An Application Specific Instruction Set Processor (ASIP) for Adaptive Filters in Neural Prosthetics

Area of Science:

• Neuroscience
• Computer Engineering
• Biomedical Engineering

Background:

• Adaptive filters are crucial for neural coding in neuroprosthetic design.
• Switching between filters is necessary due to changing neuron models or system requirements.
• Existing processors like general-purpose processors and ASICs have limitations in efficiency and flexibility.

Purpose of the Study:

• To explore an application-specific instruction-set processor (ASIP) for adaptive filters in neural decoding.
• To address the computational inefficiency and inflexibility of current hardware solutions for neural prostheses.
• To develop a processor architecture balancing flexibility and throughput for adaptive filtering.

Main Methods:

• Design and implementation of a novel ASIP architecture.
• Focus on optimizing matrix/vector operations common in adaptive filters.
• Evaluation of the ASIP's performance, area efficiency, and computational capabilities.

Main Results:

• The proposed ASIP architecture demonstrates efficient computation for time-consuming matrix/vector operations.
• The ASIP provides both flexibility in filter switching and high throughput.
• Implementation results show the ASIP is area-efficient and competitive with commercial CPUs.

Conclusions:

• The developed ASIP is a promising solution for adaptive filtering in neural decoding.
• This architecture offers a flexible and computationally efficient alternative for neuroprosthetic applications.
• The ASIP design balances performance and adaptability for advanced neural prostheses.