Search papers, labs, and topics across Lattice.
This paper introduces a silicon photonics-based accelerator designed to improve the energy efficiency and throughput of diffusion models. The accelerator targets the computationally intensive UNets and attention mechanisms within diffusion models, which typically suffer from high inference energy on electronic platforms. Experimental results show a 3x improvement in energy efficiency and a 5.5x increase in throughput compared to existing diffusion model accelerators.
Diffusion models can now run with 3x better energy efficiency and 5.5x higher throughput thanks to a silicon photonics accelerator.
Diffusion models have revolutionized generative AI, with their inherent capacity to generate highly realistic state-of-the-art synthetic data. However, these models employ an iterative denoising process over computationally intensive layers such as UNets and attention mechanisms. This results in high inference energy on conventional electronic platforms, and thus, there is an emerging need to accelerate these models in a sustainable manner. To address this challenge, we present a novel silicon photonics-based accelerator for diffusion models. Experimental evaluations demonstrate that our photonic accelerator achieves at least 3x better energy efficiency and 5.5x throughput improvement compared to state-of-the-art diffusion model accelerators.
