarly, high-surface-area mesoporous silicas were fabricated using the task-specific ionic liquid 1-alkyl-3-methylimidazolium hydrogen sulfate as the template material from the precursor material TEOS [33]. In addition, it was reported that the use of the long-chain ionic liquid 1-hexadecyl-3-methylimidazolium chloride can produce two-dimensional hexagonal p6mm and cubic *Ia*3*d* mesoporous silica structures, depending on the synthesis conditions from the precursor material TEOS [34].

#### 2.1.4. Polystyrene Colloidal Crystal Template

The use of colloidal crystals as templates for the synthesis of various mesoporous structures has gained adequate recognition due to their ordered interparticle pores and other desirable characteristics [35]. Yang et al. prepared bimodal ordered porous silica structures with mesoporous macropore walls, which were formed from polystyrene colloidal crystal templates. They used the seed emulsion polymerization method to prepare the polystyrene template films, which were used to prepare the silica structures from the precursor material TEOS [36]. Míguez and coworkers reported the successful fabrication of bimodal mesoporous silica structures within polystyrene colloidal crystal films that were formed on the surface of glass substrate from the precursor material TEOS using the spin-coating method [37]. In 2019, Peng et al. could achieve the synthesis of a hybrid nanostructure composed of a core of hollow mesoporous silica nanoparticles coated with a shell of pH-responsive and thermo-sensitive moieties as a model of drug delivery platform [38]. The authors used the soap-free emulsion polymerization method to prepare monodisperse polystyrene microspheres as a template for the fabrication of their hollow mesoporous silica nanoparticles, which were monodispersed with adjustable mesoporous silica shell thickness and particle size (depending on the fabrication parameters). The synthesis of ordered macroporous silica particles from the precursor material tetramethyl orthosilicate (TMOS) using polystyrene colloidal crystal as a template was also reported [39].

#### *2.2. Molecular Imprinting Polymer Technology*

Molecular imprinting technique offers an alternative method for deep desulfurization, where refractory organosulfur species can be easily removed under mild conditions [40]. In this technique, functionalized monomers are made to react and crosslink with template molecules. After the reaction, the template molecules are extracted leaving behind well-defined arrangement of ligands structures and tailor-made binding pockets. Compared to other structurally similar compounds, surface molecularly imprinted polymers have distinct affinity towards their templates. There are several benefits associated with their usage, such as high guest absorption and access, reduced mass transfer resistance, feasible

extraction of template molecules, rapid adsorption kinetics, structural tenability, and easy synthetic protocols [41–43]. As binding capacity of these polymers critically depends on surface area of templates, new materials with high porosity and surface areas are in demand and mesoporous silica materials are ideal candidates for this role, thanks to their high surface areas, uniform pores, and structural and chemical tunability, as well as diverse morphologies, which makes them suitable materials for molecular imprinting technology.

#### *2.3. 1D Silica Fibers and Nanowires*

Development of carbon nanotube mimicking one dimensional mesoporous silica micro and nanostructures have gained much attention in recent years for applications, such as waveguides and laser components, and several different strategies are being implemented to make these materials with a diameter less than 50 nm. Electrospinning methods can generate silica fibers with an average diameter of 40 um [26], while anodic aluminium membrane strategy [27] can help in growing fibers that have small diameters of about 250 nm. In contrast, two-phase solution methods have shown the growth of mesoporous silica fibers with average diameters in the range of 1–15 um [28], and this method can even be extended to one-phase (aqueous) hydrolysis method in the presence or absence of acidic environment to produce 1D fibrous silica-based materials with 50–300 nm diameters [29]. Transforming 1D silica fibers into materials with unique pore architecture is of prime importance for materials scientists. Judicious design of mesoporous silica materials with hierarchical pores is associated with benefits, such as efficient host guest interactions, high catalytic performance, and better mass transport properties [30]. Although high porosity and large surface areas are recognized as key features for their implementation, orientation of mesopores should also be given equal importance. Studies have shown that mesopores that are parallel with the length of 1D silica fibers can limit catalyst loading and decrease active catalytic sites by blocking the longitudinal channels, whereas mesopores aligned perpendicular to the axial direction of 1D channels permit easy catalyst access and higher effective surface area [31,32].

#### **3. Strategies to Desulfurize Fuel**

#### *3.1. Hydrodesulfurization (HDS)*

Hydrodesulfurization is one of the conventional methods to desulfurize crude oil where concentration of the sulfur containing compounds is reduced. However, HDS takes place in the presence of harsh operating conditions requiring elevated temperatures in the range of 300–450 ◦C, hydrogen, catalysts, and high pressures ranging from 3–5 MPa, making it an expensive desulfurization technique [44]. HDS is certainly effective in removing simple thiols, thioethers, and disulfides; however, it is not successful in removing other complex heterocyclic molecules, such as dibenzothiophene (DBT), 4,6-dimethyldibenzothiophene (4,6-DMDBT), and their derivatives [44]. A typical HDS catalysis generates H2S gas which is converted to elemental sulfur through a Claus process [45]. Harsh catalytic conditions cause hydrogenation of olefins, followed by a negative octane rating and increased H<sup>2</sup> consumption, whereas a mild HDS environment generates recombinant mercaptans through the reaction of H2S with olefins and are responsible for retained sulfur in final fuel. Desulfurization mechanism of DBT using hydrogen at 300 ◦C and 102 atm is suggested to proceed via two major pathways, direct desulfurization (DDS) and hydro desulfurization (HYD) pathways. The DDS pathway involves elimination of sulfur without hindering the aromatic rings, while HYD proceeds via selective hydrogenation of DBT aromatic rings, followed by desulfurization [44]. Molybdenum-, nickel-, cobalt-, and tungsten-based catalysts are commonly used and incorporated in the HDS process to reduce sulfur content in the form of disulfides, thioethers, and mercaptans. Performance of HDS process is greatly influenced by the choice of catalyst, and the efficiency has been found to be high in case of second and third row elements in the periodic table, owing to their unique electronic and structural characteristics. Several bimetallic blended catalysts, including alumina supported NiMo, NiW, CoW, CoMo, NiMo, and PtPd, have shown better catalytic activity in the past through highly selective and efficient hydrogenation [44,46–48]. In particular, nano-sized noble metals catalysts, including Pd, Pt, and Rh, have shown superior hydrogenation performance and high activities for deep hydrodesulfurization. However, their high susceptibility to sulfur poisoning due to the adsorption of H2S on noble metals restricts their widespread usage [49,50]. Additional drawbacks of using noble metals in hydroprocessing catalysts have been summarized by Marafi and Furimsky that include deactivation by coke deposition, inhibition by sulfur and oxygenates, poisoning by nitrogen and chlorides, deposition of contaminant metals, agglomeration of active metals, and leaching of active metals [51]. Different modes of deactivation may take place simultaneously, and adverse efforts are made for catalyst recovery by regeneration. The issue of sulfur poisoning is overcome by using phosphides of noble metals (Rh2P, Ru2P, Pd5P2), by using transition metal phosphides (Co2P, CoP,Ni2P, MoP, WP), or by using acidic supports (Al-SBA-15) [49,50]. Among various supports, SBA-15 has been used widely as a hydrodesulfurization support. Bimetallic catalysts impregnated SBA-15 (CoMo-SBA-15, NiMo-SBA-15) have shown better thiophene conversion compared to γ-Al2O<sup>3</sup> supported catalysts [52,53]. Some excellent reports and reviews are available on this topic and reveal that mesoporous silica frameworks, including SBA-15, MCM-41, MCM-48, and KIT-6, are promising HDS agents when they are incorporated with active species, such as WS2, NiW, Ni2-xMxP (M = Co, Fe), MoS2, and NiMoO4, or with heteropolyacids, like H3PW12O40, H3PMo12O40, and (NH4)6Mo7O24.4H2O [50,54–56]. Cost associated with HDS and stringent fuel specifications in recent times have motivated researchers to explore new innovative strategies.

#### *3.2. Oxidative Desulfurization (ODS)*

ODS has emerged as a promising desulfurization technology, owing to its mild operating conditions requiring ambient temperatures and atmospheric pressures, no H<sup>2</sup> prerequisites, and high efficiency towards the refractory thiophenic sulfides (DBTs). In a typical ODS process, oxidation of heavy sulfides takes place by incorporation of oxygen atoms at sulfur sites, without rupturing C–S bonds in the presence of judiciously selected oxidants, such as H2O2, O3, NO2, peroxy salts, or tertbutyl-H2O2. Out of these, H2O<sup>2</sup> is the most widely used oxidant; however, in recent years, transition metal-based oxidizing agents (Re, Mo/W oxides, tungstophosphoric acid, MPcS and persulfates, tetra-amido macrocyclic ligands) have become popular and have been investigated in combination with hydrogen peroxide [57–60]. The success of an ODS process depends on the textural, structural, surface, and chemical properties of an active catalyst, combined with an appropriate oxidant. Other methods of oxidative desulfurization involve irradiation techniques, ultrasonic methods, direct photo-oxidation, and photocatalytic, electrocatalytic, and plasma techniques [44,61]. Oxidation is accompanied by liquid extraction to separate the oxidized components because of their enhanced polarities. The efficiency of extraction is influenced by solvent polarity, boiling point, and freezing point, as well as surface tension. In addition to common water-soluble polar solvents, ionic liquids are employed to extract S-compounds directly or after they have been oxidized. Recently, a combination of extractive and oxidative desulfurization called extractive catalytic oxidative desulfurization (ECODS) has gained recognition as a technique for producing ultra-low sulfur clean fuels, making using of ionic liquids as extractants and H2O<sup>2</sup> as the oxidant, easily converting organosulfur compounds to sulfoxides and sulfones [62]. Issues associated with ODS, such as lack of oxidant selectivity, unwanted side reactions, undesirable residual S-containing aromatic moieties due to incorrect choice of extraction solvent, and high cost, are carefully weighed against the benefits linked with this process that cannot be achieved in conventional HDS process. For more information, we direct our readers to some new reviews that have appeared in literature focusing on ODS catalysts for clean environment, ODS of heavy oils with high sulfur contents, ODS using heteropolyacid-based catalysts, and ultrasound assisted ODS of hydrocarbon fuels [63–67].

#### *3.3. Adsorptive Desulfurization (ADS)*

There is a direct link between microporosity and adsorption capacity for efficient desulfurization. There has been a quest for materials with pores larger than 10 Å and adsorbents, such as zeolites, activated carbon, metal-organic frameworks (MOFs), aluminosilicates, and ZnO, have been tested for capturing and eliminating organosulfur compounds present in high value fuels. In addition to having excellent porosity, an ideal adsorbent has high available surface area, high selectivity, good surface chemistry, and recyclability for efficient removal of S-compounds. ADS can proceed through two main approaches: physical adsorption, a less energy intensive pathway where the S-based compounds are not chemically modified on separation; and reactive adsorption that offers benefits of both catalytic HDS and physisorption. A chemical reaction taking place between S-based moieties and solid adsorbent in the latter approach releases sulfur in the form of H2S, Sox, or elemental S. For practical industrial applications, more studies and engineering at a molecular scale is required to adjust the size of host pores to match the guest size and to avoid any steric hindrance between guest molecules.

#### *3.4. Biodesulfurization (BDS)*

With advancement in biotechnology, green processing of fossil fuels has become possible, and microorganisms with an affinity for sulfur are playing an important role in metabolizing organosulfur compounds in fuels. This technology has potential to become cost effective and efficient. Microorganisms, such as *Pantoea agglomerans*, *Alcaligenes xylos-oxidans*, *Rhodococcus erythropolis*, *Mycobacterium*, and thermophilic *Paenibacillus*, have been identified for aerobic BDS, while *Desulfovibrio desulfuricans*, *Desulfomicrobium scambium*, and *Desulfovibrio longreachii* have been for anaerobic BDS. Kirkwood et al. indicated in their studies that specificity for sulfur and metabolic pathways may not be dependent only on sulfur but, rather, on the type of species used [68–70]. Metabolism of organosulfur compounds and C–S bond cleavage by bacteria is suggested to proceed either through reduction of C–S bond and oxidation of C–S bond, or through oxidation of C–C bond (Kodama pathway) [71,72]. Although BDS has several advantages over conventional desulfurization processes, it can be uneconomical due to the additional cost associated with culturing the bacteria [73]. Other important drawbacks include phase change (fuel-aqueous phase) and stability of cells in the presence of fuel species.

#### *3.5. Other Forms of Desulfurization*

British Petroleum had developed and tested alkylation-based desulfurization on light oils, where acid catalysed aromatic alkylation is performed on thiophenic compounds with an aim to upgrade olefinic gasoline by increasing the molecular weight and boiling point of alkylated sulfur compounds, so that their separation becomes feasible [74]. Another simplified form of alkylation is S-alkylation, where thiophenic molecules are made to reaction with methyl iodide and fluoroborate salts of silver, leading to alkylated sulfonium salts which are precipitated and easily eliminated without distillation [75]. However, S-alkylation is not selective in these reagents, owing to their affinity to aromatic hydrocarbons, making this process challenging in carbon rich heavy oils.

C–S and S–S bonds can also be cleaved by chlorinolysis, where chlorine is able to bind at sulfur sites. This process takes place under moderate operating conditions (25–80 ◦C, atmospheric pressure) in fuels that can be easily homogenized with chlorine and which are corrosion resistant in the presence of chlorine. In order to remove impurities, additional steps comprising hydrolysis, oxidation of sulfur, and several aqueous and caustic washes are usually performed after chlorination.

Supercritical aqueous desulfurization or supercritical water desulfurization (SCW) has been shown to break C–S bonds in non-aromatic sulfur molecules through free radical pathway. However, some studies have shown that SCW independently cannot desulfurize fuels when not in the presence of H<sup>2</sup> and conventional HDS catalysts. There are indeed some benefits associated with SCW, such as increase in liquid yield, precipitation of sulfur saturated compounds, and H<sup>2</sup> generation through water

gas shift. To discuss further on industrially applicable methods for desulfurization is beyond the scope of this review; thus, for more information, we direct readers to some excellent reviews on this topic provided by Javadli et al. [61] and Srivastava et al. [44], as well as advances in biodesulfurization by Sadare et al. [73].

#### **4. Mesoporous Silica and Mesoporous Silica Nanoparticles in Desulfurization**

#### *4.1. Photocatalytic MSs*

Photocatalytic adsorption desulfurization (PADS) has emerged as a form of oxidative desulfurization method due to the low cost, high stability, and recyclability of photocatalytic materials [76,77]. Solar power is capable of exciting the photocatalysts and generating holes that are great oxidants. Certain aromatic sulfurous molecules, including DBT, DBTO (dibenzothiophene sulfoxide), and DBTO2, can be oxidized in the presence of free radicals, such as •OH and •O<sup>2</sup> <sup>−</sup>, and PADS is useful in deep desulfurization requirements. Recently, Zhou et al. reported on the use of mesoporous ZnO/TiO2–SiO<sup>2</sup> (ZTS) as a photocatalyst and adsorbing material towards organic sulphides [78]. ZnO is known for its high photocatalytic activity and has a bandwidth of 3.2 eV (380 nm) [79]. Previous studies have shown that performance of ZnO can be enhanced dramatically in the presence of TiO<sup>2</sup> by inhibiting the recombination of electrons and holes, extending their lifetime and enlarging the photoresponse range [80]. Inspired by these findings, Zhou et al. implemented these studies for synthesising effective desulfurizing agents [78]. ZnO were chosen as active sites in the mesoporous TiO2–SiO<sup>2</sup> catalyst, and the TiO2–SiO<sup>2</sup> precursor materials were synthesized hydrothermally in the presence of triblock copolymer (P123) as a template, so as to tune the pore size. ZnO was incorporated using impregnation techniques, and the ratio of Si/Ti was adjusted in order to achieve optimized adsorption PADS capacity. The authors showed that ZTS-3 with Si/Ti = 3 exhibited the best photocatalytic desulfurization activity. The DBT conversion rate was found to be 97% in 4 h, and maximum adsorption of DBT over ZTS-3 was 47 mgS per gram of catalyst used and the adsorption rate was much higher than mesoporous TiO2–SiO2-40 (13.7 mg S/g-cat) [81]. The photocatalytic activity of ZTS was attributed to the heterojunction formed through the interactions of ZnO with TiO2, leading to the expansion of sunlight absorption, enhancing the efficiency of charge separation, and inhibiting electron–hole recombination. Interestingly, this desulfurization could proceed without involving extra oxidants, such as hydrogen peroxide, O2, or organic oxidants, making it a cost-effective process that exhibits high PADS performance and excellent adsorption capacity. As shown in Figure 1, the photocatalytic mechanism is suggested to proceed through the formation of •O<sup>2</sup> <sup>−</sup>, which is capable of oxidizing DBT to DBTO<sup>2</sup> [78].

TiO<sup>2</sup> in varying structures and morphologies are among the most promising photocatalytic agents for different applications, such as fuel desulfurization and degradation of contaminants in various systems. In 2017, Meizhen et al. prepared a photocatalytic TiO2-modified bimodal mesoporous silica structure (TiO2/BMMS) and compared its desulfurization of dibenzothiophene efficiency to that of the mono-modal mesoporous (TiO2/SBA-15) catalyst and pure TiO<sup>2</sup> [82]. They found that the photocatalytic activity and desulfurization efficiency of TiO2/BMMS was the highest, achieving a desulfurization rate of 99.2%, followed by TiO2/SBA-15 and then pure TiO2. In 2014, Zaccariello et al. found that TiO<sup>2</sup> nanoparticles confined within mesoporous silica nanospheres exhibited enhanced photocatalytic activity due to the added adsorptive effects and thermal stability of the mesoporous silica nanospheres [83].

−

**Figure 1.** Photocatalytic desulfurization mechanism of ZnO/TiO2–SiO<sup>2</sup> (ZTS-3) catalyst under solar light [78].

#### *4.2. Redox Active MSNs*

Conventional desulfurization strategies are associated with issues that hinder their implementation in desulfurization industry, and these problems include high concentration of oxidants, such as hydrogen peroxide, and elongated reaction times [84]. In order to overcome these issues, it becomes essential to formulate suitable catalysts for an eco-friendly and efficient oxidative desulfurization process. Polyoxometalates (POMs) belong to a category of transition metal-based oxygenated anionic clusters, which have emerged in recent years as promising materials for homogeneous oxidation desulfurization (HOD) process, owing to their unique structural and chemical characteristics and their redox potential. Advances in nanotechnology has made possible the immobilization of POMs onto solid supports, such as metal organic frameworks (MOFs), zeolites, porous carbons, and porous silica materials [85–88]. This addition not only imparts stability and reusability to the POMs but also enhances their catalytic activity towards refractory S-based molecules. Specially designed dendritic mesoporous silica nanospheres with center-radial oriented large mesopores were reported earlier by Zhao et al. as potential carriers or substrates for development of new functional materials [89,90]. Interestingly, Zhang et al. took advantage of the above materials to design a molybdenum-embedded dendritic mesoporous silica sphere using Stöber approach [91]. As shown in Figure 2, they introduced active molybdenum species with a concomitant use of ionic liquid (IL) as a metal source into the mesopores of dendritic silica spheres (DSSs). The DSSs featured large specific surface areas and high pore volumes and possessed a highly dispersed molybdenum species. The authors explored optimum synthetic conditions and characterized ODS products using gas chromatography–mass spectrometry GC-MS analysis. The judiciously designed catalyst showed rapid and high catalytic activity in oxidizing 4,6-DMDBT, and the catalyst could be easily separated from a heterogeneous mixture after fulfilling its role. Under optimal conditions, elimination of 4,6-DMDBT could reach 100% in 40 min, and a low dosage of oxidant was required. The recycling performance of the DSS revealed that they could be reused nine times without an appreciable decrease in catalytic activity [91].

**Figure 2.** Schematic representation of dendritic mesoporous Mo-SiO<sup>2</sup> catalyst and its desulfurization mechanism [91].

