High-density microelectrode arrays (HDMEAs) feature thousands of recording electrodes in a single chip with an area of few square millimeters. The obtained electrode density is comparable and even higher than the typical density of neuronal cells in cortical cultures. Commercially available HDMEA-based acquisition systems are able to record the neural activity from the whole array at the same time with submillisecond resolution. These devices are a very promising tool and are increasingly used in neuroscience to tackle fundamental questions regarding the complex dynamics of neural networks. Even if electrical or optical stimulation is generally an available feature of such systems, they lack the capability of creating a closed-loop between the biological neural activity and the artificial system. Stimuli are usually sent in an open-loop manner, thus violating the inherent working basis of neural circuits that in nature are constantly reacting to the external environment. This forbids to unravel the real mechanisms behind the behavior of neural networks. The primary objective of this PhD work is to overcome such limitation by creating a fullyreconfigurable processing system capable of providing real-time feedback to the ongoing neural activity recorded with HDMEA platforms. The potentiality of modern heterogeneous FPGAs has been exploited to realize the system. In particular, the Xilinx Zynq All Programmable System on Chip (APSoC) has been used. The device features reconfigurable logic, specialized hardwired blocks, and a dual-core ARM-based processor; the synergy of these components allows to achieve high elaboration performances while maintaining a high level of flexibility and adaptivity. The developed system has been embedded in an acquisition and stimulation setup featuring the following platforms: • 3·Brain BioCam X, a state-of-the-art HDMEA-based acquisition platform capable of recording in parallel from 4096 electrodes at 18 kHz per electrode. • PlexStim™ Electrical Stimulator System, able to generate electrical stimuli with custom waveforms to 16 different output channels. • Texas Instruments DLP® LightCrafter™ Evaluation Module, capable of projecting 608x684 pixels images with a refresh rate of 60 Hz; it holds the function of optical stimulation. All the features of the system, such as band-pass filtering and spike detection of all the recorded channels, have been validated by means of ex vivo experiments. Very low-latency has been achieved while processing the whole input data stream in real-time. In the case of electrical stimulation the total latency is below 2 ms; when optical stimuli are needed, instead, the total latency is a little higher, being 21 ms in the worst case. The final setup is ready to be used to infer cellular properties by means of closed-loop experiments. As a proof of this concept, it has been successfully used for the clustering and classification of retinal ganglion cells (RGCs) in mice retina. For this experiment, the light-evoked spikes from thousands of RGCs have been correctly recorded and analyzed in real-time. Around 90% of the total clusters have been classified as ON- or OFF-type cells. In addition to the closed-loop system, a denoising prototype has been developed. The main idea is to exploit oversampling techniques to reduce the thermal noise recorded by HDMEAbased acquisition systems. The prototype is capable of processing in real-time all the input signals from the BioCam X, and it is currently being tested to evaluate the performance in terms of signal-to-noise-ratio improvement.
|Titolo della tesi:||Exploiting All-Programmable System on Chips for Closed-Loop Real-Time Neural Interfaces|
|Data di discussione:||2-mag-2019|
|Appare nelle tipologie:||Tesi di dottorato|