Supporting Information A Single Input Multiple Output (SIMO) Variation-Tolerant Nanosensor Dong-Il Moon 1,2, Beomseok Kim 1,2, Ricardo Peterson 3, Kazimieras Badokas 1,4, Myeong-Lok Seol 1,2, Debbie G. Senesky 3, Jin-Woo Han*,1,2, and M. Meyyappan 1 1 Center for Nanotechnology, NASA Ames Research Center, Moffett Field, CA 94035, USA 2 Universities Space Research Association, NASA Ames Research Center, Moffett Field, CA 94035, USA 3 Department of Aeronautics and Astronautics, Stanford University, Stanford, CA 94305 USA 4 Institute of Photonics and Nanotechnology, Vilnius University, Vilnius, LT 10257, Lithuania *Address correspondence to jin-woo.han@nasa.gov Supporting Information Contents 1. Device fabrication 2. Device characterization 3. Gas mixing system and sensor testing 4. Reliability of the fabricated SIMO sensor 5. Humidity impact of the fabricated sensor 6. Analysis and comparison of conventional sensor array and the SIMO scheme 7. Analytical modeling 8. Detection limit S1
1. Device fabrication The process steps and images of the fabricated devices are shown in Figure S1. The device fabrication totally relied on the inkjet printing technology using commercial equipment (FUJIFILM Dimatix DMP-2800). Polyimide (PI) film was selected as substrate due to its thermal stability over 200 C, good chemical resistance and excellent mechanical properties. Metallic and semiconducting material inks were used for the electrode and sensing layer respectively. A conductive Ag ink (InkTec, TEC-IJ-060) was used for metal contacts and interconnection lines. The viscosity and surface tension of the Ag ink ranged from 5 to 15 cps and 27 to 32 dynes/cm at 25 C, respectively. The sixteen individual Ag electrodes were printed in a concentric fashion. Each silver pad was 700 µm 700 µm in area with 900 µm spacing. The bulk silver resistivity was 1.6 10-6 Ω cm after the curing process at 130 C for 10 min done to obtain high electrical conductivity. In this process, the color of the Ag pattern changes into shiny metallic. Pristine SWCNT powder (Nanostructured & Amorphous Materials) was used to synthesize the semiconducting ink for the sensing material. 40 mg of purified SWCNTs was dispersed in 20 ml of deionized water. The solution was then sonicated for 2 hours to disperse and shorten the nanotubes by breaking them at any defects already present. 69.7 % wt HNO 3 was then slowly added to form a 40 ml 8 M HNO 3 /SWCNT solution. The mixture was refluxed at 120 C for 4 days. Then, the SWCNT solution was diluted with DI water, centrifuged and washed three times to remove any remaining HNO 3. The SWCNT film was then printed to overlay the Ag electrodes with the film bridging arbitrary pairs of Ag electrodes. After the SWCNT printing, thermal annealing was carried out to remove the solvent of the ink and improve the contact characteristics between Ag and SWCNTs. The magnified image of the fabricated device is shown in Figure S2. S2
Figure S1. Devices fabrication steps and final devices. Ag, CNT, and PI are used for the metal electrodes, sensing material, and substrate, respectively. S3
Figure S2. Microscope images of the printed device. The sensing area of 2.5 mm by 2.5 mm is covered by the CNT film and the magnified view of the SEM image of the CNT film is shown at the bottom. S4
2. Device characterization The fabricated bare chip was mounted on a clamp-shell type test socket for gas test, which has an open window for gas exchange. A computer-based automatic measurement system was custom built for sensor characterization. The test socket was directly linked to a multimeter (Keithley 2700) through switching matrix module (Keithley 7709) in order to serially measure multiple data, as illustrated in Figure S3. The LabView system controls the overall operation. With the automatic measurement system, 16-channel input/output data were simultaneously controlled and logged. Figure S3. Multi-channel automatic measurement system. 16-channel I/O of the test socket is connected to Keithley 2700 multimeter, which are controlled and logged by the computer system. 3. Gas mixing system and sensor testing In order to perform ammonia detection, the cover of the test socket was connected to the outlet of the gas mixer system with a rubber stopper. The rubber stopper was fixed in the open window in the center of the test socket and the tube was then fixed in the hollow space in the center of the rubber stopper. Then, ammonia from the gas mixer can react with the sensor inside the test socket. The experimental setup is shown in Figure S4. The test socket was then S5
electrically connected to the multi-channel automatic measurement system explained above. The air gas with controlled humidity was used as background carrier gas. A mixture of air and ammonia was supplied from the mixer, and the concentration of ammonia was varied from 1 to 50 ppm. Figure S4. Gas sensing experimental setup. The gas mixture is supplied from the mixer and the sensor response is monitored in real-time. 4. Reliability of the fabricated SIMO sensor Stability, repeatability and reproducibility of the sensor are addressed in Figure S5. The stability of the baseline resistance is important because the gas sensor must be able to detect reliably over a long time. When a sensor is under operation, the applied voltage and resultant current flow may alter the carbon nanotube film characteristic. Likewise, when the sensor is exposed to the ambient, the carbon nanotube film characteristics may shift. The result of any physical or chemical change in the carbon nanotube film over time may manifest in the form of S6
