Packaging · Topic 13 of 14
0%
PackagingLesson 13 of 14
HBM
High-bandwidth memory: stacked DRAM connected by a very wide interface.
What you’ll learn
- What HBM is
- Why stacking + wide bus = bandwidth
- Where HBM is used
Why it matters
HBM feeds data-hungry accelerators far faster than conventional memory.
Explanation
HBM stacks several DRAM dies vertically (using TSVs) and connects them through a very wide interface, usually placed beside the processor on a 2.5D interposer. The wide bus delivers very high bandwidth at good energy efficiency.
It trades higher cost and packaging complexity for bandwidth that AI and HPC workloads demand.
Visual explanation
Key terminology
- HBM
- High-bandwidth memory (stacked DRAM).
- Wide interface
- A very large number of parallel data connections.
Example
An AI accelerator surrounds its compute die with several HBM stacks for terabytes/second of bandwidth.
Common mistakes
Watch out for:
- Confusing HBM's bandwidth advantage with lower latency — it's about throughput.
Real-world application
HBM is central to modern AI training hardware.
Explore related concepts
Continue learning
HBM
Up next
Advanced packaging
Packaging
Why packaging is now a primary driver of system performance.
Continue learning →