田山泉 Webpage

☕️

田山泉

Google

Hi!

你好！我目前在Google工作，主要专注于大语言模型（LLM）性能优化、软硬件协同设计。

我的工作涵盖 LLM 在现有平台上的优化执行、下一代 TPU 架构的设计，以及 AI 基础设施的开发。

在加入 Google 之前，我在耶鲁大学师从 Jakub Szefer 教授，在计算机架构与安全领域开展研究，并获得了耶鲁大学的博士学位。

兴趣爱好

教育经历

Software Engineer, Google
Google 三月 2023 – 现在
- AI/ML Infrastructure Performance: Conducted analysis and HW/SW co-design for current and next-generation AI accelerators to enhance Google’s large-scale distributed data center, optimizing for emerging large language models (LLMs).
- Software development for fleet reliability monitoring: Designed and implemented a system to detect and monitor silent data corruptions (SDCs) across the production fleet, improving reliability and resilience.
Software Engineer Intern, Google
Google 六月 2022 – 八月 2022
- Designed and implemented the support to run Open Container Initiative (OCI) standard containers in Borg, which is Google’s cluster manager that runs almost all jobs.
Software Engineer Intern, Alibaba Cloud
Alibaba Cloud 七月 2020 – 八月 2020
Ph.D. Student, Yale University
Yale University 九月 2017 – 二月 2023
Security of Cloud FPGAs, FPGA and hardware security, cloud infrastructures, hardware accelerator design and RTL development for ML and cryptography.

Selected Publications

Shanquan Tian, Shayan Moini, Daniel Holcomb, Russell Tessier, Jakub Szefer (2023). A Practical Remote Power Attack on Machine Learning Accelerators in Cloud FPGAs. DATE'23.

Ilias Giechaskiel (Contributed Equally), Shanquan Tian (Contributed Equally), Jakub Szefer (2022). Cross-VM Covert-and Side-Channel Attacks in Cloud FPGAs. TRETS'22.

Ilias Giechaskiel (Contributed Equally), Shanquan Tian (Contributed Equally), Jakub Szefer (2022). Cross-VM Information Leaks in FPGA-Accelerated Cloud Environments (Best Paper Candidate). HOST'22.

Shanquan Tian, Shayan Moini, Adam Wolnikowski, Daniel Holcomb, Russell Tessier, Jakub Szefer (2021). Remote power attacks on the versatile tensor accelerator in multi-tenant FPGAs (Best Paper Candidate). FCCM’21.

Blog