August 25, 2026
Process Selection White Paper - Phase II
PHASE II
How a Fabless Semiconductor Company Should Select the Right Foundry
A product-led framework for manufacturing partner selection
CORE IDEA
Hard constraints determine feasibility, while product differentiation determines what deserves optimization.
Semiconductor Manufacturing Strategy Series
Executive Summary
Foundry selection is a highly complex decision required after a semiconductor company has completed its architecture and chosen process node. The specific foundry selected among the many available process technologies has far reaching implications. This relationship embeds itself in every aspect of the product's success: performance, cost structure, qualification history, launch schedule, supply chain, and long-term strategic flexibility. One of my colleagues used to say choosing a foundry is like getting married: an ultra-high trust and practically irreversible relationship choice.
The correct starting point, which many fabless companies miss, is to start with the product-- not “the best fab”. Just as an ideal romantic partner is unique to the individual, the ideal fab is unique to each product being developed. Before comparing foundries, a fabless semiconductor company should be able to answer a basic question: why will customers choose this product instead of an alternative?
The answer may be higher efficiency, lower system cost, better performance-per-watt, greater integration, or another source of customer value – even cost alone. Those sources of differentiation should determine which foundry characteristics deserve the greatest attention.
This creates a Product Value Hierarchy. Hard constraints (table stakes) determine technical feasibility or market access. Differentiation drivers are variables whose improvement strengthens the reasons customers choose the product or improves the company's ability to capture value. Secondary considerations matter, but further improvement beyond a sufficient threshold creates limited customer or enterprise value.
| Governing principle: Filter hard constraints. Optimize differentiation drivers. Accept sufficiency on secondary considerations. |
|---|
Even differentiation drivers should not be optimized without limit. Additional technical performance is worthwhile only while its incremental value exceeds the incremental cost, schedule, and risk required to obtain it. The same is true when lower cost is the only differentiation driver. Value may be customer-facing, such as higher performance or lower system cost, or internal, such as improved gross margin, earlier revenue, lower working capital, or reduced supply risk. The company must also determine whether the value created can be captured through higher pricing, greater share, lower cost, faster launch, or another economic mechanism.
Incremental Value > Incremental Cost + Schedule Penalty + Incremental Risk
Phase II therefore extends the decision philosophy of process selection into supplier selection. The company should define why the product wins, translate differentiation into manufacturing requirements, eliminate infeasible platforms, optimize the variables most closely tied to value, quantify execution and economics, assess ecosystem and strategic risk, and resolve only the unknowns capable of changing the recommendation.
The cheapest foundry is not necessarily the best. The most technically advanced foundry is not necessarily the best. The supplier with the highest generic score across dozens of criteria is not necessarily the best. The correct foundry is the supplier and specific platform whose capabilities most effectively support the complete product and business strategy.
Part I - Product Value and Selective Optimization
1. Start With Why the Product Wins
Foundry selection should begin before the first supplier comparison is created. A high-performance accelerator may win because it delivers more compute per watt, a power-management IC may win because greater integration reduces external components and thermal burden, etc.
These products require different manufacturing priorities even when several foundries can technically fabricate them. The accelerator may justify significant investment in transistor performance, SRAM, interconnect, advanced packaging, or digital IP depending on the architecture. The power-management IC may place greater value on high-voltage devices, analog performance, yield, reliability, or lifecycle support.
Unique selling points do not need to be purely technical either. Perhaps the business relationship is truly why the product wins. Maybe the fabless company can deliver better personal relationships, better cost, or better geopolitical stability. All these are valid unique selling points that can be optimized by foundry selection.
| Key question: Superior at what, and why does that superiority matter to this product? |
|---|
2. Separate Table Stakes From Differentiation
An attribute can be essential without being differentiating. Automotive qualification, functional-safety requirements, operating-temperature range, regulatory compliance, and required interfaces may be necessary simply to participate in a market. Once every credible competitor meets the threshold, further optimization may create little additional customer preference.
