Penetrating multichannel stimulation and recording electrodes in ...

Hearing Research 242 (2008) 31–41

Contents lists available at ScienceDirect

Hearing Research journal homepage: www.elsevier.com/locate/heares

Penetrating multichannel stimulation and recording electrodes in auditory prosthesis research David J. Anderson * Electrical Engineering and Computer Science, Biomedical Engineering, Kresge Hearing Research Institute, University of Michigan, 1301 East Ann Street, Ann Arbor, MI 48109-0506, USA

a r t i c l e

i n f o

Article history: Received 29 August 2007 Received in revised form 23 January 2008 Accepted 24 January 2008 Available online 31 January 2008 Keywords: Multichannel electrode arrays Neural stimulation Single unit recordings Array data processing

a b s t r a c t Microelectrode arrays offer the auditory systems physiologists many opportunities through a number of electrode technologies. In particular, silicon substrate electrode arrays offer a large design space including choice of layout plan, range of surface areas for active sites, a choice of site materials and high spatial resolution. Further, most designs can double as recording and stimulation electrodes in the same preparation. Scala tympani auditory prosthesis research has been aided by mapping electrodes in the cortex and the inferior colliculus to assess the CNS responses to peripheral stimulation. More recently silicon stimulation electrodes placed in the auditory nerve, cochlear nucleus and the inferior colliculus have advanced the exploration of alternative stimulation sites for auditory prostheses. Multiplication of results from experimental effort by simultaneously stimulating several locations, or by acquiring several streams of data synchronized to the same stimulation event, is a commonly sought after advantage. Examples of inherently multichannel functions which are not possible with single electrode sites include (1) current steering resulting in more focused stimulation, (2) improved signal-to-noise ratio (SNR) for recording when noise and/or neural signals appear on more than one site and (3) current source density (CSD) measurements. Still more powerful are methods that exploit closely-spaced recording and stimulation sites to improve detailed interrogation of the surrounding neural domain. Here, we discuss thin-film recording/ stimulation arrays on silicon substrates. These electrode arrays have been shown to be valuable because of their precision coupled with reproducibility in an ever expanding design space. The shape of the electrode substrate can be customized to accommodate use in cortical, deep and peripheral neural structures while flexible cables, fluid delivery and novel coatings have been added to broaden their application. The use of iridium oxide as the neural interface site material has increased the efficiency of charge transfer for stimulation and lowered impedance for recording electrodes. Ó 2008 Elsevier B.V. All rights reserved.

1. Introduction The natural parallelism and stochastic principles of the nervous system have led neurophysiologists to aspire to interacting with multiple neurons simultaneously. Auditory system physiologists comprised one of the principal driving forces behind the cascade of technological developments that has led us to the present state where massive amounts of data from the nervous system may be processed without being excessively invasive. Numerous advances in technology resulted from pursuing how precisely-timed neural discharges in auditory neurons, organized in exquisite tonotopic order through out the auditory pathways, code the acoustic environment. Advances in electrodes, amplifiers, waveform visualization tools and signal processing methods culminated in the development of specialized data acquisition equipment coupled with laboratory computers to achieve precise * Tel.: +1 734 763 4367; fax: +1 734 763 8041. E-mail address: [email protected] 0378-5955/$ - see front matter Ó 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.heares.2008.01.010

quantification of electrophysiological data. These advances were quickly assimilated. Auditory science demands precise timing of neural events because nowhere in the brain is timing more important than in the auditory system. In addition, few systems have such an easily observed organizational theme as the orderly distribution of the auditory spectrum at every nuclear level. Practitioners of auditory physiology were therefore involved in the earliest vision for multichannel recording technology because the next great challenge was understanding how neurons cooperated to extract information from the auditory environment. In the early 1960s laboratory computers were finding a perfect marriage with auditory neurophysiology. Forward thinking individuals were inspired to invent methods for study of multiple neurons recorded simultaneously. Moise Goldstein and Moshe Abeles pioneered ways to separate extracellular potentials recorded from single channel electrodes (Abeles and Goldstein, 1977), Perkel et al. (1967a,b) invented visualization and computational methods that could be used to extract information from multichannel data, and Starr et al. (1973) gathered colleagues at Stanford to exploit the

32

D.J. Anderson / Hearing Research 242 (2008) 31–41

thin-film methods of the electronics industry to realize multichannel neural recording devices (Wise et al., 1970). The first results from recording in the auditory cortex with thin-film electrodes was reported shortly after the first fabrication process was developed (Starr et al., 1973). The effort at the University of Michigan to perfect the ‘‘Michigan probe” discussed below stemmed directly from Starr’s initiative and Wise’s technical skills. The advent of the auditory prosthesis in the early 1970s with its quick advance to an array of electrodes further inspired the need to electrically visualize the spatial organization of the system and later consider stimulation of the system at other more central centers. It is interesting, and probably a comparative advantage, that neurophysiologists with interests in the auditory system were early leaders in multisite electrode arrays and multichannel processing. With the advent of the scala tympani auditory prosthesis, it was important to evaluate the response of the central nervous system (CNS) to this stimulation. Initially the work was done by passing single channel electrodes through a tonotopic region of the auditory system, usually the inferior colliculus (Merzenich and Reid, 1974). Stimulation of the peripheral auditory system became a vast and extremely successful clinical enterprise starting in the early 1970s. At the same time, little research related to stimulation in the auditory CNS was taking place. This is not surprising considering the peripheral stimulation success and the problems that would be faced in the CNS. Nevertheless, the same thin-film technology that was successful in recording can easily be applied to stimulation. Stimulation of the cochlear nucleus (CN) using silicon multichannel penetrating arrays started in the late 1980s (Anderson et al., 1989; Evans et al., 1989). More recently, these arrays have been more precisely evaluated for CN stimulation (McCreery et al., 2007), the inferior colliculus (Lim and Anderson, 2006), and the auditory nerve (Middlebrooks and Snyder, 2007). Parallel efforts were being carried on in other brain areas with other technologies. While microwires (Nicolelis et al., 2003) and the Utah probe (Rousche and Normann, 1998) are not featured herein, they have made very significant contributions to neuroscience and multichannel recording and stimulation. Progress in auditory systems neuroscience has been considerable in the last three decades but it has been a difficult science to scale up to study of the intact auditory system. Research on the code and coding mechanisms in the nervous system has remained an important subject but without tools to interrogate the massively parallel nature of the nervous system, progress will continue to be slow. Here we explore the thin-film multichannel electrode array, which is one of the devices that has shown some promise for accelerating the discovery process in auditory neurobiology by systematic multichannel recording and stimulation at several levels in the auditory pathway.

