Sweat Biomarker Sensor Incorporating Picowatt, Three-Dimensional

Jul 8, 2019 - Read OnlinePDF (4 MB) ... Abstract Image ... to simulate our 3D-EMG-ISFET; IDVG characteristics of the sensors to describe how we extrac...
0 downloads 0 Views 850KB Size
Subscriber access provided by GUILFORD COLLEGE

Article

Sweat Biomarker Sensor Incorporating Pico-Watt, 3D-Extended Metal Gate ISFETs Junrui Zhang, Maneesha Rupakula, Francesco Bellando, Erick Garcia-Cordero, Johan Longo, Fabien Wildhaber, Guillaume Herment, Hoël GUERIN, and Adrian M. Ionescu ACS Sens., Just Accepted Manuscript • DOI: 10.1021/acssensors.9b00597 • Publication Date (Web): 08 Jul 2019 Downloaded from pubs.acs.org on July 18, 2019

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 17 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Sensors

Sweat Biomarker Sensor Incorporating Pico-Watt, 3DExtended Metal Gate ISFETs Junrui Zhang 1*, Maneesha Rupakula 1, Francesco Bellando 1, Erick Garcia Cordero 1, Johan Longo 2, Fabien Wildhaber 2, Guillaume Herment 1, Hoel Guérin 2, Adrian Mihai Ionescu 1 1Nanoelectronic 2Xsensio

Devices Laboratory, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland

SA, Innovation Park, Lausanne, Switzerland

*corresponding

author: [email protected]

KEYWORDS ISFET, Ultra-low power, functionalized sensing, Wearable sweat sensor, Real time sweat monitoring, Energy harvesting ABSTRACT Ion-Sensitive-Field-Effect-Transistors (ISFETs) form a very attractive solution for wearable sensors due to their capacity for ultra-miniaturization, low power operation and very high sensitivity, supported by CMOS integration. This paper reports for the first time, a multi-analyte sensing platform that incorporates high performance, high yield, high robustness, 3D-Extended-Metal-Gate ISFETs (3D-EMG-ISFETs) realized by the post-processing of a conventional 0.18 ? CMOS technology node. The detection of four analytes (pH, Na+, K+ and Ca2+) is reported with excellent sensitivities (58 mV/pH, -57 mV/dec(Na+), -48 mV/dec(K+), -26 mV/dec(Ca2+)) close to the Nernstian limit, and high selectivity, achieved by the use of highly selective Ion Selective Membranes based on post-processing integration steps aimed at eliminating any significant sensor hysteresis and parasitics. We are reporting simultaneous time-dependent recording of multiple analytes, with high selectivities. In-vitro real sweat tests are carried out to prove the validity of our sensors. The reported sensors have the lowest reported power consumption, being capable of operation down to 2 pW/sensor. Due to the ultra-low power consumption of our ISFETs, we achieve and report a final 4-analyte passive system demonstrator including the readout interface and the remote powering of the ISFET sensors, all powered by an Radio Frequency (RF) signal. Challenges in sweat sensors Human perspiration offers a real-time, non-invasive way of personalized health condition monitoring since it has been fully demonstrated that sweat contains key biomarkers such as glucose, lactate, Na+, K+, H+, NH4+, Ca2+, Cl-, many of which are correlated with their concentrations in blood [1, 2]. Thus, monitoring these analytes is of high interest for a large variety of applications, ranging from complex body states such as hydration, to heart and metabolic diseases [1, 3, 4, 5]. The most common device architectures for wearable sweat sensors are based on Ion Selective Electrodes (ISEs) [6, 7, 8, 9, 10]. However, they usually have a large form factor (on the order of millimeters), thus, a large number of such electrodes in a sensor array could inhibit system integration. Moreover, due to their large

