Novel therapeutic applications for neural implants require miniaturized devices. Miniaturization imposes stricter requirements for reliability of materials. Pilot clinical studies suggest that rapid failure of the miniaturized neural implants in the body presents a major challenge for this type of technology. Traditional evaluations of neural implant performance over clinically relevant durations present time- and resource-intensive experiments in animals. Reactive accelerated aging (RAA) is an in vitro test platform that was developed to expedite durability testing of neural implants, as a screening technique designed to simulate the aggressive physiological environment experienced by the implants. This approach employs hydrogen peroxide, which mimics reactive oxygen species, and a high temperature to accelerate chemical reactions that lead to device degradation similar to that found with devices implanted in vivo. The original RAA system required daily manual maintenance and was prone to variability in performance. To address these limitations, this work introduces automated reactive accelerated aging (aRAA) with closed-loop monitoring components that make the system simple, robust, and scalable. The core novel technology in the aRAA is electrochemical detection for feedback control of hydrogen peroxide concentration, implemented with simple off-the-shelf components. The aRAA can run multiple parallel experiments for high-throughput device testing and optimization. For this reason, the aRAA provides a simple tool for rapid in vitro evaluation of the durability of neural implants, ultimately expediting the development of a new generation of miniaturized devices with a long functional lifespan.

Neural implants are medical devices that modulate and record the activity of the nervous system for therapeutic applications. Common clinical examples of implantable neurological devices are cochlear implants for restoration of hearing, spinal cord stimulators for pain relief, and deep brain stimulators for treatment of essential tremor.1 The repertoire of diseases to be addressed with neural implants is continuously expanding with the increased attention on innovative applications.2,3 Cortical neural implants for brain-computer interfaces (BCIs) have evolved from early pilot trials4 into multi-site clinical studies with a community-developed roadmap for broader clinical implementation.5 Advancements in peripheral nerve interfaces drive bioelectronic medicine, where targeted neuromodulation of nerves connecting peripheral organs is used to treat a variety of medical conditions with unprecedented flexibility and potential to mitigate the side effects.6 

Novel applications for neural implants require higher spatial resolution to increase the “bandwidth” of the communication between medical devices and a nervous system. Progress in neuroprosthetic BCI systems has been achieved with increased numbers of neurons that can be recorded simultaneously,7 which requires high-density arrays of microelectrodes with dimensions comparable to a size of a single neuron (<100 μm). Additionally, use of sensory feedback for neuroprosthetics relies on a high spatial resolution of electrical stimulation of neurons responsible for tactile sensation. Furthermore, a main target for bioelectronic medicine modulation is the vagus nerve, where spatially focused stimulation might be the key to eliminating stimulation-induced side effects.

An increase in spatial resolution requires miniaturization of neural implants, which is accomplished using microfabrication methods developed in the semiconductor industry. However, miniaturization of these devices brings additional challenges associated with rapid decline in device performance that has been extensively documented and considered to be a great challenge by the neuroengineering community.8,9 Discovery of new materials and development of new device designs for a reliable miniaturized neural interface requires rapid and high-throughput testing methods. However, the performance testing for traditional clinical neural implants has typically been performed in large animals1 due to the large size of the implants and requirement for clinical translation of the testing results.10 While miniaturization of neural implants enabled testing of their performance in rodents,11 animal experiments are expensive and lengthy, require highly skilled personnel, and are subject to a variability. To expedite neural implant development, we designed a reactive accelerated aging (RAA)12 system for rapid simulation of in vivo degradation of these devices using hydrogen peroxide (H2O2),13 which mimics reactive oxygen species (ROS) associated with immune system attack14 and a high temperature15 to accelerate the chemical reactions. The RAA provided valuable information on failure modes of cortical neural implants that included metal dissolution, moisture penetration, and degradation of insulation. The patterns of neural implant degradation observed after 7 days in RAA appeared very similar to data from chronic animal studies.16–18 Furthermore, a recent analysis of data on explanted cochlear implants pointed to the importance of ROS in degradation of insulation and electrodes,19 suggesting that RAA accurately simulates the in vivo environment. While reactive accelerated aging has been inspired by the testing of device shelf life that employs a high temperature to accelerate chemical reactions,15 the RAA was not designed to provide the quantitative estimates for an implant life expectancy in vivo. A very high temperature (87 °C) leads to rapid degradation, but it also changes properties of a polymer material such as the reactions that occur at this temperature are different from the reactions that occur at ambient temperature.20 Due to these limitations, the results from high-temperature RAA experiments provide qualitative insight for in vivo device degradation.