Differentiation drivers operate differently. If one automotive power IC reduces losses enough to shrink the cooling system or eliminate external components, the improvement can directly affect the customer's purchasing decision. Secondary considerations matter operationally but have limited influence beyond a sufficient level.
| Importance != differentiation. Table stakes determine whether the product can compete; differentiation determines where additional optimization may help it win. |
|---|
3. Translate Product Value Into Manufacturing Requirements
Product differentiation should not map directly to a foundry name. It should pass through a traceable chain from customer need, to product differentiation, to technical requirement, to process requirement, and finally to foundry requirement.
Consider a power-management IC whose value proposition is lower system cost. It may deliver that value through greater integration, requiring high-voltage devices, analog precision, isolation, and additional on-chip functions, etc. Phase I selects a BCD process family accordingly. Phase II then compares the foundry platforms capable of delivering the required devices, performance, yield, qualification, economics, and product lifetime. There are dozens of BCD processes available in the market – a fabless company would ideally analyze each that passes hard constraints to arrive at the optimal option to maximize a fabless companies’ success.
| Product differentiation | Foundry variables likely to matter |
|---|---|
| Performance-per-watt | Transistor efficiency, leakage, SRAM, interconnect, packaging |
| Lower system cost | Integration, die size, yield, package and test economics |
| Faster launch | PDK maturity, IP readiness, support, cycle time |
| RF performance | Devices, substrate, passives, model accuracy |
| Smaller footprint | Density, integration, packaging compatibility |
| Long lifecycle | Process longevity, fab commitment, equipment support |
Exhibit 1. Product differentiation translated into manufacturing priorities
4. The Product Value Hierarchy
Tier 1 - Hard Constraints
Hard constraints define feasibility. Fabless companies often explore foundries forgetting to separate hard and soft constraints. A product cannot be made if it does not satisfy hard constraints! Examples include required voltage, mandatory automotive qualification, substrate, export-control status, required memory or IP, and minimum lifecycle support. A candidate that fails a genuine hard constraint should be ruthlessly eliminated rather than compensated by strengths elsewhere.
Tier 2 - Differentiation Drivers
Differentiation drivers deserve selective optimization because improvement may strengthen customer or enterprise value. This is where advanced multi-variable optimization plays an important role. Differentiation drivers should be scored on importance to the end product’s unique selling points and optimized accordingly across different platforms. Examples include RDS(on) for a power IC, leakage for an always-on product, compute density for an accelerator, passive performance for RF, cycle time when early launch has substantial value, or qualified product cost in a price-sensitive market.
Tier 3 - Secondary Considerations
Secondary considerations should satisfy the product requirement without being maximized. These are soft requirements with lower importance rankings, but nonetheless may help create a superior product. Examples excessive capacity beyond forecast demand, transistor matching, or peak digital performance in an analog-limited product. It is tempting for different internal teams to mis-prioritize secondary considerations based on their own internal goals. For example, procurement may place over importance on cycle time, while engineers may claim device matching of ultra-high importance, when both simply make each other’s lives easier and do not pass much value to their end customer.
5. Avoid Over-Optimization and Manage Tradeoffs
Even a differentiation driver eventually reaches diminishing returns. An efficiency improvement that eliminates a heatsink may create substantial value. A much smaller improvement beyond that point may be technically impressive but commercially irrelevant. Products also have multiple differentiation drivers that can conflict: better performance may increase power, greater integration may increase die area, and a technically superior foundry may require higher NRE, a slower launch, or greater supply concentration.
The company must therefore optimize the complete product proposition rather than isolated specifications. It should also distinguish value creation from value capture. A technical improvement creates customer value when it makes the customer's system better. The semiconductor company captures that value only if it contributes to concrete benefits, such as higher ASP, greater market share, lower cost, earlier revenue, better margins, etc.
Part II - Build the Foundry Decision Architecture
6. Phase I Ends Where Phase II Begins
Phase I asks what process technology best supports the product's requirements and differentiation. Phase II asks which foundry and specific process platform should deliver that technology. Semiconductor companies today often place these two phases together, wasting resources as they engage foundries, or favoring a specific foundry and forgetting hard constraints and under optimizing USPs. A conclusion such as 130-180 nm BCD, 40-65 nm automotive mixed signal, 28 nm CMOS, or a mature leading-edge digital platform leaves meaningful supplier choice.
