ADI Thesis — Semiconductor Supply Chain Resilience
Nine fabs, two models — how shutdowns turn into backlogs.
What happens when a fab goes down?
When a semiconductor fabrication plant shuts down unexpectedly, managers face cascading failures: die bank inventories drain, order backlogs spike, and recovery timelines remain unclear. Existing tools offered no fast, quantitative way to test which mitigation levers — buffers, capacity reallocation, surge production — actually work, and at what cost.
This project built two decision-support tools to give managers quantitative answers.
Dual-model framework
We built a dual-model framework. The first model was an Excel Solver-based linear programme that minimised total cost across 9 fabrication plants and 3 chip types. It compared two allocation modes: fixed-mix (rule-based) versus minimum-cost (Solver-optimised), enabling instant what-if scenarios for cost attribution and capacity redistribution.
The second model was an AnyLogic discrete-event simulation modelling 9 fabs, 3 product families, die bank inventories, and order fulfilment over time. It captured week-by-week backlog dynamics, inventory depletion, and recovery trajectories — the things a spreadsheet can't show you.
We ran four categories of disruption scenarios: varying shutdown durations, critical node failures, multi-fab simultaneous shutdowns, and different restart production rates. Each scenario produced quantifiable fulfilment rates, cost impacts, and recovery timelines.
The team of 4 worked collaboratively — I contributed to both models and all scenario analysis. We presented findings directly to Analog Devices operations leadership.
What the models revealed
Scenario Insights
- Fab 3 and Fab 5 identified as critical vulnerability nodes — their loss caused structural shortages the network couldn't compensate for.
- Simultaneous outages at two fabs proved worse than a single longer shutdown — simultaneity matters more than duration.
- Simply restarting a fab is insufficient. The rate of post-restart production determines whether recovery occurs at all (98% vs 58%).
- Inventory buffers provide non-linear protection. Optimal sizing is 2-3 weeks of demand coverage — beyond that, diminishing returns.
What we told ADI leadership
- Size die bank buffers at 2-3 weeks of demand coverage — balancing holding cost against fulfilment protection.
- Pre-qualify backup capacity for critical nodes (Fab 3, Fab 5) with pre-negotiated supplier contracts.
- Plan surge production rates post-restart — higher than baseline to clear accumulated backlog.
- Stress-test for simultaneous multi-fab disruptions, not just longer single ones.
- Deploy fab-specific control towers for real-time visibility and proactive reallocation.