In a short time, the RAA received some attention21,22 due to its ability to provide data on materials’ reliability and robustness of the implant design without the need for animal experiments. However, the original RAA design12 was complex and required constant human attention, which prevented wider adaptation of the RAA. We have gone through several iterations in developing an automated RAA system, incorporating optical and electrochemical detection, and have devised what we now present as a robust and reproducible design for the automated RAA (aRAA) that is simple to build and easy to operate. The key part of automation is a feedback loop that relies on the electrochemical activity of H2O2

H2O2O2+2H++2e.

This method of H2O2 concentration measurement is ubiquitously used in enzyme-coupled biosensors including clinical devices, suggesting that it is a simple and robust analytical technique. We have developed chronoamperometric (constant potential pulses) detection of H2O2 concentration on platinum microelectrodes, which enables accurate and stable measurement of H2O2 over a long period of time (weeks). Additionally, we have modified the system and built it using off-the-shelf Arduino and Raspberry Pi based parts. The new aRAA design allows easy scaling of the RAA setup to have multiple modules running in parallel at different conditions (H2O2 concentration, temperature, and time) to precisely tune the intensity of reactive aging. To take advantage of this new feature, we performed simultaneous reactive aging of Blackrock and TDT (Tucker-Davis Technologies) microelectrode arrays (MEAs) at two different temperatures (87 °C and 67 °C) and characterized the degradation of each implant using electron microscopy and impedance spectroscopy. Data on the reactive aging-induced degradation of Blackrock microelectrode arrays confirmed that the aRAA design has the same performance as the original RAA system. Milder degradation of TDT at the lower temperature demonstrated that fine-tuning of RAA conditions is possible to better match in vitro reactive aging conditions to in vivo environment and can be easily implemented in the new aRAA design.

The automated reactive accelerated (aRAA) system was developed with two independent modules, each of which uses different sets of experimental parameters (Fig. 1). Each reaction vessel consisted of a five-neck European-style 125 ml flask (Ace Glass, Inc., Vineland, NJ), a proportional-integral-differential (PID) temperature controller, an electric heating mantle, an electrochemical H2O2 feedback controller, a three-channel peristaltic pump (400DM3, 120s, Watson-Marlow, Wilmington, Massachusetts), and a magnetic stirrer (Nuova ii, Ramsey, Minnesota). Each reaction flask had two small ports dedicated to a Pt working microelectrode and carbon rod counter electrode inserted with the stoppers that came with the flask. Left and right side ports were used to hold thermocouples and tubes for delivery of the liquids via a 14/20 polytetrafluoroethylene (PTFE) cap with an opening (Cat# F20309-1680, belart.com). Neural implants for testing were placed in a central port with Blackrock microelectrode arrays being held by a custom build PTFE holder and the TDT array being attached with PTFE tape to a PTFE rod that was fixed in an opening of a 24/40 PTFE cap (Cat# F20311-1718, belart.com).

FIG. 1.

Diagram of the automated reactive accelerated aging (aRAA) system with a feedback loop for maintenance of H2O2 concentration. Each aRAA module is a five-neck flask with a magnetic stirrer and is equipped with feedback loops to maintain the desired temperature and H2O2 concentration. The temperature feedback loop consists of a heating mantle connected to the PID temperature control unit. H2O2 concentration is measured electrochemically using the Rodeostat potentiostat, in a two-electrode scheme with the platinum microelectrode as a working electrode and the carbon rod as a counter electrode. The temperature and H2O2 concentration are sampled by the Raspberry Pi control unit that activates pumps via power relay switch to deliver stock H2O2 solution whenever H2O2 concentration falls below a threshold value and to remove waste to maintain the constant volume.