Phase I should provide Phase II with the customer value proposition, differentiation drivers, market-entry requirements, hard technical constraints, selected technology family, priority optimization variables, cost and time-to-market sensitivity, and product-lifecycle assumptions. These outputs prevent foundry selection from reopening decisions already resolved during process selection or allowing a preferred supplier to bias the technology choice.
| Product Strategy |
|---|
| Technology Requirements |
| Process Technology Selection |
| Foundry Requirement Specification |
| Foundry Selection |
Exhibit 2. Phase I to Phase II handoff
7. Build the Foundry Requirement Specification
The first formal output of Phase II should be a Foundry Requirement Specification. For each important criterion, it should record the requirement, classification, relationship to customer or enterprise value, minimum acceptable threshold, desired optimization direction, source confidence, and decision sensitivity.
| Requirement | Classification | Value link | Decision treatment |
|---|---|---|---|
| 60 V devices | Hard constraint | Enables architecture | Must pass |
| Automotive qualification | Table stake | Market access | Must pass |
| RDS(on) | Differentiation driver | Efficiency and thermal | Optimize |
| Analog performance | Differentiation driver | Product performance | Optimize where valuable |
| Wafer price | Economic variable | Margin | Evaluate with yield and TCO |
| Logic density | Secondary | Limited product impact | Sufficient capability |
| Multi-site production | Risk consideration | Continuity | Optimize only if justified |
Exhibit 3. Illustrative Foundry Requirement Specification
The specification separates feasibility from preference, establishes where sufficient performance begins, and makes uncertainty visible. A parameter should not appear as a number without indicating whether it is confirmed, supplier-reported, customer-reported, modeled, or unknown. An uncertain variable may not deserve further investigation if different answers would not change the recommendation.
8. The Foundry Selection Funnel
The Foundry Selection Funnel defines the sequence of the decision and reduces time by preventing deep diligence on candidates that should have been eliminated earlier.
1. Product Value Definition - define table stakes, differentiation drivers, secondary considerations, and enterprise objectives.
2. Technology Output From Phase I - establish the selected technology family, device requirements, performance envelope, reliability needs, and lifecycle assumptions.
3. Foundry Universe - identify all potentially viable suppliers and specific process platforms. Compare platforms, not merely company names.
4. Hard-Constraint Screening - remove candidates that fail genuine non-negotiable requirements; preserve unknown as a valid status when diligence is required.
5. Differentiation-Driver Evaluation - compare the parameters most closely linked to product value and spend less effort on secondary criteria.
6. Manufacturing and Economic Evaluation - assess yield, maturity, qualification burden, NRE, cycle time, accessible capacity, and qualified product cost.
7. Ecosystem and Supplier-Fit Evaluation - evaluate PDK maturity, IP, engineering access, customer importance, commercial flexibility, and escalation.
8. Strategic Risk Evaluation - assess concentration, process and fab longevity, disruption exposure, migration options, and recovery strategies.
9. Decision-Weighted Diligence - resolve the unknowns most likely to change the recommendation.
10.Commercial Diligence and Final Selection - validate terms and recommend the preferred foundry, assumptions, protections, and contingency actions.
9. Six Foundry Evaluation Dimensions
The funnel defines when questions should be asked. Six evaluation dimensions define where the analysis occurs.
