Photovoltaic Stimulation of Mouse and Pig Retina With Pyrolytic Carbon Microelectrodes Integrated on High‐Density Silicon Solar Cell Arrays

ABSTRACT

In this work, we introduce carbon as a novel electrode material in retinal prosthesis. We fabricate Si‐based photovoltaic retinal implants with single p‐n junctions in 40 µm and 200 µm pixels and successfully integrate pyrolytic carbon electrodes. Single pixels provide an open‐circuit voltage of up to 0.5 V under near‐infrared (NIR) illumination, demonstrating the functionality of the miniaturized solar cells despite the high thermal budget of the pyrolysis process. In electrophysiological ex vivo experiments with mouse retinal tissue, the intra‐ and extracellular spike activity significantly increased upon NIR‐induced photovoltaic stimulation with the retinal implant compared to spontaneous activity, while spike amplitudes were comparable to the natural response to white light. The absence of action potentials after subsequent blocking of Na channels with tetrodotoxin confirmed the biological nature of the signals recorded during stimulation. Furthermore, spike amplitudes of evoked potentials were slightly higher for carbon electrodes compared to Au counterparts. Photovoltaic retinal implants with 3D carbon pillar electrodes were explored for ex vivo stimulation of porcine retinal tissue, confirming their ability to evoke action potentials in retinal neurons. The results demonstrate the potential of miniaturized photovoltaic pixels with 3D pyrolytic carbon electrodes.

In this study, we fabricate photovoltaic retinal implants with pyrolytic carbon electrodes and test their ability to stimulate mouse and pig retinas when illuminated with near‐infrared light. We find increased spiking activity in both animal models compared to spontaneous activity and TTX controls and analyze the spike amplitude, latency and frequency.

1. Introduction

Retinal degenerative diseases such as retinitis pigmentosa (RP) and age‐related macular degeneration (AMD) are primary causes for visual impairment affecting millions of individuals worldwide [1, 2]. These conditions result from damaged photoreceptors cells, which are the primary signal transducers in the retina. Photopigments within the photoreceptors absorb light, initiating the phototransduction cascade. This process converts the light into electrical signals, which are then transmitted to postsynaptic cells, specifically the bipolar cells, via synapses [3]. The signal continues to propagate through the retinal circuitry to the retinal ganglion cells (RGC), the output neurons whose axons form the optic nerve [4]. The action potentials generated by the RGC travel to the brain, where visual information is processed. Consequently, when photoreceptors are degenerated, no visual information can be transmitted, leading to impaired vision and ultimately blindness [5, 6, 7].

A promising approach for vision restoration in patients suffering from RP and AMD is to use the remaining neural network, which preserves functionality with minor remodeling during degeneration [8]. Therefore, replacing the degenerated photoreceptor layers with a retinal implant that can mimic the functionalities by electrical stimulation of the bipolar cells may partly retain the natural process of signal transduction in the retina, restoring the visual information to the brain [9, 10]. Recently, the PRIMA implant, developed by the Palanker group at Stanford University, showed vision restoration and improvement in visual acuity after 12 months in a clinical study involving 38 patients [11].

Photovoltaic retinal implants typically consist of arrays of pixels, with each pixel having a solar cell with an interfacial electrode. State‐of‐the‐art implants typically use sputtered iridium oxide (SIROF) [12], PtIr [13], TiN [14], or Au‐SIROF [15] as stimulation electrodes. While such materials have shown promising stimulation behavior, their metallic nature imposes a potential risk of metal ion migration and thus biocompatibility issues [16, 17, 18]. Furthermore, the fabrication of such electrodes is limited to conventional microfabrication techniques such as sputtering and e‐beam evaporation with certain limitations on layer thickness. Therefore, there is a need to investigate non‐metallic electrode materials for retinal prosthesis. Carbon electrodes have been successfully introduced for brain probes [19, 20] or heart electro‐stimulation, demonstrating high biocompatibility and stability [21, 22]. However, for retinal prosthesis, carbon has not been investigated so far, mainly due to the lack of suitable strategies for integration of miniaturized electrodes on single pixels.

The stimulation of retinal neurons by the photovoltaic implant when located in the subretinal space targets the bipolar cells [23], and when placed epiretinally, targets the RGC [24]. Stimulation happens when illuminating the implant with near‐infrared (NIR) light, which is then absorbed in the photovoltaic implant, generating photocurrent [10, 25, 26]. Due to the distance between each stimulating electrode and the closest neuron, the electric field penetration depth is limited, which leads to a need for increased pixel size, that is, larger photocurrent per pixel. However, larger pixels result in lower resolution, which means that there is a need to increase the delivered current without increasing pixel size. A promising approach towards this balance between resolution and pixel current involves 3D interfacial electrodes [27]. Such a design enlarges the surface area available for neurons to contact, therefore enhancing stimulation efficiency. This approach further decreases the distance to the target neurons, and this may improve the electric field coverage [15, 24, 28, 29, 30]. Pyrolytic carbon (PyrC) electrodes obtained via conversion of photoresist materials at high temperature in inert atmosphere have been applied for electrochemical monitoring of neurons [31, 32, 33]. PyrC is known for its biocompatibility [34, 35, 36, 37], wide electrochemical potential window, and chemical inertness [38, 39]. We have previously characterized the capacitive behavior of PyrC electrodes with a measured charge storage capacity (CSC) of 0.21 ± 0.05 mC/cm², aligning closely with that of glassy carbon (GC) electrodes at 0.25 mC/cm² [40]. The relatively low CSC is attributed to the fact that PyrC obtained from photoresist precursors has a very low surface roughness and porosity [41]. This is on par with established electrode materials like platinum, but significantly lower than, for example, SIROF [42] and nanostructured Pt [43]. Nevertheless, we successfully demonstrated electrical stimulation of retinal neural tissue with 3D carbon electrodes [44]. Furthermore, several other studies explored the fabrication of 3D PyrC electrodes, including selective sidewall passivation [45, 46], and demonstrated enhanced interaction with cells and tissue [47].

Here, we for the first time integrated carbon stimulating electrodes on 40 and 200 µm wide pixels of high‐density photovoltaic retinal implants based on Si solar cells. The proposed electrode microfabrication employs a combination of UV photolithography and subsequent pyrolysis to convert the defined polymer microstructures into PyrC. We introduce PyrC as a novel electrode material on photovoltaic retinal implants, potentially providing excellent biocompatibility, no toxicity, possibility of 3D electrode integration with excellent dimensional control, and large‐scale production at relatively low costs. Photovoltaic pixel characterization demonstrated that integration of PyrC on Si solar cells is possible without detrimental effect on the photovoltaic performance despite the long high temperature process. Photovoltaic implants with 40 µm pixels and PyrC electrodes were used for NIR‐induced ex vivo stimulation of mouse retinas. Si solar cell arrays with 200 µm pixel diameter and 3D PyrC pillar electrodes were employed for photovoltaic stimulation of pig retinas. Robust neural activity was observed in both species. However, the response characteristics differed between the two models. In mouse retinas, photovoltaic stimulation evoked physiological RGC responses with cell‐to‐cell variability in response amplitude and latency. In contrast, pig retinas required burst stimulation to reliably evoke neural activity and exhibited greater inter‐sample variability. Despite these species‐dependent differences, the electrophysiological studies with both animal models confirmed that we can achieve effective retinal neuron activation with carbon electrodes.