FIG. 1.

Diagram of the automated reactive accelerated aging (aRAA) system with a feedback loop for maintenance of H2O2 concentration. Each aRAA module is a five-neck flask with a magnetic stirrer and is equipped with feedback loops to maintain the desired temperature and H2O2 concentration. The temperature feedback loop consists of a heating mantle connected to the PID temperature control unit. H2O2 concentration is measured electrochemically using the Rodeostat potentiostat, in a two-electrode scheme with the platinum microelectrode as a working electrode and the carbon rod as a counter electrode. The temperature and H2O2 concentration are sampled by the Raspberry Pi control unit that activates pumps via power relay switch to deliver stock H2O2 solution whenever H2O2 concentration falls below a threshold value and to remove waste to maintain the constant volume.

Close modal

Automation was implemented with the Raspberry Pi that activated the peristaltic pumps to deliver H2O2 as necessary and logged the temperature and H2O2 concentration for each reaction module. The pumps delivered concentrated H2O2 [1.5M in phosphate-buffered saline (PBS), kept in a dark glass vessel] and removed excess solution to maintain constant volume using ∼16 cm pieces of soft Viton tubing (1.59 mm ID, 3.18 mm OD, Cat# 5119K39) that was connected to PTFE tubing (1.59 mm ID, 3.18 mm OD, Cat# 5239K24) using barbed connectors (Cat# 53055K111, McMaster-Carr, Robbinsville, NJ). The Raspberry Pi was equipped with a battery-backed PCF-8523 clock (Adafruit, New York City, NY) to account for possible system power loss and to facilitate precise data logging. A 5-port powered universal serial bus (USB) extension was added to the Raspberry Pi USB ports to enable connection to two Rodeostat potentiostats, two PID temperature controllers, a keyboard, and a mouse. The main script, running on the Raspberry Pi and written in Python, maintained serial communication with the peripherals to set, read, and log process values. Finally, the Raspberry Pi was configured to initialize the script at startup in the case of power failure and to run it until the specified end date for each module was reached. All solutions were prepared with deionized water (18 MΩ cm) using PBS tablets and 30% H2O2 solution (Fisher Scientific, Hampton, NH). To initialize an experiment, flasks were filled with H2O2 solution in PBS (15 mM). The Python scripts and additional protocols are provided in a GitHub repository.23 

Each reaction vessel was equipped with a hermetically sealed PFA-jacketed thermistor (Omega Engineering, Inc., Norwalk, Connecticut) connected to a Platinum Series PID controller (Omega Engineering, Inc., Norwalk, Connecticut). PID parameters were tuned for each individual reaction module using the embedded auto-tune feature of the PID device. Each PID controller powered a DC-controlled solid state relay (Omega Engineering, Inc., Norwalk, Connecticut) using on/off control of a resistive element heating mantle (Glas-Col LLC, Terre Haute, Indiana). The PID controllers were connected to the Raspberry Pi via USB to set or record the module temperature.