Exemplary performance of monolacunary [PW11O39] 7- Keggin polyoxometalates in oxidative reactions [92–96] has inspired a few researchers to explore their performance in ODS process, too [97–99]. For example, Ribeiro et al. have worked extensively on Keggin-type materials, including PW<sup>11</sup> embedded on amine decorated SBA-15 (PW11@aptesSBA-15 and PW11@tbaSBA-15), and on cationic tma functionalized mesoporous silica supports for the desulfurization of model and real diesels (Figure 3) [100]. Oxidative desulfurization studies on amine functionalized SBA-15 supports revealed high performance in case of PW11@aptesSBA-15, which could completely desulfurize simulated diesel in solvent-free conditions with half the oxidant amount (H2O2/S = 4), whereas it showed 83.4% efficiency in case of real unthread diesel (in a biphasic system, 1:1 diesel/acetonitrile) with a recycle capacity for eight consecutive cycles. Their recent work involved impregnation of PW<sup>11</sup> on cationic group (*N*-trimethoxysilypropyl-*N, N, N*-trimethylammonium, TMA) functionalized MS supports [101]. Here, two kinds of MS supports were selected (ordered MS, SBA-15, and an ethylene-bridged periodic mesoporous organosilica, PMOE) and two catalysts were prepared, PW11@TMA-SBA-15 and PW11@TMA-PMOE. Oxidative desulfurization was carried out on simulant diesel under biphasic (1:1 diesel/acetonitrile) and solvent-free conditions. A remarkable desulfurization performance was exhibited by the PW11@TMA-SBA-15 catalyst, which could achieve ultra-low sulfur levels (<10 ppm) in both biphasic, as well as solvent-free conditions, and could be recycled six consecutive times without appreciable loss of catalytic activity. This promising catalyst was also tested on untreated real diesel provided by Spanish petroleum company, CEPSA (1335 ppm S) under biphasic system and demonstrated 90% desulfurization efficiency for 3 consecutive cycles [101].

Recently, Ribeiro et al. reported on zinc-incorporated polyoxotungstates [PW11Zn(H2O)39] 5−, PW11Zn for ODS of real, as well as model, diesels, where active species were embedded in periodic mesoporous organosilicas (PMOs) composed of walls with ethane-bridges and benzene bridges [102]. These compounds showed high desulfurization efficiency under solvent-free conditions and in the presence of H2O<sup>2</sup> with a H2O2/S ratio of 4, where ultra-low levels of sulfur could be obtained in just 1 h. PW11Zn@aptesPMOE catalyst was further investigated due to its robust nature for sulfur removal of real diesel, and an efficiency of 75.9% could be achieved after 2 h, and the catalyst could be reused.

A heteropolyacid-based adsorbent catalyst was reported by Yuzbashi et al. with an aim to decrease the leaching of active components and to increase the ODS kinetics of DBT present in model diesel fuel [59]. Phosphotungstic acid (HPW) was incorporated in a zirconium-modified hexagonal mesoporous silica (Zr-HMS) to give HPW/Zr-HMS catalyst, and its DBT removal ability was compared with Zr-HMS, HPW-HMS. Zr-HMS showed only 15% S-removal efficiency, whereas the efficiency of the two others was found to be similar; however, an efficiency of 99.4% was observed within 120 min when the composition of prepared catalyst was 20% HPW/Zr-HMS. An increase in the catalyst dosage from 0.03 to 0.05 g drastically increased S-removal efficiency from 10% to 99.4%. In the presence of H2O2, the model fuel was oxidized in less than 30 min and achieved an efficiency greater than 95% of the 350 ppm DBT. The formulated catalyst was recovered and was reused at least five times, and the leaching of HPW species was inhibited, indicating the promising nature of catalyst for ultra-deep desulfurization process [59].

**Figure 3.** Synthetic strategy to prepare Keggin's polyoxometalate anion incorporated PW11@TMA- (SBA)-15 and PW11@TMA-PMOE [101].

− There have been cases where, instead of using heteropolyacids, silicotungstic acids have been used directly for the preparation of mesoporous tungsten-silica catalysts. For example, Shen et al. reported on the formation of disordered worm-like mesoporous silica catalysts with well dispersed tungsten species using nanocellulose as templates [24]. High surface area of the formed catalysts in the range of 344–535 m<sup>2</sup> /g, their narrow pore size distributions (2 nm to 10 nm), and an optimum Si/W ratio supported high catalytic activity towards DBT and its elimination at 60 ◦C in just ten min. The catalytic activity towards S-compounds was found to increase in the following order: BT < 4,6-DMDBT < BT. GC–MS analysis of oxidized species showed that the catalyst formulations not only played their role as catalysts but also acted as efficient adsorbents. This catalyst could be recovered and be reused at least five times and still retain its desulfurization efficiency up to 94% [24]. The authors of this report had proposed a mechanism for the oxidation of DBT where the first step involves the adsorption of DBT in the pores of the catalysts, followed by its reaction with active peroxo-tungsten complex, which in turn is a direct consequence of reaction between tungsten species and H2O<sup>2</sup> as oxidant. The interactions between DBT and peroxo-tungsten complex converts DBT to DBTO2, which gets adsorbed in the mesopores of the catalyst, with the simultaneous reduction of peroxo-tungsten complex to tungsten oxide.

With advances in nanoscience, unique properties have been observed for particles with an average diameter lying in the range of 0.1 nm to 1.0 nm. This size range sets them apart from their nano counterparts, and some refer them as subnanoclusters [103,104]. A large number of catalysts are being developed using metal oxide subnanoclusters, and many materials have shown promising catalytic behavior for a large number of organic reactions. Along the same lines, the size of mesoporous silica materials is also decreasing, and new ultrasmall MSNs (UMSNs) with particle size less than 25 nm are being developed in order to reduce mass transfer resistance and enhance the catalytic activity. Some researchers, like Wang et al., reported on the synthesis of subnano-MoO<sup>3</sup> supported on ultrasmall MSN particles (~14 nm) with a peculiar "raisin-bun" structure, using reverse microemulsion technique (Figure 4a) [105]. This hybrid catalyst, subnano-MoO3/ultrasmall MSN (UMSN), was then used for ODS of a model diesel comprising of benzothiophenes, where they were oxidized to DBTO<sup>2</sup> with a 100% efficiency within 15 min (Time of flight, TOF of ODS: 53.3/h) (Figure 4b) [106]. Percentage of catalyst, reaction time, and temperature influenced the DBT conversion; however, the percentage of oxidant did not play a significant role. Surprisingly, performance of nano-MoO3/UMSN tested under similar operating conditions showed a DBT to DBTO<sup>2</sup> conversion of only 25.7% (TOF was 10.2/h), proving that subnano-MoO<sup>3</sup> was more active than nano-MoO3. It is well known that smaller size generates high catalytic active sites with higher ratio of surface to bulk atoms; however, an 80% efficiency drop was attributed to the higher binding energy associated with subnano particles (1.1 eV higher than bulk), appearing as a result of lowered core-hole screening existing in small clusters. This indicates that electronic properties show significant variation with size and can show unusual response at subnano level.

**Figure 4.** (**a**) Schematic representation for the formation of subnano-MoO<sup>3</sup> /ultrasmall mesoporous silica nanoparticles (UMSN); (**b**) catalytic mechanism of MoO<sup>3</sup> catalyzing dibenzothiophene (DBT) oxidation in the presence of TBHP (tert-butyl hydro peroxide) [106].

α − Development of 1D silica fibers with diameters of about 50 nm or less and their transformation into materials with unique pore architectures, especially with pores aligned perpendicular to the direction of fibers, has become of paramount importance for materials scientists. Aiming to improvise catalytic and sorption properties of silica materials, Dou and Zeng developed interconnected mesoporous silica nanowires with a 3D network having shallow wormhole channels aligned perpendicular to the axis of nanowires using a simple synthetic strategy, where TEOS hydrolysis took place under basic conditions (Figure 5) [107]. They explored synthetic parameters, including the effect of cosolvents, surfactants, alkalinity, duration of reaction, and optimum aging temperatures, to get an insight on the formation mechanism. α-MoO<sup>3</sup> species were immobilized on the interconnected mesoporous SiO<sup>2</sup> nanowires to form a Mo/mSiO<sup>2</sup> catalyst−adsorbent system and was examined in ODS of model diesels.

Although, in these studies, the diameter of Mo/mSiO<sup>2</sup> nanowires was significantly reduced and reached in the range of 10–20 nm, the authors were able to increase the final dimensions of the catalyst to a micron level, allowing easy recovery and separation after being used. The BET surface area of Mo/mSiO<sup>2</sup> reduced from 852 to 503 m<sup>2</sup> /g on increasing the Mo loading from 2 to 15%. The recipes containing 5 to 10% Mo showed highest conversions of DBT, owing to an optimum balance between total working catalyst and effective surface area. For a 10% Mo/mSiO<sup>2</sup> nanowire recipe, a DBT conversion efficiency remained as high as 95.4% in the first 30 min, even after seven cycles, and the sulfur concentration was 33 ppm after the seventh run.

−

α

**Figure 5.** (**a**) Structure of mesoporous silica nanowires with mesopores aligned parallel or aligned perpendicular to the axial direction of nanowires; (**b**) desulfurization mechanism showing reaction of DBT unit with catalytic species and removal of DBTO<sup>2</sup> molecules after oxidation [107].

#### *4.3. Ionic Liquid MSN Catalysts*

Ionic liquids (ILs) have emerged in the past few years as green solvents and catalysts, thanks to their unique properties that include high thermal stability, feasible design, and synthesis, as well as a wide liquid range [108]. These features allow dissolution of several catalysts in the ionic liquids providing a biphasic liquid system beneficial not only in esterification [109] and fructose dehydration [110] but also in oxidation of sulphides [111,112]. However, necessity for large amounts of ILs and difficult catalyst recovery makes it challenging to apply them in wide range of organic reactions, except when these ILs can be heterogenized. Therefore, many attempts have been exerted towards fabrication of "supported ionic liquid catalysts" (SILCs) that can immobilize homogeneous ILs on appropriate host materials, which can range from silica-based materials and MOFs to carbon nanotubes [113–118]. Advantages associated with use of mesoporous silica have compelled researchers to use impregnation and grafting methods to prepare SILCs, but these efforts are accompanied by other issues, like reduction in surface area of supports, leaching of active components, or confinement in the degree of freedom of ILs [119,120]. These limitations can be overcome by incorporation of functionalized ILs in the grafting matrix, by performing one pot hydrothermal synthesis in architectural supports, and by removing templates in subsequent reactions [121–123]. In our opinion, it is worth discussing examples where ionic liquids have been immobilized on mesoporous silica materials to be used for desulfurization process.

A few years ago, Gu et al. reported on the synthesis of hybrid mesoporous silica material using task-specific ionic liquid (TSIL) templating strategy, where POM-based IL [(n-C8H17)3-NCH3]<sup>2</sup> [W2O3(O2)4] (termed as T8W2O11) was used as template [124]. The authors blended the amphiphilic T8W2O<sup>11</sup> with silicate source tetraethyl orthosilicate (TEOS). Silicate condensation lead to the entrapment of T8W2O<sup>11</sup> species, and subsequent calcination eliminated the organic cations, leaving behind a functionalized mesoporous silica framework. Catalytic activity of the resulting product was tested in oxidative desulfurization process, where 99.6% efficiency was observed in the case of DBT sulfur removal using low catalyst dose and without using organic solvents. The efficiency of catalytic oxidation of organosulfur compounds decreased in the following order: DT > 4,6-DMDBT > DBT > BT. The efficiency of the catalyst could be retained, even after recycling eight times [124].

This work was followed by Zhang et al., who reported encapsulation of polyoxometalate-based ILs with the formula [Cxmim]3PW12O<sup>40</sup> (Cx-IL, x= 4,8, and 16) into the pores of ordered mesoporous silica through a one-pot hydrothermal reaction (Figure 6a) [125]. The as-prepared catalytic material (Cx-IL@OMS) was systematically characterized using multiple chemical and physical techniques, proving its high specific surface area and uniform dispersion in silica supports. These catalysts were employed in the desulfurization of organosulfur compounds, and [C4mim]3PW12O40@OMS was a winner among the series (Figure 6b). It showed high catalytic activity, high stability, and could be recycled multiple times. When compared with other IL@OMS, catalytic oxidation of DBT decreased in the following order: C4-IL@OMS (99.5%) > C8-IL@OMS (76.7%) > C16-IL@OMS (30.5%). High performance of C4-IL@OMS could be ascribed to its large surface area providing highly exposed active sites for oxidation of DBT and short IL chain imparting low steric hindrance and enough platform for peroxide species to interact with DBT molecule. Interestingly, catalytic oxidative desulfurization efficiency of pure C4-IL (S removal:12.7%) was much lower in comparison with C4-IL@OMS (S removal:99.5%). Additionally, even after recycling seven times, the removed sulfur in case of C4-IL@OMS exhibited an impressive 93% efficiency [125].

**Figure 6.** (**a**) Fabrication of ordered mesoporous silica encapsulating polyoxometalates (POMs)-based ionic liquids (IL); (**b**) possible desulfurization mechanism for the oxidation of DBT to DBTO<sup>2</sup> with concomitant transformation of hydrogen peroxide to water [125].

Prior studies have also shown lanthanide-based POM clusters Na7H2LaW10O36·32H2O (LaW10) as promising catalysts for ECODS process [126,127]. However, new materials are required to make ECODS more efficient and highly selective in terms catalyst separation, recovery, and reusability. As shown in Figure 7, Chen and Song impregnated LaW<sup>10</sup> on dihydroimidazolium (ionic liquid) modified mesoporous silica supports, forming LaW10/IL-SiO<sup>2</sup> [128]. The new catalyst exhibited deep desulfurization of several organosulfur compounds, including DBT, BT, and 4,6-DMBT, under mild operating conditions and in less than 30 min.

Ultra-deep desulfurization (<100 ppm) could be achieved for DBT in just 1 min for small batches, whereas scaled-up studies (1 litre model oil with S content of 1000 ppm) showed 100% sulfur removal in 25 min and at 70 ◦C. Additionally, the process was cost-effective as reusability of catalyst did not require simultaneous use of ionic liquid each time, recovery of LaW10/IL-SiO<sup>2</sup> catalyst was easy, and it could be reused at least ten times without degrading its efficiency, making it a promising material for ultra-low sulfur fuels [128].

Although some mesoporous silicas, such as MCM-41 and SBA-15, possess uniform pore size and a large internal surface area, their small pore size at times can lead to blocking and collapsing of pores during the desulfurization process, eventually giving poor recycling performance. In order to overcome these challenges, three-dimensional ordered macroporous (3DOM) materials with well-connected pores, large pore size sufficient to catalyze most fuel related organosulfur compounds, better mass transfer properties, and high surface areas are being developed [129,130]. Silica-based meso-macroporous materials (H3PW12O40/SiO2) synthesized via polystyrene colloidal crystal template

−

and using evaporation-induced self-assembly (EISA) have been reported by Lei et al. and applied in the ODS process [131,132]. In both cases, hierarchically meso/macroporous catalysts exhibit superior catalytic activity than their purely mesoporous and macroporous counterparts. This enhanced performance can be attributed to the larger mesoporous specific surface area, peculiar hierarchically meso/macrochannels structural characteristics and shorter length of mesoporous channels that facilitate mass transfer of precursors and products. Recently, Chen et al. reported on a POM-based ionic liquid supported 3DOM silica (IL-3DOM SiO2) as a catalyst for heterogeneous oxidative desulfurization [58]. The catalyst exhibited high porosity, as well as, a high specific surface area. This catalyst was prepared using the colloidal crystal template assembled from polymethylmethacrylate (PMMA), having a mean diameter of approximately 240 nm. The 3D crosslinked structure of catalyst allows guest diffusion, capture, and exposure of the active species. Sulfur removal efficiency of an IL-3DOM SiO<sup>2</sup> catalyst in the absence of H2O<sup>2</sup> was only 5.7%, while it was 100% in its presence, indicating the ability of 3D crosslinked structure to effectively adsorb organo sulfur molecules and essential presence of tungsten-based IL to activate H2O<sup>2</sup> to oxidize S-compounds. The desulfurization performance of IL-3DOM SiO<sup>2</sup> catalyst decreased in the following order: 4-MDBT > DBT > 4,6-DMDBT > 3-MBT > BT. Additionally, even after recycling 17 times, the sulfur removal efficiency could still reach 94% in the absence of a regeneration process [58]. The mechanism of DBT desulfurization by IL-3DOM SiO<sup>2</sup> is shown to proceed, firstly via adsorption of DBT in the macropores and then its reaction with peroxo species formed in turn from the combination of POM-IL with H2O2. Subsequent steps involve the oxidation of DBT to the corresponding sulfones (DBTO2), their precipitation, and presence in the catalyst phase. Towards the end of cycle, the peroxo species convert into active species, which further combine with H2O2, forming species for the next cycle [58].

− **Figure 7.** Schematic illustration for the highly selective and efficient removal of sulfur species using LaW10/IL-SiO<sup>2</sup> nanocomposites. (**a**) Extraction of the substrate by ionic liquid, (**b**) catalytic oxidation of DBT to DBTO<sup>2</sup> by LaW<sup>10</sup> in the presence of H2O<sup>2</sup> (R = −C8H17) [128].

Compared to conventional mesoporous silicas, such as MCM-41 or SBA-15, the reports on the corresponding nanoparticles as adsorbents for desulfurization are very limited. Last year, Mirante et al. reported on composite materials prepared through the immobilization of Keggin polyoxomolybdate [PMo12O40] 3- anion on MCM-48 type mesoporous silica nanoparticles functionalized with surface-tethered tributylammonium (TBA) groups [133]. The composite catalyst PMo12@TBA-MSN had high stability and was highly efficient in the oxidative desulfurization (ECODS) of a diesel simulant in the presence of H2O<sup>2</sup> (oxidant) and [BMIM]PF<sup>6</sup> (IL and solvent). The shorter channels of MSNs and cubic mesoporous framework of MCM-48 nanoparticles with branched networks seem to allow mass transfer in comparison to conventional porous silica materials with unidirectional

structures. The ECODS studies were carried out on a model diesel composed of BT, DBT, 4-MDBT, and 4,6-DMBT in n-octane with a total sulfur content of 2016 ppm. For the first ECODS cycle, the efficiency of PMo12@TBA-MSN was high and comparable to the homogeneous catalyst PMo12, where desulfurization could be achieved in 2 h. However, PMo<sup>12</sup> could not be reused in the subsequent cycles, while the hybrid catalyst PMo12@TBA-MSN exhibited higher stability and could be reused, along with ionic liquid, several times. The impregnation of PMo<sup>12</sup> on MSNs proved to be beneficial in enhancing the stability of polyoxomolybdate and in obtaining not only a robust and effective catalyst but also a recyclable material [133].

In comparison to polyoxometalates (POMs), Keggin-type heteropolyacids (HPAs) have shown high catalytic activity for oxidation studies, owing to their high chemical resistance and mechanical robustness, making them suitable for desulfurization reactions as mentioned earlier, when they were combined with meso/macroporous silica [58]. Heteropolyacids have octahedral MO<sup>6</sup> units, and their lowest unoccupied molecular orbitals (LUMO) are mainly the non-bonding metal orbitals, allowing easy participation in redox reactions [134,135]. A heterogeneous catalyst (HPMo-IL/SBA-15) was reported by Xiong et al. via impregnation of phosphomolybdic acid on the IL functionalized SBA-15 and used for the ODS process [136]. The hydrophilicity of SBA-15 was conquered by functionalization with imidazole-based IL, and the heterogeneous HPMo-IL/SBA-15 catalyst exhibited good wettability for the model oil while providing sufficient active sites for the access of organosulfur compounds. The catalytic system involved the use of H2O<sup>2</sup> with a H2O2/sulfur mole ratio of 2, which is much less in comparison to other desulfurization systems, and, in the presence of 0.05HPMo-IL/SBA-15, a S-removal efficiency of 90% could be achieved. High catalytic performance of HPMo-IL/SBA-15 is attributed to its (a) high surface area and good IL dispersion; (b) uniform pore channels allowing efficient mass transfer; and (c) high wettability towards model oil [136].

