Education

• 2022 Sept - exp. 2027

  University of Michigan

  Ph.D in Computer Science and Engineering | Computer Architecture & Systems
• 2022 Apr

  University of Michigen

  B.S.E. in Computer Engineering

  Summa Cum Laude

• 2022 Aug

  Shanghai Jiaotong Univeristy

  B.S.E. in Electrical & Computer Engineering

Industry

• 2024 May - 2024 Aug

  Meta

  Research Scientist Intern | PyTorch Team

  Build flexible and efficient attention programming model: FlexAttention.
  Work with TorchInductor and conduct performance analysis and optimizations on attention kernels.

• 2022 May - 2022 Aug

  NVIDIA

  Deep Learning Compute Architect Intern | GPU Architecture

  Model and analyze new memory features on next-gen GPUs such as distributed shared memory and TMA.
  Specialize in: GPU architecture, memory hierarchy & multi-device communication

• MLSys ‘25

  FlexAttention: The Flexibility of PyTorch with the Performance of FlashAttention

  Joy Dong *, Boyuan Feng *, Driss Guessous *, Yanbo Liang *, Horace He

  *authors contributed equally to this work.
  [poster] [arxiv] [blog] [github] [citeme]

  

  FlexAttention is a novel compiler-driven programming model for flexible and efficient attention variants implementation.
  🌟Flexible: Allow users to implements majority of attention variants in a few lines of idomatic PyTorch code.
  🌟Fast & Efficient: Achive comparable performance to expert tuned kernels via JIT torch.compile.
  🌟Block Sparsity: Leverages block sparsity to further improve performance without manual optimization for a specific mask.

• ASPLOS ‘24

  Toleo: Scaling Freshness to Tera-scale Memory Using CXL and PIM

  Juechu Dong, Jonah Rosenblum, Satish Narayanasamy

  We will present Toleo at ASPLOS'25!
  [arxiv] [github] [citeme]

  

  🌟Scale trusted memory size from hundreds of MB to tens of TB by expanding the span of trusted from a single trusted processor to an entire platform including intelligent memories.
  🌟Design a new scheme of freshness protection that reduces the space requirement by 50x.
  🌟Reduce deployment cost by spacing sharing one intelligent memory device among multiple CPUs.

• Nature Computer Science – under submission

  SECRET-GWAS: Confidential Computing for Population-Scale GWAS

  Jonah Rosenblum, Juechu Dong, Satish Narayanasamy

  [preprint] [code] [citeme]

  Develop a thousand-core platform on Azure Confidential Computing to conduct multi-institutional GWAS on millions of patients in less than a minute.
  Adapt Spark-based Hail genomic analysis framework to run on TEE under obliviousness requirement.
  Parallelize GWAS computation on 1k cores to achieve near linear speedup.

• ACM BCB ‘24

  mm2-gb: GPU Accelerated Minimap2 for Long Read DNA Mapping

  Juechu Dong *, Xueshen Liu *, Harisankar Sadasivan, Sriranjani Sitaraman, Satish Narayanasamy

  *both authors contributed equally to this work.
  [paper] [github] [slides] [blog] [citeme]

  

  Performance Boost: Accelerate bottleneck step (chaining) of state-of-art long sequence mapping tool minimap2 by 2.57x-5.33x on GPU.
  Scales well: Optimize towards ultra long reads of 50kb+ to accommodate genome sequencing technology trend.
  Open Sourced! with active maintainance and optimization! Welcome community contributions~

Leadership

• Coordinator

  Computer Engineering Lab Reading Group

  Organize weekly paper reading presentations and discussions.
  Host talks from visiting researchers and professors.

• Co-Founder & Co-President

  UM-SJTU Joint Institute Alumni Association

  Alumni Engagement: Organize alumni and student gatherings.
  Relationship Building: Involve in expanding SJTU - UM collaborations, connecting to JI sponsors, and building industry relationships.
  Career Advising: Organize students career development workshops.
  Welcoming: Host new student orientation events, organize airport pickups, and offer settle down help.
  Student Support: Support students during the stressful transition to start in a new university in a new country, and during urgent crisis.

  

Teaching

• WN2024

  Graduate Student Instructor: EECS570 Parallel Computer Architecture

  with Prof. Ronald Dreslinski @UMich
• FA2023

  Graduate Student Instructor: EECS471 CUDA Programming

  with Dr. Valeriy Tenishev @UMich
• FA2021, WN2022

  Instructional Aid: EECS470 Computer Architecture

  with Prof. Mark Brehob and Prof. Ronald Dreslinski @UMich

  Teach out of order processor design topics including branch prediction, pipelines, prefetching, caches etc. Hold lab sessions and develop exam problems regarding OoO processor design.

• SP2021

  Teaching Assistant: VE401 Probabilistic Methods in Eng.

  Instructor Dr. Horst Hohberger @SJTU-UM Joint Institute
• SU2020

  Teaching Assistant: VP260 Honors Physics

  Instructor Dr. Mateusz Krzyzosiak @SJTU-UM Joint Institue.