H2O2 concentration was measured using electrochemical detection with chronoamperometry. In this technique, voltage pulses were applied to a 25 μm platinum disk working electrode [Fig. 2(a)] and a current was measured in a two-electrode scheme with a carbon rod of 6.35 mm diameter (Electron Microscopy Sciences, Fort Washington, Pennsylvania) as a counter electrode. The platinum microelectrode was fabricated in-lab using a platinum wire and glass capillaries according to the procedure24 detailed in Sec. II D. The electrochemical experiments were performed with the Rodeostat (IO Rodeo, Pasadena, California), an open-source potentiostat based on the Arduino. The Rodeostat comes with an established Python library that was used to interface with the Raspberry Pi. The parameters for chronoamperometry were determined using cyclic voltammetry (300 samples/s, scan rate 500 mV/s, 2-electrode scheme) by finding the lowest anodic voltage at which H2O2 oxidized [Fig. 2(b)]. Microelectrodes were used to ensure that a hemispherical diffusion profile with a steady state current is rapidly established. The faradaic current from H2O2 oxidation was measured after discharge of a capacitance associated with a double layer [Fig. 2(c)]. The optimal chronoamperometric waveform for detection of hydrogen peroxide was to hold the electrode at −0.3 V for 0.5 s and for 2 s at +0.7 V pulses [Fig. 2(d)]. The current was detected by acquiring 10 samples over 1 s at the end of anodic pulses and was converted to H2O2 concentration using a calibration curve established for each individual Rodeostat and microelectrode pair. Each Rodeostat was connected to the Raspberry Pi using USB serial communication and controlled via the main Python script. To account for the noise generated in the signal by the magnetic stir bar, a running average with 10 data points of the detected current was used to calculate H2O2 concentration. The peristaltic pump on/off state was determined using a simple thresholding technique by the Raspberry Pi. When toggled on, the pump would deliver concentrated H2O2 to the module while simultaneously removing the excess solution.

FIG. 2.

Electrochemical sensing for a feedback loop to maintain H2O2 concentration in the automated RAA. Detection of H2O2 is performed with 25-micrometer platinum disk microelectrodes [(a) electron micrograph of the electrode with a higher magnification inset] using the Rodeostat potentiostat. Optimum conditions for electrochemical detection were determined using cyclic voltammetry (b). The measurement of H2O2 concentration is performed using chronoamperometry with 2 s anodic pulses of +0.7 V being applied to the microelectrode that is held at −0.3 V between the pulses (c). The current reached steady state very quickly which allowed for sampling every 3 s (d). The inset in (b) with color legend for 0 (navy blue), 5 mM (green), 10 mM (red), 15 mM (blue), 20 mM (purple), 25 mM (gold), and 30 mM (black) of hydrogen peroxide corresponds to all plots with current traces.

FIG. 2.

Electrochemical sensing for a feedback loop to maintain H2O2 concentration in the automated RAA. Detection of H2O2 is performed with 25-micrometer platinum disk microelectrodes [(a) electron micrograph of the electrode with a higher magnification inset] using the Rodeostat potentiostat. Optimum conditions for electrochemical detection were determined using cyclic voltammetry (b). The measurement of H2O2 concentration is performed using chronoamperometry with 2 s anodic pulses of +0.7 V being applied to the microelectrode that is held at −0.3 V between the pulses (c). The current reached steady state very quickly which allowed for sampling every 3 s (d). The inset in (b) with color legend for 0 (navy blue), 5 mM (green), 10 mM (red), 15 mM (blue), 20 mM (purple), 25 mM (gold), and 30 mM (black) of hydrogen peroxide corresponds to all plots with current traces.

Close modal

In the three channel pump head, one channel was for the delivery of H2O2, while two tubes were for the waste removal. The waste removal tubing was positioned in the flask such that the tip was at the same level as the surface of the solution at the target volume. This arrangement allowed the removal of liquid above the level of the exhaust tube, at a rate double that of the H2O2 delivery, thus ensuring that the solution level remained constant.

