EDGEBIC for Semiconductor Scheduling: Recipe Changeovers

EDGEBIC handles semiconductor scheduling by pricing every recipe changeover on a shared tool in a sequence-dependent setup matrix, so the scheduler knows that switching a lot from one process to another is not a fixed cost but depends entirely on what ran before, and sequences lots to stop burning capacity on avoidable requalification. EDGEBIC by User Solutions brings the same finite-capacity discipline that has served demanding manufacturers since 1991 to the recipe-changeover problem that quietly drains fab capacity. This post shows how.

The Changeover Problem in a Fab

In a wafer fab, the expensive resource is the tool, and the hidden thief of tool time is the recipe change. Switching a track, an etcher, or a diffusion furnace from one process to another is rarely a fixed cost. Going from a light process to a heavy one might mean a short gas purge. Going the other way might mean a full clean, a season run, and a requalification that eats hours.

A scheduler that treats changeover as a single flat number cannot see this. It assumes every lot-to-lot transition costs the same, so it sequences lots by due date or arrival and lets the real changeover cost fall out however it falls. On a tool running a dozen lots a shift, that indifference to sequence can silently cost hours of capacity every day. The tool looks busy. It is actually spending a chunk of its time purging and requalifying because nobody sequenced the recipes to avoid it.

The Sequence-Dependent Setup Matrix

EDGEBIC's answer is a setup matrix defined per tool. Instead of one changeover number, you record the cost of going from each process to each other process. The matrix captures the asymmetry that a flat number cannot:

Once these real numbers are in the matrix, the scheduler stops guessing. When it places the next lot on a tool, it reads the actual cost of transitioning from whatever ran last, and it can sequence lots to avoid the expensive transitions. This is the same mechanism explained in general in the EDGEBIC setup matrix, applied to recipes instead of paint colors or alloys.

Setup Families Keep the Matrix Manageable

A fab may run hundreds of recipes. A full recipe-by-recipe matrix would explode into tens of thousands of cells nobody could maintain. EDGEBIC solves this with setup families, the concept covered in what is a setup family.

You assign each recipe to a family, then record changeover times at the family level. Light-field recipes go in one family, dark-field in another, contamination-sensitive processes in a third. The matrix becomes a small grid of family-to-family transitions instead of an unmanageable recipe-by-recipe explosion. Lots in the same family incur little or no changeover; crossing a family boundary incurs the family cost. Where a specific recipe pair genuinely differs from its family, you add a product-level override for just that pair. This is exactly how a fab campaigns compatible recipes together in practice.

A Worked Example: Sequencing a Track's Shift

Consider a coat-and-develop track scheduled to run six lots in a shift, using two recipe families: light-field and dark-field. The family changeover costs are 0 minutes within a family, 60 minutes going light to dark, and 240 minutes going dark to light because the reverse requires a clean and requalification.

Suppose the lots arrive in an order that alternates families:

Light, Dark, Light, Dark, Light, Dark

With a flat changeover assumption, the scheduler would happily accept that order and quietly incur: 60, then 240, then 60, then 240, then 60. That is 660 minutes, eleven hours, of changeover crammed into a shift that is only eight hours long. The schedule would look feasible on paper and be impossible on the floor.

EDGEBIC, reading the matrix, knows the true cost of each transition. It can group the lots by family and order them light-first:

Light, Light, Light, Dark, Dark, Dark

Now the changeovers are 0, 0, then a single 60-minute light-to-dark transition, then 0, 0. Total changeover: 60 minutes. The same six lots that could not fit an eight-hour shift now fit comfortably, because the sequence stopped ping-ponging across the expensive boundary. That recovered time is real tool capacity, and it came from nothing but ordering the lots correctly.

Multi-Step Routings and Lot Streaming

Semiconductor lots do not visit one tool. They flow through photolithography, etch, diffusion, deposition, and metrology, often circling back through the same tool sets across many layers. EDGEBIC models this as a multi-step bill of routing, with each step targeting the right tool or tool group, and schedules the steps in dependency order while respecting each tool's finite capacity.

It also handles the reality that a lot does not have to finish a step entirely before the next step begins. With lot streaming, you define a transfer batch so a downstream step can start once the first sublot is ready. If a diffusion step processes a lot of 25 wafers and the next step can begin after the first 5 are through, the downstream tool starts earlier and the lot's total cycle time falls. EDGEBIC composes that overlap with the queue and transit times between tool sets, so the schedule reflects how wafers really move rather than an all-or-nothing handoff. The general mechanism is covered in EDGEBIC lot streaming.

Shared Tools and Tool Groups

Many fab tools are interchangeable within a class: several etchers that can run the same recipes, a bank of identical furnaces. EDGEBIC represents these as work center groups, a pool of interchangeable tools that a routing step can target instead of a single named tool. The scheduler picks a member of the pool based on the group's strategy, whether that is earliest completion or a designated primary tool first, and never overbooks a tool. This keeps lots moving when one tool in the class is busy, without forcing a planner to hand-assign every lot to a specific chamber.

Why This Matters for a Fab

The economics of a fab are the economics of tool time. A tool that spends its shift requalifying because lots were sequenced carelessly is a tool that is not producing wafers, and in a capital-intensive fab that lost time is expensive. The value of pricing changeovers correctly is not abstract. It is the difference between a schedule that fits the shift and one that quietly cannot.

By capturing the true, asymmetric cost of every recipe transition, keeping it manageable through families, and letting the scheduler sequence lots to avoid the costly transitions, EDGEBIC turns changeover from an invisible tax into a variable the schedule actually controls. Combined with multi-step routings, lot streaming, and interchangeable tool pools, it gives a fab a schedule that respects both the physics of recipe changes and the finite capacity of the tools. For a deeper picture of how the same engine handles other recipe-driven environments, the general semiconductor scheduling overview is a useful companion.

The starting point is always the same: get the real changeover numbers into the matrix. A fab that has been losing hours to unsequenced recipe changes usually finds that its biggest capacity gain is not a new tool. It is sequencing the tools it already has.