Benjamin Arce
← Work
2025-09

Choke Point: CoWoS/HBM Thesis Log

Tri-objective sim–opt model of the resilience–decarbonization trade-off at the CoWoS/HBM packaging chokepoint.

thesissemiconductorsresearch

This is my long-game project: less a polished product, more a running log of building something honest and narrow.

Quick scan
What it is A tri-objective simulation–optimization model focused on the CoWoS/HBM advanced packaging chokepoint, under disruption scenarios, optimizing cost vs resilience vs carbon.
Why it matters A lot of resilience literature ends with “just collaborate and share data.” For semiconductors, that’s often unrealistic. The thesis flips the assumption and asks what one enterprise can do with private levers.
Payoff Instead of a generic framework, the goal is a numeric frontier: what trade-offs are possible, what levers move the frontier, and where resilience directly conflicts with decarbonization.

What this actually is (plain language)

I’m trying to answer a specific question: if the world keeps depending on Taiwan for advanced packaging, how do we buy resilience without dumping the cost onto the climate?

More concretely: model HBM suppliers → CoWoS packaging sites → downstream demand, under disruptions (outages, spikes, logistics shocks), with three conflicting objectives: cost, resilience, and emissions.

  • Chokepoint focus: CoWoS / HBM advanced packaging
  • Disruption layer: outages, demand spikes, combined shocks
  • Objectives: total cost, service/recovery resilience, CO2e
  • Key assumption: no magical cross-firm data sharing

Why I cared enough to go this deep

I kept reading the same ending: “firms should increase transparency and collaborate.” It sounds nice, but it often ignores the politics and incentives in semiconductors.

So the thesis bakes in the disagreement: assume no magical coordination. Focus on levers a single firm can plausibly control at the chokepoint.

Where the thesis is right now

  • Motivation + context drafted (why CoWoS/HBM matters; why resilience vs decarb isn’t fake here).
  • Literature map sketched (quant resilience modeling, decarb multi-objective, semiconductor-specific resilience).
  • Research questions locked (frontier shape, lever sensitivity, conflict zones).
  • Current phase: turning sets/variables/objectives into clean math instead of vibes.

How I’m planning to attack it

  • Define a minimal but realistic network around the chokepoint (regions, capacities, lead times, risks, emission factors).
  • Optimization layer generates candidate designs along the Pareto frontier (regional split, safety capacity, mode mix).
  • Simulation layer stress-tests designs under disruption scenarios (service loss, recovery time, cost, emissions).
  • Carbon accounting: freight emissions via ton-km × mode factor; process emissions via energy/unit × grid intensity; optional carbon price scenarios.

What I expect to learn (and might be wrong about)

  • Resilience can improve surprisingly far with private levers (even without industry-wide coordination).
  • There are zones where resilience and decarb align (cleaner + diversified sites).
  • There are ugly zones where resilience means more cost and more emissions (especially with air freight under tight carbon pricing).

Technical appendix

  • Problem type: tri-objective simulation–optimization at a single chokepoint in a multi-echelon supply chain
  • Decision layer: regions r, time t, products p, modes m; decisions like x_pr (alloc shares), c_r (safety capacity), mode shares by route/time
  • Objectives: expected total cost; resilience metric (service loss / recovery time); life-cycle CO2 (process + logistics)
  • Solution idea: multi-objective heuristic (e.g., NSGA-style) to generate designs; simulation evaluates designs across scenarios
  • Emissions: logistics via ton-km × mode EF; process via energy/unit × grid intensity; optional carbon price shifts the frontier
← Back to Work