Test report DSG-1002 · Rev A · tested November 18, 2026
Foundries & ManufacturingDevice under test
DT-MATCH targets yield loss in repairable chiplet assembly
ASU and Ampere's DT-MATCH combines ILP and GRASP to match defective, repairable chiplets, balancing higher solution quality against lower runtime under assembly constraints.
- Read
- 3 min
- Words
- 616
- Node
- 45nm
- Operator
- Amara Osei

By Amara Osei
What does DT-MATCH optimize?
DT-MATCH sets two matching methods against one manufacturing objective: maximizing the number of functional chiplet-based systems assembled from defective but repairable chiplets. An integer linear programming approach delivers higher solution quality; a greedy randomized adaptive search procedure lowers runtime without guaranteeing yield optimality.
Researchers at Arizona State University and Ampere Computing published “DT-MATCH: Defect- and Topology-Aware Chiplet Matching for Yield Optimization in Heterogeneous Integration.” arXiv lists the technical paper as no. 2611.19544 under the date Nov. 14, 2026.
Xuanyi Tan, Sanmitra Banerjee, Dhruv Thapar, Sreejit Chakravarty and Krishnendu Chakrabarty authored the study.
The authors write in the abstract: “Experimental results on industrial designs demonstrate that, as expected, the ILP-based approach achieves superior solution quality.”
- Integer linear programming (ILP): A batching strategy approximates the exact formulation. This branch records superior solution quality in the industrial-design experiments.
- Greedy randomized adaptive search procedure (GRASP): Lower runtimes and solution quality comparable to ILP define this branch, but it does not guarantee yield optimality.
ILP therefore leads on solution quality, while GRASP leads on runtime.
Why do defects complicate chiplet matching?
Heterogeneous integration, or HI, and advanced packaging enable chiplet-based system-on-chip, or SoC, designs. The authors also identify manufacturing defects and complex die-to-die, or D2D, interconnect constraints as sources of yield loss.
Three constraints shape the matching problem:
- Lane-level defects
- Limited spares
- Strict structural compatibility requirements
The paper says these factors impede efficient assembly of functional systems from defective but repairable chiplets. Limited replacement inventory and incompatible die structures can narrow the available pairing options even when a chiplet remains repairable.
DT-MATCH therefore seeks to maximize functional SoC output under realistic assembly constraints rather than setting a separate objective for defect repair.
The objective is functional SoC output, constrained by defects, spares and compatibility.
What does the ILP method deliver?
The first method formulates matching as an integer linear programming problem. The researchers implemented it with a batching strategy that approximates the exact formulation.
Tests on industrial designs gave this branch superior solution quality. GRASP produced lower runtimes and solution quality comparable to the ILP-based result, establishing the framework’s central quality-versus-efficiency trade-off.
The abstract supplies no batch size, runtime value, yield percentage or numerical quality margin. It supports a relative comparison rather than a quantified production gain.
ILP leads on reported solution quality, but the abstract leaves the size of that lead unquantified.
What does GRASP change?
The second method applies GRASP, a heuristic search. The authors state: “The GRASP-based approach leads to lower runtimes while still achieving solution quality that is comparable to the ILP-based solution.”
Because GRASP does not guarantee yield optimality, the method cannot promise the best matching outcome for every industrial design. Its reported advantage concerns runtime, not a guarantee of maximum functional SoC output.
The abstract provides no acceleration factor or runtime measurement, so the speed advantage remains qualitative in the cited summary.
GRASP lowers runtime and retains comparable solution quality, without guaranteeing optimal yield.
Which method should an assembler choose?
DT-MATCH presents two operating profiles rather than one method for every assembly setting. ILP suits assemblers that prioritize higher reported solution quality. GRASP suits assemblers that prioritize lower runtime and accept the absence of a yield-optimality guarantee.
Both routes use a preprocessing pipeline to reduce the problem size. The paper positions the pair as a trade-off between solution quality and efficiency in practical chiplet assembly optimization.
For assemblers, the choice is explicit: accept ILP’s longer runtime for higher reported quality, or use GRASP for lower runtime without a guarantee of optimal yield.
via arxiv.org (Original)
More from Amara Osei
Show full bio
Staff writer covering money, funding and market shifts around hardware layer of artificial intelligence at Die Signal.
698 articles
Same lot · LOT-362E
- DSG-415445nmSamsung Foundry, Cadence Team on Chiplet Designs for Physical AI
- DSG-464514nmMATCH Act clears first hurdle, pivots export curbs from chips to tools
- DSG-824020nmAI Chips Outpace Test: Insertions Double, Silicon Dies Before Deployment
- DSG-614410nmSynopsys, TSMC extend design and IP support for AI silicon
- DSG-379245nmTSMC Recasts Advanced Packaging as a Design Software Problem