bits.

FREE HISTORY OF COMPUTING COMPANIES LESSON

1976: No nation builds a chip alone

Japan, TSMC, ASML, Arm, and recursive specialization

The semiconductor timeline in this case spans decades. The 1976 opening refers to coordinated Japanese semiconductor research; ASML's origins, TSMC's1987 founding, and Arm's1990 establishment came later. They represent different ways to divide a demanding production system. A chip design, a fabrication process, a lithography tool, and a processor instruction set are related capabilities, but they are not interchangeable products.

Begin with a simplified path. A designer chooses architecture and logic, uses design tools and process libraries to produce a physical layout, submits a design for fabrication, then packages and tests the resulting dies. Each boundary has specifications: timing models, design rules, masks, electrical interfaces, thermal limits, and test expectations. A defect in any boundary can make an otherwise sound design unusable.

TSMC's dedicated-foundry account emphasizes manufacturing for customers rather than selling a directly competing portfolio of chip designs. Arm's history describes a different boundary around processor designs and architecture. Those choices let some firms participate without owning an entire fabrication chain. They also create dependence on the accuracy, availability, and terms of the shared interfaces. A design licensed from one firm is not automatically portable to every process or manufacturing site.

Use a fictional wafer batch costing 100000 units and containing 1000 candidate dies. At 80% yield,800 good dies share the batch cost, or 125 each before packaging. At 50% yield, the same expenditure produces 500, or 200 each. A small change in defects can therefore materially affect cost. But maximizing yield alone is not the objective: a larger or faster design may command more value, and a cheaper process may not meet power or performance constraints.

Capacity is another boundary. A customer can finish a design while lacking a manufacturing slot, suitable packaging, test equipment, or sufficient power at its destination. Buying more design software does not remove a bottleneck in physical production. Equally, spare factory capacity is not a substitute for an unqualified process or missing engineering knowledge. Represent capacities by stage rather than describing the entire industry with one number.

Specialization can compound learning. A tooling supplier learns across many customers; a foundry learns from many designs; a designer can concentrate on a particular workload. It can also concentrate risk. A substitute is credible only if it meets the required specification, can produce sufficient volume, and can be qualified within the needed time. Two suppliers on a slide are not necessarily two independent recovery paths if both rely on the same upstream component.

The title should therefore be read as a systems question, not a claim that national boundaries alone determine technical competence. Map dependencies at the level of processes, equipment, skills, standards, and logistics. Corporate accounts are primary evidence of each firm's stated role, but a resilience analysis requires independent evidence about substitutability and shared dependencies. A useful answer identifies a concrete bottleneck and explains what it would take to replace it.

Reconstruction exercise

Calculate good-die cost for the fictional batch at 80% and 50% yield. A second factory offers nominally equal capacity but requires six months to qualify the design, while the product must ship in eight weeks. Explain whether it is an immediate substitute. Draw a four-stage dependency chain and identify a shared upstream dependency that can make two nominal suppliers fail together.

Worked solution

At 80% yield the batch produces 800 good dies, costing 125 units each before packaging; at 50% it produces 500, costing 200 each. The second factory is a possible longer-term option but cannot satisfy the stated eight-week deadline if qualification genuinely takes six months. A chain could be design tools and process libraries → fabrication → packaging → testing. Two fabrication suppliers might share a crucial equipment or materials source, or both require the same packaging provider. Independent names do not establish independent failure exposure. A stronger resilience plan records qualification lead time, usable capacity, compatible process requirements, and upstream overlap rather than simply counting suppliers.

Test the limits and transfer the idea

1. Capacity calculation · allow 25 minutes

Extra wafers do not remove a packaging bottleneck

A fictional chip supply chain fabricates 1000 candidate dies per week with 80% yield. Packaging can attempt 600 per week with 95% final yield. Demand is 700 shipped chips per week. Compute steady shipped capacity and backlog growth. Option A doubles fabrication; option B raises packaging attempts to 800 with the same final yield. Compare both. State the role of unfinished inventory and why a foundry's nominal capacity does not equal completed-product capacity.

Get a starting hint

Apply each stage's yield before comparing its output with the next stage's input limit.

Compare your reasoning with the worked solution

Fabrication supplies 800 good dies, but packaging accepts only 600, yielding 570 shipments. Customer backlog grows by 130 per week if demand stays 700 and no stock fills the gap; unprocessed good dies also accumulate unless fabrication is throttled. Doubling fabrication still leaves 570 shipments because packaging remains limited to 600 attempts. Packaging at 800 attempts yields 760 shipments, enough for current demand and potentially 60 units of backlog reduction per week. This assumes all other stages and product mixes are compatible. Corporate profiles explain firms' stated roles, while actual throughput needs yield, qualification, equipment and logistics records. A count of wafers or licensed processor designs cannot establish the availability of a shipped chip.

Check your reasoning

  • Computes 570 shipments and 130 weekly unmet demand.
  • Locates the bottleneck after accounting for yield.
  • Distinguishes intermediate inventory and finished-product capacity.

2. Probability and dependency analysis · allow 25 minutes

Two suppliers can share one failure domain

Two fictional suppliers each depend on the same upstream tool service. It fails during the planning period with probability 0.02, stopping both. Conditional on that service working, each supplier independently fails with probability 0.03. Calculate the probability that both suppliers are unavailable. Compare with incorrectly treating their unconditional failure events as independent. Explain why the Noyce patent and corporate role descriptions cannot supply these failure probabilities.

Get a starting hint

Condition on the common service first; do not multiply away an explicitly shared cause.

Compare your reasoning with the worked solution

Both fail with probability 0.02+0.98×0.03²=0.020882, or 2.0882%. Each supplier's marginal unavailability is 0.02+0.98×0.03=0.0494. Multiplying those marginals gives 0.00244036, about 0.2440%, a much smaller and unjustified estimate. The shared event creates dependence. Independent supplier ownership is therefore insufficient evidence of independent recovery paths. The numbers are stipulated teaching inputs, not industry measurements. A primary patent documents a disclosed device approach; corporate profiles describe business roles. Estimating joint operational risk requires incident histories, upstream dependency mapping and qualification evidence. Even those data may underrepresent rare common events, so sensitivity analysis should accompany the estimate.

Check your reasoning

  • Conditions on common-service success and failure.
  • Computes correct and naive joint probabilities.
  • Separates institutional role evidence from measured failure dependence.

Check your understanding

A second chip factory has enough nominal capacity but needs six months to qualify a design required in eight weeks. Is it an immediate substitute?

  • Yes; the designer can waive every electrical constraint.
  • Yes; capacity figures alone establish interchangeability.
  • No; two factories can never make compatible chips.
  • No; qualification lead time prevents it meeting the stated deadline.
Explanation

Substitution depends on specification, usable capacity, and time, not just the existence of another supplier.

Two chip suppliers depend on the same specialized packaging plant. What does this imply for resilience?

  • A third sales contract automatically removes the dependency.
  • They share a failure dependency that supplier count alone hides.
  • The packaging plant cannot affect chip availability.
  • Their failures must be statistically independent.
Explanation

A dependency graph must include upstream and downstream common bottlenecks.

Sources and further reading