baseline resistance drift and thus, the long-term stability becomes an issue. If the resistance changes or drifts during the standby state or after gas exposure, it is difficult to design a circuit, and it could also possibly cause a malfunction. Figure S5a shows the long-term baseline resistance of the SIMO sensor over a two-month period, demonstrating stability over time. In order to confirm the operational stability of the printed SIMO, the gas response test was intermittently conducted as shown in Figure S5b. The response characteristics are consistent over two month, and there was no signature of sensor drift. The gas response characteristics of each of the sub-sensors in the SIMO on the 56 th day of measurement are shown in Figure S5c. As expected, no degradation of the sub-sensor response over this period was noticed, which proves the long-term stability of the printed SIMO gas sensor. a b C Resistance (kω) Response (%) Response (%) 20 15 10 5 0 2.5 2.0 1.5 1.0 0.5 0.0 2.5 2.0 1.5 1.0 0.5 0.0 Gas experiment NH 3 20 ppm NH 3 10 ppm NH 3 5 ppm 0 10 20 30 Day (#) 40 50 60 NH 3 20 ppm NH 3 10 ppm NH 3 5 ppm 1 3 5 7 9 Sensor (#) 11 13 15 Figure S5. (a) Long term baseline resistance stability of the SIMO. (b) Long term repeatability of gas response of the SIMO. (c) Reproducibility of gas response across sub-sensors in the SIMO sensor after 56 days of operation. Devices show reliable operation in terms of time, electrical stress and chemical reaction. All measurements were conducted under dry conditions (RH= 0%). S7
5. Impact of humidity on sensor performance Humidity is known to limit the sensor performance as it can change the baseline resistance. The water vapor can also prevent the sensing reaction with the target gas molecules if the humidity is too high. Since CNTs show cross-response to water vapor, it is important to assess the change in sensor behavior with humidity. Ammonia detection at various humidity levels is shown in Figure S6. The response point of the sensor lags behind the target gas exposure when the humidity is high. The data suggests that the water vapor may prevent spontaneous adsorption of the gas on the nanotube surface. However, the delay time decreases as the target gas concentration increases. The sensor response becomes greater when the humidity is high. Thus, the water vapor delays the reaction between the target gas and the sensor, but does not prevent the reaction itself. Based on the fact that ammonia has a good solubility in water and that the response of the sensor is larger at higher humidity, we can conclude that ammonia is dissolved in the water vapor present on the surface of nanotubes, and that the water vapor acts as a promoter for ammonia sensing. When the humidity increases over 70 %, the sensor response starts to degrade and no sensor response is seen at humidity over 90 %. This indicates the need for deploying some type of humidity filter as a preprocessor in breath monitoring applications, which commonly feature excessive humidity of over 95%, in order to reduce the humidity to nominal manageable levels of 40-50%. a Resistance (kω) 27.5 27.0 26.5 26.0 25.5 25.0 Purge Humidity 0% Humidity 60 % NH 3 5 ppm Delay Delay Delay 0 600 1200 1800 2400 3000 Time (sec) 6 5 4 3 2 1 0 4 6 8 10 12 14 16 18 20 NH 3 concetration (ppm) Figure S6. Impact of humidity on ammonia sensing. (a) Real-time detection of NH 3 under different humidity conditions. (b) Sensor response according to NH 3 concentration and humidity. b Response (%) 10 9 8 7 Humidity 0% Humidity 50% Humidity 60 % Humidity 70 % 6. Analysis and comparison of conventional sensor array and the SIMO scheme S8
Figure S7a maps 120 combinatorial responses before (top left) and after (bottom right) 10 ppm ammonia exposure. The fingerprint patterns after gas exposure look similar to the pattern before exposure, i.e. the baseline pattern, implying that each conduction path formed in the nanotube network responds consistently. Despite the spread seen in the resistance values across each electrode pair, the resistance shift upon gas exposure is consistent. However, it does not mean that the resistance shift is quantitatively similar. The distribution of resistance responses is plotted in order to quantitatively examine the resistance shift of resistors formed across individual electrodes. All data points showed increase in resistance upon exposure to ammonia (Figure S7b). So, the sensing material is qualitatively functional but not quantitatively consistent, which provides motivation for the suggested statistical Gaussian measure. Figure S7. (a) The resistance distribution of the printed CNT network. No significant difference is observed between contour maps of the purging step (top) and sensing step (bottom). It means that individual resistance path within the SWCNT network shows similar sensing response. (b) Summary of sensing response from the SIMO scheme. The sensor response statistical analysis of the multiple array of 120 conventional twoterminal devices and the 120 data sets obtained from the proposed SIMO structure is presented in S9
