Novel multi-sided, microelectrode arrays for implantable neural applications
- First Online:
- Cite this article as:
- Seymour, J.P., Langhals, N.B., Anderson, D.J. et al. Biomed Microdevices (2011) 13: 441. doi:10.1007/s10544-011-9512-z
A new parylene-based microfabrication process is presented for neural recording and drug delivery applications. We introduce a large design space for electrode placement and structural flexibility with a six mask process. By using chemical mechanical polishing, electrode sites may be created top-side, back-side, or on the edge of the device having three exposed sides. Added surface area was achieved on the exposed edge through electroplating. Poly(3,4-ethylenedioxythiophene) (PEDOT) modified edge electrodes having an 85-μm2 footprint resulted in an impedance of 200 kΩ at 1 kHz. Edge electrodes were able to successfully record single unit activity in acute animal studies. A finite element model of planar and edge electrodes relative to neuron position reveals that edge electrodes should be beneficial for increasing the volume of tissue being sampled in recording applications.
KeywordsNeural recordingMicroelectrode arrayParyleneNeural prosthesesDrug deliveryChemical mechanical polishing
High density neural recording devices have been and continue to be a critical tool for neuroscientists studying complex neural circuits (Buzsaki 2004; Riera et al. 2009). This technology also enables potentially life-transforming clinical devices for the treatment of paralysis (Donoghue et al. 2007) or refractory epilepsy (Stacey and Litt 2008). Next generation neural probes must enable stable, higher density recordings to advance neurotechnology (Cheung 2007; Kipke et al. 2008). This work offers a new polymer-based microfabrication technique that enables a larger design space for electrode placement, electrode density, and structural flexibility.
Parylene-C (poly-chloro-p-xylylene) has desirable material properties for use in a microfabricated implantable device. It has the highest certification level of biocompatibility, USP Class VI. This biocompatibility results from it being chemically inert and synthesized using a solvent-less polymerization process (Gorham 1966). Critically important to microfabrication, one can generate conformal, pinhole-free barrier films that are subsequently dry-etched using an oxygen plasma. For these reasons, parylene-C has long been used in neurotechnology, especially as a coating. It has been successful in coating microwires (Loeb et al. 1977) and micro-machined silicon arrays (Kim et al. 2008). Researchers have also microfabricated planar devices with a parylene-C substrate in a microfluidic device with recording capabilities (Takeuchi et al. 2005), retinal stimulation array (Li et al. 2005; Rodger and Tai 2005), and a flexible nerve cuff (Spence et al. 2007).
Unfortunately, a microfabricated parylene-based neural probe, as opposed to a parylene coating on free-standing conductors, cannot be assumed to provide low noise performance. Researchers have shown that water and salts will rapidly diffuse to the various interfacial boundaries (Yasuda et al. 2001; Seymour et al. 2009). Interfacial delamination due to poor wet adhesion of parylene on metal (Sharma and Yasuda 1982; Yasuda et al. 1996) and mechanical stress may create low impedance pathways resulting in cross-talk or shunt loss. More recently, researchers have shown that thin parylene films (50-300 nm) have regular pinholes, which can be used for drug diffusion control (Zeng et al. 2005; Pierstorff et al. 2008; Robinson et al. 2008), and must be of concern in electrical insulation applications. High-density parylene-C microfabricated structures have closely spaced conductors similar to conventional silicon-based neural probes (Vetter et al. 2004; Wise 2005; Kim et al. 2008). Unlike their inorganic counterparts, parylene-C devices are prone to poor adhesion at either the dielectric to dielectric interface or at the dielectric to metal interface (Sharma and Yasuda 1982; Yasuda et al. 2001; Seymour et al. 2009).