2. Results

2.1. Photovoltaic Retinal Implant Design and Fabrication

The Si solar cell arrays with dimensions of 2 mm × 2 mm are designed for subretinal prothesis with a total size of 3 mm × 3 mm. The high‐density photovoltaic implants have 1755 pixels with 40 µm diameter and 50 µm pitch resulting in a density of 462 pixels mm⁻². This pixel density is comparable to state‐of‐the‐art photovoltaic retinal prosthesis [48]. Additionally, arrays with 65 pixels with 200 µm pixel diameter and a 250 µm pitch, corresponding to a pixel density of 18 pixels mm⁻² are designed. Each pixel features n‐ and p‐type doped regions interfaced with a central circular stimulating electrode and a circumferential annular return electrode, respectively.

The fabrication of the Si solar cell arrays was done on a Czochralski Si n‐type wafer passivated with a thermally grown 600 nm silicon dioxide layer. First, the n‐type region is defined by oxide etching (Figure 1a), formation of a phosphorus‐doped oxide phase layer using POCl₃, annealing to activate diffusion, etching of the phase layer and a drive‐in process at 950°C for phosphorus diffusion into Si (Figure 1b). This resulted in a heavily doped n⁺ region with a sheet resistance of 2.6 ± 1.2 Ω/sq. The thermal oxide formed during the drive‐in process masks the n⁺ regions during subsequent processing. Similarly, the p‐type region is defined by patterning of the oxide (Figure 1c). Pre‐deposition of a boron‐doped oxide layer and a subsequent drive‐in process at 950°C after etching of the phase layer resulted in a p⁺ region with a sheet resistance of 4 ± 2.1 Ω/sq. During the drive‐in process, a 250 ± 100 nm thin oxide is formed, passivating the surface (Figure 1d). After completing solar cell array fabrication, the oxide in the doped regions was selectively etched to enable direct electrical contact with the electrodes (Figure 1e).

FIGURE 1.

Photovoltaic retinal implant design and fabrication. Cross‐sectional illustration of a single Si solar cell pixel during fabrication with (a) oxide etching to define n⁺ region; (b) doping of n⁺ region; (c) oxide etching to define p⁺ region; (d) doping of p⁺ region; (e) oxide opening for contacts. Integration of PyrC electrodes with (f, g) SU‐8 photolithography to pattern PyrC precursor structures; (i) pyrolysis. (j) Deposition of Au electrodes with photolithography and lift‐off. SEM images of Si solar cell arrays with PyrC electrodes and pixel diameter (k, l) 40 µm and (m, n) 200 µm; Au electrodes and pixel diameter (o, p) 40 µm and (q, r) 200 µm.

For integration of PyrC electrodes, precursor microstructures with a height of 5 ± 1 µm were defined with UV photolithography using the negative epoxy‐based photoresist SU‐8 (Figure 1f). On the larger pixels with a diameter of 200 µm, a second step of photolithography was conducted to increase the height of the central pillar to 62 ± 3 µm (Figure 1g). The polymer structures were converted into carbon by pyrolysis at 1050°C in an inert atmosphere (Figure 1h,i). During this thermal process, significant vertical shrinkage occurs due to material evaporation reducing the height of both electrodes on the 40 µm pixels to 1.5 ± 1 µm, while the central pillar height on the large pixels was reduced to 26 ± 5 µm. For comparison, retinal implants with 200 nm thick Au stimulating and return electrodes were fabricated using photolithography and lift‐off (Figure 1j). A 30 nm Cr layer was deposited on the contact regions as adhesion promoter between Au and Si.

Figure 1k–n shows the photovoltaic retinal implant with the Si solar cell array and the PyrC electrodes, while Figure 1o–r shows the devices with the Au electrodes. The diameter of the central PyrC electrode on the 40 µm pixels is 10 ± 1 µm (Figure 1k,l). This is on par with the smallest stimulating electrodes for retinal prosthesis reported elsewhere [49]. For the 200 µm pixels (Figure 1m,n), the diameter of the central PyrC electrode is 50 ± 2 µm at the bottom and 40 ± 1 µm at the top. It is clearly visible that the double‐layer process with SU‐8 (Figure 1g) resulted in a 3D pillar geometry on these arrays. The annular return electrodes have a width of 3 µm for the 40 µm pixels and 15 µm for the 200 µm pixels. For the small pixels with Au electrodes (Figure 1o,p), the diameter of the central electrode was 12 ± 2 µm, while it was 60 ± 3 µm for the large pixels (Figure 1q,r). This means that the electrodes cover approximately 32% of the total pixel area, which inevitably affects the output current and voltage of the photovoltaic cells. For the 40 µm pixels the n‐ and p‐type regions were only separated by 5 µm undoped silicon (Figure 1p). The SEM images reveal the quality of the alignment between different doped regions, the oxide openings and the integrated PyrC and Au electrodes, which is essential for the functionality of the device. The doped regions are passivated with 250 ± 100 nm SiO₂ layer, while the undoped region has an oxide thickness of 533 ± 20 nm, resulting in height variations as illustrated in Figure 1e and shown in all the SEM images.

2.2. Photovoltaic Pixel Characterization

PyrC has never been used as an electrode material in solar cells before. As described above, the integration of PyrC on Si solar cell arrays involves processing at high temperatures for several hours, which might be detrimental for the photovoltaic properties. Furthermore, it was unclear if electrical resistance at the carbon‐Si interface might affect electron transfer between the solar cells and the stimulating electrodes. To investigate these aspects, test chips with single photovoltaic pixels and Au contacts for electrical probing of PyrC stimulating and return electrodes were designed for both diameter 40 and 200 µm (Figure 2a,b). Compared to the pixels in the array, the area of the p⁺ region was reduced by approximately 20% and the oxide openings etched for the PyrC‐p⁺ contacts (Figure 2c,d) were decreased by 50% to enable placement of the Au contact on the central stimulating electrode. The fabrication of the test chips followed the identical processing as for the photovoltaic retinal implants with the PyrC electrodes (Figure 1a–i) except for an additional step of photolithography and lift‐off to define the 200 nm thick Au contacts on a 30 nm Cr adhesion layer. The 1.5 ± 0.5 µm tall PyrC structure presented a challenge in the final lithography step, which was mitigated by spin‐coating a 2.2 µm thick photoresist for the lift‐off process. Figure 2e–h shows the 40 and 200 µm pixels before and after Au deposition on the test chip revealing the quality of the alignment between etched oxide openings, PyrC electrodes and Au contacts. In all areas other than the PyrC electrodes, the Au contacts are formed on oxide and hence electrically isolated from the solar cell. For comparison, identical test chips were fabricated without the PyrC electrodes, where the Au directly contacted the doped regions (Figure S1). This configuration corresponds also to the photovoltaic retinal implants with Au electrodes discussed above (Figure 1o–r).

FIGURE 2.

