Spiking Neural Networks (SNNs) are widely used in brain-inspired computing and neuroscience research. Several many-core accelerators have been built to improve the running speed and energy efficiency of SNNs. However, current accelerators generally need explicit synchronization among all cores after each timestep of SNNs, which poses a challenge to overall efficiency. This paper proposes DepAsync, an asynchronous architecture that eliminates inter-core synchronization, facilitating fast and energy-efficient SNN inference with commendable scalability. The main idea is to exploit the dependency of neuromorphic cores predetermined at compile time. We design a DepAsync scheduler for each core to trace the running state of its dependencies and control the core to safely forward to the next timestep without waiting for other cores to complete their tasks. This approach prevents the necessity for global synchronization, allowing DepAsync to minimize core waiting time facing inherent core and time imbalance in SNN workloads. The comprehensive evaluations using five SNN workloads show that DepAsync achieves 2.47x speedup and 1.55x energy efficiency compared to the state-of-the-art synchronization architectures.

The performance of visible light cameras in complex environments is often subject to many limitations, especially in the night scene, the imaging effect is significantly reduced, and it is difficult to meet the practical application requirements. In contrast, infrared imaging has become an effective complement to visible light cameras due to its good concealment and day and night capabilities. However, infrared imaging has an insufficient ability to express semantic information and low colour contrast, which limits its wide application to some extent. Consequently, we propose CDS-GAN, an unpaired infrared-to-visible (IR-to-VI) image translation method that uses a pre-trained ViT to extract marker embeddings. Our method preserves both coarse-grained and fine-grained structures via a novel layer-wise cross-domain regional similarity matching technique, which constrains the generation process at both global and local levels. We introduce identity loss to further constrain the generator, thus enhancing its ability to fit the visible image domain distribution. Compared with the SOTA method, CDS-GAN improves 15.5% and 28.3% in FID and KID metrics, respectively. Qualitative analyses also show that CDS-GAN exhibits significant advantages in IR-to-VI image translation tasks. To further promote the research in this field, we also construct and make public a challenging IR-to-VI translation dataset, CIVC, which is larger in size and richer in scenarios than the existing mainstream datasets, and can provide more powerful support for future research. CIVC can be obtained from https://drive.google.com/file/d/1bmimHWxNTW9hXebFlr4ZGk6wA0rteY5R/view?usp=sharing.