Platinum microelectrodes were fabricated in-house using a 1.5/0.84 mm OD/ID glass capillary (1B150F-4, World Precision Instruments, Sarasota, FL), 25 μm platinum (Pt) wire (Alfa Aesar, Tewksbury, MA), hook up copper (Cu) wire, and Sn63/Pb37 0.63 mm solder wire. First, one end of the glass capillary was sealed with a butane torch, making the seal as small as possible. A 1.5 cm long piece of Pt wire was degreased in acetone for 1 min and then inserted in the open end of the capillary. The sealed end of the capillary was pointed down and gently tapped on a hard surface until the Pt wire sled to the sealed bottom of the capillary. With the Pt wire at the sealed tip, the open end of the capillary was wrapped with 1-2 layers of parafilm and attached to a rubber tube of small diameter to create a good seal and then hooked to a vacuum pump. The capillary was fixed on a ring stand with the sealed end pointing downward at a 45° angle to the ground. The vacuum was turned on, and the sealed end was carefully melted using a butane torch. The glass was fused continuously for at least 2 mm from the tip of the Pt wire so as to create a good seal around the Pt wire. After the Pt wire was sealed, the working end of the capillary was ground with 600 grit sandpaper until the Pt wire was exposed. The microelectrode was then polished with a polishing kit first with 0.3 μm followed by 0.05 μm Al2O3 slurry (eDAQ, Australia). A 1 cm piece of solder was inserted in the open end of the capillary. The Cu wire was used to push the solder down the capillary until it reached the Pt wire. The capillary was slowly heated from the outside at the position of the solder using the butane torch. Once a good connection is formed between the Pt wire and Cu wire, the Cu wire was fixed to the capillary with heat shrink so as to remove mechanical stress from the soldered connection. The quality of a glass seal for the Pt disk microelectrode was verified with electron microscopy (Mira3, Tescan USA, Inc., Warrendale, PA) and via cyclic voltammetry of potassium ferrocyanide.25 

Commercial neural implants were subjected to RAA conditions to compare the performance of the aRAA system to its earlier version12 using the same characterization protocol with electron microscopy and electrochemical impedance spectroscopy. In the first aRAA vessel, the Blackrock Microsystems implant with 16 individual microfabricated electrodes coated with parylene-C in 4 × 4 configuration (Blackrock Microsystems, Salt Lake City, UT) has been exposed to 15 mM H2O2 at 87 °C. In the second aRAA module, the TDT implant (TDT, Alachua, FL) with an array of 16 polyimide insulated gold-coated tungsten microelectrodes was exposed to 15 mM H2O2 at 67 °C. Experiments were run in parallel for 7 days with the temperature and H2O2 concentration being logged with the Raspberry Pi for both modules (Fig. 3).

FIG. 3.

Performance of the automated RAA system with two independently controlled modules. (a) Temperature and (b) H2O2 concentration for two RAA modules operating simultaneously at two different conditions for seven days. Each module had a H2O2 concentration feedback controller and a PID temperature controller with Module 1 set at 67 °C and Module 2 set at 87 °C. Concentration and temperature data were logged with the Raspberry Pi, which was connected to each feedback controller. Noise in the H2O2 readings was largely due to the magnetic stirrer.

FIG. 3.

Performance of the automated RAA system with two independently controlled modules. (a) Temperature and (b) H2O2 concentration for two RAA modules operating simultaneously at two different conditions for seven days. Each module had a H2O2 concentration feedback controller and a PID temperature controller with Module 1 set at 67 °C and Module 2 set at 87 °C. Concentration and temperature data were logged with the Raspberry Pi, which was connected to each feedback controller. Noise in the H2O2 readings was largely due to the magnetic stirrer.

Close modal

The new aRAA system has redesigned heating, automated control of H2O2 concentration, and automated logging of the data. The original RAA design used oil heating with a 1000 ml jacketed flask to enable accelerated aging at high temperatures. This design was prone to failure due to rapid degradation of oil lines and eventual ruptures with oil spillage. The use of electric mantles for heating enabled miniaturization of the aRAA setup with a ten-fold decrease in the volume of a reaction flask without a change in system performance. The original RAA system did not have the ability to log the temperature, but based on day-to-day observations, it was oscillating within several degrees Celsius. In the new aRAA design, the temperature was 87 ± 1.5 °C and 67 ± 4 °C for the first and second RAA modules, respectively [Fig. 3(a)].

