Cellulose Fibers Enable Near-Zero-Cost Electrical Sensing of Water-Soluble Gases

We report an entirely new class of printed electrical gas sensors that are produced at near “zero cost”. This technology exploits the intrinsic hygroscopic properties of cellulose fibers within paper; although it feels and looks dry, paper contains substantial amount of moisture, adsorbed from the environment, enabling the use of wet chemical methods for sensing without manually adding water to the substrate. The sensors exhibit high sensitivity to water-soluble gases (e.g., lower limit of detection for NH3 < 200 parts-per-billion) with a fast and reversible response. The sensors show comparable or better performance (especially at high relative humidity) than most commercial ammonia sensors at a fraction of their price (<$0.02 per sensor). We demonstrate that the sensors proposed can be integrated into food packaging to monitor freshness (to reduce food waste and plastic pollution) or implemented into near-field-communication tags to function as wireless, battery-less gas sensors that can be interrogated with smartphones.


Table of Contents
To further discuss the accuracy of our model (a detailed description of our model is shown below in part SI-P1.2), we would like to analyze a specific example involving NH 3 and trimethylamine (TMA). According to our theoretical model, 10 parts-per-million (ppm) of ammonia should produce a similar change in conductivity to 20 ppm of TMA, with a ratio of ( : change in conductivity produced by 10 ppm of NH 3 , : the change in conductivity produced by 20 ppm TMA). Our experimental results for PEGS in Fig. 2A indicate that the ratio of change in conductivity for NH 3 and TMA is approx. 40 ( = 40). Hence, there is an order of ∆ 3 ∆ magnitude mismatch between the calculated and experimental results for the concentrations mentioned above.
To understand whether the deviation from the calculations were originating from the use of paper (and not just a body of bulk water), we have performed the following experiment: We printed two carbon electrodes on a polymer substrate (marking transparency), submerged into 3 mL of deionized water and measured the change in conductivity in the presence of 10 ppm of NH 3 and 20 ppm of TMA (Fig. S11) and took their ratio (

SUPPORTING INFORMATION
3  There may be interactions between the gases and cellulose matrix. Such evidence is found in cellulosic fiber treatment called 'mercerization' where physical and chemical properties of the cellulose fibers are modified using an alkaline solution, including ammonia. 1 These modifications may influence the ionic conductivity.
 The surface bound layer of water has slightly different chemical properties than free (bulk) water.
 Structural changes of cellulose fibers have been reported when moisture content surpasses 3.5 wt%. Additional conductivity due to dissociated ammonia: Henry's solubility constant : : : partial pressure of ammonia 3 Partial pressure : Dissociation constant : Assumption: Solve equation (3) and (5) for :

SI-P2. Data analysis of food spoilage experiments
We obtained our raw data from an Arduino DUE ADC port (10 bit) and scaled it to consider the different gain resistors we used in our transimpedance amplifying setup 1 . This gave us the data as shown in Fig. S6 A+B.
We applied a moving average filter before we looked for local maxima or minima in our data. The red circles in reaching equilibrium with the container's atmosphere. This means, external factors (e.g. added water, temperature) do not influence the sensors anymore. We normalized the data by subtracting and dividing by the . This gave data that indicated the change of the conductance over time in relation to a reference value (Fig. S6 C+D). We then normalized the response to 100g per sample. In this case the raw data was for 40g cod fish (Fig. S6 A) and 20g chicken breast (Fig. S6 B) respectively. Averaging over the four sensors in the containers with food samples and the two sensors in the water (control) boxes, gave us the data presented in Fig. 5 in the main text.

