Organization of Dissertation Work

This dissertation is organized as follows. Chapter 1 provides the introductions and motivations as well as the background on this research. A microsystem structure consists of versatile neural sensors associated with powering and signal conditioning electronics for neuroprosthesis applications is described. Detailed designs and results for the sensors and electrical systems are reported in the following chapters. Chapter 2 introduces a series of surface-mounted MEMS dry electrode (MDE) development for EEG, drowsiness and photo dynamic therapy (PDT) applications. Functions including low interface impedance, self-stability and transparent light guide are detailed in this chapter. Chapter 3 presents a flexible gird electrode array made by polymer for ECoG measurements; related device fabrication and characterization are investigated in this chapter. Chapter 4 reports a new stacking method for assembling a 3D microprobe array with great structure strength,

smaller implantable opening and simple assembly steps. Proposed method also provides design flexibility and volume usage efficiency. 3D neural signal propagation observation for human cerebral cortex layers is achieved by practical implantation in excised brain tissue after resection surgery in this study, which achieves the mapping and observation of the neural signal network among the target brain structure. Chapter 5 illustrates the integrated electronics toward microsystem. Miniaturized spiral coils as a wireless power module with low-dropout regulator circuit is developed to convert RF signal into DC voltage for batteryless implantation. 16-channel analog-front-end neural amplifier is also introduced, offering technical merits of sufficient low power and reasonable low noise performance. An integrated microsystem using the fabricated neural amplifier associated with commercial micro control unit and wireless transceiver module are described as well. Chapter 6 draws the conclusions in this work and provides possible future directions for further research.

Chapter 2

Skin Surface-Mounted MEMS Dry Electrode

2.1 Biopotential Electrode for Electroencephalography Recording

EEG signals consist of the differences in electrical potentials caused by summed postsynaptic graded potentials from pyramidal cells that create electrical dipoles between the soma (body of neuron) and apical dendrites (neural branches) [15]. The EEG is typically described in terms of rhythmic activity, which is divided into bands by frequency. For example, these frequency bands are a matter of nomenclature i.e. rhythmic activity between 8–12 Hz is described as alpha wave. Also, rhythmic activity within a certain frequency range was noted to have a certain distribution over the scalp or a certain biological significance [15]. Frequency bands are usually extracted using spectral methods as implemented by EEG software. Biopotential electrodes for EEG recording transfer signals from skin tissue to the amplifier circuit. When the electrodes placed onto the skin, say surface-mounted electrode, actually act as a voltage divider with the amplifier input resistance for signal transportation. Therefore, the most important characteristic of a biopotential electrode is low electrode-skin interface impedance so that signals can be propagated without attenuation or production of noise [16].

In this chapter, various types of MEMS-based dry electrodes (MDE) are presented for different purpose. Silicon-based MDE fabricated via micromachining technology is proposed for high-fidelity EEG sensor with low electrode-skin interface impedance.

Diamond-shaped MDE (DS-MDE) provides extra self-stability onto skin during measurement. Finally, a transparent MDE fabricated by advanced hot-embossing process is illustrated for photodynamic therapy application in this chapter as well.

2.2 MEMS Dry Electrode

When electrodes are placed on the skin of the forehead, an electrode-skin interface is constructed. The anatomy of the skin consists of three different layers: the epidermis, the

dermis, and the subcutaneous layer. The epidermis contains two further layers: the stratum

corneum (SC) and the stratum germinativum (SG). The SC consists of dead cells and thus

has electrical isolation characteristics. The SG is composed of living cells and is therefore

electrically conductive. Blood vessels and nerves are located in the dermis [17].

To overcome the electrical isolation properties of the SC, standard wet electrodes always require that the skin is prepared (abrasion of the SC) and an electrolytic gel is used.

Improper skin preparation may cause skin irritation, pain, or even infection. Using electrolytic gel is uncomfortable and inconvenient; it can cause an itchy feeling, and sometimes makes skin red and swollen when EEG measurements are made over a long period of time. Furthermore, the conductivity of gel gradually decreases because it hardens, resulting in degradation in the quality of data acquisition.

In this work, presented MEMS dry electrode (MDE) is designed to penetrate the SC layer into the electrically conductive SG layer but stop before the dermis layer to avoid pain or bleeding. Fig. 2.1 shows the basic idea and the difference between MDE, DS-MDE, transparent MDE and standard wet electrode. Since the dry electrode is expected to circumvent the high impedance characteristics of the SC, skin preparation and electrolytic gel application are thus not required. The thickness of the epidermis varies in different types of skin, but almost in the range from 0.05 mm (eyelids) to 1.5 mm (palms and soles). The

dermis also varies in thickness depending on the location of the skin. It is 0.3 mm on the

eyelids and 3.0 mm on the back [18]. The most important layer for EEG measurement, SC, is thinnest skin layer, being less than 20 um. Therefore, as long as the length of the microprobes on dry electrode are longer than the thickness of the SC, then, low electrode-skin-interface impedance can be achieved by the spiked tip of microprobe reaching the SG layer.