2. Micromachined multichannel electrode arrays Scaling the monitoring of single neurons to whole populations of neurons has been a dream of neurobiology since the early discoveries made by recording from single cells with single metal and glass electrodes. The objective is to fabricate recording electrodes with communication electronics by the same precise methods used by the semiconductor industry to create integrated circuits such as the ubiquitous microprocessor. There are several forms of micromachined multichannel electrode arrays that utilize a variety of fabrication methods but we will focus on planar recording and stimulation arrays. The basic features of a successful electrode are the substrate, exposed metal, conductors, and dielectrics. The substrate provides the mechanical support and the shape of the electrode array. Exposed metals provide the localized interface to the nervous system and the interface to the instrumenta-

tion electronics. Conductors, thin-film traces in the case of the Michigan probe, provide the electrical communication within the device. Finally, the dielectrics isolate the internal conductors from the external conducting media and thus allow safe delivery of the signal along the probe. The first version of this probe was developed by Najafi et al. (1985) and soon after reported as a tool for neuroscience (BeMent et al., 1986; Drake et al., 1988). Demonstration for use as a stimulation device in the auditory system was reported later (Anderson et al., 1989). Each of the devices was engineered as they are today to endow individual electrodes with the structure, dimensions and electrical characteristics suitable for stimulation or extracellular recording in each specific application. There have been hundreds of electrode designs fabricated over the years. In addition, several extensions to the technology have become available such as integrated silicon and polymer cables (Hetke et al., 1994; Pang et al., 2005; Yao et al., 2007), fluidic channels (Chen et al., 1997), the implementation of electronic circuits for recording (Ji and Wise, 1992) and stimulation (Tanghe and Wise, 1992) and incorporation of 2D probes into 3D devices (Bai et al., 2000). These features vastly expand the design space of the electrode arrays. While not the only viable technology, our focus will be the Michigan probe which has been the subject or our developments for more than 20 years. 2.1. Fabrication process The substrate upon which the thin-films are deposited is a mater of choice. Silicon (Wise and Angell, 1975; Edell, 1986), ceramic (May et al., 1979; Prohaska et al., 1986; Moxon et al., 2004) and polymer (Pickard et al., 1979; Rousche et al., 2001) have been used. There are positives and negatives to all of these technologies but silicon and polymer substrates have proliferated significantly. Silicon has the advantage of allowing electronics to be fabricated as part of the device as first demonstrated by Wise and Angell (1975). Our discussion will include the silicon substrate devices developed at the University of Michigan under NIH Neuroprosthesis Program (NPP) projects and the polymer substrate developed at Arizona State University. These probes are fabricated at the University of Michigan for distribution either for testing purposes through the NPP program prior to 1994, through the NIH sponsored Center for Neural Communication Technology (CNCT) from 1994 to 2004 and thereafter by NeuroNexus Technologies of Ann Arbor, Michigan. An extensive review of the silicon technology containing much of the detail was published in 2004 (Wise et al., 2004) and the polymer technology is discussed by Rousche et al., (2001). The advantages that make these devices useful and practical are batch fabrication, precision of relative site positions on the substrate, few limitations on planer substrate shape, a variety of tip configurations for different penetration needs, an integrated cable technology (Hetke et al., 1994), and strength sufficient to be inserted into the brain. The fabrication process for the Michigan probe is shown in Fig. 1. With extra fabrication steps, the arrays can be augmented with a hollow channel on the shanks (Chen et al., 1997). These can be used for either stiffening the substrate or delivery of pharmaceuticals. Two key features of this fabrication process which make it so versatile for producing neural probes are selective silicon etching and the ability to pattern the surface features to submicron precision. Selectively etching of a silicon wafer is achieved by patterned boron doping of the wafer to a depth of 5–15 lm at the beginning of the process and the use of an etchant that preferentially etches undoped silicon vs. doped silicon at the end of the process. The pattern of the doping thus produces the gross shape of the device while the depth of the doping determines the thickness and thus the strength and flexibility of the probe. Probes which must penetrate into the brain are usually 15 lm thick while flexible cables

D.J. Anderson / Hearing Research 242 (2008) 31–41

33

icon. In all cases, dielectric layers under and over the interconnects consisting of a tri-layer of silicon dioxide (SiO2), silicon nitride (Si3N4), and silicon dioxide protect the conductors from the surrounding media and balance the mechanical stress in the electrode array. The bonding pads and the tissue interface sites are exposed. In the final release process, neither the dielectrics nor the metal sites are attacked by the etchant. 2.2. Design space

Fig. 1. The process starts with a silicon wafer with an electrode shape defining mask through which a boron diffusion (light blue) occurred. Next, a series of depositions and local removals consisting of a lower passivation layer (blue–pink– blue), conductive traces (red), a top set of passivation layers (blue–pink–blue) and finely contact and sensor sites are deposited. The final step is the etching of the undoped silicon to release the finished probe.

are usually 5 lm thick. Since the doping process is a heat driven diffusion process, the resulting shapes are smooth and do not have sharp edges except adjoining the top surface of the electrode. Adjustments in tip shape and sharpness can be made to accommodate insertion needs. The pattern of the diffusion in the plan view will determine two dimensions of the tip shape while shallow vs. deep diffusion near the tip will determine the tip shape in profile. It is believed that tip shape has an effect on tissue reaction (Edell et al., 1992). Several example designs that have been fabricated are shown in Fig. 2. The general plan of a probe with no cable is a standardized broad area where gold pads are arranged conveniently for wire bonding to a circuit board and one to several thin shanks that have one to several iridium recording/stimulation sites arranged to meet the needs of the intended experiment. Site size and spacing varies among designs but sites range from 100 lm2 to 4000 lm2 with spacing as close as 25 lm. Polysilicon conductors connect the bonding pads to the sites one-to-one. When integrated cables are added to the design, the polysilicon conductors travel over a 5 lm thick long and narrow substrate area to a thick substrate bonding area. Probes designed for stimulation often have metal traces with lower resistively such as platinum or gold to provide for lower voltage drops during stimulation current passage. A difficulty with using these metals as conductors is they preclude the use of high temperature deposition methods for the top dielectrics such as low-pressure chemical vapor deposition. These films are generally more uniform and have better long-term test results. Refractory metals and their silicides are compatible and can be used to reduce the trace impedance by a factor of 20 from polysil-

The previous paragraphs describe the fabrication of a passive silicon device that has been configured in hundreds of different designs. The limits are design rules mostly defining minimum feature sizes and separations among features providing a high probability of successful yields. There are overall size constraints compatible with the wafer size and device yield per wafer as well. As external components such as connectors with small size and large contact counts become available, the site count increases to the current standards of 16 and 32 sites. The probes can be packaged for either acute use or as chronic devices. Chronic probes generally have integrated cables. Fig. 3 shows examples of both acute and chronic packaging. The emphasis of the acute package is ease of use and the availability of large numbers of designs with no particular demand on small connector size or a long-term hermetic seal. Acute probes are low cost primarily because they are reusable. An emphasis placed on chronically implanted probes by users is small size and long life in vivo. This demands connectors that are small but robust and with greater attention to hermetic seal. Because they tend to become integrated into the regrowth of tissue as the surgical field heals their position becomes stabilized. Added capabilities come from a specially designed CMOS process that makes integrated electronics possible on the same substrate with the recording/stimulation arrays. Examples of electronic circuits that have been implemented are amplifiers, selectors, time division multiplexers, programmable stimulators and communication controllers (Wise et al., 2004). More recently, the use of polymers have led to hybrid devices with silicon probes and polymer cables. The core Michigan MEMS group has been responsible for the original process and most of the extensions of the basic process that has increased the design space dimension. Due to early export of the devices to scientists for evaluation and critique, these scientists have been responsible for many of the design details that make the devices truly useful. While most probes used come from a standard catalog, individual projects are still the drivers for many of the most interesting individual design instantiations. Several of the most interesting designs have been driven by auditory prosthesis research (Bierer and Middlebrooks, 2002; Lim and Anderson, 2006, 2007; Middlebrooks and Snyder, 2007; McCreery, 2008). 2.3. Compatibility with common electrode track imaging methods Locating electrode tracks in tissue is a common method of experiment documentation which allows comparison of response with location. All the classic track tracing methods can be used. These require serial sectioning after removal of the probe prior to sectioning. Chemical markers coated on the probe or deposited from the sites and current lesioning have been shown to work well. In some instances, sectioning parallel to and down to within 100 lm of the silicon substrate provides the opportunity to observe the intact relationship of the probe to the surrounding tissue with confocal microscopy (Edwards et al., 1997). This method preserves both a precise geometric relationship between the probe and the tissue and cellular information. Another method of obtaining precise in vivo information is magnetic resonance imaging (MRI). MRI has been shown to provide detail to the pixel level

