As illustrated in Figure 1, the proposed stimulator consists of an array of 32 stimulation channels distributed in two symmetric panels, with 4 × 4 channels on each panel. Each of these stimulation channels has a μ-LED coupled with a reflective silicon cavity (Khan and Li, 2017). The μ-LED (CREE TR2227tm) has a surface area of 270 × 220 μm and a thickness of 50 μm, and the central illumination wavelength of 465 nm for excitation of optogenetic opsins. The diameter of each hemi-spherical reflective cavity is 300 μm and the distance between adjacent cavities is ~430 μm. The overall dimension of each panel is 3 × 3 mm, which is intended to cover one hemisphere of the primary visual cortex (V1) of rats. The two-panel design exclusively allows stimulation of both cortices individually. To reduce the number of interconnects to control circuitry, LEDs on each panel share a single ground wire while having separate power wires to enable individual control of one or more LEDs at a time.

Figure 2 illustrates a simplified process flow for device fabrication, which includes the following major steps.

To establish proof-of-principle for application with an optogenetic system, we assessed the efficacy of the μ-LEDs to activate an optically-induced gene expression system (with and without reflectors). Successful induction is confirmed by the visual identification of the expression of a red fluorescent reporter following genetic modification and exposure to blue light (Figure 6A). For these in vitro experiments, human embryonic kidney cells (HEK 293) were used. HEK 293 were maintained under Dulbecco's modified Eagle's medium (DMEM; Life Technologies) supplemented with 10% fetal bovine serum (FBS; Sigma) and 1% penicillin/streptomycin (Life Technologies). Colonies were passaged every 3–4 days by chemical dissociation utilizing TrypLE (Life Technologies). Cells were kept in an incubator at 37°C with 5% CO₂. Prior to light induction experiments, cells were passaged onto 24-well plates at 68,000 cells/well. Cells were then transiently transfected utilizing Lipofectamine 3000 (Invitrogen) by following the manufacturer's protocol. Each well was transfected with 200 ng of DNA including 40 ng of activator plasmid [“EL222,” a blue light-activated gene expression system; (Motta-Mena et al., 2014; Winter et al., 2017)], 20 ng of reporter gene (red fluorescence protein, RFP) and 140 ng of PUC (an empty plasmid). Twelve (12) to sixteen (16) hours after DNA transfection, cells were exposed to blue light (465 nm) from LEDs for either 240 or 480 min. Two wells were exposed to standard (bare) LEDs, two wells were exposed to reflector-coupled LEDs, and four were kept under dark conditions as a control.

After light stimulation, cells were maintained in the incubator. After 72 h, each well was imaged using a fluorescent microscope (Leica MZ10 F) to assess optical induction of the RFP reporter construct. All plasmids used in this study were provided by BiomiLab, LLC (Winter et al., 2017). Individual images received manual threshold utilizing Ocular software (QImaging) to eliminate background fluorescence. Images from all wells were further analyzed with an in-house MATLAB script to calculate the number of “zero” pixels (where a fluorescent signal was not detected) and non-zero pixels (where fluorescent signal was detected) to obtain the percentage of pixels above threshold (number of non-zero pixels divided by total number of pixels, % PAT, multiplied by a factor of 10,000 to facilitate comparison, Table 1) for each image. Lastly, an additional MATLAB script (Kozai et al., 2014; Salatino et al., 2017) was used to assess the spatial gradient of gene expression in a preliminary test by creating bins of intensity in area increments (two wells were exposed to light for 480 min: one with the bare LED stimulator and the other with the reflector coupled stimulator). To assess images spatially, bins were utilized to quantify mean intensity as a function of distance from the center of the LED location. Twelve 200 micron-width bins were analyzed which projected radially from the center of the LED site. Location of the center was manually selected to accurately represent the center of the region that received direct light exposure. The LEDs were manually positioned at the center on each well, based on visual inspection. The manual procedure for LED placement, as well as the flexibility of the LEDs, likely contributed to a slightly off-center placement relative to the center of the well (Figure 6A).

Figure 4a provides SEM images showing the morphology of the silicon cavity after wet etching. The cavity formation is the most important step for the reflector fabrication, and the surface morphology of the cavities need to be comparatively smooth to ensure efficient light reflection with minimum scattering. AFM analysis in Figure 4b shows a small mean roughness of ~72 nm, confirming the smooth surface of the etched cavity. The horizontal and vertical etching rates of silicon cavities were calculated based on the profilometry data, as shown in Figure 4c. The results indicate that the etching rates in the horizontal and the vertical direction are not the same, with the opening of the nitride mask being a limiting factor. Due to the lack of agitation, the vertical etching rate is slower than the horizontal rate and less dependent on the size of nitride opening (Figure 4d). This phenomenon forced the cavities not to be perfectly hemi-spherical, thereby creating a deviation between the theoretical concept and the realized cavity geometry. Consequently, a deviation should also be expected between the simulation results (which considers the cavity to be perfectly hemi-spherical) and the experimental results. Figure 4e illustrates a fabricated 32-channel stimulator array on a silicon chip, with illuminated LEDs on both panels. There is a slight misalignment between the LED bonding pads and cavity, due to the inaccuracy of the photolithography process. Deposition of the Parylene C layer allows the release of the array from the silicon substrate and transform it to a fully flexible free-standing stimulator array to adapt with corrugated brain surfaces by conformal contact. Figure 4f shows a released array from the substrate wafer, where the bending is due to the mechanical stress of the thin Parylene C layer after release, and could be reduced by stabilizing and flattening the array on a separate substrate using water as adhesive.

