About the role#
The AGI Computing Lab at Samsung Semiconductor is focused on solving system-level challenges for future AI and machine learning workloads. This internship involves researching collective communication algorithms for large-scale AI systems that operate on unreliable hardware. You will work with a team to design scalable platforms that balance computational demands, memory requirements, and energy efficiency.
What you'll need#
- Pursuing a master's or PhD in Computer Science or Electrical Engineering.
- Research experience in collective communications, interconnection networks, or distributed ML systems.
- Experience developing simulators for HPC systems or networks, such as ASTRA-sim, ns-3, or SST.
- Understanding of fault-tolerance techniques like error detection, checkpointing, and recovery.
- Understanding of system-level characteristics of distributed LLM training and inference workloads.
- Familiarity with PyTorch or JAX, and communication libraries such as NCCL, RCCL, or MPI.
- Strong analytical, problem-solving, and interpersonal skills.
- Ability to work independently and as part of a team.
- Published papers in Microarchitecture, Systems, Networking, or HPC conferences are preferred.
What you'll do#
- Research collective communication algorithms capable of tolerating link, switch, and accelerator failures.
- Design techniques to detect and mitigate stragglers, fail-slow links, and silent data corruption.
- Extend analytical and network simulators with fault models to quantify how failures impact performance.
- Explore hardware and software co-design of resilient interconnects, including link-level retry and multipathing.
- Collaborate with compiler and runtime teams to integrate new algorithms into the software stack.
- Contribute to architecture research and write research papers.
- Stay current with trends in ML architectures and interconnects.
Location & details#
- This role is based on-site at our San Jose, California headquarters.
- This is a full-time, paid internship with a rolling term.