ACS Paragon Plus Environment

1

ACS Sensors 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 17

size, they require the collection of relatively large amount of biofluid to cover the whole sensor area. In contrast, Ion Sensitive Field Effect Transistors (ISFETs) are very scalable as they are based on advanced CMOS technologies [11], offer for ultra-low power operation [12] and large scale integration of functionalized sensor arrays with multiple metabolite and parameters detection capability [13, 14, 15, 16]. Further, due to their small size, they require much lower sample volume and can be easily integrated with microfluidic channels [4]. E. Stern et al. [17], demonstrated label-free detection of femtomolar concentrations of antibodies as well as real-time monitoring of the cellular immune response. C. Toumazou et al. [18] in 2013, developed an ISFET-based simultaneous DNA amplification and detection sensor array. F. Bellando et al. [4] reported in 2017, an ultra-thin body SOI ISFET sensing arrays that enable nanoWatt sweat sensing together with a passive microfluidic channel, for sweat volume down to nanoliter range. All these previous works and demonstrations support the unique advantages of ISFETs as integrated biosensors and motivate the developments reported in this work. Nevertheless, the fabrication of high performance ISFETs requires dedicated non-foundry processes, which has the drawbacks of high cost, low yield, low robustness and difficulty of integration, especially when multianalyte sensing is required. Having this in mind, commercial CMOS process has been exploited in its BackEnd-of-the-Line (BEOL) for more than 2 decades [19, 20], in order to realize cost-efficient, high performance ISFETs (CMOS ISFETs fabricated in un-modified CMOS processes). However, the performance of CMOS ISFETs is limited by low sensitivity, large spread in threshold voltage (Vth), large drift rate, and high power consumption. Table 1 summarizes key performances of state of the art CMOS ISFETs in comparison with ISFETs fabricated with dedicated non-foundry process flow. Table 1. Performance summary of state of the art pH sensing ISFET devices from both un-modified CMOS processes and dedicated non-foundry processes.

Fabrication technology 0.35 ? CMOS ISFET sensing with Si3N4 [21] 0.35 ? CMOS ISFET sensing with Si3N4 [22] =2 >? CMOS ISFET sensing with Si3N4 [23] FinFET sensing with HfO2 [11] SiNW sensing with HfO2 [24] ISFET sensing with Al2O3 [25] This work (0.18 ? CMOS ISFET)

Sensitivity (mV / pH)

Drift Rate (mV / h)

Vth Spread (V)

Min. Power per Sensor

9.2

30

1.5 ± 0.9

26.95

5.8

-3.3 ± 6

NG

10×10

26.2

NG

NG

NG

10×10

57

0.3

NG

3 nW **

8×0.04×0.12

55.3

1.88

NG

NG

10×0.2×0.2

56

0.3

NG

NG

1000×100

57.2

0.67

0.354 ± 0.028

2 pW

10×20

20 ?

Device dimensions :? ;

**

5×0.35

NG: Not Given * Erased trapped charge by exposure to UV radiation for 17 hours, chips were fabricated with a quartz window. ** Calculated from published figures.

ACS Paragon Plus Environment

2

Page 3 of 17 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Sensors

Even so, thanks to the aforementioned irreplaceable advantages, researchers are still striving hard [21, 26] to remedy CMOS ISFETs’ poor performances, at the cost of system complexity and power consumption. Thus, it will be extremely interesting to have an ISFET sensor that resolves the cost, yield, robustness versus performance dilemma from device level. Moreover, complete integration of pH, Na+, K+ and Ca2+ sensors on a platform less than 2.5 mm2 for sensing body fluid, as well as complete study of cross sensitivity among themselves, has never been reported. Na+ and K+ sensors have been developed by several groups for ion sensing, both with ISFETs [3, 4, 5] and with ISEs [6, 9]. Ca2+ sensor was first demonstrated by [8] in body fluid measurements. However, only [5] and [14] have addressed study of cross sensitivity (Na+ over H+ and K+ ions that usually coexist in body fluids).

Fig. 1 a) A photo of the wearable system. b) A photo of the top-side of the NFC tag. ASIC with 3D-EMG-ISFETs is bonded on the bottom side of the tag (‘skin side’). A sweat outlet with cotton is attached and extended from the skin side to the front side of the tag through a via. c) A photo of the DC characterization measurement setup. A drop of the liquid under test (LUT) sits within an epoxy well surrounding the chip and the reference electrode (RE) tip is inserted within the drop. The rigid PCB has the same circuitry as on the NFC tag. d) Block diagram of the NFC powered sensing system. e) Cross section view of the 3D-EMG-ISFET sensor. f) 3D-EMG-ISFETs with Al2O3 as pH sensing dielectric, functionalized with Ion Selective Membranes (ISMs) for Na+, K+, Ca2+ sensing.

