The semiconductor industry is rapidly transitioning toward 2.5D and 3D heterogeneous integration as a means to sustain performance scaling beyond traditional Moore's law. This approach combines multiple chiplets—often fabricated at different process nodes and optimized for distinct functions—into a single package or stack, enabling higher bandwidth, lower latency, and improved power efficiency compared to monolithic system-on-chip (SoC) solutions.
Physical design for these advanced packaging architectures presents unique challenges, including thermal management, signal integrity across interposers and through-silicon vias (TSVs), and complex die-to-die communication. To accelerate research and development in this field, UCLA has introduced an open-source benchmark suite specifically tailored for 2.5D/3D heterogeneous integration physical design studies.
This benchmark suite provides a standardized set of design cases, netlists, and integration scenarios, allowing researchers and industry engineers to evaluate and compare novel algorithms for placement, routing, and thermal-aware optimization. The suite is designed to reflect realistic 2026-era chiplet-based systems, incorporating heterogeneous nodes, interposer routing constraints, and advanced package architectures.
Key features of the UCLA benchmark suite include:
- Modularity: Supports a range of integration styles, including interposer-based 2.5D designs and stacked 3D ICs with TSVs.
- Realistic assumptions: Incorporates parameters derived from current industry roadmaps, such as advanced node geometries, interface bandwidth targets, and thermal budgets.
- Scalability: Offers designs of varying complexity, from small test cases to large, multi-chiplet systems.
- Open accessibility: Freely available to the academic and industrial community, with documentation and examples to facilitate adoption.
The release of this suite addresses a critical gap in the EDA research ecosystem, where proprietary benchmarks have historically hindered reproducible and comparable studies. By providing a common reference point, UCLA aims to foster innovation in physical design algorithms for heterogeneous integration, ultimately supporting the industry's push toward more sophisticated, high-performance 3D systems in the coming years.