The main breakthrough of the new aRAA design that makes it simple and easy to use is a feedback loop for automated control of H2O2 concentration. Chronoamperometric pulsing protocol with 0.5 s cathodic hold at −0.3 V, followed by 2 s anodic pulse at +0.7 V, demonstrated stable reading of H2O2 over a period of time sufficient to perform the entire RAA experiment [Fig. 3(b)]. The system has not been optimized to remove or minimize electronic noise, which appears as sharp spikes in H2O2 concentration. Most of this noise has been traced to electrical interference from a magnetic stirrer. Since experiments are run on a long time scale (days and weeks) and H2O2 degradation happens relatively slow (half-life ∼ 20 min) compared to the sampling rate (1 sample every 3 s), this interference can be addressed algorithmically in a Python code ran on the Raspberry Pi that differentiates between electronic glitches and real changes in H2O2 concentration. This approach allowed us to use the readily available Rodeostat without the need to optimize detection conditions, making the setup straightforward and easy to reproduce. The initial RAA system had manually maintained H2O2 concentration with 10 mM variation over 7 days. The new aRAA design has a peak-to-peak variation in H2O2 concentration within 3 mM [Fig. 3(b)].

The simplicity of the aRAA system and precise control of the temperature and H2O2 concentration enables easy scaling of RAA experiments, with the ability to adjust the intensity of aggressive conditions to simulate different degrees of device degradation. To take advantage of this new feature, two types of neural implants were exposed to two different sets of conditions. Blackrock microelectrode arrays were exposed to conditions reported earlier (15 mM H2O2 at 87 °C),12 while TDT microelectrode arrays were exposed to milder conditions (15 mM H2O2 at 67 °C). The results obtained for RAA of Blackrock microelectrode arrays closely resemble what has been reported earlier including partial cracking and etching of parylene-C insulation [Figs. 4(a)–4(d)] as well as a drop in electrode impedance over the wide range of frequencies [Figs. 4(e) and 4(f)], likely caused by penetration of moisture through the insulation.

FIG. 4.

Degradation of the Blackrock microelectrode array in the automated RAA system at 87 °C. Electron microscopy indicates delamination and etching of parylene-C insulation [(a) vs (b) and (c) vs (d)]. Impedance spectroscopy indicates a drop in impedance (e) and shift from capacitive conduction mode into more resistive conduction mode for lower frequencies (f).

FIG. 4.

Degradation of the Blackrock microelectrode array in the automated RAA system at 87 °C. Electron microscopy indicates delamination and etching of parylene-C insulation [(a) vs (b) and (c) vs (d)]. Impedance spectroscopy indicates a drop in impedance (e) and shift from capacitive conduction mode into more resistive conduction mode for lower frequencies (f).

Close modal

Examination of TDT MEAs after RAA at the milder conditions revealed a less extreme degradation pattern than what has been observed earlier12 for a higher temperature (87 °C). TDT microelectrodes have a tungsten core that is covered with a thin gold layer and insulated with polyimide. Exposure of TDT arrays to RAA at 87 °C led to complete dissolution of a tungsten core, almost complete loss of polyimide insulation, and bending and breaking of thin-walled gold tubes. However, after RAA at 67 °C, none of the microelectrodes were broken [Figs. 5(a) and 5(b)], while cracks and loss of polyimide insulation were observed for some of the electrodes [Figs. 5(c)–5(f)]. The impedance data are consistent with earlier RAA experiments when little change in impedance was observed after RAA [Figs. 5(g) and 5(h)]. This suggests that RAA of TDT arrays at 67 °C is more similar to the degradation reported of TDT arrays after in vivo implantation,17,18 where damage to polyimide insulation was also less severe than what has been observed after RAA at 87 °C.12 The ability to tune the strength of reactive accelerated aging with better control over experimental parameters (temperature, H2O2 concentration, and time) combined with the ability to replicate the aRAA setup and run multiple experiments in parallel will enable the search for optimal conditions that more accurately match the degradation of the implants observed in vivo. Performing comparative study by accessing device degradation at different temperatures is also crucial to establish device failure mechanisms since using only high-temperature data (87 °C) can inaccurately represent device failure for certain materials. For example, acceleration of polyimide degradation at this temperature is not representative of degradation that occurs at body temperature.20 

FIG. 5.

Degradation of TDT microelectrodes array in the automated RAA system at 67 °C. Electron microscopy indicates delamination of polyimide insulation and dissolution of tungsten electrode cores [(a) vs (b), (c) vs (d), and (e) vs (f)], while impedance spectroscopy does not indicate significant change in impedance magnitude (g) or phase (h).