3D-EMG-ISFETs revolute the challenging situations in traditional ISFETs

ACS Paragon Plus Environment

3

ACS Sensors 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 4 of 17

In this work, we present a high performance, high yield, high robustness, low cost, 3D-EMG-ISFETs, for pH and multi-analyte (Na+, K+, Ca2+) sensing. Device to device variation, which has been a big issue in ISFET sensors, is eliminated through our novel post-process steps. The sensors exploit BEOL of a commercial CMOS process, making it easy to adopt different surface modification methods which not necessarily have to be CMOS compatible. For example, [27] realized heterogeneous integration of a CMOS-graphene sensor for measuring dopamine. Another example, the lowest biomarker concentration that can be detected (LOD) with a planar surface has been theoretically compared with that of a cylindrical nanowire surface [28]. The LOD of cylindrical nanowire surface is 3 orders of magnitude lower than that of planar surface. Further, femtomolar LOD is evidenced by experiments with silicon nanowires [13, 17]. Thus, similar heterogeneous integration technique could be envisioned to improve the LOD of planar ISFET sensors. We also demonstrate the potential of this sensor to be used in wearable sweat sensing (Fig. 1a). Fig. 1b depicts our fabricated sensor, which consists of (1) NFC interface chip and antenna. (2) Microcontroller. (3) 12-bit Analog to Digital Converter (ADC). (4) absorbing material for draining the sweat and ASIC (Application-Specific Integrated Circuit) containing 3D-EMG-ISFETs, 16:1 multiplexer, and onchip integrator. As depicted in Fig. 1a and Fig. 1d, once the mobile phone gets in proximity, the sensor tag is powered on by NFC energy harvesting. A constant drain-source bias voltage is generated from the ASIC, and the Liquid Under Test (LUT) is biased by a commercial Ag/AgCl reference electrode (R.E) (RE-Ag/AgCl from Micrux). The functionalized 3D-EMG-ISFET outputs a drain current (ID). ID is readout with an active integrator integrated in the ASIC, in order to reject high frequency noise. The integrator reads the current signal into an amplified, low noise voltage signal. The voltage signal is then converted by an ADC and transmitted via NFC interface to a smart phone for data processing and finally displaying the measured drain current as well as calibrated ion concentration/pH results. The wearable system on the subject’s arm shown in Fig. 1 is to give an impression of what the final sensor might look like in the future and not the embodiment used to generate the results shown in this paper. The future sensor will include properly integrated microfluidics (handles sweat samples down to 100 nL) and absorbing material for displacing old sweat. These have been implemented in our previous work [29] for ISFETs fabricated on an SOI substrate. The core of the NFC sensing tag is formed by 3D-EMG-ISFET arrays that have been carefully characterized as sensors for different ions (H+, Na+, K+ and Ca2+) and combined with an adapted microfluidic interface for sweat collection on humans. The measurement setup is shown in Fig. 1c: the R.E. is used to bias the LUT, an Agilent semiconductor parameter analyzer (HP4156) is used to apply bias voltages at the drain, source, reference terminals, and, at the same time, monitor the corresponding currents. With this measurement setup, we demonstrate the high performance (high predictability, high repeatability, high sensitivity, high selectivity and fast response) of 3D-EMG-ISFETs for pH, Na+, K+ and Ca2+ sensing, each ion sensor employing a different dedicated Ion Sensitive Membrane (ISM), as depicted in Figs. 1e-f). RESULTS AND DISCUSSION 3D-EMG-ISFET Sensing with Al2O3 The 3D-EMG-ISFETs are fabricated by post-processing MOSFET devices designed in a commercial 0.18 µm CMOS chip. The gate of the MOSFET (10 ? × 20 ? ; is vertically extended in 3D to the top metal layer through stacks of vias and metal layers with SiO2 as inter-metal dielectric (IMD), as shown in the cross section

