What if humanity forgot how to make CPUs?

Original Article ↗ Hacker News Discussion ↗

Plausibility of “forgetting” how to make CPUs

  • Many see the premise as unrealistic: to fully lose global CPU manufacturing would require extreme events (world war, civilization-scale collapse, or a magical “no more silicon” rule).
  • Critics argue that if we’re in a state where no CPUs can be made for decades, the deeper problem is societal collapse, not AWS uptime.
  • Others treat it as pure SF / thought experiment: its value is in exploring hardware lifetimes and dependencies, not realism.

Impact on civilization and population

  • Debate on how dependent 8B people are on CPUs for food production and logistics. Some think we can’t support current population without them; others think we’d adapt at lower scale.
  • Emphasis that the most critical chips aren’t in phones or PCs but in power grids, factories, transportation, and healthcare.
  • Historical analogies: Rome, “Dark Ages,” WWII logistics, and lost techniques show that advanced knowledge can disappear while civilization continues, but restarting now would be harder due to depleted easy fossil fuels and higher complexity.

Reinventing computation and fallback tech

  • Many argue we’d quickly recreate 1970s–1990s-level CPUs because:
    • We know they’re possible and have abundant documentation, die shots, and theory.
    • Numerous non–cutting-edge fabs (e.g., 180–28 nm, microcontrollers) already exist.
  • Others stress tacit, undocumented process knowledge and the singularity of EUV / ASML-like tooling; recreating cutting edge might take decades.
  • Several note fallbacks: vacuum tubes, relays, electromechanical computers, older lithography, alternative semiconductors (e.g., non-silicon materials).
  • Even crude CPUs (6502-class) or discrete-logic machines can run robots and bootstrap more factories.

Institutional knowledge and supply-chain fragility

  • Strong theme: institutional know‑how is brittle. Examples from other domains:
    • Difficulties restarting closed manufacturing lines (appliances, weapons materials, bluing formulas, CRTs).
    • Specialized processes lost or hard to reproduce (FOGBANK, CRTs, high‑end tape/CD players).
  • Concern that heavy reliance on a few private companies (TSMC, ASML, others) makes advanced nodes a strategic single point of failure, even if older nodes persist.

Hardware longevity and decay

  • Several posts focus on failure timelines:
    • Consumer gear often starts failing after 10–15 years (capacitors, plastics, drives).
    • Large-feature chips resist electromigration, but true lifetimes for 5/3 nm are unknown.
    • Keeping complex automated equipment idle often ruins it if shutdown wasn’t designed/documented.
  • Consensus: there’s enough existing hardware to bridge at least a decade or two, likely enough to rebootstrap some semiconductor capability, but not indefinitely.

Broader reflections

  • Some think a constraint on new CPUs might finally force efficient software and less wasteful computing—though others reply that CPU frequency scaling has already mostly stalled.
  • Several highlight that digital records and infrastructure are far more fragile than we assume; analog artifacts (e.g., ancient scrolls) may outlast our data centers.