Research on alternative parylene materials have promising results. Complementary reactive parylene layers of poly(4-aminomethyl-p-xylylene-co-p-xylylene) and poly(4-formyl-p-xylylene-co-p-xylylene) were discovered and synthesized in the laboratory of Joerg Lahann and found to improve adhesion in microfluidic applications (Chen et al. 2008). In another study we evaluated these same materials, PPX-CH2NH2 and PPX-CHO, and found they improved wet metal adhesion and maintained higher electrical impedance relative to parylene-C alone (Seymour et al. 2009). These reactive parylene materials are used to interface the metal traces in this work, however, more research is necessary to understand their mechanical and biological properties over time and across temperatures. Fortunately, one advantage of this process is that alternative dielectric materials can be used instead of the reactive parylene if necessary. Compatible dielectric materials are limited primarily by the temperature of deposition. Researchers have reported successful combination of polymer and low-temperature inorganic dielectrics such as PECVD SiO2, Si3N4 (Weaver et al. 2002; Lewis et al. 2004; Wuu et al. 2005) or SiC (Chiang et al. 2003; Cogan et al. 2003) all can be deposited below 220°C.
One electrode design in particular that drove the development of a new process was the concept of a double-sided edge electrode on an ultra-thin substrate. In a previous histological study (Seymour and Kipke 2007), we hypothesized that if a structural feature size is smaller than a reactive cell body (<7 μm), the resulting encapsulation would be mitigated by the prevention of cellular spreading. We investigated this relationship between size and tissue reactivity using parylene structures and found the cellular density around the sub-cellular feature was less than half of that around the probe shank. Our interpretation of the biological response suggested that an optimal electrode design is a two or three-sided microelectrode on the edge of thin parylene substrate (5-μm thick). We present here the first microfabrication approach capable of implementing such a design which we hope may be used in future long-term performance studies.
In general, the scope of this work was to develop a polymer-based technology platform that enabled a greater design space for microelectrodes and structural features. The platform needed a dynamic structural range including a thick shank for insertion strength or reservoir capabilities, as well as thin, flexible structures where electrodes could be positioned. The electrode placement also needed to be top-side and bottom-side to enable the edge electrode and higher density arrays. Finding a planarization process that enabled the implementation of double-sided electrodes was the critical process step. We used chemical mechanical polishing (CMP) of parylene-C over SiO2 to construct a planar insulation layer with electrode openings facing toward the wafer. CMP of parylene is rare, but one investigation showed that it was possible with parylene-N and resulted in only a small increase in oxygen content (Yang et al. 1997). CMP of polyimide is more common and may be a useful alternative (Hyoung-Gyun et al. 2000). We found the most significant challenge in polishing parylene was the prevention of delamination, which we discuss in detail. Finally, we successfully recorded neural spike activity in the rat neocortex in acute experiments.
The microfabrication for these structures required six masks and was conducted on 4” silicon wafers in the Lurie Nanofabrication Facility at the University of Michigan. For simplicity, we describe the process in eight steps (Fig. 1(a-h)). A variety of neural devices were included in the mask set but the cross-section shown in the illustrated process was taken from a polytrode recording array that included face-down electrodes, face-up electrodes, and edge electrodes (Fig. 3(a)).
Figure 1(a). Five microns of high temperature (910°C) SiO2 was deposited on a clean Si wafer. This layer served a dual purpose—it provided a sacrificial release layer and formed the face-down electrode/bond pad place holders. The Al mask was patterned using standard lithography and a wet etchant. The SiO2 was patterned using the Al mask and a dry C2F6 plasma etch (LAM Research TCP9400,Fremont, CA). The Al was removed and the wafer was cleaned after an etch depth of 2.8 μm was achieved.
Figure 1(b). An adhesion promoter, A-174, was dip coated on the wafer surface and rinsed. Next parylene-C was CVD polymerized (Specialty Coating System PDS2010, Indianapolis, IN) on the oxide. Deposition was 4-μm thick.
Figure 1(c). Chemical mechanical polishing of parylene-C was primarily mechanical in nature until the SiO2 was exposed. It is important to minimize shear stress in the parylene film. We discovered the likelihood of film delamination was reduced by using short polish steps (less than 4 minutes), dehydration baking between polishing, minimal carrier down force, and the prior use of an adhesion promoter. CMP was carried out with an IPEC Avanti 472 (Axus Technology, Chandler, AZ) using a KOH silica slurry (0.2 μ, pH = 10, 10% solid load provided by Planar Solutions, Adrian, Michigan). Down pressure and back pressure gradually increased to a maximum of 24.1 kPa (3.5 psi) and 10.3 kPa (1.5 psi), respectively. Parylene thickness was measured using interferometry (Nanospec/AFT, Nanometrics) and the calculated polish rate at full force was 220 nm/min. Between polish runs, wafers were cleaned using a scrubber and NH4OH solution (OnTrak Systems, Fremont, CA) then baked at 90°C for 10 min in a nitrogen purge. Planarization results on a variety of features and spacing were tested (Fig. 2). We achieved 2 μm features and larger with sufficient uniformity. Since our metallization layer was 4200 Å, the 200-1200 Å profile was sufficiently small for this process.
