Technological and Computational Approaches for Large Count High-Density Neural Probes

Abstract

Implantable neural probes with various shapes, designs, and materials are extensively used to study the brain by recording the electrical and chemical responses of neural structures and circuits. This thesis addresses four critical challenges in the development of advanced neural interfaces aimed at mapping large collections of neurons. The first challenge is to develop innovative technological methods for microfabricating high-count, high-density probes with user-defined features such as density, size, shape, and distribution for specific applications. A new class of silicon-based two-dimensional (2D) planar neural probes is developed incorporating at least four shanks, each featuring over 16 recording sites with high density (320 electrodes/mm2) and narrow vertical (6µm) and horizontal (10.5µm) separation between sites. Each shank is as narrow as 43µm and as long as 5mm. Furthermore, a new class of high-count, high-density three-dimensional (3D) non-planar silicon-based neural interfaces has been developed, allowing over 10,000 slender shanks each supporting a recording/stimulation site at its tip, and providing design flexibility in array size, density (400 electrodes/mm2), and distribution. The second challenge is the engineering of the electrical and mechanical features of individual probe shanks to make them minimally invasive and suitable for long-term measurements. This involves optimizing the probe shank design to be more mechanically compliant to reduce foreign body response while ensuring it is sufficiently stiff to be implanted and reach the targeted region with minimal buckling and without the need for large mechanical insertion shuttles. The proposed planar 2D probes have T-shaped and π-shaped cross-sections, with a top side of 43µm wide and 3µm thick, and vertical stiffeners that are 4µm wide and 10µm thick. These new geometries reduce the shank cross-section and volume by ~3 times compared to conventional thick rectangular cross-section shanks while providing similar mechanical stiffness. The proposed 3D non-planar silicon-based needle arrays have been manufactured with needles ranging in length from 0.5-1.5mm, diameters from 10-15µm, and with sharp tips < 2µm wide. Three advanced technologies are proposed to create well-defined recording and stimulating sites at the tips of these needles. Electrode robustness, insertion and recording functionality both acute and long-term have been demonstrated by mechanical and electrical in vitro tests. The third challenge involves packaging and integration. Flexible, miniaturized, and robust Parylene-C and Polyimide cables, which are more than 20mm long, ~ 1mm wide, 5-15µm thick, and carry hundreds of interconnect lines, are monolithically integrated with the 2D and 3D probes. This integration significantly improves the ease of handling and reduces mechanical tethering on probes during measurements. Especially-designed Polyimide cables are integrated with large-count 3D arrays. They are solder attached directly to external connectors or readout electronics, thus avoiding the need for traditional wire bonding. This innovative approach significantly reduces the complexity and labor associated with external connections for large-count electrode arrays, thus making the packaging process faster, more efficient, and more reliable. Finally, high-count neural recording presents computational challenges, particularly in spike-sorting. A new Python Toolbox that calculates a dynamic version of the L-ratio as a quality metric for assessing cluster isolation as well as providing additional temporal information is developed. By leveraging this new information, a neural network learning algorithm to automatically curate the spike-sorting results has been trained. This approach achieved 50-100% higher F1-score on average compared to when we trained the same algorithm with the classic L-ratio.