Cambricon-LLM: A Chiplet-Based Hybrid Architecture for On-Device Inference of 70B LLM
By Zhongkai Yu1,2,†, Shengwen Liang1,†, Tianyun Ma3, Yunke Cai1,2, Ziyuan Nan1,2, Di Huang1, Xinkai Song1, Yifan Hao1, Jie Zhang4, Tian Zhi1, Yongwei Zhao1, Zidong Du1,5, Xing Hu1,5,∗, Qi Guo1, Tianshi Chen6
1 SKL of Processors, Institute of Computing Technology, CAS, Beijing, China
2 University of Chinese Academy of Sciences, Beijing, China
3 University of Science and Technology of China, Beijing, China
4 Peking University, Beijing, China
5 Shanghai Innovation Center for Processor Technologies
6 Cambricon Technologies Co., Ltd., China
Deploying advanced large language models on edge devices, such as smartphones and robotics, is a growing trend that enhances user data privacy and network connectivity resilience while preserving intelligent capabilities. However, such a task exhibits single-batch computing with incredibly low arithmetic intensity, which poses the significant challenges of huge memory footprint and bandwidth demands on limited edge resources. To address these issues, we introduce Cambricon-LLM, a chiplet-based hybrid architecture with NPU and a dedicated NAND flash chip to enable efficient on-device inference of 70B LLMs. Such a hybrid architecture utilizes both the high computing capability of NPU and the data capacity of the NAND flash chip, with the proposed hardware-tiling strategy that minimizes the data movement overhead between NPU and NAND flash chip. Specifically, the NAND flash chip, enhanced by our innovative in-flash computing and on-die ECC techniques, excels at performing precise lightweight on-die processing. Simultaneously, the NPU collaborates with the flash chip for matrix operations and handles special function computations beyond the flash's on-die processing capabilities. Overall, Cambricon-LLM enables the on-device inference of 70B LLMs at a speed of 3.44 token/s, and 7B LLMs at a speed of 36.34 token/s, which is over 22X to 45X faster than existing flash-offloading technologies, showing the potentiality of deploying powerful LLMs in edge devices.
Related Chiplet
- UCIe AP based 8-bit 170-Gsps Chiplet Transceiver
- UCIe based 8-bit 48-Gsps Transceiver
- UCIe based 12-bit 12-Gsps Transceiver
- 400G Transmitter Chiplet for 400G, 800G and 1.6T Pluggable Transceivers
- FPGA Chiplets with 40K -600K LUTS
Related Technical Papers
- DCRA: A Distributed Chiplet-based Reconfigurable Architecture for Irregular Applications
- RapidChiplet: A Toolchain for Rapid Design Space Exploration of Chiplet Architectures
- A Heterogeneous Chiplet Architecture for Accelerating End-to-End Transformer Models
- Muchisim: A Simulation Framework for Design Exploration of Multi-Chip Manycore Systems
Latest Technical Papers
- Complex Heterogeneous Integration Drives Innovation In Semiconductor Test
- The Evolution of Photonic Integrated Circuits and Silicon Photonics
- Flexible electronic-photonic 3D integration from ultrathin polymer chiplets
- Enabling Innovative Multi-Vendor Chiplet-Based Designs
- Cambricon-LLM: A Chiplet-Based Hybrid Architecture for On-Device Inference of 70B LLM