June was a strong month at SixSense, with AI deployments delivering real gains across wafer fabs, advanced packaging, and hard disk drive (HDD) fabs. Across these lines, AI is now improving yield, speeding up inspection, and elevating quality standards faster than before.
In this month's edition, we look at how AI-ADC is closing inspection gap at HDD fabs and lifting yield toward 97%, how an advanced packaging line at a leading OSAT reached over 90% automation with zero escapes, and how AI-assisted data prep is taking model setup from four-to-six weeks down to under two days.
Alongside these results, we're proud to share a milestone: SixSense was named a World Economic Forum Technology Pioneer for 2026 — recognition for bringing AI to the physical industries the modern world runs on, including the fabs that make our chips. We also took the stage at AppWorks Demo Day #32 in Taipei, sharing our story in the heart of the global chip industry.
Recovering Yield at HDD Fabs with AI
SixSense is bringing AI-ADC (that already runs in high-volume semiconductor fabs and wafer packaging) to optical inspection at hard disk drive (HDD) fabs.
Today, these fabs scan wafers and capture thousands of defect images, but rarely classify them. The result: too many false alarms, good sliders over-rejected, real defects escaping to the backend, and most inspection data left underutilised — leaving baseline yield stuck below 96%.
By classifying every defect, SixSense closes these gaps. For a mid-to-large fab, the opportunity is significant:
- False alarms: ~20% → under 2%
- Good sliders over-rejected: 1–3% → under 0.1%
- Defects slipping through to later stages: 1–2% → ~0%
- Overall yield: 95–96% → ~97% (worth $7–19M a year)
The same AI that runs in high-volume semiconductor operations at over 99.3% accuracy and zero escapes is now proving impact at HDD fabs.
Read about AI in action at HDD fabs.
AI-ADC on a Panel-Level Packaging Line — Over 90% Automation, 0% Escapes
SixSense AI-ADC handled defect classification on a fan-out panel-level packaging (PLP) line — one of the most demanding environments in advanced packaging, where accept/reject decisions depend on a defect's size and exact position across large 510mm panels. The line had been running on 10+ operators, where operator fatigue drove both over-rejection and missed defects. Inconsistent labeling wasted engineering time and the end customer no longer accepted human judgment in the loop.
SixSense provided a solution to replace manual review with context-aware AI that weighs each defect's size, position, and surroundings, using a single recipe across multiple devices. The result is over 90% automation with zero escapes — missed defects fell from 16% to none. Accuracy was found to be >15% higher than manual review, recovering an additional >2% in yield. A line that once leaned on human judgment now will run on consistent, auditable AI decisions.
AI-Assisted Data Prep Tools for Operators — Weeks Cut to Two Days
What sets SixSense apart in defect classification isn't just a better model — it's also how the data is prepared before training. And it's proven in benchmarks: on a wafer-level bumping and RDL line, SixSense turned four-to-six weeks of model setup to under two days.
The bottleneck was data — choosing which images to train on and correcting inconsistent labels by operators. SixSense's AI-assisted tools handle both. One selects the right images (~2-4k) from more than a million with the tricky borderline accept/reject cases retained. The second presents the images as a visual map, allowing inconsistent labels to be identified and corrected in hours rather than days.
The outcome: a production-ready model in under two days, accuracy above 98%, defect escapes reduced to near zero, and retraining cut to one or two times a year. The right images, it turns out, beats more images.
Read more : link
SixSense Named a World Economic Forum Technology Pioneer for 2026
SixSense has been selected as a World Economic Forum Technology Pioneer for 2026 — one of 100 early-stage companies, from 23 countries, recognized for technologies that could reshape industries and everyday life.
The program is now in its 26th year, and its past honorees include Google, PayPal and Dropbox, each recognized in their early days. This year's group is united by one idea: applying AI not to apps and chatbots, but to the physical industries the modern world runs on — energy, space and advanced manufacturing.
That is exactly where SixSense builds. Chips are in nearly everything we use, yet the factories that make them still lean heavily on human judgment to decide what passes and what is scrapped. SixSense brings AI to that decision — helping fabs catch defects accurately and recover yield, at production speed.
This is a proud moment for us. The credit belongs to our team across Singapore, Taiwan and India, to the customers who trusted us on their lines early, and to the investors who backed a difficult, unglamorous, and important problem.
SixSense Takes the Stage at AppWorks Demo Day #32
This month, SixSense took the stage at AppWorks Demo Day #32 in Taipei — one of Asia's leading showcases for high-growth startups. In a two-minute pitch, we shared how SixSense brings AI to the semiconductor line: classifying defects, tracing root cause, and predicting yield issues across fabs and OSATs worldwide.
There was no better place to tell that story. Taiwan sits at the centre of the global semiconductor industry, and has become one of SixSense's fastest-growing markets — our equipment-agnostic AI fits the way its fabs and test houses actually run, backed by a team on the ground who understand their floor and their urgency.
It was also a fitting first stop as we begin our next phase of growth, with many more conversations to come. Best of all, our team showed up in full force.
Thanks to AppWorks for having us, and to everyone who stopped by to talk yield. Onward!