#### *4.4. Magnetic MS Catalysts*

Catalytic materials synthesized using surface molecular imprinting techniques have shown promise for deep desulfurization reactions. Surface imprinted polymers are beneficial in terms of high guest inclusion rates, reduced mass transport resistance, rapid adsorption kinetics, and feasible template extraction. Additionally, as far as complex fuel samples are concerned, magnetic separation can help to separate impurities and purify the process by adsorption of organosulfur compounds on magnetic catalysts. If molecular imprinted polymers (MIPs) are embedded with magnetic nanoparticles or porous materials doped with magnetic nanoparticles, these adsorbents can be separated in the presence of applied magnetic field [137,138]. Few researchers, including Li et al. and Men et al., have applied this strategy to prepare surface imprinted core-shell magnetic beads and magnetic MIPs for the adsorption of template molecules and their separation from the matrix [139,140]. However, there are only few reports of this strategy used for desulfurization of fuels. With an aim to develop efficient materials for deep-desulfurization of fuels, Wang et al. reported on the synthesis of molecularly imprinted polymer (MIP) coated magnetic mesoporous silica, Fe3O4@mSiO2@DT-MIP, using double-template strategy for desulfurization of DBT and 4-MDBT in model and real gasoline fuels (Figure 8) [141]. The adsorption studies revealed that Fe3O4@mSiO2@DT-MIP followed pseudo-second-order kinetics in the case of DBT and 4-MDBT, with an adsorption amount of 104.2 mg g−<sup>1</sup> for DBT, whereas it showed an amount of 113.6 mg g−<sup>1</sup> towards 4-MDBT, respectively. When the performance of molecular imprinted polymer was compared with non-imprinted polymer (NIP), it revealed that Fe3O4@mSiO2@DT-MIP had higher binding ability and significant recognition capacity for DBT and 4-MDBT, and it could be regenerated and reused at least eight times more than Fe3O4@mSiO2@NIP. When adsorption experiments were carried out on real fuel (92#gasoline), Fe3O4@-mSiO2@DT-MIP catalyst was able to reduce the amounts of DBT and 4-MDBT by at least by 69% [141].

The catalytic activity of materials is of prime importance in biphasic reaction of oil and water and most hydrophilic catalysts can be hindered on interactions with hydrophobic aromatic sulfur compounds. Ionic liquids can come to rescue in these situations by providing active sites and also making the carriers hydrophobic. An ideal ionic liquid has optimum length of carbon chain which does not cause steric hindrance and easy leaching of active species. Inspired by the superhydrophobic lotus leaves, it can be concluded that rough surfaces on carriers can change their wettability, and Zhao et al. demonstrated this by synthesizing core-shell mesoporous silica structures which had cauliflower mimicking morphology [142]. These materials were modified using kinetic-controlled interface co-assembly to produce magnetic mesoporous microspheres (MMSs) which acted as supports for POM-based short chain ionic liquids (([(C4H9)3NCH3]3PMo12O40). The contact angle measurements substantiated the wettability of these materials and their desulfurization capacity, which in turn could be regulated by their surface morphology. The catalyst, ([(C4H9)3NCH3]3PMo12O40/RS-MMS) was used for the ODS of diesel in the presence of H2O<sup>2</sup> as an oxidant and showed highest catalytic activity towards most refractory sulfur compounds.

**Figure 8.** Schematic illustration for the preparation of functionalized magnetic mesoporous silica (Fe3O4@mSiO2@DT-MIP) covered with double template molecular imprinted polymer (MIP) [141].

− A model oil consisting of 4,6-DMDBT as a representative was chosen to investigate the desulfurization efficiency of POM incorporating IL and IL/RS-MMS, and the results revealed that IL showed only 16.3% sulfur removal efficiency, whereas IL/RS-MMS showed 98.3% sulfur removal efficiency at equal time intervals and the same temperatures, which may be attributed to high surface area and better dispersion of active components in IL/RS-MMS catalyst. Sulfur removal followed the order: DBT > 4-MDBT > 4,6-DMDBT, and the effect of ODS reaction temperature for 4,6-DMDBT by IL/RS-MMS revealed an increase in desulfurization efficiency from 41.1% at 40 ◦C to 98.3% at 60 ◦C and a reduction in sulfur concentration in model oil to as low as 3.4 mg kg−<sup>1</sup> . When ODS activity of smooth surface (SS) catalysts were compared to those with superhydrophobic rough surfaces, only the catalysts with suitable amphiphilicity displayed good catalytic activity. The recyclability of ODS of 4,6-DMDBT using a IL/RS-MMS catalyst showed that the catalyst could be recovered and be reused at least five times [143].

There have been reports on using triblock copolymer templating strategy to synthesize 2D-hexagonal mesoporous silica structures under acidic conditions [26]. Recently, this method has also been used to create POM-IL-incorporated mesoporous silica nanocomposites (PMS) for desulfurization of fuels. For example, Zhang et al. used triblock copolymer P123, PMO-based IL ([C16mim]3PMo12O40), and TEOS to synthesize PMS, which was further impregnated with iron oxide nanoparticles using an equivalent-volumetric ultrasonic impregnation method to form magnetic PMS

(MPMS) [26]. The catalysts were used desulfurize and remove multiple refractory sulfur compounds (BT, 3-MBT, DBT, 4-MDBT, 4,6-DMDBT) and showed double-edged influence due to iron oxide, where the catalysts were superparamagnetic and were easily attracted by external magnets, laying the platform for magnetic fixation in the recovery steps. In particular, the 0.5-MPMS recipe exhibited excellent catalytic activity, and the desulfurization performance displayed the following order: DBT > 4-MDBT > 3-MBT > 4,6-DMDBT > BT. Furthermore, using this recipe, a desulfurization efficiency of 94% could be achieved, even after ten cycles, making it, so far, the highest recycled compound and a potential candidate for industrial applications [26].

#### *4.5. Spinel Embedded Silicoaluminophosphate Catalysts*

Spinel phase oxides with the formula AB2O<sup>4</sup> (A and B are metals) are known for their high stability and diverse framework architectures. In particular, Zn-based spinel adsorbents (ZnB2O4) with periodical [B4O4] <sup>4</sup><sup>−</sup> cubical secondary building units have been reported to considerably subdue vaporization of zinc atoms during hot gas desulfurization when temperatures of the reactors were <sup>&</sup>lt;<sup>600</sup> ◦C and demonstrates the synergistic effects of ZnO as B2O<sup>3</sup> units on reaction with H2S in desulfurization steps [144–146]. Additionally, zeolites are high porosity green materials that can prevent sintering of active guest species and have been used in multiple applications, including ion exchange, catalysis, and gas adsorption [147–151]. Silicoaluminophosphates (SAPO) belong to a series of crystalline zeolites, which contain repeating tetrahedral AlO<sup>4</sup> and SiO<sup>4</sup> units creating well-defined channels and microporous cavities and have received recent attention in desulfurization process. For example, this year, Liu et al. reported on the fabrication of SAPO-34 zeolites embedded with Zn-based spinels with the formula ZnBxB'2−xO<sup>4</sup> (B = Co, Mn, Fe; x = 0–2) with an aim for H2S removal from simulated coal gas (Figure 9) [152]. The authors prepared SAPO-34@SBA-15 (SS), SAPO-34@ZSM-5 (SZ), SAPO-34-P123 (SP123), and SAPO-34-PAA (S-PAA) as solid supports for spinel inclusion. SS had a BET surface area of 323 m<sup>2</sup> /g and was used as a representative for multiple examinations. Lattice substitution in B-sites of Zn spinels indicated the stability of ZnCo2/SS structure and efficiency in desulfurization process with a high breakthrough S capacity of 138.08 mg/g when compared to other supports impregnated with Fe or Mn. The best recipe for effective desulfurization was found to be 50 wt% ZnCo2/SS with a reaction proceeded at 550 ◦C. With successive regeneration of catalyst, the breakthrough S capacity decreased from 138.08 to 118.69 mg/g, owing to the evaporation of Zn species, generation of highly stable sulfides (ZnS and Co9S8), and partial sintering of the catalyst [152]. − −

**Figure 9.** Schematic illustration of (**a**) synthesis of silicoaluminophosphates (SAPO)-34@as-synthesized SBA-15 (silica sphere (SS) composite zeolite) using triblock copolymer (P123) templates; (**b**) ZnCo2O<sup>4</sup> structure [152].

#### *4.6. MSN-Carbon Composites*

In addition to mesoporous silica usage, hydrophobic carbonaceous materials can play an important role in desulfurization due to their high adsorption capacity, high specific surface areas and feasible recovery. Moreover, the wide abundance of carbon materials from biomass to rubber tires makes them lost cost adsorbents [153,154]. Combination of catalytic metal sites on carbonaceous materials embedded on mesoporous silica supports can provide a synergistic effect to the materials for adsorption desulfurization (ADS); however, only a few reports exist in literature on this topic. Recently, Liu et al. reported on the synthesis of monodispersed dendritic mesoporous silica/carbon nanospheres (DMS/CNs) impregnated with Ag ions and used them as sorbents for selective desulfurization of DBT (Figure 10). AgNO<sup>3</sup> was included in the catalyst through wet impregnation and thermal dispersion techniques. Adsorption dynamics simulation studies indicated that the adsorption rate increased with an increase in the carbon content of DMS/CNs and with a decrease in the particle size of the catalyst, whereas Ag impregnation improved the efficiency of desulfurization, owing to the sulfur and Ag interactions and formation of S–Ag bonds and π-complexation with thiophene rings present in DBT. When compared to non-impregnated S-100-HC (adsorption capacity, 4.02 mg S/g (no toluene), 2.71 mg S/g (toluene)), Ag/S-100-HC was a promising candidate giving an equilibrium adsorption capacity of 6.88 mg S/g without using toluene and 4.22 mg S/g in the presence of toluene. Its enhanced sulfur selectivity could be attributed to its ordered dendritic mesoporous silica/carbon nature with large center-radial mesoporous channels, providing easy access to active metal sites and a synergistic effect between carbonaceous species and embedded silver ions [155]. π

**Figure 10.** Schematic representation of the synergistic effect of DBT adsorption on dope carbonaceous mesoporous silica supports, Ag/ dendritic mesoporous silica/carbon nanosphere (DMS/CN) structure [155].

#### *4.7. Silica Gels*

High porosity, low cost, and wide availability of silica gels compared to well-ordered MS frameworks have also motivated researchers to explore these materials for desulfurization of fuels. Although few in number, some interesting examples exist in literature where efficiency of silica gel has been explored. For example, Song and co-workers reported on desulfurization of diesel fuel using silica gel impregnated with 5.0 wt% of undisclosed metal species, and a reduction in S content was demonstrated using GC chromatographic analysis by studying the composition of fuel before and after treatment [156]. The authors claimed that JP-8 fuel was used in further studies, and similar results

were obtained. Molecular orbital calculations were performed on sulfur rich species, like thiophene, BT, and DBT, and they revealed the presence of HOMO (highest occupied molecular orbital) on sulfur atoms. These studies indicate direct interactions of HOMO orbitals with the LUMO (lowest unoccupied molecular orbital), belonging to the active species present in silica gel, are responsible for selective elimination of sulfur rich molecules in fuel. Wang et al. reported on Ni-heteropolyacids supported on silica gel and used them for removal of thiophene, alkyl thiophene, BT, and DBT under Ultrasound and Ultraviolet irradiation conditions [157]. Soon after this report, Zheng et al. reported on the entrapment of polyoxometalate precursor ([C16H33(CH3)2NOH]3{PO4[WO(O2)2]4}) in the channels of silica gel and used this heterogenous catalyst to oxidize DBT in the presence of hydrogen peroxide [158]. A few years ago, Xun et al. reported on ionic liquid catalysts ([Bmim]FeCl4) embedded in silica gel, and various conditions were investigated for the removal of DBT using the catalytic oxidative desulfurization process, where the desulfurization efficiency was over 90%, even after recycling the catalyst six times [159]. This study marked a new era for the emergence of ionic liquid supported silica gels and interesting examples were reported by Zhao et al. and Safa et al. The former study reported on using ether functionalized ionic liquid/silica gel catalysts to adsorb SO<sup>2</sup> and the adsorption capacities ranged from 2.621 mol of SO2/mol for [C3O1Mim][H3CSO3] to 3.453 mol of SO2/mol for [C7O3Mim][H3CSO3], indicating an increase in adsorption capacity with elongation in chain length of the cation [160]. The latter study reported on using 1-octyl-3-methylimidazolium hydrogen sulfate ([Omim][HSO4]) in the silica-gel matrix for ODS of a model oil composed of DBT and real diesel fuel. The highest DBT removal efficiency of 99.1% was achieved when 17 wt% of ionic liquid embedded catalyst was used, and 75.7% S-removal efficiency was observed in case of desulfurization of the hydrotreated real diesel fuel [161].

#### **5. MS Frameworks and Nanomaterials in Desulfurization of JP-8 and JP-5 Aviation Fuel**

Fossil fuels are not only used in combustion engines but also as a source of hydrogen for solid-oxide fuel cells (SOFCs) to be used in vehicles, stationary power generators, or military purpose silent watch units [162]. JP-8 jet fuel is a popular fuel for military applications, where it not only functions as a hydrogen source for SOFCs but also to power the aircrafts, and it is used a fuel for heaters, stoves, and tanks. An ideal fuel should be widely available, hydrogen rich, and potentially safe to transport and stock [163]. However, presence of high content of sulfurous compounds in JP-8 fuel poison the anodes of SOFCs and cylinder heads and exhaust valves in supercharged diesel engines. Lifetime of SOFCs can be enhanced with the use of JP-8 fuel with a concentration of <1ppmwS. Deep desulfurization of lighter fuels, such as gasoline, is usually carried out using hydrodesulfurization (HDS) [162,164]. However, several issues associated with HDS, including demanding operating conditions, requiring high temperatures and elevated pressures; non-selective hydrogenation; inefficiency for less reactive aromatic sulfur compounds (DMBT, TMBT); and low octane levels, call for new methods of desulfurization that are low cost and less energy intensive [165].

Mesoporous hierarchical MCM-41 or SBA-15 frameworks embedded with metallic active species and even metal oxides have been investigated in the past for adsorptive desulfurization of JP-5 aviation fuel. MCM-41 and SBA-15 were impregnated with Cu<sup>+</sup> and Pd2<sup>+</sup> ions by heating at 550 ◦C followed by spontaneous dispersion of monolayers. Under such synthetic conditions, the formed CuCl or PdCl<sup>2</sup> coatings get homogenized with supporting frameworks [166]. The performance of Pd2+/MCM-41 was found to be better than Cu+/MCM-41 towards JP-5 as the breakthrough S capacity was 10.9 mg S/g for the former composite, while it was 7.7 mg S/g for the latter composite, and the saturation capacity of Pd2<sup>+</sup> composite was 16.0 mg S/g and 14.4 mg S/g for the Cu<sup>+</sup> composite. High desulfurization performance and better selectivity of Pd2<sup>+</sup> composite can be attributed to its strong π interactions with the aromatic organosulfur compounds than Cu+. Interestingly, breakthrough capacity of Pd2+/SBA-15 (32.1 mg S/g) composites were much higher than the MCM-41 composites (10.9 mg S/g), even when the impregnation had 16% less Pd2<sup>+</sup> in the former composites, and this higher performance can be correlated to the higher porosity and large pore size of SBA-15 compared to MCM-41 [166].

This work was followed by exploration of cuprous oxide impregnated mesoporous silica materials (MCM-41 and SBA-15) for ADS; however, the results contradicted with the previous work and claimed that large pore size and volume of SBA-15 was not beneficial [166]. Here, MCM-41 with a specific surface area of 523 m<sup>2</sup> /g outperformed as an adsorbent in comparison to SBA-15 with a lower surface area (400 m<sup>2</sup> /g) [167]. In these studies, a higher reduction temperature (700 ◦C) was perceived to be better in converting active species to Cu2O, leading to a greater adsorption capacity in case of MCM composite (12.8 mg S/g) compared to SBA-15 composite (9.6 mg S/g). The results also indicate that free transition metal ions exhibit higher performance in terms of saturation capacity and breakthrough capacity when compared to transition metal oxides. The saturation capacity of Cu+/SBA-15 was thrice that of Cu2O/SBA-15, and its breakthrough capacity was four times that of Cu2O/SBA-15. However, metal oxides could be regenerated more successfully in comparison to free metal ions due to their higher stability and strong bonding with surfaces via covalent interactions [167].

With the advent of studies on palladium catalysts, new efforts were exerted towards using precious metals ions like silver for desulfurization, and Ag<sup>+</sup> impregnated mesoporous silicas were achieved through wet impregnation techniques [168]. In this case, the catalytic activity of Ag+/MCM-41 applied on JP-5 fuel was compared with Ag+/SBA-15, and breakthrough capacity of 15.7 mg S/g was achieved for the formed composite, and a value of 10.3 mg S/g was achieved for the latter. Saturation capacity, on the other hand, was 32.1 mg S/g for the MCM composite, whereas it was 29.2 mg S/g for the SBA composite. This data corroborates with the findings presented earlier by Wang et al., proving the high surface area of MCM-41 [166]. It was observed that Ag+/MCM-41 could be thermally recovered in air and could maintain 50% of its initial desulfurization efficiency, even after second and third cycles [168].

This work on silver-containing frameworks acted as a platform for further research on these materials and similar compounds have been evaluated for desulfurization of real, as well as model, fuels. Researchers have reported on the use of Ag-impregnated bulk MCM-41 for JP-5 fuel; however, there is still a huge demand for high edge adsorbents to be applied in advanced JP-8 fuel. A few years ago, Palomino et al. reported on using mesoporous silica nanoparticles as an extension to bulk MCM-41 for ADS of JP-8 aviation fuel with a sulfur concentration of 516 ppmw S [169]. This report was the first of its kind to compare commercial bulk MCM-41 with MCM-41 nanoparticles (MSNs) for JP-8 desulfurization. The prepared nanoparticles were spherical in shape and had an average diameter of 80 nm. Ag-impregnated composite catalytic materials were prepared using MCM-41 and MSN and their adsorption capacities were compared. Adsorption capacity of Ag–MCM-41 (24.5 mg S/g) was found to be lower than Ag–MSN (32.6 mg S/g) [169]. Maximum model fuel capacity was achieved when a silver loading of 18 wt% was used for Ag-MCM-41 and 20 wt% for Ag-MSN. Adsorption capacity and breakthrough S capacity of Ag-MSN (32.6 mg S/g, 0.98 mg S/g) were much higher than that of Ag-MCM-41(25.4 mg S/g, 0.21 mg S/g), owing to the high surface area of nanomaterials. Ag-MSN could also be recovered with diethyl ether and be reused with 70% efficiency [169]. To go in depth on other materials for jet fuels is beyond the scope of this review. For more details on catalytic materials for desulfurization of aviation fuels, we direct our readers to one excellent review provided by Tran et al. [165].

#### **6. Summary**

Key qualities that are widely present among ideal desulfurization materials include: (i) high porosity and large surface areas offering significant area and numerous active sites for adsorption within small volumes; (ii) catalytic materials and their precursors are economical; (iii) porous catalysts are usually embedded with active metal sites (in zerovalent, ionic, or oxide forms) for enriched adsorption; and (iv) finally, they are highly stable and can be recycled without significant loss of catalytic efficiency. These desulfurizing agents have unique surface and structural properties, including surface acidity, magnetic nature, photocatalytic activity, or the ability to homogenously disperse active sites and aiding the adsorption process.

Table 1 provides a summary of various functionalized mesoporous silica frameworks and their desulfurization performance. According to this table, it is clear that: (1) the sulfur adsorption capacity of MS frameworks impregnated with core-shell magnetic nanoparticles is significantly higher than that of MS frameworks embedded with photocatalytic nanoparticles; (2) although complete DBT conversion occurs within short period of time using subnano-MoO3/UMSN structure, LaW10/IL-SiO<sup>2</sup> structure outperforms in terms of time and quantity for both small and large batches; (3) conversion performance of PW11@TMA-SBA-15 is remarkable in comparison to 0.05HPMo-IL/SBA-15, which shows lower desulfurization efficiency for various organosulfur compounds with higher reaction time; (4) the desulfurization performance of the hybrid periodic mesoporous silica (PW11@TMA-PMOE) is lower than that of the hybrid ordered mesoporous silica (PW11@TMA-SBA-15); and (5) a high conversion rate of an ideal catalyst should be accompanied by its high recyclability without significant loss in its efficiency.


**Table 1.** Desulfurization efficiency of different functionalized mesoporous silica materials.

#### **7. Conclusions**

Mesoporous silica frameworks and nanoparticles are the third generation of Si-based nanoporous materials that are promising for clean fuel applications. With increased research interest in the use of specialized MS frameworks for desulfurization of fuels, in this review, we discussed appropriate functionalization strategies vital to adsorb the sulfur rich molecules present in fuels and their oxidation to easily removable species. We examined how uniquely designed MS formulations are capable of encapsulating not only simple sulfur species but also bulky aromatic compounds, like 4,6-DMDBT. We compiled various templating technologies, as well as other methods, like molecular imprinted polymers, to generate catalytic MS frameworks and examined the diverse nature of how surface modification with ionic liquids, photocatalytic active species, magnetic core-shell nanoparticles, and carbonaceous active sites can significantly affect the outcome of sulfur adsorption and removal. Ultra-deep desulfurization of fuels, especially JP-8 type of aviation fuels for fuel cell applications, demand judicious methods of catalysis, where currently precious metal ions are used in the desulfurization process, and stability of porous frameworks is essential to make recovery and reuse of these materials viable. A thorough examination of not only the pore size but also the size of the bulk material, as well as their competitive and interfering interactions in the presence of other compounds in fuels, such as asphaltenes, remains to be explored and can provide insights on the performance of these particles and may provide valuable information related to the criteria of MS material design.