SI-P3. Comparison with state-of-the-art electronic nose
To better frame the effectiveness of the proposed paper sensors within state-of-the-art technologies, we tested metal oxide (MOX) gas sensors in parallel with paper sensors to track the degradation of cod over time.
We chose MOX sensors because of their well-established sensing capability, which has been widely demonstrated in a variety of applications. 5,6 The experimental setup was the same as used for the food experiment with the paper gas sensor. We kept the MOX device inside a container (180 mL) together with two paper-based sensors. Both technologies were tested under the same conditions and against the same target.
Since MOX are non-specific sensors, i.e. they respond to a broad range of chemicals, the electrical resistance of a single MOX device is not suitable to track the target in complex atmospheres, such as those developed by decomposing food. These MOX sensors were, therefore, exploited in a sensor array configuration (so called "electronic nose"). We applied temperature modulation protocols to achieve the desired selectivity and the sensor response is depicted as a Principal Component (PC) plot in Fig. S7. 7 The sensor's response was acquired every 20 seconds and each data point corresponds to a single measurement. For simplicity, a set of 20 data points acquired every 6 hours in a frame of 400 seconds is plotted. This data formed six different clusters. From the clusters' arrangement it is possible to observe a trend that is representative to the cod degradation and the capability of MOX sensors to track it. It can be observed that clusters are well separated for the first 18 hours.
The distance between consecutive clusters diminished over time until the clusters were almost overlapping after 18 hours. This indicates the capability of the MOX device to distinguish among the different degradation degrees during the initial stages and shows that the performance of the MOX sensors suffers as the spoilage gas concentration increases. This may be reasonably ascribed to saturation effects. 8 This saturation occurs when the total microbial concentration value is around 10 8 -10 9 CFU/g (colony forming units per gram), according to the microbial control experiment.

Methods for metal oxide sensors:
We used a commercial platform (Minimox from JLM Innovation GmbH) equipped with two micromachined MOX sensors: TGS8100 (Figaro) and CSS801 (CCMOSS). We applied a square wave to the sensor heaters, to give a warm period of 10 seconds at voltage V heater = 2.31 V and a cold period of 10 seconds with V heater = 1.65 V. We measured the sensor resistance with a sampling rate of 40 samples/s.
Since the resistance value of a single MOX gas sensor is unsuitable to track the complex processes underlying food spoilage, the response of MOX sensors is retrieved by periodically warming and cooling the sensors through the embedded heater. This activates and freezes the interaction between gaseous molecules and the metal oxide surface, producing a resistance vs. time curve. 9 In this work, the two metal oxide gas sensors were excited according to the same protocol: We applied a square wave of period 20 seconds, duty cycle 50% and voltage values of 2.31 V and 1.65 V. Fig. S14 shows the resistance over one wave cycle of the CSS801 sensor measured during cod spoilage experiments at the beginning of the experiment (Fig. S14 A) and after 24 hours (Fig. S14 B). The resistance increased during the cold period and decreased during the warm period. The curves differ in terms of amplitude and shape, which reflects the different composition of the gas phase developed by fish over time. To characterize the shape of these curves, we used the following parameters:  ΔR cold-hot is the resistance variation between the cold and the hot period. More precisely, it is calculated as the difference between the sensor resistance measured at the end of the cold period and the resistance measured at the start of the warm period after 0.2 seconds.

SUPPORTING INFORMATION
8  ΔR cold is the resistance variation within the cold period. It is calculated as the difference between the sensor resistance measured at the end of the cold period and the resistance measured at the beginning (after 0.2 seconds) of the same period.
We calculated these for each sensor and used a principal component analysis algorithm (PCA-function of MATLAB) on the data.

= -×
The circuit schematics for the entire read-out device are shown in Fig. S12 and S13 in detail.    In a last step we normalized to a fish sample of 100g to achieve our final plot (Fig. 5). (D) Corresponding data for chicken. For the control sensor signal we additionally used a moving average filter (n=200) before we normalized the data.     (B) to 20 ppm TMA over ca. 60 min (n=2).
The blue bands show when the test gas (NH 3 or TMA) was present.   (B) Similarly, two cycles of a hot and cold period of the MOX sensor after 24 h. The parameter ΔR cold is used to quantify the shape of the curve and ΔR cold-hot quantifies the difference in resistance between the cold and the hot period of the measurement cycle. Movie S1. We bypassed an NFC tag with a resistor and a paper gas sensor (Fig. 7). The sensor had a DC resistance of 1640 kΩ and a capacitance of 0.5 nF with no ammonia gas present. We kept the modified tag in a 180 mL container and tapped a reader (smartphone HUAWEI P9) to detect it. The communication between tag and reader worked when no ammonia gas was present. We added 15 mL of 10% ammonia solution to the container and the DC resistance and capacitance of the sensor changed to 80 kΩ and 7.5 nF. The tag was not detected by the reader anymore. This was because the integrated circuit of the tag was bypassed with a PEGS and did not receive enough power to communicate with the reader. We demonstrate the paper-based sensor as an on/off detector for ammonia gas.