Research map / Architectures and efficiency / Spiking networks
HardF-SNN: Hardware-Friendly Quantization for Spiking Neural Networks with Efficient Integer-Arithmetic-Only Inference
Hanwen Liu, Kexin Shi, Jieyuan Zhang, Yimeng Shan, Jibin Wu, Wenyu Chen, Malu Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add proportional shared-scale quantization (PSSQ) to the baseline (ablation) | HardF-SNN baseline (ResNet19, 4w-4u) · CIFAR-100 | accuracy (%): 73.15 -> 74.93 | Improved |
| add proportional shared-scale quantization (PSSQ) to the baseline (ablation) | HardF-SNN baseline (ResNet19, 8w-8u) · CIFAR-100 | accuracy (%): 74.82 -> 75.62 | Improved |
| add integer-only BN (I-BN) to the baseline (ablation) | HardF-SNN baseline (ResNet19, 8w-8u) · CIFAR-100 | accuracy (%): 74.82 -> 75.04 | Improved |
| add integer-only BN (I-BN) to the baseline (ablation) | HardF-SNN baseline (ResNet19, 4w-4u) · CIFAR-100 | accuracy (%): 73.15 -> 74.61 | Improved |
| add integer-only BN so 2-bit training converges (ablation) | HardF-SNN baseline (ResNet19, 2w-2u) · CIFAR-100 | trainability / accuracy (%): untrainable -> 73.93 | Improved |
| add PSSQ on top of I-BN (ablation) | HardF-SNN (ResNet19, 2w-2u) · CIFAR-100 | accuracy (%): 73.93 -> 74.22 | Improved |
| add both PSSQ and I-BN (ablation) | HardF-SNN baseline (ResNet19, 4w-4u) · CIFAR-100 | accuracy (%): 73.15 -> 75.35 | Improved |
| use power-of-two learnable eta instead of rounded max-abs eta (ablation) | PSSQ (ResNet19) · CIFAR-100 | accuracy (%) | No clear effect |
| reduce bit-width from 8w-8u to 2w-2u (ablation) | HardF-SNN (ResNet18) · ImageNet-1k | accuracy (%): 59.52 -> 57.71 | Hurt |
| reduce bit-width from 8w-8u to 2w-2u (ablation) | HardF-SNN (ResNet19) · CIFAR-10 | accuracy (%): 95.47 -> 94.49 | Hurt |
| lower FPGA precision from 8-bit to 2-bit (ablation) | HardF-SNN accelerator (VGG11) · CIFAR-100 FPGA | accuracy (%) / throughput (GOPS): 67.26 -> 65.49 accuracy; 297.44 -> 993.16 GOPS | Mixed |
Closest papers by meaning
- QP-SNN: Quantized and Pruned Spiking Neural Networks (ICLR 2025 · similarity 0.92)
- S$^2$NN: Sub-bit Spiking Neural Networks (NeurIPS 2025 · similarity 0.91)
- Are Conventional SNNs Really Efficient? A Perspective from Network Quantization (CVPR 2024 · similarity 0.91)
- Inference-Scale Complexity in ANN-SNN Conversion for High-Performance and Low-Power Applications (CVPR 2025 · similarity 0.90)
- Sorbet: A Neuromorphic Hardware-Compatible Transformer-Based Spiking Language Model (ICML 2025 · similarity 0.88)
- Faster and Stronger: When ANN-SNN Conversion Meets Parallel Spiking Calculation (ICML 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
