ShiftLIF: Power-of-Two Multi-Level Spiking Neurons for Efficient Edge AI
Researchers propose ShiftLIF, a novel multi-level spiking neuron that uses power-of-two quantization to enhance representational capacity in spiking neural networks (SNNs) while maintaining hardware efficiency. Standard LIF neurons transmit only binary spikes, limiting information. ShiftLIF maps membrane potentials to logarithmically spaced spike levels, providing finer resolution in the small-amplitude regime where potentials concentrate. This design eliminates costly multiplications by enabling bit-shift and accumulation operations, making it suitable for edge sensing applications. The paper is available on arXiv (2605.01866).
Key facts
- ShiftLIF is a multi-level spiking neuron using power-of-two quantization.
- It improves representational capacity over binary LIF neurons.
- It provides finer resolution in the small-amplitude regime.
- It enables multiplier-free synaptic computation via bit-shift and accumulation.
- The paper is on arXiv with ID 2605.01866.
- SNNs are promising for edge sensing due to event-driven computation.
- Existing multi-level neurons often use uniform quantization or costly multiplications.
- ShiftLIF targets hardware-friendly edge AI applications.
Entities
Institutions
- arXiv