| Dimension | Representative questions and variables |
|---|---|
| Technical Fit | Devices, performance, voltage, analog and RF capability, memory, design rules, metal stack |
| Manufacturing Execution | Yield, maturity, ramp confidence, capacity, cycle time, qualification, production stability |
| Product Economics | Wafer cost, good-die cost, NRE, masks, IP, package, test, inventory, schedule economics |
| Ecosystem and Enablement | PDK, models, EDA, IP, design partners, packaging, test |
| Supply and Strategic Risk | Concentration, roadmap, equipment lifecycle, longevity, migration, recovery |
| Commercial and Organizational Fit | Support, customer priority, pricing model, commitments, escalation, negotiating leverage |
Exhibit 4. Six foundry evaluation dimensions
Part III - Feasibility, Technical Fit, and Manufacturing Execution
10. Hard Constraints Are Filters
True hard constraints should be pass/fail. A foundry that cannot meet a mandatory technical, qualification, regulatory, or lifecycle requirement should not survive because it performs well elsewhere. Before eliminating a candidate, however, the team should challenge whether the claimed constraint is genuinely immutable. Embedded memory, package choice, integration level, voltage partitioning, or preferred geography may sometimes be architectural tradeoffs rather than absolute requirements.
| Do not average away true constraints, but do not mistake inherited preferences for immutable requirements. |
|---|
11. Compare Specific Platforms at the Parameters That Matter
Nominal node labels do not establish technical equivalence. Two 180 nm BCD platforms may differ in high-voltage devices, isolation, analog options, passives, memory, design rules, and qualification. Two mature leading-edge platforms may differ in transistor efficiency, SRAM, interconnect, standard cells, metal options, packaging, and IP. The company should compare specific process platforms and focus most deeply on parameters connected to differentiation.
12. Yield, Manufacturing Maturity, and Ramp Confidence
A technically attractive platform can still be a poor manufacturing choice. Yield affects unit cost, wafer consumption, capacity requirements, ramp duration, working capital, and supply predictability. Manufacturing maturity should be evaluated through production history, process-change frequency, qualification experience, equipment stability, installed manufacturing base, and performance on similar products.
Companies should ask themselves how much uncertainty remains between tapeout and stable production. The analysis should distinguish confirmed data, supplier-reported data, customer-reported data, modeled estimates, and unknowns. Ranges and sensitivity analysis are more credible than unsupported precision.
13. Qualified Product Economics
Wafer price should be translated first into good-die economics.
Good Die Cost = Wafer Cost / (Gross Die per Wafer x Yield)
If two foundries each produce 500 gross die per wafer, a $2,000 wafer at 75% yield (exaggerated) produces a simplified good-die cost of approximately $5.33, while a $2,300 wafer at 92% yield produces approximately $5.00. The higher-priced wafer produces the lower good-die cost. Yield and die size are also inextricably tied, larger dies means less yield, and visa versa. The unit economics should consider die size comparisons between nodes and the resulting yield implications.
The comparison should ultimately move toward qualified saleable product, including package, test, scrap, IP, masks, qualification, engineering resources, inventory, and schedule effects. The objective is not maximum model complexity. It is to identify the inputs capable of changing the decision.
Qualified Product Cost = Total Manufacturing and Allocated Program Cost / Qualified Saleable Units
| The economically relevant unit is the qualified saleable product, not the wafer. |
|---|
14. Capacity Accessibility
Foundries categorize capacity terms differently. Installed capacity is the physical capability of the fab. Available capacity is not currently utilized. Uncommitted capacity is not already allocated. Customer-accessible capacity is what the specific customer can realistically obtain. The final category matters most.
A small customer may have less supply security at a very large foundry than at a smaller supplier where its business receives greater priority. The objective is not maximum capacity. It is enough accessible capacity to support expected demand and realistic upside with acceptable allocation risk.
15. Time-to-Market
Foundry selection ties the product to many aspects that directly impact TTM: PDKs, IP availability, engineering wafer access, tapeout support, fab cycle time, respin timing, qualification, packaging, and test readiness. Schedule differences should be translated into business consequences. A supplier offering a $0.20 lower unit cost may appear attractive, but if it delays a two-million-unit-per-year product by three months, the lost contribution margin can exceed several years of manufacturing savings.
The actual value of time depends on the product. A few months may be manageable in a stable industrial market and devastating in a fast-moving compute market. Time-to-market should therefore be treated as an economic variable, not a generic score.
16. Qualification Burden
Mandatory qualification feasibility belongs in the hard-constraint filter. Among viable candidates, the company should compare qualification burden, including reliability testing, AEC and PPAP activity, package qualification, characterization, customer approval, engineering resources, schedule, and direct expense.