34

D.J. Anderson / Hearing Research 242 (2008) 31–41

Fig. 2. Several electrode designs are shown on the back of a US Lincoln Penney. Feature sizes in the micron range allow the technology to take on many detailed shapes and thus to address recording and stimulation requirements at many areas within the brain.

Fig. 3. Above are three types of acute packages and below is a 16 channel chronic package. This chronic package makes use of a flexible silicon cable but polymer cables can be used as well.

provided there are no nearby implant system components that distort the imaging system’s magnetic field (Martinez-Santiesteban et al., 2006, 2007). The materials used in components such as connectors must be selected carefully to avoid such distortions. Depending on the magnet field strength, the pulse sequences and the signal processing methods used, resolution of a silicon substrate probe can be to the single tine level. Sites may be visualized by depositing magnetic material on the site surfaces but accurate visualization of the probe body allows one to infer site locations with ease and thus avoiding the use of nonstandard metals on the sites. MR imaging presents many challenges to the design of recording and stimulation electronics. Combining fMRI with physiological recording remains a difficult barrier to surmount. Due to similar X-ray absorption coefficients, silicon does not contrast with neural tissue in X-ray imaging methods so special devices with extra metal are required.

2.4. Recording characteristics The small signal electrode model of Robinson (1968) adequately describes the recording characteristics of the Michigan probe. Fig. 4 shows current passing through a conductive media to cell membranes during action potentials, which generates a potential field in the media with respect to a distant ground. Metal electrodes connected to a high impedance amplifier placed in this field both slightly distort the potential field due to modification of the current path and record the potential. In practice, what is important for gaining information from recordings is the amplitude of the signal recorded and the noise generated by the electrode by thermal effects and unwanted signals such as the combined effect of many low level signals generated by far off neurons. The signal amplitude is dependent on the amplitude of the sink/source and the distance from the source.

D.J. Anderson / Hearing Research 242 (2008) 31–41

Fig. 4. Robinson’s recording model illustrating the flow of current from the media into a cell during depolarization thus producing a potential drop with respect to ground (en) seen at the electrode tip and detected by the recording amplifier. The impedance elements are the spreading resistance (Rs), the electrode/tissue interface (Re and Ce), the interconnect impedance to the amplifier and its shunt capacitance (Rm and Cs) and the amplifier input impedance (Za).

Signal strength is also strongly dependent on the geometries and orientation of the membrane and the electrode. Larger cells pass more current and so generate larger potentials at an equivalent distance from the membrane. Smaller electrodes (the ideal being a point) have higher sensitivity because they can be maneuvered very close to a cell and large electrode substrates shadow the current flow therefore, seeing larger potential changes on the side of the substrate facing the cell and smaller potential changes on the side away from the cell (Anderson et al., 2001). This effect is a factor in attempts to calculate the position of a cell by signal strength at several nearby sites. Noise power from an electrode in comes from a combination of Johnson noise attributed to the spreading resistance and the 1/f noise attributed to the diffusion region of the electrochemical interface (Hassibi et al., 2004). The spreading resistance of a metal disk on an infinite non-conductive substrate in a media with resistivity q and radius r is Rs = q/4r (Kovacs, 1998). Both of these noise contributions increase with diminishing electrode size thus setting up a tradeoff between noise and sensitivity (Lempka et al., 2006). We have found that many individuals who conduct comparative site size tests in their own preparations tend to choose larger site sizes than was first intuitive and improve unit selectivity through single or multiple channel signal processing. Some choose the more traditional method of lowering impedance by surface modification through activation of the iridium metal as explained in a following section. Within limitations, this method allows high sensitivity using a small electrode while having low noise through activation. One should expect a noise floor of 12 lV rms or less in the neural spike bandwidth but activation or large sites can make this number smaller. 2.5. Acute vs. chronic recording Comparable devices packaged as acute or chronic devices have identical initial recording characteristics if introduced into the tissue without damage. Acute probes can be cleaned after each experiment and can have essentially unaltered performance over dozens of uses. Accidental breakage or poor cleaning practice are the most likely failure modes. A long history of performance statistics covering several types of chronic electrodes has shown that recording characteristics diverge significantly from initial characteristics over roughly a month’s time with some deterioration becoming apparent in the first week. There has been research on both cause and counter measures relevant to this important impediment to

35

Fig. 5. The working hypothesis for fouling of recording and stimulation sites involves the deposit of several reactive products on the electrode surface which, because of lower conductivity compared to the surrounding media, interrupt signals to or from the electrode.

successful long-term research recording protocols and eventual clinical applications. Most researchers believe that a layer obstructive to current flow forms near the sensor surface as seen in Fig. 5 (Biran et al., 2005). Reaction to implants is intensified when the implant is tethered to a fixed reference such is the skull (Kim et al., 2004). There have been a number of countermeasures against this obstruction proposed including rejuvenation by passing a small electrical current through recording sites. Weiland and Anderson (2000) have shown the effect with chronic stimulation using impedance spectroscopy while extensive recording data with models have also been reported (Otto et al., 2006). When chronic systems allow the movement of probes, recording quality can be maintained by periodic small displacements of the probes (Bragin et al., 2000). Insights into what is happening to reduce the signal and how the signal can be restored can be obtained from the classic electrical/tissue interface model shown in Fig. 4. Any obstruction or diversion of cell membrane current on the left from the main pathway to the sensor amplifier input on the right makes the system less sensitive. Moreover, the equivalent impedance of dissipative elements in the pathway contributes to signal corrupting noise and reduced bandwidth. There is strong evidence that countermeasures taken at the time of implant can be effective. Skillful insertion as indicated above for acute electrodes is very important for chronic implants as well. Almost every user reports improved performance of implants with experience and perfection of technique. We and others have experience using the steroid dexamethasone (DEX) to interrupt the inflammation process. Fig. 6 shows the results of a month long comparison of guinea pigs systemically treated with dexamethasone for just 1 week beginning just before the implant procedure. The location of the implants was the IC. Spataro et al. (2005) has shown the histological substrate for this result. More local methods of DEX administration such as release from the probe substrate may produce the same effect as systemic administration without the side effects (Kim and Martin, 2006). Knowledge of the interaction between matrix proteins and sensors has revealed a general ‘sensor fouling’ problem which is being approached in a number of ways: (1) understanding of the tissue reaction sequence (Szarowski et al., 2003; Kim et al., 2004) which leads to changes in the protein population in the tissue and a subsequent layering of species around the electrode, (2) surgical procedure and taking care with insertion and stabilization to prevent continuous irritation of the tissue (Edell et al., 1992; Vetter et al., 2004), (3) material science solutions to reducing tissue reaction, (4) electrode surface or systemic drug therapy that reduces tissue reaction (Shain et al., 2003), (5) surface technology that minimizes adsorption of proteins on the surfaces of sensors and (6) delivery of antifouling agents by the sensor itself.

