2.1 Fluid Production

Due to the cost prohibitive nature of purchasing Galinstan (>$5.60/g or ~$36/mL), it was decided to produce the alloy from the individual metals^[7]. Individual metals at 99.99 pct purity were purchased with the total metal cost coming to approximately $2500 to produce slightly over 1L of fluid.

The metals were received as an ingot and, to allow for easier composition control, the indium and tin were pelletized. To pelletize each metal, the ingot was placed in a fused quartz crucible and then heated in a glass reaction vessel within a heating mantle to 175 °C and 300 °C for the indium and tin, respectively. Temperature was monitored using a Type-K thermocouple placed outside of the crucible at roughly half the crucible height but inside the glass reaction vessel. As these temperatures are above the melting point for their respective metals, a relatively short melting time was needed for the ingots to be fully melted. Once melted, the liquid was poured through a stainless-steel kitchen strainer into a water-ice bath. The strainer was held slightly above the water surface so as to produce small pellets without the liquid metal solidifying on the screen. Pellets were then filtered in a vacuum assisted Buchner funnel and stored in HDPE lab bottles in a desiccator.

For fluid production, the gallium ingots were melted in their plastic shipping bottles in a drying oven and then poured into a clean fused quartz crucible. Based on the gallium mass, the required indium and tin masses were determined and added to the crucible. The entire crucible was then heated to 175 °C and held for 2hrs to homogenize the alloy followed by cooling to room temperature. A slag layer formed on the melt, likely from oxidation of the indium and tin during the pelletization process and this was skimmed off before the liquid alloy was poured into HDPE lab bottles.

2.2 Density

The fluid density was determined using a predetermined fluid volume and mass measurement. A 1mL insulin syringe with indications every 0.02mL was used to draw up 0.5mL of fluid. The syringe mass was measured both empty and after drawing up the fluid; a lint-free paper was used to wipe away any external fluid on the syringe. The mass difference between the empty and filled syringe was adjusted according to the air mass displaced by the fluid before use in density determinations. Air density was determined using the ideal gas law shown in equation (1), with temperature measurements taken using a Type-K thermocouple and pressure reported by the National Weather Service. In the ideal gas law, the air composition was assumed to be 79 pct N₂ and 21 pct O₂. In equation (1), ρ is the gas density, P the gas pressure, MW_air the molecular weight of air, which is assumed to be the weighted average of the molecular weight of O₂ and N₂, R the universal gas constant, and T is the temperature.

DI water, and N.4 viscosity standard were used to assess the validity of the measurement technique before measurement with the GaInSn alloy, and both have a known density or density relationship with temperature.

2.3 Viscosity

Viscosity was measured using an Ubbelohde size-0 viscometer manufactured by Cannon Instrument Company. Other approaches considered included a torque measurement in a Couette rheometer or with viscosity (flow) cups, however, a compatible spindle could not be found, nor a viscosity cup that matched the expected fluid viscosity. The small capillary in the viscometer used was found to plug easily when testing the GaInSn alloy; this was believed to be due to oxide formation. Hydrochloric acid was originally used to clean the viscometer following this capillary plugging, but after observing the effectiveness of sulfuric acid, it was used for treatment instead with a drop placed on top of the alloy to prevent oxidation during testing. While this did increase the hydrostatic pressure on the liquid metal, due to the large density difference between the alloy and acid, it is not believed this significantly altered the determined viscosity. Measurements without an acid cover were initially attempted however the capillary plugging due to oxide formation prevented both accurate measurement and comparison with results obtained using an acid cover.

The presence of acid over the alloy is believed to have some influence over the alloy wetting on the viscometer walls, particularly in the vicinity of the alloy meniscus. To mitigate the effect of the acid on the measured viscosity, the acid and alloy were drawn above the viscometer measurement indication. This allowed for a stable acid-alloy interface to form before timing commenced. Due to the density difference between the alloy and acid, and in-situ optical observation, it is believed the acid was primarily present at the previous alloy-atmosphere interface and therefore did not affect wettability within the capillary itself. This theory was supported by optical observation of the alloy-viscometer interface in the initial, un-covered, and later, acid-covered, cases with no difference detected. Although meniscus behavior was observed to change following acid treatment, as the timing marks were located well above the capillary entrance, this change in behavior had minimal influence on the result since the only requirement is that the meniscus maintain a consistent shape for the duration of the test.

