Skip to content

LSI-Contest 2026: Generative Adversarial Networks（GANοΌ‰

  • Title: SpikeGAN: An Energy-Efficient Spiking Generative Adversarial Network Design
  • Members: Yuga Hanyu, Atharv Sharma and Aruki Komatsuzaki
  • Abstract: The use of generative AI in our society has been increasing rapidly. At the same time, generative tasks, such as image generation, are computationally expensive, leading to major carbon emissions due to powering the machines. However, there is a possibility to reduce carbon emissions from generative AI by adapting Spiking Neural Networks (SNNs), thanks to their event-driven nature and energy efficiency. In this study, we developed a framework that converts a Generative Adversarial Neural (GAN) Network consisting of Artificial Neural Networks (ANNs) into an SNN, which can be deployed on an FPGA and implemented in ASIC 45nm CMOS technology. The output images from software and RTL simulations show great correspondence (3.02e-05 to 1.00e-04 MSE values).