36

D.J. Anderson / Hearing Research 242 (2008) 31–41

Fig. 6. A one week systemic Dexamethasone administration improves the long-term viability of multichannel electrode arrays. Here non-treated animals show a steady reduction in the number of effective remaining channels while the reduction in effective channels in treated animals is arrested even for times beyond the treatment period. An effective channel is one that exceeds a threshold SNR as defined by Snellings et al. (2006).

The number of factors that enter into the success of a chronic implant system take the design of such systems far beyond the probe technology and much work remains before the problems are fully understood and neutralized. The current status of chronic implants is at a stage where systems are well within the range of usefulness by the neuroscience research community but are not satisfactory for long-term experiments or any life-time neuroprosthesis requiring recording. Stimulation devices are at a stage where very long implants can be contemplated. 2.6. Stimulation characteristics The use of iridium as the site metal for the Michigan probe was introduced based on the work in Brummer’s laboratory at EIC (Robblee et al., 1983). Platinum or platinum/iridium are the most used metals for electrical stimulation but iridium has a much higher safe charge delivery when activated either by triangular or pulsatile voltage cycling. The activation process converts some surface iridium to complex oxide states thus setting up the basis for a very favorable redox reaction. When stimulation current is applied the oxides transform reversibly to deliver charge to excitable tissue. The electrical result is what appears to be a large capacitor that can deliver about 3 mC/cm2 in a 200 ls pulse while under the same conditions platinum can deliver about 50–100 lC/cm2. In the recording mode the voltages are much lower than required to cause transformation in the oxide states but there still are advantages to activation. The activation process produces a recording electrode that has reduced impedance compared to the polished native state of iridium that is present immediately after deposition. The reduced impedance is attributable to a modified surface texture with higher real surface area. If a probe is to be used for both recording and stimulation, the desired charge transfer can be achieved with a smaller site size thus improving the recording characteristics relative to that of a less efficient electrode. It was decided to fabricate most electrodes in this fashion for uniformity of fabrication and general utility in both recording and stimulation modes. Anderson and Weiland have reported the on the charge transfer properties of iridium oxide on Michigan neural probes (Anderson et al., 1989; Weiland and Anderson, 2000).

2.7. Fouling and stimulation Weiland and Anderson (2000) performed impedance spectroscopy and modeling on stimulation electrodes used for daily stimulation trials. They found lowered serial resistive components in the high frequency model immediately after stimulation but after an over night period of no stimulation, the serial component of the impedance increased to the previous day’s value. Over the same intervals, the low frequency model which involves the diffusion layer of the electrode and has frequency dependency (described by the Warburg model) was not substantially modified by the daily cycling between periods of stimulation and rest. It has since been shown the same types of changes with current passage have a direct relationship to recording signal quality (Johnson et al., 2005). 2.8. Chemical delivery probes Chemical delivery in conjunction with recording and stimulation probes has been achieved using a number of methods including micro-fluidic channels fabricated in both silicon and polymer devices (Chen et al., 1997; Rathnasingham et al., 2004), whole probe coatings (Mensinger et al., 2000), wells formed through the substrate and loaded with gels (Kipke et al., 2003), and electrolytically deposited carriers on the sites (Cui et al., 2001). The promise for this technology is considerable including counter measures against tissue reaction, neural growth factor introduction for associating neurons more closely to the probe, marking probe location for histology, regulating neural activity in circuits, and analysis or therapeutic applications for human disorders. 3. Signal processing methods for array electrodes Auditory physiologists have utilized the silicon probe technology in numerous ways. The ability of the probes to spatially sample a linear distance in the brain has been, perhaps, the most productive use (Bierer and Middlebrooks, 2002; Snyder et al., 2004). This method follows directly from a long history of sequential mapping used to explore auditory representation in numerous species. The linear array has multiplied the productivity of these methods by

D.J. Anderson / Hearing Research 242 (2008) 31–41

at least the number of sites on the probe and perhaps much more because of the need to assure that data are stationary over the interval of a long experiment. While the parallel data acquisition method compresses the time required for this important measurement by multiplexing the data acquisition over many sites, it also multiplies the experimental yield by the number of sites allowing an advantageous compromise between shorter experiments and more data. Signal processing methods that utilize spatial filtering are dependent on simultaneous recording and vastly improve data quality by noise reduction through averaging and cancellation of events shared by neighboring electrodes. The ultimate recording electrode would be one that has very low noise sensors while the number of sensors outnumbers the number of signals present in the media. Spatial signal processing methods that derive new information available with simultaneous measurement, such as current source/sink density calculations, add new value to array technology and improve their effectiveness as the channel count increases. Closely-spaced stimulation arrays also offer improvement over isolated single stimulation electrodes. Selectivity of excitation by tissue volume can be improved over monopolar stimulation by ‘current steering’ methods such as bi-polar, tri-polar and more generalized multi-polar stimulation. These methods have a power cost which is very important for prosthesis devices but offer further opportunities to optimize performance. 3.1. Local and wide area noise and its reduction by array signal processing The ability to distinguish discrete neural events recorded by an array is dependent on the size and distribution of the signals and the quantity and the distribution of the noise being recorded with the signal. Unwanted noise signals in neural tissue are of three major types: 1. Local thermal noise generated by the electrode and dissipative impedances near the electrode is seen by the recording amplifier as a combination of white and 1/f shaped spectrum. The noise on each electrode is independent of all the other electrode sites and therefore not correlated across the array.

