Skip to content
0MESubscribe
Chips17 Nov 20262 min read

Ampere and ASU build DT-MATCH to rescue defective chiplets and raise yields

Researchers from Arizona State University and Ampere Computing have introduced DT-MATCH, a framework to pair repairable chiplets and boost manufacturing yields.

Listen to this storyAbout 3 min, read by your device
Picture made with AI

Heterogeneous integrationHeterogeneous integrationCombining different small chips into a single package so they work together as one system. Open in glossary lets engineers build large processors by combining smaller silicon components known as chiplets. This approach lowers manufacturing costs, but combining multiple pieces into one system-on-chip creates fresh assembly challenges. When defects hit the tiny lanes connecting these dies, builders struggle to make functional products from parts that have repairable flaws.

The problem with chiplet assembly

Modern advanced packaging links several dies together using dense interconnects. Even small manufacturing defects on individual data lanes can break a whole system. Many designs include spare lines to route around broken links. However, physical layout rules and limited backup lanes make it hard to decide which imperfect parts can safely work together. Without smart sorting, factories discard components that could still function in the right pairings.

How DT-MATCH pairs parts

To solve this waste, researchers from Arizona State University and Ampere Computing developed a framework called DT-MATCH. The system checks both physical layouts and defect locations to assemble as many working systems as possible. It shrinks the scale of the matching problem first through a shared preparation pipeline, then applies one of two methods:

  • An integer linear programming approach uses batching to find the best possible pairings with top yieldYieldThe share of chips on a wafer that work. Open in glossary quality.
  • A greedy randomized search procedure runs much faster, cutting calculation times while maintaining output close to the exact method.

Tests on industrial chip designs showed that both methods help makers balance processing speed and overall yield during final packaging.

How does this story land?

One tap per reader.

Quick poll

Will chiplet matching software become essential for advanced packaging plants?

One tap. No sign-up.

Ask this story

Still have a question?

Answers use only what is in this story. They are written by AI and can be wrong.

Sources

  1. Semiconductor Engineering, Defect-Aware Chiplet Matching For Heterogeneous Integration Yield (ASU, Ampere)
Companies in this storyAmpere ComputingWords explainedHeterogeneous integrationYield
Up next in ChipsSlipStream cuts flash memory delay to boost AI chip capacity by 16x2 min read

Keep reading

All Chips stories