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.
3.1. 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(54, 55) (see the Supp. Material). The exposure dose ranges
from an underexposure dose to an overexposure dose; 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. 2 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. 2d) and the positive AZ photoresist (Fig. Supp. S1).
3.2 Dense gold microelectrode on SU8: 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 (44), resulting in electrode detachment and rupture during the lift-off process(56) (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, 44) by breaking the epoxy rings, resulting in hydroxyl and carboxyl edge groups (17, 57, 58). 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 (10sec) 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 (54), which has the advantage of creating clean and sharp electrode edges during the lift-off process. 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. 3.a-b), which results in continuity of the metal deposition and increases its lateral surface tension(59) (Supp. Fig. S4), eventually causing tears in the gold electrode during the lift-off process (Fig. 3c); this leaves the so-called “ear-pattern” gold residuals at the electrode edge (Fig. 3d). We opted to use the positive resist and to achieve the desired "negative-like" profile by using a bi-layer lift-off process (Fig. 3e-h) (60–62). 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 (120oC for 1 min), as proposed by Wilson et al. (2015) (62), we could control the dissolution rate and the desired undercut profile. Figure 3e-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. 3e-f); this leads to a complete intact circularly shaped electrode (Fig. 3g) with a clear-cut sharp profile (Fig. 3h).
3.3 Avoiding thermal stress
During the curing steps a thermal stress is prone to develop in the SU-8 due to a mismatch between the thermal expansion coefficient (CTE) of the Si wafer (2.6ppmC-1)(63) and the SU-8 (52ppmC-1)(64), resulting in cracks within S-8. We solved this issue by two different approaches. First, we iteratively optimized the cooling gradient following curing. More importantly, we added an intermediate layer of Crumuim-Nickel (13.3 ppmC-1)(65) between SU-8 and the gold layers (see Fig. 1). 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.
3.4 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.
3.5. 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 (66), 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 (67) at the end of the process. This, however, results in a yellowish SU-8, which was further prevented by the use of ammonium peroxydisulfate salt (Merck, Germany) 10%v/v instead of the acid. Fig. S7 in the Supp. Material presents the electrode adhesion integrity before and after the optimization.
3.6. 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 the cell adhesion to the surface (68, 69). 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 two retinal relevant cells (ARPE cells and photoreceptor precursors (46)) 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. 4). It can be seen that the adherence ratio (see the Methods) was significantly larger when it 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. 4c).
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 (100W, 2 min 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’s 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). As an additional qualitative measure for studying bio-functionalization, we found that PRP cells showed a preference for plasma (O2 or N2)-treated implants, as is shown in Supp. Fig. S9d-i. The positive impact of both plasma treatments (N2 and O2) can already be observed 24 hours after seeding, whereas the untreated SU-8 is clearly non-biocompatible at all the investigated time points.
3.7. Final fabrication process
The following section describes the final fabrication process (Fig. 5), which was obtained following the extensive optimization processes and is further described in detail in the supplementary material. This procedure is a sequential process involving three photolithography steps utilizing a silicon wafer surface coated with 200nm Ni (used as a sacrificial layer), followed by Cr/Au electrode metallization, a bi-layer lift-off process, wet etch release (with NHO3 21%), and RGD bio-functionalization of the electrodes.
Finally, bio-functionalization of the gold electrodes is performed by immersing the device in RGD solution (1mM, Adar Biotech, Rehovot, Israel) overnight, to stabilize the semi-covalent bonds, after which it is rinsed with DDW to remove any un-bonded residues.
3.8. Characterization of the Retinal Implant
Figure 6a 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. 6b) 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. 6c) depict a single micro-well with a gold electrode at the bottom (the arrow with good structural integrity of the various implant components).
3.9. In-vitro subretinal stimulation
Our investigation regarding the implant feasibility to serve as a subretinal neurostimulation is presented in Fig. 7. To this end, the implant was fabricated on a glass and connected to a multi-electrode stimulation system (Multichannel Systems, Harvard Bioscience, Inc.) (Fig. 7a). Transgenic rat retinas expressing the calcium indicator GCaMP in their RGC were placed on the implant (Fig. 7b and c). The retinal ganglion cell responses to the implant electrical stimulation were observed through calcium imaging experiments. Figure 7d shows a robust significant repetitive fluorescence signal change indicating the successful subretinal stimulation of the isolated retina. Furthermore, experiments investigating the activation charge threshold highlighted the activation thresholds of 0.156mC/cm2 per phase, comparable to values reported in the literature(70–73) (Fig 7d). 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 200µM). Upon the addition of this blocker, all activity was diminshed and was succssfully restored upon washout (Supp. Fig. S10), further validating the physiological nature of the observed fluorescence change.
3.10. In-vivo characterization
In order 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. 8) 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. 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. 9, 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 (74).