Solvent-induced electrochemistry at an electrically asymmetric carbon Janus particle

Chemical doping through heteroatom substitution is often used to control the Fermi level of semiconductor materials. Doping also occurs when surface adsorbed molecules modify the Fermi level of low dimensional materials such as carbon nanotubes. A gradient in dopant concentration, and hence the chemical potential, across such a material generates usable electrical current. This opens up the possibility of creating asymmetric catalytic particles capable of generating voltage from a surrounding solvent that imposes such a gradient, enabling electrochemical transformations. In this work, we report that symmetry-broken carbon particles comprised of high surface area single-walled carbon nanotube networks can effectively convert exothermic solvent adsorption into usable electrical potential, turning over electrochemical redox processes in situ with no external power supply. The results from ferrocene oxidation and the selective electro-oxidation of alcohols underscore the potential of solvent powered electrocatalytic particles to extend electrochemical transformation to various environments.


1-2. Purification of SWNT powder
Further purifications were done using an extraction method to remove a water-soluble impurity and an acid purification method to decrease a catalyst residue. Firstly, the extraction of watersoluble impurity is done using hexane/water system. After SWNT powder is dispersed in pure water by bath sonication, hexane is added to water/SWNTs solution, then it is shaken for 1 minute to extract all the SWNTs from water phase to hexane phase. After the water phase is disposed, fresh pure water is added again and shaken to extract the water-soluble impurity more. This process is done 3 times, and then SWNT powder is collected from hexane phase using glass filter with reduced pressure. SWNT powder is ground and dried on the glass filter for 15 minutes, then transferred into a vial and dried further for overnight under high vacuum condition. Secondly, acid purification is done using non-oxidative 37% hydrochloric acid (Aldrich). After adding the hydrochloric acid to SWNTs powder, the solution is bath-sonicated for 15 min to disperse SWNT powder. Then, solution is kept at 45°C for 2 hours with stirring. The hydrochloric acid solution is diluted with pure water and SWNTs is collected using glass filter with reduced pressure. The SWNT powder is washed by pure water 10 times and by acetonitrile once, then ground and dried for 15min on the glass filter. Grinding process is important to make a uniform SWNT network by hot-press method. The powder is transferred into a vial and dried further for overnight under high vacuum condition.

1-3. Characterization of SWNT powder
Purified powder samples were examined by XPS, BET, and UV-VIS-NIR for a characterization.
In order to confirm the non-oxidation purification process, oxygen atomic concentrations on both pristine and purified powder samples were checked using XPS (ULVAC-PHI, INC. PHI VersaProbe II) with a monochromated Al Kα source. The oxygen content value was calculated as an atomic percentage from the integration of O1s peak and that of C1s peak in the high-resolution scans as shown in Table S1-1. It was confirmed that there was no obvious increase of oxygen content due to the oxidation during this purification process. Chlorine and Iron contents were also checked by XPS as shown in Table S1-1. BET measurements with N2 was done using an Accelerated Surface Area and Porosimetry System from Micromeritics (ASAP 2020). Samples are degassed at 150 °C, then surface area normalized by weight was measured at -196°C. Results are shown in Table S1-2. SWNT powder used in this study showed more than 500m 2 /g BET surface area which typically well purified SWNTs have, so successful purification was confirmed by this BET measurement. UV-VIS-NIR chirality characterization was also performed. Initially, SWNT sample was dispersed in 2 wt% SDS solution for 20 hours using tip sonicator. Then, SWNT solution was centrifuged at 32000 rpm for 4 hours. Only supernatant was measured by UV-VIS-NIR scanning spectrophotometer (Shimadzu, UV-3101PC). After taking a spectrum, baseline was subtracted from the spectrum using the method of Kevin and Rishabh et al. 1 Spectra after subtraction were shown in Figure S1-1. Peak deconvolution of each chirality were also done to calculate the Langmuir surface area m²/g 913.5 BJH adsorption surface area m²/g 360.4 BJH desorption surface area m²/g 412.9 average pore size nm 9.0 chirality information by Lorentzians peak fitting using the method of Kevin and Rishabh et al. 1 Figure S1-1． UV-VIS-NIR spectra in 2wt% SDS solution of HiPCo SWNT powder.

1-4. Preparation of SWNT network
SWNTs network was made by hot-press method using Laboratory Press (Carver, Inc. Model #3912). 30 mg of SWNT powder was put on the Teflon sheet and 50μL of water was added to make the network rigid, then another Teflon sheet covers the SWNT powder. Hot-press was done by 5 tons at 50 °C for 10 min. After the pressing, SWNT network was dried for more than overnight, then it was cut into specific sizes.

