A Neuromorphic Brain Interface based on RRAM Crossbar Arrays for High Throughput Real-time Spike Sorting

Area of Science:

• Neurotechnology
• Neuromorphic Engineering
• Materials Science

Background:

• Real-time spike sorting is essential for brain-machine interfaces and neural prosthetics.
• High-density electrode arrays increase data demands on hardware.
• Existing hardware faces challenges in bandwidth and computational complexity.

Purpose of the Study:

• To develop a specialized real-time hardware for on-the-fly neural spike sorting.
• To achieve high throughput and minimal power consumption for spike sorting.
• To overcome limitations of current hardware implementations.

Main Methods:

• Utilized high-density copper oxide (CuO_x) resistive crossbars for in-memory spike sorting.
• Developed a CMOS BEOL-compatible fabrication process for CuO_x devices.
• Implemented a template matching algorithm directly mapped onto RRAM crossbars.
• Characterized device switching and statistical variations.

Main Results:

• Demonstrated energy-efficient spike sorting with high accuracy using synthetic and in vivo data.
• Achieved substantial improvements in area (~1000×), power (~200×), and latency (4.8μs for 100 channels).
• Showcased massively parallel processing capabilities of the RRAM crossbar architecture.

Conclusions:

• The developed neuromorphic interface offers significant advantages for real-time spike sorting.
• This technology enables more efficient and compact hardware for neural signal processing.
• The in-memory computing approach using CuO_x crossbars is a promising direction for future neural interfaces.