1. Introduction
Silicon serves as a pivotal functional material; it is the bedrock for microelectronics and solar cells and is continually gaining technological significance owing to its abundance on Earth and cost-effectiveness. However, its use as a thermoelectric material is hampered by inherent limitations. Despite exhibiting relatively high power factors (
) at elevated carrier densities, the silicon thermoelectric figure of merit
(where
is electrical conductivity,
is the Seebeck coefficient,
T is absolute temperature, and
is thermal conductivity) remains meager, reaching approximately 0.01 at room temperature [
1], primarily due to its high thermal conductivity of around 120 W/mK at room temperature.
To mitigate this thermal conductivity challenge without compromising electrical properties, one viable avenue involves the creation of dimensionally constrained nanostructures (NSs) [
2]. Silicon nanowires (NWs) with diameters below the phonon mean-free path (MFP), ≈200 nm in Si at room temperature [
3], exhibit substantially reduced thermal conductivity (<5 W
) [
4,
5,
6]. Similarly, silicon nanolayers (NLs) also manifest lower thermal conductivity, contingent upon the nanolayer thickness [
7]. In both cases, dimensional constraints cause incoherent phonon scattering at the NW/NL walls. Since, in silicon, thermal conductivity is largely due to heat transported by long MFP phonons, this leads to a major
reduction. An alternative strategy involves top-down control to generate nanopores in ‘holey’ silicon, reducing
without adversely affecting the power factor [
8,
9].
Regardless of the chosen strategy for minimizing thermal conductivity, two critical considerations must be addressed to ensure the efficacy of the resulting nanomaterial for heat harvesting or refrigeration. Firstly, manufacturing methods should facilitate the cost-effective fabrication of the final device. Equally crucial is the need to enhance efficiency (coefficient of performance) without sacrificing power density (cooling power). This is particularly pertinent in the case of NWs, where Si NWs are typically supported on insulating templates when fabricated using standard integrated-circuit technologies [
10]. In most instances, however, Si NWs fabricated on insulating templates result in a low ratio of thermoelectric active cross-sections to the device footprint, significantly diminishing the accepted input thermal power and consequently limiting the output power density. We will demonstrate that metal-assisted chemical etching (MACE) can potentially surmount such limitations, concurrently addressing the imperatives of low production costs and high filling factors.
Arguably, the initial application of MACE in producing Si NWs for thermoelectric purposes was documented in one of the foundational papers on thermoelectric NWs, published back-to-back in
Nature in 2008. In their seminal works, Boukai and colleagues achieved notable thermoelectric performance in Si NWs prepared by extreme lithography techniques [
5]. Hochbaum and collaborators demonstrated comparable outcomes by employing MACE to craft NWs protruding from a silicon substrate [
4]. Hereafter, we will refer to them as nanopillars (NPs). Subsequently, MACE has undergone extensive exploration, driven not only by its application in Si NPs for thermoelectricity but also due to Si NP usability in diverse areas such as anti-reflective finishing in solar cells [
11], chemical sensing elements [
12], and nanoantennas [
13], among other examples. Compared to extreme lithography, the main advantage of MACE is its simplicity, as the method does not require high vacuum and it is easily scalable to obtain NP forests over large areas. However, NPs are randomly distributed over the Si surface, which could be an issue for specific applications.
MACE can be carried out in two ways (
Figure 1). In the two-pot MACE method, silicon undergoes brief exposure to a solution containing metal ions (including
,
, and
) and HF [
14]. Their reduction leads to the formation of metallic nanoparticles, sparsely coating the Si surface. The wafer is then immersed in a second solution, still containing HF and an oxidizing agent, typically
. Silicon preferentially oxidizes at the Si–metal interface. Thus, metal nanoparticles penetrate the wafer. The remaining unetched Si materials form the NPs. Conversely, the one-pot MACE technique employs a single solution where
ions serve as both precursors for catalytic nanoparticles and as the oxidizing agent. In this approach as well, Si NPs are the unetched Si portions.
Although elemental metals are the most common choice as catalysts, other materials have been considered. This is especially worthwhile when NPs are to be used in microelectronics, where transition metal contamination is to be avoided. As an example, titanium nitride was shown to lead to localized etching and the formation of Si NPs, although MACE had to be performed in the vapor phase [
15]. In all implementations, MACE emerged as an uncomplicated yet finely controlled method for producing single-crystalline Si NPs, affording excellent control over the wire length, orientation, doping type, and level, thereby enabling the achievement of high NP densities.