Qualification history can make one foundry materially easier to use even when several platforms are technically capable. It also increases future switching cost because migration may require renewed characterization, reliability work, package validation, and customer approval.
Part IV - Economics, Ecosystem, and Supplier Fit
17. Total Product Economics
Foundry economics should be divided into four categories. Upfront costs, variable costs, operational costs, and strategic costs. Upfront costs include masks, process NRE, IP, design enablement, characterization, qualification, and tooling. Variable costs include wafers, probe, packaging, final test, scrap, and yield losses. Operational costs are the costs throughout the duration of the project which the price may be fixed such as design resources, respins, yield learning, inventory, working capital, schedule delay, and supplier-management complexity. Strategic costs quantify risks such as migration, requalification, process discontinuation, supply interruption, and roadmap changes.
The goal is not to assign a perfectly precise value to every possibility. It is to identify which economic differences are decision-sensitive. For a high-volume product, a few cents of qualified product cost can become strategically significant. For a low-volume specialty product, upfront costs of masks and qualification may play a larger role in total product economics than variable production costs such as wafers.
18. Economics Change Significantly with Volume
Fixed costs dominate many low-volume programs, which may place greater value on low NRE, MPW access, flexible minimum orders, engineering support, and limited upfront commitments. As volume rises, yield, qualified product cost, pricing stability, and production consistency become more important. At high volume, small variable-cost differences compound into substantial annual value, increasing the importance of wafer economics, contractual capacity, multi-site production, and long-term agreements.
19. Ecosystem and Enablement
A foundry provides access to a design and production ecosystem, not merely wafers. Digital products may depend on standard cells, SRAM compilers, processor and interface IP, SerDes, EDA flows, and package integration. Analog and specialty products may depend more on device-model accuracy, passive characterization, specialized devices, reliability data, documentation, and access to device engineers.
PDK quality and EDA support are economic infrastructure. Poor enablement can create excessive design margin, verification work, delayed debug, and additional silicon iterations even when it never appears in the foundry quotation.
| The relevant ecosystem is the one that accelerates and de-risks the parts of the product that create value. |
|---|
20. Engineering Support, Customer Importance, and Commercial Fit
Supplier differences become most visible when, not if, the program encounters a problem. The company should evaluate FAE quality, access to process and device experts, response time, yield-debug support, failure analysis, etc. The value of support depends partly on the customer's internal capabilities.
Customer importance also matters. The same annual wafer opportunity may be strategically minor at one foundry, meaningful at another, and a lighthouse account at a third. This can influence engineering attention, executive access, commercial flexibility, escalation, and capacity allocation.
The commercial model must also fit the company. Relevant terms include minimum wafer commitments, long-term agreements, prepayments, engineering fees, payment terms, etc. A technically ideal foundry can be a poor fit for a startup if its commercial structure assumes high, predictable volume and a strong balance sheet.
| Key question: How well does this foundry fit this specific customer, not just this specific product? |
|---|
Part V - Strategic Risk and Resilience
21. Supply Continuity Risk
A technically and economically attractive foundry can still expose the product to unacceptable supply risk. Risk analysis should evaluate probability, consequence, recovery time, and mitigation cost. Relevant risks include geographic concentration, political instability, export controls, natural disasters, utilities, water, logistics, aging equipment, supplier financial distress, fab closure, and process discontinuation.
Geography should not be used as a proxy for total risk. A politically stable fab may depend on aging tools with limited spare parts, while a geographically concentrated supplier may have strong infrastructure and recovery plans. Risk tolerance should be product-specific. A 15-year automotive product with difficult customer requalification may justify significant resilience spending; a short-lifecycle consumer product may not justify the cost and schedule of a fully qualified second source.
| Risk should be managed to the level justified by the product economics, not minimized at any cost. |
|---|
22. Resilience Strategies
True second sourcing is only one option. A second manufacturing flow can provide rapid recovery but may require duplicated masks, qualification, yield learning, and ongoing production at both suppliers. Migration-aware architecture can preserve future portability by avoiding unnecessary process-specific dependencies. Portfolio diversification can distribute different product families among suppliers without dual-sourcing every product. Strategic inventory may provide more economical protection for predictable, long-lived products. Contractual capacity can address the risk that physically available capacity will not be accessible to the customer.