Photovoltaic pixel characterization. Design of test chips with photovoltaic pixels with diameter (a) 40 µm and (b) 200 µm. For I‐V measurements, Au contacts are deposited on the PyrC electrodes fabricated on the doped regions. Optical microscope images of (c) 40 µm and (d) 200 µm pixels after etching of contact openings in SiO₂. SEM images (e, f) before and (g, h) after deposition of Au contacts on PyrC electrodes for both pixel sizes. Comparison of I‐V curves recorded under illumination with visible light in AM 1.5G spectral conditions at 100 mW cm⁻² for single photovoltaic pixels with (i) 40 µm and (j) 200 µm diameter with PyrC and Au electrodes. (k) experimental setup for NIR illumination. I‐V curves recorded under NIR illumination with 940 nm wavelength at different intensities for (l) 40 µm pixels and (m) 200 µm pixels with PyrC electrodes.

Initially, the photocurrent‐voltage (I‐V) characteristics of single pixels were measured in a solar simulator for conventional testing of solar cells with visible light in AM 1.5G spectral conditions at 100 mW cm⁻² (Figure S2) by sweeping the external voltage from ‐3 V to 3 V and recording the resulting current. In dark conditions, expected diode behavior was observed with minimal leakage current indicating good electrical isolation and a successful fabrication process (Figure 2i,j). The open circuit voltage V _oc for pixels with diameters of 40 µm was 0.46 ± 0.02 V and 0.42 ± 0.06 V with and without PyrC electrodes, respectively (Figure 2i). For the larger pixels with diameter of 200 µm with and without PyrC electrodes V _oc was 0.50 ± 0.04 V and 0.48 ± 0.05 V, respectively (Figure 2j). These measurements demonstrate that both Si solar cells with PyrC electrodes and with Au electrodes were functional with good quality of p‐n junction formation. Most importantly, the consistency of V _oc indicates no detrimental effect of the pyrolysis process on photovoltaic performance, a well optimized electrode‐semiconductor interface and negligible influence of the electrode material itself on the built‐in potential of the device.

The ability of a solar cell to convert the generated current and voltage into useful power is described by the fill factor (FF). Interestingly, the FF was significantly higher for photovoltaic pixels with a PyrC electrode compared to pixels with Au electrodes under visible light illumination. For 200 µm pixels, FF was 0.31 with PyrC and 0.19 with Au, while the difference was even larger for 40 µm pixels with FF 0.72 and 0.12 for PyrC and Au, respectively. This suggests that PyrC electrodes provide better charge collection and lower series resistance than Au, enhancing the overall efficiency of the solar cell. It should be noted that Au was chosen as reference here, mainly due to its use in some retinal implants [29] and its well‐known inert and stable electrical properties. As a conventional solar cell contact material, Au is not ideal, due to the formation of a Schottky barrier in contact with Si [50], also in the case of an intermediate Cr layer [51]. This might explain the lower performance of photovoltaic pixels with Au.

For photovoltaic stimulation in retinal prosthesis, typically NIR light is preferred over visible light due to less activation of remaining photoreceptors in the retina of patients with some residual vision. Therefore, photovoltaic pixel characterization was repeated using a custom‐made experimental setup for NIR illumination with light emitting diodes (LED) at wavelengths of 880 and 940 nm (Figure 2k). These wavelengths were selected, since 880 nm has typically been used for subretinal photovoltaic retinal implants [9, 10], while 940 nm is a realistic alternative wavelength with similar absorption properties in Si and was the one selected for the ex vivo studies with mouse and pig retinas later on. The 940 nm LED available in the setup for photovoltaic pixel characterization further was able to reach higher light intensities of up to 124 mW cm⁻². Figure 2l,m show the I‐V curves recorded under increasing NIR illumination with 940 nm wavelength for single photovoltaic pixels with diameters of 40 and 200 µm and PyrC electrode contacts. Table 1 summarizes the relevant photovoltaic properties extracted from the I‐V curves for single pixels with both PyrC and Au contacts and the two diameters. The general observations in the I‐V characteristics under NIR illumination with 940 nm for these pixels and the ones with Au contacts (Figure S3) were identical as for visible light (Figure 2i,j) confirming the functionality of the photovoltaic pixels. Diode behavior was observed in dark conditions. For increasing NIR illumination intensities the short circuit current I _sc increased linearly for devices with both PyrC and Au contacts as expected for a photovoltaic cell (Figure S4). At the highest intensity (124 mW cm⁻²) the I _sc for PyrC contacts was ∼ 42 times larger for the 200 µm pixels compared to the 40 µm pixels. This is partly explained by the 25x larger pixel area available for light absorption and generation of charge carriers, but also with other effects of downscaling such as unwanted dopant diffusion, higher risk of misalignment during fabrication and small leads with higher resistances. V _oc initially increased with illumination intensity and then stabilized independently of the electrode material at values of approximately 0.46 and 0.49 V for 40 and 200 µm pixels, respectively. In I‐V measurements with 880 nm wavelength similar trends were observed and V _oc values reached 0.42 V for the small pixels for both Au and PyrC, and 0.46 V for PyrC compared to 0.49 V for Au at maximum intensities of 17 mW cm⁻² (Figure S5 and Table S1).

TABLE 1.

Fill factor FF, short circuit current I _sc and open circuit voltage V _oc extracted from the recorded I‐V curves under 940 nm NIR illumination at different intensities for pixels with PyrC and Au electrodes.

Irradiance 940 nm (mW cm−2)	40 µm pixels						200 µm pixels
	FF		Is c (nA)		V oc (V)		FF		Is c (µA)		V oc (V)
	Au	PyrC	Au	PyrC	Au	PyrC	Au	PyrC	Au	PyrC	Au	PyrC
3	0.39	0.73	8	5	0.343	0.343	0.38	0.50	0.09	0.25	0.382	0.382
8	0.33	0.72	20	15	0.383	0.382	0.33	0.42	0.24	0.68	0.423	0.423
16	0.29	0.71	39	26	0.383	0.383	0.31	0.39	0.45	1.21	0.423	0.423
38	0.22	0.71	110	81	0.423	0.423	0.17	0.27	1.15	3.52	0.463	0.463
93	0.17	0.65	275	187	0.463	0.463	0.15	0.18	1.35	8.10	0.491	0.488
124	0.16	0.63	365	238	0.463	0.463	0.15	0.16	1.50	10.20	0.493	0.494

The FF indicates the electrical ideality of the photovoltaic cell with respect to resistance. At higher irradiance the FF decreases due to increased recombination and resistive losses [52, 53]. As shown in Table 1, the pixels with PyrC electrodes demonstrate superior FF performance and stability upon NIR illumination, with the FF decreasing less drastically with increasing irradiance, compared to similar pixels with Au electrodes. This suggests that pixels with a PyrC electrode mitigate the series resistance effect at higher light intensities better than similar pixels with an Au electrode. This might be attributed to the formation of Schottky barriers when Au/Cr is in contact with Si, which apparently is less of an issue for PyrC. As such, the main finding here is mainly the lack of Schottky barriers between PyrC and Si, rather than the expected behavior of Au contacts.

2.3. Photovoltaic Stimulation of RGCs in Mouse Retina

The performance of the photovoltaic implants with pixel diameters of 40 µm was evaluated with extracellular ex vivo experiments in mouse retina. More specifically, the ability of the solar cell arrays to activate RGC, which are the retinal output neurons, was characterized in electrophysiological measurements. The neural activity from RGC during NIR illumination was recorded using a multielectrode array (MEA). For this purpose, murine retina isolated from 8‐to‐24‐week‐old wild‐type mice of either sex was mounted on 64 channel probes with the photoreceptor side up (Figure 3a). The photovoltaic implant was placed with the p‐n junctions and the PyrC electrodes facing downwards onto the excised retinal tissue. For photovoltaic stimulation the stack of implant and retina was illuminated by 940 nm NIR from the top.

