ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

Semiconductor Technical Assessment — Illustrative Sample

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

What this document is. A format sample showing how a CFS Semiconductor Technical Assessment is structured, how findings are labelled and cited, and how uncertainty is recorded.

What this document is not. It is not a record of work performed for any customer. “Northlight Systems, Inc.” and “NL-V1” are fictional names created for this sample, and the company, product, requirements and figures are invented for illustration. Any resemblance to a real organisation, product or person is coincidental and unintended. CFS makes no claim, here or elsewhere, to have performed this or any paid customer engagement, and names no real company, product, foundry, process, supplier or person.

No commitment is given. This sample provides no tape-out authorisation, no foundry commitment, no capacity or pricing commitment, no qualification, and no performance, yield, reliability or schedule guarantee. It is engineering analysis for illustration only, not certification, and nothing in it may be relied upon.

HOW TO READ THIS DOCUMENT

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

Every statement in a CFS assessment is one of four kinds, and is marked as such. The distinction is the point of the document: it lets a reader see exactly how much weight any given sentence carries.

Each finding also carries a confidence label:

1. PROBLEM DEFINITION

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

Northlight Systems (fictional) is developing the NL-V1, an always-on edge-AI inference SoC for industrial machine vision. The part is intended to run continuously in a sealed enclosure with no forced-air cooling.

The team must decide between two process options, described here generically and without naming any foundry or process node owner:

• Process Option A — a fully-depleted SOI-class process at approximately the 22 nm generation
• Process Option B — a FinFET-class process at approximately the 16 nm generation

A second question follows: whether a standard laminate package is viable, or whether a higher-conductivity package is required.

The decision this unblocks: which class of process to pursue in supplier discussions, and whether the mechanical enclosure design can be frozen.

Stated constraint: the fictional team reports an eight-week internal milestone requiring a process direction.

2. SUMMARY OF FINDINGS

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

Finding 1 — Process Option A is the better fit for this workload. [Assessed] The governing constraint is sustained average power in a passively cooled sealed enclosure, not peak throughput. [ASSUMPTION] At the stated duty cycle, static leakage over a 24-hour average dominates the energy budget, which inverts the usual “finer node is better” logic. [Assessed]

Finding 2 — Body biasing is the decisive technical differentiator. [Verified] [PUBLIC FACT] Fully-depleted SOI-class processes provide a back-gate terminal allowing dynamic threshold-voltage adjustment: forward bias to raise performance during inference bursts, reverse bias to suppress leakage while idle. FinFET-class bulk processes do not offer an equivalent lever. For a low-duty-cycle always-on workload this directly addresses the dominant energy term. [Assessed]

Finding 3 — The package question cannot be closed on the information available. [Unverified] [ASSUMPTION] Package selection is governed by total thermal resistance from junction to ambient. In a sealed enclosure, the enclosure-to-ambient term typically dominates the package’s own junction-to-case term. [PUBLIC FACT] Because the enclosure’s thermal boundary conditions were not supplied, no package recommendation can be made. A standard laminate package may well be sufficient — but this is a hypothesis, not a finding.

Overall recommendation. [RECOMMENDATION] Proceed toward Process Option A and open supplier discussions. Do not freeze the enclosure design until Experiment E-1 (§8) completes. The thermal work, not the process decision, is the schedule risk against the stated milestone.

3. TECHNICAL ANALYSIS

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

3.1 The workload sets the constraint

[ASSUMPTION A-1] Sustained package power approximately 2 W. [ASSUMPTION A-2] Inference duty cycle below 15%; idle otherwise.

[CALCULATION] With a duty cycle d, average energy over a period is approximately d × (dynamic + static) + (1 − d) × static. At d = 0.15, roughly 85% of wall-clock time contributes static power only. Static leakage therefore governs the 24-hour average, and any process advantage in dynamic energy is diluted by a factor approaching d. [Assessed]

This is the hinge of the whole assessment. [RECOMMENDATION] If A-2 is materially wrong — if the true duty cycle is, say, 50% — the conclusion in §2 must be revisited before any supplier commitment. Confirming A-2 costs days and is the single highest-value action available.

3.2 Why body biasing matters here

[PUBLIC FACT] In a fully-depleted SOI-class process, a back-gate bias shifts device threshold voltage, trading drive current against leakage at run time. [Assessed] For a workload spending most of its time idle, reverse body bias during idle reduces precisely the term §3.1 identifies as dominant, and forward bias during bursts recovers performance when it is actually needed.

[Assessed] A FinFET-class process at a finer generation offers higher transistor density and better dynamic energy per operation. Neither is the limiting factor in this design as specified — [ASSUMPTION A-1, A-2] — and the finer process’s higher static leakage is continuously present.

3.3 Cost and access

[Assessed] Mask-set and design-enablement costs rise substantially with each finer generation, and the difference between a ~22 nm-generation and a ~16 nm-generation process is material relative to a pre-Series-A budget.

[ASSUMPTION A-3] No specific cost figures are given in this sample, and none should be inferred. Actual costs vary by supplier, volume, IP content and commercial terms, and CFS has not established them here. Publishing a number CFS cannot support would be exactly the kind of unsupported claim this document exists to avoid. Northlight would obtain figures directly from suppliers under NDA.

3.4 Thermal

[PUBLIC FACT] Total junction-to-ambient thermal resistance is the sum of junction-to-case, case-to-enclosure and enclosure-to-ambient terms in series. [Assessed] In a sealed, passively cooled enclosure, the enclosure-to-ambient term is commonly the largest, which means enclosure design usually constrains the outcome more than package selection does.

[RECOMMENDATION] This drives sequencing rather than a component choice: the thermal simulation must precede the enclosure freeze, because if the enclosure cannot dissipate the power, no package choice rescues it.