37

2. Signals from nearby neurons that may not be distinguishable by single channel spike sorting methods. Many of these signals may be represented on nearby electrodes and therefore may be locally correlated on patches of the array. This noise can be spatially filtered from the data on one electrode by subtraction or cancellation with selected members of the array. 3. Broad area potentials that are due to a large number of single neural discharges or by currents generated by dendritic fields. These evoked potentials may be highly correlated across the array and can be compensated by a reference signal extracted from the array or obtained from an additional reference electrode. Fig. 7 describes the recording process graphically. The noise sources described above become more and more manageable with increased noise correlation among channels while, as will be seen in the next section, spike sorting is improved by redundant recording in an array. To give the concepts of signal processing needed to retrieve neural signals from an array of electrodes, we create the matrix notation Y = AX + N where if X is an M row neural signal, A is a L  M matrix which maps the M neural signals on an L channel electrode array and N is the L row noise signal seen by each channel. Since some noise signals may be considered relevant, components of the noise may be placed in the X vector. In any case, Y is the signal sampled by the data acquisition system and is the starting point for analysis of a multichannel record. Since the vector X contains the relevant information, we seek a solution of the form e ¼ BY where B is a filter which transforms the vector Y to an estiX mate of X taking into account the mapping matrix A and the nature of N. Since A and the nature of N are not initially known, this problem is termed the ‘blind source problem’ in the signal processing literature. Fortunately, because of the compactness of the neural signal, there are relatively simple ways to discover A (Oweiss and Anderson, 2007) and the covariance matrix of N can also be calculated over regions of the signal where no spike activity occurs. A pseudo-inverse of A(B) can then be calculated which produces neural data channels from the recorded data set. One particular inverse is the Best Linear Unbiased Estimator (BLUE) which partially removes cross channel interference among neural spikes and maximizes the signal-to-noise ratio on all channels (Stark and Woods 1994). The formula for this special ‘BLUE’ inverse is

Fig. 7. Large numbers of cells contribute to the signals recorded from a multichannel electrode array. Some cells (in red) are close to the recording sites and have high signalto-noise ratios (SNR). Others of this group are farther from individual sites but are recorded on more than one site. Combining the signals can enhance the SNR. Many cells (in green) are not distinguishable as individual signals but are represented by slow phase signals or correlated noise.

38

D.J. Anderson / Hearing Research 242 (2008) 31–41

B = (A0 K1A)1A0 K1, where K is the noise covariance matrix. This method works well in the rare case where the number of neural channels is less than the number of electrode channels but breaks down for either badly distributed spikes across channels or an excessive number of neurons contributing to the signal. The second is Independent Component Analysis (ICA) (Brown et al., 2001). ICA discovers independent components embedded in data flowing from sensor arrays and provides an inverse matrix for extracting them. The approach ICA uses to discover independent channels is minimizing mutual information across the derived array. We have found that ICA will improve SNR, spatially sort spikes when possible and often discover meaningful slow wave signals (Snellings et al., 2006). 3.2. Spike sorting by close spaced array methods Since single channel functional decomposition methods such as principle components and singular value decomposition are largely limited to single channels and have been well established for more than 40 years, we will limit signal processing method to inherently multichannel methods. The methods in the previous section are designed to recover the underlying neural signals that overlap electrode sites. In situations where the number of sensors is greater than the number of signals, sorting is complete and event times can be recovered with simple detection methods. In addition, overlaps in spike timing can be resolved thus creating a more valid data record particularly for highly correlated spike times. In most cases, however, neural signals outnumber the number of sites and supplemental procedures are required that can resolve overlaps. Such sorting methods depend on the distribution of signal strength for each recorded neuron. These data are the same as those contained in the A matrix in the section above where the distribution for A is stored in the column reserved for it. Since sorting methods for tetrode configurations have been well established (McNaughton et al., 1983; Gray et al., 1995), they are very appropriate for use on partitions of high count silicon substrate electrodes. Graphical methods can be used to separate neurons from a small number of signal channels but when the array is large and the A matrix can not be partitioned easily, more sophisticated methods of spatial spike separation are required (Oweiss and Anderson, 2007). 3.3. Frequency-intensity rate functions The most consistent hall mark of the auditory pathway is tonotopic organization. In a progression of investigations since Helmholtz hypothesized a frequency analyzer in the fully evolved inner ear and suggested mechanisms for it, tonotopic organization in the auditory system has been found almost universally. Throughout the intense descriptive period of auditory neurophysiology and anatomy in the second half of the 20th century starting with von Bekesy’s experimental validation of Helmholtz’s theory, every nucleus and pathway of the auditory system has been shown to be organized tonotopically. Early mapping with single site microelectrodes following a trajectory through the tissue depended on time sequential measurements as the electrode was lowered by several microns at a time. Depending on the desired resolution, a set of measurements could take hours. If the dimensionality of the experiment for each point is two or more, then the time per point expands. Statistically stationary results are difficult to obtain under these conditions and the number of trajectories that could be used is limited by the time constraints. Recently, auditory electrophysiologists from a number of institutions have been taking advantage of array technology to explore single and multiple linear trajectories through neural structures. In addition when an array can be used for stimulation as well as

recording and a second recording array is placed in a higher center; specificity from neural site to neural site can be measured. An example of this method is taken from our own work on the inferior colliculus and its cortical connections. A brief description of the method is the following: 1. Arrays are placed along the tonotopic dimensions of the IC and the primary auditory cortex of the guinea pig. 2. The arrays are frequency calibrated with tone pulses of several frequencies and intensities to assure the two arrays cover the same acoustic range. 3. The role of the IC electrode is then reversed and it is used as a stimulation electrode to measure the parameters of the IC to cortex connections. 4. Within the same animal and over several animals, different IC electrode trajectories can be tested against the same and different cortical trajectories. One could envision this experimental procedure being made even more efficient with the additions of still more recording and stimulation sites in the IC and cortex. As more sites are added the complexity of the physiological experiment remains modest but the need for more instrumentation and computing resources and the drive toward size reduction and integration increases. Devices with more than one parallel array make it possible to experiment on multiple trajectories simultaneously. Use of a high resolution mapping electrode with 32 sites not only allows simultaneous measurement of several identical experiments but allows spatial correlation of probabilistic data such as neural spikes across the many sites. An example of a two shank device that can compare two trajectories over most of the IC frequency range of the guinea pig is shown by Lim and Anderson (2006). 3.4. Current source density analysis There are several examples in the literature of the use of electrode arrays to estimate the location of current sources and sinks in neural tissue and attempt to make estimates of electrical properties of the brain at high resolution. With modest activation the recording properties can be extended to the lower frequencies so local field potentials (LFPs) can be recorded. The method of converting LFPs to determine current source density (CSD) had been used in several studies to derive information about charge emerging from a source such as ionic currents crossing a cell membrane, flowing through the extracellular media and sinking into another structure within the system. The objective of the method is to derive the location and intensity of the sources and sinks by analysis of the extracellular potentials. Nicholson and Freeman, in two 1975 papers, clarified the method by stating the physical assumptions and principles behind it and applying it to additional problems (Buzsáki, 2004; Nicholson and Freeman, 1975; Prechtl et al., 2000). Without repeating the detail of the classical derivation and assuming n0o magnetic fields and a homogenous isotropic media (conductivity r is a constant scalar), the current source density in space determines the divergence of the current vector ðCSDx;y;z ¼ r  JÞ, the current vector is proportional to the electrical field (J = rE) and the electric field is the negative gradient of the electrical potential ðE ¼ rUÞ resulting in Poisson’s equation ðCSDx;y;z ¼ rr2 Ux;y;z Þ. Limiting the calculation to a single dimension ðCSDx ¼ r@ 2 Ux =@x2 Þ and making x a discrete spatial variable spaced and indexed by the potential sensors, the final approximation is

