Neuromorphic Cochlear Implant

Abstract

The first commercial Cochlea implant was released in 1982. In the following ten years the implants were significantly improved, and has now become a common aid for individuals with severe to profound hearing loss. Briefly a cochlea implant is a hearing aid that stimulates the nerve-cells in the inner ear using electrical pulses, and it can be used if the nerve cells connected to the inner ear are reasonable intact. A central part of these implants are the processing unit which generates the electrical pulses. The main design challenges are related to generation of a meaningful real time stimulus for the inner ear using a minimum of power with simple and compact electronics.

The thesis “Neuromorphic Cochlear Implant” presents a processing unit where the design is inspired by the human nervous system. Most likely this will lead to a reduced current consumption compared to common implementation strategies used in current implants. At the same time the sound perception could be experienced as more natural, and the device could be made more compact. To support the arguments, a test unit has been implemented in microelectronics which demonstrates that such a unit could be mass produced using common technology.

A central part of the thesis focuses on nerve-cell modulation in microelectronics and usage of these models for advanced signal processing. This part of the work can open for real-time signal processing in applications where battery capacity is now a limiting factor, not only for Cochlear implants but even for general low-power electronics. Fundamental for this is a clear similarity between the most important properties of the human neurons and Delta Sigma analog to digital converters which commonly are used for conversion between for instance sound/picture and digital signals in CD/DVD players, digital radio (DAB) and television.

The above is based on a general idea used throughout the work to use biology as inspiration. Assuming evolution creates efficient solutions optimizing performance given limited resources, it is likely that the methods found in the human body could be successfully transferred to microelectronics to create really computationally efficient solutions. The thesis is strongly based on this assumption, and future applications exploring neuromorphic silicon is envisioned.