Journal of Neuroscience Methods 131 (2003) 9–26
Fabrication and characteristics of an implantable,
polymer-based, intrafascicular electrode
Stephen M. Lawrence, Gurpreet S. Dhillon, Kenneth W. Horch∗
Department of Bioengineering, University of Utah, 50 S. Central Campus Dr., Salt Lake City, UT 84112, USA
Received 27 November 2002; received in revised form 8 July 2003; accepted 11 July 2003
Abstract
We describe new manufacturing techniques and physical properties of an improved polymer-based longitudinal intrafascicular electrode
(polyLIFE). Modifications were made to correct: (1) poor metal film adhesion and fatigue resistance, (2) inconsistent insulation adhesion and
control over recording/stimulation zone length, and (3) insufficient tensile strength for clinical use. Metal adhesion was significantly improved
by both plasma treatment and fiber rotation (about the long axis) during metal deposition. Fatigue resistance was improved by reduction in
sputtering energy (time × power) combined with long axis rotation, resulting in thin metal films that were 250 times more resistant to cyclic
bending fatigue. Insulation adhesion was enhanced with the application of an adhesion-promoting silicone (MED2-4013, Nusil), while the
recording/stimulation zone length was controlled to 1 ± 0.2 mm (mean ± S.D.). The polyLIFE was made more robust by the inclusion of three
individually metallized fibers, improving its tensile strength by a factor of 4 while producing minimal changes to its overall stiffness. However,
the metallized fiber redundancy did not significantly affect fatigue resistance. The manufacturing changes described in this study enable the
construction of more mechanically robust polyLIFEs, which should provide greater success when chronically implanted in peripheral nerves.
© 2003 Elsevier B.V. All rights reserved.
Keywords: polyLIFE; Intrafascicular electrode; Polymer
1. Introduction
The design and development of peripheral nerve interfaces has been a major goal of the researchers interested in
aiding patients with motor deficits such as foot drop, paralysis, and limb amputations. In order to provide more natural
and fine control over prosthetics that utilize these neural
interfaces, a concerted effort has been made to increase
electrode selectivity for both stimulation and recording. Selectivity for stimulation has been accomplished to a certain
degree by modifying cuff technology to add more active
sites, but recording selectivity has not been reported (Grill
and Mortimer, 1996; Rozman et al., 1993; Veraart et al.,
1993; Walter et al., 1997).
A much larger increase in selectivity has been obtained by
the use of interfaces whose recording/stimulation zones (active sites) have either been placed within individual fascicles
or rely on neural regeneration at a nerve stump. Several different designs incorporating this intraneural approach have
∗
Corresponding author. Tel.: +1-801-585-1981.
E-mail address: k.horch@m.cc.utah.edu (K.W. Horch).
0165-0270/$ – see front matter © 2003 Elsevier B.V. All rights reserved.
doi:10.1016/S0165-0270(03)00231-0
been developed including silicon-based (Akin et al., 1994;
Branner et al., 2001; Edell, 1986; Kovacs et al., 1992;
Rutten et al., 1991; Veltink et al., 1989) and polyimide-based
(González and Rodrı́guez, 1997; Stieglitz and Meyer, 1997)
microelectrode arrays as well as longitudinally implanted
microwires (Bowman and Erickson, 1985; Malagodi et al.,
1989). In acute experiments, these intraneural interfaces
have allowed activation of axons at low current densities
and recording from small subsets of neurons as opposed
to the relatively high current densities required with cuffs
and mass activity obtained with extraneural recordings.
Key design requirements for intrafascicular devices include
well defined, low impedance active sites, small dimensions
in order to minimize damage during insertion and prevent
long-term foreign body reaction, and the ability to penetrate the tough epineurial and perineurial connective tissue.
While silicon-based microelectrode arrays for peripheral
nerves take advantage of long-proven micromachining techniques to design devices with high densities of active sites,
on-chip circuitry, and manufacturing repeatability, they have
achieved limited success in long-term recordings.
The failure of silicon-based probes to provide long-term
peripheral nerve recordings is thought to be due in part
10
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
to their high stiffness compared to the surrounding neural tissue (mechanical mismatch), causing large foreign
body reactions and extensive encapsulation (Stensaas and
Stensaas, 1978). Long-term intrafascicular recordings (>4
months) from intact nerves have only been reported with the
use of longitudinally implanted intrafascicular electrodes
constructed from Teflon-coated, 25 m diameter Pt-Ir wire
(Lefurge et al., 1991). These longitudinal intrafascicular
electrodes (LIFEs) demonstrated selective recordings of
afferent activity from the radial nerve of cats for up to 6
months with good biocompatibility. However, we believe
that the mechanical stiffness of the Pt-Ir wire caused gradual drift of these electrodes within the implanted fascicle
over time, leading to recording instability and dense encapsulation tissue formation around the electrode. Therefore,
we feel that further minimization of mechanical mismatches
between the implanted electrodes and the surrounding
neural tissue will increase the biocompatibility and recording stability of intrafascicular technologies. Towards this
end, our lab has developed a polymer-based longitudinal
intrafascicular electrode (polyLIFE), constructed by metallizing single 12 m diameter Kevlar® fibers and insulating
them with silicone (McNaughton and Horch, 1996). These
polyLIFEs have shown similar acute recording capabilities
and tensile strength to Pt-Ir LIFEs, but have a 60 times
greater flexibility.
In addition to their acute recording capabilities, the
polyLIFEs have demonstrated long-term biocompatibility
in both peripheral nerve (Lawrence et al., 2002) and dorsal
rootlets (Malmstrom et al., 1998), but have yet to achieve
the reliability required for chronic recording. Previous work
has identified problems associated with the conducting and
insulating layers, including poor adhesion of the metal to
the Kevlar® fiber, limited mechanical fatigue resistance,
inconsistent adhesion of the silicone to the metal layer, and
inadequate control over active site length (unpublished observations). Clinical experimentation has also demonstrated
a need to increase the tensile properties of the original
polyLIFE design in order to better survive surgical handling
and percutaneous applications (preliminary unpublished
data from clinical trials).
The aim of the present study was to improve mechanical
and electrical properties of polyLIFEs by modification of
their fabrication processes. The effectiveness of these modifications was evaluated by several bench tests. The result
was a fabrication process that can produce more mechanically robust polyLIFEs with specifically tailored electrical
properties.
2. Materials and methods
Fig. 1. General design of a polymer-based intrafascicular electrode
(polyLIFE). The polyLIFE consists of a 12 m diameter Kevlar® fiber,
metallized with sputter-deposited Ti, Au, and Pt and insulated with silicone (metal and silicone layers not drawn to scale, but drawn to relative thickness). The recording/stimulation zone consists of approximately
1 mm non-insulated portion of the metallized fiber (diameter:length ratio
of ∼500:1).
lated with silicone (Fig. 1). polyLIFE manufacture involves
four steps: (1) fiber preparation and mounting, (2) surface
treatment, (3) metal deposition, and (4) insulation. In order to produce the most mechanically and electrically robust
polyLIFE possible, variations of each manufacturing step
were conducted and subsequently evaluated with bench-top
testing methods. Certain processes within each step were
identical to all constructed test samples and are listed below. Other procedures within each of the four manufacturing
steps were conducted on some samples but not others, and
are described in detail in the following sections.
Step 1 (fiber mounting): Kevlar-119® fibers were obtained
from Dupont in yarn form (∼1500 denier). Individual fibers
of 1 m length were teased from these bundles while soaking
in alcohol (to help prevent fiber-to-fiber abrasion). These
fibers were then wrapped under constant tension onto metal
looms and secured with cyanoacrylate.
Step 2 (surface treatment): Once the fibers were secured
to the loom, they were washed by sequential rinsing in four
baths for 3 min per bath: acetone, methanol, isopropanol,
and DI water.
Step 3 (metal deposition): Metal deposition was accomplished by dc sputter deposition (Denton, Discovery 18 sputter system) at vacuum of <0.67 × 10−3 Pa. Three metal
layers, Ti, Au, and Pt, were sequentially sputtered onto the
loom-bound fibers. This sputter system contains three sputter guns, enabling deposition of up to three different metals
onto a substrate without breaking vacuum.
Step 4 (insulation): The metallized fibers were then transferred to another metal loom, secured with cyanoacrylate
and insulated with a dip-coating process. In this process,
a drop of medical-grade silicone, at the end of a 2 mm diameter steel rod, was applied to the metal clad fiber as it
was drawn slowly along the fiber’s longitudinal axis. Reproducible control of dip-coating speed was provided by a
modified perfusion pump motor.
2.2. Manufacturing process modifications and LIFE
construction
2.1. General design
The basic structure of the polyLIFE consists of a Kevlar®
mono-filament core, clad with three metal layers and insu-
2.2.1. Fiber mounting
Due to the cylindrical geometry of the fibers, it is challenging to obtain uniform coating thickness, not only around
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
the fiber’s circumference, but also along its longitudinal
axis. Gas-phase scattering makes it possible to coat surfaces
that do not view the source; however, these “shadowed” surfaces receive metal at a much reduced sputtering flux, causing non-uniformities in thickness (Fancey and Matthews,
1991). Furthermore, sputtering flux is inversely proportional
to the distance from the sputtering source. An additional
geometrical concern involves the angle of incidence of
impinging atoms. Sputter deposition at normal angles of incidence and higher substrate temperatures tends to produce
more dense films with lower surface roughness (according to simulations; Dong et al., 1996), whereas sputter
deposition at oblique angles of incidence tends to cause
columnar development (a common microstructural feature
of thin-films deposited at low pressures and temperatures)
(Nieuwenhuizen and Haanstra, 1966). This columnar morphology can significantly affect thin films in terms of
mechanical integrity (crack propagation is frequently seen
along weak, low-density intercolumnar regions), electrical
properties, and surface roughness (Ohring, 2002).