**Author Contributions:** S.M. and A.A.A.A. sketched the outline of the manuscript, collected the scientific materials, reviewed the literature and wrote the manuscript. S.M. drew all the figures in the manuscript. S.M. and A.A.A.A. revised the manuscript for content and scientific quality. Both authors reviewed and approved the submitted manuscript and its revisions. All authors have read and agreed to the published version of the manuscript.

**Acknowledgments:** Authors wish to thank the University of Calgary's Canada First Research Excellence Fund (CFREF) program, the Global Research Initiative for Sustainable Low-Carbon Unconventional Resources, for financial support.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Derivation of Luminescent Mesoporous Silicon Nanocrystals from Biomass Rice Husks by Facile Magnesiothermic Reduction**

**Sankar Sekar 1,2 and Sejoon Lee 1,2, \***


**Abstract:** High-quality silicon (Si) nanocrystals that simultaneously had superior mesoporous and luminescent characteristics were derived from sticky, red, and brown rice husks via the facile and cost-effective magnesiothermic reduction method. The Si nanocrystals were confirmed to comprise an aggregated morphology with spherical nanocrystals (e.g., average sizes of 15–50 nm). Due to the surface functional groups formed at the nanocrystalline Si surfaces, the Si nanocrystals clearly exhibited multiple luminescence peaks in visible-wavelength regions (i.e., blue, green, and yellow light). Among the synthesized Si nanocrystals, additionally, the brown rice husk (BRH)-derived Si nanocrystals showed to have a strong UV absorption and a high porosity (i.e., large specific surface area: 265.6 m2/g, small average pore diameter: 1.91 nm, and large total pore volume: 0.5389 cm3/g). These are indicative of the excellent optical and textural characteristics of the BRHderived Si nanocrystals, compared to previously reported biomass-derived Si nanocrystals. The results suggest that the biomass BRH-derived Si nanocrystals hold great potential as an active source material for optoelectronic devices as well as a highly efficient catalyst or photocatalyst for energy conversion devices.

**Keywords:** biomass rice husk; silicon; nanocrystals; luminescence; high porosity

#### **1. Introduction**

Silicon (Si) is one of the most powerful semiconductors that have led to the strong advancement of modern electronics. However, bulk Si is inadequate as an active material (i.e., a core part for visible light emission or detection) in optoelectronic devices because of its indirect bandgap with an infrared energy gap of 1.12 eV at 300 K [1,2]. One effective way to overcome this issue is the nanocrystallization of Si, which can allow us to create visible light emission and detection characteristics, attributable to the quantum confinement effect and the size effect in Si nanocrystals [3–7]. Therefore, the fabrication of Si nanocrystals has attracted tremendous attention in wide scientific and technologic communities because of their vast application fields in optoelectronics as well as electronics. For instance, nanofloating gate flash memory devices [8–11], field-effect electroluminescence devices [12,13], tandem solar cells [14], and optical waveguides [15,16] are typical examples that can use the quantum-confined electronic energy system in Si nanocrystals. Owing to the high porosity of Si nanocrystals, furthermore, they are also very useful as an active source material in energy storage and conversion devices. For example, Si nanocrystals could be used as an effective catalyst for the hydrogen evolution reaction [17], and be utilized as an anodic source material for highly energy-efficient lithium-ion and sodium-ion batteries [18–20].

To obtain highly porous and/or highly luminescent Si nanocrystals, many researchers have contrived and designed various experimental methods, e.g., laser ablation [21], nonthermal plasma processes [22], pulsed laser deposition [23], chemical doping [24], electrochemical etching [25], chemical vapor deposition [26], annealing of borophosphosilicate

**Citation:** Sekar, S.; Lee, S. Derivation of Luminescent Mesoporous Silicon Nanocrystals from Biomass Rice Husks by Facile Magnesiothermic Reduction. *Nanomaterials* **2021**, *11*, 613. https:// doi.org/10.3390/nano11030613

Academic Editors: Céline Ternon and Sotirios Baskoutas

Received: 31 December 2020 Accepted: 22 February 2021 Published: 1 March 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

glasses [27], and laser pyrolysis [28]. However, these methods require expensive equipment, complex procedures, and high thermal budgets (also see Table S1, Supplementary Materials). Therefore, a facile and cost-effective approach is necessary for the mass production of Si nanoparticles. Considering both the cost-effectiveness and the eco-friendliness, biomass wastes are truly fascinating resources that can provide us with natural siliceous constituents. Accordingly, various biomass resources (e.g., sugarcane bagasse [29], bamboo leaves [29,30], beach sand [31], corn leaves [32], and rice husks (RHs) [33–37]) were used in earlier studies for the derivation of high-quality Si nanocrystals (also see Table S2 for the comparison of Si production from various biomass resources by using several experimental techniques, Supplementary Materials). Among them, RHs are one of the most prominent siliceous precursors because of their huge availability and high silica contents [18,38,39]. These provide us with a good hint to produce a large amount of Si nanocrystals via the recycling of biomass RHs. Despite such benefits, to our best knowledge, the synthesis of RH-derived high-quality Si nanocrystals with both high porosity and high luminescence has not been reported to date. Furthermore, the coexistence of both mesoporous and luminescent characteristics in a single material system is truly helpful for future energy technology; for example, the photocatalytic hydrogen evolution reaction [40–42] and oxygen evolution reaction [43–45].

We, therefore, investigated the facile derivation of mesoporous-and-luminescent Si nanocrystals from various RHs (i.e., sticky RHs, red RHs, and brown RHs) through the magnesiothermic reduction process, which can be simply performed in an inert atmosphere without toxic gases and vacuum facilities. Herein, we report on a comprehensive study from the synthesis to the characterization of RH-derived mesoporous luminescent Si nanocrystals. The kinetics of magnesiothermic reduction for Si nanocrystal production is discussed, and the structural, morphological, optical, and textural properties of the synthesized Si nanocrystals are thoroughly examined in detail.

#### **2. Experimental Section**

#### *2.1. Material Preparation*

The biomass sticky rice husks (S-RHs) were collected from Gyeonggi, South Korea, and the red rice husks (R-RHs) and brown rice husks (B-RHs) were collected from Perambalur, Tamil Nadu, India. Hydrochloric acid (HCl, 37%), hydrofluoric acid (HF, 48%), and magnesium (Mg, 99% purity) powders were purchased from Sigma-Aldrich (St. Louis, MO, USA) and used with no additional purification.

#### *2.2. Synthesis of Si Nanocrystals*

The derivation of the Si nanocrystals via magnesiothermic reduction can be described by the following chemical reactions:

$$\text{SiO}\_2(\text{Ashes}) + \text{HCl} \underset{27 \stackrel{\circ}{\text{C}} \text{(2 h)}}{\text{SiO}\_2(\text{2 h})} \text{SiO}\_2(\text{Colloidal}) + \text{H}\_2\text{O} + \text{MCl} \tag{1}$$

$$\text{SiO}\_2(\text{Colloidal}) \underset{700 \text{ } ^\circ \text{C (2 h)}}{\text{SiO}\_2(\text{Nanopparticles})} \text{(2)}$$

$$2\text{ SiO}\_2(\text{Naonparticles}) + 2\text{Mg}^{-} \underset{700^{\circ}\text{C}\ (2\text{ h})}{\rightarrow} \text{Si}(\text{Naonocycles}) + 2\text{MgO} \tag{3}$$

$$\text{Si(Nanocrystals)} + 2\text{MgO} + 4\text{HCl} \underset{27\,^\circ \text{C} \,(10\,\text{h})}{\text{Si} \,(\text{Nanocrystals})} + 2\text{MgCl}\_2 + 2\text{H}\_2\text{O} \quad \text{(4)}$$

M in Equation (1) is the possible precipitates from raw bio-silica in the biomass RHs (e.g., Na, K, Ca, Fe, and Mg), which are normally removed as MCl after the HCl treatment. To investigate the dependence of the biomass RH resources, we used three different types of RHs, i.e., S-RHs, R-RHs, and B-RHs. As schematically illustrated in Figure 1, initially, all three types of RHs were carbonized at 500 ◦C for 2 h under an air atmosphere to obtain their ashes. Then, 3 g of each RH ash was stirred in a 10% HCl solution for 2 h to eliminate metal ions and contamination (e.g., Equation (1)). After HCl leaching, the

samples were rinsed with deionized water (DI), filtered, and dried at 150 ◦C for 15 h in an electric oven. Subsequently, the samples were transferred to an alumina crucible and were further calcinated at 700 ◦C for 2 h under an air atmosphere in a muffle furnace (e.g., Equation (2)). During this calcination step, the silica nanopowders were obtained from the HCl-leached RH ashes. Next, the SiO<sup>2</sup> nanopowders were reduced into the Si nanocrystals through magnesiothermic reduction. To achieve the reduction reaction, as a primary task, each type of SiO<sup>2</sup> nanopowder (2 g) was mixed with the Mg powders (0.5 g). Then, the mixture powders were annealed at 700 ◦C for 2 h under an Ar atmosphere in a tube furnace (e.g., Equation (3)). The obtained products were stimulated with 1 M HCl (HCl:H2O:EtOH = 0.66:4.72:8.88 molar ratio) for 10 h to remove MgO (e.g., Equation (4)). After the HCl treatment, the colloidal solutions were reacted with 5% HF for 1 h to eliminate the residual SiO<sup>2</sup> inside the magnesiothermically reduced Si nanopowders. Finally, the obtained Si nanopowders were washed in DI water, filtered, and dried at 80 ◦C for 12 h under vacuum. Through these sequences, we were able to obtain the powder type of the Si nanocrystals. For convenience, we denote the three different types of the Si nanocrystals as S-Si, R-Si, and B-Si, which were derived from S-RHs, R-RHs, and B-RHs, respectively.

**Figure 1.** Schematic illustration of the magnesiothermic reduction process for synthesizing the Si nanocrystals by using biomass sticky rice husks (S-RHs), red rice husks (R-RHs), and brown rice husks (B-RHs).

#### *2.3. Characterization of Material Properties*

– The morphological and the compositional properties of the Si nanocrystals were monitored by field-emission scanning electron microscopy (FE-SEM) using an Inspect F50 system (FEI Co., Mahwah, NJ, USA) and its in situ energy dispersive X-ray (EDX) spectroscopy, respectively. The structural and the vibrational properties of the samples were characterized by Raman scattering spectroscopy using a LabRAM HR800 system (HORIBA Jobin Yvon Inc., Edison, NJ, USA) and X-ray diffractometry (XRD) using a D2 Phaser system (Bruker, Madison, WI, USA), respectively. The functional groups of the nanocrystals were examined by Fourier transform infrared (FTIR) spectroscopy using a Spectrum-100 system (Perkin Elmer, Shelton, CT, USA). The optical absorption and emission characteristics were evaluated by UV–VIS spectroscopy using an S-3100 system (Scinco, Seoul, Republic of Korea) and photoluminescence (PL) spectroscopy using a Cary Eclipse Fluorescence Spectrophotometer (Agilent Technologies, Santa Clara, CA, USA), respectively. The textural properties were analyzed by nitrogen absorption–desorption isotherms (N2-ADI) using a BELSORP-mini II system (MicrotracBEL, Osaka, Japan).

#### **3. Results and Discussion**

#### *3.1. Morphological and Compositional Properties*

Figure 2 shows the FE-SEM images of the S-Si, R-Si, and B-Si nanocrystals. The S-Si sample exhibited cylindrically interconnected spherical nanocrystals (Figure 2a). However, the R-Si and the B-Si samples displayed a nanosponge-like morphology, where a lot of small spherical nanocrystals were densely aggregated (Figure 2b,c).

–

**Figure 2.** Field-emission scanning electron microscopy (FE-SEM) images of the (**a**) S-Si, (**b**) R-Si, and (**c**) B-Si nanocrystals and energy dispersive X-ray (EDX) spectra of the (**d**) S-Si, (**e**) R-Si, and (**f**) B-Si nanocrystals. The inset in each EDX graph summarizes the compositional properties of the Si nanoparticles.

(,)

– – Here, it should be noticed that the average crystal size of B-Si (~15 nm) is much smaller than those of R-Si (~35 nm) and S-Si (~50 nm). We believe such a discrepancy is attributable to the smaller contents of Si species in the raw sources of the B-RH ashes (Si~2.46%) than the R-RH (Si~5.95%) and S-RH ashes (Si~24.62%) (see Figure S1, Supplementary Materials). In other words, during the acid treatment and the calcination step (i.e., Equations (1) and (2)), the size of the colloidal SiO<sup>2</sup> should be smaller for the B-RH case than the others because the lower quantity of Si species in B-RH (i.e., raw bio-silica in the biomass resource) may increase the segregation of the silica nanoparticles [46]. According to previous literature [47,48], using the Fokker–Planck equation [49–51], the size distribution of the Si nanoparticles (*C*(*i,t*)) can be described as follows:

(()(,))

$$\frac{\partial \mathcal{C}(i, t)}{\partial t} = -(N(t) - N\_\mathbf{e}) \frac{\partial (k(i)\mathcal{C}(i, t))}{\partial i} + (N(t) + N\_\mathbf{e}) \frac{\partial^2 (k(i)\mathcal{C}(i, t))}{\partial^2 i^2} \,, \tag{5}$$

where *i* is the number of Si atoms, *N*e is the equilibrium concentration of impurity atoms in the substance matrix, and *N*(*t*) is the number of Si atoms in the nanocrystal at each moment of time. The kinetic coefficient *k*(*i*) in Equation (5) is proportional to both the diffusion coefficient of the Si atoms in the matrix *D* and the Si nanocrystal radius (*R*(*i*)):

$$k(i) = 4\pi DR(i)\tag{6}$$

<sup>2</sup>(()(,))

$$R(i) = b(i+m)^{\alpha},\tag{7}$$

where *b* is the distance parameter of the nanocrystals, *m* is the size homogeneity factor, and *α* is the geometry factor (= 1/3 for spherical nanocrystals). Hence, the smaller size of B-Si can be interpreted as resulting from the lower concentration of Si species in B-RH. We therefore conjecture that the size of the Si nanocrystals could be automatically controlled by choosing the type of the biomass raw resources. *α*

() = 4π()

Next, the compositional properties of the samples were evaluated by EDX. As shown in Figure 2d–f, all the prepared samples were composed of the main species of Si and O, arising from the body and the surface of the nanocrystals, respectively. The additional component of Pt is thought of as sprouting from the conductive coating layer for the FE-SEM measurements. –

#### *3.2. Structural and Vibrational Properties*

The crystallographic properties of the S-Si, R-Si, and B-Si samples were characterized by XRD. As shown in Figure 3a, all of the three samples exhibited the typical diffraction patterns of crystalline Si at 28.4 ◦ , 47.4 ◦ , 56.1 ◦ , 69.1 ◦ , 76.4 ◦ , and 88.2 ◦ , which correspond to the (111), (220), (311), (400), (331), and (422) Si planes (JCPDS No. 27-1402 [18,33,52,53]), respectively. This means that all three, S-Si, R-Si, and B-Si, were well crystallized via magnesiothermic reduction from the biomass resources of the S-RH, R-RH, and B-RH ashes, respectively. By using the Scherer formula [54–56], the average crystallite sizes of the S-Si, R-Si, and B-Si nanocrystals were determined to be 33, 28, and 22 nm, respectively. This corroborates the dependence of the Si nanocrystal size on the kind of biomass raw source; i.e., the Si nanocrystal size relies on the different Si contents in each RH, as confirmed in Figure 2. The nanocrystallization of the samples was further elucidated by Raman spectroscopy measurements. As shown in Figure 3b, the Raman spectra of all three samples revealed a similar feature of the typical Raman vibration modes from crystalline Si. Namely, the sharp peak at 519 cm−<sup>1</sup> (i.e., the first-order transversal optical (TO) mode [18,57,58]) and the broad hump at 957 cm−<sup>1</sup> (i.e., the second-order TO mode [18,57,58]) are clearly observable in all the samples, while no other Raman bands are visible. This demonstrates that the high-purity Si nanocrystals were effectively derived from the biomass RHs through the magnesiothermic reduction process. – −1 −1

**Figure 3.** (**a**) X-ray diffractometry (XRD) patterns, (**b**) Raman spectra, and (**c**) Fourier transform infrared (FTIR) spectra of the S-Si, R-Si, and B-Si nanocrystals.

For the nanostructured materials, the functional groups of the elemental species and molecular states depend on the shape and the size of the nanomaterials because they rely on the bonding states at the surface terminals. To examine the functional groups of the samples, FTIR measurements were carried out. As shown in Figure 3c, the samples displayed several FTIR features at 795, 869, 956, 1070, 1377, 1643, 2884, 2977, and 3648 cm−<sup>1</sup> , all of which are closely relevant to the Si nanostructure. In other words, the transmission band at 795 cm−<sup>1</sup> is ascribed to the Si–C stretching mode [36], and the vibrational band at

869 cm−<sup>1</sup> is attributed to the Si–N stretching mode [59,60]. Similarly, the band at 956 cm−<sup>1</sup> arose from the Si–H bending mode [61]. Additionally, the bands at 1070, 1377, 2884, and 2977 cm−<sup>1</sup> correspond to the Si–O bending, CH<sup>3</sup> bending, symmetric CH<sup>2</sup> vibration, and CH<sup>3</sup> stretching modes, respectively [59,62,63]. The bands at 3648 and 1643 cm−<sup>1</sup> are responsible for the Si–OH vibrations [64]. <sup>−</sup> – − – <sup>−</sup> – − –

−

#### *3.3. Optical Properties*

Figure 4a shows the UV−VIS absorption spectra of the S-Si, R-Si, and B-Si samples. The B-Si nanocrystals revealed strong UV adsorption, while the S-Si and the R-Si nanocrystals exhibited predominant visible-light adsorption characteristics. For B-Si, particularly, two distinct Si nanocrystal-related absorption bands are observable at *A*1~270 nm and *A*2~340 nm. Namely, the *A*<sup>1</sup> peak and the *A*<sup>2</sup> shoulder are associated with the *L*−*L* and the Γ−Γ transitions in nanocrystalline Si, respectively [6,36]. Since both the *A*<sup>1</sup> and *A*<sup>2</sup> adsorption intensities indicate the degree of nanocrystallization [65], one can observe that the B-Si sample was well crystallized with a smaller size than the others, as confirmed by FE-SEM and XRD. Figure 4a shows the UV−VIS absorption spectra of the S − *Γ*−*Γ*

<sup>−</sup> –

**Figure 4.** (**a**) UV-VIS absorption spectra, (**b**) photoluminescence (PL) spectra, and (**c**) light emission mechanism of the rice husk (RH)-derived biomass Si nanocrystals.

– – – – According to earlier literature [3–5], those transitions originate from the discrete energy states within the modified electronic band structure of the Si nanocrystal. In other words, when the nanocrystal size becomes smaller than the exciton Bohr radius (~4 nm for Si), the subbands above the conduction band and below the valence band could be altered because of the quantum confinement effect. Then, discrete energy states would be created inside the modified electronic band structure. Furthermore, since the Si nanocrystal surface is typically terminated by H, C, and O (also see the FTIR for our samples in Figure 3c), the overlap of electron and hole wave-functions would become significant [65]. These subband modulation effects can be elucidated by PL. As shown in Figure 4b, the samples emitted visible light at *P*1~485 nm, *P*2~530 nm, *P*3~545 nm, and *P*4~573 nm. The strong *P*<sup>1</sup> emission is reported to emerge from the radiative optical transitions between the discrete energy states that are created at the Si-OH surface functional groups [66] (also see Figure 4c). The other peaks at *P*2, *P*3, and *P*<sup>4</sup> are also well known to arise from the radiative optical transitions between the energy states that are created by the surface functional groups of Si–O [67], Si–C [5], and Si–H [68,69], respectively. Figure 5 shows the excitation-dependent PL spectra of the S-Si, R-Si, and B-Si samples. As the excitation wavelength (*λ*ex) increased, the peak position of the light emission (*λ*emit ) tended to shift to the longer wavelength region (i.e., red shift). Such a *λ*emit dependence on *λ*ex was present in all the samples, and the *λ*emit positions were almost identical, regardless of the raw source of the RHs. These findings depict that the PL emission in all the samples originated from the surface functional group-related subband modulation rather than the quantum confinement effect.