CSDðnDxÞ ¼ 

Uððn þ 1ÞDxÞ  2UðnDxÞ þ Uððn  1ÞDxÞ ðDxÞ2

D.J. Anderson / Hearing Research 242 (2008) 31–41

Assuming that each spatial sample is also a time series, each CSD measure in space is also a time series. One- and two-dimensional arrays have also been used to study layered and very systematic neural structures such as the cerebellum and the hippocampus over broad regions (Kandel and Buzsaski, 1997). With close spaced sensors detailed studies at the cellular level can be accomplished (Blanche et al., 2005). One-dimensional arrays of modest density can be used to access dendritic activity and quickly locate cortical layers (Muller-Preuss and Mitzdorf, 1984). 3.5. Stimulating the tonotopic gradient The central auditory system’s fundamental principle of organization is its dedication to orderly spatial distribution of the audible spectrum. This makes it possible to design recording and stimulation site distributions on a single substrate that ‘‘fit” the spatially distributed spectrum for maximum efficiency of nucleus coverage in any specified species. Investigations conducted by Lim and Anderson (2006, 2007) to stimulate the inferior colliculus at several points along its tonotopic gradient used parallel tines of a silicon electrode each with 16 recording/stimulation sites to first map the natural center frequency of each site as determined by acoustic means then used the same electrode to stimulate the tissue surrounding the sites. A second electrode placed along a tonotopic axis of the auditory cortex recorded the resulting response patterns. Additional details of these experiments are included in this volume (Lim et al., 2008). Stimulation using Michigan type probes has also been accomplished in the cochlear nucleus (Evans et al., 1989; McCreery et al., 2007; McCreery, 2008), scala tympani (Wise et al., 2008) and the auditory nerve (Middlebrooks and Snyder, 2007; Badi et al., 2003). Other technologies have been used to stimulate all of these structures.

39

developments will allow extensions of deep structure research to larger mammalian chronic preparations and full exploitation of cortical areas in area and depth. While scala tympani stimulation has been extremely successful, there are both clinical and technical reasons to move more centrally. The initial clinical reasons center on destruction of or the loss of communication between the end organ and the CNS. The technical reasons are the better use of high channel count devices because of close proximity to the excitable tissue. This can lead to better frequency discrimination, larger dynamic range and lower power. It is not known at this time if the nervous system will be able to adapt to more central stimulation but if there are positive results in clinical trials and the risk/benefit ration is controlled, CNS implants may become more prominent or even dominate. Other branches of neural science have made use of methods that could offer opportunities to auditory science. Among these are signal-to-noise optimization by array processing, CSD calculations, fluid delivery systems, and network analysis using stimulation in conjunction with recording and sophisticated statistical methods for discovering associations among neurons. Acknowledgements Although this paper was not supported by any specific award, the author wishes to thank the NIH for the many years of development support for the Michigan probe through the NINDS’s Neuroprosthesis Program and ten years of P41 support for the University of Michigan Center for Neural Communications Technology from NCRR and NIBIB. It should be noted that the author is a principle of NeuroNexus Technologies of Ann Arbor and stands to benefit financially from investigator use of the electrodes described herein. References

3.6. Stimulation current focusing using arrays The focusing of stimulation current to increase selectivity has been of interest for some time and was discussed in detail by Jolly et al. (1996) for cochlear implants. A similar method has been analyzed for cuff electrodes and multipolar stimulation has been suggested for the CNS (Gingerich et al., 2001). Stimulation selectivity of excitable tissue is generally greater for single penetrating electrodes placed within a few microns of the tissue but there are gains to be obtained from multipolar stimulation. The gains are clear for electrodes placed at a distance from the excitable tissue such as the scala tympani cochlear implant. The use of multipolar current patterns can provide still more focus and/or steering. Tripolar stimulation in scala tympani greatly restricts the cortical response image (Bierer and Middlebrooks, 2002; Middlebrooks and Bierer, 2002) and the inferior colliculus response image (Snyder et al., 2004) compared to monopolar stimulation. 4. Discussion and conclusion Large site count recording and stimulation arrays batch processed for repeatability and precision offer many opportunities for auditory scientists and perhaps future prostheses. The design space of thin-film electrodes includes almost any two-dimensional shape and the number of recording sites is steadily on the increase. The most prominent applications are mapping of nuclear and cortical areas and multiunit recording using such structures as the tetrode. These applications allow simultaneous recording from several points on the tonotopic axis giving easy assessment of the function of auditory prostheses in animals. Two important extensions of the technology design space are access to deep structures beyond 10 mm and truly three-dimensional arrays. These

Abeles, M., Goldstein, M.H., 1977. Multispike train analysis. Proc. Inst. Electr. Electron. 65, 762–773. Anderson, D.J., Najafi, K., Tanghe, S.J., Evans, D.A., Levy, K.L., Hetke, J.F., Xue, X., Zappia, J.J., Wise, K.D., 1989. Batch-fabricated thin-film electrodes for stimulation of the central auditory system. IEEE Trans. Biomed. Eng. 36, 693– 704. Anderson, D.J., Oweiss, K.G., Bierer, S.M., 2001. Sensor arrays in the microenvironment of the brain. In: Acoustics, Speech, and Signal Processing, (ICASSP’01), vol. 6, pp. 3433–3436. Badi, A.N., Kertesz, T.R., Gurgel, R.K., Shelton, C., Normann, R.A., 2003. Development of a novel eighth-nerve intraneural auditory neuroprosthesis. Laryngoscope 113, 833–842. Bai, Q., Wise, K.D., Anderson, D.J., 2000. A high-yield microassembly structure for three-dimensional microelectrode arrays. IEEE Trans. Biomed. Eng. 47, 281– 289. BeMent, S.L., Wise, K.D., Anderson, D.J., Najafi, K., Drake, K.L., 1986. Solid-state electrodes for multichannel multiplexed intracortical neuronal recording. IEEE Trans. Biomed. Eng. 33, 693–704. Bierer, J.A., Middlebrooks, J.C., 2002. Auditory cortical images of cochlear-implant stimuli: dependence on electrode configuration. J. Neurophysiol. 87, 478– 492. Biran, R., Martin, D.C., Tresco, P.A., 2005. Neuronal cell loss accompanies the brain tissue response to chronically implanted silicon microelectrode arrays. Exp. Neurol. 195, 115–126. Blanche, T.J., Spacek, M.A., Hetke, J.F., Swindale, N.V., 2005. Polytrodes: high-density silicon electrode arrays for large-scale multiunit recording. J. Neurophysiol. 93, 2987–3000. Bragin, A., Hetke, J., Wilson, C.L., Anderson, D.J., Engel Jr., J., Buzsaki, G., 2000. Multiple site silicon-based probes for chronic recordings in freely moving rats: implantation, recording and histological verification. J. Neurosci. Methods 98, 77–82. Brown, G.D., Yamada, S., Sejnowski, T.J., 2001. Independent component analysis at the neural cocktail party. Trends Neurosci. 24, 54–63. Buzsáki, G., 2004. Large-scale recording of neuronal ensembles. Nat. Neurosci. 7, 446–451. Chen, J.K., Wise, K.D., Hetke, J.F., Bledsoe Jr., S.C., 1997. A multichannel neural probe for selective chemical delivery at the cellular level. IEEE Trans. Biomed. Eng. 44, 760–769. Cui, X., Hetke, J.F., Wiler, J.A., Anderson, D.J., Martin, D.C., 2001. Electrochemical deposition and characterization of conducting polymer polypyrrole/PSS on multichannel neural probes. Sensor Actuat. A – Phys. 93, 8–18.

