Journal of Neuroscience Methods 153 (2006) 147153

A MEMS fabricated exible electrode array for recording surface eld potentials
Brian A. Hollenberg a , Cecilia D. Richards a , Robert Richards a , David F. Bahr a , David M. Rector b,
b a School of Mechanical and Materials Engineering, Washington State University, Pullman, WA 99164, USA Department of Veterinary Comparative Anatomy Physiology and Pharmacology, Washington State University, 205 Wegner Hall, Pullman, WA 99164, USA

Received 15 June 2005; received in revised form 1 September 2005; accepted 24 October 2005

Abstract We developed a method to microfabricate exible electrode arrays on a thin Kapton substrate, which was engineered to minimize trauma when inserted between the dura and skull to obtain surface EEG recordings. The array consisted of 64 gold electrodes, each 150 m in diameter on a 750 m spaced 8 8 grid. Using photolithographic procedures, any arrangement of electrodes can be implemented. We used the electrode array to record evoked response signals to create topographical maps of the whisker barrels on the cortical surface with excellent signal stability over a period of 8 h. The materials used for this fabrication are potentially biologically inert and, with some additional modications to the design, can be chronically implanted with minimal side effects. Retinal prosthesis, human neurosurgery, and neurological research are all limited to some degree by the resolution and biological compatibility of the implants used. This type of array could greatly enhance the spatial resolution, signal quality, and stability of implantable surface electrode arrays. 2005 Elsevier B.V. All rights reserved.
Keywords: Human neurosurgery; EEG; Cortical mapping; Microfabrication; Kapton; Evoked response

1. Introduction Cortical surface eld potentials are a powerful tool for mapping electrical activation non-invasively across large tissue regions with high temporal resolution. A number of studies have taken a detailed look at localized EEG using high density electrode arrays for surface potential mapping (Jones and Barth, 1999; Huber et al., 2004; Rector et al., 2005) which demonstrate novel concepts in the local structure of cortical activity. Additionally, recent developments in high density electrode arrays for cortical mapping and ne-wire recordings from freely moving animals underscore a major need for many electrophysiological channels in neural network analysis and especially in behavioral studies (Bragin et al., 2000; Patil et al., 2004). For human neurological imaging, EEG scalp recordings typically utilize 64256 electrode positions. However, brain mapping using external electrodes has poor spatial resolution for

Corresponding author. Tel.: +1 509 335 1587; fax: +1 509 335 4650. E-mail address: drector@vetmed.wsu.edu (D.M. Rector).

source localization. Unfortunately, better source localization is achieved only by placing the electrodes closer to the tissue: either directly in contact with the tissue surface, or penetrating into the tissue. For example, neurosurgery often requires detailed maps of brain surface function before resections of damaged or tumor tissue. For this application, surgeons use surface electrode arrays embedded in exible plastic sheets placed on the brain surface. In order to access large brain surface regions, signicant portions of the skull must be removed, introducing additional trauma to the tissue. If a surface electrode array could be inserted through a slot cut into the bone, then the trauma to the patient would be minimized. Many studies have shown that surface eld potentials, using 200 m or smaller electrodes, from rodent brains can effectively map the representations of cortical activation and EEG surface eld potentials with high spatial and temporal resolution (Jones and Barth, 1999). A number of electrode arrays have been developed for both acute and chronic/awake preparations in animals (Rousche et al., 1999; Kipke et al., 2003). However, all of these electrodes are either penetrating unit electrodes, or require large portions of the skull to be removed in order to expose enough

0165-0270/$ see front matter 2005 Elsevier B.V. All rights reserved. doi:10.1016/j.jneumeth.2005.10.016

148

B.A. Hollenberg et al. / Journal of Neuroscience Methods 153 (2006) 147153

of the brain surface for mapping (Jones and Barth, 1999). Large skull openings, however, are difcult to maintain chronically in small rodents and many stainless steel EEG screws can be problematic in thin bone. In either case, it is difcult to maintain such chronic preparations for extended time periods, and another type of electrode array is needed. Several laboratories have developed electrode arrays using exible substrates, (Takeuchi et al., 2004; Grumet et al., 2000) primarily for insertion into cortical tissue to minimize the trauma associated with placing rigid objects into tissue. These approaches utilize polyimide compounds as exible substrates, but are not capable of recording from tissue surfaces. Our objective was to develop a device which could record from a large surface of cortical tissue with minimal trauma for long term recordings. Since the dura is relatively thin in rodents, such surface measurements would be less prone to volume diffusion of the signal. We tested the array on the surface of the rat somatosensory cortex and created two dimensional maps of the whisker cortical columns (barrels). 2. Materials and methods The fabricated electrode array was an 8 8 grid, covering a 5.25 mm 5.25 mm area, using three basic components: a Kapton substrate, a patterned electrical circuit, and a photoimagable lm to protect and electrically insulate the leads, where needed. A polyamide, Kapton lm (Dupont, Wilmington DE) was chosen as the substrate due to its exibility, excellent chemical resistance, and retention of mechanical and electrical properties over a wide temperature range. The electrical circuit was constructed from patterned gold that connects the 64 electrodes to 64 contact pads which are protected by a photoimagable,

