| Chip Learning：从芯片设计到芯片学习 |
| |
| 引用本文： | 陈云霁,杜子东,郭崎,李威,谭懿峻.Chip Learning：从芯片设计到芯片学习[J].中国科学院院刊,2022,37(1):15-23. |
| |
| 作者姓名： | 陈云霁 杜子东 郭崎 李威 谭懿峻 |
| |
| 作者单位： | 中国科学院计算技术研究所 计算机体系结构国家重点实验室 北京 100190;中国科学院计算技术研究所 计算机体系结构国家重点实验室 北京 100190;中国科学院计算技术研究所 计算机体系结构国家重点实验室 北京 100190;中国科学院大学 计算机科学与技术学院 北京 100049 |
| |
| 基金项目： | 国家重点研发计划（2017YFA0700902、2018AAA0103300），国家自然科学基金（61925208、61906179、62002338、62102399），北京市自然科学基金（JQ18013），北京智源人工智能研究院及北京市科技新星计划项目（Z191100001119093），中国科学院稳定支持基础研究领域青年团队计划（YSBR-029），中国科学院青年创新促进会（2019105），“科学探索奖” |
| |
| 摘 要： | 芯片是现代信息社会的关键基础设施,未来人机物三元融合的智能万物互联时代将需要大量不同种类的专用体系结构芯片。然而,芯片设计本身代价很高,具有设计周期长、过程非常复杂、专业门槛高的特点。因此,智能万物互联时代芯片需求多和芯片设计代价高之间产生了巨大的矛盾。文章提出利用芯片学习(Chip Learning)来取代芯片设计以解决上述矛盾,即采用学习的方法来完成芯片从逻辑设计到物理设计的全流程。芯片学习的目标是通过学习使得芯片设计完全不需要专业知识和设计经验,可以在短时间、无人参与的情况下高效完成。
|
| 关 键 词： | 芯片设计 人工智能 芯片学习 设计自动化 |
| 收稿时间： | 2021/12/21 0:00:00 |
From Chip Design to Chip Learning |
| |
| CHEN Yunji,DU Zidong,GUO Qi,LI Wei,TAN Yijun.From Chip Design to Chip Learning[J].Bulletin of the Chinese Academy of Sciences,2022,37(1):15-23. |
| |
| Authors: | CHEN Yunji DU Zidong GUO Qi LI Wei TAN Yijun |
| |
| Institution: | State Key Laboratory of Computer Architecture, Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China;State Key Laboratory of Computer Architecture, Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China; State Key Laboratory of Computer Architecture, Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China;School of Computer Science and Technology, University of Chinese Academy of Sciences, Beijing 100049, China |
| |
| Abstract: | Chip is the foundation of the modern information society. As the world is entering a new era of human-cyber-physical ternary computing, with diverse intelligent applications over trillions of devices, chip with specialized architecture will be heavily demanded in both numbers and types. However, chip design is very costly, which usually requires a long design cycle, complicated process, and high professional developers. Hence, there is a large gap between the need of tremendous chips and the high cost of chip design in the new era. This study proposes Chip Learning, a learning-based method to perform the entire chip design, including logic design, circuit design, and physical design. As an alternate to chip design, Chip Learning aims to remove the barrier of professional knowledge and experiences for effective chip design with a no-human-in-the-loop process in a short time. |
| |
| Keywords: | chip design artificial intelligence chip learning design automation |
|
| 点击此处可从《中国科学院院刊》浏览原始摘要信息 |
| 点击此处可从《中国科学院院刊》下载免费的PDF全文 |
|