For geometrically complex surfaces (such as miniature diameter fibers), control of various thin-film properties can be
11
exercised by both optimization of the source-metal to substrate distance, reduction of the shadowing effect, and minimization of oblique angles of incident atoms. This optimization might be accomplished by continuous rotation and short
sample lengths. However, in our sputter system, rotation was
possible only around one axis, and maximization of fiber
length was a critical goal. Thus, positioning of the mounted
fibers inside the sputter system could significantly affect the
mechanical performance of the deposited metal film.
In order to ascertain the optimal mounting strategy,
fibers were mounted on two different types of looms: (1)
a flat, rectangular-shaped, metal frame (referred to as the
“flat loom” shown in Fig. 2A) and (2) a cylindrical-shaped
metal frame (referred to as the “cylindrical loom” shown in
Fig. 2B).
2.2.2. Surface treatment
Plasma treatment of the surfaces has long been known
to enhance the adhesion of metal to polymers. At high
pressures (>9.3 Pa) and with plasmas created from reactive gases such as O2 , dry chemical etching (commonly
known as reactive-ion etching) of the substrate can enhance
Fig. 2. Loom orientation during sputtering. (A) The geometrical condition during sputtering of fibers mounted on a flat loom. In this case, rotation of
the loom is accomplished by rotation of the entire platform. (B) The geometrical condition during sputtering of fibers mounted on a cylindrical loom.
In this case, rotation of the loom is accomplished via a feed-through to a motor external to the sputter chamber. The loom axle is connected to the
feed-through by a flexible rotation coupler.
12
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
adhesion by chemically modifying, removing, or roughening the surface (Lamendola et al., 1999; Sugawara and
Stansfield, 1998). For instance, reactive-ion etching of
Kevlar-149® fibers with 100 W of O2 plasma demonstrated
significant increases in oxygen content of the surface within
15 s and visible surface roughening after 10 min (Sheu
and Shyu, 1994). At low pressures (<0.93 Pa) and with
chemically inert plasmas, physical sputtering (commonly
known as sputter etching) of the substrate can enhance adhesion by removing or dispersing contaminants, creating
dangling bonds or free radicals, and roughening the surface (Baglin, 1995; Ohring, 2002; Sugawara and Stansfield,
1998).
In this study, four different surface treatment processes
were evaluated: (1) no treatment, (2) reactive-ion etch (30 s
of 100 W O2 plasma at 27 Pa with an Oxford, Plasmalab
80 Plus), (3) sputter etch (6 min of 250 W rf power of an
N2 plasma at <1.4 Pa with a Denton, Discovery 18 sputter
system), and (4) a reactive-ion etch and a subsequent sputter
etch. Each surface treatment process will be referred to as
“S#”, where S stands for surface treatment and # stands for
the specific process (e.g. S2 stands for surface treatment
process #2, Table 1).
2.2.3. Metal deposition
The mechanical properties of vacuum deposited thin films
have been shown to be quite different from bulk properties.
For instance, depending on the deposited film microstructure or morphology and particular testing methods, different
researchers have found widely different Young’s modulus
for Au films (53–55 GPa, Espinosa and Prorok, 2001 or
30–78 GPa, Nix, 1989) as compared to bulk gold (78 GPa,
Nix, 1989). Other mechanical properties such as yield stress
have been found to be strongly related to film thickness
(170 MPa for 0.5 m thick Au films versus 50–55 MPa for
1.0 m thick Au films) (Espinosa and Prorok, 2001) and
crystallographic texture (Thompson, 1993). Furthermore,
the electrical properties can be related to the film density, where films with greater density can exhibit bulk-like
resistivity (Maissell, 1973).
The mechanical and electrical properties of thin metal
films can be tailored by controlling sputtering parameters
Table 1
Surface treatment processes
Process
Etch
S1
S2
S3
S4
None
SP
RIE
RIE
SP
Pressure (Pa)
–
<1.4
27
<1.4
27
Power (W)
Gas
Time (min)
–
250
100
250
100
–
N2
O2
N2
O2
–
6.0
0.5
0.5
6.0
Reactive-ion etching (RIE) was conducted in an Oxford Plasmalab 80
Plus. Sputter etching (SP) was conducted in a Denton Discovery 18 sputter
system immediately prior to metal deposition. In the case where both
sputter and reactive-ion etching were performed, the reactive-ion etching
was performed prior to sputter etching.
such as deposition rate and film thickness. Both deposition
rate and film thickness affect microstructural properties such
as density and grain size. For instance, higher deposition
rates tend to result in greater film density and larger grain
size (Ohring, 2002). Each metal has its own power/rate relationship that is defined by the number of sputtered metal
ions per bombarding argon ion (or sputtering yield) (Stuart,
1983). The substrate, platform, and source-metal target heat
up during sputtering in relation to power and time (Lamont,
1979). Density and grain size have been shown to increase
with greater film thickness and substrate temperature during
metal deposition (Ohring, 2002). Higher substrate temperatures, directly related to higher deposition rates (Maissell,
1973), can also directly affect adhesion, nucleation, and microstructure of growing thin films.
Metal deposition for this study involves several features:
(1) setup/positioning of the loom inside the sputter chamber, (2) application of continuous rotation, and (3) deposition parameter setup (source-metal, power, session time,
and number of repetitions). Due to geometrical constraints
inside the Denton sputter system, flat and cylindrical looms
were positioned inside the chamber with different methods.
Flat looms were positioned such that the centroid of the
loom was placed over the center of the substrate platform
(Fig. 2A). For the cylindrical loom, a special holder was fashioned that would allow rotation about the loom’s radial axis
without contacting the substrate platform. The cylindrical
loom was placed in the holder and centered on the substrate
platform. The cylindrical loom system was oriented such
that one end was positioned equidistant from the Ti and Au
sputter guns and the other end was below the Pt sputter gun
(Fig. 2B). The three sputter guns were positioned equidistant
from each other and oriented towards the center of the substrate platform at an angle 30◦ from normal to the platform
surface.
Rotation was supplied to the flat loom via rotation of the
entire substrate platform at a speed of 60◦ /s (Fig. 2A). For
the cylindrical loom, the substrate platform remained fixed;
only the loom was rotated. In this case, a vacuum-sealed,
translation feed-through was used to supply rotation at a
speed of 24◦ /s (Fig. 2B).
The deposition parameters were subdivided into six categories: (1) source-metal, (2) session power (where one
session is defined as period between creation and extinction of the plasma), (3) session time, (4) number of session repetitions, (5) session energy (session time × session
power), and (6) total sputter energy (sputter energy for entire
process).
For this study, each layer of metal was typically completed in one session; however, in order to minimize
substrate temperatures, some processes incorporated multiple sessions per metal layer, where a cooling period
of 3 min would follow each session. Process energy is
reported as the sum of all session energies per process
and is related to the amount of metal deposited. Table 2
summarizes the parameters involved for each of the
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
13
Table 2
Metal deposition processes
Process
Loom
Metal
Power (W)
Time (min)
Energy (kJ)
M1
Flat
Ti
Au
Pt
750
500
380
2
6
4
90
180
91.2
1
1
1
361
M2
Flat
Ti
Au
Pt
750
250
250
2
4
2
90
60
30
1
2a
2a
270
M3
Flat
Ti
Au
Pt
750
250
250
2.5
16
4
113
240
60
1
1
1
413
M4
Cylindrical
Ti
Au
Pt
50
50
50
4
3b
6
12
9
18
1
20a
1
210
M5
Cylindrical
Ti
Au
Pt
50
50
50
4
3b
6
12
9
18
1
18
1
192
M6
Cylindrical
Ti
Au
Pt
250
100
–
0.5
15
–
M7c
Cylindrical
Ti
Au
Pt
50
50
50
7.5
90
–
4
50d , 3b
3–6
12
150d , 9
9–18
Reps (#)
Process energy (kJ)
1
1
–
97.5
1
1d , 18–60a
1
129–309
A “session” is defined as the period between plasma ignition and extinction while sputtering from a single metal source. The “Reps” category indicates
how many sessions consisting of the same power/time parameters were completed. Session energy was calculated as powermetal × timemetal and process
energy was calculated as the sum of energy from all sessions in a single process.
a A break in vacuum was made to flip or reorient the loom.
b A 3 min stoppage or “cooling” period was made immediately following each session.
c M7 is a collection of several processes that were mainly used to help demonstrate the relationship between longitudinal resistance and process
energy for metallized fibers.
d A single process identical to that of M6 with the exception of session power and times.
metal deposition processes which are referred to as “M#”
where M stands for metal deposition and # stands for
the specific process (e.g. M2 stands for metal deposition
process #2).