*λ*

*λ*

*λ λ*

*λ λ*

*λ λ*

*λ λ*

**Figure 5.** Excitation-dependent PL spectra of the (**a**) S-Si, (**b**) R-Si, and (**c**) B-Si nanocrystals.

#### *3.4. Textural Properties*

– – – – The nanocrystallization of Si would have sturdily affected the porosity of the entire material system because the locally crystallized small nanocrystal has a high surface-tovolume ratio. In short, the nanocrystals must form structural voids at the surface area, giving rise to an increase in the porosity of the material. To assess the porosity of the S-Si, R-Si, and B-Si samples, thus, the textural characteristics were evaluated by the Brunauer– Emmett–Teller (BET) and the Barrett–Joyner–Halenda (BJH) analysis methods. Firstly, the specific surface area (*S*ss) was determined by N2-ADI measurements. As shown in Figure 6, all the samples exhibited Type-IV isotherm curves (classified according to IUPAC), representing the distinctive mesoporous characteristics of the materials [33,70]. Through the BET analysis, the *S*ss values of the S-Si, R-Si, and B-Si nanocrystals were calculated to be 168.1, 212.4, and 265.6 m2/g, respectively. – – – –

– **Figure 6.** Nitrogen absorption–desorption isotherm (N – <sup>2</sup> -ADI) characteristics of the (**a**) S-Si, (**b**) R-Si, and (**c**) B-Si nanocrystals.

Next, the pore distribution characteristics were examined by BJH measurements (Figure 7). The pore surface areas (*S*ps) were determined to be 149.3, 196.9, and 218.5 m2/g for the S-Si, R-Si, and B-Si nanocrystals, respectively, and the total pore volumes (*V*tp) were calculated to be 0.4103, 0.4201, and 0.5389 cm3/g for S-Si, R-Si, and B-Si, respectively. Compared to S-Si and R-Si, the B-Si sample had larger magnitudes of *S*ps and *V*tp because of both the smaller nanocrystal size and the uniform distribution. Accordingly, the average pore diameter of B-Si (*d*ap~4.91 nm) was also smaller than those of S-Si (*d*ap~9.76 nm) and R-Si (*d*ap~7.82 nm).

**Figure 7.** Pore size characteristics of the (**a**) S-Si, (**b**) R-Si, and (**c**) B-Si nanocrystals.

Finally, we compared the textural and the optical characteristics of our samples with various biomass-derived Si nanocrystals previously reported. As can be confirmed from Tables 1 and 2, the B-Si nanocrystals possessed a superior porosity compared to the others, and had a great potential for multiple light emissions. Based on all the above results, therefore, one can surmise that the magnesiothermically reduced biomass B-Si nanocrystals hold great promise in various applications such as energy storage/conversion devices and optoelectronic devices.

**Table 1.** Comparison of the luminescence characteristics for various biomass-derived Si nanostructures.


**Table 2.** Comparison of the pore characteristics for various biomass-derived Si nanostructures.


#### **4. Conclusions**

High-quality Si nanocrystals that simultaneously showed both strong luminescence and high porosity were successfully derived from various RHs through the facile magnesiothermic reduction method. Owing to the different quantities of raw bio-silica in each RH, the size of the Si nanoparticles could be automatically varied from 15 to 50 nm. Due to the existence of the surface functional groups at the nanocrystals, the samples showed multiple light emissions in the visible-wavelength regions (i.e., blue, green, and yellow). Among the prepared samples, the B-Si nanocrystals exhibited a higher UV absorption and

a superior porosity. The results depict that the B-RH-derived Si nanocrystals can play a crucial role as high-performance electrocatalysts, photocatalysts, and light emitters.

**Supplementary Materials:** The Supplementary Materials are available online at https://www.mdpi. com/2079-4991/11/3/613/s1: Chemical Composition of Rice Husk Ashes; Comparison of Various Methods for Silicon Production; Figure S1. EDX spectra of (a) S-RH, (b) R-RH, and (c) B-RH ashes. Note that Pt in each raw source material arose from the conductive coating of Pt for better focusing and imaging during SEM and EDX measurements, Table S1. Summary of silicon synthesized from various resources through several experimental methods, Table S2. Summary of silicon synthesized from various biomass resources through several experimental methods.

**Author Contributions:** Investigation and writing—original draft, S.S.; conceptualization, supervision, funding acquisition, and writing—review and editing, S.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was supported by the National Research Foundation (NRF) of Korea through the Basic Science Research Programs (2016R1A6A1A03012877 and 2019R1A2C1085448) funded by the Korean Government.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Article* **Ultrabright Fluorescent Silica Nanoparticles for Dual pH and Temperature Measurements**

**Saquib Ahmed M. A. Peerzade 1 , Nadezhda Makarova <sup>2</sup> and Igor Sokolov 1,2,3, \***


**Abstract:** The mesoporous nature of silica nanoparticles provides a novel platform for the development of ultrabright fluorescent particles, which have organic molecular fluorescent dyes physically encapsulated inside the silica pores. The close proximity of the dye molecules, which is possible without fluorescence quenching, gives an advantage of building sensors using FRET coupling between the encapsulated dye molecules. Here we present the use of this approach to demonstrate the assembly of ultrabright fluorescent ratiometric sensors capable of simultaneous acidity (pH) and temperature measurements. FRET pairs of the temperature-responsive, pH-sensitive and reference dyes are physically encapsulated inside the silica matrix of ~50 nm particles. We demonstrate that the particles can be used to measure both the temperature in the biologically relevant range (20 to 50 ◦C) and pH within 4 to 7 range with the error (mean absolute deviation) of 0.54 ◦C and 0.09, respectively. Stability of the sensor is demonstrated. The sensitivity of the sensor ranges within 0.2–3% ◦C −1 for the measurements of temperature and 2–6% pH−<sup>1</sup> for acidity.

**Keywords:** pH sensor; temperature sensor; dual sensor

#### **1. Introduction**

Knowledge of temperature and acidity at the nanoscale is of interest from both applied and fundamental points of view. For example, understanding the distribution of those parameters inside of the biological cell is key to understanding complex biochemical processes that are occurring in a highly heterogeneous environment of the cell. Cellular functions like gene and protein expression and protein stability are strongly temperaturedependent [1]. It is also known that cell migration, cell proliferation, wound healing [2], protein denaturation, protein folding and protein stability [3] are strongly pH-dependent. From a fundamental point of view, physical and chemical processes in the nanoscale are still not well understood in general because temperature and acidity were not measured at that scale.

Using fluorescence as an indication of pH and temperature is an attractive option because it can be accessed remotely and in three dimensions across a volume of interest. There were attempts to make a complex fluorescent molecule that would be sensitive to the change in both pH and temperature [4]. However, the fluorescence of this molecule was excessively sensitive to the changes in the ionic composition of the medium. This effectively prohibits the direct use of this molecule as a sensor. Encapsulation of sensitive molecules inside nanoparticles can decrease the dependence of the fluorescence from the medium surrounding nanoparticles, and thus, make it a sensor.

There were multiple attempts to measure both pH and temperature sequentially. For example, it was done to study biological cells, specifically, the investigation of mitochondrial acidification due to real-time monitoring of mitochondrial pH using fluorescent

**Citation:** Peerzade, S.A.M.A.; Makarova, N.; Sokolov, I. Ultrabright Fluorescent Silica Nanoparticles for Dual pH and Temperature Measurements. *Nanomaterials* **2021**, *11*, 1524. https://doi.org/10.3390/ nano11061524

Academic Editor: Céline Ternon

Received: 17 May 2021 Accepted: 5 June 2021 Published: 9 June 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

probes [5]. In this study, for measuring intracellular pH, the fluorescent probe was used that shows pH response based on a single peak [5]. For making measurements ratiometric, cells were post-stained with mitotracker. Measurements of pH were done by using a ratio of mitotracker and molecular probe. It was assumed that mitotracker and fluorescent probe both are homogeneously distributed. Here the concentration of mitotracker should be known and kept constant with calibration and cell measurements. The cell density on the glass-bottomed dishes for both calibration and temperature distribution measurements should be constant. Moreover, two wavelengths were used for exciting mitotracker and molecular probe. Local heat production by mitochondria inside the living cells has also been detected before using intracellular fluorescent temperature mapping [6]. Temperature mapping was performed using fluorescence lifetime imaging microscopy (FLIM), which is typically a time-consuming imaging.

Fluorescent nanoparticles changing their fluorescence spectrum as a function of acidity have previously been reported [7]. FITC, a pH-sensitive dye, and rhodamine B was used as a reference. Although rhodamine B shows a temperature dependent fluorescence, it was not discussed, and only pH measurements were demonstrated. In another study, two dyes FITC and Rb were used for measuring both pH and temperature simultaneously [8]. However, only relative changes in the pH and temperature were measured in this approach. In addition, the reported particles were of micron size, not nano. As a result, the particles have a rather limited range of applications. The fluorescent nanosensors capable of simultaneous measurements of actual accurate values of the pH and temperature have not been reported (see more comparison to the state-of-the-art in the Discussion section).

Here we report on the synthesis of fluorescent nanoparticles that can serve as nanoscopic sensors capable of measuring the values of both temperature and acidity simultaneously. The sensors are based on the encapsulation of three fluorescent dyes within the mesoporous silica matrix of nanoparticles. As was shown previously [9–11], such encapsulation results in obtaining ultrabright fluorescence of nanoparticles. The phenomenon of ultrabrightness was attributed to preserving the quantum yield of the encapsulated dyes while attaining a very high concentration of the dye inside the particles. The close proximity of the encapsulated dye molecules allows for the Förster's resonance energy transfer (FRET) between the neighbor dye molecules. This allows us to build effective ultrabright fluorescent ratiometric sensors, which are excited with just one excitation wavelength. The possibility to use the FRET approach to build ultrabright fluorescent nanosensors of temperature has been recently demonstrated [12]. Our fluorescent nanosensors are ratiometric to avoid dependence on the intensity of the excitation light. This is important because it is practically impossible to control the intensity of the excitation light in an optically inhomogeneous medium. It should be noted that we previously demonstrated encapsulation of multiple organic dyes in the reported nanoparticles [13] for the purpose of multiplexed detection. To the best of the authors' knowledge, the present work is the first report on ultrabright mesoporous silica nanoparticles for sensing both pH and temperature simultaneously.

In the presentation here, nanosensors were allowed to find the temperature and pH by using two fluorescent ratios. It should be noted that such an approach was reported for a silica-based pH and oxygen sensor [14]. Both fluorescent ratios reported in that work were independent of each other. In our approach, we do observe crosstalk between the two ratios used. To define both temperature and pH independently, we developed an algorithm that allows for decoupling the crosstalk, and as a result, minimizes the error in the definition of temperature and pH. Our algorithm allows for finding the particular wavelengths of the fluorescence spectrum to maximize the signal to noise ratio of the spectral signal. We demonstrated that using our approach, one can attain quite small errors (mean absolute deviation, or the average of residuals at different pH and temperature) of 0.54 ◦C and 0.09 in the definition of temperature and pH, respectively. The sensitivity of the sensor ranges within 0.2–3% for ◦C and 2–6% for pH−<sup>1</sup> . This is comparable with ranges reported in the literature [15].

#### **2. Materials and Methods**

#### *2.1. Materials*

Tetraethylorthosilicate (TEOS, ≥99%, GC, Acros Organics, Fair Lawn, NJ, USA), triethanolamine (TEA, reagent grade 98%, Sigma Aldrich, St. Louis, MO, USA), cetyltrimethylammonium bromide (CTAB, High Purity Grade, Amresco, Solon, OH, USA), ethyltriethoxysilane (ETES, 96%, Frontier Scientific, Logan, UT, USA), fluorescein isothiocyanate (FITC, Exciton, Dayton, OH, USA), rhodamine B (Rb, Exciton, Dayton, OH, USA) and Nile blue perchlorate (Nb, 95%, Sigma Aldrich) were used. RC membrane (RC membrane, Spectra/Pore, Rancho Dominguez, CA, USA) with 10–15 kDa MW was used. Deionized water was used in all synthesis.

#### *2.2. Synthesis of Nanosensors*

The previously reported procedure [13,16] was modified to assemble the presented pH and temperature sensor based on mesoporous silica particles. The molar ratio of chemicals in the synthesizing bath was 1 TEOS: 8.2 TEA: 0.23 CTAB: 142 H2O: 0.1 ETES. The ratio of FITC: Rb: Nb was 1:4:55. The mixture of TEOS (1.71 g, 8.2 mmol) and TEA (10 g, 67 mmol) was stirred for one minute and kept at 90 ◦C under quiescent conditions for 20 min. Another mixture of CTAB (0.69 g, 1.9 mmol), FITC (0.001 g, 0.0026 mmol), Rb (0.005 g, 0.013 mmol) and Nb (0.05 g, 0.14 mmol) and H2O (21 mL) was stirred for 1 min in 21 mL water and kept at 60 ◦C for 40 min. The CTAB, dye and water mixture were stirred at room temperature for another 15 min and kept in a cold bath for 5 min. After 5 min the mixture of TEOS and TEA was then added to the aqueous solution of CTAB and dye. ETES (130 uL, 0.8 mmol) was added after 10–15 min and stirred for another 40 min in the cold bath. After 40 min, the synthesis mixture was diluted with 30 mL water and the excess reagents were removed by dialyzing with water using the membrane of MW 10–15 kDa until no fluorescence was obtained from the dialysate (several (2–3) days). The pH of the mixture after dialysis was ~9. HCl was added to neutralize the mixture.

#### *2.3. Characterization Techniques*

Dynamic light scattering (DLS): DLS was used to measure the particle size and zeta potential of the nanoparticles using Zetasizer Nano ZS by Malvern Instruments Ltd., Malvern, UK. DLS uses the laser light of 633 nm and the backscattered light is monitored at an angle of 173◦ . The intensity-average size (Z-average) and most probable size (mean of number weighted distribution) was the average of three measurements.

Optical measurements: Cary 60 UV–Vis spectrometer (Agilent, Santa Clara, CA, USA) was used to measure the absorbance with an averaging time of 0.1 s and a scan rate of 600 nm/min. Fluorescence was measured using a Horiba Fluorelog 3 (Horiba, Japan) using a 2 nm slit width with an integration time of 0.1 s and a scan rate of 600 nm/min.

AFM imaging of sensors: An Icon Atomic Force Microscopy (AFM, Bruker, Inc., Santa Barbara, CA, USA) with NanoScope V controller with a ringing mode add-on (NanoScience Solutions, Inc., Arlington, VA, USA) was used to image the nanoparticles.

#### **3. Results**

#### *3.1. Characterization of the Nanosensors*

The DLS measurements of the particle size are shown in Figure 1a. The most probable particle size is ~51 nm. This is confirmed by direct imaging of the particles with AFM, Figure 1b. From the material point of view, the synthesized particles are identical to those previously reported by us in [11,13,16,17]. As was shown there, the internal structure of the particles is not changed when dyes are added, while the overall size of the particles can change. Thus, one can check the internal structure (measured by TEM and N<sup>2</sup> absorbance) of these particles in [11,13,16,17].

**Figure 1.** (**a**) Dynamic Light Scattering (DLS) measurements of the particle size distribution and (**b**) Atomic Force Microscopy (AFM) of nanosensors.

Three fluorescent dyes were encapsulated inside each nanosensor. Nile blue (Nb) was used as the reference, FITC was used for pH sensing, and rhodamine B (Rb) for temperature sensing. The spectral characteristics of synthesized nanosensors are shown in Figure 2 together with the spectral characteristics of individually encapsulated dyes. Figure 2 shows the (A) absorbance and (B) fluorescence spectra (excited at 488 nm) of the nanosensors. The dotted black line represents the fitting after demultiplexing (the addition of the individual dye components). Normalized absorbance and fluorescence spectra of individual FITC, Rb and Nb dyes are shown in Figure 2C.

**Figure 2.** Spectral characteristics of nanosensors. (**A**) Normalized absorbance and (**B**) normalized fluorescence spectra of nanosensors. Individual component spectra of dyes used for fitting are shown. (**C**) Normalized absorbance/emission spectra of FITC, Rb and Nb dyes.

The fluorescence brightness of the synthesized nanosensors was measured as described in the Equations (S15) and (S18), see reference [18] for detail. Because of the complex fluorescence composition, the brightness is calculated with respect to each of the three encapsulated dyes. The results are presented in Table 1, which shows the fluorescence characteristics of the synthesized particles, including the fluorescent brightness and quantum yield. It should be noted that the definition of the quantum yield of a fluorescent nanoparticle is somewhat ambiguous. Here we use the definition in which the quantum yield is assigned per single encapsulated dye molecule, see references [9,10] for detail. Brightness is given in the relative MESF (Molecules of Equivalent Soluble Fluorochrome) and absolute M−<sup>1</sup> cm−<sup>1</sup> units. MESF units relative to the corresponding dye are broadly used in flow cytometry and the characterization of particle brightness [19–24]. The absolute units are used to compare the brightness with the particles with any other fluorophore.

**Table 1.** Fluorescence properties of ultrabright pH and temperature-sensing nanoparticles. Relative fluorescence brightness of ultrabright mesoporous silica nanoparticles in both MESF units (relative to one free dye molecule) and M−<sup>1</sup> cm−<sup>1</sup> , and the number of dye molecules per particle. Note that all the calculations were performed using the number weighted average particle size of 51 nm.


One can see that particles showed the brightness of ~150 and 390 relative to the free FITC and Rb dye molecules, respectively, when exciting with 490 and 550 nm. Brightness relative to Nile blue is the highest (~2000 in MESF units) when particles are excited with 630 nm wavelength. The number of FITC, Rb and Nb molecules encapsulated per particle is 180 ± 10, 390 ± 30 and 2500 ± 200, respectively. FRET efficiency and distance between dye molecules are given in Table S4 [25,26].

To find the absolute fluorescent brightness, the brightness in MESF units was multiplied by the absolute brightness of corresponding fluorescent molecules. The values of the extinction coefficient of FITC dye at 490 nm, Rb dye at 550 nm and Nb dye at 630 nm wavelength were taken from the literature, 70,000 M−<sup>1</sup> cm−<sup>1</sup> [27], ~100,000 M−<sup>1</sup> cm−<sup>1</sup> [28] and ~76,000 M−<sup>1</sup> cm−<sup>1</sup> [29,30], respectively. The quantum yields of free FITC, Rb and Nb dyes were used as reported in the literature 0.93 [31], 0.31 [32] and 0.27 [29], respectively. The quantum yield of the encapsulated dye was calculated in the Supplementary Information (S16) and (S17). Thus, the absolute brightness of nanosensors was (10 ± 1) × 10<sup>6</sup> , (12 ± 1) × 10<sup>6</sup> and (40 ± 3) × 10<sup>6</sup> M−<sup>1</sup> cm−<sup>1</sup> , respectively.

The number of Nile blue dye molecules is the highest, 2500 ± 200 Nile blue dye molecules per particle. This corresponds to a 60 mM concentration of Nb dye molecules inside the particles. Thus, it corresponds to the ratio of the number of the dye molecules inside the particles FITC: Rb: Nb 1:2:14. It is interesting to compare the ratio of dyes used for synthesis, FITC: Rb: Nb 1:5:55. The difference is presumably due to the different solubility of the dyes inside the silica matrix.

When considering fluorescence sensing, it is important to verify that the fluorescence does not depend on the other parameters of the environment. In the case of the most probable use of the sensors, biomedical applications, the environment can contain complex ions, which potentially can influence the fluorescence of the encapsulated dye. Figure 3 shows the ratio of fluorescent intensities, which are used for the determination of temperature and pH (see later), in the presence of various ions typical for physiological buffers. For this purpose, the selectivity measurements were done in the presence of monovalent (K<sup>+</sup> and Na<sup>+</sup> ) and divalent (Ca2+ and Mg2+) ions. The stability of the ratios for determining pH and temperature in the presence of ions is shown in Figure 3. Ratios I(525 nm)/I(537 nm) and

I(581 nm)/I(611 nm) were used for calculating temperature and pH, as shown in Figure S3. The ions K<sup>+</sup> , Ca2+, Mg2+ and Na<sup>+</sup> were added consecutively. It can be seen that the ratios stayed virtually constant after the addition of the ions. The standard deviations across all the different addition of ions were ±0.2 ◦C and ±0.03 for temperature and pH, respectively.

