During the optimization of the complex fabrication process, several challenges that affected the implant quality and feature resolution were overcome. The main challenges include the optimization of UV exposure for achieving a high aspect ratio of SU-8 wells, the formation of dense gold microelectrode arrays with sharp edges and with strong bonding to the SU-8 surface, avoiding thermal stress development within the SU-8, avoiding the creation of "streaming lines" due to a multilayer lithography process, implant release, and bio-functionalizing.

Optimization of the UV exposure in a high aspect ratio device

Optimization of UV exposure during a complex lithography process of a 3D high-aspect ratio device is affected by numerous factors such as the type of photoresist (positive or negative), geometry, the reflections due to the presence of metal electrodes, and more [29, 49] (see the Supp. Material). The exposure dose ranges from underexposure to overexposure; thus, in order to avoid undesired structural defects, the optimal exposure dose needs to be determined. To this end, for each photoresist, substrate and thickness manual optimization was carefully performed using the designated hashtag #, as is described in the Supp. Material and can be seen in Fig. 3 and Figs. S1 and S2. Using these optimization steps, we were able to achieve perfectly circular shaped micro-wells in the desired dimensions for the negative SU-8 photoresist (Fig. 3d) and the positive AZ photoresist (Fig. Supp. S1).

Fig. 3

UV dose effect on the SU-8 patterning. a) An extreme underexposure dose leading to low mechanical strength and rupturing of the film. b) Underexposure dose leading to the expansion (22 µm) of the micro-wells and the fusion of adjacent wells. c) Extreme overexposure, leading to the closing and reduced diameter of the micro-wells. d) Optimal UV dose resulting in perfectly circularly shaped micro-wells of the desired diameter (20 µm). Scale bar - 50 µm in all figures

Dense gold microelectrode on SU-8: strong attachment, high resolution, and sharp profile electrodes

The electrode fabrication process involves thin film metallization of gold by spattering onto the patterned AZ photoresist, followed by a lift-off process that removes the undesired metal (gold) at the unexposed photoresist areas. Two main factors affect this method and limit a successful fabrication of sharp, flat, and clear-cut metal electrode edges: the strength of the bond between the deposited metal (gold) onto the surface (SU-8) and the patterned photoresist profile.

The deposited gold adheres weakly to the SU-8 polymeric surface due to the gold’s inertness and its low surface energy in conjunction with the poor wettability (hydrophobicity) of the SU-8 [40], resulting in electrode detachment and rupture during the lift-off process [49] (Fig. S3). To increase the electrodes’ adherence to the SU-8 surface, we applied several methods aiming to increase the SU-8 surface energy: dry plasma etching (O2, Ar, or N2) treatments for various times and powers (1min to 5min and 100W to 300W, respectively) were applied onto the SU-8 before metallization to modify the SU-8 surface [17, 40] by breaking the epoxy rings, resulting in hydroxyl and carboxyl edge groups [17, 50, 51]. The various dry etch plasmas’ impact on the SU-8 hydrophilicity was evaluated by contact angle measurements (Supp. Fig. S9a-c). In addition, argon (Ar) ion milling was applied prior to the sputter deposition process, in the same chamber, to further increase the SU-8 surface energy. Moreover, we investigated the use of the metal adhesion layers (Ti and Cr) at various thicknesses (5nm-20nm) with the aim of bridging the hydrophobic SU-8 nature and the gold’s inertness (data not shown). Following the optimization process, we concluded that O2 plasma (150W, 5min), combined with (Ar) ion-milling (10s) and the use of a Cr adhesion layer resulted in the best adhesion.