chemically amplied, epoxy-based negative photoresist, SU-8 (MicroChem, Newton, MA). SU-8 was chosen for its high aspect ratio imaging, thermal stability, and good chemical resistance. The procedures in this process were carried out in a class 1000 clean room to promote adhesion between layers and reduce surface contamination. Using Kapton tape, a 50 m thick Kapton substrate was attached to a glass slide to provide a rigid structure to endure the fabrication processes. The substrate was then prepared by an acetone, isopropanol, deionized water wash, and an RF plasma etch (PE-2000 Plasma Etching system, South Bay Technology INC, San Clemente, CA). A DC Magnetron system (Edwards Auto 306, Wilmington, MA) was used to sputter a 5 nm layer of a titanium-tungsten alloy onto the Kapton substrate, followed by a 300 nm thick gold layer. The titaniumtungsten alloy layer was used as an intermediary layer between the Kapton lm and the gold to promote adhesion. The sample was then plasma etched again. Photolithography was used to pattern the gold to produce the circuit. First a thin layer of hexamethyldisilazane (Fluka, Switzerland) was applied on the Kapton substrate using a spin coater. This was followed by a layer of AX5214-EIR (Clariant, Somerville, NJ) photoresist. The laminate was then baked on a hotplate. The hexamethyldisilazane was used as an adhesion promoter layer between the photoresist and the gold. The patterning of the photoresist was performed by ooding the photoresist with ultraviolet light through a mask which changes the chemistry of the photoresist making it soluble for etching. The unwanted photoresist was removed using a bath of AZ-400K Developer (Clariant Somerville, NJ). The unwanted gold was removed using a gold etch bath (Gold Etch-Type TFA, Transene Company, Danvers, MA). The titaniumtungsten alloy was removed by hydrogen peroxide bath in the same pattern

Fig. 1. Resulting gold circuit on Kapton substrate. The electrode array mask shown is the resulting gold circuit that was left on the Kapton substrate. The large circles on the upper left and upper right are the alignment targets. These targets are used to align the electrode array into a drilling jig. The smaller circles arranged at the bottom of the mask in an 8 8 grid are the electrodes. SU-8 was blanketed over the entire gold circuit. It was then patterned using photolithography to expose the electrodes and the connection leads at the top of the mask.

B.A. Hollenberg et al. / Journal of Neuroscience Methods 153 (2006) 147153

149

as the gold. Finally, an acetone wash was used to remove the photoresist leaving behind the completed gold circuit shown in Fig. 1. We used targets on both sides of the electrodes for alignment purposes. Photolithography was also used to pattern the SU-8 electrical insulating layer. To prepare the Kapton and gold sample for SU-8, the sample was plasma etched, washed with acetone, isopropanol, and deionized water, and then baked to remove all remaining organics. A 13 m layer of 2010 SU-8 photoresist was spun on top of the Kapton and gold circuit, using a spincoater, then baked to crosslink the SU-8 bers. A mask was used to expose everything with UV light except the gold connection pads at the top, and the smaller gold electrodes at the bottom. The sample was again baked and the unwanted SU-8 was removed using SU-8 developer bath (MicroChem, Newton, MA). The sample was nally washed with acetone, isopropanol, and deionized water, dried, and baked again to insure the thermal properties of the SU-8. A schematic representation of the clean room fabrication process is shown in Fig. 2. A close-up photograph of the electrode array surface is shown in Fig. 3. All layers resulted in a total device thickness of 75 m. Once the cleanroom fabrication of the electrode array was nished, an electrical connection between the electrode array and the test equipment was made. To accomplish this connection, a corresponding circuit board was constructed that matched the exposed, gold connection pads. Electrical connection between the electrode array and the corresponding circuit board was obtained by using a Carbon Zebra Elastomeric Connector (Fujipoly, Tokyo, Japan). The Zebra connector consisted of layers of conductive material and non-conductive elastomeric material alternating at a pitch of 50 m. The Zebra connector was sandwiched between the circuit board and the electrode array. Fiducial holes drilled in both the circuit board and the Kapton of the electrode array were tted rmly around alignment pins on a clamping jig to ensure that the ne lines of the electrode array and the circuit board matched up correctly. The whole assembly was then clamped together. A photograph of the nished electrode array is shown in Fig. 4. The electrical characteristics and the integrity of the electrodes were evaluated by measuring the impedance of each of the 64 gold traces from electrode to connection pad in a 0.9% saline solution. These electrical attributes were determined prior to experimentation with rodent subjects. The average impedance was 225 k at 1 kHz with a standard deviation of 90 k . In addition, impedance measurements between traces were conducted to ensure electrical isolation. To ensure that there was correct alignment of the zebra strip between the electrode array and the circuit board, a probe station linked to an impedance analyzer was used. Each lead of the electrode array was checked to guarantee that the channels were properly aligned to the traces of the circuit board. 3. Experiment Four male SpragueDawley rats (300400 g) were implanted with an electrode array for testing. All procedures were approved by The Animals Care and Use Committee at Wash-

