C-DNN and C-Transformer: Mixing ANNs and SNNs for the Best of Both Worlds

Sangyeob and his team have developed a C-DNN processor that effectively processes object recognition workloads, achieving 51.3% higher energy efficiency compared to the previous state-of-the-art processor. Subsequently, they have applied C-DNN not only to image classification but also to other applications, and have developed the C-Transformer, which applies this technique to a Large Language Model (LLM). As a result, they demonstrate that the energy consumed in LLM can be reduced by 30% to 72% using the C-DNN technique, compared to the previous state-of-the-art processor. In this talk, we will introduce the processor developed for C-DNN and C-Transformer, and discuss how neuromorphic computing can be used in actual applications in the future.