| Invented scenario | Result | Boundary |
|---|---|---|
| 100 starts;80 first-pass;10 recovered | 80% first-pass;90% final accepted | Rework cost separate |
| 100 × $10 plus90 × $5;81accepted | $17.90 per accepted circuit | Staged costs;no overhead or margin |
| 10panels ×20positions;1panel lost;9individual rejects | 171accepted;85.5% overall | Distinct loss counts |
| $1000 fixed batch cost;80/90/95accepted | $12.50/$11.11/$10.53 | Sensitivity cases,not predictions |
Make the yield denominator part of the quotation
A ceramic PCB cost estimate becomes difficult to compare when two suppliers use the same word, yield, for different quantities. One may mean accepted individual circuits divided by started circuits. Another may mean panels that reach final inspection, and a third may include successfully reworked parts. Before entering a percentage into a cost model, name the unit, process boundary and acceptance decision. The ceramic PCB cost guide covers the broader commercial structure. This article addresses the manufacturing losses and recovery assumptions behind that structure, so purchasing and engineering can distinguish a useful estimate from a number that merely appears precise on a quotation spreadsheet.
Keep geometry and process acceptance separate
The number of circuit positions that fit on a manufacturing panel is a geometric input. The fraction of those positions that becomes acceptable product is a different input. A panel with more positions can have a lower cost per started circuit without having a better manufacturing yield. Conversely, a less dense arrangement may become attractive if it changes handling or inspection outcomes, but that improvement needs evidence. Use the panelization guidance to define the intended arrangement, then keep the accepted-circuit count in a separate field. This prevents a nesting improvement from being counted once as better utilization and again as an unsupported increase in process yield.
Distinguish first-pass yield from final accepted output
ASQ defines first-pass yield around conforming output without corrective repetition or repair. Final accepted output can include material that required an approved recovery route, so it answers a different question. For an invented batch of one hundred started circuits, suppose eighty pass initially, ten more become acceptable after authorized rework and ten are scrapped. First-pass yield is eighty percent, while final accepted output is ninety percent of starts. Neither figure alone describes the cost of the batch. Record rework effort, additional inspection and any delay separately. A quotation that reports ninety percent acceptance should not silently assume that all ninety circuits consumed only the original process resources.
Define what counts as an accepted ceramic circuit
Tie acceptance to the controlled drawing and agreed inspection plan. Relevant requirements might include circuit dimensions, metallization condition, electrical continuity, insulation checks or ceramic edge condition, depending on the product. The important point is not to impose every possible test, but to keep the acceptance definition consistent across the data used in the estimate. A historical yield measured against a different edge criterion or test scope may not represent the new order. Document any changes in requirements alongside the yield assumption. Otherwise a seemingly improved process can actually be a changed counting rule, and a cost comparison can reward a proposal that is delivering a different level of verification.
Model losses at the stage where they occur
Suppose an illustrative process starts one hundred circuits at ten dollars each. Ninety pass the first stage and each of those ninety receives a second operation costing five dollars. Eighty-one circuits are finally accepted. The modeled batch cost is one thousand dollars plus four hundred fifty dollars, or fourteen hundred fifty dollars. Dividing by eighty-one gives about seventeen dollars and ninety cents per accepted circuit. This staged calculation differs from charging both operations to every start and then dividing by final yield, which would give about eighteen dollars and fifty-two cents. The distinction matters whenever rejected material stops consuming downstream processing resources.
Use linked stage counts instead of unrelated percentages
For the same example, the first stage passes ninety of one hundred starts, and the second passes eighty-one of those ninety survivors. Multiplying those conditional stage fractions gives an overall acceptance fraction of eighty-one percent. This arithmetic follows the same cohort through the stated route. It does not justify multiplying percentages copied from unrelated products, different months or incompatible inspection boundaries. If rework, branching or returned material is present, draw those flows and reconcile the counts explicitly. A model can remain simple, but each numerator should connect to the appropriate downstream denominator so that units do not disappear, reappear or get counted twice in the cost calculation.
Separate panel-level and circuit-level losses
Some losses remove an entire panel, while others reject particular positions after a panel completes processing. Keep these categories separate when the distinction affects the estimate. For example, ten panels with twenty circuit positions each provide two hundred starts. If one panel is lost completely and nine individual circuits are rejected from the remaining panels, accepted output is one hundred seventy-one circuits. Overall acceptance is eighty-five and a half percent, not the sum of a panel-loss percentage and a circuit-reject percentage computed against a different denominator. Preserve the panel identity in the production record so the model can distinguish distributed rejects from a shared event affecting many positions together.
Treat small prototype samples as limited evidence
A prototype lot in which every inspected circuit passes is useful evidence about those circuits, but it does not establish that future production has zero loss. NIST discusses confidence intervals for a binomial proportion and the need to use an appropriate statistical method. Before applying such a model, consider whether observations reasonably represent independent outcomes from a common process. Ceramic circuits from one panel may share conditions, making an unqualified independent-item assumption questionable. Report the sample size, panel count, process revision and observed results instead of replacing a zero-reject observation with a guaranteed production yield. When a formal bound is needed, document the statistical method and its assumptions with the estimate.