This paper reports on recent progress aimed at using MACE as a technology to obtain low-cost, large-area, efficient thermoelectric devices, exploiting the advantages of nanotechnology in the making of integrated micro-thermoelectric harvesters and macroscopic thermoelectric generators (TEGs). Specifically, it will be shown how NP forests can be obtained from both p- and n-type silicon, with doping levels up to
, preserving their bulk crystallinity in all cases. Two procedures for establishing electrical contacts on NPs will be described. First, following an ingenious procedure developed by Pennelli and co-workers [
16], it will be shown how copper can be electrochemically plated under non-equilibrium conditions, namely by applying large current densities. Contacts are always non-rectifying due to defect injections at NP tips occurring during MACE. As a second possibility, we embedded NP forests into a polymeric matrix, obtaining mechanically stabilized elements that could be used as conventional thermoelectric legs in TEGs and thermoelectric coolers (TECs). It should be mentioned that alternate approaches have been reported in the literature [
17], which will be discussed and compared to electroplating and embedding.
4. Discussion
In this section, we first summarize the mechanism of MACE, emphasizing the model that explains bore protection during one-pot MACE. Such a mechanism is a key element used to explain the observed characteristics of metal contacts.
4.1. Electrochemistry of MACE
In either one- or two-pot implementation, MACE involves an electrochemical reaction (reduction of the oxidant and the oxidation of Si) and a chemical process, namely the etching of oxidized silicon. The electrochemistry of MACE has been addressed by many scholars. The use of
/HF solutions on surfaces that were previously patterned either with Ag or Au has received the most substantial attention [
25,
26,
27,
28], as it may also be used to fabricate ordered nanopillar arrays. In many contexts, however, randomly distributed NPs are of use, and one-pot MACE provides an easier way to obtain NP forests. In both cases, the anodic reactions of the electroless process are the same [
29]. The oxidant injects holes into Si, where two processes take place. In the first, silicon is directly oxidized to
as
and then
is etched by HF:
Thisprocess is often referred to as electropolishing. When etching occurs homogeneously over the whole surface area, it results in a smooth, crystalline, defect-free surface. When etching is localized, it leads to the formation of single-crystalline NPs.
A two-electron process may also occur, where Si is initially oxidized to Si(II) and dissolved as such [
30]:
Oxidationof Si(II) is then completed in the solution:
also with the formation of
. This second mechanism always results in the formation of a porous silicon layer or, when localized, silicon NPs with porous surfaces [
30].
As of the cathodic reaction, in one-pot MACE,
ions are reduced to Ag
either at the bare Si surface or at the Ag surface. They are the only oxidants sustaining the redox reaction. In two-pot MACE silver, nanoparticles are nucleated in the first stage, and then Si is typically oxidized by hydrogen peroxide, as follows:
The literature indicates that the two processes take place simultaneously during the electrochemical etching of silicon. In an electrochemical cell, electropolishing becomes dominant at elevated current densities, surpassing a critical current density,
, of approximately 0.5 A/
[
29].
The role of the threshold current can be explained by considering the rate-limiting step of the electrochemical etching of silicon. At lower potentials, the rate of oxide formation is slow, leading to the immediate etching of as it forms. Consequently, local oxide passivation of the surface (resulting in porous Si through the two-electron mechanism) occurs when oxide formation proceeds slowly, namely the oxidation rate is minimal. Rather, at a higher potential, oxide accumulates across the entire surface, as its removal by HF is now a slow step. Thus, the entire surface is uniformly coated with the oxide, and its removal by HF occurs simultaneously across the entire surface. This results in Si electropolishing.
