Computing with Spikes: The Advantage of Fine-Grained Timing

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Neural-inspired computing utilizes spike-based communication for potential energy and time efficiency.
• Fundamental questions persist regarding the quantifiable advantages and optimal application scenarios for spike-based computation compared to conventional methods.
• Directly translating existing algorithms to a spike-based medium does not inherently guarantee performance gains.

Purpose of the Study:

• To investigate and demonstrate the performance advantages of spike-based communication and computation within algorithms.
• To identify specific circumstances where spike-based approaches offer a comparative advantage over traditional computing methods.
• To present novel spiking algorithms for fundamental computational tasks and an image processing application.

Main Methods:

• Development and analysis of several spiking algorithms for core computational operations.
• Implementation of a spiking median-filtering algorithm for image processing.
• Comparative analysis of throughput and energy efficiency for spike-based versus conventional approaches.

Main Results:

• Demonstrated that spike-based communication and computation can increase throughput.
• Showcased cases where spike-based algorithms decrease energy consumption.
• Presented successful spiking algorithms for sorting, finding extrema, and median calculations.
• Developed a low-energy, parallel spiking median-filtering approach for image processing.

Conclusions:

• Spiking algorithms can offer significant performance advantages in specific computational contexts.
• Efficient computation of fundamental operations using spiking mechanisms is achievable.
• The presented algorithms and analyses support the utility of spike-based computing for complex applications.