Network Design for Wafer-Scale Systems with Wafer-on-Wafer Hybrid Bonding
This addresses communication limitations in wafer-scale AI systems, offering incremental improvements in network design for enhanced performance.
The paper tackles the problem of communication bottlenecks in wafer-scale systems for large language models by investigating how reticle placement affects network topology, achieving up to 250% higher throughput, 36% lower latency, and 38% lower energy per byte.
Transformer-based large language models are increasingly constrained by data movement as communication bandwidth drops sharply beyond the chip boundary. Wafer-scale integration using wafer-on-wafer hybrid bonding alleviates this limitation by providing ultra-high bandwidth between reticles on bonded wafers. In this paper, we investigate how the physical placement of reticles on wafers influences the achievable network topology and the resulting communication performance. Starting from a 2D mesh-like baseline, we propose four reticle placements (Aligned, Interleaved, Rotated, and Contoured) that improve throughput by up to 250%, reduce latency by up to 36%, and decrease energy per transmitted byte by up to 38%.