1-5. Preparation of devices
The cut SWTNs network was placed on across two copper electrodes then were fixed with another copper to make a device for voltage and current measurement. Polymer coating (polyvinylalcohol) on the copper electrodes were added in order to protect the connection between SWNTs and electrodes from liquid acceptors.
2. Fabrication procedure of asymmetric SWNT particles SWNT particles were prepared using the method below, shown in Figure 2-1. To create particles capable of solvent-induced galvanic potential, we hot-pressed purified and oxidized SWNT powder into 500 µm thick sheets, with one side covered with a barrier polymer material such as Nafion, polyvinylalcohol (PVA), or polytetrafluoroethylene (PTFE). Dicing these sheets into 250 µm cuboids creates carbon Janus particles, leaving only the exposed (unprotected) surface with direct access to the surrounding solvent. If drying is performed on the particle preserved in solvent before actual use, then the shelf life should be months if not years.
Here PCT denotes the probability of carrier transfer across the metal-SWNT contact; A represents the overlapping integral between the two wavefunctions; k and α are scaling parameters.
In the case of Asymmetric Chemical Doping, upon lattice withdrawal of electron from the dopant exposed side of the SWNT, the desired carrier transport process is the drift-diffusion from the undoped side to the doped side of the SWNT. This is a productive process that will increase the observed voltage output. However, simultaneous carrier transport from the metal electrode into the doped SWNT occurs as well, diminishing the observed voltage. The rate of carrier transport of this undesired process depends on energy gap (between the metal SWNT work functions) and the overlapping states (in a probabilistic sense) ( Fig. S3-1). The combined rate of carrier accumulation within doped SWNT can be expressed as: where N doped is the carrier concentration on the side of the doped SWNT; EF denotes the Fermi level; e is the elementary charge; F is Faraday's constant; η is carrier mobility; L is SWNT conduit length; and D is carrier diffusivity; k, α, and β are left as fitting parameters. Note that both the work function of the metal (EF elec ) and its binding energy with the SWNT (Ebinding) contributes to the rate of carrier transport across the metal-SWNT junction, with the work function affecting the barrier height for the activated carrier transport process and the binding energy affecting the probability of that process. With this model, we can simulate carrier dynamics within the SWNT.
To relate back to voltage output, we calculate the number density of electrons for (n, m)-SWNT, Ne (n,m) , needed to lower the SWNT's Fermi level by 1 meV upon removal: Here kB is the Boltzmann constant and T is system temperature, which remains constant at 298 K.
ΔE represents the energy difference between the original and the reduced Fermi levels of the SWNT due to ET, which is set to 1 meV by definition. Ni (n,m) and Nh (n,m) represent the corresponding density of holes before and after the Fermi level reduction in the (n, m)-SWNT valance band (VB), respectively. Ni (n,m) is a strong function of the density of state (DOS) of (n, m)-SWNT, and is estimated following a procedure by Marulanda et. al. 2 From our simulation results, we observe that the voltage output in ACD experiments is critically affected by Ebinding ( Fig. S3-2). In both cases depicted in Figure S3-2, the metal electrode has a work function of 4.9 eV. On the left we simulated a scenario where Ebinding is 0.6 eV, whereas on the right we simulated the case when Ebinding is increased to 1.4 eV. We see that the resulting measurable voltage in both cases are 600 mV and 10 mV respectively (defined here as the difference between the terminal Fermi energy of the doped SWNT and the metal work function). Using this model, we surveyed a wide range of E binding values and plotted the simulated short circuit current output as a function of these values ( Fig. S3-3a) to highlight this dependence of voltage (and hence current) output as a function of E binding between metal and SWNT. A literature survey was also performed to estimate the binding energy between SWNT and different metal electrodes ( Fig. S3-3b). We fabricated ACD devices using HiPCO SWNT with various metal electrodes (Al, Ti, Ag, Cu, Ni, Au) and experimentally measured the closed-circuit current outputs (Rext = 1 kΩ) for these devices (Figure S3-4a). In order to exclude potential galvanic contribution to the observed current, we did a control study in which the contact between SWNT and the electrode metal were cut off ( Figure S3-4b). It is confirmed (from the equivalent circuit analysis (Fig. S3-5) that the galvanic effect contributes to less than 10% of the measured ACD current output.  SWNT network using polar aprotic organic solvents follow a convoluted Gaussian curve as a function of the solvents' lowest unoccupied molecular orbitals (LUMOs). This was explained as an electron transfer limited process, following a modified Marcus electron transfer model (see Section 3-5 for more details). In addition, the liquid should be a good solvent for the reaction that will be powered by the Janus particle. The dopant molecule has to be small enough to enter the inter-tube spaces in SWNT bundles. Overall, acetonitrile (CH3CN) is an appropriate solvent that satisfies all requirements well, and was hence chosen in our study.
Since we used hot-pressed HiPCO SWNT networks following a previously reported procedure 3 to fabricate the Janus polymer coated SWNT particles in the present study, these particles were tested 13 in a selected group of common organic solvents (selected from the same reported study) for voltage production in order to assess their abilities to produce self-generated voltage for electrochemical transformations when immersed in these solvent environments. We measured open circuit output to access the maximal electrical potential (and hence estimate the over potential available for electrochemical reactions) available in these devices, closed circuit measurements are also performed for a subset of the configurations tested, as shown in the main text and subsequent sections.
Using pristine (unoxidized, 10 % oxygen atom concentration by XPS characterization, see Section Voltage output through CH3CN doping on SWNT/polymer particle devices was tuned by the oxidation level of SWNT used for particle fabrication, experimentally tuned using acid treatment and pyrolysis. The acid treatment can increase the oxygen moieties and pyrolysis can remove them from SWNT surface. It is in principle possible to multiply the voltage of the particles by connecting them in series. However, this might have to be done by connecting several particles in separate solvent baths through external circuits, just like connecting enclosed batteries in series. We have preliminary data suggesting this is indeed possible, which we will reserve as the focus for follow up studies. For this study, the ability to tune the generated voltage of individual particles by varying the oxygen content is very convenient.
Acid treatment was done by soaking the SWNT powder into the mixture of sulfuric acid / nitric acid. We controlled the degree of modification using the concentration of acid solution. Then samples were transferred into closed vials which were filled with nitrogen gas in order to prevent re-oxidation.