**Figure 3.** Stability of the ratios for nanosensors toward metal ions (1 mM) K<sup>+</sup> , Ca2+, Mg2+ and Na<sup>+</sup> .

#### *3.2. Temperature and pH Calibration*

Figure 4 demonstrates the changes in the fluorescence spectral behavior of the nanosensors to the change of pH in the range of 4–7 and the temperature in the range of 25–45 ◦C. FITC dye is known to be sensitive to both temperature and pH while rhodamine B dye is sensitive to only temperature [8,33]. One can see that the rhodamine B peak (~570 nm) decreases with the increase in temperature, whereas the FITC peak at 515 nm decreases with the decrease of pH (increase in acidity) and increased with increase in temperature at pH 6.8 and 6.3 (Figure 4a,b). This is consistent with the reported response of FITC and rhodamine B dyes to the change of pH and temperature, respectively.

It should be noted that the presence of FRET can be seen through a relatively high fluorescence of rhodamine B and Nile blue dyes. The excitation wavelength was 488 nm. The absorbance of rhodamine B and Nile blue dyes at this wavelength is much lower compared to the absorbance of FITC. In the present configuration, FITC plays the role of the donor dye. The change of the fluorescence spectra shown in Figure 4 are due to the change of fluorescent properties of the dyes as described above, and presumably not due to the change in FRET efficiency. The latter might happen due to the change of distances between the dye molecules because of the temperature expansion of the silica matrix. However, a simple estimation shows that this effect is negligible.

The calibration of the sensors was done for a number of fixed conditions: five different temperatures (24.17, 29.63, 34.74, 39.77 and 44.8 ◦C) and six different acidity (pH 6.8, 6.3, 5.8, 5.3, 4.8 and 4.3). The fluorescence spectra were recorded for each of these 30 conditions. These calibration points were used to fit the polynomial functions of the algorithm as described in the next sections.

**Figure 4.** Temperature and pH responses of nanosensors to the change in temperature (shown for 25, 30, 35, 40 and 45 ◦C and the change in pH: (**a**) 6.8, (**b**) 6.3, (**c**) 5.8, (**d**) 5.3, (**e**) 4.8 and (**f**) 4.3, respectively. The excitation wavelength is 488 nm.

#### *3.3. Method of Simultaneous Measurements of Temperature and Acidity*

As one can see in Figure 4, the combination of encapsulated dyes shows a rather nontrivial dependence on both temperature and pH. There is substantial crosstalk between the responses to the change of temperature and pH. Here we present an algorithm, which allows us to decouple this complex behavior of the fluorescent spectra to define the temperature and acidity separately and with relatively high precision. We demonstrate that it is sufficient to have two ratios of fluorescence intensities at different wavelengths to determine the temperature and acidity quite accurately. To do it, first, we will find the optimal values of the wavelengths to be used to calculate the intensity ratios. Then we suggest simple polynomial functions to determine the temperature and acidity using those ratios.

#### *3.4. Finding the Optimal Values of Wavelengths to Measure Temperature and Acidity Simultaneously*

To define both pH and temperature in a ratiometric manner, two different fluorescence ratios corresponding to two different wavelengths should be chosen. Our approach is based

on Equation (1) minimization of the ambiguity of finding both pH and temperature (due to existing crosstalk between responses to the change in pH and temperature) and Equation (2) minimization of the error of the measurements (because the fluorescence intensities have different signal-to-noise ratio at different wavelengths). To minimize the complexity of such an analysis, it is preferable to have a linear relationship between the sought ratios and pH and temperature. To find the linear relations, regression analysis was used [34]. The Pearson correlation coefficient r indicates the proportion of variance in variable Y represented by a linear function of X. This coefficient is given by the following formula:

$$\mathbf{r}^2 = \frac{\mathbf{S}\_{\mathbf{xy}}^2}{\mathbf{S}\_{\mathbf{x}}\mathbf{S}\_{\mathbf{y}}} = \frac{\left(\sum\_{i=1}^n (\mathbf{X}\_{\mathbf{i}} - \overline{\mathbf{X}})(\mathbf{Y}\_{\mathbf{i}} - \overline{\mathbf{Y}})\right)^2}{\sum\_{i=1}^n \left(\mathbf{X}\_{\mathbf{i}} - \overline{\mathbf{X}}\right)^2 \sum\_{i=1}^n \left(\mathbf{Y}\_{\mathbf{i}} - \overline{\mathbf{Y}}\right)^2},\tag{1}$$

where, Sxy is the covariance of X and Y, S<sup>x</sup> is the standard deviation of X, S<sup>y</sup> is the standard deviation of Y. The linear relation implies this coefficient to be close to 1.

Combining the standard deviation of pH and the ratio R and covariance of pH and R, as given in Equations (S2), (S3) and (S5), one can find the Pearson correlation coefficient for pH as follows:

$$\mathbf{r}\_{\rm pH}^{2} = \frac{\mathbf{S}\_{\rm pH,R}^{2}}{\mathbf{S}\_{\rm pH}\mathbf{S}\_{\rm R}} \mathbf{r}\_{\rm pH}^{2} = \frac{\mathbf{S}\_{\rm pH,R}^{2}}{\mathbf{S}\_{\rm pH}\mathbf{S}\_{\rm R}} = \frac{\left(\sum\_{\rm pH\_{1}}^{\rm pH\_{n}} \left(\left(\rm pH - \rm pH\_{\rm mean}\right) \left(\sum\_{\rm T\_{1}}^{\rm T\_{n}} \left(\frac{R}{n\_{\rm T}}\right) - \overline{R}\right)\right)\right)^{2}}{\sum\_{\rm pH\_{1}}^{\rm pH\_{n}} \left(\rm pH - \rm pH\_{\rm mean}\right)^{2} \sum\_{\rm pH\_{1}}^{\rm pH\_{n}} \left(\sum\_{\rm T\_{1}}^{\rm T\_{n}} \left(\frac{R}{n\_{\rm T}}\right) - \overline{R}\right)^{2}},\tag{2}$$

Similarly, combining the standard deviation of T, R and covariance of T and R given by Equations (S6), (S7) and (S9), one can find the Pearson correlation coefficient for temperature T given by the following formula:

$$\mathbf{r}\_{\mathbf{T}}^{2} = \frac{\mathbf{S}\_{\mathbf{T},\mathbf{R}}^{2}}{\mathbf{S}\_{\mathbf{T}}\mathbf{S}\_{\mathbf{R}}} = \frac{\left(\sum\_{\mathbf{T}\_{1}}^{\mathrm{T}\_{\mathrm{n}}} \left(\left(\mathbf{T} - \mathrm{T}\_{\mathrm{mean}}\right) \left(\sum\_{\mathrm{pH}\_{1}}^{\mathrm{pH}\_{\mathrm{n}}} \left(\frac{\mathrm{R}}{\mathrm{n}\_{\mathrm{pH}}}\right) - \overline{\mathbf{R}}\right)\right)\right)^{2}}{\sum\_{\mathbf{T}\_{1}}^{\mathrm{T}\_{\mathrm{n}}} \left(\mathbf{T} - \mathrm{T}\_{\mathrm{mean}}\right)^{2} \sum\_{\mathbf{T}\_{1}}^{\mathrm{T}\_{\mathrm{n}}} \left(\sum\_{\mathrm{pH}\_{1}}^{\mathrm{pH}\_{\mathrm{n}}} \left(\frac{\mathrm{R}}{\mathrm{n}\_{\mathrm{pH}}}\right) - \overline{\mathbf{R}}\right)^{2}}\,\tag{3}$$

Figure 5 shows the plots of the values of the Pearson coefficients for pH and temperature calculated using the fluorescence spectra shown in Figure 4. Using Equations (2) and (3), the Pearson coefficients are calculated for different wavelengths used to calculate the ratio of fluorescent intensities. One can see that Figure 5a,b shows linearity (the Pearson coefficients close to 1) for both pH and temperature for a wavelength range of 500–550 nm. The range of 550–650 nm is also highly linear with respect to the change of temperature with the Pearson coefficient >0.996 (Figure 5b). Comparing Figure 5a,b, one can see that the ratio dependencies are more linear for temperature (r 2 T > 0.996) compared to pH (r 2 pH > 0.95). This is in agreement with previously reported FITC-based pH sensors, which are known to be nonlinear with respect to the change in pH [35]. The regions of wavelengths, in which the dependence of temperature and pH on the corresponding fluorescense ratios is quite linear are considered further to choose the wavelengths for the ratio to obtain the minimum error of measurements.

Although we see the linear relation between the parameters, pH and temperature, and the intensity ratios, it is instructive to add a few nonlinear polynomial terms because the Pearson coefficient is not exactly 1, which indicates it may be small, but still a deviation from linearity. In addition, it will help us to verify that the nonlinear terms are small. Hereafter, we suggest the Equations (4)–(9) comprising the polynomial of the third order and a linear crosstalk product of two ratios:

$$\text{T(R}\text{-}\text{R}\_{\text{pH}}\text{)}=\text{At}\times\left(\text{R}\_{\text{pH}}\right)^{3}+\text{Bt}\times\left(\text{R}\_{\text{pH}}\right)^{2}+\text{Ct}\times\text{R}\_{\text{pH}}+\text{Dt}\times\text{R}\_{\text{T}}\times\text{R}\_{\text{pH}}+\text{Et}\,,\tag{4}$$

$$\text{pH} \left(\text{R}\_{\text{T}}, \text{R}\_{\text{pH}}\right) = \text{ApH} \times \left(\text{R}\_{\text{pH}}\right)^{3} + \text{BpH} \times \left(\text{R}\_{\text{pH}}\right)^{2} + \text{CpH} \times \text{R}\_{\text{pH}} + \text{DpH} \times \text{R}\_{\text{T}} \times \text{R}\_{\text{pH}} + \text{EpH},\tag{5}$$

r୮ୌ ଶ

where At, Bt, Ct, Dt Et, ApH, BpH, CpH, Dt and Et are constants, RpH and R<sup>T</sup> are ratios of fluorescence intensities of the chosen reflexes (i.e., RpH = I(λ1)/I(λ2) and R<sup>T</sup> = I(λ3)/I(λ4)).

r ଶ

**Figure 5.** The Pearson correlation coefficients are shown for different wavelengths used to calculate the ratio of fluorescent intensities. The dependence between the model parameters (temperature and pH) and the corresponding ratios is close to linear when the Pearson coefficient is close to 1. The results are shown for pH (**a**) and temperature (**b**) measurements.

Now we can focus on maximizing the signal-to-noise ratio of the fluorescence signal that is used to find temperature and pH, i.e., the ratio of the fluorescence intensities. It is useful to simultaneously maximize the sensitivity to the change of temperature and pH. Therefore, we suggest maximizing the signal-to-noise ratio of the following signal:

$$\mathbf{S(A,B)} = \left| \frac{\left( \begin{pmatrix} \mathbf{I\_1} \\ \mathbf{I\_2} \end{pmatrix}\_{\mathbf{T}\mathbf{1}} - \begin{pmatrix} \mathbf{I\_1} \\ \mathbf{I\_2} \end{pmatrix}\_{\mathbf{T}\mathbf{2}} \right)}{\left| \begin{pmatrix} \frac{\begin{pmatrix} \mathbf{I\_1} \\ \mathbf{I\_2} \end{pmatrix}\_{\mathbf{T}\mathbf{1}} + \begin{pmatrix} \mathbf{I\_1} \\ \mathbf{I\_2} \end{pmatrix}\_{\mathbf{T}\mathbf{2}} \end{pmatrix} \right|} \right| = \left| \begin{pmatrix} \begin{pmatrix} \mathbf{A - B} \\ \mathbf{A - B} \end{pmatrix} \end{pmatrix} \right|, \tag{6}$$

R୮ୌ + EpH λ λ λ λ where I<sup>1</sup> and I<sup>2</sup> are the fluorescence intensities at wavelength 1 and 2, respectively, A and B are the ratio of intensity at temperatures T<sup>1</sup> and T2, respectively.

According to the formula of error propagation, the error in signal S (A,B) is given by

$$\delta \mathbf{S(A,B)} = \sqrt{(\partial\_{\mathbf{A}} \mathbf{S(A,B)})^2 (\delta \mathbf{A})^2 + (\partial\_{\mathbf{B}} \mathbf{S(A,B)})^2 (\delta \mathbf{B})^2} \tag{7}$$

$$=\sqrt{\left(\left(\partial\_{\mathbf{A}}\left(\frac{\mathbf{A}-\mathbf{B}}{\left(\frac{\mathbf{A}+\mathbf{B}}{2}\right)}\right)\right)^{2}(\delta\mathbf{A})^{2}+\left(\partial\_{\mathbf{B}}\left(\frac{\mathbf{A}-\mathbf{B}}{\left(\frac{\mathbf{A}+\mathbf{B}}{2}\right)}\right)\right)^{2}(\delta\mathbf{B})^{2}}=4\sqrt{\frac{\mathbf{B}^{2}(\delta\mathbf{A})^{2}+\mathbf{A}^{2}(\delta\mathbf{B})^{2}}{\left(\mathbf{A}+\mathbf{B}\right)^{4}}},\tag{8}$$

where,

$$\text{\ $A(I\_1, I\_2) = A\sqrt{\left(\frac{\delta I\_1}{I\_1}\right)^2 + \left(\frac{\delta I\_2}{I\_2}\right)^2}, \text{\$ B(I\_1, I\_2) = B\sqrt{\left(\frac{\delta I\_1}{I\_1}\right)^2 + \left(\frac{\delta I\_2}{I\_2}\right)^2} \tag{9}$$

Similar formulas can be derived for the signal defining pH.

Pearson correlation coefficient (PCC) and signal-to-noise ratios (SNR), S(A, B)/δS(A, B) calculated for the fluorescent nanosensors are shown in Figure 6 and Figure S2. One can see still a sufficiently broad range of wavelengths to choose from when PCC is close to 1.

The freedom of choice of the optimal wavelengths can further be reduced by finding the regions of the wavelengths in which the Pearson correlation coefficient is sufficiently close to 1 and the signal-to-noise ratio is sufficiently high. In addition, to reduce the ambiguity and cross-correlation between pH and temperature, one ratio should depend on only one parameter and should be independent of the other parameter. For finding the two ratios, where one ratio depends on only one parameter and is independent of the other, the region of interest was identified such that the ratio of two wavelengths have high linearity (or dependence) to pH and low linearity (or independence) for temperature, and similarly, the region of interest for temperature was identified such that the second ratio of other two wavelengths has high linearity (or dependence) to temperature and low linearity (or less dependence) to pH. Specifically, we find two wavelengths for pH such that the PCC of pH was taken greater than 0.7 while that of temperature was taken less than 0.5 (Figure 6c). Figure 6e shows the SNR of pH for the region of interest where PCC of pH is greater than 0.7 while PCC of temperature is less than 0.5. Similarly, for finding two wavelengths for temperature, for PCC of temperature were set such that the PCC of temperature was taken greater than 0.95 while that of pH was taken less than 0.5 (Figure 6d), and Figure 6f shows the SNR of temperature for the region of interest where PCC of temperature is greater than 0.95 while PCC of pH is less than 0.5. The ratio of intensities at wavelength 582 to 614 nm showed high SNR and PCC for temperature while the ratio of intensities at wavelength 523 to 536 nm showed high SNR and PCC for pH. Hence R<sup>T</sup> and RpH were taken as the ratio of intensities at wavelength 582 to 614 nm and 523 to 536 nm, respectively.

#### *3.5. Algorithm for Further Finding the Optimal Values of Wavelength*

In principle, the accuracy of the measuring temperature and acidity could further be improved using the algorithm shown in Figure 7. According to the algorithm, for the wavelengths in the range of λ1 − n to λ1 + n, λ2 − n to λ2 + n, λ3 − n to λ3 + n and λ4 − n to λ4 + n, the first four wavelengths (i.e., λ1 − n, λ2 − n, λ3 − n and λ4 − n) were chosen and fitted to Equation (4) and values of temperature were determined. The residual (the absolute difference between the actual values of temperature and the values measured using our nanosensors) was calculated for 30 data points (i.e., five data points for different temperatures and six data points for different pH) and the maximum of all the residuals were calculated. If the maximum residual was less than 1.2 ◦C for temperature, the four wavelengths and their residual were stored and displayed, and if it was greater then, one of the wavelengths was iterated and the loop continued until all the range of the wavelengths were iterated (i.e., until λ1 + n, λ2 + n, λ3 + n and λ4 + n). In this way, the four wavelengths that show a maximum residual of less than 1.2 ◦C were identified. For finding temperature, since the wavelengths determined using PCC and SNR were 523, 536, 582 and 614 nm for λ1, λ2, λ3 and λ4, respectively, the range of the wavelengths used was 519–527 nm for λ1, 532–540 nm for λ2, 578–586 nm for λ3 and 610–618 nm for λ4 (considering n = 4). The range was four wavelengths below and above the wavelengths determined using regression and SNR analysis. Since the temperature error should be less than 1.2 ◦C, the expected residual in temperature was chosen to be less than 1.2 ◦C. Out of all the wavelengths, the wavelengths 525, 537, 581 and 611 nm showed the lowest maximum residual of ~1.1 ◦C for temperature. Using the same four wavelengths, the maximum residual for pH was determined to be equal to 0.3.

Table S1 shows the ratios at different pH and temperature. These ratios were further used for multiparametric fitting. From Figure S1, it can be seen that the ratios change repeatably when pH is constant and the temperature is changed and also vice versa. This makes the pH and temperature sensor repeatable.

**Figure 6.** The Pearson correlation coefficient and signal-to-noise ratios for different wavelengths. The Pearson correlation coefficients for pH (**a**) and temperature (**b**) measurements. Pearson correlation coefficient for (**c**) pH when PCC of pH is greater than 0.7 while PCC of temperature is less than 0.5 and for (**d**) T when PCC of temperature is greater than 0.95 while PCC of pH is less than 0.5. SNR for (**e**) pH when PCC of pH is greater than 0.7 while PCC of temperature is less than 0.5 and for (**f**) T when PCC of temperature is greater than 0.95 while PCC of pH is less than 0.5.

λ λ λ λ **Figure 7.** Flowchart outlining the procedure for obtaining four wavelengths for temperature. Note that the fitting for temperature was performed using Equation (4) and RpH and R<sup>T</sup> represent I(λ1)/I(λ2) and I(λ3)/I(λ4), respectively; X represents 1.2 ◦C for temperature.

#### *3.6. Multiparametric Fit for pH and Temperature*

Using the wavelengths that were found, we fit the experimental data (shown in Figure 4 for a set of different temperatures and pH) with the help of Equations (4) and (5). Table 2 shows the fitted coefficients used in Equations (4) and (5). One can see that the coefficients corresponding to the nonlinear terms are indeed rather small. Both the results of the fitting and the experimental data are shown in Figure 8. One can see a rather good fit.

**Table 2.** Fitted coefficients used in Equations (4) and (5) when using the optimal wavelengths of 525, 537, 581 and 611 nm.


− −

− −

−

**Figure 8.** The results of the multiparametric fitting of experimental data (shown by dots) using Equations (4) and (5) to measure measuring temperature and pH, respectively. The measurements of the (**a**) temperature and (**b**) pH as a function of two ratios, R<sup>T</sup> and RpH.

To estimate the errors of the measurements of temperature and pH using the functions of fitted Equations (4) and (5), we conducted four independent measurements of fluorescence at temperature 24.17, 29.63, 34.74, 39.77 and 44.8 ◦C and pH 6.8, 6.3, 5.8, 5.3, 4.8 and 4.3. Measuring multiple spectra and using the error propagation (see the Supplementary Information), we find the errors of measurements of both temperature and pH. The examples are shown in Table S2 (assuming fixed pH) and Table S3 (assuming fixed temperature). Corresponding ratios R<sup>T</sup> and RpH are shown in the Supplementary Information.

Table S1 shows the ratios that are calculated using the intensities at wavelengths 525 to 537 nm, and 581 to 611 nm wavelengths for pH and temperature, respectively. The accuracy determined using this formula was 98.4% and 98.36% for pH and temperature, respectively.

#### **4. Discussion**

Here we discuss the results obtained in comparison with the current state-of-the-art. Because the achievement reported is based on multiple pillars, we describe each of them separately below.