The correct strategy depends on product volume, margin, lifetime, interruption cost, qualification burden, and migration difficulty. Second sourcing is expensive and often slows down TTM at the expense of pleasing internal supply chain teams, however, it has it’s place for certain products.
| Resilience does not necessarily require identical second-source silicon. |
|---|
23. Platform and Fab Longevity
The company must determine whether the selected platform will remain economically and operationally viable for the product's expected life. Mature nodes are constantly being end of life’d. Process maturity does not guarantee longevity. An older platform may face declining investment and reduced engineering support, while a newer platform may have less historical production experience but stronger long-term capital support.
Useful diligence questions include whether the process is still promoted for new designs, whether capacity is being maintained, whether new IP is being developed, whether the flow is supported at more than one fab, what discontinuation notice is available, and whether the foundry is encouraging migration. The required longevity should match the economic life of the product.
Part VI - Decision-Making Under Disordered Information
24. Foundry Selection Is an Information-Asymmetry Problem
Foundries generally know more than customers about the process technology. This asymmetry cannot be eliminated entirely. The correct objective is not perfect certainty, but a disciplined distinction among confirmed facts and known risks.
A weak process hides uncertainty by placing precise-looking numbers into a scorecard. A stronger process asks which uncertainties could change the recommendation and what is the most efficient way to reduce them.
25. Decision-Weighted Diligence
An unknown deserves greater attention when it could invalidate a hard constraint, materially affect differentiation, or create substantial downside if the assumption is wrong. High-decision-value unknowns may include: device reliability, accessible ramp capacity, production-ready IP, yield sufficient to meet the cost target, or long-term platform support. Moderate-value unknowns may change contractual structure or economics without changing feasibility. Low-value unknowns should not delay the decision.
For each unresolved question, the team should ask: what decision would change if the answer were different, how large could the impact be, how costly is the information to obtain, can the risk be addressed contractually, and is the decision reversible?
| The value of information depends on whether it can change the decision. |
|---|
26. Why Generic Weighted Scorecards Fail
Weighted scorecards are the go-to for many fabless companies today. They are a simple framework that grades foundries on several metrics, and make sense at first glance. Scorecards useful for organizing observations but dangerous frameworks for process selection. Hard constraints can be averaged away, weights are often arbitrary, related variables can be double-counted, uncertainty disappears, and marginal value is ignored. A process may receive a higher score for capability the product cannot use or monetize.
The decision hierarchy should be: define product value; separate table stakes from differentiation; translate differentiation into manufacturing variables; identify hard constraints; eliminate infeasible platforms; quantify technical, economic, and execution consequences; resolve decision-sensitive unknowns; and use multi-variable optimization for residual qualitative tradeoffs.
| Constraints first. Value second. Economics third. Optimization last. |
|---|
27. Common Failure Modes
Fabless companies simply do not have the resources to perform detailed process selection among the hundreds of candidate options. Common errors include choosing by wafer price or node label, treating every criterion equally, confusing table stakes with differentiation, optimizing specifications end-customers do not value, over-optimizing beyond marginal value, treating preferences as constraints, confusing installed capacity with accessible capacity, underestimating qualification, ignoring customer importance, accepting incompatible commercial terms, demanding uneconomic second sourcing, ignoring lifecycle, and collecting information incapable of changing the decision.