A comprehensive examination of the intricate interplay between the defectivity of NP surfaces and doping levels and types was carried out [
18]. Transmission electron microscopy investigation, aligning partially with prior studies [
26,
31,
32], reveals that at low doping levels (either a p-type or n-type), NPs maintain their crystallinity without any discernible defect. Conversely, at higher doping levels (both n- and p-types), surface damage becomes evident. It is crucial to emphasize that in the context of one-pot MACE, only the outer surfaces of NPs exhibit porosity, with the bulk of the NPs retaining their pristine crystalline structures. This observation lines up seamlessly with the underlying etching mechanism. As the NPs constitute the unetched segment of the original wafer, the increasing influence of the two-electron mechanism exclusively impacts the portion of NPs exposed to the etching solution, resulting in porosity solely on the outer surface while the bulk retains its crystallinity.
It is intriguing to note that, as dopant concentrations increase, irrespective of whether they are of p- or n-types, the etch rates exhibit a decay. The activation energy of MACE undergoes a decrease when transitioning from
to
substrates [
18]. Additionally, surface porosity manifests even in lightly doped substrates at elevated MACE temperatures, where etch rates (or injected current densities) are comparatively higher. The parallel trend observed in etch rates and the prevalence of two-electron over four-electron processes suggest that there is no simple relationship between etch rates and surface porosity. For a deeper understanding of the hole injection mechanism, delving into the electronic structure of Si and Ag, along with their interface, proves insightful. Considering that polycrystalline Ag deposited onto silicon at low temperatures yields an experimental Ag work function as low as 4.26 eV [
33], the existence of a barrier opposing hole injection becomes questionable. It appears that the crucial determinant for hole injection through a non-rectifying contact is the material resistivity,
, governing the electromotive force,
experienced by Si. Since
(with
ℓ denoting the distance between cathodic and anodic regions) and acknowledging that the current density,
j, is proportional to the etch rate, it is evident that surface-porous NPs are achieved only under low voltage conditions. This observation agrees with the findings of Zhang and co-workers [
34] concerning the electrochemical formation of porous silicon. The impact of MACE temperature also aligns with this model, as higher temperatures escalate the rate of the chemical step in the etching process. Consequently, the range of potential differences where HF removal is not the limiting reaction step widens, leading to the observation of surface-porous NPs at higher
.
4.2. Etching Localization and Lateral Bore Protection
A pivotal aspect defining any MACE process is the localized etching of Si. On (100) Si surfaces, etching rates remain uniform across all surfaces. Consequently, one might anticipate lateral bore etching as holes injected into Si diffuse to neighboring regions. However, this expectation is contradicted, as protective mechanisms—either chemical or electrochemical—are in effect.
In the case of two-pot MACE, bore protection arises from the heightened etching rate at the metal–silicon interface. As metal nanoparticles embed themselves into Si, etching primarily occurs at the bottom of the bores [
31]. Conversely, in one-pot MACE, bores are overfilled with metallic silver. Pal, Ghosh, and Giri [
35] attributed bore protection to the decreased stability (higher reactivity) of Si surfaces at the bottom of the bore, induced by defects stemming from the ongoing oxidation process. Wall protection by hydrogen released through two-electron MACE has also been considered, although this mechanism is evidently unsuitable for explaining directional etching leading to single-crystalline NPs [
36]. An alternative mechanism, based on the analysis of Ag particle dynamics in
/HF MACE, offers a fresh perspective. Drawing inspiration from a model proposed by Peng et al. for two-pot MACE [
37], we posited [
18] that diffusion hindrance by Ag dendrimers prevents
hole transfer at the submerged end of the Ag aggregate. Consequently, the silver reduction must take place at the outer dendrimer end, propelling momentum transfer at the solution-Ag interface. This process causes Si to be extruded at the surface, where oxidation takes place (
Figure 7). This dual functionality explains both the directional etching (and, hence, the lateral bore protection in one-pot MACE) and the formation of Si flakes enveloping Ag nanoparticles [
18].
4.3. Filling factors and Contact Resistances
It should be stressed that any attempts to visually count tips and their areas are subjected to large incertitude; however, the agreement between electrical and visual filling factors lend credibility to each other.
Comparable filling factors before and after contact formation show that both approaches—electroplating and polymer encapsulation—caused no significant alteration of NP orientation. This is not obvious, as both methodologies exposed NP forests to chemicals that could have modified their arrangement. Concerning contact resistances, at present, electroplating outperforms polymer embedding. Residual polymer layers on top of NP ends that survive polishing can be ruled out since insulating layers between Si and the metal layer would lead to non-linear current–voltage characteristics independently of the doping type. It cannot be excluded that, during polymerization, the polymer might also partially cover the sample’s backside, decreasing the real back-contact area and, therefore, introducing a large apparent contact resistance. Nonetheless, embedding NPs remains a key issue in granting due mechanical stability to thermoelectric legs. Therefore, additional research will be needed to optimize the process.
