Neural interfaces consist of systems to observe and manipulate neural activity, such as electrophysiological recording and stimulation. They have enormous potential for experimental neuroscience, clinical diagnostics and therapies, and tool use. In this dissertation, we develop systems and methods for utilizing ultrasonic wireless powering and communication for ultra-miniaturized wireless neural implants. We present a fully realized system for a stimulating device and methods for recording devices.
We present the design, realization, and in vivo validation of a 1.7 mm^3 all-wireless neural stimulation system capable of current-controlled stimulation and pulse repetition frequency up to 2.2 kHz, and demonstrate controlled peripheral nerve recruitment in vivo. This demonstrates the feasibility of ultrasonic wireless powering sufficient for high frequency and moderate charge delivery for stimulation purposes. It demonstrates backscatter uplink in a digital form and demonstrates the feasibility of this functionality in a very small package.
We develop methods to improve the capabilities of ultrasonic wireless recording devices and demonstrate those in simulation and in vitro. In order to achieve extremely small size in wireless recording devices, a previously developed backscatter impedance modulation scheme is employed, which enables simple, small, low-power devices which operate in pulse-mode. We study the challenges of this system, which include relatively low modulation sensitivity, discriminating multiple devices, separating sensor modulation from motion artifact, and detecting backscatter modulation through a skull. We present methods to overcome these challenges.
We develop a signal model of ultrasonic backscatter which first identifies features of backscatter along each backscatter pulse. We then model the backscatter as a DC offset at each sampling point with linear modulation at each sampling point imparted by the input signal. With this model, we can separate out the `DC' effect from the modulation effect. Once we isolate the modulation effect, solving for multiple sensors becomes a matrix decomposition problem. While this matrix decomposition problem is underconstrained, we can use certain statistical properties of the data set, for example, that neural spikes are sparse in time and our sensors are sparse in space, to put priors on solving the underconstrained matrix decomposition problem. We demonstrate the use of this to extract modulation signal simultaneously from multiple sensors through an ex vivo macaque skull. Furthermore, we show that using a robust sensor position detection algorithm, it is possible to build a model of backscatter DC offset and modulation which incorporates sensor position information, and fit this model from large amounts of online backscatter data. We demonstrate in a simulation that this method shows promise for separating motion artifact from sensor modulation signal in the same frequency band.
This work demonstrates the utility of ultrasonic backscatter powering and communication for ultra-miniaturized implants and paves the way for improving the performance of recording devices thus enabling a truly free-floating distributed wireless neural interface. This type of neural interface is anticipated to open new avenues for neural stimulation, for example deep brain stimulation, as well as recording across disparate neuroanatomical areas, for example a brain-machine interface which acquires information from both motor and speech areas. This work presents early versions of solutions to some of the challenges needed to achieve recording devices that are small enough for embedding in the cortex at high densities. This work is aimed to help foster commercializable neural interface implementations which improve the utility-to-risk ratio and increase the size of the patient population which can benefit from this technology.
Updates and errata to this dissertation can be found at: https://github.com/piech/piech_dissertation_notes.
