Reactor Downtime Optimization

A semiconductor manufacturer faced costly reactor aborts losing over $19,000 each and spent 4–6 hours manually investigating root causes per incident.

The Challenge

Minimizing Downtime in Semiconductor Manufacturing

One of our clients in the semiconductor space operates multiple chemical vapor deposition (CVD) reactors to deposit critical material layers during chip production. When a reactor unexpectedly stops mid-process, known as a "run-abort", the impact is immediate and costly. Each incident leads to material waste, production delays, and an average loss of over h.

In a high-demand environment, every minute of downtime matters. Restarting a reactor requires identifying the root cause of failure, a process that currently consumes 4–6 hours of manual data collection per incident. Quality and process engineers must sift through disparate systems, form hypotheses, and run repeated analyses, often under intense time pressure.

We help streamline this process by building digital solutions that accelerate root cause analysis, reduce unplanned downtime, and drive higher yield, giving engineers the tools they need to respond faster and smarter.

The Solution

Accelerating Root Cause Analysis with a Secure GenAI Assistant

The CVD process is highly complex, with numerous potential failure modes that can trigger a run-abort. Identifying the cause requires gathering and analyzing large volumes of data, often from multiple systems and in varying formats. This manual process is time-consuming and inefficient.

To solve this, we developed a secure GenAI assistant that allows engineers to interact using natural language. Instead of presenting raw data, the assistant delivers structured and relevant insights instantly.

Engineers can now access critical information such as recipe changes, material numbers, log entries, temperatures, and pressures within minutes, simply by asking a question.

The solution is powered by Azure OpenAI and enhanced with LangChain for seamless, conversational interaction. We integrated Azure Web Services with Azure Entra ID to ensure secure access, and implemented Langfuse to provide full transparency into how the assistant generates its responses.

To solve this, we developed a secure GenAI assistant that allows engineers to interact using natural language. Instead of presenting raw data, the assistant delivers structured and relevant insights instantly.

Engineers can now access critical information such as recipe changes, material numbers, log entries, temperatures, and pressures within minutes, simply by asking a question.

The solution is powered by Azure OpenAI and enhanced with LangChain for seamless, conversational interaction. We integrated Azure Web Services with Azure Entra ID to ensure secure access, and implemented Langfuse to provide full transparency into how the assistant generates its responses.

The results

Shortened Payback Time

With the secure GenAI assistant, engineers now gather data in 5 to 10 minutes, eliminating 25 hours of manual work every week. On top of that, the reactor can restart sooner, which saves valuable production time. And with every abort prevented, 18k worth of raw material is saved.

We’ve got more great projects