Fig. 2. Process steps in fabrication of the electrode array. (A) The 50 m Kapton substrate was cut into 75 mm 25 mm rectangle to match a glass slide (not shown here), which was used as a rigid structure for fabrication purposes. (B) A 5 nm titaniumtungsten alloy was sputtered onto the Kapton substrate, followed by a 300 nm layer of gold. (C) The gold and titaniumtungsten alloy was then patterned using standard photolithography processes. The resulting gold circuit, shown in Fig. 1, consists of an 8 8 grid of 64 electrodes and 64 connection pads. (D) SU-8 was then applied using a spin coater over the entire sample. (E) SU-8 was then patterned to expose each of the electrodes and their corresponding connection pads. This layer was used to electrically isolate each one of the electrodes of the electrode array.

ington State University. The animals were anesthetized with ketamine/xylazine (100 and 10 mg/kg, respectively) and body temperature maintained at 37 C. Twenty percent of the initial dose was administered as supplemental anesthesia when needed. The cardiac activity and respiration was also monitored during the experiment through a pair of multi-stranded stainless steel wires that were inserted subcutaneously on both sides of the lower rib cage. The animals were then placed in a standard sterotaxic frame, and we drilled a 6 mm 1 mm slit in the rostro-caudal direction between bregma and lambda just lateral to the midline placement of the electrode array (Fig. 5). The midline edge of the craniotomy was beveled to assist in the insertion of the probe. Using a thin, exible, plastic knife inserted under the bone, the dura was separated from the skull to promote an easy electrode array insertion. Occasionally, some minor bleeding

150

B.A. Hollenberg et al. / Journal of Neuroscience Methods 153 (2006) 147153

Fig. 3. A close up of electrode array surface. The lighter color in this gure shows the gold traces. Each of the circles at the end of the gold traces are isolated electrodes. The darker region of this picture is Kapton substrate. There is a blanketed SU-8 layer over the entire surface, except for each of the exposed circle electrodes. The electrodes are an array of 150 m spots with 750 m spacing.

Fig. 4. Electrode set up. The basic components of the electrode array consist of the circuit board, the Kapton electrode array, the zebra connector and the clamp. The assembly was clamped together using an aluminum and acrylic aligning jig.

would occur by severing vessels that lead from the bone to the dura. This bleeding stopped within a few minutes. The electrode array was then slipped between the dura and the skull over the whisker barrels somatosensory cortex on right side of the animal. During a 4 h recording each animal was exposed to repetitive deection of a single whisker on the contralateral side

using a 0.2 ms, 012 V signal pulse to an 8 speaker-based whisker actuator. The whisker was placed into a 10 cm length of hypodermic tubing which was in turn attached to the speaker. Each whisker twitch was randomized between 1 and 2 s to avoid accommodation. Surface eld potentials recorded by the 64 channel electrode array were continuously collected with a 16 bit digitizer at 20 kHz (Rector and George, 2001). Typically, 20 responses were averaged and stored for each whisker tested. Using a laser based optical lever deectometer (OFV-5000 Vibrometer Controller and OFV 511 Fiber Interferometer,

Fig. 5. Schematic diagram of electrode array placement. A 6 mm 1 mm slit was drilled in the rostro-caudal direction in the animals skull between the bregma and lambda, just lateral of the midline. The dura was separated from the skull using a thin plastic knife to ease insertion of the electrode array. The electrode array was inserted to covered whisker barrels on the right side of the animal, in the region between the temporal ridge and the midline.