ACS Paragon Plus Environment

4

Page 5 of 17 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Sensors

in Fig. 1e. A Reactive Ion Etching (RIE) post-process is utilized to open the passivation layers (Si3N4 and SiO2) sitting above (details shown in the supplementary, Fig. S3 and Fig. S4) the top metal. The exposed top metal (Aluminum) is oxidized to form a thin Al2O3 layer which is used as a pH sensing layer. These postprocess steps can easily be replaced by using a passivation opening mask in the layout design phase before sending to foundry fabrication, thus making the fabrication of 3D-EMG-ISFET an unmodified commercial CMOS process. Sensitivity: Limit of ISFET Devices The potential of the sensing surface ( S) is modified when the target ions are captured (Fig. 1f). logarithmically on the target ion concentration ( 0) given by [30]: 1

S depends

( 0)

(1)

where k is the Boltzmann constant, T is the absolute temperature, z is the valence of the target ion, q is elementary charge. Since the sensing surface is electrically connected to the gate of the MOSFET, modification of results in change of the drain current, . Hence, the sensitivity of an ISFET is defined in two ways: Voltage sensitivity

= [

Current sensitivity

= [

[

(unit: mV/decX)

(2)

(unit: dec/decX)

(3)

( 0)] ( )] ( 0)]

where decX stands for decade-of-target-ion-concentration. All the sensors in this paper will be operated in the subthreshold region, where ISFET, the variation of

in response to the reference electrode voltage (

region is called the subthreshold swing (SS) [31]:

$)

(

!"#

). For an

$

change, in the subthreshold

$

= [ ( )] (unit: mV/dec). MOSFET based devices share a physical limit, which is usually derived from the Boltzmann statistics, used as approximation of the FermiDirac statistics in the classical regime. Theoretically, there is a lower limit for the SS (SS > 59 mV/dec) of a MOSFET device at room temperature. Lower SS (abrupt transition from off to on states of transistor conduction) is desired for a FET based device, since it fundamentally results in lower voltage and lower power consumption [32]. There is also an upper limit for the voltage sensitivity S , originally derived from Gibbs free energy [33], called the Nernstian limit (0 < S < 59 mV/decX at room temperature). ISFETs exceeding Nernstian limit sensitivity (super Nernstian) have been reported by several recent papers, however, it has then been demonstrated by comprehensive theoretical analysis [34] and exhaustive experiments in [35], that they contribute no improvement to the signal to noise ratio. Super Nernstian sensitivity reported so far is based on circuit amplification, rather than fundamentally surpass of the Nernstian limit.

ACS Paragon Plus Environment

5

ACS Sensors

The limitations for

can be related to the subthreshold swing of the MOSFET, via the relation:

$

= ( , where > 59 mV/dec. It follows that, in practice 0 < < 1 and best SI can be obtained in MOSFETs with subthreshold swing close to the thermionic limit of 59 mV/dec at T=300K. pH 4

5

6

0.8 0.6

S=57.1 mV/pH S=59.3 mV/pH

0.4

S=58.8 mV/pH

0.2 4

0.8

5

6

7

0.6

0.75

0.4 0.70 pH NaCl 0.65 KCl CaCl2 NH4Cl 0.60 0.1

2.7 mV/pH -56.9 mV/dec 2.3 mV/dec 2.8 mV/dec -7.1 mV/dec

0.0 1

10

(a) 5

pH 6

7

4

0.75

0.4 6.4 mV/pH 8.6 mV/dec -48.1 mV/dec 0.8 mV/dec -3 mV/dec

0.2

Sensitivity, Vref, [V]

0.6

Cross sensitivity, Vref, [V]

0.80

0.0 1

10

5

6

7 0.6

0.54

0.8

pH NaCl 0.70 KCl CaCl2 NH4Cl 0.65 0.1

100

(b)

pH 4

0.2

Electrolyte concentration, [mM]

pH

0.85

7

0.80

S=50.6 mV/pH

Sensitivity, Vref, [V]

Sensitivity, Vref, [V]

