Advanced Memory Technology
Next-generation memory systems for NNSA leadership-class supercomputers
The Advanced Memory Technology (AMT) program is a tri-lab initiative led by Sandia National Laboratories in partnership with Lawrence Livermore and Los Alamos National Laboratories, sponsored by the NNSA Advanced Simulation and Computing (ASC) program (NA-114). The program addresses a decades-long disparity between processor performance and memory capability — memory bandwidth has increased at an order of magnitude less per year than compute — by collaborating directly with industry to develop next-generation memory technologies targeting future NNSA HPC and AI system procurements.
Andrew serves as Program Manager for Sandia’s Advanced Memory Technology activities, overseeing research into emerging memory systems and their integration into future leadership-class supercomputers. The AMT program lead is James H. Laros III.
The program’s goal is a 40x application performance improvement over current exascale systems. Key industry partnerships include:
-
Intel — R&D on re-thinking fundamental DRAM organization and coupling with compute platforms to achieve orders-of-magnitude performance gains from basic memory design.
-
AMD — Collaborative R&D on advanced memory technologies to accelerate high-performance simulation applications in support of the stockpile stewardship mission.
-
Cerebras Systems — Work was named a Gordon Bell Prize finalist after achieving 699 kilosteps per second in molecular dynamics — 457x faster than the Frontier supercomputer, exceeding the program’s 40x goal by more than 10 times.
-
NVIDIA — Targeting memory technologies that deliver higher bandwidth at lower energy consumption, coupling advanced memory with GPU architectures for continued U.S. HPC leadership.
Research areas:
- High-bandwidth memory (HBM) characterization and workload optimization
- Memory-centric and near-memory computing architectures
- Memory hierarchy modeling, simulation, and workload profiling
- CXL (Compute Express Link) and disaggregated memory systems
- Co-design of memory hardware with HPC and AI software applications