Even though rotation of the flat loom was intended to
minimize the shadowing effect, a small portion of the fiber
circumference on the bottom side of the loom was continuously shadowed. Therefore, the flat loom was manually
flipped over midway through deposition of the Au and Pt
layers for process M2. Due to the geometry of the samples,
standard methods of measuring film thickness were not possible. Hence, we were not able to estimate deposition rate or
film coating uniformity in a quantitative way. Based on several longitudinal resistance measurements taken at various
levels along samples from cylinder looms, a non-uniformity
in coating thickness was assumed. In an attempt to minimize
this thickness non-uniformity, the cylindrical loom was reoriented end-for-end by breaking vacuum midway through
deposition of the Au layer in the process M4.
After deposition, several metallized fibers constructed
with the process M4 were annealed at 45, 100, 130, 150,
and 170 ◦ C for the purpose of determining the effect of
temperature on longitudinal and interfacial impedance of
the deposited metal film.
2.2.4. Insulation
All medical-grade silicones used in this study were provided by Nusil Silicone Technology© . Three insulation
processes were employed: (1) several coats of MED-6015
(an optically clear potting and encapsulating silicone elastomer) cured at 150 ◦ C for 2 h, (2) an adhesion coating
of MED2-4013 (a fast cure silicone adhesive) with an
additional three coats of MED-4210 (silicone elastomer)
cured at 45 ◦ C for 12 h, and (3) an adhesion coating of
MED2-4013 with an additional three coats of MED-4210
cured at 150 ◦ C for >30 min (Table 3). Silicone coatings
were applied with a coating speed less than 5.5 cm/h. Vacuum deaeration was not conducted prior to curing, and may
Table 3
Insulation of the various medical-grade silicones consisted of a dip-coating
procedure at a controlled rate <5.5 cm/h
Process
Base layer
silicone
(type)
Top layer
silicone
(type)
Coats
(#)
Cure
temperature
(◦ C)
Cure
time
(h)
I1
I2
I3
MED-6015
MED2-4013
MED2-4013
MED-6015
MED-4210
MED-4210
2
3
3
100
45
150
2.0
12.0
1.5
14
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
account for observed pinholes. Each insulation process is
referred to as “I#”, where I stands for insulation and #
stands for the specific process (e.g. I2 stands for insulation
process #2). The insulation processes have been ordered in
terms of improving the adhesion performance.
2.2.5. polyLIFE construction
Two groups of polyLIFEs were constructed to evaluate the
effect of insulation and the specific insulation process on the
mechanical fatigue properties of complete electrodes. The
first group was constructed with surface treatment, metal deposition and insulation processes S4·M4·I2, and the second
group was constructed with processes S4·M4·I3.
In order to increase the tensile strength and electrical redundancy of the polyLIFE, several electrodes were assembled with three individually metallized fibers. These three
metallized fibers were wound around each other and subsequently insulated with process I3. These polyLIFEs will be
referred to as (S4·M4)R ·I3, where R indicates the metallized
fiber redundancy.
2.2.6. Miniature wire LIFE construction
As an alternative to the polyLIFE and the multi-stranded
polyLIFE, two groups of solid conductor LIFE electrodes
were constructed. One group of LIFEs was constructed
from Teflon insulated, 25 m diameter Pt-Ir wire (Lefurge
et al., 1991). A second group of LIFEs was constructed
from 14 m diameter Au wire buttressed with two strands
of bare Kevlar® and subsequently insulated with process I3.
Throughout the following text, the LIFEs described in this
section will be referred to as Pt-Ir LIFEs and Au LIFEs.
approximate 1.5 cm long section of the metallized fiber to
electrical tape that had been laid out on a flat surface, (3)
pulling the fiber off the tape normal to the tape’s surface,
and (4) measuring the final resistance of the metallized
fiber. The fiber was adhered to the tape using a pressure
of 206 ± 20 kPa (n = 20, pooled data) and pulled off the
tape with 23 ± 7 mg of force (n = 20, pooled data). The
adhering pressure and pull-off force was measured with a
mass balance on two separate days (n = 10 for each day).
Failure for the tape test was defined in the same manner as
the wipe test. Samples from all groups were pooled together
and subsequently tested in random order.
Samples subjected to the tape test were also inspected for
specific adhesion failure modes using a scanning electron
microscope.
2.3.2. Metallized fiber longitudinal resistance and
interfacial impedance
The longitudinal resistance (/cm) of metallized fibers
was tested using an ohm-meter. Interfacial impedance
(k/mm) of the metallized fibers was estimated by immersing the fiber in a 4 mm diameter drop of saline and
measuring the resulting impedance to a 1 kHz application
of a 10 nA constant-current sine wave. The return electrode consisted of a stainless steel needle whose surface
impedance was below the sensitivity of the measurement.
Statistics used for these tests were two-tailed Student’s
t-tests in which unequal variance was assumed.
2.3. Material properties characterization
2.3.3. Metal film fatigue resistance
Fatigue resistance of the metallized fibers was tested by
repeated cycling around a 500 m diameter hypodermic needle (Fig. 3). Fatigue resistance was defined as the number of
2.3.1. Metal film adhesion
Adhesion performance of the sputter-deposited metal
layer was evaluated using methods that would challenge the
interface with forces comparable to forces that would be
expected during surgical handling and long-term implantation. Towards the end, we subjected the metallized fibers
to a wipe test and a tape test. The wipe test consisted of
approximately 5 cm length of the metallized fiber drawn
between two dry fingers five times and measuring the resistance before and after the test. The applied pressure during
the wipe tests was held as consistent from test to test as
possible. Applied pressure was measured with a manometer
on two separate days (n = 10 for each day). According to
the manometer, the pressure used for the wipe tests was
22.7±1.0 kPa (pooled data). Failure on this test was defined
as a measured longitudinal resistance greater than 5 k.
Electrode impedances greater than 30 k result in poor signal to noise ratios. As a typical recording zone impedance
ranged from 15 to 25 k, a maximum lead resistance of
5 k was chosen to maximize the potential signal to noise
ratio. The tape test consisted of four steps: (1) measuring
the fiber’s initial longitudinal resistance, (2) adhering an
Fig. 3. Fatigue test setup. The fatigue test consisted of continuously wrapping and unwrapping the test electrode around an approximate 500 m
diameter hypodermic needle until electrode failure (where failure meant
that the electrode was “non-conductive at rest”). Fatigue resistance was
defined as the number of cycles until the test electrode failed. Each cycle
was comprised of a 180◦ counter-clockwise rotation and a 180◦ clockwise rotation at 240◦ /s. Longitudinal resistance was evaluated by measuring the voltage drop (Ve ) across a test electrode placed in series with
a 70 k resistor and a 25 V supply. During cycling a constant axial load
of ∼3 g was applied to the test electrode with a micro-clip dangling from
the electrode’s tip end. This micro-clip was also used to complete the
current path.
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
cycles until the electrode’s longitudinal resistance became
greater than 5 k (tests were terminated at 200 kilocycles).
In order to measure the metallized fiber’s resistance, the
fiber was placed in series with a 70 k resistor and a 25 V
dc power supply. The resistance of the metallized fiber was
measured at the end of each cycle when the cycled portion
was not subjected to dynamic or static bending. The axial
load consisted of approximately 3 g of micro-clip connected
to the tip-end of the fiber. This micro-clip, in combination
with a flexible seven-stranded stainless steel cable, was also
used to complete the current path for the resistance measurements. Each cycle consisted of a +180◦ rotation followed
by a −180◦ rotation at 360◦ /s. Rotation speed was chosen
such that a 180◦ rotation occurred at least once every second, a value within the physiological rate of movement for
humans. The diameter of the hypodermic needle was chosen so that the metallization layer was subjected to bending well in the plastic range (strain >2%) of each metal’s
stress–strain curve.
Due to the non-normal distribution of the data, the
Mann–Whitney U-test was used to compare fatigue resistance between the tested groups. Variability within tested
groups was evaluated with a two-tailed F-test.
2.3.4. Insulation adhesion and leakage impedance
Adhesion performance of the insulation layer and leakage
impedance were measured. Adhesion of the insulation layer
was tested by the wipe and tape tests described above. The
wipe test consisted of wiping saline wetted fingers along the
insulated fibers and confirming adhesion with a light microscope and surface impedance measurements before and
after the test. Insulation leakage impedance (M/cm) was
measured by immersing a 4 mm length of the insulated fiber
in saline and measuring the resulting impedance at 1 kHz.
Failure on the wipe and tape tests were defined as a resulting
leakage impedance of <0.1 M/cm (a value greater than two
orders of magnitude above uninsulated surface impedances
capable of recording neural activity and greater than three
orders of magnitude greater than the longitudinal impedance
of the metallized fibers).
2.3.5. Improved mechanical properties with insulation
and redundancy
Electrode toughness for this study was based on polyLIFE
fatigue resistance, tensile strength, and long-term leakage
impedance stability. The fatigue test was used to evaluate
changes to fatigue resistance of single and redundant metallized Kevlar® fibers after application of the insulation layer.
For comparison, Pt-Ir LIFEs and Au LIFEs were also subjected to the fatigue test. A uniaxial tensile test (described
elsewhere) (McNaughton and Horch, 1996) was used to
evaluate changes to tensile strength. The flexibility of the
multi-stranded polyLIFE was measured as described elsewhere (McNaughton and Horch, 1996) and compared to the
flexibility of the Pt-Ir LIFEs in order to determine the relative
change in stiffness due to redundancy. Long-term leakage
15
impedance (M/cm) was tested on several multi-stranded
polyLIFEs by soaking them in saline for a 6-month period.