B.A. Hollenberg et al. / Journal of Neuroscience Methods 153 (2006) 147153

151

Polytec, Waldbronn, Germany), the absolute deection of the hypodermic tubing was measured during stimuli. In order to assess the effectiveness of various stimulus amplitudes, we recorded the evoked response after driving the speaker with 0, 2.0, 3.7, 5.3, 7.0, 8.7, 10.3, and 12 V amplitude pulses of 0.2 ms width. We then plotted deection versus drive voltage, and evoked response amplitude versus deection distance. Evoked responses were also recorded while driving the speaker, but without the whisker in the tubing to test possible auditory stimulation. 4. Results Using the electrode array to map the whisker barrels, we selected an electrode which exhibited the largest and earliest response when whisker Bl was twitched. Using this electrode

channel, an averaged whisker response during various stimulus amplitudes is shown in Fig. 6. The traces show response characteristics for the P1, N1, P2, and N2 components of the response which are clearly visible with good signal-to-noise ratio. The speaker deection distance was directly proportional to the drive voltage up to 10.3 V after which the deection amplitude appeared to saturate. Evoked response amplitude was also proportional to the deection amplitude, however, the P1 response saturated at 30 m deection while the Nl response continued to rise until 100 m deection. Since the stimulus pulse width was kept constant, the deection amplitude was also proportional to the deection velocity which may correlate better to neural responses (Shoykhet et al., 2000). Fig. 7 shows 2D maps of evoked responses across all electrodes at the time of maximum P1 amplitude.

Fig. 6. Averaged evoked responses. Time triggered average evoked responses with increasing deection amplitudes show that the evoked response potential is proportional to the deection amplitude. Since the stimulus pulse width was kept constant, the deection amplitude was also proportional to the deection velocity. Each averaged response was generated from 20 stimuli. When the speaker was driven with 12 V alone, as an auditory control without twitching the whisker, a small response can be observed, possibly resulting from auditory detection of the speaker noise. Since the animal had stereotaxic ear bars in place, we expect minimal auditory contributions. The bottom trace shows the absolute deection of the hypodermic tubing during a typical 7.0 V stimulus. The speaker displacement is much longer lasting than the 0.2 ms drive pulse, and a low frequency damped oscillation follows for 200 ms after the pulse. In the upper right corner, we plot speaker drive voltage vs. deection amplitude showing that the peak-to-peak speaker response is linear with the drive voltage up to 10.3 V, when the speaker response saturates. In the lower right corner, we show that the evoked response is proportional to the deection amplitude, with the P1 component reaching saturation at 30 m deection and the N1 response saturating at 100 m deection.

152

B.A. Hollenberg et al. / Journal of Neuroscience Methods 153 (2006) 147153

Fig. 7. Spatial amplitude maps of averaged evoked responses for 24 whiskers. The averaged evoked response of the P1 surface potentials were gathered for each of the individual electrodes of the 8 8 electrode array after twitching 24 different whiskers. The amplitude distributions of evoked responses were plotted in a 3D mesh form to create maps of the cortical surface, used to establish the location of the whisker barrel being activated. The z-axis shows the cortical voltage potentials in microvolts. The x- and y-axis are the rows and columns of the electrode array.

5. Discussion We demonstrated the design, fabrication, and experimental use of a exible polymer electrode array on the surface of the rat cortex. For high density mapping of surface eld potentials, this technique represents an effective method with less potential trauma than existing methods which must remove signicant regions of the skull. The photolithographic process also allows virtually any contact pattern over the two dimensional array, as well as, variable size, and shape. Further, it is expected that these devices can be mass produced at low cost. Maintaining gold adhesion during the fabrication of the ducial holes (i.e., drilling) was the greatest challenge in the production of these prototypes. In the original design, a Zebra connector was not used. The electrical connections were made by drilling next to a gold connection pad on the electrode array, pressing a pin through the Kapton, and using conductive epoxy to complete the connection between the electrode array and the connection pins. The thin gold lm often peeled off and yield was low. In general, the results using this technique were very inconsistent. Implementation of the Zebra connector eliminated these difculties and resulted in high-yield, consistent devices. Gold was used for the electrode array because of its stability in tissue and ease of manufacturing. It may be possible to improve the impedances of the connections by using platinum black deposits for the electrodes contact area (Grumet et al., 2000). The precise location of the electrodes may be determined, after the experiment, by passing a small amount of current to make a mark by cauterizing the tissue at the electrode.

