Research: Design and Evaluation of Low-Power Spiking Neural Network Hardware for Event-...

Research question: How can a neuromorphic hardware architecture based on spiking neural networks be designed and optimized for efficient real-time processing of event-based vision data?

Neuromorphic computing leverages spiking neural networks (SNNs) to emulate the event-driven, energy-efficient computation of biological brains. In event-based vision, sensors such as dynamic vision sensors (DVS) generate sparse, asynchronous data streams, which pose unique challenges for traditional frame-based processing and offer significant advantages in latency and power consumption for real-time applications.

Despite recent advances, there is a lack of systematic research into hardware-accelerated SNN architectures tailored specifically for event-based vision workloads. Most existing studies focus on either algorithmic development or software simulation, leaving a gap in understanding practical, low-power VLSI implementations and their trade-offs with accuracy and efficiency.

This project will review state-of-the-art SNN models and neuromorphic hardware, select model architectures suitable for event-based vision, and design hardware prototypes using HDL or FPGA platforms. The approach includes benchmarking against public DVS datasets, measuring latency and energy consumption, and analyzing accuracy for vision tasks such as object recognition.

Success will contribute to the emerging field of intelligent edge devices by informing design principles for robust, low-power sensory processing hardware. This research is crucial for applications in robotics, surveillance, and autonomous systems where real-time, energy-efficient computation is essential.