Figure 1(d). An oxygen plasma was used to clean and roughen the parylene-C undercoating (Jun et al. 2001; Pornsin-Sirirak et al. 2002). This plasma cleaning step occurred before each CVD step throughout the process to promote cleanliness and adhesion between layers. The first interfacial parylene layer, PPX-CH2NH2 was polymerized using custom built CVD equipment described previously (Nandivada et al. 2005), although it could be deposited with commercially available equipment as well. The target thickness of PPX-CH2NH2 was 100 nm (80-130 nm typical). The precursor was provided by Joerg Lahann (University of Michigan) with synthesis details described elsewhere (Klee et al. 2004). Peaks at 3361 and 3301 cm-1 in the IR spectrum were used to confirm the presence of the N-H bonds after deposition. Electrode or bond pad openings in the PPX-CH2NH2 layer were created with a photoresist mask and an oxygen plasma etch (Unaxis Plasmatherm 790).
Figure 1(e). Next, metal lift-off (SPR220) was used to pattern the Cr/Au/Cr (100/4000/100 Å). The SPR220 resist surface was treated with tetramethylammonium hydroxide (TMAH) to create an inhibition layer (Redd et al. 1999). Metal was deposited using an e-beam evaporator (Enerjet). The resist was removed with acetone. Following lift-off, wafers were immediately cleaned in fresh acetone, IPA, then spin rinsed for 5 minutes.
Figure 1(g). Titanium was deposited 1000-Å thick (evaporator) and patterned (20:1:1 DI:HF:H2O2) to provide a mask during the final oxygen plasma etch of each structure (performed in Fig. 1(h)). SU-8-2025 (Microchem, Newton, MA) was spun on to 40-μm thick and patterned with SU-8 developer (Microchem) to create the core of the probe shank. The SU-8 was cured for 15 min at 150°C. A final parylene-C layer was CVD polymerized (2 μm) which encapsulated the SU-8. We then spun an 80-μm thick AZ-9260 layer (2 step spin) and patterned to create a mask for the thick shank structures. Fine features were achievable on the thin parylene platform (5 μm thickness) using the Ti mask while low resolution features were masked with 9260 resist.
Figure 1(h). The parylene stack was etched using an oxygen plasma to finalize the structural features. 9260 and the Ti mask were removed. Cr on the electrodes and bond pads was also wet etched to expose the Au. Finally, the entire structure was released in buffered hydrofluoric acid. SiO2 removal required about 3 hours then devices were filtered and thoroughly rinsed.
Heat treatment and assembly
Released devices were baked at 140°C for 3 hours to improve adhesion at the surface interfacial boundaries. This heat treatment was previously shown to create an imine bond at the PPX-CH2NH2 to PPX-CHO interface (Chen et al. 2008) and to improve polymer-metal adhesion (Seymour et al. 2009). Neural devices were mounted on a custom PCB employing gold ball studs via thermosonic bonding (Meyer et al. 2001). A protective coating of silicone (MED-4211, Nusil Technology) and epoxy were carefully applied to protect the bond region and PCB.
2.2 Electrode preparation and PEDOT-PSS electrodeposition
Variability in electrode impedance indicated organic residue or metal oxidation was an issue with bare gold electrodes. We used chemical and electrochemical cleaning to lower impedance and reduce variability with a protocol developed for gold electrodes (Rafaela Fernanda Carvalhal et al. 2005). Devices underwent a hot piranha etch for 10 min followed by “electrochemical polishing” in a 0.5 M H2SO4 solution. The polishing protocol consisted of 15 sweeps from -0.1 to 1.2-V vs. SCE (100 mV/s). Finally each device was dipped into ethanol (95%) for 30 minutes before PEDOT deposition.