Oxygen atomic concentration was investigated by XPS (ULVAC-PHI, INC. PHI VersaProbe II)
with a monochromated Al Kα source. The survey scans of samples with different oxygen content.
By the techniques of the acid treatment and pyrolysis, the oxygen content can be changed effectively from 4% to 27%. The oxygen content value was calculated as an atomic percentage from the integration of O1s peak and that of C1s peak in the high-resolution narrow scans in XPS ( Fig. S3-7).
Example XPS scans of o-SWNT/Nafion Janus particle. Top spectrum is for the Nafion coated half and the bottom spectra is for exposed bare oxidized SWNT. Inset is an optical image of the sample taken prior to the XPS scans. For the exposed oxidized SWNT side of the particles, we also found a linear correlation between the oxidation level of the o-SWNT with its sulfur atom content using XPS ( Fig. S3-8b), but not with nitrogen atom concentration ( Fig. S3-8a), indicating that the oxidized SWNT could bare functional groups (covalently bonded with SWNT carbon) such as sulfonates (R-SO3), etc. We also found a linear correlation between the oxidation level of the SWNT with its sulfur atom content using XPS ( Fig. S3-9b), but not with nitrogen atom concentration ( Fig. S3-9a), indicating that the oxidized SWNT could bare functional groups (covalently bonded with SWNT carbon) such as sulfonates (R-SO3), etc.
In terms of the open circuit voltage output of these devices, we noticed a super-linear dependence of maximum voltage reached by the asymmetric device (when inserted into CH3CN), as shown in Figure S3-3.3a. This trend was observed previously, and the reason for such increased voltage potential was attributed to the increased surface area accessible to SWNT for CH3CN adsorption.
The correlation between the SWNT oxidation level and voltage output (of the device) highlights the ability to tune to a desired voltage by dialing in a single material parameter (oxygen atom content).
We have found that the voltage creation comes from the gap of Fermi-level between doped and undoped sides. If the surface area at specific volume is larger, more solvent molecules can deprive more electrons from CNT (more electron is withdrawn per carbon in the CNT lattice), this in turn creates a larger Fermi-level gap and a larger voltage.

3-4. Electrical characterizations of asymmetric o-SWNT devices at different aspect ratios
We characterize device power outputs as a function of their aspect ratios (defined as the ratio between particle cross-sectional area and its axial length, perpendicular to the Janus faces, Fig. 3-10b). During the fabrication of such Janus o-SWNT particles, individual SWNTs are physically compressed into solid sheets with chosen thickness, as well as packing density and then, diced into rectangular particles with a laser beam ( Fig. 3-10a), depending on the desired size and form factors.
Various external loads are tested to map out the power curves of each device ( Fig. 3-10c), indicating a linear enhancement in power output with increased particle cross-sectional area ( Fig.  The external resistance used in these closed circuit measurements are: 5000 kΩ, 2000 kΩ, 500 kΩ, 200 kΩ, 100 kΩ, and 10 kΩ, and voltage signals across the external resistors are measured as a function of time ( Fig. S3-11), where the peak voltages are used for power calculation. In all measurements, the voltage generation persists over several minutes. Note since the device internal resistance scales inversely with its aspect ratio (given the same device volume), the linear scaling of the maximum device power output as a function of device aspect ratio points towards a fixed voltage generation mechanism with current output inversely scaling with the device's internal resistance. The negative voltage polarity is due to the connectivity of the Janus particle to the oscilloscope. Negative voltage response corresponds to exposed o-SWNT connected with the positive electrode, and vice versa.