Select historical data by relevance to the new design
Ask which past jobs support the quoted assumption. Material family alone may be too broad a comparison if the new circuit has different thickness, feature dimensions, metallization route or inspection requirements. Prefer a documented explanation of similarity over a large pooled number that hides different processes. A supplier may reasonably avoid sharing sensitive production records, but can still identify whether the estimate comes from a directly comparable process, a related construction or an engineering allowance awaiting validation. Give those evidence levels different confidence in the purchasing model. This helps the team decide whether to accept the estimate, request a pilot run or change a feature that drives uncertainty.
Show a yield sensitivity range without inventing probability
A planning estimate can display several assumed yields even when there is not enough information to assign a probability distribution. For a simplified fixed batch cost of one thousand dollars and one hundred starts, eighty accepted circuits imply twelve dollars and fifty cents per accepted circuit. Ninety accepted circuits imply about eleven dollars and eleven cents, while ninety-five imply about ten dollars and fifty-three cents. These are scenarios, not confidence limits or predictions. Label them accordingly and state which costs are held constant. If downstream processing cost changes with survivor count, use the staged model instead of presenting this fixed-cost illustration as a complete manufacturing quotation.
Avoid adding a yield penalty twice
Determine whether the supplier's price is per started circuit, per processed panel or per accepted delivered circuit. A firm price per accepted unit may already include the supplier's allowance for normal manufacturing loss. Dividing that price by an assumed yield again can overstate the buyer's expected purchase cost. Conversely, a panel-processing quote may leave the buyer exposed to variation in accepted output. Record the charging basis beside the quantity and acceptance terms. Use the cost estimator for transparent scenario work, but ensure that its inputs represent the commercial arrangement being evaluated rather than applying the same yield multiplier to every quotation regardless of how it is priced.
Put rework into an explicit approved route
If recovery is permitted, specify who authorizes it and what evidence establishes conformity afterward. Do not assume that a ceramic crack, conductor defect and surface condition all have an equivalent recovery option. The quotation model needs the expected number entering rework, the extra cost per attempt and the expected acceptable output from that route. It also needs a limit on repeated attempts if those attempts affect delivery or qualification. Keep recovered circuits identifiable in the underlying records even when they meet the final acceptance criteria. That distinction supports process improvement and later investigations without treating an approved recovered part as automatically unacceptable or economically identical to a first-pass part.
Connect the output model to delivery planning
Dividing an order quantity by an assumed yield gives an expected number of starts under a simplified model; it does not guarantee that the order quantity will be delivered on time. Manufacturing quantities may be constrained to whole panels, and actual rejects can vary across lots. Ask the supplier how it plans starts, handles shortfalls and communicates a partial delivery. Keep those commercial commitments distinct from the engineering yield estimate. For critical schedules, identify the time required for an additional lot or an approved alternate route. A slightly cheaper expected unit cost can be less useful if the quotation leaves the response to a shortfall entirely undefined.
Reconcile actual results after the pilot lot
After a pilot run, compare started counts, stage survivors, recovered units, final accepted output and actual processing effort with the estimate. Investigate differences using the same acceptance definitions and process boundaries used during quoting. If the design or process changed during the run, separate the affected material rather than presenting one blended figure as a stable production result. Record whether a discrepancy came from the assumed yield, the cost of a process step or an unplanned activity such as added inspection. This turns the pilot into usable estimating evidence and makes the next quotation revision explainable without requiring the supplier to reveal its entire internal cost structure.
Choose improvements by their effect on accepted output
An engineering change should be evaluated against the failure or cost mechanism it is intended to address. Enlarging a feature, changing a separation approach or modifying inspection access may alter both process cost and accepted output. Compare the complete revised route with the baseline instead of assuming that a simpler drawing must be cheaper. Some changes reduce rejects but add an operation, and others save processing time while increasing uncertainty elsewhere. Request a bounded trial when the claimed benefit is not supported by comparable production evidence. Preserve the functional requirements of the circuit while making the cost tradeoff visible, so acceptance criteria are not quietly weakened to improve a yield figure.
Release the estimate with its assumptions attached
A reviewable yield-based estimate identifies the delivered unit, process route, started quantity, stage acceptance fractions, recovery route, cost basis and evidence supporting each uncertain input. It includes a sensitivity range where appropriate and clearly separates estimated manufacturing behavior from the supplier's delivery commitment. Keep the estimate revision linked to the drawing and inspection plan, because both can change the meaning of accepted output. The objective is not to obtain a supposedly universal ceramic PCB yield percentage. It is to understand how this construction, through this route and under these acceptance requirements, converts purchased manufacturing effort into usable circuits with an uncertainty that the project can manage.
Engineering example
All costs and yields are invented arithmetic examples, not supplier prices or industry yield benchmarks.
Before you release the design
- State the unit and acceptance boundary.
- Separate nesting from process acceptance.
- Track stage survivors and recovery explicitly.
- Avoid applying yield twice to accepted-unit pricing.
- Record sample size and model assumptions.
Sources and further technical reading
Manufacturer references support the material and process context. Worked examples and checklists are engineering guidance; they are not test results or supplier guarantees.