A customizable, low-power, wireless, embedded sensing platform for resistive nanoscale sensors

Customizable, portable, battery-operated, wireless platforms for interfacing high-sensitivity nanoscale sensors are a means to improve spatiotemporal measurement coverage of physical parameters. Such a platform can enable the expansion of IoT for environmental and lifestyle applications. Here we report a platform capable of acquiring currents ranging from 1.5 nA to 7.2 µA full-scale with 20-bit resolution and variable sampling rates of up to 3.125 kSPS. In addition, it features a bipolar voltage programmable in the range of −10 V to +5 V with a 3.65 mV resolution. A Finite State Machine steers the system by executing a set of embedded functions. The FSM allows for dynamic, customized adjustments of the nanosensor bias, including elevated bias schemes for self-heating, measurement range, bandwidth, sampling rate, and measurement time intervals. Furthermore, it enables data logging on external memory (SD card) and data transmission over a Bluetooth low energy connection. The average power consumption of the platform is 64.5 mW for a measurement protocol of three samples per second, including a BLE advertisement of a 0 dBm transmission power. A state-of-the-art (SoA) application of the platform performance using a CNT nanosensor, exposed to NO2 gas concentrations from 200 ppb down to 1 ppb, has been demonstrated. Although sensor signals are measured for NO2 concentrations of 1 ppb, the 3σ limit of detection (LOD) of 23 ppb is determined (1σ: 7 ppb) in slope detection mode, including the sensor signal variations in repeated measurements. The platform’s wide current range and high versatility make it suitable for signal acquisition from resistive nanosensors such as silicon nanowires, carbon nanotubes, graphene, and other 2D materials. Along with its overall low power consumption, the proposed platform is highly suitable for various sensing applications within the context of IoT.

Using the superposition principle applied on the linear op amp configuration, the V bias5 potential can be 10 programmed in the [−2V bat. … V bat. ] range with a 3.65 mV resolution as: 11 where Vbat. = 5 V is the battery voltage supply,# =12 bits is the DAC resolution and # the input 12 code respectively. 13 By choosing the integration period and integration capacitance , the FS is programmable in the 14  The practical relevance of the proposed platform is highlighted by the example of the daily NO2 average 31 value acquired by NABEL station [2] for two of Switzerland's biggest cities: Zurich and Lugano as 32 presented in Figure S1. 33

CNT characterization 36
A pre-characterization of the CNT nanosensor is performed before exposure to 2 gas analyte. In Figure  37 S2a. the transfer characteristic of KTDS15, mounted in the test chamber under atmospheric conditions, is 38 presented at different VDS bias levels. The CNT nanosensor shows no pronounced hysteresis, which is 39 expected for suspended CNT device architecture [3]. The output characteristics of the CNT nanosensor is 40 presented in Figure S2b

59
The measurement baseline is defined as being the nanosensor current level after a reset phase by the help 61 of elevated bias voltage, which leads in the self-heating of the CNT device as Joule a thermal effect. During 62 the set of experiments presented in the manuscript, the CNT device has been exposed to a gradually decreasing 63 NO2 gas concentration followed by a reset state. One can quantize the reset efficiency by evaluating the 64 offset of the baseline current after each reset phase. By exposing the CNT device from high to low NO2 gas 65 concentration one can observe a proper reset phase given by a straight baseline. This means that the reset 66 phase duration and bias levels are properly tuned to reach the same baseline current after both high and low 67 NO2 gas exposure. For example, a long reset phase would lead in a baseline current undershoot after a low 68 gas concentration exposure. In contrast, short reset phase would lead in an incomplete desorption after a 69 short reset phase at a high gas concentration exposure. This can be highlighted by two additional 70 experiments presented in Figure R5 and Figure R6, which compares the resulting currents levels 71 immediately after the reset and exposure phase at different bias levels and reset time intervals. In Figure  72 R5, at a sensor bias of 100-200 mV, an optimal sensor response is observable. Furthermore, at a Self-73 Heating time period of 45 minutes we can observe optimal sensor reset (see: Figure R6 ). These parameters 74 were utilized for performing the measurement in

Sampling Frequency Power Consumption Overhead 80
Each sample of the 3 SPS in the current implementation is composed of 32 averaged samples, with 81 4 samples intentionally discarded during the first integration cycle. Figure S6 shows the detailed capture of 82 the DDC114 sampling signal. Note: each edge represents one sample [1]. 83 Figure S6 A) The current sampling scheme for the DDC114 resulting in 3 SPS programmed by the FSM.
B) The detailed sampling structure of the 3 SPS burst signal. C) The zoom-in signal structure of an individual sample presented in detail.

87
In this case, the DDC114 power consumption overhead is not of concern. Measurement results of the 88 DDC114 at different sampling rates are presented in Figure S7. The measured values are for all the four channels of the DDC114. Note: the measured sampling rate matches 90 with the values reported in Table 4 of the datasheet [1]. The minor difference compared to B) is given by 91 the analog voltage reference REF3140 [6], which is required for the DDC114 to operate. 92 levels to the CNT nanosensor. Over extended period of time, these can lead i) to chemical oxidation in 96 presence of O2 and ii) an increased electronic noise due to the thermal agitation of charge carriers. An 97 additional measurement set describing these effects is presented in Figure S8. In the experimental data shown in Figure S8, the CNT nanosensor is successively exposed to a constant 99 NO2 gas concentration of 100 ppb after a preliminary DA (Dry Air) exposure. After a CNT nanosensor 100 reset phase (not shown in this graph), the successive gas exposure is performed at a gradually increased 101 VDS in steps of 150 mV. It can be observed that the current signal value and it's corresponding slope 102 increases with VDS values up to VDS = 450 mV. However, for the responses with VDS > 450 mV, an 103 increasing current is noticeable but the corresponding slope decreases. This could be explained by 104 considering the thermal heating effects leading to the desorption of NO2 molecules from the CNT 105 nanosensor surface. As highlighted earlier, we can observe an increased noise level of the current signal at 106 VDS > 450 mV. This experiment has been repeated thrice for consistency. 13 9. AlphaSense signal response acquired by the proposed platform when exposed to NO2 gas analyte 108 Figure S9 AlphaSense signal response acquired by DDC114 on exposure to NO2.

Experimental determination of LOD and R 2 109
The reported response time of 12 minutes corresponds to the lowest LOD=23 ppb (3σ) and the highest 110 linearity evaluated by the help of R 2 linear fit parameter. However, this can be decreased down to 5 minutes, 111 as presented in Figure S10 with an LOD of  90 ppb (3σ). Herein still comparable to other sensing materials 112 reported by the articles highlighted in Table 1 of the manuscript. 113 Figure S10 LOD (red filled diamond) and R 2 coefficient (blue circle) vs. Slope Detection (SD) time window.
Note: The SD time of 12 minutes was identified as good choice and applied in Figure 5A of the main text.

Humidity Cross-Sensitivity 115
In Figure S11, the influence of relative humidity pulses has been evaluated in comparison to dry air 116 conditions. Reduced cross-sensitivity to humidity is observable for gas flow conditions with 0 and 117 Figure S11 CNT nanosensor signal response at 0 and 100 ppb NO2 gas concentration in the absence and presence of humidity

Power Consumption 119
The reported power consumption of main components in Active and IDLE mode is calculated as in Table  120 S1. 121