3-5. Electron transfer model on Asymmetric Chemical Doping voltage generation
We note that this work aims to provide mechanistic insights of a liquid-solid doping phenomenon (e.g., CH3CN-SWNT), which involves an electron transfer (ET) process from SWNT to the liquid dopant(s). The resulted species (a radical CH3CN anion and a hole (h + ), which presumably delocalizes in the SWNT lattice) form a tightly bound electron-hole pair due to Coulombic interactions (much like that of an exciton), instead of a free radical species (e.g., CH3CN -) that roams around in the solvent. This actually is how the origin of the term "Asymmetric Chemical Doping" comes about -for we envision this to be analogous to the process in which an intercalated boron atom p-dopes a silicon lattice (where the free electrons in the Si lattice migrate and localize in the empty p orbitals of the boron atoms). Only, in our case, we use a liquid p-dopant as opposed to a solid-state material that intercalates in the material that is being doped. However, as is the case for boron in silicon, which cannot leave its interstitial location bearing the attracted free electron (as a B -), the CH3CNshould not, in principle, leave the SWNT surface without returning the electron. As we will show, the fact that this radical anion never leaves the SWNT surface actually plays an important role in the overall energy (or current) generation mechanism. Of course, subsequent electron transfers between "doped" SWNTs (or as the reviewer described "wet SWNTs") and "undoped" SWNTs are certainly warranted. In fact, this type of ET is instrumental for the observation of an electrical potential between the two electrodes. To be specific, the very reason any ET would occur between a "wet" SWNT and "dry" SWNT is due to the electron withdrawing of the p-type liquid dopant from the SWNT in the first place, and this "electron withdrawing" process is seen to be rate-limiting and can be described by a generalized Marcustype ET theory.
We first summarize the detailed molecular picture towards the full energy generation process ( Figure S3-12). We will use CH3CN as the example p-dopant, without loss of generality. All details of this tis mechanism described below are now included in the revised manuscript.  The DBE acetonitrile anion CH3CNcan be produced by Rydberg electron transfer 4,5 or by relaxation of charge-transfer-to-solvent (CTTS) excited states of a binary iodide-acetonitrile complex I -(CH3CN). 6 The DBE acetonitrile dimer anion (CH3CN)2has also been synthesized by photoexcitation and CTTS relaxation of the ternary iodide-acetonitrile cluster I -(CH3CN)2, and the resulting dimer anion was postulated to have a linear head-to-tail structure NCCH3---NCCH3 -. The lowest unfilled valence π*(C-N) orbitals of acetonitrile are relatively high in energy (ca. 2.8 eV higher than the highest occupied molecular orbital), 7 and the VBE acetonitrile radical-anion CH3CNexists only as a metastable species in the gas phase. 8 Experimental studies of γ-or X-ray irradiated solid 9,10 and liquid 11,12 acetonitrile suggest that, in a polar medium, the CH3CN molecules bind an excess electron into valence orbitals. Two distinct species can be produced in solid acetonitrile, depending on the crystal structure: the (CH3CN)2dimer is formed in αacetonitrile, whereas the monomeric radical anion (CH3CN)is formed in β-acetonitrile. 13 Since CH3CNreadily reacts with a neutral CH3CN molecule to form (CH3CN)2 -, it can only be observed in solid β-acetonitrile, whose crystal structure precludes dimerization. In liquid acetonitrile, an excess electron may exist as either a VBE in (CH3CN)2like in solid acetonitrile or a classical solvated electron in dynamic equilibrium. The latter is a separate entity in a solvent cavity, which is stabilized by the aggregate field of the solvent and can be considered as a condensed-phase analog of the DBE. Photoelectron spectroscopy studies of negatively charged acetonitrile clusters (CH3CN)nhave shown that the DBE and VBE forms coexist for cluster with n = 11-100, but for clusters with n >=13, the VBE form prevails. 14,15 DFT calculations suggest that, for n = 4-6, the VBE form becomes thermodynamically stable. 16 Interestingly, the experimentally observed addition of a hydrogen atom to an acetonitrile molecule inside water cluster anions has been proposed to involve transfer of the excess electron to the acetonitrile molecule and subsequent reaction of CH3CNwith a water molecule. 17 More recently, researchers have started to examine the ultrafast relaxation dynamics of excess electrons injected into liquid acetonitrile. 18,19 Unlike water, when CH3CN molecules bend, their electron affinity increases, so that a molecular anion can be stabilized in which the excess electron forms a covalent bond between the cyano-carbons of two bent, antiparallel CH3CN molecules.
Thus, when excess electrons are introduced into liquid CH3CN, two species are formed. 20,21 One of these species has an absorption spectrum in the near-IR that is identical to that of solvated electrons in solvents with similar polarity, 22 and this species has been assigned to be a typical DBE.
The assignment of the other species, which absorbs weakly in the visible region of the spectrum, has been controversial, but current consensus suggests that this entity is a solvent-stabilized VBE (CH3CN)2 -. Time-resolved photoelectron spectroscopy experiments by Neumark and co-workers found a greater population of the weakly bound species for small (CH3CN)nclusters and more of the deeply bound species in larger clusters. [23][24][25] This brief literature review suggests that the acetonitrile radical anion does form quite readily in liquid form, and CH3CN could potentially be considered as an electron sink for SWNTs, which certainly plays a key part of our proposed mechanism.
To probe the existence of this "unpaired" electron in the doped SWNT sample, we performed  We summarize the preliminary results of the EPR analyses here (Figure S3-14). We feel these data strongly support our proposed mechanism of the generated CH3CNradical anion Basically, at room temperature, there exists a broad EPR signal for only experiment C), which is shortly (< 30 s) after CH3CN is injected into the HiPCO-SWNTs. We attribute the broad signal to (i) this experiment being performed at a relatively high temperature (298 K), (ii) SWNT powder being a strong microwave absorber, resulting in insufficient sample tuning before the experiment, and perhaps most importantly (iii) there being a plethora of SWNTs with different chiralities and orientations, and therefore, different kinds of CH3CN -/(n, m)-SWNT pairs, as seen by the EPR producing a broad signal.   by Raman spectroscopy). This is another strong indication that the SWNT/CH3CN doping system does not generate any free radical in solution that is potentially useful for chain polymerization.
The dissociation mechanism of the CH3CNradical anion has been deeply investigated experimentallyever since it was first discovered. As early as the 1968 Nature report, it has been established that the photo-excitation of the (CH3CN)2radical anion can be readily reversed via thermal means. 13 Later, researchers established that both the (CH3CN)2radical anion and the multimer anion (solvated electron in acetonitrile) can undergo photo-excited fragmentation back to its neutral state. 11 It has been reaffirmed that the dipole-bound radical anion species can lose the negative charge and return to the bulk solvent, both through thermal and photo-excited means.
We launched most of the experiments in dark condition as much as possible to avoid unexpected effect of ambient light. The ferrocene oxidation was done in wells covered by aluminum foil, as well as experiments in Figure 3. The reaction systems were only briefly exposed to light when taking photos and doing UV-vis characterization. To test whether the tightly-bound CH3CN -/SWNT pair can dissociate with visible light irradiation, we performed the following dark room short-circuit measurement: while the SWNT flake is generating a short-circuit current inside a dark chamber, we pulse laser in the visible/UV range to the flake, and see if any photo-induced current is generated (Figure S3-17). Using a tunable filter, we were able to scan through a large range of excitation wavelengths. We programmed the laser such that it shines onto the SWNT sample for 30 s then off for 30 s before it switches to another wavelengthand repeats the process. The results we obtained are very interesting (Figure S3-18, S3-19). First of all, we note that the current being generated by the Asymmetric Chemical Doping (ACD) process is around 0.2 µA, and the pulsatile laser irradiation has caused local perturbation to the generated current, leading to the "saw-toothed" pattern ( Figure S3-18). If we assign the ACD generated current as the baseline current, and subtract that out from the combined measurement, we obtain the pure "photonic" contribution to the current generation process (Figure S3-19). Even though detailed analysis needs to be performed for this photo-current to yield quantitative information of our system, it is obvious that photo-excitation can result in perturbations of the current generation process, which we attribute to the increased rate of CH3CN -/SWNT pair dissociation.

