18.1 An Optically-Addressed Nanowire-Based Retinal Prosthesis with 73% RF-to-Stimulation Power Efficiency and 20nC-to-3μC Wireless Charge Telemetering

Recent approaches toward a functional retinal prosthesis to restore vision in neurodegenerative patients have been limited by the number of pixels that can be individually stimulated. The conventional approach, shown in Fig. 18.1.1 (left), uses an external camera to capture video, whose information is then wirelessly delivered to a hermetic housing (typically a titanium can) for conversion to N distinct stimulation pulses fed transocularly to an $N -$ channel microelectrode array (MEA) placed epi- or sub-retinally. While this approach has proven long-term biocompatibility, translation to 1000s of channels with hermetic and durable transocular connections is not achievable with current technology, limiting existing designs to at most 256 channels, barely sufficient for 20/400 vision restoration [1] –[3]. Instead, recent work has suggested a full-CMOS solution, where an image sensor, stimulation circuits, and an MEA are integrated directly onto a single CMOS chip, thereby eliminating the interconnect challenge and providing a pathway to scalability [4], [5] (Fig. 18.1.1, middle). However, co-location of photoreceptors, amplifiers, and stimulation circuits imposes difficult fill-factor issues, while also significantly increasing heat dissipation near sensitive ocular and neural tissue due to the need to perform voltage regulation, amplification, and charge-balanced stimulation waveform generation (typical end-to-end efficiency of $\sim20$%) right next to the retina. Most importantly, since the CMOS chip must also include electrodes and yet be sufficiently thin to fit epi- or sub-retinally, current encapsulation methods rely on thin-film deposition to enable patterned electrode windows, which has unproven long-term hemiticity or biocompatibility when encapsulating CMOS which has known biological toxins (e.g., copper), limiting practical implementation.