The second challenge for the successful patterning of metal electrodes using a lift-off process is to achieve a proper patterned photoresist profile. Using a negative photoresist for deposition has the advantage of creating the desired trapezii profile [52], which has the advantage of creating clean and sharp electrode edges during the lift-off process, however it is limited in resolution and expensive. On the other hand, a positive photoresist is usually used for high-resolution features, but has the drawback of creating a typical concave (bowl-like shape) profile (Fig. 4a-b), which results in continuity of the metal deposition and increases its lateral surface tension [53] (Supp. Fig. S4), eventually causing tears in the gold electrode during the lift-off process (Fig. 4c); this leaves the so-called “ear-pattern” gold residuals at the electrode edge (Fig. 4d). We opted to use the positive resist and to achieve the desired "negative-like" profile by using a bi-layer lift-off process (Fig. 4e-h) [54,55,56]. The bi-layer process utilizes an additional layer of a fast-developing resist (e.g., PMGI, LOR) under the positive photoresist. This layer dissolves faster than the patterned photoresist during the photoresist development after UV exposure, therefore resulting in an "undercut" profile and thus efficiently separating the desired regions from the undesired metal regions as in the negative "trapezii" shape. In order to achieve the desired undercut profile, various materials with different thicknesses and dissolution rates (such as LOR10B, PMGI sf3, and PMGI sf6) were investigated. Briefly, by implementing a second cycle of a curing step at a temperature higher than the photoresist (i.e., AZ) glass transition temperature (Tg) and lower than that of the dissolved layer (120°C for 1 min), as proposed by Wilson et al. (2015) [56], we could control the dissolution rate and the desired undercut profile. Figure 4e-h. presents the results of the optimal bi-layer lift-off process with the additional curing step. The desired discontinuity between the gold layer and the photoresist can be seen in the FIB/SEM images (Fig. 4e-f); this leads to a complete intact circularly shaped electrode (Fig. 4g) with a clear-cut sharp profile (Fig. 4h).

Fig. 4

Optimization of gold microelectrode fabrication. a-d) A ruptured electrode with an "ear pattern" profile fabricated using the conventional AZ monolayer lift-off process and developed for 1 min. a) Top view SEM imaging (the Pt line is deposited for FIB/SEM and is not part of the fabrication process). b) FIB/SEM zoom in the cross section of the area demarcated with a rectangle revealing the continuity of the photoresist. c) Gold electrodes (80 µm diameter, 200 nm height) are broken and disconnected from the substrate. d) The electrodes’ edge cross section profile showing the “ear pattern” edge (arrows). e–h) An intact electrode with a sharp profile fabricated using the bi-layer lift-off process (AZ and LOR) developed for 1 min, 5 min curing at 185 °C and another development phase of 1 min. e–f) SEM imaging of the positive photoresist profile optimization process of a Bi-layer lift-off with LOR10B. e) An SEM image of the top view of a photoresist developed for 1 min. f) An FIB/SEM cross section and a zoom-in on the white rectangular area demarcated in) (e) (where the LOR10B was developed for 1 min, followed by an additional AZ curing step (120ºC). The photoresist has the desired undercut configuration, which is denoted by *. g) Intact gold electrodes (80 µm diameter, 200 nm height) with good attachment to the SU-8 surface. h) An electrode edge cross section profile showing the sharp edge (arrows)

Avoiding thermal stress

During the curing steps a thermal stress is prone to develop in the SU-8 due to the mismatch between the thermal expansion coefficient (CTE) of the Si wafer (2.6ppmC−1) [57] and the SU-8 (52ppmC−1) [58], resulting in cracks within S-8. We solved this issue by two different approaches. First, we iteratively optimized the heating and cooling gradient following curing (7°C min−1). More importantly, we added an intermediate layer of Crumuim-Nickel (13.3 ppmC−1) [59] between SU-8 and the gold layers (see Figs. 1 and 2). Indeed, these steps significantly eliminated the thermal stress, as shown in Fig. S5 in the Supp. Material; the optimized protocol is described in the Supp. Material.

Multilayer lithography process: streamlines and mechanical stress

While fabricating a 3D multilayer device via a layer-by-layer photolithography process, the cumulative effects of each of the previous steps affect the proceeding ones. One such challenge is the patterned substrate topography, which results in a "streamline" effect of the current photoresist layer, which prevents uniform coating through standard spin coating. Briefly, in order to overcome this challenge, we optimized the spinning protocol to include several steps, which led to a homogeneous uniform coating of the various photoresist layers. The method we used to overcome this is described in the Supp. Material and Fig. S6.