The final failure mode is taking too long. Foundry selection has long-term consequences, but delay also has consequences. The purpose of structure is to accelerate a high-quality decision, not to institutionalize indecision.
| Rigor should reduce uncertainty efficiently. It should not create organizational paralysis and lost time-to-market. |
|---|
Part VII - Case Study I: Mature-Node BCD Foundry Selection
28. Automotive Power-Management IC
Consider a simple example with an automotive power-management IC designed to deliver efficient and reliable power conversion with lower system complexity and thermal burden. Its differentiation is higher efficiency, lower external component count, a smaller thermal solution, and reliable automotive operation. Phase I has already selected a 130-180 nm BCD technology family. The remaining decision is which foundry platform should manufacture it. All assumptions below are illustrative and do not represent proprietary foundry data. This exercise points out a few decision criteria and points out the changing economics with product volume—complete process selection is much more detailed in optimizing among the many other variables.
| Variable | Foundry A: Mature 200 mm | Foundry B: 300 mm strategic | Foundry C: Smaller specialist |
|---|---|---|---|
| Estimated die size | 28 mm2 | 22 mm2 | 25 mm2 |
| Gross die / wafer | 870 | 2,750 | 980 |
| Expected yield | 90% | 88% initial; 93% mature | 80-84% |
| Illustrative wafer price | $1,800 | $4,800 | $1,650 |
| Masks, NRE, qualification | $1.8M | $6.2M | $2.8M |
| Cycle time | 13 weeks | 16 weeks | 10 weeks |
| Customer importance | Moderate | Low | High |
| Strategic profile | Mature, limited expansion | Strong roadmap and scale | Flexible, concentrated footprint |
Exhibit 5. Illustrative BCD foundry profiles
Simplified front-end good-die cost is approximately $2.30 for Foundry A, $1.98 initially and $1.88 at maturity for Foundry B, and $2.00-$2.10 for Foundry C. Foundry B has the highest wafer price but the lowest expected good-die cost. That does not make it the universal winner.
Scenario A - Lower Volume and High ASP
At roughly 1.5 million units per year and seven years of production, Foundry B's larger upfront cost offsets much of its unit-cost advantage. Foundry A's mature qualification history, lower capital requirement, and stronger first-pass confidence make it the preferred choice.
Scenario B - High Volume and Strong Cost Pressure
At 12 million units per year over ten years, Foundry B's NRE amortizes to only a few cents per unit. Its smaller die, mature yield, capacity scale, and roadmap can generate substantial lifetime savings. Foundry B becomes the preferred choice.
Scenario C - Long Lifecycle and Supply Sensitivity
At four million units per year over fifteen years, supplier attention and enforceable lifecycle support become more valuable. If Foundry C provides reserved capacity, a long-term support commitment, discontinuation notice, last-time-buy rights, and executive escalation, it can rationally win despite greater geographic concentration and modestly higher unit cost.
| The optimal foundry changes because volume, capital, lifecycle, and supply consequences change which enterprise-value drivers deserve optimization. |
|---|
Part VIII - Case Study II: Mature Leading-Edge Digital Platform
29. High-Performance Accelerator
Consider a high-performance accelerator whose customer value is more compute per watt and per unit of system footprint. Its differentiation is performance-per-watt, compute density, rapid iteration, and high-performance package integration. Phase I has selected a mature leading-edge class such as 7 nm. All assumptions below are illustrative.
| Variable | Foundry A: Ecosystem leader | Foundry B: Strong alternative | Foundry C: Strategic IDM option |
|---|---|---|---|
| Relative performance / watt | 100 | 94 | 97 |
| Estimated die size | 450 mm2 | 475 mm2 | 465 mm2 |
| Expected yield | 82% | 78% | 65% initial; 75% target |
| Illustrative wafer price | $16,000 | $12,500 | $11,500 |
| Masks, IP, process NRE | $48M | $29M | $21M |
| Production readiness | 18 months | 15 months | 19-23 months |
| First-year committed units | 500,000 | 900,000 | 650,000 |
| Customer importance | Low | High | Very high |
Exhibit 6. Illustrative mature leading-edge foundry profiles
Simplified front-end good-die cost is approximately $163 for Foundry A, $143 for Foundry B, and $154 initially or $133 at target yield for Foundry C. Packaging and memory integration may equal or exceed front-end die cost, but the comparison illustrates the tradeoff between technical leadership, execution, and capital.