4. DECISION MATRIX

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

Weights reflect the fictional customer’s stated priorities. [ASSUMPTION] Scores are 1–5, higher is better, and are [Assessed] engineering judgement for illustration only.

[CALCULATION] Weighted totals: Option A = 4.15, Option B = 2.95.

Option A: (5×0.30) + (3×0.15) + (5×0.20) + (4×0.15) + (3×0.10) + (3×0.10) = 1.50 + 0.45 + 1.00 + 0.60 + 0.30 + 0.30 = 4.15

Option B: (3×0.30) + (5×0.15) + (2×0.20) + (2×0.15) + (3×0.10) + (3×0.10) = 0.90 + 0.75 + 0.40 + 0.30 + 0.30 + 0.30 = 2.95

Weights sum to 1.00. Every term is shown so the result can be re-run and checked independently.

[RECOMMENDATION] A decision matrix is a way to expose reasoning, not a way to decide. Two criteria are scored neutral only because information was missing; the outcome rests mainly on the first and third rows, both of which depend on [ASSUMPTION A-2]. Confirm A-2 before treating this ranking as settled.

5. ASSUMPTIONS AND MISSING INFORMATION

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

Missing information materially limiting this assessment: enclosure thermal boundary conditions (A-4); measured or simulated power profile (A-1, A-2); supplier-specific enablement terms (A-3, A-6).

[RECOMMENDATION] Nothing in §5 is filled in with a plausible-looking value. An undisclosed assumption is how an assessment becomes wrong in a way nobody catches, so gaps are shown as gaps.

6. RISK REGISTER

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

[RECOMMENDATION] R-4 is the risk most likely to cause real damage, and it is organisational rather than technical: a fixed milestone plus an unresolved thermal question usually resolves in favour of the milestone.

7. RECOMMENDED ENGINEERING ACTIONS

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

8. RECOMMENDED NEXT EXPERIMENTS

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

E-1 — Enclosure thermal simulation - Hypothesis: the sealed enclosure can dissipate the stated sustained power within junction-temperature limits using a standard laminate package. [Unverified] - Method: three-corner thermal simulation at stated power plus 25% margin, using enclosure geometry and specified worst-case ambient. - Success criterion: junction temperature within specification at the worst-case corner with ≥ 10 °C margin. - Effort: approximately one week. - If it fails: enclosure redesign or a higher-conductivity package — both far cheaper to discover before the enclosure is frozen than after.

E-2 — Duty-cycle characterisation - Hypothesis: the application profile yields a duty cycle below 15%. [Unverified] - Method: instrument the reference application over a representative operating period. - Success criterion: measured duty cycle and its variance across operating modes. - Effort: approximately two days. - If it fails: §2 Finding 1 is re-opened and the decision matrix re-scored.

9. UNCERTAINTY STATEMENT

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

What this assessment does not establish:

• Whether the enclosure can dissipate the required power. Not determinable without A-4. This is the largest open item and it is deliberately left open.
• Actual cost difference between process classes. Not established; no figure is offered.
• IP and supply availability. Not established.
• Whether the recommendation survives a corrected duty cycle. It may not (R-1).
• Any performance, yield, reliability or schedule outcome. Outside scope entirely.

On a real engagement, an outcome of “this cannot be determined without testing” is a legitimate deliverable, stated plainly with the test that would resolve it, and it is billable. Customers are told this before committing, so that the commercial arrangement never pressures CFS toward a confident answer it cannot support.

10. HUMAN REVIEW AND PROVENANCE

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

How a real deliverable would be produced. AI assistance is used for retrieval, structuring and drafting. It is never the sole basis for a conclusion that reaches a customer. A named human engineer reviews and signs every deliverable before issue, with no low-risk exception.

Recorded on every real deliverable: source references with access dates; model and prompt version; assumptions relied upon; confidence for each finding; the named reviewer and review date; the number of AI-drafted conclusions changed in review; creation and revision timestamps; and the assessment’s limitations.

For this sample specifically: no customer engagement occurred, no customer material was reviewed, no supplier was contacted, and no named reviewer is recorded because there is no real deliverable to sign. The reviewer line in the header is a placeholder showing where a signature would appear.

Illustrating what review does. On a real engagement, §2 Finding 3 is exactly the kind of conclusion that changes in human review: an initial draft can easily present a package choice as resolved, and review is where it is corrected to [Unverified] pending E-1. CFS records such changes internally, because the record of what review caught is the evidence of whether the workflow is trustworthy.

11. NOTICES

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK

• This is an illustrative sample. “Northlight Systems”, “NL-V1” and all requirements and figures are fictional names and values created for this sample. Any resemblance to a real organisation, product or person is coincidental and unintended. CFS does not claim to have performed this or any paid customer engagement.
• No tape-out authorisation is given. Nothing here authorises, approves or supports a tape-out decision.
• No foundry or supplier commitment is given. No capacity, pricing, schedule or availability is committed by CFS or by any third party. No foundry, supplier or process owner is named, endorsed or represented.
• No performance guarantee is given — no yield, reliability, thermal, timing, power or schedule outcome is warranted.
• Analysis, not certification. CFS provides engineering analysis. It does not certify manufacturability, reliability, safety or regulatory compliance.
• CFS does not own or operate a wafer fab. Fabrication, packaging and test are performed by qualified partners.
• No commercial interest disclosed because no real vendor is named. On a real deliverable, any CFS commercial interest in a recommended vendor is disclosed in writing.
• Not legal, patent or investment advice.
• Not to be relied upon. This document exists to show format and reasoning style. No engineering decision should be based on it.

ILLUSTRATIVE SAMPLE — NOT CUSTOMER WORK