Implant release

Since SU-8 is an epoxy polymer, it tends to adhere to surfaces during polymerization, preventing the release of the device from most wafer substrates. In some processes, a striping technique using the commercially available Omnicoat has been introduced [15]; however, it could not be used in our complex multiple step process because of its interference with the following steps. Therefore, we adopted a “sacrificial” layer approach [60], whereby a sacrificial metal layer is deposited onto the substrate wafer below the SU-8 and is wet-etched by acid at the conclusion of the fabrication process. However, since the same acid can also etch the gold electrode or the Cr/Ti adhesion layer, the process had to be optimized to prevent electrode detachment. Briefly, we found that the best release can be achieved using a 200nm Ni or Cu as a sacrificial layer that is wet etched by 21% NHO3 overnight, similar to Feiner et.al [3] at the end of the process. This, however, results in a yellowish SU-8, which was further prevented using ammonium peroxydisulfate salt (Merck, Germany) 10%v/v instead of the acid. Figure S7 in the Supp. Material presents the electrode adhesion integrity before and after the optimization.

Bio-functionalization of the implant by adhesion molecules and surface treatment

Since the bio-functionalization of the implant plays a critical role in enhancing cell-electrode coupling, we investigated the effect of the functionalization of gold electrodes with a linear RGD pentapeptide and the effect of dry etch plasma on cell attraction and adhesion to the electrode surface. RGD is one of the extracellular matrix (ECM) cell recognition motifs that connect cellular integrins; thus, incorporating this molecule in a device increases cell adhesion to the surface [61, 62]. The presence of the RGD on the electrodes was verified by XPS chemistry (Fig. Supp. S8). The effect of RGD functionalization on the electrode cell interface was studied by seeding retinal relevant cells (ARPE) on electrodes functionalized with RGD and comparing the results to cells cultured on electrodes coated with Matrigel or using uncoated gold as the control (Fig. 5). It can be seen that the adherence ratio (see the Methods) was significantly larger when the electrode was coated with RGD, compared with Matrigel and the control, suggesting the facilitation of cell adherence to the functionalized gold electrodes (p < 0.05, unpaired Students t-test, Fig. 5c).

Fig. 5

Effect of various coatings on cell adhesion to gold electrodes. a-b) SEM image of ARPE cells seeded on gold electrodes (40 µm in diameter, white arrows), used to quantify the effect of various coatings on cell adhesion and the preference of gold electrodes without RGD (a) and with RGD (b); the scale bar = 200 µm. It can clearly be seen that although the cells tend to be repelled by bare gold electrodes (a), the cells tend to adhere to RGD-coated gold electrodes (b). c) The obtained adhesion of cells to gold electrodes for the various coatings defined as the prevention of the electrode area occupied by cells divided by the percentage of the area surrounding the electrodes occupied by cells. Values larger than the one obtained for the RGD coating reveal the cell’s preference for electrodes following this treatment, compared to the control

In addition to electrode functionalization, we addressed the SU-8 surface, which is known to repel cells, because of the hydrophobic epoxy surface structures. Aiming to further bio-functionalize the implant, we used dry plasma etching (150W, 3min of O2 or N2), which breaks the epoxy rings while creating reactive group chains; thus, oxidizing the surface and increases its hydrophilicity [17]. The plasma treatment effect on the surface wettability was assessed by contact angle measurements (Supp. Fig. S9a-c), which showed increased wettability in both the O2 and N2 plasma treatments (a higher contact angle compared with untreated SU-8). Twenty four hours post seeding rPRP cells density was significantly higher on N2 plasma treated SU-8 compared to O2 plasma (24.0 ± 8.6 cells/1000µm2 vs 16.1 ± 5.7, average ± STDEV, p = 0.02, for N2 plasma and O2 plasma, respectively) and compared to no treatment (13.0 ± 16.2 cells/1000µm2 vs 17.4 ± 8.0 cells/1000µm2, P = 0.002, for no treatment and N2 plasma, respectively). Similar results were found for 72h post seeding, as is shown in Fig. 6b (and Supp S9d-i).