40

D.J. Anderson / Hearing Research 242 (2008) 31–41

Drake, K.L., Wise, K.D., Farraye, J., Anderson, D.J., BeMent, S.L., 1988. Performance of planar multisite microprobes in recording extracellular single-unit intracortical activity. IEEE Trans. Biomed. Eng. 35, 719–732. Edell, D.J., 1986. A peripheral nerve information transducer for amputees: longterm multichannel recordings from rabbit peripheral nerves. IEEE Trans. Biomed. Eng. 33, 203–214. Edell, D.J., Toi, V.V., McNeil, V.M., Clark, L.D., 1992. Factors influencing the biocompatibility of insertable silicon microshafts in cerebral cortex. IEEE Trans. Biomed. Eng. 39, 635–643. Edwards, C.E., Finger, P.A., Anderson, D.J., Wiler, J.A., Hetke, J.F., Altschuler, R.A., 1997. A technique for in vivo morphological evaluation of chronically implanted neuronal silicon substrate electrodes for confocal microscopy. Abstracts for Microscopy and Microanalysis, Cleveland, OH, p. 351. Evans, D.E., Niparko, J.K., Miller, J.M., Jyung, R.W., Anderson, D.J., 1989. Multiplechannel stimulation of the cochlear nucleus. Otolaryng. Head Neck 101, 651– 657. Gingerich, M.D., Hetke, J.F., Anderson, D.J., Wise, K.D., 2001. A 256-site CMOS microelectrode array for multipoint stimulation and recording in the central nervous system. In: Digest, International Conference on Solid-State Sensors and Actuators, pp. 416–419. Gray, C.M., Maldonado, P.E., Wilson, M., McNaughton, B., 1995. Tetrodes markedly improve the reliability and yield of multiple single-unit isolation from multiunit recordings in cat striate cortex. J. Neurosci. Methods 63, 43–54. Hassibi, A., Navid, R., Dutton, R.W., Lee, T.H., 2004. Comprehensive study of noise processes in electrode electrolyte interfaces. J. Appl. Phys. 96, 1074– 1082. Hetke, J.F., Lund, J.L., Najafi, K., Wise, K.D., Anderson, D.J., 1994. Silicon ribbon cables for chronically implantable microelectrode arrays. IEEE Trans. Biomed. Eng. 41, 314–321. Ji, J., Wise, K.D., 1992. An implantable CMOS circuit interface for multiplexed microelectrode recording arrays. IEEE J. Solid-State Circ. 27, 433–443. Johnson, M.D., Otto, K.J., Kipke, D.R., 2005. Repeated voltage biasing improves unit recordings by reducing resistive tissue impedances. IEEE Trans. Neuro. Syst. Rehabil. 13, 160–165. Jolly, C.N., Spelman, F.A., Clopton, B.M., 1996. Quadrupolar stimulation for cochlear prostheses: modeling and experimental data. IEEE Trans. Biomed. Eng. 43, 857– 865. Kandel, A., Buzsaski, G., 1997. Cellular-synaptic generation of sleep spindles, spikeand-wave discharges, and evoked thalamocortical responses in the neocortex of the rat. J. Neurosci. 17, 6783–6797. Kim, D.-H., Martin, D.C., 2006. Sustained release of dexamethasone from hydrophilic matrices using PLGA nanoparticles for neural drug delivery. Biomaterials 27, 3031–3037. Kim, Y.T., Hitchcock, R.W., Bridge, M.J., Tresco, P.A., 2004. Chronic response of adult rat brain tissue to implants anchored to the skull. Biomaterials 25, 2229– 2237. Kipke, D.R., Vetter, R.J., Williams, J.C., Hetke, J.F., 2003. Silicon-substrate intracortical microelectrode arrays for long-term recording of neuronal spike activity in cerebral cortex. IEEE Trans. Neuro. Syst. Rehabil. 11, 151–155. Kovacs, G.T.A., 1998. Micromachined Transducers Sourcebook, first ed. McGraw Hill, New York. p. 688 (Chapter 8). Lempka, S.F., Johnson, M.D., Barnett, D.W., Moffitt, M.A., Otto, K.J., Kipke, D.R., McIntryre, C.C., 2006. Optimization of microelectrode design for cortical recording based on thermal noise considerations. In: Proceedings of the Institute of Electrical and Electronics 28th EMBS Conference FrC14.3. Lim, H.H., Anderson, D.J., 2006. Auditory cortical responses to electrical stimulation of the inferior colliculus: implications for an auditory midbrain implant. J. Neurophysiol. 96, 975–988. Lim, H.H., Anderson, D.J., 2007. Antidromic activation reveals tonotopically organized projections from primary auditory cortex to the central nucleus of the inferior colliculus in guinea pig. J. Neurophysiol. 97, 1413–1427. Lim, H.H., Lenarz, T., Anderson, D.J., Lenarz, M., 2008. The auditory midbrain implant: Effects of electrode location. Hear. Res. 242, 74–85. Martinez-Santiesteban, F.M., Swanson, S.D., Noll, D.C., Anderson, D.J., 2006. Magnetic resonance compatibility of multichannel silicon microelectrode systems for neural recording and stimulation: design criteria, test, and recommendations. IEEE Trans. Biomed. Eng. 53, 547–558. Martinez-Santiesteban, F.M., Swanson, S.D., Noll, D.C., Anderson, D.J., 2007. Magnetic field perturbation of neural recording and stimulating microelectrodes. Phys. Med. Biol. 52, 2073–2088. May, G.A., Shamma, S.A., White, R.L., 1979. A tantalum-on-sapphire microelectrode array. IEEE Trans. Electron. Dev. 26, 1932–1939. McCreery, D., Lossinsky, A., Pikov, V., 2007. Performance of multisite silicon microprobes implanted chronically in the ventral cochlear nucleus of the cat. IEEE Trans. Biomed. Eng. 54, 1042–1052. McCreery, 2008. Cochlear nucleus auditory prostheses. Hear. Res. 242, 64–73. McNaughton, B.L., O’Keefe, J.O., Barnes, C.A., 1983. The sterotrode: a new technique for simultaneous isolation of several single units in the central nervous system from multiple unit records. J. Neurosci. Methods 8, 391– 397. Mensinger, A.F., Anderson, D.J., Buchko, C.J., Johnson, M.A., Martin, D.C., Tresco, P.A., Silver, R.B., Highstein, S.M., 2000. Chronic recording of regenerating VIIIth nerve axons with a sieve electrode. J. Neurophysiol. 83, 611–615. Merzenich, M.M., Reid, M.D., 1974. Representation of the cochlea within the inferior colliculus of the cat. Brain Res. 77, 397–415.