FIGURE 3.

Solar cell‐induced retinal activity. (a) Schematic of multi‐electrode array (MEA) recording neural response of retinas stimulated by the photovoltaic implant with 410 µm pixels and PyrC electrodes (10 µm each). (b, c) Firing activities during NIR light stimulation in a retina (b) without and (c) with implant, respectively. Each square indicates a signal measured with an individual recording electrode in the MEA (8×8 electrodes). Colors indicate signal intensity. (d) Example of recorded cell action potentials with (top) and without (bottom) implant. (e) Comparisons of the signal width (left) and amplitude (right) between with and without photovoltaic (PV) implant. (f) Heatmaps showing the maximum firing rate of the peristimulus time histograms (PSTHs) for each of the 64 recording channels. Following spike sorting, PSTHs were generated for all identified single units, and the unit exhibiting the highest firing rate in each channel was selected for visualization. (g) Example of NIR light‐induced cell firings in murine retinal tissue on a PV implant with PyrC electrodes. Top, raster plots. Bottom, peri‐stimulus time histograms. Gray shade, phase of light stimulation. (h) and (i) NIR light‐induced (f) and white‐light‐induced (g) firings in cells without PV implant. (j) Population PSTHs (line, an average; shade, SD) of cells with (top; 98 cells, 5 retinae) and without PV implant (bottom; 67 cells, 4 retinae). (k) Comparisons of firing rates among cells with PV implant (98 cells, 5 retinae), cells without PV implant during NIR, and without PV implant during white light (67 cells, 4 retinae). The same retinal preparations were used for the NIR and white‐light recordings in the non‐implant condition.

NIR wavelengths lie outside the visible spectrum for mouse photoreceptors. Indeed, NIR illumination without photovoltaic implant did not induce retinal activity serving as negative control (Figure 3b). Notably, significant responses to NIR light illumination were observed when the photovoltaic implant was placed on the retina (Figure 3c; amplitudes 40.5 ± 18.9 µV; peak firing rates 36.8 ± 11.3 Hz). The actual shape of the action potentials recorded with and without implant was nearly identical (Figure 3d,e), indicating that photovoltaic‐induced signals were biological in nature. The firing rate during NIR illumination significantly increased when the photovoltaic implant was present (Figure 3f,g), while it remained comparable to spontaneous activity without it (Figure 3f,h). RGC responded to white light (Figure 3i–k), confirming that the retinal tissue remained viable with the PV implant and that NIR illumination itself did not activate RGC sensitive to white light, thereby ruling out direct NIR sensitivity.

Here it should be noted that the individual spikes in the MEA recordings likely represent single‐cell action potentials rather than compound activity. Given that RGC somata typically measure 20‐40 µm, MEA electrodes are 20 or 50 µm in size, and the inter‐electrode distance is 100‐150 µm, each electrode is expected to record activity from only one nearby RGC (Figure 3a).

Next, photovoltaic implant‐induced response kinetics were analyzed. For both PyrC and Au electrodes, the signals induced by the photovoltaic implant displayed larger signal intensity (Figure 4a–c), smaller peak firing rate (Figure 4d,e), and longer response latency (Figure 4f,g) compared with the control signal of responses induced by white light illumination without the implants. There was a significant negative correlation between response latency and signal amplitudes (Figure 4h; r = −0.29, p < 0.05), indicating that stronger responses were associated with shorter latencies, while weaker responses had longer latencies. This may reflect the efficacy of RGC activation during photovoltaic stimulation, which could be influenced by several factors, such as distance between individual cells and stimulation electrodes, local electric field strength, and cell types. Importantly, the photovoltaic implant‐induced signals were suppressed after addition of tetrodotoxin (TTX) for pharmacological blocking of Na channels, which supports that the observed signal is biological (Figure 4i).

FIGURE 4.

Response kinetics in the photovoltaic (PV) implant‐induced signal. (a, b) Histograms of signal intensity. Top, NIR‐induced signals in cells with implants having a) PyrC electrodes (blue; 98 cells, 5 retina), (b) Au electrodes (light cyan; 76 cells, 5 retina) on PV pixels with 40 µm sides. Bottom, white‐light‐induced signals in cells without (W/O) implant (light gray and dark gray). Insets: Detected action potentials (gray, all action potentials; black, average). (c) Comparison of cumulative histograms of signal amplitudes under three conditions: NIR‐induced signal in PV implants with PyrC (blue), PV implants with Au (light cyan), and white‐light‐induced signal without the implants (dark gray; 67 cells, 4 retina). Kolmogorov–Smirnov test. (d) Example PSTHs obtained for four different conditions: NIR‐induced signal in PV implants with PyrC (blue), PV implants with Au (light cyan) and without the implants (light gray; 67 cells, 4 retina), and white‐light‐induced signal without the implants (dark gray). (e) Comparisons of peak firing rates for the four different conditions. Mann‐Whitney‐Wilcoxon test. (f) Histograms of response latency. Top, NIR‐induced firings in cells with implants with PyrC electrodes (blue). Middle, NIR‐induced firings in cells with implants with Au electrodes (light cyan). Bottom, white‐light‐induced firings in cells without implants (dark gray). (g) Comparisons of response latency. Kolmogorov‐Smirnov test. (h) Relationship between response latency and signal amplitudes in PV implant‐induced responses. Gray dots, individual cells. Black line, linear regression. (i) Comparison of NIR‐induced firings in cells with photovoltaic implants with PyrC and Au electrodes before (control) and after TTX application (Nav blockings). Wilcoxon signed‐rank test.

Furthermore, it was found that responses induced by photovoltaic implants with PyrC electrodes and the reference with Au electrodes exhibited distinct properties: The implant with PyrC electrodes elicited higher signal intensities (Figure 4a–c), higher instantaneous firing rates (Figure 4d,e), and shorter latencies (Figure 4f,g) compared to the implant with Au electrodes. These results collectively indicate basic differences in the ability to induce retinal activity, suggesting that the implant with PyrC electrodes may excite the retina more effectively than the implant with Au electrodes.

To determine whether the observed photovoltaic implant‐induced signals originated from RGC, loose‐patch recordings from single cells in the ganglion cell layer (putative RGC) were performed (Figure 5a). The isolated mouse retina was mounted in a recording chamber with the ganglion cell side up, and ganglion cell layer cells were targeted by a patch pipette. The NIR light stimulus was projected with a custom‐made setup consisting of a digital light projector (DLP). The size of the static spot during illumination was gradually increased from 30 to 800 µm. Figure 5b,c shows that photovoltaic implants with PyrC electrodes were able to induce NIR illumination‐induced responses in putative RGC (Figure 5d). Furthermore, the signal wave forms were not significantly different for spontaneous single cell spikes before and spikes evoked during NIR illuminations (Figure 5e,f), supporting the conclusion that the observed activity represents genuine biological responses.

FIGURE 5.

