Wafer fabs: yield learning on the most expensive floor in industry

Fixed cost means a fab cannot afford to idle

Depreciation, cleanroom operation, chemical supply and skilled staff continue whether or not wafers move, so under-loading destroys margin faster than price competition does. That drives behaviour outsiders find strange: continuing to run when prices fall below full cost, taking low-margin filler work to hold utilisation, and pricing long capacity agreements aggressively. It also makes the decision to build nerve-racking, since a fab commissioned into a downturn carries full cost while earning little. Capacity announcements consequently cluster, and the oversupply that follows is a structural feature of the industry rather than a forecasting failure.

Yield learning is the first year's real output

A new process does not arrive working. Early wafers yield poorly, and improving that through defect reduction, tool tuning and design adjustment is where value gets created. The learning is cumulative, partly tacit, and difficult to transfer between sites even inside one company, which is why a copied fab still takes time to match its parent. It also explains the sensitivity around engineering staff and the reluctance to share process data. For a buyer, the implication is that a supplier running a mature process at a mature site quotes from a fundamentally different cost position.

Contamination control shapes every building decision

Particles invisible elsewhere destroy devices here, so air handling, water treatment, gas distribution, vibration isolation and gowning discipline form part of the process rather than facilities overhead. Ultrapure water and specialty gas supply must be continuous and on specification, which turns utility reliability into a production issue: a brief power disturbance can scrap work in progress worth far more than the disturbance suggests. Site selection follows, favouring stable power, abundant treated water, seismic stability and a trained workforce, and those attributes are considerably scarcer than open land with good road access.

Foundry, integrated and fabless are distinct businesses

An integrated manufacturer designs and makes its own products, capturing more value while carrying the full capital burden. A foundry sells process capacity to others and needs both a broad customer base and a design ecosystem around it. A fabless firm designs and buys manufacturing, avoiding capital but competing for capacity whenever the market tightens. Each is coherent, with different cash characteristics and different failure modes. The commonest strategic error is drifting between them: keeping a fab that no longer justifies investment while trying to outcompete firms whose capital goes into design.

Where the cash goes, and when any of it returns

Money leaves years before revenue arrives. A fab is committed, built, equipped, qualified and ramped over a long period, and recovery depends on holding high utilisation across a cycle that will certainly include a downturn. Equipment dominates the cost, and much of it cannot be moved to a different process without significant expense. Assessing such a business means examining the depreciation profile, how much capacity sits committed under long agreements, and whether the process portfolio spans several end markets instead of depending on one that happens to be strong. Dependence on a single dominant customer is what most often turns a good fab into a stranded asset.

Frequently asked questions

Why do chip shortages and gluts alternate so sharply?

Because capacity is added in large indivisible increments after a long lead time, while demand signals arrive quickly and are amplified by customers ordering defensively. By the time new capacity produces, the shortage that justified it may have passed, and the fixed cost structure means everyone keeps running rather than idling. Inventory held throughout the chain then unwinds at once. The cycle is structural, following from long build times, fixed costs and the impossibility of storing capacity for later use.

Can a fab switch between different types of product?

Within a process family, yes, and foundries do so constantly. Between substantially different processes it is much harder, because tool sets, materials and contamination controls differ and requalification takes time during which the fab earns nothing. Some equipment can be repurposed and some cannot. This is why older fabs keep producing mature products for many years instead of being upgraded, and why a shortage in one process type cannot be relieved by spare capacity in another.

What does qualifying a new fab as a supplier involve?

For most customers, technical qualification of the process against their device requirements, reliability testing, and assessment of the supplier's quality and change control systems. Automotive and medical buyers add further requirements and expect notification and approval before any process change. The exercise runs from months to years depending on the product, which is why second-sourcing silicon is planned long before it is needed and why customers resist moving volume once a source is qualified.