We selected poly(3,4-ethylenedioxythiophene) (PEDOT) and anion polystyrene sulfonate (PSS) to modify the microelectrodes and lower site impedance. Electropolymerization was carried out in a solution of 0.01 M EDOT monomer (Sigma-Aldrich) and 0.1 M PSSNa (Acros, M.W. 70,000). A current density of 2 pA/μm2 was applied (galvanostatic mode) until a charge density of 1 nC/μm2 was delivered for each site. An Autolab PGSTAT 12 (Eco Chemie, Netherlands) with associated GPES software was used for electropolymerization control.
2.3 Electrochemical impedance spectroscopy
The electrochemical impedance measurements were performed using an Autolab PGSTAT 12 with associated FRA software. The spectral range used for testing interdigitated electrodes (Fig. 5) was 100,000 to 1 Hz and with a single sinusoid at once. All voltage waveforms had an amplitude of 25 mVrms. All impedance measurements were carried out in a Faraday cage in order to minimize external interference. Soak testing was conducted using phosphate buffered saline (PBS, 1X which contains 137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, 2 mM KH2PO4) between 55-65°C for one hour. Heated water was continually monitored and exchanged with water in a large bath surrounding the dish containing PBS, and test device. A two-electrode setup was used for the interdigitated electrodes.
2.4 Edge electrode finite element model
Using COMSOL version 4.0a, we created a 3-D geometry including an insulating substrate, a current sink (modeling a neuron), the extracellular space (modeling tissue), and three different electrodes. Three electrodes were tested: (1) a planar electrode 0.5-μm thick, 15-μm square and positioned 25 μm from the edge, (2) an edge electrode (0.5-μm thick, 7.5-μm wide and 15-μm tall and exposed on both sides, and (3) an edge electrode (5.0-μm thick). We measured the potential at the planar and edge electrodes while moving the spherical current sink to various angles. A neuron was modeled as a 450 μA/cm2 current sink which is the maximum conductance of a Na ion channel, (Lewis 1968)) and positioned 50 μm from the electrode and substrate at increments of 45 degrees (Fig. 6(e,f)). The current sink was 20 μm in diameter (Haug 1987). The tissue was modeled as a large cube (3,000-μm on edge) and was given a conductance of 0.405 S/m (Logothetis et al. 2007). The ground (V = 0) was defined to be the surface of a cube defining the tissue space. The electrode material was assigned a conductance of 8.9e6 S/m (typical for platinum) and defined as having a floating potential. The substrate or shank was 2000-μm long, 100-μm wide, and 5-μm thick. The substrate was defined as an electric insulator. The planar electrode was placed on the surface of the substrate rectangle (Fig. 6(a)) and the edge electrodes were placed inside of a recess at the edge of the substrate (Fig. 6(b-c)) similar to the edge electrodes described here (Fig. 4(c)).
2.5 Acute surgery and recordings
Neural probes were surgically implanted in male Sprague Dawley rats (300–400 g). Anesthesia was administered using intra-peritoneal injections of a mixture of ketamine, xylazine, and acepromazine. The craniotomy spanned approximately 4 mm in the anterior–posterior direction, and 4 mm in the medial–lateral direction and was centered over the M1 and M2 motor cortex (Ludwig et al. 2006). A stereoscope was used to ensure a nearly orthogonal trajectory and avoid any visible surface blood vessels. Immediately before a probe was inserted via a manual stereotaxic drive, the anesthetized subject was moved to a Faraday cage, and then the probe tip was driven to a depth between 1 and 2 mm. All procedures strictly complied with the United States Department of Agriculture guidelines for the care and use of laboratory animals and were approved by the University of Michigan Animal Care and Use Committee.
Recorded neural signals were acquired using a TDT RX5 multi-channel acquisition system (Tucker-Davis Technologies). Neural electrophysiological recordings for all sixteen channels were amplified and bandpass filtered (450-5000 Hz) and sampled at ~25 kHz. Objective classification of high speed neural activity was performed using an algorithm based on an automatic threshold and principal component analysis (Ludwig et al. 2008).