1.0

0.85

1 nA 10 A

Cross sensitivity, Vref, [V]

20 pA 100 nA

100

0.52

0.4 0.50

pH 0.48 NaCl KCl CaCl2 0.46 NH4Cl 0.1

Electrolyte concentration, [mM]

(c)

0.3 mV/pH -2.4 mV/dec -6.3 mV/dec -25.7 mV/dec 0.16 mV/dec

0.2

Cross Sensitivity, Vref, [V]

1.2

Sensitivity, Vref, [V]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 17

0.0 1

10

100

Electrolyte concentration, [mM]

(d)

Fig. 2 a) pH sensitivities extracted from Fig. S2a at different drain current levels. b)-d) Sensitivity and Cross sensitivity plot for the Na+, K+, Ca2+ sensing 3D-EMG-ISFETs, extracted from Fig. S2b-S2d, respectively, at ID = 10 nA.

Pico-Watt 3D-EMG-ISFET for pH and Multi-Ion sensing The accurate modeling and experimental model calibrations for sensors are important for the both the predictability of sensor performance and the co-design with read-out circuitry. A SPICE behavioral model [36, 37], able to capture analytically the static and dynamic behavior of our ISFET sensors is described in Supplementary Figure S1, and used to simulate the pH sensing characteristics of the 3D-EMG-ISFET.

ACS Paragon Plus Environment

6

Page 7 of 17 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Sensors

Simulated IDVG characteristics agree well with measurement results (Fig. S2a, to be detailed later). However, for many reported state of the art ISFETs, their threshold voltages are highly unpredictable due to (i) immature process flow of dedicated non-foundry processes (process variations), (ii) trapped charge problem at the oxide interfaces and/or in oxides in foundry process, that affect the sensor characteristics, essentially mirrored in hysteresis. In this work, we have introduced a simple and effective way (see in Fig. S5) to eliminate the trapped charge effect in the 3D-EMG-ISFETs. The threshold voltage spread has been reduced from 4.39V (in the standard process) to 56 mV (in our modified process, with control and elimination of trapped charges). The achieved post-processing variability of the ISFETs threshold voltage values is comparable to that of metal gate MOSFET devices with the same architecture and technology node (with a threshold voltage spread of 12 mV). State of the art attempts to remove the trapped charge using UV radiation have given satisfactory results [38], however, with a device-to-device calibration process (17 hours of UV radiation for each device). The experimental achievements in trapped charge elimination reported here, supported by SPICE calibrated behavioral model simulations results, are expected to greatly enhance the predictability of high performance ISFETs, which is a big step towards ISFET sensor system mass production. The drain current (ID) to Reference electrode voltage (VG) transfer characteristics simulation results, together with the measurement results, are shown in Fig. S2a. This is done by biasing the VDS at 100 mV, and sweeping the VG from 0 V to 1 V, for each pH buffer. The pH sensor exhibits an on-off current ratio (Ion/Ioff) of about 106. Subthreshold swing in liquid conditions is 85.3 mV/dec. According to Eq. (3), low Subthreshold swing is good for high current readout sensitivity, while large Ion/Ioff is responsible for the dynamic range when the ISFET is operated in current readout mode. The same procedure is taken to measure the IDVG characteristics for 3 times. The pH sensitivity is extracted from the IDVG curves, in terms of shift in VG at various constant ID in weak inversion and shown in Fig. 2a, with mean value and error bars indicating the standard deviation. The sensitivity is close to the Nernstian limit, with no degradation down to 20 pA of operation current. In order to make the 3D-EMG-ISFETs selectively sensitive to salt ions, they have been functionalized with dedicated Ion Selective Membranes (Methods section), to become 3D-EMG-(Na+ / K+ / Ca2+) sensitive FETs. IDVG characteristics for 3D-EMG-(Na+ / K+ / Ca2+) sensitive FETs are shown in Figs. S2b, S2c S2d, respectively. Using the same procedure as with the pH sensing 3D-EMG-ISFET, we measure the IDVG characteristics for 3 times and extract the sensitivities of each sensor to the corresponding ion species at ID = 10 nA. The sensitivities are shown in Figs. 2b, 2c, 2d, respectively. Excellent sensitivities (-56.9mV/decX for Na+, -48.1 mV/decX for K+, -25.7 mV/decX for Ca2+) are achieved with the 3D-EMG-(Na+ / K+ / Ca2+) sensitive FETs, respectively. The “low” sensitivity of the Ca2+ sensitive FET is due to its divalent nature, resulting in a Nerstian limit of 29.5 mV/decX, instead of 59 mV/decX for monovalent ions (H+, Na+, K+, etc). The same method is applied to extract cross sensitivities among the 3 ions as well as with pH and NH4+. Excellent cross sensitivities without any need for post-sensing computation are also shown in Figs. 2b, 2c, 2d. High cross sensitivity, or selectivity, is very important for sweat sensing application where many interfering ions coexist with the main ion in the LUT. It is worth to note that different groups of researchers have achieved ISFETs with comparable or lower cross sensitivity [4, 5], at the cost of more area (for a control electrode) and extra computation, e.g. differential measurement. The 3D-EMG-ISFET’s dynamic response in pH sensing is studied by measuring time dependent ID response of the sensor, at constant bias voltages: VDS = 100 mV and VGS = 500 mV. As shown in Fig. 3a, the pH sensing exhibits excellent repeatability and fast response (< 5 s). Similar dynamic measurements for the 3D-EMG(Na+ / K+ / Ca2+) sensitive FETs are shown in Figs. 3b – 3d, respectively. It is clear from Fig. 3c that the 3D-

