Hardware Accelerated Brain-Computer Interfaces for Real-Time Neural Decoding

Abstract

Every detail of our perception emerges from our brain. Yet, despite being under study for thousands of years (Minagar et al. 2003), many of the precise physiological mechanisms behind our experiences remain a mystery. A critical challenge in advancing our understanding of fundamental neuroscience is the ability to isolate and monitor specific structures in the brain to determine the roles they play in cognition, sensation, and behavior. In this work, we present a fully integrated closed-loop Brain-Computer Interface (BCI) system designed to support real-time communication with the brain for controlled experimentation. Traditional BCIs are unable to meet the latency constraints on the signal decoding pipeline required to react to neural activity. Our system accelerates the decoding algorithm on a Field Programmable Gate Array (FPGA) to process high-resolution neural data with low latency, and stimulates the brain using optogenetics. We demonstrate the system’s latency characteristics, marking a significant speedup over traditional CPU and GPU-based decoding pipelines, power consumption, and decoding accuracy of the quantized decoder implemented on the FPGA.