Researchers from Arizona State University (ASU), in collaboration with Ampere Computing, have revealed a new technical framework called "DT-MATCH," which aims to address one of the most complex problems in the modern semiconductor industry: optimizing the yield of heterogeneous integration processes.
Manufacturing Challenges in the Chiplet Era
As the industry moves toward designing Systems-on-Chip (SoC) using Chiplet technology and advanced packaging, new challenges regarding manufacturing yield have emerged. These problems primarily arise from manufacturing defects and complex constraints in die-to-die interconnects.
The research explained that lane-level defects, limited spares, and strict structural compatibility requirements are all factors that hinder the efficient assembly of functional systems from chips that may contain manufacturing defects but remain repairable or usable in specific contexts.
DT-MATCH Methodology: Balancing Precision and Speed
The DT-MATCH framework presents two integrated approaches to optimize the chiplet matching process to ensure obtaining the largest possible number of functional Systems-on-Chip (SoCs):
- Integer Linear Programming (ILP) approach: This approach relies on a "batching" strategy to approximate exact mathematical formulas and is characterized by its ability to provide the highest possible solution quality, ensuring maximum yield.
- Greedy Randomized Adaptive Search Procedure (GRASP) approach: This approach is designed to provide significant acceleration in execution time; although it does not guarantee reaching the mathematically optimal solution, experimental results showed solution quality very close to the ILP approach, making it a practical choice for applications requiring processing speed.
Experiments on industrial designs have proven that both approaches provide an effective trade-off between solution quality and runtime efficiency, serving the needs of real-world chip assembly processes.
Why Does This Development Matter?
In light of the global trend toward advanced packaging technologies, the challenge is no longer limited to designing the chip itself, but rather how to efficiently assemble several chiplets together. The core importance of DT-MATCH lies in its ability to "rescue" chips that contain minor defects; instead of discarding the entire chip due to a defect in one of the lanes, the framework intelligently matches it with other chips that are structurally and functionally compatible to create a complete, efficiently working system.
This shift means reducing silicon waste, lowering overall manufacturing costs, and increasing the number of saleable units from each silicon wafer, which is vital for semiconductor companies seeking to reduce production costs amidst the complexity of modern designs.