Fig. 6

Effect of plasma surface treatment on SU-8 biocompatibility. Rat PRP cell density on N2 or O2 plasma-treated cured SU-8 (150W, 3 min) compared to not treated cured SU8. 24 h (a) and 72 h (b) post seeding. * Denotes p < 0.05

Final fabrication process

The final fabrication process is depicted in Fig. 2 and is detailed in the Supp. Material.

Characterization of the Retinal Implant

Figure 7a depicts a bright-field microscopy image of a completed 1mm diameter implant with a dense circular micro-well electrode array. Further characterization of the implant at higher resolution using SEM (in Fig. 7b) revealed the dense micro-well-like structures and the electrodes with clear-cut features (20μm in diameter, 23μm pitch to pitch). Cross sections obtained through FIB/SEM (Fig. 7c) depict a single micro-well with a gold electrode at the bottom (arrow) with good structural integrity of the various implant components.

Fig. 7

Images of a completed retinal implant (1mm in diameter) with a gold electrode array. a) Color image; the top view of a full SU-8-gold retinal implant; scale bar = 0.5mm. The insert is a zoom-in on the area demarcated by the rectangle (scale bar = 20µm). b) SEM images; the top view of the implant as in a); scale bar = 0.5mm. In the insert, a zoom-in on the area demarcated by the rectangular; scale bar = 20 µm. c) A FIB/SEM cross section image of the 3D well-like structure encapsulating the electrode; scale bar = 10µm. The black pillars are the SU-8 micro-wells walls (*) and the gold electrode (the white arrow). Scale bar = 10µm

Ex-vivo subretinal stimulation

The investigation of the implant feasibility to serve as a subretinal neurostimulator is presented in Fig. 8. As described above, the implant was fabricated on glass and mounted in a multi-electrode stimulation system (Multichannel Systems, Inc.) (Fig. 8a). Transgenic rat retinas expressing the calcium indicator GCaMP6 in their RGC were placed on the implant RGC facing up (Fig. 8b and c), and retinal ganglion cell responses to the implant electrical stimulation were observed through calcium imaging experiments. Figure 8d shows a robust significant repetitive fluorescence signal change indicating the successful subretinal stimulation of the isolated retina. Increasing the stimulus charge resulted in the expected sigmoidal increase in the RGC responses (Fig. 8e). Experiments investigating the activation charge threshold revealed the activation thresholds of 0.156mC/cm2 per phase, comparable to values reported in the literature [63,64,65,66] (Fig. 8d). To validate the nature of the observed activity and to rule out potential artifacts, we added the voltage-gated calcium channel blocker Verapamil (at a concentration of 200 µM). Upon the addition of this blocker, all activity was diminished and was successfully restored upon washout (Supp. Fig. S10), further validating the physiological nature of the observed fluorescence change. Furthermore, using a high-resolution ex-vivo prototype of the implant, we could stimulate localized area of excised retinas (Supp Fig. S11a,b); the activation threshold showed a characteristic strength-duration function (Supp Fig. S11c).

Fig. 8

Ex-vivo retinal stimulation proof of concept. a) The conceptual implant is placed in a dish with Ringer’s medium. b) A fluorescence image of the isolated retina harvested from transgenic GCaMP6f-Thy1 rats mounted on the implant. Arrows point to the micro-wells. c) The same as in b with the focal plane adjusted to show the fluorescent RGCs and axons (arrows). d) Average fluorescence change in response to electrical stimulation with increasing current density. e) Average fluorescence change in response to electrical stimulation with increasing charge density per phase, indicating an activation threshold below 0.156 \(\frac^phase}.\)

