Completing AI: Physical intelligence and the role of memory
According to SK hynix Newsroom: As AI moves from datacenters into physical devices like robots and smart glasses, computing power alone is no longer enough. According to SK hynix and UIUC Professor Nam-sung Kim, data movement and storage have become the primary bottlenecks due to strict limits on battery, heat, and latency.

Original image temporarily unavailable
According to SK hynix Newsroom:
The Shift to Physical AI
As AI expands into real-world devices such as autonomous vehicles and industrial drones, offloading all decisions to a datacenter is impractical due to latency and connectivity risks. These physical AI systems must operate under strict constraints including limited battery capacity, tight thermal budgets, and restricted physical space. Consequently, system-level integration must begin with memory capacity and data transfer paths rather than raw compute alone.
Algorithmic Changes Driving Memory Design
New software techniques are altering how hardware handles information. Methods like quantization and Engram reduce computational loads by using conditional memory lookup tables instead of repetitive neural network calculations. However, this shift increases the demand for memory capacity and bandwidth, making efficient data delivery crucial for real-time inference.
A Unified Memory Hierarchy
To address these physical constraints, SK hynix outlines a vision combining custom HBM, 3D-stacked DRAM, Processing-in-Memory (PIM), CXL, and High Bandwidth Flash (HBF). Rather than relying on isolated components, the semiconductor industry is moving toward a full-stack approach where memory architecture is co-designed to match specific workloads and form factors.
View selection rationale
Importance 75/100
About this brief
- Content type
- News summary
- Source type
- First-party / official source
- Original source
- SK hynix Newsroom ↗
- Source published
- Process
- AI-assisted discovery, translation, summarisation, classification and scoring; standards and corrections are maintained by the editorial desk.
- Verification
- The original URL and publication metadata are retained. This brief is not independently fact-checked unless explicitly stated.
- Published by
- Polyemergence Editorial Desk · PolyEmergence Pte Ltd.
- Conflict disclosure
- No financial or commercial conflict declared.