4.4. Nanopillar Defectivity
Comparison between current–voltage characteristics of electroplated and encapsulated devices based on lightly doped NPs ( ), and p- and n-types, provide insights about their defectivity.
First, the striking difference between contacts made with Cu on encapsulated or electroplated p-type NPs is consistent with the mechanism of NP formation by MACE. Although MACE implies faster etching at the metal-Si interface, etching also occurs over the clean Si surface. Both two- and four-electron etching mechanisms inject defects into Si-inducing localized states at the Si-
interface (as detected by photoluminescence spectroscopy [
38]). Their formations sensibly result from the strain generated during MACE. As reported, in one-pot MACE, the anodic process takes place at the extruded Si flakes, where strain accumulates and exceeds the elastic regime, causing the formation of point and extended defects. Localized states associated with such surface defects explain the lack of rectification in electroplated samples. Tunneling through a thin potential barrier at the metal–semiconductor junction is responsible for the linear current–voltage characteristics. This is also in agreement with the initial rectifying back-contact, which could be made linearly by injecting defect states through mechanical damage.
When encapsulated, polishing needed to uncap NP ends results in the removal of the pristine tips. Thus, the top surface exposed to the metal by NPs was never subjected to any chemical etching and is, therefore, virtually defect-free, since polishing (down to a residual roughness of 0.3 m) on bulk silicon has been shown not to introduce a significant density of localized states on the surface.
The picture for n-type NPs is completely different. Aluminum contacts (with an Al work function of 4.08 eV) are sufficiently electropositive to avoid setting any barrier on n-type Si. Thus, both electroplated and encapsulated NPs display ohmic characteristics.
5. Conclusions
We presented and discussed strategies to obtain metal contacts on Si nanopillars prepared by metal-assisted chemical etching. Two methodologies were considered, namely direct electroplating, where copper contacts were fabricated using electrochemical deposition, and polymer encapsulation followed by the standard e-beam-assisted evaporation of metal contacts. In both cases, we showed how optimization of the preparation conditions is strictly entangled with accurate control of the NP defectivity and factors ruling their mutual interactions. To this aim, results achieved over the years by the present authors and by other scholars were summarized. Building upon such findings, we showed how NP surface termination rules NP bundling, not only at the end of the MACE process but also in the subsequent stages of contact fabrication. This enabled the preparation of contacted NP forests preserving high filling factors, which is a prerequisite for using NP forests as active components (legs) in TEGs and TECs.
When contacts were obtained by electroplating, low contact resistances could be obtained along with non-rectifying metal-NP junctions on both n- and p-type NPs at any doping level. Additional work is needed when NPs are embedded into an insulating matrix. While the filling factor remains large, contact resistance is large, possibly because of the improvable polymerization and polishing procedures. As an alternate strategy, Amat et al. reported on the successful fabrication of metallic contacts on polymer-embedded NPs by using reactive-ion etching (RIE) to uncap NP tips [
17]. Upon accurate optimization of RIE conditions (power, duration, and cycles),
plasma enabled the selective removal of the polymer (a photolithographic resist) while preserving Si NPs. Although RIE is a technology that is suitable for processing large-area samples, the last step of contact deposition was reported to require the use of (non-scalable) e-beam lithography. Nonetheless, replacing or complementing mechanical polishing with oxidative etch-back is an intriguing approach that should be further explored.
Concerning contact linearity, mechanical polishing that uncaps NP ends also removed the pristine NP tips, where defects were injected upon MACE processing. Thus, paradoxically enough, the lower defectivity of the metal–silicon interface did not allow for obtaining non-rectifying contacts on p-type NP forests. Strategies, including low-temperature metal alloying, to obtain tunneling linear junctions on p-type NPs are under evaluation.
In all cases, we showed how the wise control of defectivity is of paramount importance to understanding and controlling contact quality toward the full qualification of NP forests as Si-based thermoelectric legs with high figures of merit to be deployed in thermoelectric devices.