Qualcomm Unveils HBC Near-Memory AI Architecture with AI250 and AI350 Accelerators
Qualcomm's New AI Architecture
Qualcomm has revealed its HBC (Hybrid Bandwidth Core) near-memory AI architecture, accompanied by the new AI250 and AI350 accelerators. The architecture is designed to address the growing demand for efficient AI processing by bringing computation closer to memory.
Performance Claims
According to Qualcomm, the HBC architecture offers two significant advantages over existing solutions:
- 6x higher bandwidth-per-watt compared to traditional HBM (High Bandwidth Memory)
- 200x greater capacity compared to on-chip SRAM
These improvements aim to reduce the energy bottleneck commonly associated with AI workloads, where moving data between memory and processors consumes substantial power.
Near-Memory Computing Approach
The HBC architecture implements a near-memory computing approach, which places processing elements closer to memory banks. This design minimizes the distance data must travel during AI computations, potentially improving both efficiency and latency for inference and training tasks.
Market Context
The announcement comes as demand for specialized AI accelerators continues to grow across data centers, edge devices, and mobile platforms. Qualcomm's focus on power efficiency aligns with broader industry trends emphasizing sustainable AI deployment.