Leakage impedance measurements were conducted on day 0,
10, 90, 150, and 180 of the soak test. Leakage impedance was
estimated by immersing the fiber in a 2 mm diameter drop
of saline and measuring the resulting impedance (at 1 kHz).
Fatigue resistance differences were tested for statistical significance (α = 0.05) with the Mann–Whitney U-test. Ultimate tensile strength differences were tested for statistical
significance (α = 0.05) with a two-tailed, Student’s t-test.
3. Results
3.1. Modified polyLIFE material properties
3.1.1. Metal film adhesion
It has been found that adhesion performance is directly
affected by certain mechanical properties of the metal film
such as internal residual stress, thickness, and Young’s modulus. Barenblatt (1962) found that adhesion performance is
negatively affected by thin films with larger residual stresses,
greater thickness, and lower values of Young’s modulus.
Wipe tests and tape tests were conducted on six groups of
metallized fibers (n = 7 for each group), manufactured with
processes S1·M1, S3·M2, S3·M3, S4·M4, S2·M5, and S3·M6.
Except for one sample from group S1·M1, all groups passed
the wipe test. However, groups S1·M1, S3·M2, and S3·M3,
failed the tape test 100, 43, and 29% of the time, respectively.
Among the groups that passed the tape test, the cylindrical
loom groups experienced a significant lower mean resistance
change (percentage ± S.E.) due to the tape test than the flat
loom groups (P < 0.05 for pair-wise comparison of fibers
in groups S3·M2, S3·M3, S4·M4, S2·M5, and S3·M6 whose
longitudinal resistance had changed 14 ± 10%, 66 ± 27%,
1 ± 0.6%, 2 ± 0.7%, and 0 ± 0.1%, respectively, due to the
tape test). In general, metallized fibers from flat looms faired
worse on the wipe and tape tests than those from cylindrical
looms. Likewise, metallized fibers that had been fabricated
with higher process energies also tended to perform worse
on the wipe and tape tests.
Fig. 4 shows scanning electron micrographs of an exemplary fiber from each group subjected to the tape test. These
micrographs reveal that the tape test caused micro-cracks
in the metal layer. These cracks appear to have propagated
along the metal layer’s grain boundaries. Fibers from flat
looms showed tendencies to crack not only circumferentially, but longitudinally, leading to the failure and peeling
of large (>50 m long) sections of the metal film. However,
fibers mounted on flat looms were always metallized with
high process energies (>270 kJ), which could have resulted
in thicker films and higher substrate temperatures during the
metal deposition process. Damage due to the wipe or tape
tests on fibers from groups S4·M4, S2·M5, and S3·M6, appeared as circumferential or spiral fractures that did not result in a significant increase in longitudinal resistance.
16
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
Fig. 4. Scanning electron micrographs of metallized fibers subjected to the tape test. Metallized fibers from flat looms where delamination adhesion
failures are visible: (A) S1·M1, (B) S3·M2, (C) S3·M3. Metallized fibers from cylindrical looms where no cracking or adhesion failure is visible: (D)
S4·M4, (E) S2·M5, and (F) S3·M6 (gray bars: 10 m).
From the micrographs, it was difficult to determine in
some cases whether complete delamination was prevented
by better adhesion of the metal to the Kevlar® or by better cohesion of a non-adhered metal tube surrounding the
Kevlar® fiber.
3.1.2. Metallized fiber longitudinal resistance and
interfacial impedance
Longitudinal resistance and interfacial impedance were
measured on metallized fibers from groups manufactured
with processes S1·M1, S3·M2, S3·M3, S4·M4, S2·M5, and
S3·M6 (n = 7 for each group). Longitudinal resistance was
also measured from a single metallized fiber from each of the
metal deposition processes S2·M7. Longitudinal resistance
was found to be related to process energy used to metallize
these fibers (Fig. 5A). The interfacial impedance was found
to be logarithmically related (r 2 = 0.997) to the session
time and session energy for Au deposition (Fig. 5B).
3.1.3. Annealing effects on longitudinal resistance and
interfacial impedance
While no relationship was found between the annealing temperature and the longitudinal resistance (Fig. 6A),
annealing caused both positive and negative effects on
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
Fig. 5. (A) Process energy vs. longitudinal resistance. For details on
processes M1–M7, see Table 2. M7∗ is comprised of several separate processes conducted solely to help demonstrate the process energy/longitudinal resistance relationship. (B) Session time × session energy vs. interfacial impedance. A logarithmic relationship exists between
(session time × session energy) and interfacial impedance. This graph
suggests that interfacial impedance is related to the time taken to deposit
a certain amount of metal.
interfacial impedance (Fig. 6B). For instance, the mean surface impedance was reduced by factors of 33% (P < 0.01)
and 47% (P < 0.01) for annealing temperatures of 45 and
100 ◦ C, respectively, as compared to the control values.
However, the mean surface impedance was increased by
350% (P ≪ 0.01) and 930% (P ≪ 0.01) for annealing
temperatures of 150 and 170 ◦ C, respectively, as compared
to control values.
3.1.4. Metal film fatigue resistance
Fatigue tests were conducted on groups of metallized
fibers manufactured with processes S1·M1 (n = 8), S3·M2
(n = 10), S3·M3 (n = 6), S4·M4 (n = 7), S2·M5 (n = 4),
and S3·M6 (n = 4). Fig. 7 shows median fatigue resistance
(±S.E.) from each group and the median fatigue resistance
from Pt-Ir LIFEs (n = 12). Regardless of process energy
used for metal deposition, fibers mounted on flat looms did
not demonstrate significantly better fatigue resistance (P >
0.05) than Pt-Ir wire. However, fibers mounted on cylindrical looms demonstrated a fatigue resistance of greater than
20 (P ≪ 0.01), 250 (P < 0.01), and 490 (P < 0.01) times
17
Fig. 6. (A) Annealing temperature vs. longitudinal resistance for metallized fibers. (B) Annealing temperature vs. mean interfacial impedance
(±S.E.) for metallized fibers. Metallized fibers made with process M4
were annealed at 45, 100, 130, 150, and 170 ◦ C post-deposition, and subsequently evaluated in terms of longitudinal and interfacial impedance.
that of the Pt-Ir wire for groups S4·M4, S2·M5, and S3·M6,
respectively. In terms of flat loom mounted fibers, fatigue resistance was not correlated with either process energy (P >
0.34 for S1·M1 versus S3·M3 and P > 0.63 for S3·M2 versus S3·M3) or attempts to improve circumferential thickness
uniformity (P > 0.57 for S1·M1 versus S3·M2). For fibers
mounted on cylindrical looms, fatigue resistance was not
correlated with process energy (P > 0.2 for group S2·M5
versus S3·M6), but was negatively correlated with attempts
to improve the longitudinal thickness uniformity (P < 0.01
for group S4·M4 versus S2·M5). However, fibers mounted
on cylindrical looms not only improved the metal film’s fatigue resistance, by a factor of 140 (S2·M5, P < 0.05) and
280 (S3·M6, P < 0.05), compared to fibers mounted on flat
looms, but dramatically decreased the within-group variability for groups S2·M5 (P < 0.05), S3·M6 (P < 0.05), and
S4·M4 (P < 0.01), respectively versus pooled flat loom data.
Fig. 8 shows scanning electron micrographs of the representative metallized fibers that had been subjected to fatigue
tests. The micrographs from flat looms demonstrate both longitudinal and circumferential cracking, as well as large delaminations. Fig. 8A shows an extreme example from group
S1·M1 where the metal film failed after two order of magnitude more cycles than would have been expected according
to the group’s median value. The question here is why this
18
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
Fig. 7. Fatigue resistance of metallized fibers from groups S1·M1, S3·M2, S3·M3, S4·M4, S2·M5, and S3·M6 compared to fatigue resistance of Pt-Ir wire.
fiber remained conductive when it is obvious from inspection of the micrograph that no continuous metal layer exists.
In fact, this fiber was conductive even though only islands
existed next to large patches of delaminations. Presumably,
electrical conduction was possible through contact between
the metal islands when the fiber was in the straightened condition (i.e. at the time the resistance measurement was taken
during the fatigue test). It is believed that this behavior seen
with metallized fibers from flat looms causes the extreme
variability in fatigue tests within groups as compared to the
Pt-Ir LIFEs (P ≪ 0.01 two-tailed F-test). Compared to fatigued fibers from cylindrical looms, the length of uncracked
metal was greater with fibers from flat looms. For example,
fatigued fibers from group S4·M4 showed circumferential
cracks that were spaced at substantially closer levels than
the circumferential cracks displayed in fatigued fibers from
Fig. 8. Scanning electron micrographs of metallized fibers subjected to the fatigue test. Metallized fibers from flat looms where both longitudinal and
circumferential cracks are visible: (A) S1·M1 failed after 153 kilocycles, (B) S3·M2 failed after 7.7 kilocycles, (C) S3·M3 failed after 2.2 kilocycles.
Metallized fibers from cylindrical looms where only circumferential cracks are visible: (D) S4·M4 failed after 9.3 kilocycles, (E) S2·M5 failed after 25
kilocycles, and (F) S3·M6 test stopped before failure at 200 kilocycles (gray bars: 10 m).