Viscosity was determined using equation (2) with the viscometer constant being given by the manufacturer and confirmed with the N.4 viscosity standard. In equation (2), ν is the kinematic viscosity while C is the viscometer constant (9.746*10^-4 cSt/s), and t is the time required for the fluid meniscus to descend from the upper to lower timing mark.

2.4 Electrical Conductivity

An improvised 4-point probe constructed of tungsten electrodes and a potentiostat (Gamry Interface 1000e) was used to measure electrical conductivity. The 4-point probe approach is a commonly used technique for liquid metal electrical conductivity measurements while tungsten electrodes were chosen for their known stability with the gallium alloys^[8-10]. The electrodes were forced through a rubber stopper to ensure a consistent geometry and the portion of the electrodes on the opposite side of the stopper as the fluid were electrically isolated from each other using electrical tape and connected to the potentiostat cables with insulated copper wire. A glass vial was used to hold the fluid during testing, and the fluid volume was fixed to ensure the same electrode area in contact with the fluid. A picture of the setup, disconnected from the potentiostat, is shown in Fig. 1. Once the probe was connected to the potentiostat and the fluid added to the vial, the setup was placed inside a Faraday cage.

Fig. 1. The constructed 4-point probe used for liquid metal electrical conductivity measurements.

The linear sweep voltammetry (LSV) program was run from the Gamry potentiostat controller to measure the electrical resistance. To calculate the geometric factor for the cell, which would enable conversion of the cell conductance to fluid electrical conductivity, different fluids with known electrical conductivity were tested. DI water was first tried followed by sulfuric acid though it was found that aqueous systems reacted with the tungsten electrodes as seen in Figure 2. From the color of the electrode, it is assumed that WO₂ was formed on one electrode when sulfuric acid was used while WO_2.9 or WO_2.722 was formed on the other electrode with both the DI and sulfuric acid systems^[11].

Fig. 2. Photos of tungsten electrodes of the 4-point probe after calibration in (a) a DI water and (b) a sulfuric acid solution.

A proposed mechanism for the formation of WO_2.9 at the anode is shown in equations (3) and (4). The formation of WO_2.722 follows a mechanism similar to that for a different O₂ to W ratio in equation (4), reflecting the different stoichiometry of the oxide.

The difference in discoloration between the DI and sulfuric acid tests suggest that different phases and possibly reaction mechanisms occurred. The Pourbaix diagram for the tungsten system at different tungsten ion concentrations, in Fig. 3.a and Fig. 3.b, helps to explain this difference. For the DI system, the electrode is initially converted to soluble WO₄ ^2- by equation (4.17) thereby leaving only tungsten on the electrode surface. As the applied voltage increases, more WO₄ ^2- forms shifting the WO₄ ^2- stability region to the right, transitioning as the tungsten concentration increase changes, from Fig. 3.a to Fig. 3.b. In this manner, the tungsten is able to transition to WO_2.9 without producing intermediate oxides.

Fig. 3. Pourbaix diagram for the W-H₂O system at 294.15K and a tungsten ion concentration of (a) 10^-6 M and (b) 2.5 M which is near the solubility limit for Na₂WO₄ in water at 294.15K^[12]. Dashed lines indicate the water stability region.

Conversely, in sulfuric acid, WO_2.722 is produced as a film on the electrode surface. When the sign of the applied cell potential flips from negative to positive, the WO_2.722 is able to react according to equations (4.18) and (4.19) for the sulfuric acid system, while equation (4.19) does not occur in the DI system due to the lack of tungsten oxide on the surface, and the formed WO₄ ^2- diffuses into the solution. The DI water was also scanned over a larger voltage range than the sulfuric acid, -0.5 to 0.5 V vs -3*10^-5 to 3*10^-5 V, which may also explain the formation of WO_2.9 or WO_2.722 and the lack of WO₂ in the DI solution.

Given the reactivity of the electrodes in aqueous systems, the use of paraffin (mineral) oil was tested for calibration. This fluid did not display a linear current-voltage relationship and subsequent attempts, such as chronoamperometry and electrochemical impedance spectroscopy (EIS), did not illuminate the cause for this obscure behavior. Plots for the LSV, chronoamperometry, and EIS for the paraffin oil are shown in Figure 4.

