CMOS-based deep brain imager with a monolithically integrated metasurface

Monitoring brain activity with light-mediated methods presents significant advancement over conventional electrophysiological methods, including cell-type specificity, high spatiotemporal resolution over large fields of view, and minimal signal interference. However, an implantable form factor with penetrating shanks is required to reach arbitrary depth in the brain due to highly scattering and absorptive nature of the surrounding tissue. In this paper, we present an implantable optical brain interface based on a scalable complementary metal-oxide semiconductor (CMOS) process. The device incorporates 4000 photodetectors interleaved with 40 micro-LED (μLED) sites, spectral filters and an ultracompact metasurface in a sub-100-μm profile to reach single-neuron resolution at depth with minimal tissue damage. A machine learning- based optimization algorithm is used to design the metasurface with a working distance of less than 50-μm, enabling direct integration with the CMOS imager array. Adaptation of the mixing matrix through wavefront shaping generated by the 25-μm-thick metasurface allows for reconstruction of the volume above the μLED sites with high precision. The imager is able to localize multiple 10-μm fluorescent beads with frame rates up to 400 frames/sec within a field of view of 6300×200×200μm³.