Validation of equivalency between large and small particles on the experiment of Ferrocene oxidation
In standard particle fabrication, we compress vacuum dried o-SWNTs in a hot press (at 50 °C) with a polymer backing to yield a Janus sheet of polymer protected o-SWNT (Fig. S4-1a). Further dicing of the sheet into rectangular shape gives rise to Janus particles of different aspect ratios. The depth of the particles (or the o-SWNT/polymer sheets) are controlled via packing density of o-SWNT prior to the hot-pressing step. We can dice the particles to small pieces commensurate to the width of a human hair (Fig. S4-1b). These smaller particles (volume ≈ 360 µm (length) × 200 µm (width) ×100 µm (depth) = 7.2 × 10 6 µm 3 ) exhibit the same SWNT and polymer micro structures seen in their larger counterparts (4 mm 3 volume, Fig. S4-1c). When subjected to the same reaction conditions with TBAP as electrolyte and Ferrocene in CH3CN as solvent, the smaller particles (560) works just as well as one larger particle (same overall o-SWNT volume) in turning over the oxidation reaction ( Fig. S4-1 d, e). It is worth noting that the smaller particles appear to be have faster kinetics of oxidation of ferrocene to ferrocenium (Fig. S4-1e), and this can be attributed to the increased overall surface area for the smaller particles than the larger one. Overall the yield of the reaction is quantitative for both cases.

5-1. Assessment of the role of o-SWNT surface catalysis
Detailed kinetic analyses reveal that the initial rate of the electrochemical process is limited by ferrocene adsorption on the anodic (exposed) o-SWNT surface of the Janus particles. We used UV-Vis kinetic data to back out the ferrocene oxidation reaction mechanism (a catalytic process on the o-SWNT surface). We considered a three-step reactant (ferrocene) adsorption -surface reaction -product (ferrocenium) desorption model and use experimental means to decide which step is rate limiting and use subsequent fittings to elucidate kinetic parameters.
Without the presence of a catalyst, the ferrocene ↔ ferrocenium redox pair is an equilibrium reaction. We first asked the question whether a catalytic model even necessary? In other words, can the measured kinetic data be described by a simple 1 st order forward and backward reaction?
Suppose A is ferrocene, B is ferrocenium, the measurement we have here is [B] as a function of t: 1 1 0

with forward and backward rate constant k and k and [ ] [ ] [ ]
The solution for this reaction system is (with the initial In this expression, there is only a single fitting parameter β, with all the other ones easily measured or calculated. We plotted the kinetic data (blue circles) and used MATLAB to perform least square customized equation fitting (Fig. S5-1). (2-parameter fit), and a one-parameter fit that follows the derived equation exactly. As we can see, while both the three-and two-parameter exponentials fit the kinetics data fairly well, the one-parameter fit is very poor and cannot describe the measured kinetics. This analysis in a good indicator that the mechanism of Janus particle assisted ferroceneoxidation is a Janus particle assisted process and cannot be described without considering the catalytic surface of the o-SWNTs that make up the Janus particle. If we define equilibrium constants KA, KS, and KD the same way we did Keq, the rate of these three equilibrium rates can be express as: At steady state, there is no accumulation of reaction species on the anodic surface, that is Here we consider three scenarios that takes each of these steps as rate limiting (while the other two are still in equilibrium). The motivation is that we know from prior analysis that these steps cannot all be in equilibrium with each other (which would have recovered a single parameter exponential concentration profile.
If adsorption if rate limiting, kA is small compared to kS and kD, therefore 0  This means that if we plot the initial rate as a function of the reactant initial concentration, an adsorption limited mechanism would yield a linear dependence.
If, on the other hand, the surface reaction is rate limiting, similar analysis yields: In other words if we plot the initial rate as a function of the reactant initial concentration, an adsorption limited mechanism would yield a plateaued increase dependence.
Lastly, for the desorption limiting scenario, the analysis yields: In other words if we plot the initial rate as a function of the reactant initial concentration, an desorption limited mechanism would yield no dependence of the initial rate on the initial reactant concentration.