Fig. 4. (a) Linear sweep voltammetry, (b) chronoamperometry, and (c) electrochemical impedance spectroscopy for paraffin oil when evaluated at room temperature using the constructed 4-point probe with tungsten electrodes.

Lacking a linear current-voltage relationship meant that a single conductance value could not be determined and therefore additional fluids were explored to evaluate the cell geometric constant. A small volume, approximately 15mL, of GaInSn was purchased from an industrial vendor but this arrived heavily oxidized and could not be used to confirm the electrical performance of the produced alloy. Instead, Woods metal, Bi₅₀Pb₂₅Cd_12.5Sn_12.5 (wt. pct), was used as a calibration fluid. The melting temperature of Woods metal is 70 °C and so the cell had to be moved from the Faraday cage to the glass reaction vessel and a heating mantle used during fluid production^{[2, 13]}. Since the heating mantle did not fit within the Faraday cage, a grounded aluminum foil Faraday cage was employed to produce a similar effect. The electrical conductivity of Woods metal was determined via regression of the data in Verma et al. to yield equation (8) which had an R² value of 0.994; σ is the electrical conductivity in siemens per meter while T is the temperature in Kelvin^[2]. Using equation (8) and the conductance of the Woods metal cell, the geometric factor for the cell was determined. During all LSV measurements, currents exceeding ±1amp were discarded as ±1amp was the reported limit for the potentiostat used.

To assess whether the volume of fluid had a significant effect on measured conductivity, the fluid volume of the GaInSn alloy was varied between 4.8mL and 6.8mL and the conductivities were compared. It was also necessary to check whether the aluminum foil provided shielding comparable to the commercially available Faraday cage and so a comparison of the data for GaInSn was also made under these conditions.

2.5 Fluid Oxidation Treatment

The issue of fluid oxidation has been previously noted in literature and was observed during the viscosity testing, so a method to mitigate oxidation was deemed necessary^{[5, 14]}. Approaches explored in the literature include keeping the oxygen content around or below 1 ppm or potential treatment with hydrochloric acid^[15-17]. It was suggested that a possible treatment with sulfuric acid would be able to impact the level of oxidation provided the pH was < 3; this was the approach investigated here^[17]. 0.5mL of metal was placed in a glass vial with 1.0mL of acid at various concentrations as a cover fluid and left to age for approximately one month. A sample with DI water and another that was uncovered were used as reference points in the study. After one month, the droplets were analyzed via scanning electron microscope (SEM) energy dispersive X-ray spectroscopy (EDS). Metal for these tests was taken from the volume rather than surface of the alloy to avoid the possibility of including the aqueous phase, which could damage the SEM. The thickness of the oxide layer was not predicted and therefore a qualitative surface tension assessment was performed by optical inspection while laying the vial on its side before any fluid was withdrawn^[17-19].

2.6 Compatibility with PVC

In order to evaluate the stability of GaInSn with PVC, PVC tubing was cut and capped with a PVC end cap and partially filled with fluid. The surface of the PVC was visually inspected after one week and one month to observe if discoloration had occurred, which would serve as an indicator of corrosion. A mass-based approach was originally considered but the pipe and end cap were sealed with a PTFE (Teflon) spreadable paste with an acetone base which was not given proper time to cure before GaInSn was injected, leading to an unaccounted-for mass change.

