DT-MATCH targets higher yield in heterogeneous chiplet assembly

November 17, 2026

DT-MATCH targets higher yield in heterogeneous chiplet assembly

Image: ChatGPT

Researchers from Arizona State University and Ampere Computing have presented DTdynamic test-MATCH, a chiplet matching framework aimed at improving yield in heterogeneous integration. The work addresses a central packaging problem: defective but still repairable chiplets can only be assembled into working systems when defects, interconnect topology, and structural compatibility are considered together.

Heterogeneous integration and advanced packaging have made chiplet-based system-on-chip designs more practical, but they also create new yield bottlenecks. In this case, the relevant constraints include lane-level defects, limited spare resources, and strict requirements on how dies can be connected. DTdynamic test-MATCH is built to account for all of these factors at once instead of treating matching as a purely combinatorial assembly task.

The framework offers two complementary optimization approaches. One is based on integer linear programming and uses batching to approximate an exact formulation. The other is a greedy randomized adaptive search procedure, which is intended to accelerate the search process even though it does not guarantee an optimal yield result.

Both approaches use a preprocessing pipeline to reduce the problem size before optimization. That step is part of the practical design of the method, since chiplet assembly problems can become large and difficult to solve directly when multiple defect and topology constraints must be satisfied simultaneously.

Experimental results on industrial designs show the expected trade-off between solution quality and runtime. The integer linear programming approach delivered the best solution quality, while the greedy randomized adaptive search procedure reduced runtime and still reached results that were comparable to the ILPinteger linear programmingInteger linear programming is an optimization method for linear models with integer variables.-based solution.

The paper therefore positions DTdynamic test-MATCH as a way to balance yield optimization and computational efficiency in chiplet assembly planning. The reported contribution is not a new package structure or device, but a matching methodology for assembling functional SoCs from imperfect chiplets under realistic heterogeneous integration constraints.

Source: Semiconductor Engineering

More background: Advanced Packaging

← All posts