5-3. Experimental validation of rate law of ferrocene electro-oxidation
In the previous section we developed an analytical model of how the initial rate of ferrocene electro-oxidation would depend on the initial ferrocene concentration. In this section, we used UV-Vis spectrometer to perform the initial rate kinetic studies (Fig. S5-2a). Different concentrations of ferrocene were prepared, so were the Janus o-SWNT/polymer particles. The kinetics measurements were performed as soon as the baking of the particles are done. Samples were kept under innert (N2) environment for maximum reproducibility. Before each measurement, we recorrect the UV-Vis baseline using pure CH3CN, scan the whole spectrometer range to locate the forrocene and ferrocenium peaks. We the initial rate measurements, we take 622 nm (ferrocenium wavelength of the absorption peak) absorbance repeatedly at 2 Hz. We always start spectrum collection before inserting the particle into the quartz cuvette. For measurements over two minutes, the cuvette is covered with a plastic film to prevent evaporation of solvent evaporation.
As seen from the initial rate data as a function of the initial reactant concentration (Fig. S5-2b), the initial rate follows a linear trajectory within the concentration window of interest. This indicates that the o-SWNT Janus particle assisted ferrocene oxidation is rate limited by the adsorption of ferrocene onto the anodic o-SWNT surface. This experiments further indicates that the reaction rate can be accelerated by increasing surface are of the particles (and hence its accessibility to ferrocene). This finding supports our explanation to the observation (Section 4) that smaller Janus particles with equivalent volume (but increased overall surface area) perform the oxidation of ferrocene towards the same yield but at faster rates.  (Fig. S6-1a).

6-2. Polymer coating on SWNTs network
Half of the o-SWNT network was coated by PVA to prevent acetonitrile access, using dip coating method with PVA/water solution. PFMMA was dissolved to THF, and then Janus o-SWNT/PVA network was coated with PFMMA solution using dip coating method and dried at room temperature. The dip coating -drying process was repeated 3 times (Fig. S6-1b).