Scenario A - Performance Dominates
Foundry A requires roughly $19 million more upfront investment than Foundry B and about $20 more front-end cost per unit. Across two million units, it must create and capture about $59 million of incremental value. If its performance-per-watt advantage supports sufficient ASP, share, or customer system value, Foundry A is rational.
Scenario B - Launch Schedule and Capacity Dominate
If first-year demand reaches 800,000 units and contribution margin is about $500 per unit, Foundry A's 500,000-unit commitment can expose the company to far more lost value than its technical advantage creates. Foundry B's earlier launch and 900,000-unit capacity make it the preferred choice.
Scenario C - Capital Efficiency Dominates
For a capital-constrained startup with 500,000 lifetime units, Foundry C preserves $27 million of upfront capital relative to Foundry A. If staged payments, milestone-based engineering, capacity commitments, and yield contingencies contain the execution risk, Foundry C can become rational despite greater uncertainty.
| At mature leading-edge nodes, the correct supplier is not necessarily the platform with the best transistor or lowest wafer price. It is the platform whose combination of performance, ecosystem, execution, capacity, and capital requirements creates the strongest probability-adjusted business outcome. |
|---|
Part IX - Practical Decision Heuristics
30. Eight Rules for Foundry Selection
1. Start with why the product wins.
2. Separate table stakes from true differentiation.
3. Hard constraints define feasibility; do not average them away.
4. Optimize differentiation only while marginal value exceeds marginal cost, time, and risk.
5. Accept sufficient performance on secondary considerations.
6. Compare qualified product economics, accessible capacity, and execution reality - not headline specifications.
7. Focus diligence on information capable of changing the decision.
8. Select the foundry that best enables the product and business strategy, not the one with the highest generic score.
Part X - Complete Phase I and Phase II Framework
31. From Customer Value to Manufacturing Strategy
| Customer Need, Market Entry, and Product Unique Selling Points |
|---|
| Product Architecture and Technical Requirements |
| Product Value Hierarchy and Process Technology Selection |
| Foundry Requirement Specification and Candidate Universe |
| Hard-Constraint Screening and Differentiation Evaluation |
| Manufacturing, Economics, Ecosystem, Strategic Risk, and Diligence |
| Foundry Selection, Design, Qualification, and Production |
| Customer and Enterprise Value Delivered |
Exhibit 7. End-to-end semiconductor manufacturing decision architecture
The final recommendation should include more than a supplier name. It should document why the selected platform fits the product strategy, which assumptions drive the result, which risks are being accepted, how those risks will be mitigated, which contractual protections are required, and what events would trigger reconsideration or migration.
Conclusion - Select the Foundry Around the Product
A fabless semiconductor company should not begin foundry selection with the most advanced process, the lowest wafer price, or the highest weighted scorecard. It should begin with the product and the business: why customers will buy, which requirements establish market access, and which attributes create genuine differentiation. Those sources of value should be translated into technical and manufacturing requirements. True hard constraints eliminate infeasible options; among viable platforms, optimization should stop when further capability costs more in capital, schedule, complexity, or risk than the value it can create and capture.
The decision must then account for manufacturing maturity, qualified product cost, accessible capacity, time-to-market, qualification, design enablement, engineering support, commercial structure, supply continuity, and platform longevity. None has a universal priority. Their importance follows from the product's differentiation, volume, lifecycle, financing, and market timing.
The same methodology can therefore recommend different suppliers under different assumptions: a mature low-NRE platform for lower volume, a capital-intensive 300 mm platform at scale, or a premium leading-edge foundry when technical differentiation can be monetized. The purpose is not to eliminate uncertainty, but to expose the assumptions that matter, resolve the unknowns capable of changing the recommendation, and make the remaining tradeoffs explicit.
| Foundry selection is an exercise in selective optimization. Hard constraints determine feasibility. Product differentiation determines where additional capability creates value. Economics, time, risk, and the company's ability to capture that value determine how far optimization should go. |
|---|
When process and foundry selection are treated as one connected strategic discipline, manufacturing becomes an intentional extension of product strategy and a potential source of durable competitive advantage.