Product Carbon Footprint for Semiconductor Assembly and Materials Plants: Method and Practice

Why an OSAT plant's PCF is harder than general manufacturing

Front-end wafer fabs concentrate energy in cleanrooms and process tools, with very high value per die. Assembly, test and electronic materials are the opposite: high batch counts, low value per batch, dispersed processes, and a wide range of chemicals and gases. The emissions structure of these product types differs enough that one template applied across all of them usually gets rejected at review.

Step 1: Set the calculation boundary

The boundary determines which processes you must collect data for, and therefore which parts can never be filled in later. Produce a one-page boundary statement at kick-off that lists which processes are inside the boundary, which are the upstream supplier's responsibility, and which are excluded with a stated reason.

Step 2: Where activity data comes from, and how to make it credible

Activity data is where most problems appear and where sampling is most likely. A reviewer typically traces one data point upward through three layers: reported value, summary table, original evidence. If any layer breaks, that number is downgraded to an estimate.

Ranked from highest to lowest data quality, the sources seen in practice are:

1. Continuous metering (electricity meters, flow meters, weighing systems) that is already inside a calibration programme.
2. Data recorded automatically by a device or system and exportable as a raw file — instrument software measurement records, MES process records, LIMS test reports.
3. Purchase and requisition records with clear evidence (invoices, issue slips, warehouse movements).
4. Engineering estimates or output-based allocations (must state method, assumptions and uncertainty).

The first two are more reliable because they carry a timestamp and a device identifier, which answers "who measured this, on which instrument, and when". If the test data also carries calibration status and quality-control results, it answers a further question: "is this number trustworthy?" That is the real value of on-site measuring instruments in a PCF project — not a more precise number, but a number that can be attributed.

• Energy: take it from the facility system or billing meter, not from "rated power × run hours". The latter is for cases where no metering exists, and must be disclosed.
• Materials: confirm the weight basis matches the product basis (net weight versus including packaging), and avoid unit or form mismatch.
• Test-derived data (material composition, moisture, carbon content): record the sampling method, the reporting basis (as-received, dry, dry-ash-free) and the test method standard. Any conversion between bases must state the formula.
• Cross-system extracts need an interface log or a reconciliation record. Manually transcribed tables should retain the original export file and a conversion note.

Step 3: Emission factors — selection and version control

Emission factors almost never come from the company's own measurement. They come from databases, suppliers, trade associations or government publications. Once chosen, a factor should be recorded together with the standard version, publisher, year and scope of applicability.

• Avoid mixing several sources for the same factor type within one report. Where mixing is unavoidable, annotate each factor and state the reason.
• For supplier-provided factors, establish whether they are measured or industry averages, and whether their boundary matches yours (for example, whether upstream transport is already included).
• Select the grid factor version and region required by the customer. When comparing across years, do not quietly swap factor versions — it hides real reduction or inflates it.
• For fuels, distinguish net from gross calorific value, and keep that choice consistent with how your own calorific data is used.

Step 4: Calculation, allocation and uncertainty

The calculation is not complicated: emissions per process step equal activity data multiplied by the emission factor, summed to the product unit. What needs real design work is allocation and how uncertainty is described.

• Allocation: state the parameter used (labour hours, output, area, equipment time) and show that it is related to the emissions driver of that product.
• Shared utilities: allocate compressed air, chilled water, ultrapure water and lighting by metered use where possible; where not metered, disclose the assumption.
• Uncertainty: statistical rigour is not required, but data sources should be graded (metered, system-recorded, evidenced, estimated) so a reviewer can see which numbers hold and which are assumptions.
• Sensitivity: run a sensitivity note on the two or three largest inputs (for example, the effect of a 10 % change in the grid factor). In customer review this is often more persuasive than the headline total.

Step 5: Reporting output and internal retention

Design the customer report and the internal evidence file separately. The customer report carries conclusions and method; the internal file carries evidence. It is the internal file that a reviewer actually requests in volume.

• Customer report should contain: objective and boundary, standard and version, data source description, factor list, allocation method, results and units, uncertainty note, change log.
• Internal file should contain: raw metering and export files, test reports and instrument calibration status, summary sheets and calculation formulas, version and approval records, explanations of year-on-year differences.
• Version control must answer "why did this change from the previous version", not merely retain the final version.
• Retention periods follow the customer contract and local requirements; if uncertain, retain at least until the next external review closes.

One practical suggestion: turn each year's calculation into a reusable template and work list rather than rebuilding the spreadsheet. From the second year onward, most of the effort should go into data updates and explaining differences — not into re-arguing methodology.

Implementation path: from zero to a reviewable first report

If the plant has not started, follow this order. It avoids buying software before the data habits exist.

1. Clarify the customer requirement: standard, boundary level, report format, submission date.
2. Pick one pilot part number with stable output and comparatively complete data; run the full flow before scaling.
3. Build the data list: map every required input to a specific device, system or document, and mark the gaps.
4. Close the gaps: fix metering and recording habits first. Systemising unstable data only produces wrong numbers faster.
5. Set up internal review: separate the data owner, the calculator and the reviewer, and keep the review record.
6. Subject it to an external read: even an informal customer pre-review surfaces most problems early.