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
19
Fig. 9. Fatigue resistance of polyLIFEs. Results are from two groups of miniature wire LIFE electrodes (Pt-Ir LIFE and Au LIFE) and four groups of
polyLIFEs (S4·M4, S4·M4·I2, S4·M4·I3, and (S4·M4)R ·I3).
group S3·M2 where the process energy used for each group
was similar (210 and 270 kJ, respectively).
3.1.5. Insulation adhesion performance
and leakage impedance
In order to choose an effective combination of silicones
for insulation of the polyLIFE electrode, adhesion performance and leakage impedance were tested on metallized
fibers insulated with processes I1, I2, and I3 (Table 2). Results from the wipe test indicated that insulation process
I1 provided poor adhesion and process I2 demonstrated
variable adhesion. This variable adhesion could be in part
due to an incomplete cure of the MED2-4013 layer. Nusil
had suggested a cure temperature greater than 100 ◦ C, but
a cure temperature of 45 ◦ C was used to avoid interfacial
impedance increases of the metal film. Metallized fibers
coated with insulation process I3 showed no decrease after the wipe test (n = 6) or tape test (n = 6). Since
only process I3 passed the wipe and tape tests, the result suggest that a complete cure (curing the adhesion silicone MED2-4013 above 100 ◦ C) is necessary for improved
adhesion.
According to the leakage impedance tests, insulation
processes I2 and I3 were sufficient to produce leakage
impedance values greater than 2 M/cm. For insulation
process I3, pinholes were discovered upon inspection with
the scanning electron microscope when fewer than three
coatings of MED-4210 were applied. When applying process I3, the insulation layer was estimated to be 3–5 m
thick as determined from scanning electron micrographs.
3.1.6. Improved mechanical properties of polyLIFEs
due to insulation and redundancy
Two groups of miniature wire LIFE electrodes and
four groups of polyLIFEs were subjected to the fatigue
test: (1) Pt-Ir LIFEs (n = 12), (2) Au LIFEs (n = 7),
(3) non-insulated polyLIFEs made with processes S4·M4
(n = 7), (4) insulated polyLIFEs made with processes
S4·M4·I2 (n = 7), (5) insulated polyLIFEs made with
processes S4·M4·I3 (n = 13), and (6) insulated, redundant polyLIFEs made with processes (S4·M4)R ·I2 (n = 7).
Fig. 9 shows median fatigue resistance (±S.E.) of the two
groups of miniature wire LIFEs and all tested polyLIFEs.
All insulated polyLIFE groups demonstrated a fatigue resistance of greater than 40 times than that of Pt-Ir LIFEs
(P ≪ 0.01) or Au LIFEs (P < 0.01). The fatigue resistance of non-insulated polyLIFEs was improved by 6.8
times (P < 0.05) and 3.7 times (P < 0.01) when insulated
with processes I2 and I3, respectively. No significant relation was found between cure temperature (group S4·M4·I2
versus S4·M4·I3) and fatigue resistance (P > 0.36). While
insulated, redundant polyLIFEs improved the median fatigue resistance of non-insulated polyLIFEs by greater than
2.5 times (P < 0.05 for group (S4·M4)R ·I2 versus group
S4·M4), their fatigue resistance was not significantly different than that of insulated polyLIFEs (P > 0.73 for group
(S4·M4)R ·I2 versus S4·M4·I3).
The insulated, redundant polyLIFEs’ ultimate tensile
strength (1050 ± 187 mN, mean ± S.D.) was a factor of
approximately four times greater than insulated polyLIFEs
(P < 0.0001 group (S4·M4)R ·I2 versus S4·M4·I3) and
approximately three times greater than Pt-Ir wire (P <
0.0001). Furthermore, the insulated, redundant polyLIFE remained more flexible (60 ± 19 nN m) than the Teflon-coated
Pt-Ir wire (330 ± 80 nN m) by a factor of 5 (P < 0.01).
Fig. 10 demonstrates that long-term leakage impedance
remained stable over the 6-month soak test, with all samples
(n = 10) exhibiting values >5 M/cm. Mean active zone
length (±S.D.) of the insulated, redundant polyLIFEs was
1.01 ± 0.2 mm, as measured from scanning electron micrographs. Some longitudinal wicking of the silicone between
metallized fibers (perhaps due to capillary action) was also
observed, which could account for some of the measured
variability in active zone length.
20
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
Fig. 10. Mean leakage impedance (±S.D.) during a 6-month saline soak test of polyLIFEs (n = 10). The polyLIFEs were manufactured with processes
(S4·M4)R ·I3 (i.e. three metallized fibers made with processes S4·M4 were first wound around each other and subsequently insulated with process I3).
4. Discussion
4.1. Metal–polymer adhesion
Adhesion performance of the metal sputtered onto
Kevlar® fibers mounted on flat looms was significantly improved by reactive-ion etching with O2 , as demonstrated by
the wipe and tape tests. Other studies have shown similar
metal–polymer adhesion performance improvements not
only by plasma treatments, but also by substrate heating
(during and after deposition, Baglin et al., 1991; Faupel
et al., 1999) and/or inclusion of a “glue” metal layer such as
Ti or Cr (Lee, 1991). Covalent bonding of Ti to polyimide
substrates (Ti-C) with evaporation deposition has shown
to be promoted at increased coverage (Ohuchi and Fellich,
1986) or after high energy (2 keV) Ar+ ion bombardment of the substrate prior to metal deposition (Bodö and
Sundgren, 1988). While we did not attempt direct adhesion
of the Au layer to Kevlar® , our application of the Ti (“glue”)
layer without additional plasma treatment was shown to be
insufficient. Specifically, tenuous adhesion may have been
achieved between the Ti and Kevlar® during sputtering, but
compressive residual stresses achieved in the film during
cooling may have led to adhesion failure (as evidenced
in scanning electron micrographs of flat-loom mounted
fibers). This residual internal stress was most likely caused
by thermal expansion coefficient differences; the transverse
thermal expansion coefficient of Kevlar® (60 × 10−6 ⑀/◦ C)
is nearly six times that of any of the deposited metals (8.8,
14.1, and 9.1 × 10−6 ⑀/◦ C for Ti, Au, and Pt, respectively).
Differential thermal expansion of the metals and polymer
during metal deposition may have caused significant compressive internal residual stresses within the metal film
after cooling, thus contributing to the poor adhesive and
mechanical performance observed from fibers from the flat
loom.
In addition to thermal residual stresses, Barenblatt (1962)
suggests that adhesion failure of thin film metal is affected
by other factors such as Young’s modulus and film thickness. In our case, it is not likely that Young’s modulus
had any effect on adhesion performance as it has been
shown to be independent of film thickness (Espinosa et al.,
2003). Likewise, it is unlikely that increased thickness is
the cause of the adhesion failure exhibited in our system.
For instance, group S3·M2 demonstrated worse adhesion
performance than group S3·M3 (a group for which higher
process energies presumably resulted in thicker metal films).
Furthermore, fibers from group S3·M3 likely experienced a
higher temperature and thermal expansion during sputtering than fibers from S3·M2. Therefore, we have attributed
the poor adhesion performance of fibers in group S3·M2 to
increased thermal residual stresses in the metal film due to
the multiple heating and cooling cycles (caused by breaking
vacuum and flipping the loom to promote circumferential
thickness uniformity). We believe that the improved adhesion performance for metallized fibers from flat looms can
be attributed to the minimization of thermal residual stress
in combination with plasma surface treatments.
Adhesion performance of fibers mounted on cylindrical
looms was excellent, as evidenced by minimal changes in
resistance in the tape test. However, although no delamination was visible in the scanning electron micrographs from
the tested fibers, we cannot conclude that adhesion of the
metal to the polymer was responsible for the improved performance. Although no correlation was found between adhesion and any of the surface treatment processes (S2, S3,
or S4) when used with cylindrically mounted fibers, this result could have been due more to an improvement in the
mechanical properties of the deposited metal film. For instance, rotation of the cylindrically mounted fibers would
have minimized the oblique angle of incident atoms around
the circumference, perhaps accounting for the absence of
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
longitudinally directed cracking due to the tape test. One
might propose the thinner Ti layer was responsible for the
improved adhesion performance; however, this trend was not
seen with fibers from the flat loom. The lower session energies experienced by the cylindrically bound fibers should
have contributed to the lower substrate temperatures, which
in turn may have resulted in lower residual stresses and improved adhesion performance.
Various “glue” layers, such as Ti, adhere best to polymers
when their surfaces have been oxidized (Ohring, 2002).
Reactive-ion etching and sputter etching worked equally
well in improving the adhesion performance for fibers
bound to cylindrical looms; however, we suggested using
reactive-ion etching with an O2 plasma over sputter etching as the latter process affects surfaces in a non-specific
way and requires high process power, which may lead to
excessive substrate heating.
21
The lower interfacial impedances observed in fibers
made with higher process energies could be attributed to increased surface area; caused by increased surface roughness
or larger dimensions. The larger diameters observed with
flat loom-bound fibers could partially account for the lower
interfacial impedance. However, the group S3·M6 demonstrated a much lower interfacial impedance than either
group S4·M4 or S2·M5 even though they had comparable
fiber diameters. Likewise, increased surface roughness must
also be excluded as a primary cause of increased interfacial
impedance as group S3·M6 is observed to be much smoother
than any of the flat loom groups. Metallized fibers from
group S3·M6 exhibited a similar surface roughness to the
two other cylindrical loom groups (by visual inspection of
scanning electron micrographs), but demonstrated the second lowest interfacial impedance. These results suggest that
the surface impedance of these sputter-deposited metal layers depends more on film density than on surface roughness.