3.1 Melting Temperature

Although not measured in this study, the literature disagrees on the melting temperature value (and other properties) of the GaInSn alloy composition used in this study, and similar alloys. Consequently, a brief discussion of the literature inconsistency around melting temperature is given here to inform latter discussions. Plevachuk et al. used Ga₆₇In_20.5Sn_12.5 (wt. pct) and reported a melting temperature of 283.7K (10.55 °C) which is in agreement with Dubovikova et al. (exact composition unspecified), and Wang et al. (Ga₆₈In₁₂Sn₂₀)^{[3, 21, 22]}. Guo et al. reported a melting temperature of 10.5 °C for a Ga₆₇In_20.5Sn_12.5 alloy while Afrin et al. suggested 11 °C for the same composition that was used in the present study^{[8, 23]}. Scharmann et al. did not specify their composition but reported a melting temperature of 11 °C^[14]. Sarfo et al. suggested a liquidus temperature of 11 °C and solidus of -19 °C for Ga_68.5In_21.5Sn₁₀ while Li et al. reported -19 °C as the melting temperature for the same alloy^{[24, 25]}. Cadwallader reported a freezing temperature of ~ -20 °C for Ga_66.0In_20.5Sn_13.5 in alignment with the solidus reported by Sarfo et al. though for a slightly different alloy^{[5, 24]}. Anwar et al. used the commercial alloy, Galinstan, produced by Geratherm Medical AG and reported a melting temperature of -19 °C while the company American Elements produces an alloy of unspecified composition (Ga₆₂In₂₂Sn₁₆, Ga₆₆In_20.5Sn_13.5, or Ga₆₅In₁₉Sn₁₆) that shares this melting temperature^{[26, 27]}. The metals supplier, RotoMetals, produces an alloy of the same composition as the one used in this study and reported a melting temperature range of -22 to 104 °C^[28].

From this literature array, it seems reasonable to suggest a melting temperature between 11 °C and -19 °C, though determining the precise temperature within this range would likely benefit from further study.

The alloy composition used in this study is reported to be the eutectic alloy and, as such, should have a converged liquidus and solidus temperature rather than a range. Impurities in alloy preparation could result in a slight shift of the eutectic composition or melting temperature. The observed range though the magnitude, 30 °C, seems large unless significant impurities were present in the alloys.

3.2 Density

The density of GaInSn was determined to be 6.488g/cm³. A singular literature value for the density was difficult to establish, since posted values disagreed. Plevachuk et al. proposed equation (9) for calculating the density^[3]. However, this was found to be a poor fit for the data presented in Plevachuk et al.’s own paper and so a regression was performed on their data which resulted in equation (10). In both equations (9) and (10), ρ is the alloy density in g/cm³ while T is the temperature in Kelvin. Based on these equations, a density of 6.57 (listed regression) or 6.35g/cm³ (regression performed) should be expected at 20 °C, the latter being in close agreement with Jian and Karcher, who used a Ga₆₈In₂₀Sn₁₂ alloy^[29]. Wang et al. likewise reported a value of 6.363 g/cm^{3 [22]}.

It is interesting to note that suppliers suggest the value should be closer to 6.4 g/cm³ at 20 °C which is close to the value observed^{[27, 28]}. Due to the crudeness of the method used to determine density, it was decided to conduct a propagation of uncertainty analysis. The formula for error propagation is given in equation (11) which, when applied to the density formula, equation (12), results is equation (13). For equation (11), σ_x is the uncertainty in property x while x_i is a variable going into the calculation of property x. σ_xi is the uncertainty in the measurement itself. Equation (12) has ρ as the density, Δm as the difference in mass between the empty syringe and fluid containing syringe, and V is the volume of fluid in the syringe (0.5mL). Variables in equation (13) with a bar over them are the average across multiple tests, STDEV is the standard deviation of the measurements, and σ_balance and σ_syringe are the balance and syringe readability respectively.

The results of this uncertainty analysis as well as the confidence interval are shown in Table 1. Although the literature value is outside of the 95 pct confidence interval, it does fall within the uncertainty bounds, which can be considered a better measure of accuracy, while the confidence interval is a measure of precision.

As a validation of the method for density analysis, measurements were also taken for the N.4 viscosity standard, used for calibration of the Ubbelohde viscometer, as this has a well-defined density, and DI water. A regression of the N.4 density data provided by the manufacturer yielded equation (14), R² = 0.99997; this equation was used to calculate the literature value reported in Table 2. For equation (14), ρ is the density in g/cm³ while T is the temperature in Celsius not Kelvin. Although the confidence interval around the N.4 standard density does not contain the literature prediction, the uncertainty bound does, as was the case for GaInSn.

For DI water, the measured density was 0.998g/cm³ with a confidence interval of 0.007g/cm³; the density of water taken from Koretskey predicts a value of 0.998g/cm^{3 [30]}.