ACS Paragon Plus Environment

7

ACS Sensors 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Paragon Plus Environment

Page 8 of 17

Page 9 of 17 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Sensors

in Fig. S7. Shown in Fig. S7a is a 0.18 ? foundry fabricated CMOS sensor chip containing five groups of 3D-EMG-ISFET sensors, each group contains nine ISFETs, dedicated to sense one type of ion. The drain, source, and gate terminals of all the ISFETs are connected through metallic wires and vias to pads in the peripherals of the chip. The bulk terminals are all connected to the p-doped substrate and biased through the aluminum base shown in the inset of Fig. S7c. Fig. S7b shows two groups of ISFETs covered by K+ and Na+ ISMs, respectively. The ISMs are casted on top of the sensing area, but we are also working on inkjet printing of ISMs, in order to make the fabrication of ion sensors more controllable and occupy less area. The measurement setup is shown in Fig. S7c. The LUT is biased with a commercial Ag/AgCl RE (16-702, Microelectrodes Inc.) through plastic tubes. A flow cell (Fig. S7c inset) is used to deliver the LUT to the surface of the sensors. The flow cell is constructed with PMMA by laser cut fabrication. There is a cavity at the end of the flow cell allowing the LUT to cover the area of the sensors. The cavity is filled with 6 ?% of LUT when a measurement is undergoing. We chose one of the 3D-EMG-ISFET sensors in each group covered by the ISMs, and measured their dynamic response by probing the corresponding drain and source teminals through the pads. Setting VS = VB = 0 mV, VD = 100 mV and VRE = 500 mV, the dynamic drain currents ID,Na+ and ID,K+ are recorded simultaneously, and shown in Fig. 3e, which to our knowledge is one of the first experiment of its kind reported in multiple analyte detection with ISFETs. The LUTs are prepared with two salts (NaCl and KCl) of various quantities dissolved in DI water. From this plot, the sensors show rapid and repeatable response to the concentration changes in the main ions, while at the same time exhibit negligible response to the interfering ions. In-vitro Sweat Test We have designed an in-vitro sweat test experiment to prove the validity of our sensors in sweat sensing application, using the same setup as described in Fig. S7c to measure samples of unspiked sweat. The detailed experiment steps and calibration methods are described in the Methods section. The dynamic measurements of artificial sweat buffers and unspiked sweat samples are shown in Figs. 4a and 4c. The calibration curves are extracted and shown in Figs. 4b and 4d, together with error bars showing the mean value and standard deviations from the dynamic measurement data. We can see that the drain current of our 3D-EMG-ISFET sensors are, as expected from Eq. (3), exponentially related to pH or, in general, to the logarithm of ion concentrations. Na+ concentrations and pH values of the sweat samples are calculated with the aid of the calibration curve. The dynamic measurement results for extracting calibration curve and calculating sweat electrolyte concentration show a maximum time-dependent standard deviation of 1.95% in 3D-EMG-Na+ sensor, and 1.36% in the 3D-EMG-pH sensor, with respect to their average value (Fig. S8). The corresponding calibration curves are shown in Fig. 4. Calculated sweat pH and Na+ concentrations are listed in Table 2. We measured the same sweat samples with three 3D-EMG-Na+ sensors and three 3D-EMG-pH sensors. The standard-deviation-to-average-value-ratios of the measurement results from sensor to sensor are characterized to be 9.3% for our Na+ sensors, and 2.3% for our pH sensors. Compared to the results from the commercial Na+ meter and pH meter, our sensors’ measurement results deviate from them by maximum 12% and 2.3%, respectively. These standard deviations and variations from commercial sensors are small enough to be neglected compared to the large range of Na+ concentration and pH in sweat during daily activities.