Fig. 2.1 Implementation diagram of applying standard wet electrode, MDE and DS-MDE onto skin surface

The relating electrode-skin interface cross section sketch and the related electrode-skin-interface equivalent circuit are illustrated in Fig. 2.2. The model for the standard electrode includes a capacitance Cd and a resistor Rd at the electrochemical electrode–electrolyte interface of the standard electrode and the electrolytic gel. The resistor RS reflects the electrical resistivity of the electrolytic gel. The SC can be considered to be a semi-permeable membrane. If there is a difference in ionic concentration across this

membrane (e.g., the ion concentration of the electrolytic gel and the ion concentration of the SG do not match perfectly), there is a potential difference Ese. Furthermore, the electrolytic gel soaked SC shows distinctive resistive and capacitive behavior, represented by Ce and Re. Finally, Ru is the resistivity of the SG and underlying tissue. The model of the dry spiked electrode is less complex. The conductive metal coating of the spikes is in direct contact with the SG, hence, an electrode–electrolyte interface, Cd and Rd, is created. Rm is the resistivity of the SG and underlying tissue [19].

Fig. 2.2 Electrode-skin interface comparison between spiked dry and electrode standard wet electrode.

2.3 Diamond-Shaped MDE

The MDE designs [20] use electrically conductive microprobe to penetrate the high electrically resistive outer skin layer to obtain better electrical conductivity than wet electrodes. However, the conical shape microprobe array lacks the position stabilization capability. This limitation is due to the force created by the compressed skin tissue that continually counteracts the conical shape microprobes into outward direction and degrades the bio-signal recording quality directly. To overcome this drawback, a self-stabilized, DS-MDE is developed. The proposed diamond-like design has a wider neck and a narrower bottom than previous conical shape dry electrodes. This configuration provides an external force to stabilize the probe that can remain in the skin when the tissue counteracts the probe continually. In contrast with a related work [21] which also demonstrates the mechanical attachment capability of a body surface electrode, the proposed DS-MDE without sharp barded edges does not damage the skin tissue after removing the electrodes from skin. The DS-MDE can thus provide satisfying self-stability capability and superior electric conductivity when attached onto skin without additional tissue injury.

Fig. 2.3 (A) illustrates the forces applied on the probe after the DS-MDE was placed onto skin tissue. Fb and Fp, caused by the compressed tissue, are the normal force applied on the face of tip and the face of shaft. Fr is the resultant force of |Fb + Fp| with direction towards the skin. Fc and Fv denote the frictional force and viscous force, respectively. Fs is the

minimal force that required for dragging out the probe from skin tissue. Clearly, the minimal Fs for dragging out the probe is



_{c v}



r

s

F F F

F   

(2-1) Since that Fc and Fv are small, difficult to estimate and always help the probe to stay in the tissue when the probes are dragging out of the tissue, they are ignored in the simulation. To quantitatively determine the improved self-stability of the presented DS-MDE comparing with MDE, stability factor (SF) is simply defined as the Fr, the resultant force of |Fb + Fp|. It is clear that if Fr can provide an inward force into the tissue, the stability of the probe can be improved.

The force Fb and Fp are positively proportional to the compressed tissue volume, which can be calculated from the dimension of probe (a, b, c and d) illustrated in the force diagram.

Thus, the resultant force Fr is calculated from the angle relationship in the force diagram.

For the general MDE, its vertical shaft wall (a = b) and conical tip result in a Fr with outward direction from the tissue which can not help the probe stay in the skin. Comparing with the MDE, DS-MDE with an invert-triangle shaft wall makes the resultant of Fb with the direction towards into the tissue, which helps the probe stay in the skin.

The calculated result indicates that the narrower the bottom width a, the higher stability can be achieved. Simulation result is shown in Fig. 2.3 (B).Note that with tip length and neck diameter are 50μm, SF reaches the maximum value if the bottom diameter is less than 30μm and shaft length is longer than 150μm. As a result, a stable probe configuration with 250μm probe length has been designed to reach the SG layer.

Fig. 2.3 Stability modeling

2.3 Fabrication and Characterization

The fabrication process of MDE and DS-MDE is illustrated as Fig. 2.4. The microfabrication process consisting of ion-etching with inductive coupled plasma (RIE-ICP) etching process and sputtering metallization technology was developed. In this process, a 6μm thick photoresist film was patterned as circular hard-mask for the isotropic etching process to produce the probe tip. Next, we proceeded with the anisotropic etching process to

form the probe shaft with high aspect ratio. According to the etching parameter control, different style of shafts including cylindrically shaped and inverted-triangle shaped are fabricated for MDE and DS-MDE, respectively. Then, the hard mask at the probe tip was released by sulfuric acid wet-etching. Finally, the probes were subsequently coated with titanium and platinum using sputtering technique to achieve electrical conductivity and bio-compatibility.