4.2. Electrical properties of sputter-deposited metal
4.3. Annealing effects on electrical properties
Previous work has shown that low film densities, high
dislocation content, greater amount of trapped gases, voids,
and defects lead to poor electrical properties (Ohring, 2002).
In films containing a predominantly columnar microstructure (typically observed in films deposited at low pressures,
low substrate temperatures, and oblique angles of incident
atoms), large voids and higher concentrations of precipitates
can also be found in the intercolumnar regions, which in turn
contribute to poor electrical properties. As our films are deposited at low gas pressures, relatively low substrate temperatures (TS /TM < 0.3, where TS is the substrate temperature
and TM is the melting temperature of the sputtered metal)
and highly oblique angles of incidence, it can be assumed
that our metal deposition processes result in low density
films with columnar microstructure (qualitatively observed
in scanning electron micrographs of our metal films).
Only process energy had any significant effect on longitudinal resistance, presumably due to the increased amount
of metal deposited with higher process energies, with no
changes evident due to post-deposition annealing. However,
interfacial impedance was logarithmically related to the session energy, indicating that it is important to control the time
taken to sputter a given amount of metal. Longer times may
have allowed the substrate temperature to rise thereby increasing the incident atoms lateral mobility, which resulted
in greater film density and lower interfacial impedance.
For the purpose of recording neural activity, previous
studies have suggested that interfacial impedances of the
recording/stimulating zone should be lower than 30 k/mm
(Malagodi et al., 1989). Therefore, we suggest avoiding short
session times combined with low session energies (<9 kJ).
The lowest values for interfacial impedance were found with
session times >10 min in combination with higher session
energies. It would appear that time was a more important
factor than session energy, indicating that perhaps substrate
temperatures were related more to time than to power.
Interfacial impedance was affected by annealing temperatures. In the case where high interfacial impedance was
observed after metal deposition, annealing resulted in even
higher interfacial impedances. Since only one group was
tested for annealing effects, it is not known if a similar
trend of increasing interfacial impedance with increasing annealing temperature would be found with the other groups.
The exact cause of the increasing interfacial impedance due
to higher annealing temperature is unknown, but potential
causes could be microstructural or chemical in nature.
For instance, cracking due to differential thermal expansion between the metals and polymer (Owusu-Boahen and
King, 1998) has been shown to result in increased surface
roughness and presumably lower interfacial impedance.
Likewise, elevated temperatures known to cause grain
growth, and hence removal of grain boundaries, would
have resulted in densification of the film. However, grain
growth at substrate temperatures below 0.3 (TS /TM ) is unlikely (Hentzell et al., 1984). Therefore, surface cracking
is the most likely cause of decreased interfacial impedance
between 22 and 130 ◦ C.
At temperature higher than 130 ◦ C, increased interfacial
impedance could have been caused by chemical modification of the metal surface. Chemical modification of the surface could be directly related to diffusion characteristics of
the metal film. For thin metal films, the diffusion coefficient
can be 1015 times greater than in bulk due to the high density of defects, grain boundaries and vacancies, making diffusion times through the small dimension films characteristically short. In thin gold films with ∼500 nm grains and
at low temperatures (<0.3 TS /TM ), diffusion tends to occur mainly through the grain boundaries (Gupta and Asai,
1974). Any precipitates or oxide compounds formed at the
grain boundaries can out-diffuse, thus chemically modifying the surface (Ohring, 2002). This situation is amplified
22
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
by the larger surface-to-bulk ratios involved in thin films.
Although no proof exists that would directly point to any
of the specific mechanisms mentioned, it is our belief that
diffusion is in some way related to the increased interfacial impedance observed after heating above 130 ◦ C. Surface analysis tools such as electron spectroscopy for chemical analysis (ESCA) or secondary ion mass spectrometry
(SIMS) should help reveal the exact compositional changes
caused by post-deposition heating.
4.4. Metal film fatigue resistance
The greatest fatigue resistance was found with fibers that
had been bound to cylindrical looms and metallized with
lower process energies. Fibers bound to flat looms and
metallized with higher process energies, not withstanding
some extreme cases, tended to show worse performance and
demonstrated higher variability. Additionally, attempts to
create circumferential or longitudinal thickness uniformity
either made no difference (flipping of flat loom) or were
detrimental (end-for-end reorientation of cylindrical loom)
to the metal film’s fatigue resistance.
The improved fatigue resistance of metallized fibers
bound to cylindrical looms could be explained by their presumably lower film thickness, higher density, or crack propagation morphology. As mentioned earlier, thinner films
exhibit higher yield stresses, at the cost of a more brittle
behavior. However, lower film thickness, as well as lower
incident atom mobility (assumed to be the case for metal
deposition with lower session and process energies of the
cylindrical bound fibers) has a tendency to negatively affect
density. Therefore, it is not likely that density had a significant effect on fatigue resistance, at least for group S2·M5.
In terms of crack morphology, crack initiation is promoted
by rough surfaces, corrosive environments, and mechanical failure along grain facets or boundaries (Suresh, 1991).
For instance, it has been shown that the microscopic valleys
or notches of irreversibly roughened metal surfaces cause
stress concentrations and additional crack nucleation (Wood,
1958). Likewise, corrosive environments such as oxygen
(Gough and Sopwith, 1932; Thompson et al., 1956) or seawater (Suresh, 1991) tend to promote surface roughening,
by reacting with freshly revealed material or metal protrusions at the surface, caused by plastic deformation. Once the
protruding metal has reacted or oxidized, reverse movement
back within the bulk is impeded, resulting in surfaces with
greater roughness. Grain boundaries tend to separate highly
misoriented grains, and crack initiation is promoted when
the deformation within a single grain is directed at the intersection of two neighboring grains located at the film surface
(Porter and Levy, 1960). Additionally, preferential oxidation
at grain boundaries (Duquette, 1979) and void formation
around precipitates found in grain boundaries (Vasudévan
and Doherty, 1987) can cause microscopic stress concentrations due to notch development or intergranular cavitation.
These stress concentrations can in turn initiate cracks and
promote crack propagation along the weak grain boundaries.
Perhaps the most likely cause for the improved fatigue resistance of groups S2·M5 and S3·M6 are the relatively smoother
surfaces observed (as compared to flat loom-bound fibers)
and more densely packed grains. In addition to the inhibited
crack initiation and propagation morphology, improved fatigue resistance could be attributed to lower film thickness,
improved adhesion performance, and lower residual stresses
in the cylindrically mounted fibers. A qualitative comparison of the effects of thickness on metal film morphology due
to externally imposed bending is provided in Appendix A.
4.5. Insulation adhesion performance and
leakage impedance
Insulation process I3 provided the best adhesion consistency and leakage impedance. The improved adhesion was
due to the use of MED2-4013 as a base or “glue” layer as
it contained adhesion promoters. However, use of multiple
cycles of >100 ◦ C is not recommended due to the detrimental effect on the interfacial impedance, mechanical integrity,
and adhesion of the metal film. Since a complete cure of
MED2-4013 is only possible at or slightly above 100 ◦ C,
one might attempt to obtain a silicone that includes adhesion
promoters, but with a low temperature cure cycle. We have
obtained a low temperature cure version of MED2-4013 in
the past, but the work time of this silicone was insufficient
for our needs. It was also found that at least three layers of
MED-4210 were necessary to prevent pinholes.
4.6. Improved mechanical properties of the polyLIFE
due to insulation and redundancy
Our study demonstrates improved fatigue resistance by
the application of silicone or fiber redundancy. The silicone layer may inhibit crack formation at the metal film
surface by the mechanism of either smoothing the surface
(a phenomena known to increase the fatigue life of metals) (Basinski et al., 1983) or creating a passivation layer
that inhibits continual surface oxidation, thus allowing reversibility of plastic deformations. Alternatively the silicone
might hold islands of metal in tight physical contact. The redundancy technique most likely improves fatigue resistance
by the mechanism of probability. For instance, the greater
number of metallized fibers used for a single electrode improves the chances that one will last to the measured fatigue
limit. Observations from scanning electron micrographs of
three metallized fibers that had been wound around each
other, but not insulated, demonstrate fiber-to-fiber abrasion,
resulting in significant damage to the metal film. Fatigue
test results demonstrate that the damage resulting from the
act of winding the metallized fibers did not negatively affect
fatigue resistance of insulated, redundant polyLIFEs. However, no significant positive effect on fatigue resistance was
observed either. The most significant effect of redundancy
was an increase to the electrode’s tensile strength without
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
a substantial decrease in flexibility, the original polyLIFEs’
primary design advantage over the Pt-Ir LIFEs.
5. Conclusions
We have demonstrated improvements to the metal film adhesion, fatigue resistance, insulation adhesion consistency,
leakage impedance, and mechanical properties of finished
polyLIFE designs. Metal adhesion was improved both by
reactive-ion etching of the bare Kevlar® fibers, adjustments
to the geometry, and rotation of the fibers during sputtering.
The mechanical quality of the deposited films was better for
fibers that were bound to cylindrical looms that were axially
rotated. It is presumed that this type of rotation and mounting helped minimize the abundance of weak intergranular
regions. We feel that crack propagation along weak grain
boundaries is the main failure mechanism in loss of adhesion and fatigue failure. Cylindrically bound fibers also performed better on fatigue testing due to thinner films, and
minimization of residual internal stresses caused by differential thermal expansion.