ACS Paragon Plus Environment

9

Page 10 of 17

100 Drain current, ID, [nA]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Average drain current, ID, [nA]

ACS Sensors

10

10

1

0.1

10 100 Na concentration, [mM]

5

6

7

8

pH

(a)

(b)

Fig. 4 Calibration curves of three a) 3D-EMG-Na+ sensors and b) 3D-EMG-pH sensors.

Table 2. Na+ concentrations in two sweat samples measured by three 3D-EMG-Na+ sensors and a commercial Na+ Meter; pH values of the same samples measured by three 3D-EMG-pH sensors and a commercial pH Meter. Na+ Meter results were converted from open-circuit-potential readings.

sw #1 sw #2

3D-EMGNa+ sensor 1 74.7 33.9

[Na+] (mM) 3D-EMG3D-EMGNa+ sensor Na+ sensor 2 3 69.2 68.2 29.7 35.7

pH HORIBA Na+ Meter

3D-EMGpH sensor 1

3D-EMGpH sensor 2

3D-EMGpH sensor 3

HORIBA pH Meter

77.5 33.2

7.76 5.28

7.69 5.24

7.43 5.16

7.65 5.16

Wearable potential demonstration through an NFC platform The NFC platform’s power budget is about 1.2 mW and has been evaluated per each of these system components (1150 ? for the microcontroller, 6 ? for the ASIC and about 69 ? for the ADC). The NFC interface chip adopted is able to provide energy harvesting power supply up to 22.5 mW. The block diagram of the NFC platform is shown in Fig. 1d, and more detailed description of the readout circuitry is discussed in supplementary Fig. S9. The significantly high power consumption of the demonstrator NFC platform may give an impression to the readers that there’s no point of minimizing the power consumption of individual sensors, and hence our 2 pW consumption sensor would be pointless. However, looking into the future, the system power consumption will need to be significantly lowered, to a level that can be operated continuously with harvested energy. This motivates the quest for a lower power budget of sensors arrays. According to a recent publication, Douthwaite, M., et al. [15] demonstrated a pH sensor array powered with thermoelectric generator, which goes along the stringent requirement of nanoWatt power for sensors. The NFC platform is applied to perform pH sensing, as well as Na+ sensing, the drain current ID value is recorded by a corresponding android application, and plotted in Fig. S10, in comparison with the same device measured with precision semiconductor analyzer. An on-chip miniaturized Ag/AgCl quasi reference electrode (QRE) has been fabricated (fabrication process described in Methods section). The fabricated QRE is shown in Fig. S11.

ACS Paragon Plus Environment

10

Page 11 of 17

Table 3. State-of-the-Art (SoA) in electrochemical ion sensors

Na

K

Ca

CMOS Int4

H

SoA

DR3

Ion

RT2

Cross sensitivity: 3

FoM

min. Power1

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Sensors

NG

NG 2.3

NG 2.8 0.8

0.59 0.78 0.7 0.5

2 pW