Photovoltaic implant‐induced putative retinal ganglion cell responses. (a) Schematic of loose‐patch recording on single RGC cells in mouse retina with PV implant. (b) NIR illumination (spot diameters, 30–800 µm)‐induced spikes recorded from two putative RGC. Gray, phases of NIR stimulation. (c) Example NIR illumination‐induced spikes in a putative RGC (Cell#1) by 200 µm NIR spots (3 trials). Top, raster plots. Bottom, peri‐stimulus time histograms. Gray shade, phase of NIR stimulation. (d) Comparison of firing rates before and during NIR illumination, with and without (W/O) PV implants (5 cells, 5 retina). Light gray indicates individual cells, and black represents the average. (e) Spike waveforms before (spontaneous) and during (evoked) NIR stimulation. (f) Evaluation of waveform similarity between NIR‐evoked and spontaneous spikes. Similarity was quantified using Pearson correlation coefficients between NIR‐evoked and spontaneous spike waveforms recorded from the same cell (“Same cell”). As a control, shuffled correlations were calculated between NIR‐evoked waveforms from one cell and spontaneous spike waveforms recorded from different cells (“Shuffled”). Light gray indicates individual cells, and black represents the average.

2.4. Photovoltaic Stimulation of Porcine Retinal Tissue

The porcine eye is anatomically closer to the human eye compared to that of a mouse. Therefore, ex vivo electrophysiological recordings of neural activity during photovoltaic stimulation with the Si‐based implant were performed on porcine retinal tissue. For these experiments, implants with 200 µm pixel diameter and integrated 35 µm tall PyrC pillar electrodes (Figure 1m,n) were used. In a previous study we have shown that 3D pillar electrodes with similar dimensions enabled electrical stimulation of pig retina [44]. Segments of neuroretinal tissue with a size of approximately 3 mm × 10 mm were excised from eyes of Danish Land Race pigs (Sus domesticus) 2–3 mm from the optical disk. The tissue samples were placed on the photovoltaic implant with the photoreceptor layer facing down towards the PyrC pillar electrodes (Figure 6a). A 32 channel MEA was positioned on the RGC side of the tissue for signal recording. A 940 nm NIR laser was placed below the assembly for illumination of the photovoltaic implant and the retinal tissue.

FIGURE 6.

Photovoltaic stimulation of porcine retinal tissue. (a) Schematic of experimental setup for photovoltaic stimulation of porcine retina. The dissected neuroretina (top left) was placed with the photoreceptor side downwards on a photovoltaic implant in a 3D‐printed holder (top right). An O‐ring was used to seal the stimulation compartment and the holder top mounted with screws. The MEA was inserted through a hole in the top facing the RGC side of the tissue. A NIR‐laser illuminated the implant from below (bottom). Created by Biorender.com. (b) Left, raw data trace of single MEA channel recording during stimulation (bottom) compared to spontaneous activity (top). Right, example spike waveforms detected in the (top) spontaneous and (bottom) stimulation data trace, where gray traces represent all spikes and the blue trace is the average. (c) Mean relative spike counts, (d) mean firing frequencies and (e) average spike amplitudes for all eyes (N = 3). Gray indicates individual eyes and black represents the average. Bars indicate standard deviations.

For each tissue sample, a baseline of spontaneous activity was established by a 10 s recording prior to stimulation. Figure 6b shows an example of a raw data trace for one MEA channel recorded for 120 ms after a pulse of NIR stimulation compared to spontaneous activity of porcine retinal tissue. The stimulation trace displays post stimulation spiking scattered across the whole duration, while the spontaneous activity trace shows typical activity without stimulus including a single spike in this case. Figure 6c–e shows quantified data from three porcine retinal tissue samples each obtained from a different eye. Figure 6c displays the mean spike counts relative to stimulation spike counts for all samples. Relative data representation was chosen to minimize effects of inter‐tissue sample variation, which can be substantial. The spike count increased roughly fivefold to ∼225 Hz and mean amplitudes of ∼400 µV when stimulated via the photovoltaic chip with carbon electrodes and 940 nm illumination compared to ∼50 Hz and ∼60 µV for spontaneous activity. This demonstrates the ability of the photovoltaic implant with the 3D PyrC pillar electrodes to activate porcine neuroretina. To test whether recorded spikes were biological in origin TTX was added to the well after stimulation within the setup [54, 55]. TTX is a voltage‐gated Na⁺ channel antagonist and will prevent cells in the retina from generating action potentials. After administration of TTX, the tissue was allowed to rest for 5 min, followed by photovoltaic stimulation with the same parameters as before. After addition of TTX, the spike count decreased significantly to less than 10% of the counts during initial photovoltaic stimulation (Figure 6c), which confirms the biological nature of the signals evoked during NIR illumination. Furthermore, the photovoltaic implant stimulated tissue activity also displayed significantly higher firing frequencies of recorded spikes compared to spontaneous and stimulation after addition of TTX (Figure 6d). Finally, both relative (Figure S6) and absolute peak‐to‐peak spike amplitudes (Figure 6e) were significantly higher when recorded during stimulation compared to spontaneous activity and stimulation after TTX addition. After stimulation, spike amplitudes exceeded 400 µV and were significantly higher than spontaneous and post‐TTX spikes.

3. Discussion

The first objective of this study was to investigate if integration of PyrC as a novel electrode material on functional Si‐based photovoltaic implants is feasible. The main concern was that the long high‐temperature process at 900°C used for conversion of polymer precursor structures into PyrC electrodes might affect the photovoltaic properties of the solar cell pixels in the high‐density array. The additional thermal process might alter the intended p‐ and n‐doped regions, that is, further drive‐in dopant atoms. While this presumably still occurs to some degree, the I‐V curves in Figure 2 confirm acceptable photovoltaic behavior indicating that doped regions are resulting in the diode behavior needed for an effective photovoltaic response to NIR light. Furthermore, the photovoltaic performance of pixels with PyrC electrodes was comparable to those with Au electrodes, which were not exposed to the additional high‐temperature process.

The second major research question was whether photovoltaic stimulation of retinal tissue using the photovoltaic implant with the PyrC electrodes is possible. The electrophysiological ex vivo measurements on retinal tissue from mice and pigs (Figures 3, 4, 5, 6) demonstrate effective stimulation when the photovoltaic implants with PyrC electrodes were illuminated by NIR light with a wavelength of 940 nm. For mouse retinas, spike rate and amplitude resulting from stimulation via the NIR illuminated photovoltaic implant were significantly higher than without implant given the same NIR illumination and on par with the response of the same tissue to white light illumination without the implant. This result shows the apparent effect of not just the presence of either the photovoltaic device or the NIR illumination, but of NIR illumination leading to photovoltaic stimulation of retinal neurons by the implant.