Middlebrooks, J.C., Bierer, J.A., 2002. Auditory cortical images of cochlear-implant stimuli: coding of stimulus channel and current level. J. Neurophysiol. 87, 493– 507. Middlebrooks, J.C., Snyder, R.L., 2007. Auditory prosthesis with a penetrating nerve array. JARO 8, 258–279. Moxon, K.A., Leiser, S.C., Gerhardt, G.A., Barbee, K.A., Chapin, J.K., 2004. Ceramicbased multisite electrode arrays for chronic single-neuron recording. IEEE Trans. Biomed. Eng. 51, 647–656. Muller-Preuss, P., Mitzdorf, U., 1984. Functional anatomy of the inferior colliculus and the auditory cortex: current source density analyses of click-evoked potentials. Hear. Res. 16, 133–142. Najafi, K., Wise, K.D., Mochizuki, T., 1985. A high-yield IC-compatible multichannel recording array. IEEE Trans. Electron. Dev. 32, 1206–1211. Nicholson, C., Freeman, J.A., 1975. Theory of current source-density analysis and determination of conductivity tensor for anuran cerebellum. J. Neurophysiol. 38, 356–368. Nicolelis, M.A., Dimitrov, D., Camena, J.M., Crist, R., Lehew, G., Kralik, J.D., 2003. Chronic, multisite, multielectrode recordings in macaque monkeys. PNAS 100, 11041–11046. Otto, K.J., Johnson, M.D., Kipke, D.R., 2006. Voltage pulses change neural interface properties and improve unit recordings with chronically implanted microelectrodes. IEEE Trans. Biomed. Eng. 53, 333–340. Oweiss, K., Anderson, D.J., 2007. Tracking signal subspace invariance for blind separation and classification of nonorthogonal sources in correlated noise. EURASIP J. Appl. Sig. P, 20p, Article ID 37485. Pang, C., Cham, J.G., Nenadic, Z., Musallam, S., Tai, Y.-C., Burdick, J.W., Anderson, R.A., 2005. A new multi-site probe array with monolithically integrated parylene flexible cable for neural prostheses. In: Proceedings of the Institute of Electrical and Electronics 27th EMBS Conference, pp. 7114–7117. Perkel, D.H., Gerstein, G.L., Moore, G.P., 1967a. Neuronal spike trains and stochastic point processes: II. Simultaneous spike trains. Biophys. J. 7, 419–440. Perkel, D.H., Gerstein, G.L., Moore, G.P., 1967b. Neuronal spike trains and stochastic point processes: I. The single spike train. Biophys. J. 7, 391–418. Pickard, R.S., Joseph, P.L., Collins, A.J., Hicks, R.C.J., 1979. Flexible printed-circuit probe for electrophysiology. Med. Biol. Eng. Comput. 17, 261–267. Prechtl, J.C., Bullock, T.H., Kleinfeld, D., 2000. Direct evidence for local oscillatory current sources and intracortical phase gradients in turtle visual cortex. PNAS 97, 877–882. Prohaska, O.J., Olcaytug, F.P., Pfundner, P., Dragaun, H., 1986. Thin-film multiple electrode probes: possibilities and limitations. IEEE Trans. Biomed. Eng. 33, 223–229. Rathnasingham, R., Kipke, D.R., Bledsoe Jr., S.C., McLaren, J.D., 2004. Characterization of implantable microfabricated fluid delivery devices. IEEE Trans. Biomed. Eng. 51, 138–145. Robblee, L.S., Lefko, J.L., Brummer, S.B., 1983. Activated iridium: an electrode suitable for reversible charge injection in saline solution. J. Electrochem. Soc. 130, 731–733. Robinson, D.A., 1968. The electrical properties of metal microelectrodes. Proc. IEEE 56, 1065–1071. Rousche, P.J., Normann, R.A., 1998. Chronic recording capability of the Utah intracortical electrode array in cat sensory cortex. J. Neurosci. Methods 82, 1–15. Rousche, P.J., Pellinen, D.S., Pivin, D.P., Williams, J.C., Vetter, R.J., Kipke, D.R., 2001. Flexible polyimide-based intracortical electrode arrays with bioactive capability. IEEE Trans. Biomed. Eng. 48, 361–371. Shain, W., Spataro, L., Dilgen, J., Haverstick, K., Retterer, S., Isaacson, M., Saltzman, M., Turner, J.N., 2003. Controlling cellular reactive responses around neural prosthetic devices using peripheral and local intervention strategies. IEEE Trans. Neuro. Syst. Rehabil. 11, 186–188. Snellings, A., Anderson, D.J., Aldridge, J.W., 2006. Improved signal and reduced noise in neural recordings from close-spaced electrode arrays using independent component analysis as a preprocessor. J. Neurosci. Methods 150, 254–264. Snyder, R.L., Bierer, J.A., Middlebrooks, J.C., 2004. Topographic spread of inferior colliculus activation in response to acoustic and intracochlear electric stimulation. JARO 5, 305–322. Spataro, L., Dilgen, J., Retterer, S., Spence, A.J., Isaacson, M., Turner, J.N., Shain, W., 2005. Dexamethasone treatment reduces astroglia responses to inserted neuroprosthetic devices in rat neocortex. Exp. Neurol. 194, 289–300. Stark, H., Woods, J.W., 1994. Probability, Random Processes, and Estimation Theory for Engineers, second ed. Prentice Hall, Upper Saddle River, NJ. Starr, A., Wise, K.D., Csongradi, J., 1973. An evaluation of photoengraved microelectrodes for extracellular single-unit recording. IEEE Trans. Biomed. Eng. 20, 291–293. Szarowski, D.H., Andersen, M.D., Retterer, S., Spence, A.J., Isaacson, M., Craighead, H.G., Turner, J.N., Shain, W., 2003. Brain responses to micro-machined silicon devices. Brain Res. 983, 23–35. Tanghe, S.J., Wise, K.D., 1992. A 16-Channel CMOS neural stimulating array. J. SolidState Circ. 27, 1819–1825. Vetter, R.J., Williams, J.C., Hetke, J.F., Nunamaker, E.A., Kipke, D.R., 2004. Chronic neural recording using silicon-substrate microelectrode arrays implanted in cerebral cortex. IEEE Trans. Biomed. Eng. 51, 896–904. Weiland, J.D., Anderson, D.J., 2000. Chronic neural stimulation with thin-film, iridium oxide electrodes. IEEE Trans. Biomed. Eng. 47, 911–918. Wise, K.D., Angell, J.B., 1975. A low-capacitance multielectrode probe for use in extracellular neurophysiology. IEEE Trans. Biomed. Eng. 22, 212–219. Wise, K.D., Angell, J.B., Starr, A., 1970. An integrated-circuit approach to extracellular microelectrodes. IEEE Trans. Biomed. Eng. 17, 238–247.

D.J. Anderson / Hearing Research 242 (2008) 31–41 Wise, K.D., Anderson, D.J., Hetke, J.F., Kipke, D.R., Najafi, K., 2004. Wireless implantable microsystems: high-density electronic interfaces to the nervous system. Proc. IEEE 92, 76–97. Wise, K.D., Bhatti, P.T., Wang, J., Friedrich, C.R., 2008. High-density cochlear implants with position sensing and control. Hear. Res. 242, 22–30.

41

Yao, Y., Gulari, M.N., Casey, B., Wiler, J.A., Wise, K.D., 2007. Silicon microelectrodes with flexible integrated cables for neural implant applications. In: Proceedings of the Institute of Electrical and Electronics EMBS Conference on Neural Engineering, Article ID. FrD3.20.