Yield is a supply-chain problem too

When a new fab struggles to raise its yield, the instinct is to look at the process tools and the recipes that run on them. That is often where the answer is. But in a new fab, in a new country, a large share of yield problems start somewhere else: in the materials, chemicals, gases, water and services that feed the process.

Where yield losses come from

Yield is lost when defects or variation damage the circuits on a wafer. Some of that comes from the process recipes and tools. Much of the rest comes from inputs:

Chemical purity. Trace metals or particles in a wet chemical can contaminate every wafer it touches. Purity specifications are measured in parts per trillion.
Gas quality. Moisture or impurities in process gases affect film quality and etch results.
Lot-to-lot consistency. A slurry, resist or chemical that is in specification but varies from one lot to the next creates variation that is hard to trace.
Parts and consumables. Poorly cleaned chamber parts and worn quartz shed particles.
Utilities. Ultrapure water quality, power stability and exhaust performance all show up in yield data.
Response time. A tool that waits a week for a spare part or service engineer is not producing, and a rushed restart after a long wait carries its own risk.

Why new fabs are more exposed

Mature fabs in established clusters have spent years tuning their supply base. They know which suppliers are reliable, they have long data histories on each material, and they have engineers at the supplier who understand their process. A new fab in a new location starts without most of that. Its inputs often travel long distances, through more hands, with less history. When something drifts, it takes longer to find out why and longer to fix.

In a new fab, the supply chain is part of the process. It should be managed with the same discipline.

Managing supply as a yield lever

Qualify for consistency, not just specification. Ask suppliers for lot-level data and statistical capability, not only certificates of analysis.
Connect supplier data to yield data. Link material lot numbers to wafer histories so that a yield excursion can be traced to an input quickly.
Monitor the subfab continuously. Chemical delivery, gas systems, abatement and water treatment can be monitored in real time. Drift found there costs far less than drift found at wafer sort.
Shorten the distance. Local supply and service do not just save freight. They cut the time from problem to answer, which is what drives yield learning.
Plan for the ramp, not the steady state. During ramp, consumption changes quickly and qualification activity is intense. Supply plans should be built for that, with buffers where the risk is highest.

The role of AI

Fabs generate enormous amounts of data from inspection, metrology, tools, the subfab and suppliers. Much of it sits in separate systems. Machine learning is increasingly used to classify defects, detect drift early and find links between inputs and yield that people would not see. The value comes from connecting the data, and the subfab and supply chain are where the connection is most often missing.

Our view

Every new fab plans its yield ramp. Fewer plan the supply chain as part of it. For India's first fabs, treating materials, chemicals, utilities and services as yield levers from day one could be worth months of ramp time. That is time, and money, that no amount of tool tuning will recover later.