3.3 Viscosity

Kinematic viscosity of the fluid was calculated using equation (2) with the literature value reported in Table 1 calculated from the dynamic viscosity and density reported by Plevachuk et al. which is for a Ga₆₇In_20.5Sn_12.5 alloy rather than the Ga_68.5In_21.5Sn₁₀ composition used in this study^[3]. The dynamic viscosity was converted to the kinematic viscosity via equation (15); ν is the kinematic viscosity, μ the dynamic viscosity, and ρ the density. Jian and Karcher reported a kinematic viscosity of 0.34cSt at 20 °C, Wang et al. measured 0.4cSt, and Khondoker and Sameoto measured 0.37cSt, suggesting some variability in the literature, although still below the lower bound of the uncertainty interval^{[17, 22, 29]}.

Unlike the density measurements, for the viscosity both the confidence interval and uncertainty bounds failed to include the literature value. A proposed explanation for this is the oxidation of GaInSn. Initially, large solid particles could be observed on the fluid surface, which caused the capillary of the viscometer to plug. A larger capillary could not be used because the expected viscosity was small, and this meant that a method to prevent oxide formation was required. Sulfuric acid was used for reasons elaborated later in this work, but it is theorized that some particles may not have been treated and, while not causing complete blockage of the capillary, blocked enough of it to reduce the effective cross-sectional area. This then increased the resistance to flow enough to give the measured viscosity.

Oxide particles are known to create a viscous film which could also increase the apparent viscosity, as the acid treatment only affected the upper surface of the fluid, while oxides could have formed when loading the viscometer and collected on the exterior of the fluid volume beyond the upper surface^[5].

Additionally, the Einstein equation for the viscosity of a dispersion/suspension, equation (16), indicates that even oxide particles not large enough to obstruct the capillary are still capable of increasing the viscosity^[31]. In this equation, μ is the apparent viscosity, μ₀ the viscosity of the pure fluid, α a fitting coefficient, and F the volume fraction of solids. The Einstein-Roscoe equation, equation (17), which is often used for calculating slag viscosity, is a further generalization of equation (16) with the additional fitting parameter, n^[32]. In either case, it can be seen that a small fraction of solid particles has the ability to quickly increase the apparent viscosity of the fluid, even in a flow which is not significantly obstructed.

Conducting measurements in either an inert atmosphere or via a different method, such as a Couette rheometer constructed of appropriate materials, are suggested methods to avoid the issue of solid particles increasing the observed viscosity. The uncertainty bound for the kinematic viscosity given in Table 1 was determined by applying equation (11) to equation (2) with the result being equation (18). σ_ν is the uncertainty in the kinematic viscosity, $\bar{t}$ is the average time required for the meniscus to drop from the upper to lower timing mark, C the viscometer constant, σ_expanded the expanded uncertainty given by the manufacture to be 0.16 pct (k=2), StdDev_t the standard deviation of the timing measurements, and σ_timer the readability of the stopwatch used. A confidence interval was not determined for the dynamic viscosity, but the uncertainty bound was assessed using equation (11) with a rearranged equation (15) which resulted in equation (19). For equation (19), σ_μ is the uncertainty in the dynamic viscosity, σ_ν the uncertainty in the kinematic viscosity given by equation (18), σ_ρ the uncertainty in the density given by equation (13), and ν and ρ are the measured kinematic viscosity and density given by equations (2) and (12) respectively.

The N.4 viscosity standard was used to confirm the calibration of the viscometer provided by the manufacturer, with the result being shown in Table 2. Kinematic viscosity data provided by the manufacturer was regressed to give the viscosity as a function of temperature shown in equation (20) (R² = 0.9981). For equation (20) ν is the kinematic viscosity in cSt while T is the temperature in °C.

For the N.4 standard, the value predicted by equation (20) does fall within the uncertainty bound, suggesting that the deviation from literature reported for GaInSn was not due to user or device error and was instead caused by either fluid properties differing from those in the literature, or, the more likely case, of oxidation causing minor capillary obstruction which was unnoticed by the naked eye while performing tests.

3.4 Electrical Conductivity

The LSV data provides the resistance/conductance of the system, and so the geometric factor was required to determine the fluid property; equation (21) gives this relationship. To get the geometric factor, the Woods metal conductance was compared with the literature conductivity given by equation (8); G is the measured conductance, c the cell geometric factor, and σ is the electrical conductivity.