Figure 5A shows the optical intensity enhancement due to the coupled reflector, which was measured experimentally with respect to a bare (no-reflector-coupled) optical stimulator. The reflector-coupled devices achieved the minimal and maximal enhancement of 49 and 65%, respectively, verifying the effectiveness of the reflector to collect the rear surface emission of the LED. For these benchtop experiments, the LEDs were coupled with 300-μm-diameter reflectors that ensure the optimized intensity enhancement as suggested by the simulation above. The experimental data was slightly lower than simulated values, which could be attributed to several factors: fabrication imperfections, oval-shaped cavity profile, inaccuracy of manual alignment between the LED and reflector layers, and high contact impedance of the LED-solder-contact interface. Regarding in-vivo experiments, a common excitatory opsin is channelrhodopsin (ChR2) that can be activated by blue light with threshold of 1 mW/mm² (Aravanis et al., 2007). Preliminary benchtop measurements indicate that our intensity enhanced stimulators surpass this threshold even at a low applied current of <2.5 mA. It is of note that the detector and LED surface had a separation of 7 mm due to the use of a probe station. The measured intensity might be further improved if the separation could be reduced by improving the measurement setup.

Although the brain is only 2% of the body's mass, 20% of the cardiac output is delivered toward it, indicating a high metabolic demand in the brain (Wolf, 2008). A high resting blood flow clears the heat generated by the brain. Adverse effects of increased temperature due to implants, on top of this high metabolism activity have been observed in the central nervous system (CNS) where breakdown of blood-brain barrier was observed after 60 min at 42°C along with some neuronal death (6%) after 60 min of heating at 40.5°C (Goldstein et al., 2003). A maximum tolerable heat dose of 42.0–42.5°C for 40–60 min or 43°C for 10–30 min (Haveman et al., 2005) was established after a review study on regional exposure of the animal brain and spinal cord. The American Association of Medical Instrumentations (ANSI/AAMI) has a standard limit of 2°C increase for chronic biomedical implants for neuro-stimulators (ISO-14708-1). A change in temperature was calculated by subtracting the initial temperature (~22°C when the LED was off) from the operating temperature of the LED measured by the infrared camera. A temperature increase of <1°C was observed when our stimulators were operated under continuous current inputs (Figure 5B), even at higher light intensities (>8 mW/mm²). It could be noted that when compared with bare LEDs, the reflector-coupled stimulators dissipate less heat at >8 mW/mm² intensities, potentially due to the decreased thermal conductivities of silicon and PDMS at higher temperatures (Silicon Wafer — Thermophysical Properties, n.d).

In vitro tests confirmed proof-of-principle for the successful induction of red fluorescent reporter expression with μ-LEDs, as well as a more robust and spatially localized area of induction with the use of reflectors (Figure 6; Table 1). Fluorescent microscope images of each well were analyzed for either %PAT or intensity (see Methods) at each of the various duration settings chosen for light exposure. Qualitatively, light-induced gene expression was visually observable following 240 min of exposure (Figure 6A). Table 1 shows %PAT values obtained from our first in-house MATLAB script and represents average values of %PAT for two images per condition (n = 2, where the entire image area is quantified to deliver a single output value) for a blue-light exposure duration of 240 min and %PAT values for one image per condition (n = 1) for a blue-light exposure duration of 480 min. As expected, light-stimulated wells registered higher %PAT values than non-exposed wells, verifying successful optical induction of the reporter gene (Table 1). Due to the residual dark-state binding of the activator, some fluorescence in dark conditions was observed and served as a baseline for quantification. At 480 min, higher %PAT values were measured for the cells exposed to both the reflector coupled LEDs and for the bare LEDs when compared to the 240 min exposure duration (Table 1). Additionally, reflector-coupled LEDs registered slightly elevated %PAT values for both timepoints (Table 1; values multiplied by 10,000 to facilitate detection of differences between values).

Utilizing our second MATLAB script, preliminary observations suggest a more spatially localized pattern of reporter expression at 480 min (white circle in Figure 6A) potentially related to focused illumination enabled by the reflectors. To analyze the expression pattern as a function of distance from the center of induction, mean intensity per bin was calculated and normalized by using the corners of each image. This confirmed elevated expression in comparison to standard LEDs within the highlighted region of interest (Figure 6B). Future work will need to confirm this result with an expanded data set.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.