#### *4.1. Use of Mesoporous Silica Versus Solid Silica as the Material of the Nanothermometers*

Gold–FITC encapsulation of dye molecules in solid silica matrix has been previously reported. For example, silica nanoparticles had physically encapsulated rhodamine B dye while the silica shell was modified with APTES followed by conjugation with FITC [7]. This post-conjugation of silica particles with APTES and further attachment with FITC is not only an additional and time-consuming process, but it also reduces the quantum yield of the encapsulated dye molecules. The particles obtained served as a pH sensor due to the presence of pH-sensitive FITC dye, whereas rhodamine B served as a reference [7]. Although rhodamine B is temperature-sensitive, no temperature measurements were reported for that work because there was no third reference dye involved (a third dye is needed to make the sensor ratiometric). In another paper, a ratiometric pH sensor using FITC and Nile blue was reported [36]. Nile blue was used as a reference. FITC and Nb were conjugated on the surface of carbon dots. However, ratiometric pH and dual temperature sensors based on two ratios (three dyes) have not been made before.

The use of mesoporous silica particles as hosts for dye molecules creates a novel possibility to create ratiometric fluorescent sensors coupling two different dyes with FRET [12,13]. For extending the application of sensors to understand cellular processes, it is important

that the sensors can operate at the nanoscale. Physical encapsulation of dyes in mesoporous silica could be used for encapsulating multiple dyes that are sensitive to different parameters (for example, pH and temperature in this case) in a single nanoparticle. In this work, sensor-based on ultrabright mesoporous silica nanoparticles is synthesized incorporating three dyes. Herein, FITC and rhodamine B were used for pH and temperature sensing, respectively, while Nile blue was used as a reference. This is the first demonstration of pH and temperature nanosensors based on the physical encapsulation of three dyes. The particles reported here are free of the issues mentioned above and can be used as sensors.

#### *4.2. The Problem of Ambiguity of Simultaneous Measurement of Temperature and Acidity*

Gold–FITC based nanoclusters have been reported for pH and temperature sensing [37]. This dual sensor was built using two fluorescence peaks coming from gold and FITC. However, for unambiguous determination of both pH and temperature, the magnitude of the intensity ratio or the initial values of pH and temperature should be known. In other work, a pH and temperature sensor was made using a co-assembly of temperaturesensitive polymer PNIPAM, pH-sensitive polymer PNIPAM-co-PAA, photoluminescence sources europium and quaternary ammonium tetraphenyl ethylene derivatives [38]. However, simultaneous temperature and pH measurements were not conducted due to the ambiguity of cross-correlation between the fluorescence dependence of the temperature and pH. The temperature measurements were performed at a constant pH 7 only. As we demonstrated in the present work, we use two different fluorescence ratios to avoid the need of the above ad hoc knowledge and solve the problem of cross-correlation.

#### *4.3. Multiparametric Sensor*

The other pillar used in this work is the use of multiparametric algorithms, in particular, the use of polynomial sensing Equations (4) and (5). Ideally, in a dual sensor, the two parameters should be independent of each other. Linear multiparametric sensors, a polynomial of the order of one, have been studied in [39]. Nonlinear polynomial fit for calibrating pH sensor is reported [35]. However, to the best of our knowledge, a nonlinear polynomial fit for sensing both pH and temperature has not been previously reported. In addition, the multiparametric sensor, in which the two parameters are related to each other, has not been previously reported. In this study, a nonlinear multiparametric sensor in which the two parameters are cross-correlated but avoid ambiguity in measuring both parameters (temperature and pH) is reported.

#### *4.4. Stability of the Sensor*

Because of a potential concern of leakage of the encapsulated dye and photobleaching, it is important to demonstrate that the sensor is stable and the readings are repeatable. The repeatability of the described sensors is shown in Figure S1. Fluorescent spectra from nanosensors were measured at pH 7 for temperatures 35 and 50 ◦C and, similarly, at pH 4 for temperatures 35 and 50 ◦C for four cycles of temperature. The errors defined as the standard deviation across four measurement cycles are shown in Figure S1. One can see that the error to signal ratio is less than 1%, Table S5.

Spectral stability with respect to the presence of different ions in the solution is shown in Figure 3 and Figure S3. Measurements were done in the presence of monovalent (K<sup>+</sup> and Na<sup>+</sup> ) and divalent (Ca2+ and Mg2+) ions. The choice of these particular ions was dictated by the immediate application of the reported sensors to measure the temperature and pH in biological tissues and cells. These ions are the most abundant in biological buffers. It should be noted that the applicability of sensors to other environments has to be tested for each specific environment. Because the number of cases in which the sensors described can be used is practically unlimited, the study of spectral stability in other buffers is beyond the scope of this work. Furthermore, if this problem appears, it could potentially be addressed by additional nonporous silica coating.

#### **5. Conclusions**

In the present work, we describe a dual acidity (pH) and temperature sensor based on ultrabright fluorescent nanoparticles. Simultaneous, multiparametric and nonlinear pH and temperature sensing were successfully demonstrated using FITC (a pH-sensitive dye), rhodamine B (a temperature-sensitive dye) and Nile blue (a reference dye) physically mixed in mesoporous silica particle.

To measure both pH and temperature simultaneously, using two ratios of fluorescent intensities measured at four wavelengths was suggested. A regression analysis was used to find regions of four wavelengths that are (almost) linear with pH and temperature. This allowed us to find the wavelengths to optimize the signal-to-noise ratio and reduce cross-correlation between acidity and temperature dependencies, which in turn decreased the error of the measurements when using the sensors. The regression analysis and algorithm described may further be used in multiparametric systems or sensors in which the parameters are cross-correlated and there is a need to reduce the cross-correlation. The analysis of the discrepancy between the equations and experimentally measured pH and temperature showed that the method suggested can be used to measure pH and temperature with an accuracy of greater than 98%. The repeatability of measuring pH and temperature was demonstrated by multiple cycling of temperature and acidity. Finally, the stability of the sensor in the presence of different ions typical for biological media was demonstrated. Due to a high brightness, sensing with ultrabright nanoparticles could be done at a low dosage and low intensity of the excitation light. When applied to biomedical detection, this will decrease not only the chemotoxicity but also the phototoxicity.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/ 10.3390/nano11061524/s1, Figure S1: Repeatability of measurements at (a and b) constant pH and different temperature and at (c and d) constant temperature and different pH. Figure S2: (a) PCC of pH when PCC of pH is greater than 0.7 while PCC of temperature is less than 0.5, PCC of temperature when PCC of temperature is greater than 0.95 while PCC of temperature is less than 0.5 and (b) SNR of pH when PCC of pH is greater than 0.7 while PCC of temperature is less than 0.5, (c) PCC and (d) SNR of temperature when PCC of temperature is greater than 0.95 while PCC of pH is less than 0.5. Figure S3: Temperature and pH response of nanosensors towards metal ions (1 mM) K<sup>+</sup> , Ca2+ , Mg2+ and Na<sup>+</sup> . Table S1: Ratios of fluorescence intensity at wavelengths of 525 and 537 nm (RpH) and 581 and 611 nm (RT) for measuring pH and temperature. Error represents the standard deviation of four measurements. Table S2: Temperature and standard deviation calculated using Equation (4) (considering pH fixed). Both temperature and pH in the table headings are measured using standard tools (a thermocouple and pH meter). Table S3: pH and standard deviation calculated using Equation (5) (considering temperature fixed). Table S4: FRET efficiency and distance between different dye molecules. Both temperature and pH in the table headings are measured using standard tools (a thermocouple and pH meter). Table S5: Repeatability of the sensor ratios.

**Author Contributions:** Conceptualization, I.S. and S.A.M.A.P.; methodology, S.A.M.A.P.; synthesis, S.A.M.A.P.; data analysis, S.A.M.A.P.; AFM measurements, N.M.; writing—original draft preparation, S.A.M.A.P.; writing—review and editing, I.S.; supervision, I.S.; funding acquisition, I.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** The study was funded by NSF grants CBET 1605405 and CMMI 1937373 (I.S.).

**Acknowledgments:** Authors would like to thank Kyle Monahan for help with Python code.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


## *Article* **Effect of Size and Shape on Electrochemical Performance of Nano-Silicon-Based Lithium Battery**

**Caroline Keller 1,2 , Antoine Desrues 3 , Saravanan Karuppiah 1,2 , Eléa Martin 1 , John P. Alper 2,3 , Florent Boismain 3 , Claire Villevieille 1 , Nathalie Herlin-Boime 3 , Cédric Haon <sup>2</sup> and Pascale Chenevier 1, \***


**Abstract:** Silicon is a promising material for high-energy anode materials for the next generation of lithium-ion batteries. The gain in specific capacity depends highly on the quality of the Si dispersion and on the size and shape of the nano-silicon. The aim of this study is to investigate the impact of the size/shape of Si on the electrochemical performance of conventional Li-ion batteries. The scalable synthesis processes of both nanoparticles and nanowires in the 10–100 nm size range are discussed. In cycling lithium batteries, the initial specific capacity is significantly higher for nanoparticles than for nanowires. We demonstrate a linear correlation of the first Coulombic efficiency with the specific area of the Si materials. In long-term cycling tests, the electrochemical performance of the nanoparticles fades faster due to an increased internal resistance, whereas the smallest nanowires show an impressive cycling stability. Finally, the reversibility of the electrochemical processes is found to be highly dependent on the size/shape of the Si particles and its impact on lithiation depth, formation of crystalline Li15Si<sup>4</sup> in cycling, and Li transport pathways.

**Keywords:** silicon nanoparticles; silicon nanowires; synthesis; high energy density; lithium-ion batteries; high-capacity anode; VLS; laser pyrolysis; size effect; shape effect

#### **1. Introduction**

High energy density in lithium-ion batteries (LiB) requires active materials with enhanced lithium absorption capacity and a large potential gap. On the anode side, silicon appears as the most promising material to enhance the specific capacity for several reasons [1]. Si is an earth-abundant, low-cost, and non-toxic element, with a mature industrial knowledge base from its application in electronics, optics, and photovoltaics. Furthermore, its low potential, close to Li/Li + , makes it suitable for composites with graphite, the current commercial anode material [2–4]. Finally, Si shows a high Li-alloying capacity but, consequently, an undesired high volume change in cycling [5]. This deleterious volume change stimulated the development of Si nanomaterials, because Si at the nano scale can withstand swelling without fracture during LiB cycling. This has been demonstrated for a variety of morphologies including Si nanoparticles (SiNP) [6–11], Si nanowires (SiNW) [3,12–14], Si nanotubes [15,16], and Si porous nanomaterials [17–19]. However, nanostructuration can be a costly process that needs to be carefully optimized for a dedicated application.

The size of the nano-Si and its impact on lithiation mechanisms has been the subject of numerous papers. Several in situ TEM investigations have shown that fracture of crystalline SiNWs [20,21] and SiNPs [21,22] occurs above a critical diameter of ≈150 nm. A critical diameter was also predicted by a mechanical numerical model [23] at 90 nm for SiNPs

**Citation:** Keller, C.; Desrues, A.; Karuppiah, S.; Martin, E.; Alper, J.P.; Boismain, F.; Villevieille, C.; Herlin-Boime, N.; Haon, C.; Chenevier, P. Effect of Size and Shape on Electrochemical Performance of Nano-Silicon-Based Lithium Battery. *Nanomaterials* **2021**, *11*, 307. http:// doi.org/10.3390/nano11020307

Academic Editor: Jie Wang Received: 23 December 2020 Accepted: 21 January 2021 Published: 25 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

and 70 nm for SiNWs. By contrast, several electrochemical studies in LiB investigated the impact of particle size on Si-based anode cycling for small [6,8] and large [7–9] SiNPs, with the best results at diameters of 20–50 nm. As for SiNWs, some studies have suggested the optimal diameter as being around 30 nm [10,24]. These size limits in batteries look much lower than the safe nano-silicon size inferred from calculations or TEM observations. However, these electrochemical studies use different sources of Si with heterogeneous size distributions and within a narrow range of average sizes, making the optimum estimation quite challenging. Additionally, to date, no study has compared different shapes of nanosilicon cycled in the same conditions to enable direct comparison of 0D and 1D structures.

SiNPs and SiNWs can be obtained with a very high control on diameter from welldescribed synthesis methods, such as colloidal nanocrystal synthesis [25] and chemical vapor deposition (CVD) growth [26,27]. However, these highly controlled growth processes produce insufficient (~1 mg) quantities of material for assembling a lithium-ion battery (requiring >100 mg for typical coin cell electrodes). Typical scalable processes developed for industrial synthesis of SiNPs [28,29] and SiNWs [9,30,31] produce nanomaterial either with a very wide size distribution or with a limited ability to change the diameter. The synthesis of a series of nano-silicon materials of fine-tuned size in a large scale is, thus, a challenge for materials chemists.

In the present paper, we present optimized, scalable processes to obtain large batches of SiNPs and SiNWs allowing a comparison of Si nanomaterials in shape (spherical or cylindrical) and size. All materials were tested in half-cell Li-ion batteries in the same conditions to investigate the impact of shape/size on the electrochemical performance and reaction mechanisms.

#### **2. Materials and Methods**

Chemicals: Calcium carbonate (CaCO<sup>3</sup> nanopowder, 50 nm), sodium citrate, ascorbic acid, and carboxymethylcellulose sodium (CMC) were purchased from Sigma Aldrich. Gold tetrachloroauric acid (HAuCl4), tetraoctylammonium bromide (TOAB), dodecanethiol, and sodium chloride were purchased from Alfa Aesar. Carbon black Super-P (CB) was purchased from Imerys. Diphenylsilane was purchased from Chemical Point. Silane (high purity, 98%) was purchased from Messer.

Silicon nanoparticle synthesis: The silicon nanoparticles (NPs) were synthesized by laser pyrolysis as already described [28,32]. Briefly, a flow of gaseous silane intersects with the beam of a high power CO<sup>2</sup> laser. In the interaction zone, silane absorbs the laser radiation. Dissociation of the precursor followed by nucleation and growth of Si nanoparticles occurs with appearance of a flame. Key parameters to control the nanoparticle size are the residence time in the laser beam, the nucleation, and the flame temperatures. The main control parameters are the gas flow rates, the dilution of active gases, and the laser power.

Gold nanoparticle synthesis for silicon nanowire growth: 1–2 nm gold nanoparticles (AuNP) are synthetized by the Brust method [33]. Briefly, HAuCl4·xH2O dissolved in water is transferred in toluene using tetraoctylammonium bromide, then reduced with NaBH<sup>4</sup> and stabilized with dodecanethiol (dodecanethiol/gold molar ratio 2:1), leading to 1.5–2 nm diameter AuNPs quantitatively. The organic gold nanoparticles are purified by precipitation in ethanol and redispersed in chloroform. Twelve nanometers AuNPs are synthetized using the Turkevich method [34]. Briefly, HAuCl4·xH2O dissolved in water is mixed with a sodium citrate solution at 100 ◦C and agitated for 30 min. Higher diameter AuNPs are made following the Ziegler and Eychmüller method [35], by regrowing from the 12 nm AuNPs used as gold seeds. Briefly, a suspension of 12 nm AuNPs in water (120 mL, 90 mg/L) is slowly mixed with an aqueous solution of HAuCl<sup>4</sup> (60 mL, 3.4 mM), sodium citrate, and ascorbic acid (60 mL, 6.3 and 4.6 mM, respectively). The aqueous AuNPs are purified by centrifugation after precipitation with ethanol.

Silicon nanowire synthesis: SiNWs are synthetized in a 150 cm<sup>3</sup> home-built stainlesssteel reactor designed to withstand pressure above 50 bars and temperatures above 500 ◦C. AuNPs are deposited by drop drying either on a NaCl micropowder if the colloid solvent is organic, or on a CaCO<sup>3</sup> nanopowder if the colloid is in aqueous solution (50 mg AuNPs on 25 g NaCl or 3 g CaCO3). After drying, the powder is placed in a 2 cm diameter alumina crucible in the reactor with 12 mL diphenylsilane under vacuum. The reactor is heated to 430 ◦C within 40 min and kept at 430 ◦C for 80 min. After cooling, the powder is washed with water if grown on NaCl, or with 4 M aqueous HCl if grown on CaCO3, and with dichloromethane. To check for the effect of NaCl or CaCO<sup>3</sup> on the growth process, some 12 nm AuNPs were transferred in toluene using ligand exchange with hexadecylamine in toluene [36], then replacing hexadecylamine with dodecanethiol and subsequently depositing on NaCl powder by centrifugation. The SiNWs grown from 12 nm AuNPs on NaCl did not show any change in shape and size as compared to the SiNWs grown from 12 nm AuNPs on CaCO3.

Materials characterization: Phase identification was performed with powder X-ray diffraction (XRD) technique on a Bruker D8 advance diffractometer θ–2θ configuration with a Cu anticathode (Bruker AXS, Karlsruhe, Germany). The scanning step used was 0.02◦ with a counting time of 1.2 s per step. Scanning electron microscopy (SEM) was performed on a Zeiss Ultra 55 microscope (Zeiss, Oberkochen, Germany) at an accelerating voltage of 5 kV and working distance of 5 mm. A JEOL 2010 (JEOL, Tokyo, Japan) high-resolution transmission electron microscope (HRTEM) operated at 200 kV was used for TEM and HRTEM observations. For TEM measurements, the powder was dispersed in ethanol and nanoparticles separated with intensive ultrasound using the Hielscher Ultrasound Technology VialTweeter UIS250V (Hielscher, Teltow, Germany). Then, the dispersion was dropped on a grid made of a Lacey Carbon Film (300 mesh Copper—S166-3H). The Brunauer, Emmet, and Teller (BET) method was used to measure the SBET specific surface of the different samples using Micromeritic apparatus Tristar II and Flowsorb 2300 (Micromeritics, Norcross, GA, USA). Electrochemical studies, including electrochemical impedance spectroscopy, were carried out using a Biologic VMP3 multichannel potentiostat (Biologic, Seyssinet-Pariset, France) and an ARBIN charge–discharge cycle life tester (Arbin, College Station, TX, USA).

Lithium battery assembly and test: SiNPs or SiNWs are mixed in an ink containing 50 wt. % of active material, 25 wt. % of carbon black, and 25 wt. % of CMC dispersed in distilled water. The resultant slurry is coated using the doctor blade method on thin copper foil (12 µm), dried at 80 ◦C overnight (mass loading 0.19–0.46 mgsi cm−<sup>2</sup> , dry thickness around 20 µm), and calendared at 1 ton. A celgard separator was soaked with 150 µL electrolyte of ethylenecarbonate/diethylenecarbonate 1/1 *v*/*v* containing 1M LiPF6, 10 wt. % fluoroethylene carbonate, and 2 wt. % vinylenecarbonate. Two thousand and thirty-two coin cells were assembled with lithium metal as reference and counter electrode in an argon-filled glove box and crimp sealed. Electrochemical properties of the half-cells were evaluated in the potential window between 0.01 and 1.0 V vs. Li+/Li. All potentials reported below were measured in a half cell configuration in reference to the Li metal counter electrode and are, thus, expressed as vs. Li+/Li. The first cycle rate is C/20. Later, cycles are performed at C/5 rate, with a floating time at the end of lithiation until a current of C/100 and C/50, respectively, under 0.01 V. All the capacity values shown in this paper are based on the mass of silicon in the electrode.

#### **3. Results**

#### *3.1. Tuning the Size of SiNPs: Silane Concentration and Residence Time*

Batches of SiNPs with different diameters were synthesized by laser pyrolysis. Table 1 presents the main experimental parameters and size measurements of the samples used in this study. The samples are labeled SiNP<sup>X</sup> where X states the average SiNP diameter. A flow of silane (silicon precursor) is diluted in He with slightly different ratios to control the nucleation and growth. The silane flow rate is varied from 50 to 200 sccm with a total gas flow rate of 1100 ± 100 sccm. Increasing the SiH<sup>4</sup> to He ratio results in samples with diameters regularly increasing from 30 to 87 nm. The samples SiNP<sup>43</sup> and SiNP<sup>53</sup> were

synthesized with the same SiH<sup>4</sup> flow and differ by the He flow rate (1000 vs. 1100 sccm). Finally, in order to reach higher SiNP diameters, we modified the reactor to increase the residence time. The reactant inlet tube close to the laser beam was enlarged from 2 to 4 mm diameter, thus dividing the gas velocity by a factor of 4 and increasing the time of residence in the reaction zone. The synthesis of SiNPs with a diameter of 110 nm could, thus, be achieved, although with a broader size distribution.


