A silicon carbide array for electrocorticography and peripheral nerve recording.

OBJECTIVE: Current neural probes have a limited device lifetime of a few years. Their common failure mode is the degradation of insulating films and/or the delamination of the conductor-insulator interfaces. We sought to develop a technology that does not suffer from such limitations and would be suitable for chronic applications with very long device lifetimes. APPROACH: We developed a fabrication method that integrates polycrystalline conductive silicon carbide with insulating silicon carbide. The technology employs amorphous silicon carbide as the insulator and conductive silicon carbide at the recording sites, resulting in a seamless transition between doped and amorphous regions of the same material, eliminating heterogeneous interfaces prone to delamination. Silicon carbide has outstanding chemical stability, is biocompatible, is an excellent molecular barrier and is compatible with standard microfabrication processes. MAIN RESULTS: We have fabricated silicon carbide electrode arrays using our novel fabrication method. We conducted in vivo experiments in which electrocorticography recordings from the primary visual cortex of a rat were obtained and were of similar quality to those of polymer based electrocorticography arrays. The silicon carbide electrode arrays were also used as a cuff electrode wrapped around the sciatic nerve of a rat to record the nerve response to electrical stimulation. Finally, we demonstrated the outstanding long term stability of our insulating silicon carbide films through accelerated aging tests. SIGNIFICANCE: Clinical translation in neural engineering has been slowed in part due to the poor long term performance of current probes. Silicon carbide devices are a promising technology that may accelerate this transition by enabling truly chronic applications.

Sub-mm functional decoupling of electrocortical signals through closed-loop BMI learning.

Volitional control of neural activity lies at the heart of the Brain-Machine Interface (BMI) paradigm. In this work we investigated if subdural field potentials recorded by electrodes < 1mm apart can be decoupled through closed-loop BMI learning. To this end, we fabricated custom, flexible microelectrode arrays with 200 µm electrode pitch and increased the effective electrode area by electrodeposition of platinum black to reduce thermal noise. We have chronically implanted these arrays subdurally over primary motor cortex (M1) of 5 male Long-Evans Rats and monitored the electrochemical electrode impedance in vivo to assess the stability of these neural interfaces. We successfully trained the rodents to perform a one-dimensional center-out task using closed-loop brain control to adjust the pitch of an auditory cursor by differentially modulating high gamma (70-110 Hz) power on pairs of surface microelectrodes that were separated by less than 1 mm.

A comparison of optimal MIMO linear and nonlinear models for brain-machine interfaces.

The field of brain-machine interfaces requires the estimation of a mapping from spike trains collected in motor cortex areas to the hand kinematics of the behaving animal. This paper presents a systematic investigation of several linear (Wiener filter, LMS adaptive filters, gamma filter, subspace Wiener filters) and nonlinear models (time-delay neural network and local linear switching models) applied to datasets from two experiments in monkeys performing motor tasks (reaching for food and target hitting). Ensembles of 100-200 cortical neurons were simultaneously recorded in these experiments, and even larger neuronal samples are anticipated in the future. Due to the large size of the models (thousands of parameters), the major issue studied was the generalization performance. Every parameter of the models (not only the weights) was selected optimally using signal processing and machine learning techniques. The models were also compared statistically with respect to the Wiener filter as the baseline. Each of the optimization procedures produced improvements over that baseline for either one of the two datasets or both.

Simultaneus prediction of four kinematic variables for a brain-machine interface using a single recurrent neural network.

Implementation of brain-machine interface neural-to-motor mapping algorithms in low-power, portable digital signal processors (DSPs) requires efficient use of model resources especially when predicting signals that show interdependencies. We show here that a single recurrent neural network can simultaneously predict hand position and velocity from the same ensemble of cells using a minimalist topology. Analysis of the trained topology showed that the model learns to concurrently represent multiple kinematic parameters in a single state variable. We further assess the expressive power of the state variables for both large and small topologies.

Artificial ears for a biomimetic sonarhead: from multiple reflectors to surfaces.

This work presents an evolutionary approach to pinna design. Narrowband echolocating bats move the pinna to alter the directional sensitivity of their perceptual systems. Adding pinnae to RoBat--a biomimetic sonarhead mounted on a mobile robot--is the goal of this work. After a description of the earlier work on artificial pinnae consisting of multiple reflectors around the transducer, an acoustic model, inspired by a physical model of sound diffraction and reflections in the human concha, is described and revisited as the model to use for evolving complex shapes. A genetic algorithm evolved the shape of the pinnae with respect to desired features of the directivity pattern of the receiver transducers. Some interesting paraboloid shapes for specific echolocating behaviors were evolved, improving performance with respect to the bare transducer's performance.