Modeling Chiplet-to-Chiplet (C2C) Communication for Chiplet-based Co-Design
By Fabian Schatzle, Carlos Falquez, Nam Ho, Andre Zambanini, Johannes van den Boom and Estela Suarez
Forschungszentrum Julich GmbH
Abstract
Chiplet-based processor design, which combines small dies called chiplets to form a larger chip, enables scalable designs at economical costs. This trend has received high attention such that standards for chiplet design have rapidly established, including packaging, protocols, and Chiplet-to-Chiplet (C2C) interfaces. With numerous well-defined chiplet options available, hardware architects would leverage on the co-design process to make optimal decisions on design parameters.
An important performance limitation in multi-chiplet designs come from the protocol translation in the C2C communication, needed to maintain cache coherency and avoid risk of deadlocks. When integrating multiple chiplets, deadlocks can happen from both protocol and routing, making deadlock-free designs important.
This paper tackles these challenges by introducing a Chiplet-to-Chiplet Gateway (C2CG) architecture, a C2C interface that bridges two chiplet protocols and ensures deadlock-free C2C communication. We also extend the Coherent Hub Interface (CHI) protocol to support cache coherent data sharing among cores across chiplets. The complete design is implemented in the gem5 simulator, constructing a modeling tool for chiplet-based co-design targeting next-generation High-performance Computing (HPC) processors. We demonstrate the benefit of the model through a design space exploration of three 64-core Armv8 HPC processor configurations: monolithic, two- and four-chiplet. The exploration, using representative HPC benchmarks, provides insights into C2C parameters and studies the impact of Non-Uniform Memory Access (NUMA) configuration, giving valuable co-design feedback for hardware architects.
Index Terms—High-Performance Computing, Chiplet, Cache coherency
To read the full article, click here
Related Chiplet
- DPIQ Tx PICs
- IMDD Tx PICs
- Near-Packaged Optics (NPO) Chiplet Solution
- High Performance Droplet
- Interconnect Chiplet
Related Technical Papers
- Cambricon-LLM: A Chiplet-Based Hybrid Architecture for On-Device Inference of 70B LLM
- MFIT : Multi-FIdelity Thermal Modeling for 2.5D and 3D Multi-Chiplet Architectures
- Communication Characterization of AI Workloads for Large-scale Multi-chiplet Accelerators
- Chiplever: Towards Effortless Extension of Chiplet-based System for FHE
Latest Technical Papers
- Rethinking Compute Substrates for 3D-Stacked Near-Memory LLMDecoding: Microarchitecture–Scheduling Co-Design
- DeepStack: Scalable and Accurate Design Space Exploration for Distributed 3D-Stacked AI Accelerators
- Mapping Space Exploration for Multi-Chiplet Accelerators Targeting LLM Inference Serving Workloads
- Escaping Flatland: A Placement Flow for Enabling 3D FPGAs
- 3D optoelectronics and co-packaged optics: when solving the wrong problems stalls deployment