**Table 1.** Main experimental parameters and characterizations of the Si nanoparticles (SiNP) samples.

The typical morphology and size analysis of the different samples presented in Figure 1 shows the quality of this synthesis approach, with a very narrow size dispersion obtained for most SiNP samples. The SiNP size was measured both from the BET specific area and estimated from TEM images. The BET diameter is consistently larger than the TEM average diameter because of the morphology of the SiNPs forming small agglomerates. The surface area lost in interparticle contacts decreases the measured BET surface and, therefore, increases the estimated diameter. In the same way, the crystallite size deduced from XRD measurements (Figure S1) is always smaller than TEM indicating that the SiNPs are not monocrystalline, as also seen by HRTEM in Figure S2.

#### *3.2. Tuning the SiNW Diameter: Catalyst Size and Silane Partial Pressure*

Free-standing SiNW growth is achieved by a process similar to the vapor–liquid–solid (VLS) mechanism, using a sacrificial porous support as described in our previous work [30]. The SiNWs are grown at low temperature (430 ◦C) using gold nanoparticles (AuNPs) as a catalyst. The sacrificial powder of NaCl or CaCO<sup>3</sup> covered with the AuNPs is heated in a closed stainless-steel reactor with diphenylsilane as a Si source. At the end of the synthesis, the sacrificial template is removed by washing with water or aqueous HCl, respectively. The optimized process yields up to 0.5 g of SiNWs of 10 nm diameter with a low size dispersion, from AuNPs of 2 nm [30].

It has been demonstrated that diameter control in similar CVD processes can be realized by controlling the catalyst size as was demonstrated for carbon nanotubes [37,38] and for SiNWs [39,40]. Mechanistic studies on SiNW CVD growth show that thermodynamic constraints control the nanowire diameter. It was demonstrated that a high partial pressure of silane (90 Pa) at 400–500 ◦C induces the growth of small diameter SiNWs (2–20 nm), whereas a low partial pressure (5 Pa) at 550–650 ◦C allows the growth of SiNWs of 50 to 1000 nm diameter [26,27,41,42]. However, our closed reactor does not allow for a direct control on the silane partial pressure.

In our process, the diphenylsilane evaporates then decomposes into silane and tetraphenylsilane, following a disproportionation reaction [43–45]. The silane partial pressure is, therefore, controlled by the kinetic balance between its formation from diphenylsilane and its consumption by the SiNW growth. The diphenylsilane disproportionation is slower but starts at a low temperature [44] (from 200 ◦C), while the silane decomposition on gold is fast but requires a high enough temperature to form the Au-Si eutectic [27] at 363 ◦C. Thus, a SiH<sup>4</sup> stock has time to build up in the reactor before the AuNP catalyst turns active for silane decomposition. From diphenylsilane disproportionation rate estimates [44], we can infer a partial pressure of SiH<sup>4</sup> of 100–500 Pa at the onset of gold catalyzed silane

decomposition. At such a high silane pressure, SiNWs can grow as fast as 200 nm/min, and the thinnest SiNWs are favored [26].

‐

‐ ‐

‐

**Figure 1.** TEM images (**a**) and corresponding size analysis (**b**) of the SiNPs samples described in Table 1. The histograms were obtained from at least 100 particle measurements.

‐ ‐ μ We, therefore, developed a novel strategy to achieve size tuning of SiNWs by locally reducing the silane partial pressure. The dense CaCO<sup>3</sup> nanopowder (50 nm) used as a porous growth support is placed in a cylindrical crucible. The powder compacity generates a gradient of partial pressure from the surface to the bottom of the crucible, as illustrated in Figure 2a. Silane entering the crucible is quickly consumed by the top AuNPs, so that only limited silane can diffuse down in the powder. Therefore, the average silane partial pressure inside the powder is lower than at the surface, which drives the growth of bigger SiNWs if proper catalysts are present. We first demonstrated this effect using a 1 cm<sup>3</sup> carbon foam cube (Figure 2b). Analysis of the SiNW diameters at the cube surface and deep in the cube show a significant difference of 30%, as presented in Figure 2c. It can be concluded that the chemical rate of silane consumption for SiNW growth is much faster than the rate of silane gas diffusion, although the porosity of the carbon foam is very large (400 µm pores). A much higher silane depletion is expected in the CaCO<sup>3</sup> nanopowder, in which the pores are thinner (<100 nm) and the pathway more tortuous.

‐ **Figure 2.** Scheme of the partial pressure reduction in the porous growth support, CaCO<sup>3</sup> powder (**a**) and carbon foam cube (**b**). The SiH<sup>4</sup> pressure is represented in red shades. (**c**) Diameters of Si nanowires (SiNWs) grown from 1–2 nm gold nanoparticles (AuNPs) in a 1 cm<sup>3</sup> carbon foam cube on the surface ("top") and 2 mm below the surface ("inside"). Inside, SiNWs had the same diameter distribution at all depths down to the center.

‐

‐

‐ Once able to lower the silane pressure, gold catalysts with the right size are still needed for large SiNW growth. Three colloidal AuNP growth methods [33–35] were necessary to access AuNPs with a wide range of sizes. Small AuNPs (1–3 nm) are efficiently grown by dodecanethiol stabilization [33] in an organic solvent, while bigger AuNPs (12 nm) can only be obtained stabilized by citrate [34] in aqueous solution. The latter can be enlarged in a controlled way using an established regrowth method [35] in water in the range 15 to 120 nm (see SEM images of the AuNPs in insets Figure 3a and Figure S3).

**Figure 3.** SEM images (**a**) and corresponding diameter histograms (**b**) of SiNWs. Insets: SEM images of the AuNP catalysts used for each SiNW growth. Histograms are calculated on >250 counts.

We could check the combined effect of AuNP size and thin porosity of the support on the growth of SiNWs from the 12 nm AuNPs (Table 2, Figure S4): SiNWs grown in the CaCO<sup>3</sup> nanopowder have a diameter of 18 nm, 35% higher than the 13 nm SiNWs grown in the carbon foam. Then, by using AuNP catalysts of increasing size, it was possible to obtain a series of SiNWs of increasing diameter (Figure 3, Figure S3). Note that the growth is not homogeneous, as the thin SiNWs grown on the top constitute a minor population in all samples. This is clearly shown in Figure 3b for SiNW<sup>55</sup> with a small population at 10 nm, aside the main population at 50 nm. Although numerous, the thin SiNWs represent

only a small fraction of the Si volume in the material. Some large SiNWs show a more tortuous shape, and the number of kinks in the SiNWs increases with their diameter. This worm-like morphology, already described in SiNW CVD growth, is typical of an unfavored growth [46,47]. ‐ ‐


**Table 2.** Characteristics of the SiNW samples.

Figure 4a displays the SiNW diameter distribution as a function of the size of the AuNPs used as catalyst. The minor population of small SiNWs is shown with circles. The diameter of the main population of SiNWs nearly matches the size of the AuNPs, as reported for the VLS mechanism [47]. Our strategy for a catalyst-size-directed growth of SiNW, thus, proves efficient, as the SiNW diameter increased over a decade by this method. The distribution width of the main population (measured as the full width at half maximum) is 30–40% for all samples. ‐‐ ‐ 

‐ **Figure 4.** (**a**) Average diameter of the main (diamond) and minority (circles) SiNW populations as a function of the size of the catalyst AuNPs, as measured by SEM (diameter histograms in Figure 3 and Figure S3). The shaded zones show the full width at half maximum and the 90% limit for both SiNW diameter peaks. (**b**) Specific area of the SiNWs as measured by the BET method as a function of the size of the AuNP catalysts.

 The specific area measured by the BET method on the SiNW samples (Figure 4b) drops consistently when increasing the size. The BET and SEM data are in agreement for the smallest SiNWs (SEM estimated specific area of 150 m2/g for a BET surface of 194 m2/g). However, for larger SiNW samples, they diverge, showing an underestimated count of the small SiNW population by SEM.

#### *3.3. Electrochemical Performance in Li-Ion Batteries*

The electrochemical performance of SiNPs and SiNWs in lithium batteries are displayed in Figure 5. All materials were tested with 50 wt. % active material in the anode. The

quantity of carbon black (25 wt. %) and binder (25 wt. %) were quite high to ensure good electronic conductivity and mechanical stability during electrochemical measurements for all materials. The contribution of the carbon black to the specific capacity, measured independently, is a constant 100 mAh/g. ‐ ‐ 

‐

*‐*

‐ ‐ ‐ **Figure 5.** Electrochemical performance of SiNPs and SiNWs measured in half-cell configuration at C/20 rate (activation), then C/5. Specific charge capacity (mAh per gram of silicon) as a function of the cycle number obtained from (**a**) SiNP and (**c**) SiNW electrodes. Corresponding Coulombic efficiency for (**b**) SiNP and (**d**) SiNW electrodes. The results presented here are the most representative of at least 3 repeatable cells (standard deviation 10%). Data for all SiNP sizes are available in Figure S5.

‐ ‐ μ ‐ The specific capacity is higher and more repeatable for SiNPs (3000–3500 mAh/g) than for SiNWs (2500–3000 mAh/g). All SiNP anodes give a similar specific capacity. For SiNWs, we observe a minimum of specific capacity for the medium size, SiNW42. The most favorable trade-off between specific capacity and first irreversible capacity is observed for the largest size, SiNW<sup>55</sup> (Figure 5c and Figure S5). However, among all samples, only the smallest SiNW anode, SiNW9, attains the Coulombic efficiency of 99.5% required for long term cycling (Figure 5d). The lower specific capacity of the SiNWs may be due to difficulties in dispersion of the material during the slurry preparation. The growth in a porous support favors SiNW entanglement, and the obtained 1–10 µm sized agglomerates do not fully separate during slurry process, thus leading to a poorer wetting by the electrolyte in the agglomerates (Figure S6). During the potentiostatic step applied at the end of full lithiation, the equilibrium is not reached, especially for large SiNWs (Figure S7). This indicates residual capacity still available in the SiNW electrode and a kinetic limitation to lithiation in the agglomerates.

‐ ‐ The loss in the first Coulombic efficiency (CE) is due to irreversible processes happening in the first lithiation (also called activation). For nano-silicon anodes, it is mostly related to the solid–electrolyte interphase (SEI) formation. It depends on the surface area of the material as demonstrated for graphite particles [48]. Figure 6a presents the first and fifth CE as a function of the BET surface measured for SiNPs and SiNWs. As can be seen for the SiNP electrodes, the excellent correlation shows that the first CE depends linearly on the specific area. The capacity loss in the first cycle is, thus, due to the homogeneous coverage of the silicon surface by a SEI passivation layer. SiNWs follow the same trend, but with a poorer correlation due to a less controlled wetting in the SiNW agglomerates. To the best of our knowledge, this linear correlation has never been reported for silicon. In the subsequent cycles (fifth CE shown on Figure 6a), the CE rises above 95%, and the linear correlation with specific area fades away. On Figure 6b, the first CE is presented as a function of SEM/TEM diameters, *d*, and shows the expected 1/*d* evolution. The SiNW diameter reported here is the average of the most abundant SiNW population. The presence of the small SiNWs brings a large additional surface area and, thus, a higher irreversible capacity in the first cycle. ‐ ‐ 

‐ **Figure 6.** Coulombic efficiency as a function of BET specific area (**a**) and statistical SEM/TEM diameter (**b**) for SiNPs (full circles) and SiNWs (empty circles) on the 1st cycle (black) and the 5th cycle (blue).

‐ Upon long-term cycling, the SiNW<sup>9</sup> electrodes outperform the other materials due to their enhanced stability and very high CE (Figure 5d). On the contrary, the SiNP specific capacity suffers from a faster fading. The size of SiNPs also has a strong impact on the evolution of the Coulombic efficiency upon cycling. Indeed, for the smallest SiNPs, the CE presents a progressive increase, whereas the larger SiNPs show a fast initial increase then a decrease with a minimum around 10 cycles. This phenomenon could be attributed to an electrochemical sintering [49–51] leading to larger SiNPs. Microscopy studies on SiNP anodes in cycling showed general sintering, forming large networks of sintered SiNPs in the anode [49]. Such large Si structures are sensitive to mechanical pulverization in the subsequent cycles, leading to a loss in Coulombic efficiency.

‐ ‐ To better assess the impact of the Si shape and size during electrochemical cycling, we plotted the normalized galvanostatic cycle as a function of the cycle number. During the first lithiation (Figure S8), the very long potential plateau at ca. 100 mV is shifted toward lower potential for larger SiNPs, indicating polarization. This is in agreement with a slower Li diffusion along longer distances within large SiNPs [52]. In the following cycles (Figure 7), we can see that polarization during lithiation is increasing as a function of the cycling number for the large SiNPs. For the SiNWs, the potential plateau in the first cycle is also low, indicating anode polarization, which is independent of size (Figure S8), and in the subsequent cycles, polarization does not increase (Figure 8). This discrepancy indicates an influence of the Si shape on the Li diffusion.

‐ Stronger differences can be seen during the delithiation. For SiNPs, a very long potential plateau at ca. 450 mV appears during the first cycles (second or third depending of the samples) and disappears after more than 30 cycles. This potential plateau is generally ascribed to the biphasic delithiation of the crystalline Li15Si<sup>4</sup> phase [53] and was reported to be size-dependent (appearing with large particles [50,54,55]) or linked to the stress distribution in the case of thin films [56]. The appearance of this plateau indicates that a large part of the Si is fully lithiated to the crystalline Li15Si4, even if the crystallization of this phase is reported to be more difficult in small SiNPs [50,54,57]. According to the literature, this phase also reduces the Coulombic efficiency [58], which is in agreement with the observed CE evolution of the SiNPs in our study (Figure 5b).

Thus, the CE fluctuations observed for larger SiNPs would correlate with the pronounced potential plateaus in delithiation, i.e., with the larger amount of crystalline Li15Si<sup>4</sup> formed during lithiation. Deep lithiation has been reported to favor the electrochemical sintering of SiNPs in lithium battery anodes, because the swelling of Si in the form of an amorphous Li alloy induces the formation of necks between neighboring particles [49,51,59]. Within this soft structure, the crystallization of Li15Si<sup>4</sup> in the points of deepest lithiation might strengthen these simple contacts by growing crystals through the necks. Indeed, our SiNPs are connected in a necklace morphology from the growth (Figure 1a), which could enhance this phenomenon.

‐ ‐

‐

‐

Surprisingly, this crystalline Li15Si<sup>4</sup> delithiation process is much less visible in the SiNW electrodes, with very small potential plateaus at 450 mV (Figure 8d–f). This shows that the crystalline Li15Si<sup>4</sup> phase is present in a much smaller quantity, or that the lithiated phase remains mostly amorphous [53]. Only the mid-sized SiNW<sup>42</sup> shows a significant plateau, a fact that can be correlated with its low Coulombic efficiency as compared to the other SiNW samples (Figure 5d). A reason why SiNWs do not undergo electrochemical sintering as easily as SiNPs might be related to their elastic "spring" behavior. Even after grinding and calendaring, SiNWs in the anode remain stiff. Agglomerates of SiNWs contain a nanoscale porosity that was clearly imaged in our recent FIB-SEM study [3]. Such porosity allows swelling during lithiation with a low risk of sintering with neighboring SiNWs. ‐ ‐ ‐ 

**Figure 7.** Normalized galvanostatic curves in lithiation (**a**–**c**) and delithiation (**d**–**f**) as a function of the cycle numbers for SiNP<sup>30</sup> (**a**,**d**), SiNP<sup>71</sup> (**b**,**e**), and SiNP<sup>87</sup> (**c**,**f**).

**Figure 8.** Normalized galvanostatic curves in lithiation (**a**–**c**) and delithiation (**d**–**f**) as a function of the cycle numbers for SiNW<sup>9</sup> (**a**,**d**), SiNW<sup>42</sup> (**b**,**e**), and SiNW<sup>55</sup> (**c**,**f**).

#### **4. Discussion**

‐ The discrepancy in the ability of the nano-silicon material to lithiate into crystalline Li15Si<sup>4</sup> may have a direct impact on the specific capacity retention of the LiB, as shown on Figure 9. The smallest SiNWs show a much higher specific capacity retention over the fourth to 80th cycles, while most of the capacity fading for the SiNPs happens in the cycles 4–30, i.e., when lithiation into crystalline Li15Si<sup>4</sup> is observed. ‐

**Figure 9.** Capacity retention for small (black circles) and large (red triangles) SiNPs (full symbols) and SiNWs (empty symbols).

‐ ‐ The influence of the shape/size of the nano-silicon on the reaction mechanisms in cycling and the electrochemical performance can be summarized as follows:


#### **5. Conclusions**

This paper presents the first extensive comparison of size/shape of nano-silicon (nanoparticles and nanowires) used as anode materials in lithium-ion batteries. The main challenge was to obtain each nanomaterial with a good size and shape control, and in gramscale quantities for reliable LiB electrochemical tests. We used two independent synthesis processes: laser pyrolysis for SiNPs (now an industrial process) and gold-catalyzed VLS growth in porous media for SiNWs (a scalable patented process). Tuning the synthesis parameters, we obtained SiNPs in a 25–110 nm diameter range. By tuning the gold catalyst size and the porosity of the support, SiNWs with diameters ranging 9–93 nm were realized. An important difference between the two nano-silicon morphologies is that the SiNP material, grown in a gas flow, consists of a fine nanopowder, while the SiNW material, grown in a porous support, consists mostly of 1–10 µm sized agglomerates. Although an advantage in dealing with nano safety, this turns to a disadvantage in the making the electrode, leading to a less homogeneous slurry and to a part of the Si being inaccessible to the electrolyte. Unsurprisingly, the 1D shape appears more difficult than 0D to control and produce at a large scale.

The matching series of SiNP and SiNW materials was implemented in lithium batteries in the same conditions and cycled the same way. Their electrochemical behavior shows several consistent effects of size and shape. First, the initial specific capacity of the material

depends on its shape: it is close to the theoretical 3579 mAh/g for SiNPs, while SiNWs provide an about 25% lower capacity due to the entanglement of SiNWs during the growth. Second, the irreversible specific capacity at the first cycle is linearly correlated to the specific surface area of the materials, thus to the nano-silicon size. Third, the SiNPs prove much more prone to lithiation down to the crystalline Li15Si<sup>4</sup> phase, and probably to electrochemical sintering, than SiNWs. Finally, in the long run, the smallest SiNW electrodes show a better capacity retention, with a Coulombic efficiency above 99.5% after 43 cycles. A size effect is again observed as smaller SiNWs and SiNPs show a lower polarization and, therefore, a better capacity retention. Improvements are currently under investigation, using carbon coating in the case of SiNPs to reduce irreversible capacity and using direct growth on graphite in the case of SiNWs to make more efficient composites.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/2079-499 1/11/2/307/s1, Figure S1: X-ray diffractograms of the SiNPs; Figure S2: HRTEM images of SiNP30; Figure S3 SEM images of AuNPs and SiNWs; Figure S4: SEM images/diameters of SiNWs grown from 12 nm AuNPs on CaCO<sup>3</sup> nanopowder or in carbon foam cube; Figure S5: Specific capacity and coulombic efficiency for all SiNPs; Figure S6: SEM image of an electrode with SiNW<sup>9</sup> ; Figure S7: Specific capacity in the galvanostatic/potentiostatic steps; Figure S8: Cell potential vs. normalized galvanostatic capacity.

**Author Contributions:** A.D., F.B. performed synthesis of SiNPs; C.K., E.M., P.C. performed synthesis of SiNWs; A.D., N.H.-B., C.H., J.P.A. performed characterization and analysis of results on SiNPs; C.K., S.K. performed characterization of SiNWs; C.K., C.V., N.H.-B., C.H., P.C. finalized the analyses and wrote the paper. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was funded under the CEA Silicon for Lithium-Ion Batteries (SiBaLi) project, the French National agency under the HELIOS project (ANR-13-PRGE-0006), and the H2020 European funding Flagship Graphene Core 2 (grant agreement 785219). J.P.A. is grateful for support of Eurotalents postdoctoral fellowship.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author.

**Acknowledgments:** N.H.-B., F.B., A.D. gratefully thank Dominique Porterat for his constant help in solving all kinds of experimental trouble. A.D. and N.H.-B. thank Eric Larquet at Ecole Polytechnique for TEM and HRTEM measurements. A.D. and C.H. thank Daniel Tomasi for battery making and cycling. C.K. and P.C. thank Annette Delices for AuNP size tuning and synthesis. C.K. and C.H. thank Agnès Brun for BET analysis. C.K. and P.C. thank Gérard Lapertot for his help in designing the stainless-steel reactors.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