Based on the tests conducted using Woods metal, the geometric factor for the cell was determined to be 2.91 cm with a 95 pct confidence interval of 0.08 cm. To ensure that the shielding provided by the aluminum foil cage was sufficient when compared to the commercial Faraday cage, the data was compared with that of GaInSn under the same conditions, with the result being that the 95 pct confidence intervals for each condition contained the average of the other. This then suggests that the shielding provided was sufficient to establish that the geometric factor was not a function of the shielding method used.

Using the geometric factor and equation (21), the conductance determined via the LSV data for GaInSn was converted to electrical conductivity, and the result is shown in Table 1. Literature values ranged from 3.4*10⁶ S/m at 20 °C as reported by Afrin et al. to 3.5*10⁶ S/m as reported by Dubovikova et al., although the most commonly reported value was 3.46*10⁶ S/m by Sarfo et al.^{[8, 21, 24]}. Values around 3.3*10⁶ S/m, which is lower than those otherwise seen in the literature, were reported by Plevachuk et al. and Wang et al.^{[3, 22]}. It is theorized these results were impacted by oxidation of the gallium which created a less electrically conductive film despite the use of an argon atmosphere^[15].

The fluid volume used impacted the electrode area in contact with the alloy, and therefore it was theorized that the geometric constant for the cell would be a function of the fluid volume used. Since it was assumed the alloy properties were homogenous, the LSV slope (the system resistance) was used as a proxy for the geometric constant. Using GaInSn, the fluid volume varied from 4.8 mL (the volume corresponding to the volume of Woods metal used in the cell calibration) to 6.8 mL. The slope was observed to slightly increase across this range, although the 95 pct confidence intervals for these data sets did overlap. Even though there was overlap between the 95 pct confidence intervals of the slope at both fluid volumes, comparison with the theoretical impact reveals a likely statistically significant influence.

Owing to the high electrical conductivity of GaInSn, the majority of current is believed to be carried in the top few layers of the fluid. The difference in fluid volume therefore primarily impacts the length of the electrode that the current must first pass through. A shorter electrode length, corresponding to a larger fluid volume, would decrease the measured resistance. The height difference for a 2 mL volume change is approximately 1 cm in the glass vial used, resulting in an overall conductivity path length change of 2 cm. Presuming the electrode is pure tungsten, this would result in a 10-12 μΩ change in the cell resistance. In comparison, the 4.8 and 6.8 mL trials showed a difference of 10 μΩ, aligning well with the prediction. If the geometric factor was assumed to be constant and did not account for the change in fluid volume, a larger fluid volume would exhibit a higher electrical conductivity than predicted with a volume matching that of calibration. Since the fluid volume did influence results, only experiments with volumes matching that of the calibration with Woods metal were included when determining conductivity.

3.5 Fluid Wetting

The issue of oxidation in GaInSn and similar alloys has been reported in the literature^{[3, 5, 14, 18, 19]}. As a treatment approach, sulfuric acid was studied in this work. Without treatment, Cadwallader reported modest oxide formation as a film on GaInSn but did not specify a film thickness, while Khondoker and Sameoto suggested a thickness of 0.7 nm in vacuum conditions and 1-3nm in a laboratory environment ^{[5, 17]}. Scharmann et al. found the oxide thickness to be approximately 1.9 nm in a 9 pct relative humidity atmosphere, while 2.5 nm in a 95 pct relative humidity atmosphere, which agrees with Khondoker and Sameoto for the thickness expected in laboratory conditions^{[14, 17]}. Touronen et al. reported the rapid formation of a 1-2 nm thick passivation layer while Bo et al. reported a thickness of 0.5-3 nm^{[18, 19]}.

Qualitative analysis of GaInSn after aging in various conditions is shown in Figure 5. Although the shape of the GaInSn surface does not show significant deviation between the different storage conditions, the fluid oxidation can be inferred by the color of the surface. Additionally, although not at a level that impacts the GaInSn-storage fluid interface, surface irregularities can be seen on the GaInSn surface in all cases, except when stored in sulfuric acid with a pH close to 0, as seen in Figure 5.c. This case also showed the highest flowability when the vial was shaken.

Fig. 5. Macro fluid behavior of GaInSn on a 45 ° incline with various cover fluids. For aging, (a) had no cover fluid while others were covered with (b) DI water, (c) sulfuric acid with pH 0, (d) sulfuric acid with pH 2.1, (e) sulfuric acid with pH 3.7, and (f) sulfuric acid with pH 5.0.

