While GPUs have the spotlight, high bandwidth memory (HBM) has an Oscar nomination for best supporting actor – but what's the state of the storage/memory market; what are the innovations therein; and who are the major players?
A trip down memory lane:
Memory vs compute
NAND/SSD and data lakes (HDD need not apply)
AI at the edge (client-side NVM SSDs)
HBM3, 3D DRAM, and LLMs (stacked packaging)
Future compute-in-memory architectures (memristor crossbar arrays)
Memory and manufacturing competitive landscape
See also:🧵https://x.com/AnthPB/status/1658939488729460737
This podcast is AI-generated with NotebookLM, using the following sources, research, and analysis:
AI Workloads and Data storage solutions requirements and recommended options (Medium, Sugur, 2024.02.21)
AI at the Edge: Future of memory and storage in accelerating intelligence (IoT Now, 2024.02.07)
Flash forward: The need for AI storage and security in 2024 (Fast Company, Wu, 2024.02.29)
GDDR5 vs GDDR5X vs HBM2 vs GDDR6 vs GDDR6X Memory Comparison (Graphics Card Hub, 2023.10.29)
HBM3 Update to High Bandwidth Memory Standard (JEDC, 2022.01.27)
HBM: Memory Solution for Bandwidth-Hungry Processors (Kim/Kim, 2014.08)
High Bandwidth Memory (Wikipedia, 2024.10.23)
Historical trends in technical themes: Analog systems (ISSCC, Analog and Data Converters Subcommittees, 2014)
Memory Is All You Need: An Overview of Compute-in-Memory Architectures for Accelerating Large Language Model Inference (Wolters/Yang/Schlichtmann/Suzumura, 2024.06.12)
Micron Commences Volume Production of Industry-Leading HBM3E Solution to Accelerate the Growth of AI (Micron, 2024.02.26)
Micron earnings 2024q3 (CNBC, 2024.09.25)
SK Hynix earnings 2024q3 (CNBC, 2024.10.23)
Samsung earnings 2024q3 (CNBC, 2024.10.07)
Staying Curious: Developments in Memory Chip Technology (Telecom Review, 2024.06.19)
The Value of Data Storage for the AI Era (Janukowicz, 2024.05)
The history of SK Hynix 1983-2023 (SK Hynix, 2024.10.24)
What changes in storage will AI drive? (Micron, 2023.11)
What’s Next For High Bandwidth Memory: Different approaches for breaking down the memory wall (Semiconductor Engineering, 2019.12.17)
Not investment advice; do your own due diligence!
# GPU IoT HCP hyperscale cloud datacenter edge computing Micron Samsung SK Hynix Taiwan Semiconductor $mu $tsm