For mouse retinas, the average magnitudes of neural activity induced by the photovoltaic stimulation ranged from 20 to 60 µV in signal amplitudes and from 20 to 50 Hz in instantaneous firing rates (Figure 3). These values are consistent with those reported in previous studies of mouse RGC activity [56, 57, 58], supporting the biological validity of our recordings. Indeed, the responses observed with the photovoltaic chip during NIR illumination fall within this physiological range (median firing rate: 40.83 Hz for NIR with implant; 42.67 Hz for white light without implant). This consistency also indicates the ability of the implant to elicit responses in mouse retinas under NIR illumination similar to their normal behavior under broadband visible light illumination. However, differences in response properties such as signal amplitude and latency between photovoltaic stimulation and white light stimulation were observed (Figure 4). Specifically, there was a negative correlation between latency and signal amplitude (Figure 4f), indicative of differences in the efficacy of RGC activation by photovoltaic stimulation. For example, cells located closer to the PyrC electrodes are more strongly activated electrically, rather than relying solely on synaptic transmission as in light stimulation [59]. In contrast, cells located more distant from the PyrC electrodes tend to be less effectively activated, resulting in lower response fidelity and longer latencies. Furthermore, photovoltaic stimulation can spread spatially across the retinal neural network. Such widespread activation may recruit non‐physiological lateral modifications mediated by interneurons, such as horizontal cells and amacrine cells, leading to prolonged response latency through enhanced inhibition in RGC. It is possible that certain cell types receive stronger electrically induced inhibition and therefore exhibit longer latency, while others do not. Indeed, previous studies have shown that changes in response properties under electrical stimulation are cell‐type dependent, with some cells displaying markedly prolonged latency [60, 61]. In future studies, cell type‐dependent evaluations of the effect of photovoltaic implant induced stimulation should be conducted.

For pig retinas, spike rate and amplitude were approximately four and three times higher, respectively, compared to spontaneous activity from similar tissue without the presence of implant and NIR light. When adding TTX, both spike rate and amplitude were reduced, even compared to spontaneous activity. This significant reduction of activity caused by TTX application confirmed the biological nature of the neural response evoked by NIR illumination. While these results are very clear and encouraging, a few differences in the measurement setup compared to that of the mouse study have to be noted: (i) Spontaneous activity from pig retinas was recorded with ambient background light from the surrounding lab, while in the mouse study tissue was deliberately illuminated with a white light source in a dark room and (ii) for the pig retinas, photovoltaic stimulation was performed with “bursts” of NIR light, that is, several pulses within a short duration, rather than just a single NIR pulse. Single‐pulse stimulation did not result in significant spiking activity, which is why it was required to shift to “burst” stimulation. Whether this is due to the pig retina requiring a larger electrical input compared to mouse retinas, or simply a difference in the two setups, for example, involving the MEA‐tissue‐chip connection, remains an open question. A possible explanation could also be that a relatively large fraction of possible spikes in the pig retinas was not recorded or counted. As seen in Figure 4, a large fraction of the recorded spikes in the mouse retinas had amplitudes < 30 µV, which is the approximate noise level of the pig retina MEA recordings. Furthermore, many spikes of this amplitude would not be counted by the automated counting process used for the pig retinal data, as the amplitude of such spikes are below the RMS square criterion used. For pig retinas, others reported that wild type retinas exhibited biological activity when varying pulse duration from 0.5 to 2 ms and pulse amplitudes from 10 to 30 µA with 50 Hz stimulation frequency [62]. This is comparable both in duration and frequency to the data presented here. The stimulation pulses reported in [62] were generated by one of the electrodes of the MEA, whereas in this study a photovoltaic implant was employed, thus indicating that the implant can produce consistent electrical pulses that lead to RGC spiking.

Similar to what others have reported for porcine neuroretina [54] the inter‐tissue sample activity variation was significant in this study. This is expected and caused by several factors including variability in the dissection of the eye from the animal during slaughter, transportation time and storage before use in the lab, dissection of the retinal sample from the eyes, and the specific degree of contact achieved between the implant, tissue and MEA in each individual case. Finally, slaughter pigs are not bred for biomedical experimentation and have far greater genetic and phenotypic variation between individuals than purpose bred laboratory animals. However, it should be noted that significant inter‐tissue variability is also expected clinically amongst AMD and RP patients. It should also be stated that not all channels of the MEA recorded spikes for all eyes. Data from the channels where no spikes were observed during stimulation were discarded in the analysis. The variation in spikes recorded across channels within single tissue samples is likely caused by factors pertaining to the density and condition of excitable cells near a given channel and again the degree of contact between implant, tissue and MEA in the specific area. These sources of variation in the porcine experiments should be addressed in future experiments.

Since this study includes both a detailed characterization of the photovoltaic device itself and electrophysiological measurements with retinal tissue, it permits to quantify the required current and voltage levels for this type of photovoltaic retinal stimulation. The maximum photocurrent and V _oc of the implant with 200 µm pixels and PyrC electrodes at 940 nm wavelength at maximum intensity was around 10.2 µA and 0.49 V, respectively. For the smaller pixels with PyrC electrodes the maximum photocurrent was 238 nA and the V _oc 0.46 V. In the electrophysiological experiments at the same wavelength, such current and voltages are shown to be sufficient to elicit responses on par with natural retinal response to white light in case of mice, and significantly higher than spontaneous activity in pigs. In previous work an apparent threshold voltage level of 0.5–0.6 V for electrical stimulation of pig retina via similar 3D carbon electrodes has been shown [44]. The results presented here indicate that significantly increased spiking activity occurs at even lower voltages using the photovoltaic implant, which is promising for the use of single‐junction Si‐based solar cells for retinal prosthesis. Given the simplicity of the presented photovoltaic implants, that is, only one pn‐junction per pixel, non‐optimized surface passivation, emitter doping and anti‐reflective coating, and presumably a yet suboptimal electrical connection between Si substrate and PyrC electrodes, the presented biological responses are very encouraging for future work on photovoltaic retinal prosthesis.

The use of PyrC as the stimulating electrodes confirms the ability of this material to be successfully incorporated into a Si‐based photovoltaic implant and transfer sufficient electrical charge to the retinal neurons to elicit biological spiking responses in both mouse and pig retinas. The proposed strategy for integration of PyrC electrodes in retinal prosthesis is highly scalable because it employs methods that are available from established microelectronic device fabrication. The use of photolithography to pattern the precursor structures provides excellent design flexibility and the option to further explore 3D electrode fabrication. Furthermore, no toxic materials are involved and PyrC provides excellent biocompatibility. These are all aspects that identify PyrC as highly promising novel electrode material for retinal implants and should be investigated further in future work.

4. Methods

4.1. Silicon Solar Cell Array Fabrication

Hundred‐millimeter double‐side polished Czochralski Si n‐type wafers with resistivity of 1–20 Ωcm and a thickness of 350 ± 15 µm were used as substrates. The mean minority carrier lifetime of the wafers before any fabrication done measured with a MDP mapper from Freiburg instruments was 262 µs. A 600 ± 50 nm SiO₂ layer was thermally grown in a Tempress tube furnace. The n‐type region was defined by maskless lithography (MLA150, Heidelberg Instruments, Germany) with a 1.5‐µm‐thick AZ MIR 701 photoresist and dry oxide etching with C₄F₈ and O₂ plasma chemistry in an advanced oxide etcher (AOE, STS MESC Multiplex ICP). A 90 ± 5 nm thick phase layer of phosphorus‐doped oxide was formed in a Tempress diffusion furnace using POCl₃ at 950°C for 30 min, followed by 20 min annealing in N₂ atmosphere to activate the atoms for diffusion. The doped oxide phase layer was etched in buffered hydrofluoric acid (BHF) for 10 min. The drive‐in process of the phosphorus atoms into the Si was conducted at 950°C for 70 min. The p‐type region was defined by maskless lithography and oxide etching in an identical process as for the n‐type region. Solid boron source wafers (B₂O₃) and oxidation in O₂ atmosphere at 1125°C for 30 min were used to form a 80 ± 13 nm boron‐doped oxide layer. After annealing for 60 min at 900°C, the boron phase layer was etched in BHF for 10 min. Next, the drive‐in process was run for 90 min at 950°C. Another step of UV photolithography was performed for selective etching of the thermal oxide formed during the drive‐in process using oxide etching with C₄F₈ and O₂ plasma to open electrode contacts. Resistivities of doped regions were measured with a 4‐point probe on test wafers from the same batch.