Electrical properties such as longitudinal resistance were
related to the process energy for metal deposition, while
the interfacial impedance was found to be logarithmically
related to the session energy. Although high temperatures did
not affect longitudinal resistance, they did cause significantly
elevated interfacial impedances; a situation we would like
to avoid in order to ensure good recording properties.
The insulation adhesion consistency and leakage
impedance was improved with process I3, however the
high temperatures necessary to cure the adhesion layer
may be detrimental to the surface impedance of the metal
film. Application of the insulation layer caused greater fatigue resistance for the finished polyLIFEs, as might have
been expected; surprisingly, redundancy did not improve
fatigue resistance. However, we feel that the improved tensile strength of the multi-fiber polyLIFE design provides
sufficient rationale to recommend redundancy for chronic
applications over the single fiber design.
Alternative flexible intrafascicular electrodes could involve silicon- or polyimide-based substrates, or the use
of intrinsically conductive fibers. Since the silicon- and
polyimide-based interfaces are founded on time-tested micromachining techniques they have an advantage over hand
made electrodes in that one has the ability to quickly modify designs (fast prototyping) and can assert precise control
over active zone size and repeatability. For silicon to obtain
any degree of flexibility, the dimensions must be reduced to
such a degree that the final product is mechanically fragile.
A silicon-based ribbon cable developed for high flexibility
demonstrated successful function after chronic implantation
of up to 1 year (Hetke et al., 1994), but this interconnect
cable was not designed with the highly mobile peripheral
nerve in mind and is thought by the authors to be too stiff
and brittle to be practical for intrafascicular application
23
(Najafi and Hetke, 1990; Najafi et al., 1990; Petersen, 1982).
Several highly flexible polyimide-based devices have been
developed for cortex (Boppart et al., 1992; Rousche et al.,
2001) or peripheral nerves (González and Rodrı́guez, 1997;
Rodrı́guez et al., 2000; Stieglitz and Beutei, 1997), but are
not directly applicable to intrafascicular peripheral nerve
recording/stimulation, or rely on the regeneration of nerve
stumps. Mechanically, the metal layers of these devices
have demonstrated resistance to fatigue, but seem to have
been tested only in the elastic strain range. Therefore, in addition to the geometrical differences, a direct comparison of
fatigue resistance between metal layers deposited on planar
polyimide substrates versus Kevlar® fibers was not possible.
To the authors’ knowledge, only one polyimide-based
longitudinal intrafascicular device has been developed, but
it has somewhat larger dimensions and higher interfacial
impedance (Yoshida et al., 2000). As a substrate material, polyimide possesses some advantages over Kevlar®
in that its thermal expansion coefficients more closely
match those of the deposited metals. However, polyimide’s
tensile strength and Young’s modulus (55.8 MPa and
3.2 GPa, respectively) are substantially lower than Kevlar’s®
(3500–3600 MPa and 59–124 GPa, respectively) (Wen,
1996), and consequently, devices with similar dimensions to
our 12 m diameter fibers would not likely be strong enough
in tension to be inserted within fascicles. Metal layers on
polyimide substrates have been applied with evaporation
techniques, generally known to give greater film thickness
variability, and worse mechanical and adhesion performance. This thickness variability and worse mechanical performance is primarily due to “spitting” and lower energy of
impinging atoms as compared to sputtering (Ohring, 2002).
Intrafascicular electrodes made from intrinsically conductive fibers could potentially combine high flexibility with
near infinite fatigue resistance. Carbon fibers have demonstrated sufficient intraneural recording capabilities, but their
brittle nature prohibits chronic implantation (McNaughton
and Horch, 1994). On the other hand, intrinsically conductive polymers (polyaniline, polythiophene, polypyrrole) with
sufficiently low longitudinal resistance, have been found to
lack sufficient tensile strength for our application (Andreatta
et al., 1989; Hsu et al., 1993; Moulton and Smith, 1992;
Nemoto and Marks, 1991).
Future development of metal clad Kevlar® fibers would
involve further adjustment to the surface treatment or metal
deposition processes as well as modification of redundancy
methods. Our reactive-ion etch time could be increased in
order to more extensively roughen and oxidize the surface to
promote better adhesion performance. However, extremely
long etching times could detrimentally affect the fiber’s tensile strength and its integrity at the interfacial region. This
weakening effect has previously been seen with wet chemical etching processes of long duration (Sheu et al., 1994).
In order to increase the density of the deposited metal film,
a negative substrate bias greater than −200 V could be applied. The negative bias changes the electric field near the
24
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
substrate, thus enhancing the flux and energy of charged
incident atoms (Ohring, 2002). The increased energy of
incident atoms is thought to improve adhering atoms lateral
mobility and promote resputtering that can remove adsorbed
gas atoms and eliminate the phenomena of void formation
common at low pressures and substrate temperatures. With
the use of bias sputtering, a myriad of metal film properties
can be tailored such as increased density and step coverage, decreased resistivity, modification of hardness and
residual stresses, modification of columnar microstructures,
and improved adhesion performance. Longitudinal coating
thickness uniformity could be improved by combining phasic linear translation with the long axis rotation already
achieved with the cylindrical loom. This phasic longitudinal motion may also minimize the shadowing effect due
to oblique angles of incident atoms experience at either
end of the fiber. The interfacial impedance of the films
might be minimized by adjusting the session time and energy. Finally, one might enjoy better fatigue resistance with
multi-fiber polyLIFE designs if the fibers were not wound
around each other, thus avoiding excessive metal-to-metal
abrasion.
Fig. 11. Effects of metal film thickness on deformation behavior. Due to externally imposed bending, (a) thinner films show brittle fracture, while (b)
thicker films show ductile deformation. The greater film thickness in part b is also evidenced by the larger grains.
Fig. 12. Effects of thermal residual stresses on metal film integrity. Extremely long session times result in high substrate temperatures and subsequent
thermal stress. Due to externally imposed bending of similar thickness films, (a) thermally stressed films demonstrate catastrophic failure, and (b) thermally
unstressed films demonstrate ductile deformation. For parts a and b, the session times for Au deposition were 50 and 3 min, respectively.
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
Acknowledgements
This study was supported by NINDS of the US National
Institutes of Health.
Appendix A
Metal film mechanical properties are directly affected by
both metal film thickness and session time, as observed
from scanning electron micrographs (Figs. 11 and 12). Even
though gold is commonly thought to be a highly ductile material, brittle behavior is seen in thinner films (Fig. 11a). As
metal film thickness increases, plastic or ductile deformation precedes ultimate failure (Fig. 11b). During microbeam
load-deflection experiments, Espinosa et al., found this transition from brittle to ductile deformation behavior occurred
in beams between 0.5 and 1.0 m thick (Espinosa et al.,
2003). Fig. 12a shows the effects of residual thermal stress
on film integrity. Presumably, an excessively long session
duration resulted in a high substrate temperature. The differential expansion and contraction of the involved materials
caused excessive thermal stress that, upon bending, caused
catastrophic failure in the metal film. With much shorter session times, substrate heating is minimized during sputtering
and this failure mode can be avoided (Fig. 12b). With the
aid of scanning electron microscopic analysis, we can tailor the metal deposition process to achieve either ductile or
brittle failure (according to our specific needs) and prevent
residual thermal stresses.
References
Akin T, Najafi K, Smoke RH, Bradley RM. A micromachined silicon sieve
electrode for nerve regeneration applications. IEEE Trans Biomed Eng
1994;41:305–13.
Andreatta A, et al. Electrically conductive polyblend fibers of polyaniline
and p-phenylene terephthalamide. Polym Commun 1989;31:275–8.
Baglin JEE. Interface design for thin film adhesion. In: Lee L-H, editor.
Fundamentals of adhesion. New York: Plenum Press; 1991. p. 363–82.
Baglin JEE. Interface structure, adhesion, and ion beam processing. In:
Pauleau F, editor. Materials and processes for surface and interface
engineering. The Netherlands; 1995. p. 111–49.
Barenblatt GI. Mathematical theory of equilibrium cracks in brittle fracture. Adv Appl Mech 1962;7:55–129.
Basinski ZS, Pascual R, Basinski SJ. Low amplitude fatigue of copper
single crystals I. The role of the surface in fatigue failure. Acta Metall
1983;31:591–602.
Bodö P, Sundgren J-E. Ion bombardment and titanium film growth on
polyimide. J Vacuum Sci Technol 1988;A6:2396–402.
Boppart SA, Wheeler BC, Wallace CS. A flexible perforated microelectrode array for extended neural recordings. IEEE Trans Biomed Eng
1992;39:37–42.
Bowman BR, Erickson RCI. Acute and chronic implantation of coiled
wire intraneural electrodes during cyclical electrical stimulation. Ann
Biomed Eng 1985;13:75–93.
Branner A, Stein RB, Normann RA. Selective stimulation using a
slanted microelectrode array in cat sciatic nerve. J Neurophysiol
2001;85:1585–94.
25
Dong L, Smith RW, Srolovitz DJ. A two-dimensional molecular dynamics
simulation of thin film growth by oblique deposition. J Appl Phys
1996;80:5682–90.
Duquette DJ. Environmental effects I: general fatigue resistance and
crack nucleation in metals and alloys. In: Meshii M, editor. Fatigue
and microstructure. Metals Park: American Society for Metals; 1979.
p. 335–62.
