Micron Plans to Stack GDDR Modules Like HBM to Serve AI Workloads

Micron is developing a method to vertically stack GDDR memory modules to serve AI workloads.

The initial GDDR stacking is expected to use about four layers and samples could arrive as early as next year.

Micron says the stacked GDDR would offer higher capacities but would not match HBM performance.

Thermal management and signal integrity are cited as primary technical challenges for GDDR stacking.