In-vivo characterization

To investigate the integration of the device within the retina, it was implanted in the subretinal space of Long Evans rats. Fundus camera imaging and optical coherence tomography (OCT) were performed at 30-days following implantation. The images (Fig. 9) revealed the good anatomical integration of the implant in the sub-retinal space. OCT imaging further highlighted the good proximity between the device and the inner nuclear layer (INL), where the target cells (bipolar cells) are located (Fig. 9, insert). The animal was then euthanized, and the whole mount eye was fixated, treated for nuclear staining (Hoechst), and the bipolar cell marker PKC alpha, and then imaged by confocal microscopy. As shown in Fig. 10, the implant is located in the desired location of the subretinal space (arrow), with some bipolar cells migrating into the micro-wells (insert), as was previously reported [67].

Fig. 9

Integration of the implant in the sub-retinal space. Fundus image of the implanted device (white arrow) demonstrates the good placement near the optic disk; scale bar = 1mm. In the inset, an optical coherence tomography cross section reveals good integration of the implant in the sub-retinal space under the inner nuclear layer (INL), ONL – Outer Nuclear Layer; scale bar = 200µm

Fig. 10

Histology of a flat mount retina implanted with the retinal device. a) Confocal imaging of the implanted retinal device (arrow) showing good anatomical integration in the subretinal space; scale bar = 100µm. The insert shows a high magnification of the micro-wells in the demarcated area showing proximity between the electrodes and bipolar cells entering the micro-wells; scale bar = 100µm. b) A cross section of the implanted retina showing the proximity between the implant and the BPC layer with some of the cells migrating towards the micro-wells (arrows). c) A cross-section at the micro-wells’ mid-height (reference point: the white dashed line in b) revealing the presence of bipolar cells within the micro-wells. Scale bar = 50µm. GC—ganglion cells, BPC—bipolar cells. Green—PKCα, blue—Nuclei

Four animals were followed up for up to 3 months and repeated OCT imaging showed the long-term implant structural stability (Supp Fig. 11a, b). Furthermore, long-term incubation of the implant in saline 0.9% at 37°C showed no significant effect on the electrode impedance (n = 7), as was measured at 1kHz using the NanoZ (Multi Channel System) (p > 0.1) (Fig. 11c, d).

Fig. 11

Implant Stability. a) OCT imaging performed 3 months following transplantation, revealing the structural integrity of the implant and its correct location in the subretinal space. b) The implant impedance at 1kHz was measured before and after a 20-day incubation, showing no significant changes in the electrodes’ impedance (p > 0.2)

To address the important issue of immune cell response, we performed the commonly used immune-cell antibody staining Iba-1 (see Supp. Material for the methods), which stains for microglia and microphages in the retina. We found no differences in the immune response of the implanted retina compared to control eyes (17.0 ± 3.2 vs. 19.8 ± 3.6, average ± STD, p = 0.18 per 500µm retinal slide, for control and implanted eyes, respectively (Fig. 12a-b).

Fig. 12

IBA-1 staining—Cryosections of a control retina (a) and an implanted retina (b) stained for stained for Hoechst (blue), IBA-1 antibody (yellow) and Rhodamine B (red), which visualized the nuclei in the various retinal layers, microglia and microphages, and the implant respectively showing no difference in the immune system response

Moreover, using proTUNEL staining, the implant was found to be biocompatible, as inferred by the similar number of viable ARPE-19 cells following culturing on the implant, compared to those cultured on the control surfaces (Supp Fig. S12).

Discussion and future directions

This work involved the fabrication of a conceptual 1mm circularly shaped complex 3D implant aimed at neural stimulation of the retina. The implant consists of gold micro-electrodes located at the bottom of micro-well structures, with a high aspect ratio and is fabricated using an SU-8—metal—SU-8 configuration. This 3D microwell or “honeycomb” geometry was suggested to enhance cell-electrode coupling by vertically align the electric field and thus decouple the field penetration depth from the pixel width [67].