Edell DJ. A peripheral nerve information transducer for amputees:
long-term multichannel recordings from rabbit peripheral nerves. IEEE
Trans Biomed Eng 1986;BME-33:393–403.
Espinosa HD, Prorok BC. Size effects on the mechanical behavior of gold
thin films. In: Proceedings of the symposium on mechanical properties of MEMS structures, the ASME 2001 international mechanical
engineering congress and exposition, New York; 2001.
Espinosa HD, Prorok BC, Fischer M. A methodology for determining
mechanical properties of freestanding thin films and MEMS materials.
J Mech Phys Solids 2003;51:47–67.
Fancey KS, Matthews A. Plasma assisted physical vapor deposition. In:
Rickerby DS, Matthews A, editors. Advanced surface coatings. Glasgow: Blackie; 1991.
Faupel F, Thran A, Azporojtchenko V, Kiene M, Strunskus T, Behnke
K. Nucleation, growth, interdiffusion, and adhesion of metal films on
polymers. In: Kraft O, editor. Stress induced phenomena in metallization. Stuttgart, Germany: Fifth International Workshop; 1999.
González C, Rodrı́guez M. A flexible perforated microelectrode array
probe for action potential recording in nerve and muscle tissues. J
Neurosci Methods 1997;72:189–95.
Gough HJ, Sopwith DG. Atmospheric action as a factor in fatigue of
metals. J Inst Met 1932;49:93–122.
Grill Jr WM, Mortimer JT. Quantification of recruitment properties of
multiple contact cuff electrodes. IEEE Trans Rehabil Eng 1996;4:49–
62.
Gupta D, Asai KW. Grain-boundary self-diffusion in evaporated Au films
at low temperatures. Thin Solid Films 1974;22:121.
Hentzell HTG, Grovenor CRM, Smith DA. Grain structure variation
with temperature for evaporated metal films. J Vacuum Sci Technol
1984;A2:218.
Hetke JF, Lund JL, Najafi K, Wise D, Anderson DJ. Silicon ribbon cables
for chronically implantable microelectrode arrays. IEEE Trans Biomed
Eng 1994;41:314–21.
Hsu CH, et al. Polyaniline spinning solution and fibers. Synth Met
1993;59:37–41.
Kovacs GTA, Stormet CW, Rosen JM. Regeneration microelectrode array
for peripheral nerve recording and stimulation. IEEE Trans Biomed
Eng 1992;39:893–902.
Lamendola R, Matarrese E, Creatore M, Favia P, d’Agostino R. Plasma
treatment of polymers for improving Al adhesion. Materials Research
Society Symposium Proceedings. Stuttgart, Germany: Materials Research Society; 1999.
Lamont LT. Thermal history of substrates during sputtering and sputter
etching. Solid State Technol 1979;22:107–12.
Lawrence SM, Larsen JO, Horch KW, Riso R, Sinkjaer T. Long-term biocompatibility of implanted polymer-based intrafascicular electrodes. J
Biomed Mater Res 2002;63:501–6.
Lee L-H. Fundamentals of adhesion. New York: Plenum Press; 1991.
Lefurge T, Goodall E, Horch K, Stensaas L, Schoenberg A. Chronically implanted intrafascicular recording electrodes. Ann Biomed Eng
1991;19:197–207.
Maissell LI. Sputtering. An introduction to thin films. New York: Gordon
and Breach Science Publishers; 1973. p. 39–59 [Chapter 3].
Malagodi MS, Horch KW, Schoenberg AA. An intrafascicular electrode
for recording of action potentials in peripheral nerves. Ann Biomed
Eng 1989;17:397–410.
Malmstrom JA, McNaughton TG, Horch KW. Recording properties and
biocompatibility of chronically implanted polymer-based intrafascicular electrodes. Ann Biomed Eng 1998;26:1055–64.
26
S.M. Lawrence et al. / Journal of Neuroscience Methods 131 (2003) 9–26
McNaughton TG, Horch KW. Action potential classification with
dual channel intrafascicular electrodes. IEEE Trans Biomed Eng
1994;41:609–16.
McNaughton TG, Horch KW. Metallized polymer fibers as leadwires and
intrafascicular microelectrodes. J Neurosci Methods 1996;70:103–10.
Moulton J, Smith P. Gel processing of electrically conductive blends of
poly(3-octylthiophene) and ultrahigh molecular weight polyethylene.
J Polym Sci 1992;30:871–8.
Najafi K, Hetke JF. Strength characterization of silicon microprobes in
neurophysiological tissues. IEEE Trans Biomed Eng 1990;37:474–81.
Najafi K, Ji J, Wise KD. Scaling limitations of silicon multichannel
recording probes. IEEE Trans Biomed Eng 1990;37:1–11.
Nemoto H, Marks TJ. Conductive polymer-high modulus polymer alloys. A solution route to Kevlar-polypyrrole hybrid fibers. Synth Met
1991;41–43:415–9.
Nieuwenhuizen JM, Haanstra HB. Microfractography of thin films. Philips
Tech Rev 1966;27:87–91.
Nix W. Mechanical properties of thin films. Metall Trans 1989;20A:2217–
45.
Ohring M. The materials science of thin films: deposition and structure.
San Diego, CA: Academic Press; 2002.
Ohuchi FS, Fellich SC. Metal polyimide interface: a titatium reaction
mechanism. J Vacuum Sci Technol 1986;A4:1039–45.
Owusu-Boahen K, King AH. TEM study of yielding in polycrystalline
gold thin films, vol. 505. Materials Research Society. Mater Res Soc
Symp Proc 1998:383–8.
Petersen KE. Silicon as a mechanical material. Proc IEEE 1982;70:420–
57.
Porter J, Levy JC. The fatigue curves of copper. J Inst Met 1960;89:86–9.
Rodrı́guez FJ, Ceballos D, Schuttler M, Valero A, Valderrama E, Stieglitz
T, et al. Polyimide cuff electrodes for peripheral nerve stimulation. J
Neurosci Methods 2000;98:105–18.
Rousche PJ, Pellinen DS, Pivin J, Williams JC, Vetter RJ, Kipke DR.
Flexible polyimide-based intracortical electrode arrays with bioactive
capability. IEEE Trans Biomed Eng 2001;48:361–71.
Rozman J, Sovinec B, Triep M, Zorko B. Multielectrode spiral cuff
for ordered and reversed activation of nerve fibers. J Biomed Eng
1993;15:113–20.
Rutten WLC, Wier HJv, Put JHM. Sensitivity and selectivity of intraneural
stimulation using a silicon electrode array. IEEE Trans Biomed Eng
1991;38:192–8.
Sheu GS, Shyu SS. Surface modification of Kevlar-149 fibers by gas
plasma treatment. J Adhesion Sci Technol 1994;8:531–42.
Sheu GS, Lin TK, Shyu SS, Lai JY. Improvement of adhesion of Kevlar
fiber to epoxy by chemical modification. J Adhesion Sci Technol
1994;8:511–30.
Stensaas SS, Stensaas LJ. Histopathological evaluation of materials implanted in the cerebral cortex. Acta Neuropathol (Berl) 1978;41:145–
55.
Stieglitz T, Beutei HMJU. A flexible, light-weight multichannel seive
electrode with integrated cables for interfacing regenerating peripheral
nerves. Sens Actuators 1997;A60:240–3.
Stieglitz T, Meyer JU. Microtechnical interfaces to neurons. Microsyst
Technol Chem Life Sci 1997:131–62.
Stuart RV. Sputtering. Vacuum technology, thin films, and sputtering. New
York: Academic Press; 1983. p. 3–135 [Chapter IV].
Sugawara M, Stansfield BL. Plasma etching: fundamentals and applications. New York: Oxford University Press; 1998.
Suresh S. Fatigue of materials. Cambridge: University Press; 1991.
Thompson C. The yield stress of polycrystalline thin films. J Mater Res
1993;8:237–8.
Thompson N, Wadsworth NJ, Louat N. The origin of fatigue fracture in
copper. Philos Mag 1956;1:113–26.
Vasudévan AK, Doherty R. Grain boundary ductile fracture in precipitation
hardened aluminum alloys. Acta Metall 1987;35:1193–219.
Veltink PH, van Alste JA, Boom HB. Multielectrode intrafascicular and
extraneural stimulation. Med Biol Eng Comput 1989;27:19–24.
Veraart C, Grill WM, Mortimer JT. Selective control of muscle activation
with a multipolar nerve cuff electrode. IEEE Trans Biomed Eng
1993;40:640–53.
Walter JS, Griffith P, Sweeney J, Scarpine V, Bidnar M, McLane J, et al.
Multielectrode nerve cuff stimulation of the median nerve produces
selective movements in a raccoon animal model. J Spinal Cord Med
1997;20:233–43.
Wen J. Some mechanical properties of typical polymer-based composites. In: Mark JE, editor. Physical properties of polymers handbook.
Woodbury, NY: American Institute of Physics; 1996. p. 371–7.
Wood WA. Formation of fatigue cracks. Philos Mag 1958;3:392–9.
Yoshida K, Pellinen D, Pivin D, Rousche P, Kipke D. Development of
the thin-film longitudinal intrafascicular electrode. In: Proceedings of
the Fifth Annual Conference of the International Functional Electrical
Stimulation Society. Aalborg, Denmark; 2000. p. 279–81.