3 Results and discussion
3.1 Parylene-based neurotechnology platform
An important goal of this research was to create an edge electrode having a top and bottom surface and the potential for a third surface area (at the edge) through electrodeposition (Fig. 3(i,j)). This feature allowed us to create the smallest footprint (85 μm2) of a functional recording electrode to date. This new electrode design should theoretically increase signal amplitude by reducing the spatial footprint (Moffitt and McIntyre 2005) while achieving a greater surface area than a conventional planar geometry would allow (Fig. 3(j)). In addition to edge electrodes, we provide several examples of single or double-sided electrode geometries (Fig. 3(e-i)) that may also be useful when placed on a thin parylene platform.
3.2 Microelectrode impedance
Electrode Geometry and Impedance (1 kHz)
Au Sitea, MΩ
POST Implantc, kΩ
N = 48
N = 33
N = 15
Planar 17 × 17 μm
5.4 ± 3.7
240 ± 110
380 ± 82
Edge 17 × 5 × 2 μm
6.0 ± 3.1
210 ± 64
460 ± 72
The increase in geometric and electrochemical surface area provided by electrode surface modification is critically important on this size scale. Thermal noise is proportional to the square root of the real part of the impedance, therefore, a surface modification technique is required to decrease the noise level when attempting to discriminate single unit activity. In a few attempts to record single units with bare gold sites of either 280 μm2 or 170 μm2, we found the thermal noise to be more than double the PEDOT modified sites (discussed below). A number of other techniques to increase the electrochemical surface area of an electrode have been reviewed elsewhere (Cogan 2008) and could be used here.
3.3 Interconnect impedance
In a previous study, we presented soak data at 37°C for 60 days using the same parylene-C barrier layer and reactive parylene interfacial layers tested here. After one hour the impedance dropped to its approximate minimum value (Seymour et al. 2009) and actually increased slightly over the 60 day period (Fig. 5(b)). Other researchers have estimated that water molecules will diffuse 15-μm/min in parylene (Yasuda et al. 2001). Our results here also show a rapid, asymptotic decline of low frequency impedance as well. While the magnitude of the low frequency change is large, the final modulus remains stable and much higher than the electrode impedance. In the frequency range critical for neural recording, 500-5000 Hz, the impedance modulus decline was small.
3.4 Edge electrode finite element model
The double-sided edge electrode is a unique alternative to conventional planar microelectrodes and has the potential to double the microelectrode density. To achieve a third side or thick, laterally facing edge requires an extra post-fabrication step to electrodeposit a thicker electrode material. One may wonder if a thicker, laterally-facing surface would increase sensitivity to lateral neural activity? Would a thicker electrode increase the viewing angle? A three-dimensional finite element model was created to evaluate the affect electrode position and electrode thickness has on the sensitivity of relative neuron placement. We discuss our finite element model and previously reported microelectrode field models.
The model results suggest that there is no advantage of a three-sided edge electrode, i.e. one with thick electroplating beyond the intention to decrease impedance and therefore reduce thermal noise. The model also suggests a potential tradeoff exists between two or three-sided edge electrodes versus planar electrodes that should be further investigated. If this phenomenon is confirmed experimentally, a wide viewing angle may be advantageous in some applications whereas the larger signal amplitude for fewer neurons may be preferred in others.
3.5 Neural recordings
We present a CMP-parylene process that was successfully employed for the microfabrication of a variety of neural probe designs. A large design space was created most notably by providing a dynamic range of structural stiffness and enabling electrode placement on the front, back, or edge. Preliminary recording data indicates that the small edge electrode design is functional and may be an improved alternative to the conventional planar microelectrode. The implementation of novel edge microelectrode arrays will be used in future testing with the hope of significantly increasing the recording volume and performance over time.
We are grateful to Prof. Joerg Lahann and Dr. Yaseen Elkasabi for providing the reactive parylene dimer and for conducting the CVD polymerization in their laboratory. Ning Gulari provided the critical idea of using CMP in this process and other helpful conversations. Drs. Pilar Herrera-Fierro, Hung-Chin Guthrie, and Ramin Emami shared their considerable CMP expertise which was vitally important. Dr. Kip Ludwig shared his method and software for automatic neural spike sorting. Dr. Mohammad Abidian and Eugene Daneshvar engaged in many helpful discussions regarding conductive polymers. We gratefully acknowledge support from the NIH P41 Center for Neural Communication Technology (EB002030) through the NIBIB.