Scheduling at Full-Factory Scale
How INFICON solved the back-end assignment problem.
Back-end semiconductor manufacturing is a scheduling problem that is easy to underestimate until you work through the numbers.
A single lot in a large back-end factory may have hundreds of qualified tools (and in some cases more than a thousand possible tools) that can process it. Multiply that across thousands of work-in-progress lots, then add in changing priorities, tool downs, setup constraints, and customer delivery commitments, and the result can be more than 1 million possible lot-to-tool assignments before the main scheduler makes its first decision.
That scale creates a real operational problem. If the scheduler evaluates every possible assignment on every run, runtime grows to the point where the schedule is stale before it is finished. If assignments are cut too aggressively to bring runtime down, the scheduler loses the choices that matter for cycle time, throughput, on-time delivery, and setup stability. The search space must be made smaller, but it cannot be made smaller carelessly.
The Assignment Preprocessor
INFICON developed a production solution to this problem as part of its Intelligent Manufacturing Systems software suite. The Assignment Preprocessor runs ahead of the main factory scheduler and narrows the lot-to-tool assignment space before scheduling begins, preserving the candidates most likely to drive good factory outcomes while eliminating assignments that will not.
In production use at a large back-end factory in Asia, the preprocessor reduces more than 1 million possible lot-to-tool pairings to roughly 20,000 to 50,000 high-value candidates. It completes this in approximately 60 seconds. The scheduler then runs on fresher factory data, covers a tighter, more meaningful decision space, and produces results faster.
See INFICON at SEMICON West
Visit Booth 5268 to discover how INFICON solutions reduce yield loss, shorten cycle times, provide actionable process insights, and maximize equipment utilization.
How It Works
The preprocessor is not a simple filter. It combines live manufacturing data, factory-specific business rules, and optimization to construct a focused candidate set that reflects the full operational picture.
Lots are grouped by key characteristics including product, route, process step, and urgency. The system then builds a candidate assignment set that accounts for tool qualifications, current factory state, setup considerations, delivery priorities, and operational constraints. It captures what matters, not just what is technically possible.
A snapshot layer makes these candidates available when the scheduler runs, with automatic fallback logic to protect production robustness if optimized results are stale or unavailable for any reason. This is a production system, not a research tool, and it is designed to behave gracefully when conditions are not ideal.
What Production Deployment Produces
The results of this architecture are not abstract. Tighter candidate sets mean the scheduler spends its computational time on decisions that affect factory performance rather than evaluating assignments that were never going to be selected.
The downstream gains translate to tighter cycle time control, better on-time delivery performance, improved setup stability, and throughput improvements. In a back-end environment where customer delivery commitments are a direct business concern and equipment utilization drives profitability, those outcomes are measurable.
Back-end factories are entering a serious phase of modernization, driven by the growth of advanced packaging, heterogeneous integration, and high-volume OSAT operations. Better scheduling is a significant part of that modernization opportunity. But scheduling at back-end scale requires confronting the complexity that makes it hard in the first place. The Assignment Preprocessor is how INFICON addressed that problem in production, not in principle.