IBM planned chips with a solvable MIP instead of a perfect impossible model

IBM Systems and Technology Group had to plan an extended semiconductor supply chain at a scale that made clean textbook optimization difficult. The obvious choices were both flawed: oversimplify the model until it solved, or build a faithful model that planners could not run.

The team kept the important integer decisions and designed heuristics around them. IBM Research says the model and method were used daily for operational and strategic planning and saved substantial costs.

This is a quiet supply-chain invention: not a new container or warehouse, but a decision machine that accepted the real discreteness of chip planning without letting computational scale defeat the business.

Keep the discreteness, solve around it

The model preserved discrete manufacturing choices that mattered operationally.

Three heuristics were driven by practical applications, not abstract elegance alone.

Daily use made the method part of planning, which is harder and more valuable than a one-time optimization run.

For messy industrial planning, genius is often choosing which exactness to keep and which computation to approximate.

The 2006 Interfaces paper won attention as a practice case, and RePEc lists later citations in semiconductor planning and supply-chain optimization. IBM Research keeps the publication in its archive with the DOI and abstract.