FIG. 5.

Degradation of TDT microelectrodes array in the automated RAA system at 67 °C. Electron microscopy indicates delamination of polyimide insulation and dissolution of tungsten electrode cores [(a) vs (b), (c) vs (d), and (e) vs (f)], while impedance spectroscopy does not indicate significant change in impedance magnitude (g) or phase (h).

Close modal

Automated reactive accelerated aging (aRAA) is a simple setup that can be used for rapid screening of new neural implants to establish potential failure modes without the need for lengthy and expensive animal experiments. The new version of the RAA system is fully automated and implements a feedback loop to maintain H2O2 concentration using a simple electrochemical detection protocol. The aRAA design employs readily available components and open-source solutions and can easily be replicated on a modest budget. The simplicity of the aRAA system and precision in control over the temperature and H2O2 concentration enables easy scaling of RAA experiments and tuning of the intensity of aggressive conditions to simulate different degrees of device degradation.

This work was sponsored by the Defense Advanced Research Projects Agency (DARPA) BTO under the auspices of Dr. Doug Weber through the DARPA Contracts Management Office. Grant/Contract: Inter Agency Agreement with U.S. Food and Drug Administration. We are grateful to Dr. Yong Wu and Dr. Jiwen Zheng of Advanced Characterization Facility (US FDA) for technical assistance with electron microscopy. We thank Dr. Katherine Vorvolakos (US FDA) for review and valuable input to the manuscript.

The views, opinions, and/or findings contained in this article are those of the authors and should not be interpreted as representing the official views or policies of the Department of Defense or the U.S. government. The mention of commercial products, their sources, or their use in connection with material reported herein is not to be construed as either actual or implied endorsement of such products by the Department of Health and Human Services.