6-3. Characterization of PFMMA oxidation state
To further explore the detailed mechanism of the ferrocene-to-o-SWNT electron transfer, and how it couples with the driving ACD process, we investigate on which side of the o-SWNT/PVA Janus particle does the ferrocene oxidation occur. This was probed by grafting a thin layer of polyferrocenylmethyl-methacrylate (PFMMA) over the entire particle surface and subjecting the PFMMA-grafted o-SWNT/PVA into a CH3CN/electrolyte solution. Since the only ferrocene molecules available for electrochemical oxidation are covalently linked to the PFMMA backbone, they are spatially fixed to the particle surface.
We took advantage of the spatial resolution of X-ray Photoelectron Spectroscopy (XPS) to analyze the PFMMA coated Janus o-SWNT/PVA particles after immersion of the particle into CH3CN for 30 minutes (Fig. S6-2a). As always, extensive baking (25 minutes at 200 °C) was performed prior to each experiment to avoid any acid contamination (e.g., H2SO4, HNO3) to the o-SWNT assisted oxidation of ferrocene moieties fixed spatially in PFMMA.
We first performed a control study, in which PFMMA (Fe 2+ ) was chemically oxidized to the oxidized (Fe 3+ ) version of PFMMA ("PFMMA-ox") using HNO3. Both PFMMA (blue) and PFMMA-ox (red) were examined under XPS (Fig. S6-2b). We temporarily assign the 706 eV binding energy peak to the Fe(II)2p3/2 orbital in PFMMA and the peak around 718 eV to the Fe(II)2p1/2 orbital ( Fig. S6-2b, blue). The XPS scan for the PFMMA-ox sample, on the other hand, returns a much broadened spectrum with less pronounced features ( Fig. S6-2b, red). When compared to the reduced PFMMA sample, however, the lack of the previously identified Fe(II)2p3/2 orbital peak centered around 706 eV is conspicuous.
With this in mind, we took again a survey scan of the post reaction PFMMA coated Janus o-SWNT/PVA particle with XPS and present the spectrum collected on the exposed anodic o-SWNT side (red) as well as that of the PVA protected side (blue) in Figure S6-2c. While both scans appear to be less clean with respect to the control samples, we do observe a 705.5 eV peak for the PVA protected (cathodic) side PFMMA, suggesting perhaps the presence of Fe(II) species in this region.
Notably the peak is right-shifted a little when compared to the control sample, potentially due to the conductive carbon layer (o-SWNT) underneath. This 705.5 eV peak, however, is notably absent from the PMFFA spectrum the exposed (anodic) o-SWNT side, suggesting the lack of Fe(II) species. Instead, the broad peak with higher binding energy than the assigned Fe(II)2p3/2 orbital is reminiscent of the Fe(III) feature seen in the PFMMA-ox control sample ( Fig. S6-2c, red).
Admittedly, the XPS analyses did not provide conclusive evidence that the oxidation of PFMMA to PFMMA-ox occurred exclusively on the exposed anodic o-SWNT side of the Janus particle, it however did provide support for the idea that the Janus carbon particle resulted in a spatial asymmetry for the oxidation state of the Fe in the PFMMA layer coated outside of the particle. Janus o-SWNT/PVA particle after immersed (reacted) in CH3CN for 30 minutes. The PVA side is blue, and the exposed o-SWNT side is red.
In the next set of experiments, we used UV-Vis spectrometry to identify the oxidation state for PFMMA and correlated the PFMMA oxidation state as a function of its location on the Janus o-SWNT/PVA particle. Because UV-Vis measures bulk chemical information without high spatial resolution, we physically cut the PFMMA coated Janus o-SWNT/PVA particle in half (the PVA cathodic half and the exposed anodic half) after its reaction in CH3CN, extract the PFMMA polymer from the surface of the particle using THF for both sides separately, and analyze their respective oxidation states under UV-Vis (Fig. S6-3). As a negative control, we also performed the same particle separation (into two halves), THF extraction, and UV-Vis measurement for PFMMA coated Janus particles that did not go through the CH3CN immersion (hence ACD mediated electricity generation and subsequent electro-oxidation reaction of ferrocene). We first collected the UV-Vis spectra for the as prepared PFMMA (Fe 2+ ) and the chemically oxidized PFMMA-ox (Fe 3+ ) (Figure S6-4a). The absorption peak centered around 447 nm wavelength for the reduced PFMMA (Fe 2+ ) is recognized as its spectrometric feature (Figure S6-4a, red), which is notably absent from the oxidized PFMMA-ox (Fe 3+ ) sample ( Figure S6-4a, blue). For the PFMMA extracted from the negative control particle, both the PVA protected and the exposed side exhibit the reduced form of PFMMA ( Figure S6-4b), which is not surprising, because without CH3CN, no electrical potential will be generated across the Janus o-SWNT/PVA particle, and hence there is no overpotential that drives the thermodynamically unfavorable ferrocene oxidation process. This can also support the electricity mediated oxidation mechanism, for without CH3CN, a chemically mediated oxidation that directly oxidize the ferrocene moieties in PFMMA should still be possible (and since in our control, the PFMMA coated particle are put in THF to extract the polymer, the lack of solvent to facilitate a direct chemical oxidation should also not be the problem. Also note, THF is known to not generate ACD voltage with o-SWNT, highlighting once more the important role an appropriate molecular dopant such as CH3CN plays in this electro-oxidation reaction. After the control group, we characterized particles in the experimental group that was immersed into CH3CN for 30 minutes (Figure S6-4c). Upon reaction, we dry the CH3CN off, divide the Janus particle into the PVA protected half and exposed (bare) half (Fig. S6-4c), and extract the PFMMA coated on each half with THF ( Figure S6-4d). It is evident from the color of the THF solution that the PFMMA coated on the PVA protected (cathodic) half is in its reduced form (yellow solution), whereas the exposed o-SWNT (anodic) half has lost that yellow color. This can also be seen in the UV-Vis spectra with a corresponding reduction in the 447 nm absorption peak previously assigned to the reduced PFMMA (Fe 2+ ) species (Figure S6-4f), a strong indication of oxidation of PFMMA only occurs on the anodic side of the Janus o-SWNT/PVA particle.
Both the XPS and UV-Vis post-reaction analysis reveals that only the ferrocene molecules bound to the bare o-SWNT side were oxidized into ferrocenium, leaving those on the PVA protected side mostly in their original reduced state. This reaffirms the molecular picture that ferrocene exclusively oxidizes on the electron deficient, unprotected o-SWNT surface (i.e., the anode), driven by the ACD process that lowered the EF on that side. An interesting corollary of this mechanism states that the electrochemical redox reaction consumes the ACD generated electron flow, and necessarily reduces the solvent-powered closed circuit current, which we explore in the next section.

Keythley measurement with changing ferrocene concentration
Short circuit current was measured using Keythley Current Meter with changing ferrocene concentration. Basic setup we used is shown in Figure S7-1a. Keytheley Current Meter was connected with Janus particle in series, which was soaked in the prepared MeCN solutions with different concentration of ferrocene from 0 to 20 mM. The resulted current trend in Figure S7-1c, d clearly shows that the short circuit current through the Janus particle decreases as concentration of ferrocene increases. This indicates that the short circuit current resulted from the MeCN doping is a measure of free electron available to perform the chemical reaction as shown in Figure S7-1b, and its magnitude directly competes with the reaction rate.
We believe the reduced current in the external circuit should be completely due to the increased rate of redox reaction. Any continuous DC current flowing through an electrolyte must be Faradic, and must correspond to redox reaction on electrodes. The only possible non-Faradic DC current is the capacitive current in the electrical double layer, which is minimum in long time.
Therefore the reduced current in Fig. 3a is almost exclusively due to increased consumption from redox reaction.  Table S8.1., was added to prepare the solution for CV measurement. The measurement was scanned by 50 mV/s, using glassy carbon electrode and Ag/AgCl electrode for working and reference electrode, respectively. Figure S8.1(a) shows the result of CV on Ferrocene (reference), and oxidation potential was determined as 380 mV from its oxidation peak on CV curve. Since Janus particle prepared in this study typically has 500 mV voltage in MeCN, this 380 mV of ferrocene oxidation potential implies it can successfully oxidize ferrocene in MeCN. Oxidation potentials of 9 different Ferrocene derivatives were obtained by same method and are listed in Table S8-1.  Half of the o-SWNT network was coated by Nafion to prevent acetonitrile access, using dip coating method with Nafion solution (as purchased from Sigma Aldrich) at room temperature.
The dip coating -drying process was repeated 3 times (Same method with the procedure in S6-2).

