Recently, hardware creation platform Efabless announced the selection results of its first "AI Generated Open-Source Silicon Design Challenge". The RISC-V International Open Source Laboratory (RIOS Lab) team from Tsinghua University, based on the support of Amazon Cloud Technology's cloud scientific research computing platform, won the runner-up with the RISC-V-based CPU design of the entry project "CyberRio", becoming the first in the world The CPU design was implemented with Generative AI as the first contributor and successfully taped out in a fully open source manner.
This selection is organized by Efabless and is an international competition for software and hardware design engineers, enthusiasts and researchers around the world. It aims to stimulate the innovative power of artificial intelligence and machine learning in the field of IC manufacturing and promote the open source hardware community and artificial intelligence. of coordinated development. Contestants need to use generative AI to generate complete Verilog models for digital designs.
RIOS team members Wang Xinze, Yin Guohua, and Zhu Yifei successfully designed a design process for processor application scenarios based on the support of the Amazon Cloud Technology cloud scientific research computing platform and the large language model, and used this process to generate a comprehensive HDL (Hardware Description Language) , completed the five-level pipeline architecture, supported RV32I non-privileged ISA and machine mode CPU design, passed basic simulation and verification, and was finally successfully tape-out based on the 130nm process. This marks that the world's first CPU whose main design is completed by large language model (LLM) has been actually manufactured, opening up new possibilities for the application of LLM in the field of IC design.
Testimonials from RIOS Lab
Amazon Cloud Technology Innovation Support Program
Dr. Wang Xi, head of the Amazon Innovation Support Project from RIOS Lab, said: "I am very grateful to Amazon Cloud Technology for assisting in the AI computing power and innovation support program, which played a key role in our project, providing powerful computing power and rich Tool support. The use of these services enabled the team to complete complex computing tasks in a short time, thereby achieving this important technological breakthrough. Relying on this platform, the tape-out was successfully completed and won the runner-up in the relevant competition. The team also recently prepared Together with New York University, we have become the first explorers in this field. We would like to express our gratitude to Amazon Cloud Technology and believe that future cooperation will bring more innovations and breakthroughs."
Amazon Cloud Technology's scientific research cloud workbench (Service Workbench), a self-service open source platform based on cloud computing, helps RIOS researchers realize the visualization of collaborative research. Researchers can quickly find and obtain and use machine learning analysis data of interest by accessing the user interface. Service Workbench currently provides three types of data sets (Studies): personal data sets, organizational data sets and public data sets, making it easier for researchers to share and collaborate with other members of the organization. Together with the latest generation memory-optimized high-performance instance R7g equipped with DDR5 memory used in this project, it helps RIOS accelerate the processing efficiency of calculation and analysis. This instance uses a customized Graviton 3 processor based on ARM architecture. Compared with the previous generation Graviton 2 processor, the memory bandwidth, computing performance and floating point performance are improved by 50%, 25% and 200% respectively. Machine learning inference performance has been improved by 3 times, and it has been widely used in memory-intensive workloads including open source databases, memory caches, and real-time big data analysis scenarios.
The innovation support plan mentioned by Dr. Wang Xi is an Amazon Cloud Technology plan aimed at strengthening cooperation and exchanges with universities and scientific research institutions on technological innovation, including:
1) Support the construction of a cloud scientific research and R&D environment, including cloud deployment and task execution of scientific research software and R&D software;
2) Perform proof-of-concepts, benchmarks, and performance tests to evaluate the performance improvements of migrating research workloads or public data sets to the cloud;
3) Cultivate the broader community on cloud research workloads through workshops or training.
About the chip industry of Amazon Cloud Technology
Solutions and High Performance Computing (HPC)
As the world's leading cloud service provider and semiconductor industry leader, Amazon Cloud Technology not only continues to develop and mass-produce a variety of high-end process chips, but also provides a secure, agile and scalable platform, including comprehensive service integration solutions. to enable high-performance design, verification and smart manufacturing.
Facing the rise of chiplets in the field of semiconductor chip design, AI/ML technology to improve IC design efficiency and the three major trends of high-performance computing power on the cloud, Amazon Cloud Technology is based on its own multiple high-performance chip design experience and many of the world's leading chip design companies as customers Continuously release and update high-performance computing (HPC) solutions for electronic design automation (EDA) to achieve the best match. Including highly flexible automatic integrated computing solutions (SOCA) suitable for EDA design, helping customers to provide one-stop solution for semiconductor chips from UI user interface, cloud cluster pre-processing and post-processing visualization, real-time analysis, custom code and automation capabilities. Design customer pain points.
Get insights faster and get ideas and products to market quickly with nearly unlimited compute capacity, high-performance file systems, and high-throughput networking. Based on relevant Amazon cloud technology solutions and partner solutions, we have successfully helped customers such as MediaTek, ARM, NXP, and Qualcomm complete relevant cloud design workloads.