1.
D.
Kumsa
,
G.
Steinke
,
G. F.
Molnar
,
E. M.
Hudak
,
F. W.
Montague
,
S. C.
Kelley
,
D. F.
Untereker
,
A.
Shi
,
B. P.
Hahn
,
C.
Condit
,
H.
Lee
,
D.
Bardot
,
J.
Centeno
,
V.
Krauthamer
, and
P. A.
Takmakov
,
Neuromodulation: Technol. Neural Interface
21
,
117
(
2017
).
2.
K.
Birmingham
,
V.
Gradinaru
,
P.
Anikeeva
,
W. M.
Grill
,
V.
Pikov
,
B.
McLaughlin
,
P.
Pasricha
,
D.
Weber
,
K.
Ludwig
, and
K.
Famm
,
Nat. Rev. Drug Discovery
13
,
399
(
2014
).
3.
R. A.
Miranda
,
W. D.
Casebeer
,
A. M.
Hein
,
J. W.
Judy
,
E. P.
Krotkov
,
T. L.
Laabs
,
J. E.
Manzo
,
K. G.
Pankratz
,
G. A.
Pratt
,
J. C.
Sanchez
,
D. J.
Weber
,
T. L.
Wheeler
, and
G. S. F.
Ling
,
J. Neurosci. Methods
244
,
52
(
2015
).
4.
M.
Gay
,
The Brain Electric: The Dramatic High-Tech Race to Merge Minds and Machines
(
Farrar, Straus and Giroux
,
2015
).
5.
K.
Bowsher
,
E. F.
Civillico
,
J.
Coburn
,
J.
Collinger
,
J. L.
Contreras-Vidal
,
T.
Denison
,
J.
Donoghue
,
J.
French
,
N.
Getzoff
,
L. R.
Hochberg
,
M.
Hoffmann
,
J.
Judy
,
N.
Kleitman
,
G.
Knaack
,
V.
Krauthamer
,
K.
Ludwig
,
M.
Moynahan
,
J. J.
Pancrazio
,
P. H.
Peckham
,
C.
Pena
,
V.
Pinto
,
T.
Ryan
,
D.
Saha
,
H.
Scharen
,
S.
Shermer
,
K.
Skodacek
,
P.
Takmakov
,
D.
Tyler
,
S.
Vasudevan
,
K.
Wachrathit
,
D.
Weber
,
C. G.
Welle
, and
M.
Ye
,
J. Neural Eng.
13
,
023001
(
2016
).
6.
E.
Waltz
,
Nat. Biotechnol.
34
,
904
(
2016
).
7.
I. H.
Stevenson
and
K. P.
Kording
,
Nat. Neurosci.
14
,
139
(
2011
).
8.
P. A.
Takmakov
,
Electrochem. Soc. Interface
26
,
49
(
2017
).
9.
S. M.
Wellman
,
J. R.
Eles
,
K. A.
Ludwig
,
J. P.
Seymour
,
N. J.
Michelson
,
W. E.
McFadden
,
A. L.
Vazquez
, and
T. D. Y.
Kozai
,
Adv. Funct. Mater.
28
,
1701269
(
2017
).
10.
J. T.
Hachmann
,
J. H.
Jeong
,
P. J.
Grahn
,
G. W.
Mallory
,
L. Q.
Evertz
,
A. J.
Bieber
,
D. A.
Lobel
,
K. E.
Bennet
,
K. H.
Lee
, and
J. L.
Lujan
,
PLoS One
8
,
e81443
(
2013
).
11.
S.
Vasudevan
,
K.
Patel
, and
C.
Welle
,
J. Neural Eng.
14
,
016008
(
2017
).
12.
P.
Takmakov
,
K.
Ruda
,
K. S.
Phillips
,
I. S.
Isayeva
,
V.
Krauthamer
, and
C. G.
Welle
,
J. Neural Eng.
12
,
026003
(
2015
).
13.
E.
Patrick
,
M. E.
Orazem
,
J. C.
Sanchez
, and
T.
Nishida
,
J. Neurosci. Methods
198
,
158
(
2011
).
14.
K. A.
Potter-Baker
and
J. R.
Capadona
,
ACS Macro Lett.
4
,
275
(
2015
).
15.
D. W. L.
Hukins
,
A.
Mahomed
, and
S. N.
Kukureka
,
Med. Eng. Phys.
30
,
1270
(
2008
).
16.
J. C.
Barrese
,
N.
Rao
,
K.
Paroo
,
C.
Triebwasser
,
C.
Vargas-Irwin
,
L.
Franquemont
, and
J. P.
Donoghue
,
J. Neural Eng.
10
,
066014
(
2013
).
17.
A.
Prasad
and
J. C.
Sanchez
,
J. Neural Eng.
9
,
026028
(
2012
).
18.
A.
Prasad
,
Q.-S.
Xue
,
R.
Dieme
,
V.
Sankar
,
R. C.
Mayrand
,
T.
Nishida
,
W. J.
Streit
, and
J. C.
Sanchez
,
Front. Neuroeng.
7
,
1
(
2014
).
19.
J. T.
O’Malley
,
B. J.
Burgess
,
D.
Galler
, and
J. B.
Nadol
,
Otol. Neurotol.
38
,
970
(
2017
).
20.
B.
Rubehn
and
T.
Stieglitz
,
Biomaterials
31
,
3449
(
2010
).
21.
R.
Caldwell
,
R.
Sharma
,
P.
Takmakov
,
M. G.
Street
,
F.
Solzbacher
,
P.
Tathireddy
, and
L.
Rieth
,
J. Neurosci. Methods
293
,
210
(
2018
).
22.
J.
Pfau
,
T.
Stieglitz
, and
J.
Ordonez
, in
2017 8th International IEEE EMBS Neural Engineering Conference NER
(
IEEE
,
2017
), pp.
166
169
.
24.
D. J.
Caruana
and
J. V.
Bannister
,
J. Electroanal. Chem.
424
,
197
(
1997
).
25.
U. K.
Sur
,
A.
Dhason
, and
V.
Lakshminarayanan
,
J. Chem. Educ.
89
,
168
(
2012
).