4.2. Pyrolytic Carbon Electrode Fabrication

A 5 ± 1 µm thick SU‐8 2005 photoresist was spin‐coated and soft‐baked for 15 min at 50°C. This uniform layer was then patterned by maskless lithography (MLA 100, 365 nm laser) and post exposure baked for 2 h at 50°C. On the 200 µm pixels, a 62.3 ± 3 µm thick layer of SU‐8 2075 was spin‐coated and soft‐baked for 5 h at 50°C. SU‐8 was exposed in MLA100 and post exposure baked for 2 h at 50°C. The SU‐8 structures were developed in propylene glycol methyl ether acetate (PGMEA) for 5 min (40 µm pixels) or 30 min (200 µm pixels) and hard‐baked for 15 h at 90°C. The SU‐8 precursor structures were pyrolyzed in an ATV furnace (PEO‐604) at 1050°C in N₂ atmosphere for 3 h with subsequent heating and dwell steps as described elsewhere [44].

4.3. Au Electrode Fabrication

A 2.2‐µm‐thick layer of AZ 5214E photoresist was spin coated and patterned using MLA150 followed by a post exposure bake, flood exposure, and development revealing the contact regions. Subsequently, a 30 nm Cr layer and 200 nm Au were deposited on the contact regions via e‐beam evaporation (Temescal). The photoresist with Au is lifted‐off in Microposit Remover 1165 resulting in Au contacts on the doped regions.

4.4. Photovoltaic Pixel Characterization

The experimental setup for AM 1.5 conditions consisted of a Newport solar simulator (Figure S1) equipped with a xenon arc lamp and an AM 1.5G specific filter. It delivered a maximum output intensity of 100 mW cm⁻² at the plane of the solar cells. The temperature was kept to approximately 25°C. A Keithley 2400 source meter was used to perform the I‐V measurements. Probes allowing for four wire measurements were used to contact the photovoltaic test chips. The experimental setup for NIR studies (Figure 2E) consisted of an LED from Thorlabs positioned 2 cm above the solar cells and emitting at wavelengths of 880 nm and 940 nm. The setup included a custom built LabView (National Instruments) data analyzer that sets up the I‐V sweep and reads the data from the Keithley system. Prior to the experiment, the illumination intensity was calibrated using a reference Si solar cell (ESTI). For NIR measurements, the I‐V response for different intensities at 880 and 940 nm was recorded for all pixel types. This setup allowed for comprehensive analysis of the device performance by extracting the V _oc, I _sc, and FF of the solar cell [53]. The measurements were conducted on n = 6 different chips for each pixel design.

4.5. Preparation of Mouse Retina

Wild‐type mice (C57BL/6J) were obtained from Janvier Labs. 8‐ to 24‐week‐old mice of either sex were used. Mice were group‐housed throughout and maintained in a 12 h/12 h light/dark cycle with ad libitum access to food and water. One retina was sampled from one mouse. Five retinas (98 cells) were used for PyrC electrode experiments, 5 retinas (76 cells) for Au electrode experiments, 4 retinas (67 cells) for recording without photovoltaic implant, and 5 retinas (5 cells) for single‐cell recording experiments. All animal experiments were performed according to standard ethical guidelines and were approved by the Danish National Animal Experiment Committee (Permission No. 2015−15−0201−00541; 2020‐15‐0201‐00452) and National Institute of Genetics (R7‐1). Retinas were isolated from the left eye of mice dark‐adapted for 1 h before experiments. During the procedure, the retina was illuminated by dim red light (KL 1600 LED, Olympus) filtered with a 650 ± 45 nm band‐pass optical filter (ET650/45×, Chroma) and bathed in Ringer's medium (all concentrations in mM): 110 NaCl, 2.5 KCl, 1 CaCl₂, 1.6 MgCl₂, 10 D‐glucose, 22 NaHCO₃ bubbled with 5% CO₂, 95% O₂. The retina was kept at 35°C–36°C and continuously superfused with oxygenated Ringer's medium during recordings. For pharmacological experiments, TTX (1 µM, Tocris) was used by bath application.

4.6. Electrophysiological Recordings on Mouse Retina With MEA

The isolated mouse retina was mounted on 64 channel probes (MED‐P210, 20 µm × 20 µm size, 100 µm intervals; MED‐P515, 50 µm × 50 µm size, 150 µm intervals; Alpha MED) with the photoreceptor side up. Spike discharges of RGC were amplified and acquired (MED‐A64MD1; MED‐A64HE1; Alpha MED). Recorded spike discharges were band‐pass filtered between 100 and 3000 Hz and sorted into single unit activities by principal component analysis (PCA) and the template‐matching method with custom programs using MATLAB (Mathworks) [63, 64, 65]. Briefly, candidate spike events were first detected from the raw traces using a threshold criterion (mean ±3SD). The detected events were then projected into a two‐dimensional feature space using PCA (Figure S6) and clustered using the k‐means algorithm. The optimal number of clusters was determined based on the Bayesian information criterion (BIC). For each cluster, a spike template was generated by averaging the waveforms of all events assigned to that cluster. Template matching was subsequently performed by calculating the Pearson correlation coefficient between each template and candidate event waveform. Events that matched a given template were assigned to the corresponding single unit. To assess sorting quality, we computed the auto‐correlogram of each isolated unit and confirmed the absence of spikes within the refractory period (e.g., within 3 ms time delay), indicating appropriate unit isolation. For the photovoltaic implant stimulation, the retina placed on the top of the recording probe was illuminated by 940 nm NIR (150–200 mW cm⁻², 2 s illumination) controlled by an Arduino board (ELEGOO UNO R3). For white light illumination, a static spot (50% contrast, 2 s illumination, 500 µm diameter) generated by Psychtoolbox3 on MATLAB was used.

4.7. Single‐Cell Spike Recordings on Mouse Retina

The isolated retina was mounted with the ganglion cell side up on a small piece of filter paper (MF‐membrane, Millipore), in which a 2 mm × 2 mm window was cut. Loose patch recording of spikes from RGC was performed. Signals were acquired using an Axon Multiclamp 700B amplifier (Molecular Devices) and customized software on LabVIEW (National Instruments) developed by Zoltan Raics (SENS Software) and digitized at 10 kHz. Borosilicate glass micropipettes pulled by a micropipette puller (P‐97, Sutter Instrument), filled with the Ringer's medium, were used for recordings. The visual stimulation was generated via custom‐made software (Python and LabVIEW) developed by Zoltan Raics (SENS Software). The stimulus was projected through a DLP (NP‐V311X, NEC). The stimulus was focused on the photoreceptor layer of the mounted retina through a condenser (WI‐DICD, Olympus). The light was filtered with a band‐pass optical filter (ET745/40m, Chroma). A static spot (50% contrast, 2 s illumination) with different diameters from 30, 50, 100, 200, 400, and 800 µm was used. To assess responsiveness, the number of spikes recorded before and during light stimulation was compared across repeated presentations of the light stimulus (3–5 trials) using a Wilcoxon signed‐rank test. Cells were classified as responsive when a significant difference was detected. To determine whether NIR‐evoked spikes exhibited waveform characteristics consistent with physiological spontaneous spikes, waveform similarity was quantified using Pearson correlation analysis. For each cell, the correlation coefficient between NIR‐evoked and spontaneous spike waveforms recorded from the same cell was calculated. As a control, a shuffled distribution was generated by calculating correlations between NIR‐evoked waveforms from one cell and spontaneous spike waveforms recorded from different cells. Then the correlation coefficients obtained from the within‐cell and shuffled comparisons were statistically compared. A significantly higher correlation in the within‐cell condition would indicate that NIR‐evoked spikes more closely resemble the cell's own physiological spontaneous spikes than spike waveforms recorded from other cells.

