HBM Yield Is the Bottleneck
Nihar Shah@nehharshah
The idea
I believe the real AI memory bottleneck isn't making more HBM wafers, it's keeping yields up as stacks get taller and thinner, since one bad die can scrap an expensive stack. So this owns the inspection, metrology and test companies that catch defects early, weighted most heavily toward the die-level test and packaging-inspection names already benefiting today, with smaller stakes in less direct yield analytics and an upstream burn-in bet. It skips memory makers and bonding equipment directly, and keeps a small cash buffer.
What it owns
- Tools that find defects in thinned dies and bonded stacks prevent one flaw from ruining an entire high-value HBM package. · Onto Innovation Inc.Sells inspection and measurement tools for advanced packaging where thinner dies and more bonding steps create new failure modes., Camtek Ltd.Focused on finding defects on dies and bumps before stacking, with most revenue now tied to HBM and advanced packaging customers., KLA CorporationProcess control leader whose packaging and memory inspection sales grow as chipmakers prioritize detecting defects over simply adding capacity., Nova Ltd.Measures film thickness and shapes on wafers, becoming more important as dies thin and stacking tolerances tighten in advanced packages.56%
- Testing each die before stacking becomes more critical as memory stacks grow taller and each defect costs more to fix inside an expensive package. · FormFactor, Inc.Makes the probe cards that test memory dies before stacking, catching defects at the cheapest point as HBM layers multiply., Teradyne, Inc.Builds test systems for memory chips where HBM stacks require more testing per layer as designs stack higher.33%
- Software that identifies where yield is lost in manufacturing becomes more valuable as each defect in advanced packaging costs more to discover late. · PDF Solutions, Inc.Sells software that consolidates factory data to help chipmakers find yield losses in complex packaging processes.3%
- Stress-testing chips at the wafer stage before assembly could become standard for AI packages if it proves cost-effective versus fixing failures after packaging. · Aehr Test SystemsIt makes wafer-level burn-in systems that test chips before packaging, positioning it to benefit if this quality control step becomes standard in AI and memory chip production.3%
- Cash5%