Table S1. It also includes comparison of three different SIMO sensors. All sensors, regardless of the multiple array or the SIMO, show similar mean response as the sample size of 120 is sufficient enough to converge into Gaussian distribution. However, the response distribution of the three SIMO sensors is narrower than that of conventional sensors. This is attributed to the fact that the SIMO shares 16 electrodes and one nanotube film to generate 120 data points whereas the 120 individual devices in the conventional array sensor use 240 electrodes and 120 nanotube films. Despite the spread in individual sub-sensor values from 2.63 % to 5.51%, the population mean value of the SIMO sensors becomes reproducible to 3.87 ± 0.44%. In addition, the sensing results from the three SIMO devices are relatively uniform. Conv. SIMO_1 SIMO_2 SIMO_3 SIMO_avg Number of data 120 120 120 120 360 Mean of response 3.28 % 3.88 % 3.83 % 3.91 % 3.87 % Standard deviation 0.98 % 0.55 % 0.37 % 0.4 % 0.44 % Standard error of mean 0.089 % 0.05 % 0.033 % 0.036 % 0.023 % Minimum 0.7 % 2.63 % 2.76 % 2.8 % 2.63 % Median 3.3 % 3.88 % 3.82 % 3.9 % 3.86 % Maximum 5.05 % 5.49 % 4.8 % 5.51 % 5.51 % Table S1. Summary of gas sensor response. The results of conventional sensors (conv.) are based on 120 individual sensors. SIMO provides 120 data points in one device. The average of the SIMO (SIMO_avg) includes three SIMO devices. All data in Table S1 are from the 10 ppm NH 3 experiment under a relative humidity of 50 %. The standard deviation, error of mean, minimum, median, and maximum are from the distribution of sub-sensors of a given sensor. 7. Analytical modeling Analytical modeling was carried out to support the experimental results. The model is based on the total conductance of a bundle of nanotubes where the diameter, metal-tosemiconductor ratio and total tube counts are considered as random variables but captured by a Gaussian distribution. The resistance of an individual nanotube is derived from the diameter of S10
the nanotube. The metal-to-semiconductor ratio can be determined by the purification process, and the total number of nanotubes to form a network can be determined by the concentration of the nanotube ink and deposition process. Thus, the mean and standard deviation of the ratio and total number of nanotubes were chosen to fit the experimental result. The resistance of the nanotube ensemble to form a sensor is then derived from random samples created from these three variables As unpurified nanotubes were used in the experiments, the approximate metal-tosemiconductor ratio (r m/s ) was assumed to be 1/3. The number of conducting modes for every nanotube shell can be defined to be N modes/shell (d) ~ (a d + b) r for d > 6 nm or 2 r for d > 6 nm. Herein, the coefficients a = 10.1836 nm -1 and b = 1.275 are used as fitting parameters. When the CNT length is greater than the mean free path, the conductance of each shell is defined as G shell (D out )=G 0 N modes/shell (d i ), where G 0 = (2q 2 /h) (λ/l cnt ) is coefficient, L cnt is nanotube length, λ = ν F d/α T is the mean free path, ν F = 10 6 m/s is Fermi velocity of graphene, and α = 6.25 m/sk is the scattering coefficient. The conductance of a nanotube can be obtained as = (1) This conductance is for the nanotube of particular diameter. In order to obtain the distribution density function of the nanotube, the number of nanotubes is given by the following Gaussian distribution = (2) where N network is the total number of nanotubes, is the average diameter, σ Dout is the standard deviation of diameter. Herein, N network = 120, = 2.5 nm, and σ Dout = 1 nm were used. Then, the conductance of a random network of nanotubes can be calculated by S11
= (3) The resistance can be calculated as R network = 1 / G network. The model outlined above was iterated for many times to simulate multiple samples and fit them to the experimental data. The impact of 10 ppm of NH 3 was expected to shift 9.1 % of the resistance of a semiconducting nanotube. Figure S8a and S8b shows the probability distribution of nanotube density and metalto-semiconductor ratio obtained to fit the experimental results of ammonia response. The mean value and standard deviation are 2500 and 750 (nanotube density) and 1/3 and 0.0667 (metal-tosemiconductor ratio), respectively. a PDF (10-3 ) 4 3 2 1 0 0 1 2 3 4 5 6 Number of CNTs (10 3 /sample) b PDF 6 5 4 3 2 1 0 0 1 2 3 4 5 6 CNT ratio (10-1 /sample) Figure S8. Probability density function (PDF) of (a) the nanotube density and (b) metal-tosemiconductor ratio used in the modeling. 8. Detection limit The lowest detectable gas concentration is limited by the noise from the experimental setup. The standard deviation of averaged sensor noise (σ noise ) from the baseline data is 5.515 10-3 when the signal-to-noise ratio equals 3. S1 Therefore, the detection limit can be extrapolated from the linear calibration curve (Figure S9). S12
= 3 The NH 3 detection limit calculated to be 0.115 ppm from the above equation. Figure S9 Calibration curve (red line) from the experiment data (square dot): (a) measured response for NH 3 up to 20 ppm and (b) extrapolated response for NH 3 close to 0 ppm. (S1) Currie, L. A. Pure Appl. Chem. 1995, 67, 1699. S13