Even though the method described here does not allow single unit recording as with penetrating wire electrodes, we can nevertheless, still record high frequency bursts from the cortical surface which correlate to population spikes generated below the surface (Jones and Barth, 1999). If clear substrates are used in the manufacturing process, optical techniques can be used in combination with electrode arrays for simultaneous surface eld potential mapping and optical mapping of neural responses using spectroscopy, scattering, or vital dyes. Without penetrating the dura, tissue trauma is reduced, and probes can potentially remain viable for extended time periods. However, there are several limitations in chronic applications of the probe. First, a better selection of materials may lead to increased bio-compatibility. While the bio-compatibility of SU8 and other types of coatings on Kapton have shown excellent results (e.g. Cosofret et al., 1994; Voskerician et al., 2003), further testing is required in long term studies to demonstrate that there are no long term effects of these compounds. Second, since the circuit board is somewhat bulky, a truly chronic implementation cannot be achieved until the amplier and multiplexer chips can be bonded onto the Kapton array. Several groups are currently fabricating such amplier systems with promising results (Guo et al., 2004; Wise and Naja, 1991). The exible technology developed here also has many attractive applications for clinical use. Neurosurgical planning and mapping with electrical eld potentials can be performed without removing large skull regions, as is necessary for existing plastic arrays with embedded electrode buttons. Additionally, higher density arrays are possible for more accurate mapping.

B.A. Hollenberg et al. / Journal of Neuroscience Methods 153 (2006) 147153

153

Prosthetic devices such as cochlear and retinal implants will also benet from further development of this technology which will characterize materials for long term chronic applications. Acknowledgments The authors would like to acknowledge the help of Mark D. Fuller and Brian J. Stephenson, as well as NIMH 60263 for funding of this project. References
Bragin A, Hetke J, Wilson CL, Anderson DJ, Engel Jr J, Buzsaki G. Multiple site silicon-based probes for chronic recordings in freely moving rats: implantation, recording and histological verication. J Neurosci Meth 2000;98(2):7782. Cosofret VV, Erdosy M, Buck RP, Kao WJ, Anderson JM, Lindner E, Neuman MR. Electroanalytical and biocompatibility studies on carboxylated poly(vinyl chloride) membranes for microfabricated array sensors. Analyst 1994;119(11):228392. Guo H., Champion C.L., Rector D.M., La Rue G.S. A low-power low-noise sensor IC. Microelectronics and Electron Devices, 2004 IEEE Workshop on 16 April 2004, p. 603. Grumet AE, Wyatt JL, Rizzo JF. Multi-electrode stimulation and recording in the isolated retina. J Neurosci Meth 2000;1010(1):3142.

Huber R, Ghilardi MF, Massimini M, Tononi G. Local sleep and learning. Nature 2004;430:7881. Jones MS, Barth DS. Spatiotemporal organization of fast (>200 Hz) electrical oscillations in rat vibrissa/barrel cortex. J Neurophysiol 1999;82(3):1599609. Kipke Dr, Vetter RJ, Williams JC, Hetke JF. Silicon-substrate intracortical microelectrode arrays for long-term recoding of neuronal spike activity in cerebral cortex. Trans Neural Syst Rehabil Eng 2003;11(2):1515. Patil PG, Camena JM, Nicolelis MAL, Turner DA. Ensemble recordings of human subcortical neurons as a source of motor control signals for a brainmachine interface. Neurosurgery 2004;55:2738. Rector DM, George JS. Continuous image and electrophysiological recording with real-time processing and control. Methods 2001;25(2):15163. Rector DM, Topchiy IA, Carter KM, Rojas MJ. Local functional state differences between rat cortical columns. Brain Res 2005;1047(1):4555. Rousche PJ, Petersen RS, Battiston S, Giannotta S, Diamond ME. Examination of the spatial and temporal distribution of sensory cortical activity using a 100-electrode array. J Neurosci Meth 1999;90(1):5766. Shoykhet M, Doherty D, Simons DJ. Coding of deection velocity and amplitude by whisker primary afferent neurons: implications for higher level processing. Somatosens Mot Res 2000;17(2):17180. Takeuchi S, Suzuki T, Mabuchi K, Fujita H. 3D exible multichannel neural probe array. J Micromech Microeng 2004;14:1047. Voskerician G, Shive MS, Shawgo RS, von Recum H, Anderson JM, Cima MJ, Langer R. Biocompatibility and biofouling of MEMS drug delivery devices. Biomaterials 2003;24(11):195967. Wise KD, Naja K. Microfabrication techniques for integrated sensors and microsystems. Science 1991;254(5036):133542.