Fig. 2.4 Fabrication process flow

Fig. 2.5 shows the microphotograph of the fabricated results: (a) microprobe array of MDE, (b) tip of the microprobe, (c) inverted-triangle shaped microprobe array of DS-MDE, (d) close view of the DS-MDE. The fabricated MDE and DS-MDE are 20×20 array in 4×4 cm2 with 250μm in probe height. The MDE is 30μm probe width, while the DS-MDE is 50μm and 17μm in neck and bottom width, respectively. Notably, the peak of the probes on the DS-MDE are not perfectly sharp (< 10 μm) but tiny enough to penetrate the outer skin layer.

Fig. 2.5 Fabrication results

To characterize the electrode-skin interface impedance effect, two electrodes were lined up on the forehead with a distance of 4 cm apart to perform an electrode-skin-electrode interface (ESEI) experiments. The ESEI acts as two electrode-skin interface in series.

Therefore, lower ESEI performance implies lower electrode-skin impedance. A circuit [19]

was used to determine the ESEI impedance and reduce the risk of harming the test person during biopotential recordings [22]. A total of 19 tests were performed, involving 5 subjects to evaluate the performance of different types of electrode. According to Fig. 2.6, in the interest frequency range for EEG (0.5~100 Hz), MDE shows much smaller impedance performance then standard wet electrodes under condition without skin preparation (use of gel, abrasion of the SC). Experiment results demonstrate that the impedance of presented MDE was superior to that of the wet electrodes.

Fig. 2.6 Electrode-skin interface impedance characterization

A recording example of the MDE and standard wet electrode is shown in Fig. 2.7. The recorded EEG signal (α rhythm, 8-12Hz, seen in normal relaxed adults) with the peak-to-peak magnitude of approximate 0.5V and 0.95V for MDE and wet electrode. Note that both wet electrode and MDE are placed at the neighboring location with identical amplifier circuit. The relationship of electrode/skin interface impedance and the signal intensity proved the advantage of the MDE design.

Fig. 2.7 Recording example of alpha rhythm by MDE and standard wet electrode To compare the EEG signals acquired by MDE and standard wet electrodes, five MDE/wet electrode pairs are placed at the frontal of head. The MDE/wet electrode pair 1 and 5 is placed at Fp1 and Fp2 according to the international 10-20 electrode placement system [1].

Three more electrode pairs are also evenly placed between pair 1 and 5 labeled as MDE/wet electrode pair 2, 3, and 4, respectively. The distance between MDE and wet electrode in the MDE/wet electrode pair is about 1 cm. We used the average EEG signals at A1 and A2 as the reference. EEG signals at MDE and wet electrodes were then simultaneously recorded by the Scan NuAmps Express system (Compumedics Ltd., VIC, Australia). Before data acquisition, the contact impedance between each dry and wet electrode was calibrated to be less than 5 kΩ. The EEG data were recorded with a 16-bit quantization level at a sampling rate of 500 Hz and then re-sampled down to 250 Hz to simplify data processing. Each amplifier circuit channel had a differential-input instrumentation amplifier as the first amplifier stage, followed by a 0.5–100 Hz band pass filter and a 60 Hz notch filter.

We recruited three volunteers to take part in the experiment. In each experiment, the EEG signals were simultaneously recorded for 1000 seconds from the two pairs of disposable dry and wet electrodes, including eyes-open and eyes-closed conditions as shown in Fig. 2.8 (A).

Using the comparison method [23], a segment of a time-series recording comparing the signals obtained for Subject 1 from the pair of disposable dry and wet electrodes respectively at Fp1 is shown in Fig. 2.8 (B). Also included on the plot is the Pearson correlation over 0.25-s segments. Although only a single 5-s segment is shown, the recording quality was consistent for all subjects without obvious changes in signal quality or noise levels. In Fig. 2.8 (C), statistics showed that the average Pearson correlation of all subjects was over 90%, and in 91% of the recording data it was higher than 92%. Very significant correlations, in excess of 90%, are evident throughout the record.

Fig. 2.8 (A) Simultaneously EEG recording from MDE and wet electrode (B) 5-s segment comparison of EEG signals. Upper line shows the Pearson correlation data calculated using

0.25-s segments. (C) Statistics analysis of the average correlation for all subjects.

