← Products
01Overview

DutchBoy SEMI · Semiconductor Manufacturing

From tool data to test results, find anomalies and root causes in one place

DutchBoy SEMI analyzes tool sensor traces, equipment logs, and metrology and test results with a single time-series engine. It catches anomalies first, then links them to likely causes, supporting evidence and corrective actions. Your FDC, YMS, MES and other existing systems stay as they are.

Applicable Processes

EtchDepositionCMPCleaningIon implantationWafer testPackage assemblyFinal test

Data Analyzed

P

Pulse

Tool sensor traces, recipe steps

D

Dot

Test results · bins · STDF

T

Table

Metrology · MES process history

02Challenges on the Floor

What existing systems tell you — what happened. What's still up to people — why, and what's next.

Challenge 1

Plenty of alarms, but people still hunt for the cause

Alarms fire only when control limits are crossed, and engineers dig for the cause by hand across FDC, sensor databases, MES and YMS.

Challenge 2

Subtle changes slip through

Summary-statistic control charts miss level shifts inside a step and slow, creeping drift.

Challenge 3

Every tool is a little different

The same recipe yields different results across chambers and tools, and finding and closing the gaps takes time.

Challenge 4

Yield problems surface too late

Test yield and bin issues are analyzed after the lot finishes, then traced back to the tool responsible.

03What Gets Analyzed, What Comes Out

The data you already have, read as one — anomalies, root causes and actions, delivered.

Input · existing systems and data

PTool sensor traces · recipe steps
PEquipment logs · alarms · settings
DTest results · bins · STDF
TMetrology · MES process history
FDC · SPC · YMS · results

DUTCHBOY ANALYSIS

1

Collect & clean

Align scattered data on one timeline

2

Detect anomalies

Score deviations from the normal baseline

3

Find root causes

Where, what and why it differed

4

Connect to action

Next steps backed by past cases and evidence

Sensor, waveform, measurement and image data are all converted to time-series and analyzed by one engine — no need to build a new model for each data type.

Output

  • Anomaly scores and risk ranking by wafer and chamber
  • Where it happened (step, segment) and top contributing sensors
  • Differences against the normal baseline and the reference tool
  • Real-time yield and bin shifts linked to the tools behind them
  • Similar past cases and the actions taken
  • Evidence charts and automated reports
04How This Differs From Existing Systems

Your existing systems, unchanged — just the 'why' and 'what next' added on top.

FDC

Limitation

Engineers set and maintain control limits sensor by sensor. Alarms fire only when a limit is crossed; the 'why' is left to them

What DutchBoy Adds

Automatically picks a normal baseline with no limit setup, scores deviations, and shows where they occurred and which sensors contributed

SPC

Limitation

Summary-statistic control charts miss subtle changes inside the waveform

What DutchBoy Adds

Analyzes raw waveforms together with step-level statistics to catch level shifts and drift

YMS

Limitation

Yield and bin are analyzed after the lot finishes

What DutchBoy Adds

Detects yield and bin shifts during test and ties them to tool status and alarms

MES · alarm rules

Limitation

Each system raises its own alarms, and rule changes go live without knowing their impact

What DutchBoy Adds

Brings alarms and history into one place and validates rules before they go live

Engineering know-how

Limitation

Corrective know-how stays with individuals

What DutchBoy Adds

Builds up a history of defects and actions to reuse when similar situations occur

05Application Scenarios

One approach, different problems — solved in every process.

Front-end

Chamber anomaly root cause

  1. 1Sensor anomaly score rises
  2. 2Check the abnormal step and contributing sensors
  3. 3Look up similar past cases
  4. 4Re-verify after the fix
Front-end

Tool-to-tool matching

  1. 1Select a reference tool
  2. 2Auto-compare sensor patterns, parameters and settings
  3. 3Rank the differences
  4. 4Compare results after adjustment
Back-end

Responding to test yield drops

  1. 1Detect yield and bin shifts
  2. 2Compare testers, sites and conditions
  3. 3Link to the tool or process behind it
  4. 4Turn it into an alarm rule
06How the 6 Modules Work Together

Only the modules you need, added on top of what you have — Core to start, automated reporting and conversational AI as Add-ons. Every module running on the plant data Weave designs.

Agent · Query & reasoning
Visualize · Evidence & reporting
Sense
Recall
Watch
Weave · Factory data design

Your existing systems — unchanged

MESFDCSPCYMSEESAPCTesters

Weave

Data design

Collects tool trace, test files and MES history, aligns and organizes recipe/step master data

Core

Sense

Detection & RCA

Detects wafer and chamber anomalies, ranks contributing sensors, compares tools, flags yield and bin shifts

Core

Recall

Recall & response

Searches past cases and actions by product and defect type, and verifies results after a fix

Core

Watch

Integration & alarms

Brings FDC, SPC, YMS and MES results and alarms into one place, with rule-based alarms

Core

Visualize

Evidence & reporting

Generates evidence charts and daily/weekly reports automatically

Add-on

Agent

Query & reasoning

'Find the cause of the chamber 3 anomaly' → calls the modules and answers with evidence

Add-on

Why an LLM alone isn't enough — Time series from just 150 sensors add up to millions of numbers (about 5.4 million in our technology video example), far more than an LLM can read directly. DutchBoy's manufacturing-specific tools condense them into a few lines of key evidence — severity, pattern, scope of impact and past cases — and hand them to the Agent.

Could this fit your process?

From feasibility to expected impact, the AIBIZ team will walk you through it directly.

Book a Consultation