9-2. Reaction
Half-coated o-SWNT network was soaked in 3 kinds of metal salts (shown in Figure S9-1) in acetonitrile with TBAP(50mM) to launch reactions for 1 hour. During the reactions, metal salts were reduced, as shown in Figure 3 c) in the main text, proposed to be adsorbed on the Nafioncoated side of SWNT.

9-3. XPS analysis
In order to identify where actually reduced metal salts are, metal atomic concentrations on both sides of o-SWNT and Nafion-coated SWNT were checked using XPS (ULVAC-PHI, INC. PHI VersaProbe II) with a monochromated Al Kα source. The reduced metal salts content was calculated as an atomic percentage from the integration of Cu 2p peak, Co 2p peak, and Ag 3d peak, respectively in the high-resolution scans. As shown in Figure S9-1, all of three metal salts shows higher concentration on the Nafion-side, which reduction of metal salts occur as explained in the main text. These data strongly support our hypothesis where reduction and oxidation occur on the half-coated SWNT.

9-4. SEM analysis
The half-coated SWNT was imaged using a Zeiss Merlin field emission scanning electron microscopy (SEM) and elementally analyzed using an Energy Dispersive X-ray Spectroscopy (EDS). As shown in Figure S9-2, there appeared to be a brighter fiber structure on the upper side due to the Nafion-coating, while there is darker fiber structure on the lower side with bare o-SWNT network. On the bare o-SWNT side, several brighter particles are observed as shown in shown in the elemental maps in Figure S9-2c, d. Figure S9-2b indicates that this region reduced more copper due to large amount of Nafion, compared to the rest of portion. Thus, these SEM and EDS analyses on the half-coated SWNT elucidate our hypothesis where reduction and oxidation occur furthermore.

Procedure of resistance measurement of polymer-coated SWNT
The half-polymer-coated SWNT for the resistance measurement was prepared as followings.
Firstly, half portion of the cut SWNTs network was dipped into polymer solution and dry. This procedure is repeated several times to make sure enough thick to prevent MeCN penetration. Then, the edge of polymer coating side is cut diagonally in the thickness direction by razor and crosssectional naked SWNT is placed on the copper electrode to make a good contact between them.
The other bare side, without any coating, of SWNT is also placed on the copper electrode, bridging across two copper electrodes then were fixed with another copper.
Resistance across SWNT was measured through copper electrodes by multi-tester with twoprobe method. 11. Hypothesis on mechanism of long-lasting voltage creation in the reaction solution.
Although ADC voltage creation lasts only for a few seconds when electrons are consumed as an electrical current doing some work, ie., flow the external circuit, observed electrochemical reaction, in the main text, lasts for 30 minutes or more. In order to understand this long-lasting phenomenon, we propose a hypothesis to give an adequate explanation for this long-lasting reaction phenomenon here.
At a first process, as we propose in the previous works, voltage creation occurs due to the asymmetric MeCN doping on SWNT which creates Fermi level gap in SWNT, leading the potential bias across the SWNT (Figure S11-1a) with negative voltage on the bare SWNT side.
Then secondly, this voltage makes chemicals reacted electrochemically as shown in Figure S11-1b. Since this reaction puts electrons on the bare SWNT side and takes electrons from PTFE SWNT side which makes voltage decrease, these chemical reactions should have occurred quickly and voltage disappeared in short time without extra mechanism to sustain voltage.
However, in our experiment, most of the reactions went for a long time more than 30 minutes, so here we propose one additional mechanism concomitantly from our ADC scheme in electrochemical reactions written in the main text.
As shown in Figure S11-1b, once Ferrocene oxidation occurs on the bare side of SWNT, counter reaction, hydrogen generation, occurs on the other side as well. When this counter reaction goes further, there creates a COOH/COOgradient across the SWNT as shown Figure   S11-1b. Liu et al., have showed a long-lasting current generation using protein and gold electrodes. In their work, they proposed that the water gradient from surface to bottom, created due to its asymmetric structure, brings the proton gradient due to the ionization of carboxylic acid and concomitant diffusion of proton, which drives a long-lasting electric current on their system. Similar phenomenon should happen in our system, because there should be a COOH/COOgradient after a while reaction goes as shown in Figure S11-1b. This COOH/COOgradient promote a proton diffusion across the SWNT through hopping of proton on COOH.
During the proton diffusion, bare SWNT (left) side of Figure