Samsung · Filed Aug 13, 2025 · Published Jul 16, 2026 · verified — real USPTO data

Samsung Patents an AI System That Fills In Missing Data During Chip Manufacturing

When a sensor on a chip-making machine goes silent, the data it should have recorded is gone, and that gap can mean a bad batch of chips nobody catches until it's too late. Samsung's new patent describes an AI system that reconstructs what that sensor would have recorded, filling the hole before it becomes a problem.

Samsung Patent: AI Fills Data Gaps in Chip Manufacturing — figure from US 2026/0203658 A1
Figure from the official USPTO publication.
Publication number US 2026/0203658 A1
Applicant Samsung Electronics Co., Ltd.
Filing date Aug 13, 2025
Publication date Jul 16, 2026
Inventors Minseok KIM, Gilhwan KIM, Sangbeom PARK, Junhyeok PARK, Yoonsang LEE, Yohwan JOO, Ahryeon CHOI, Sangmin HWANG
CPC classification 706/12
Grant likelihood Medium
Examiner CENTRAL, DOCKET (Art Unit OPAP)
Status Docketed New Case - Ready for Examination (Sep 4, 2025)
Document 20 claims

What Samsung's chip-factory data recovery system actually does

Imagine a factory making computer chips where dozens of sensors are constantly tracking every step of the process. If one sensor drops out, or its data gets corrupted, you suddenly have a gap in your records. That gap can hide a defect that ruins thousands of chips, or it can trip up the software trying to monitor quality.

Samsung's patent describes an electronic device that watches over the chip-making machines in real time. When it detects that some data is missing or corrupt, it builds a model to reconstruct what that data should have been, based on what all the other sensors recorded during that same window.

The goal is to keep the monitoring picture complete even when individual data streams fail. Rather than flagging an error and stopping production, the system fills in the gap and hands back a complete picture of what happened inside the machine.

How the restoration model reconstructs lost operation data

The patent describes a device that connects to a semiconductor exposure machine (the tool that uses light to etch circuit patterns onto a chip, sometimes called a lithography system). That machine has many moving parts, called assemblies, each generating its own stream of data.

The device does four things in sequence:

  • Collect: It gathers operation data from all those assemblies during an exposure run on a wafer.
  • Identify: It spots what the patent calls a loss operation, meaning a moment when data from one or more assemblies is missing or corrupted.
  • Build a model: It generates a restoration model, essentially an AI trained on the surrounding good data, designed to predict what the missing values should look like.
  • Restore: It runs the restoration model to produce restoration operation data, a synthetic but informed replacement for the lost readings.

The patent does not specify which AI architecture the restoration model uses, so the approach is intentionally broad. The core idea is that because all the assemblies in one machine are physically related, data from healthy sensors can reasonably predict what a failed sensor would have recorded.

What this means for semiconductor yield and quality control

Semiconductor manufacturing is one of the most data-intensive industrial processes on the planet. A single wafer passes through dozens of steps, and equipment vendors, fabs, and quality teams all depend on complete logs to trace defects back to their source. When data is missing, engineers either have to scrap that wafer's traceability record or make guesses. Both options cost time and money.

For Samsung, which operates its own large-scale chip fabs and supplies chips to external customers, better data continuity means fewer unexplained yield drops and faster defect investigations. If this system works as described, it could reduce the number of wafers discarded because their process history has a gap, which adds up quickly when you are running millions of wafers per year.

Editorial take

This is a practical patent, not a flashy AI play. The problem it solves, missing sensor data in chip factories, is real and expensive, and the approach of using surrounding good data to reconstruct bad data is sensible. Whether the restored data is accurate enough to actually trust is the engineering challenge the patent does not fully answer, but the problem statement alone makes this worth watching as chip fabs push toward more automated quality control.

Which company should we read for you?

We track 17 companies here. Pro is the same weekly breakdown for any company you choose, delivered privately. Type a name and we'll scope it and send you a quote.

Get one Big Tech patent every Sunday

Plain English, intelligent commentary, no hype. Free.

Source. Full patent text and figures from the official USPTO publication PDF.

Editorial commentary on a publicly published patent application. Not legal advice.