When aged in air, as shown in Figure 5.a, the sample formed an oxide layer on the surface but did not seem to heavily oxidize based on observable surface tension wrinkles and color. However, the water aged sample, Figure 5.b, underwent significant color change and no surface remained which resembled the metallic one in Figure 5.c. Storage in sulfuric acid at pH 2.1, Figure 5.d, produced a similar stationary surface appearance as storage in air, Figure 5.a, though storage in acid led to less wetting on the glass and no residual oxide film on the glass once the bulk fluid was removed from contact. Storage in pH 3.7 and 5, Figure 5.e and f respectively, showed heavier oxidation of the GaInSn than storage in air though, like the case with the pH 0 and 2 storage, the cover fluid was able to clean the residual oxide from the glass surface. This “cleaning” effect may have been largely due to the mechanical effect of the fluid movement on the oxide layer rather than the chemical treatment.

While the macroscopic surface behavior is shown in Figure 5, Figure 6 gives the SEM results used to assess the alloy condition when the sample was withdrawn from below the surface. As the fluid drawn up for the SEM imaging was not from the fluid surface (to avoid having aqueous solution in the low vacuum the microscope), the SEM results are not indicative of the degree of surface oxidation.

Fig. 6. SEM images for the fluid oxidation study. For aging, (a) had no cover fluid while others were covered with (b) DI water, (c) sulfuric acid with pH 0, (d) sulfuric acid with pH 2.1, (e) sulfuric acid with pH 3.7, and (f) sulfuric acid with pH 5.0.

When looking at Figure 5.e and f, it is expected that the less concentrated acid cover would exhibit more oxides under the SEM; this is not the case as shown in Figure 6.e and f. This observation indicates that oxide formation is localized to the fluid surface and oxide mobility is high in the alloy, as otherwise particles would be within the volume drawn up when the vial was mixed. In Figure 6.a and e, the large structure is the remnant of the transfer needle exit location; that the structure can be seen in these results suggests a decreased surface tension. Surface tension is expected to decrease with increasing oxidation, and from the macroscopic observations in Figure 5 it would then be predicted that the water cover and lowest sulfuric acid concentration should also have these structures^[33]. Under the water cover, Figure 6.b, the alloy surface shows a grain-like structure rather than continuous metal, with some precipitates as seen in the other images. However, if this were a continuous layer of oxides, and particularly gallium oxides, the oxygen content when assessed by EDS should be elevated relative to the other instances. However, as Table 3 shows, this was not the case. In fact, the water cover had the lowest oxygen content according to the EDS results even though the composition in both cases is similar. Since standards weren’t available to calibrate the method, it is possible the variation is within the uncertainty of the instrument.

The images in Figure 6 show that the oxide presence seems to manifest in three mechanisms. In Figure 6.d and f, solid particles are seen on the surface. Figure 6.a and e show a large-scale structure, which is the remnant of the transfer needle withdrawal that can only occur when the surface tension is decreased. Likewise, Figure 6.a and c show ripples in the surface which are only possible with a reduced surface tension. And finally, Figure 6.b seems to show a grain-like structure. Further investigation of the oxide morphology under different conditions is suggested, to determine whether it behaves as solute in the alloy or as a distinct phase on the surface, either as distinct particle or as a protective film. A film would be expected to increase viscosity but stop further oxidation, while solid particles, if small enough, would not necessarily increase the viscosity significantly but wouldn’t physically obstruct further oxidation.

3.6 Compatibility with PVC

Results of the PVC compatibility study are shown in Figure 7 below. As can be seen, the fluid did form a film on the PVC, but this film was easily removed with sulfuric acid, at approximately pH 0, and the PVC surface was then not optically distinguishable from the pre-treated surface with the naked eye. No discoloration of the PVC was observed following the acid bath after one week or one month, which would suggest no metallic enrichment in the PVC itself. There was also no observable geometric effect on the PVC such as pitting. These qualitative tests confirm the reported stability of GaInSn with PVC^{[8, 16, 17]}.

Fig. 7. PVC holder (a) prior to GaInSn filling and post fluid removal after 1 week (b) and 1 month (c).