3DLS: A 3D Logic-Stacked Architecture for Disaggregated LLM Serving
By Jaehun Lee, In-Jun Jung, Joo-Young Kim
KAIST, South Korea

Abstract
Large language model (LLM) serving increasingly combines prefill-decode (PD) disaggregation with tensor parallelism (TP) to support large models and long contexts. In conventional 2D/2.5D chiplet architectures, layer-wise prefill-todecode KV-cache transfer decode-side TP collectives share the same lateral die-to-die (D2D) interconnect, creating mixed-traffic contention on the decode critical path. This contention increases communication latency, prolongs token generation intervals, and degrades end-to-end serving performance. We propose 3DLS, a logic-on-logic 3D-stacked chiplet architecture that separates traffic classes by routing KV-cache transfers through vertical interconnects while preserving decode-side TP collectives on the lateral D2D fabric. 3DLS achieves up to 1.49× throughput and 60.2% lower end-to-end (E2E) latency over the shared-fabric planar baseline, and still achieves up to 1.17× throughput and 31.4% lower E2E latency over a workload-aware priority-managed planar baseline. These results highlight that physical isolation is an important design principle for future chiplet-based PD-disaggregated LLM serving systems.
Index Terms—Large Language Model, LLM Serving, Disaggregated Serving, Chiplet, KV cache, 3D integration.
To read the full article, click here
Related Chiplet
- eFPGA Chiplet
- DPIQ Tx PICs
- IMDD Tx PICs
- Near-Packaged Optics (NPO) Chiplet Solution
- High Performance Droplet
Related Technical Papers
- Exploring the Efficiency of 3D-Stacked AI Chip Architecture for LLM Inference with Voxel
- AMMA: A Multi-Chiplet Memory-Centric Architecture for Low-Latency 1M Context Attention Serving
- 3D Stacked HBM and Compute Accelerators for LLM: Optimizing Thermal Management and Power Delivery Efficiency
- CHICO-Agent: An LLM Agent for the Cross-layer Optimization of 2.5D and 3D Chiplet-based Systems
Latest Technical Papers
- Chiplet-Escape: An Efficient Obstacle-Avoiding Escape Routing Method for Die-to-Die Interconnections in Chiplet-Based Designs
- Shared timing resources for multiple chiplet interfaces
- 3DLS: A 3D Logic-Stacked Architecture for Disaggregated LLM Serving
- WarpagePINN: Thermal Warpage Prediction in Advanced Packaging via a Two-Stage Physics-Informed Neural Networks
- A Comprehensive Design Framework for Vertical Power Delivery in High-Performance Computing