AIG-CIM: A Scalable Chiplet Module with Tri-Gear Heterogeneous Compute-in-Memory for Diffusion Acceleration

By Yiqi Jing¹, Meng Wu¹, Jiaqi Zhou¹, Yiyang Sun¹, Yufei Ma^1,2, Ru Huang¹ , Le Ye^1,3 , Tianyu Jia^1
1 School of Integrated Circuits, Peking University, Beijing, China, ² Institute for Artificial Intelligence, Peking University, Beijing, China, ³ Advanced Institute of Information Technology of Peking University, Hangzhou, China

ABSTRACT

The emergence of Diffusion models has gained significant attention in the field of Artificial Intelligence Generated Content. While Diffusion demonstrates impressive image generation capability, it faces hardware deployment challenges due to its unique model architecture and computation requirement. In this paper, we present a hardware accelerator design, i.e. AIG-CIM, which incorporates tri-gear heterogeneous digital compute-in-memory to address the flexible data reuse demands in Diffusion models. Our framework offers a collaborative design methodology for large generative models from the computational circuit-level to the multi-chip-module system-level. We implemented and evaluated the AIG-CIM accelerator using TSMC 22nm technology. For several Diffusion inferences, scalable AIG-CIM chiplets achieve 21.3× latency reduction, up to 231.2× throughput improvement and three orders of magnitude energy efficiency improvement compared to RTX 3090 GPU.

Click here to read more ...