4.8. Preparation of Porcine Retina

Eyes from Danish land race pigs (Sus domesticus) were collected at a slaughterhouse (Danish Crown, Horsens, Denmark) immediately after the animals had been stunned with carbon dioxide and euthanized by exsanguination. For storage and transportation, physiological saline solution (PSS) was used (concentrations all in mM): 118 NaCl, 4.8 KCl, 1.14 MgSO₄, 25 NaHCO₃, 5 Hepes, 1.5 CaCl₂, 5.5 glucose. The eyes were transported to AUH laboratory in PSS at 4°C, and the time from the collection of the eyes to the commencement of the dissection procedure never exceeded 1 h. For dissection, PSS0.0 was prepared: 119 NaCl, 4.7 KCl, 1.17 MgSO₄, 25 NaHCO₃, 5 Hepes, 5.5 glucose, 1.18 KH₂PO₄, 0.026 EDTA. Dissection was performed in 4°C PSS0.0 as follows: Each eye was bisected at the equator with a double‐edged razor blade, and the anterior segment was removed. The posterior segment containing the optical disk was placed under a stereomicroscope and the vitreous body was removed. 2‐3 mm from the optical disk a segment of approximately 3 mm × 10 mm neuroretina tissue was cut out using a self‐locking chisel blade handle (VWR International, Herlev, Denmark) equipped with a 30° microblade (BD Beaver, D.J. Instruments, Billerica, USA). Care was taken to avoid large blood vessels in the selected segment.

4.9. Electrophysiological Recordings on Pig Retina With MEA

A 3D printed holder was designed and printed with a Formlabs 3D printer [44]. It was designed to accommodate the photovoltaic implant with the PyrC electrodes in the holder base. The retinal tissue samples were placed on the implant with the photoreceptor layer facing down towards the PyrC pillar electrodes. A silicone O‐ring was placed on top of the chip for sealing purpose. PSS1.6, which refers to PSS0.0 with 1.6 mM CaCl₂, was then added to the reservoir. PSS1.6 was heated to 37°C and oxygenated by bubbling with a gas mixture of 95% atmospheric air and 5% CO₂ prior to all experiments, to maintain the tissue at close to physiological conditions during electrical measurements. A 32 channel Blackrock MEA was positioned on the RGC side of the tissue through the well after the lid was fastened to the holder bottom. Data was recorded using the 32 channel MEA connected to a Cereplex data acquisition unit (DAQ) via a Cereplex M headstage. The setup was surrounded by a Faraday cage. Data was sampled at 30 kHz and high pass filtered at 250 Hz before being stored for further analysis. A Thorlabs 940 nm NIR laser was placed below the holder for illumination of the PV implant inside. The maximum light intensity of the NIR light source (124 mW cm⁻²) was used for all porcine retina experiments. Each stimulation run lasted for 10 s with stimulation pulses at 10 Hz resulting in a total of 100 stimulations. Stimulation characteristics varied between tissue samples, as the stimulation regimen needed for successful activation of the tissue changed between samples. All stimulation consisted of a series of tightly spaced (buzz) light pulses, with 1.5 ± 1 ms duration, 1.5 ± 1 ms inter‐pulse‐interval with a total stimulation time between 1 and 30 ms. TTX citrate was obtained from Hello Bio Reagents (Bristol, United Kingdom, HB1035). 5 µM TTX solutions were prepared from a stock solution in PSS1.6 of 1.0 M concentration. One milliliter aliquots of 2.25 mM concentration were prepared for use in the experiments. All TTX‐solutions were stored at −20°C.

4.10. Spike Analysis (Mouse Retina)

To quantitatively analyze the signal waveforms, first waveform segments were extracted corresponding to spike events identified by spike sorting. For each event, the waveform segment was defined using the time points corresponding to one time constant (τ) before and after the signal peak, where τ was determined as the time required for the waveform to reach 63.2% of its peak amplitude during the rising and decaying phases, respectively. To ensure that the entire signal waveform was included, the segment boundaries were extended by an additional 2 ms before the pre‐peak τ point and 2 ms after the post‐peak τ point (Figure S7). Signal width was defined as the time interval between the waveform reaching 63.2% of its peak amplitude during the rising phase (t0) and the waveform decaying to 63.2% of its peak amplitude during the falling phase (t1). For amplitude quantification, the baseline was defined as the mean signal during the first 1 ms of the extracted segment, and the signal amplitude was calculated as the difference between this baseline and the peak value. To quantify firing rates, peri‐stimulus time histograms (PSTHs) were constructed from the sorted spike trains across repeated presentations of the light stimulus (3–5 trials) using a bin size of 200 ms. The firing rate for each cell was defined as the maximum value of its PSTH. Response latency was defined as the time interval between the onset of the light stimulus and the time point corresponding to the peak firing rate in the PSTH.

Author Contributions

Akihiro Matsumoto: methodology, writing – review and editing, visualization, writing – original draft, investigation. Pratik Kusumanchi: investigation, writing – original draft, methodology, visualization. Jesper Guldsmed Madsen: investigation, writing – original draft, methodology. Asbjørn Cortnum Jørgensen: writing – review and editing, methodology, visualization. Gisele Alves dos Reis Benatto: methodology, validation, writing – review and editing, investigation. Sune Thorsteinsson: investigation, writing – review and editing, methodology. Peter Behrensdorff Poulsen: writing – review and editing, supervision, resources, methodology. Keisuke Yonehara: conceptualization, investigation, funding acquisition, writing – original draft, writing – review and editing, methodology, project administration, supervision. Toke Bek: conceptualization, funding acquisition, writing – review and editing, methodology, project administration, supervision. Stephan Sylvest Keller: conceptualization, funding acquisition, writing – original draft, investigation, methodology, writing – review and editing, supervision, project administration. Rasmus Schmidt Davidsen: conceptualization, investigation, funding acquisition, writing – original draft, writing – review and editing, methodology, project administration, supervision, resources.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File: adhm71651‐sup‐0001‐SuppMat.docx.

Matsumoto A., Kusumanchi P., Madsen J. G., et al. “Photovoltaic Stimulation of Mouse and Pig Retina With Pyrolytic Carbon Microelectrodes Integrated on High‐Density Silicon Solar Cell Arrays.” Advanced Healthcare Materials 15, no. 36 (2026): e71651. 10.1002/adhm.71651

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.