The self-stabilized capability of the DS-MDE was tested by a micro force testing system

(MTS Corp., USA). MDE and DS-MDE with same area (4x4 mm²), each consists of 20x20 micro probe array, are gluing to a PMMA holder for the pulling force test. A pig skin fixed on another PMMA plate was used for the test tissue. The DS-MDE was pressed into the pig skin tissue under testing by the holder beam of the force testing system. The force with which the DS-MDE was pressed into the tissue was not measured. After ensuring the micro probes penetrated the test tissue, the micro force testing system started to pull off the DS-MDE. During testing, the applied force and displacement were recorded. Fig. 2.9 shows the testing results. The average required pulling force for the general MDE and DS-MDE were 1.705N and 3.160N, respectively. Notably, both MDE and DS-MDE had the same chip area, number of probes and probe length.

In contrast with prior art, proposed DS-MDE without sharp barded edges does not damage the tissue after removing from skin. The DS-MDE can thus provide satisfying self-stability capability and superior electric conductivity without additional tissue injury.

Fig. 2.9 Pulling test result. The average required force for the general MDE and DS-MDE were 1.705N and 3.160N, respectively.

2.4 Drowsiness Monitoring with MDE

In order to demonstrate the potential applications of the MDE sensors during long and routine recording in operational environments, the drowsiness monitoring was investigated according to the EEG signals recorded by proposed MDE placed at Fp1 and Fp2 in an attention-demanding driving experiment. Preventing accidents caused by drowsiness is greatly desirable but requires techniques of continuously monitoring drivers’ drowsiness levels and delivering effective feedback to avoid dangerous situations at the wheel [24]. An EEG-based drowsiness estimation system that continuously estimates drivers’ drowsiness levels in a virtual-reality (VR) based driving simulator is used here [25]. The VR based highway-driving environment provides the study of drivers’ cognitive change during a long-term driving. Also, a lane-keeping driving error experiment is defined as the drowsiness level used to verify the estimated drowsiness level produced by the EEG power

spectrum analysis. The power spectrum level of theta wave and alpha wave defined the drowsiness index as shown in Fig. 2.10. During the period of increasing drowsiness but when subjects are still responding, delta and theta power increased, whereas alpha decreased [26]. Therefore, the drowsiness monitoring by MDE can be achieved by combining the EEG spectrum analysis and VR based driving error.

Fig. 2.10 Drowsiness level detection by EEG power spectrum observation

The EEG signals recorded by five MDE sensors are fed into an EEG-based drowsiness estimation system to indirectly estimate the driving drowsiness levels. The recorded driving performance time series were smoothed using a causal 90-s square moving-averaged filter [27] advancing at 2-s steps to eliminate variance at cycle lengths shorter than 1–2 min, since the fluctuations in drowsiness levels had a cycle length over 4 min [28]. The EEG data recorded by five MDE are first preprocessed using a simple low-pass filter with a cut-off frequency of 50 Hz to remove the line noise and other high-frequency noise. After moving-average power spectral analysis, we obtained EEG log power spectrum in time-series from the EEG sensors, with a frequency range from 1 to 40 Hz [29]. Then, Karhunen-Loeve Principal Component Analysis (PCA) is applied to the resultant EEG log spectrum to extract the directions of the largest variance for each session. Projections (PCA components) of the EEG log spectral data on the subspace formed by the eigenvectors corresponding to the largest 50 eigenvalues are used as inputs to a multiple linear regression model [30] to estimate the time course of driving errors for each subject. Each model is trained only using the features extracted from the training session and tested on a separate testing session.

Fig. 2.11 shows the performance comparison of drowsiness estimation either using MDE sensors or standard wet electrodes. As illustrated in the figure, the blue line and red line represent the driving errors acquired by VR system and estimated by EEG, respectively. Fig.

2.11(A) shows the estimated driving error correlation of Subject 1 in Session 2 by using EEG recorded from the wet electrodes, where the estimators are trained from Session 1.

Conversely, Fig. 2.11 (B) shows the estimated driving error of Subject 1 using EEG data

from wet electrodes, where Session 2 acts as training dataset and Session acts as testing session. Also, Fig. 2.11 (C) and (D) show the estimated and actual errors made by Subject 2.

Similarly, Fig. 2.11 (E), (F), (G) and (H) display the estimated driving error correlation made by MDE sensors with varied training session and testing session.

Fig. 2.11 EEG estimated (red line) and actual VR driving error (blue). (A)-(D): Estimated by standard wet electrodes (E)-(H) Estimated by MDE sensors

Table 1 shows the comparison of the correlation coefficients between the actual and estimated driving error time series using MDE and wet electrodes. As shown in Fig. 2.11

Table 1 shows the comparison of the correlation coefficients between the actual and estimated driving error time series using MDE and wet electrodes. As shown in Fig. 2.11

在文檔中 腦神經訊號感測元件開發與系統整合應用 (頁 13-0)