The optimal fabrication process consisted of three sequential cycles of photolithography formed on a silicon wafer coated with a thin layer of nickel, which was used as a sacrificial layer, followed by spatter deposition of chromium-gold metallization for electrodes patterned by a bi-layer lift-off process, implant release, and RGD bio-functionalization. Although this conceptual implant lacks a current source for neural stimulation (photovoltaic, wired or inductive), it was used for the detailed layer-by-layer optimization of the complex fabrication process, and thus can serve as a guide for fabricating similar devices.

The first challenge was overcoming the thermal stress, which develops in the SU-8 volume during the process, when applied on a silicon wafer. Similarly, to what Guo et al. [68] reported, we concluded that to overcome the effect arising from the different CTE values of the SU8 and the Si substrate, a slow and gradual rise in temperature (7°C min−1) is required, in addition to the use of high-quality masks. Moreover, an additional intermediate layer with an intermediate CTE value further reduced the thermal stress, similar to Abgrall [15]. In our work, the nickel layer (which has a CTE value of 13K−1) served both as a sacrificial and a thermal intermediate layer. Indeed, taking these measures resulted in a reduction of the thermal stress and prevented cracks in the SU-8 layer.

The second challenge we addressed during this process was the formation of the so-called “stream-lines”, which occur during fabrication of a multilayer photolithography process, due to the multilevel effects of each of the previous steps on the proceeding ones. This challenge was resolved by adopting a uniform coating technique, as described above. Next, we tackled with the optimization of the UV exposure dose required for the patterning of the photoresist. The main parameters determining the UV exposure dose are the photoresist height and the substrate reflectance, which create standing waves on its surface, preventing the proper exposure of the resist on its interface [49]. Although this can be partially solved by use of anti-reflecting coatings, the common materials used for this purpose fall short of resolving this issue because they lack sufficient mechanical and chemical properties necessary for integrating permanently in the device and thus become a potential failure point. Thus, we resorted to a step-by-step manual optimization of the dose through the use of fine-resolution marks.

Another major challenge was the poor attachment between the gold electrodes and SU-8 [69]. This obstacle was overcome by adding the steps of surface treatment (O2 plasma and Ar ion milling) to the SU-8 substrate after curing, and by adding an adhesion layer of chromium, which all improved the gold adhesion. More importantly, the electrode integrity was further significantly improved by generating an undercut, a “trapezii”-like, patterned photoresist profile by applying a bi-layer lift-off process, where control over the dissolution rate of the faster developed material (LOR) was achieved by an additional curing step to a temperature that is higher than the Tg value of the photoresist (i.e., AZ) and lower than that of the LOR itself.

Implant release from the wafer was achieved by using a sacrificial layer approach. Since the commonly used Omnicoat (MCC, USA) could not be used in our process, because of the technical constraints described above, we used a metal sacrificial layer. However, this approach, which utilizes wet etching, had to be optimized for preserving the metal electrodes and adhesion layers by optimizing the proper material combination and adjusting the height contrast between the electrode adhesion layer and the sacrificial layer.

Finally, in order to increase the bio-functionalization of the electrodes to attract the retinal neurons and increase the neuron-electrode coupling, the gold electrodes were functionalized by an RGD oligopeptide, a motif of the extracellular peptide fibronectin, which connects to the cell integrin peptides [69]. This RGD monolayer forms spontaneously on the gold surface via a thiol group (SH) when immersing the gold electrodes in aqueous RGD solution overnight at room temperature. This bio-functionalization was performed at the conclusion of the fabrication process and was shown to increase the attraction of neurons to the electrodes in vitro.

The feasibility of using the final implant was demonstrated by successful localized ex-vivo retinal stimulation and was further explored by implanting the device in the sub-retinal space of rats. The later experiments revealed a good anatomical integration of the implant, which was demonstrated by OCT and by histology showing close proximity between the bipolar cells and the electrodes. Our findings additionally demonstrate the migration of bipolar cells into the micro-wells, consistent with a previous study [68], which further confirms the successful integration of the implant into the retinal tissue. Moreover, our results indicate the long-term durability of the implant in both in-vitro and in-vivo settings, with no notable indications of a substan