How do we make future neurally integrated prosthetic devices speak the same language as the nervous system

It is postulated that future brain-machine interfaces (BMI) will not only be used to infer the host's intentions from brain activities (e.g. motor decode) or to provide the host with extrinsic information (e.g. sensory feedback), but may provide extensions, enhancements and replacement of parts of the brain itself. For example, the BMI implant may replace injured parts of the brain after traumatic brain injuries (TBI), extend the host's memory with a hippocampus implant, provide enhanced visual capabilities by allowing the host to see different parts of the optical spectrum, or provide bat-like echolocation capabilities to blind humans. These applications of BMI are clearly in the realm of science fiction at this point and time, however, progress is being made to model and mimick parts of the brain so that we are getting closer to making these dreams a reality. We assert that such BMI devices will not be truly realized until the electronics “speak the same language” as the biological brain. That is, the signaling interface between the electronic and biological systems must match and the computation performed by the electronic device must also be implemented similarly to biological computation. Hence, the complete system must be neuromorphic, implying that it must replicate both the form and function of the brain [1]. This is a departure from the current approach in the BMI community, which wraps translational circuitry (neural amplifies and biphasic current stimulators) around traditional digital CPU based processing when interfacing with the nervous system. Our approach will still make use of the interface circuits, however, the computation will be performed with analogs of neurons, synapses, axons and dendrites in silicon. As new methods for interfacing to the nervous system emerge (i.e. carbon nano-tubes and memristers [2]), the interface circuits will also be replaced with more biologically plausible versions. This abstract summarizes the computation inside future neural prosthesis systems that take the neuromorphic engineering approach.