DFT · Chapter 1 · Manufacturing Test Mindset

DPPM & Quality Goals

DPPM, defective parts per million, is the industry's quality metric: the escape rate expressed as how many defective parts reach the customer per million shipped. It is the customer-facing number that a car maker or a phone maker writes into a supply contract. This lesson ties the chapter's chain together: fault coverage drives the escape rate, the escape rate becomes DPPM, and DPPM is the field-quality number. A quality goal is a DPPM target set by the market, so consumer parts tolerate a higher DPPM while automotive, medical, and aerospace demand single-digit, zero-defect quality. The target works backward to set a required coverage target given the defect density. Two nuances matter: DPPM depends on both coverage and defect density, and 100 percent coverage does not equal zero DPPM because of unmodeled defects, which is why safety parts add burn-in and on-chip test.

Foundation13 min readDFTDPPMQuality GoalsCoverageFunctional Safety

Chapter 1 · Section 1.5 · Manufacturing Test Mindset — chapter closer

Project thread — the 4-bit counter, now shipped in millions and measured by its DPPM.

1. Why Should I Learn This?

DPPM is the number the customer cares about, and it closes this chapter's whole argument:

DPPM = Defective Parts Per Million — escapes normalized per million shipped.
The chain: coverage → escape rate → DPPM → customer quality.
A quality goal is a DPPM target — consumer (higher) vs automotive/medical (single-digit, zero-defect).
The target works backward: DPPM goal → required coverage target (Ch6).
DPPM depends on coverage and defect density; 100% coverage ≠ 0 DPPM (unmodeled defects).

2. Real Silicon Story — the automotive DPPM clause

A team shipping a part into consumer products ran comfortably at their DPPM target for years. Then the same silicon won an automotive design-in — and the customer's contract specified a single-digit DPPM requirement backed by ISO 26262.

The existing coverage — fine for consumer — produced an escape rate that translated to a DPPM far above the automotive clause. The team had to work the chain backward: the DPPM target dictated a much higher required coverage, so they raised ATPG coverage (Chapter 6), tightened the fault models (Chapter 2), and — because even maxed-out coverage can't catch unmodeled defects — added system-level test and burn-in (1.3) plus on-chip BIST hooks (Chapters 8–9) as insurance for the zero-defect market. Lesson: DPPM is a contractual, market-set number; it drives coverage backward, and hitting an aggressive DPPM takes coverage plus extra test insurance, not coverage alone.

3. Concept — DPPM and the quality chain

DPPM — the customer-facing quality metric:

DPPM = (defective parts shipped) / (total parts shipped) × 1,000,000. It's the escape rate per million.
A part with 10 escapes per million shipped is at 10 DPPM. Lower is better; 0 DPPM is the aspiration.

The chain this whole chapter built:

Fault coverage (how many faults the test detects) → sets the escape rate (undetected faults on defective dies escape) → normalized to DPPM → experienced as customer field quality.
Raise coverage → fewer escapes → lower DPPM → better quality. That's the causal spine of DFT.

Quality goals are market-set DPPM targets:

Market	DPPM expectation	Mindset
Consumer	Higher DPPM tolerated	rare dead part = annoyance
Industrial	Lower	uptime matters
Automotive / medical / aerospace	Single-digit, → 0	zero-defect, safety (ISO 26262)

The target works backward — DPPM goal sets the coverage target:

Given the process defect density, a DPPM target implies a required fault coverage (Chapter 6 signoff).
Tighter DPPM → higher required coverage → more DFT effort (and test time, 1.4).

Two nuances you must carry (accuracy):

DPPM depends on coverage and defect density. A mature, low-defect process plus high coverage yields low DPPM; a noisy process needs even higher coverage to hit the same DPPM.
100% coverage ≠ 0 DPPM. Coverage is measured against modeled faults; unmodeled defects (real physical failures no model represents) can still escape. So zero-defect markets add insurance: SLT, burn-in (1.3), and on-chip/in-field BIST (Chapters 8–9).

Figure 1 — the quality chain and its backward-set goal (representative). Forward causality (left to right): FAULT COVERAGE sets the ESCAPE RATE, which normalized per million is DPPM, which the CUSTOMER experiences as field quality. The QUALITY GOAL (a market-set DPPM target) closes the loop by working BACKWARD (dashed): the DPPM target dictates the REQUIRED COVERAGE target (Ch6 signoff), given the process defect density. Higher coverage -> fewer escapes -> lower DPPM. Note: 100% coverage still != 0 DPPM because of UNMODELED defects, so safety markets add SLT/burn-in/BIST insurance.

4. Mental Model — a factory's defect-per-million promise

Picture signing a contract that promises a customer 'no more than N bad units per million shipped.'

DPPM is that promised number. It's what the customer measures you against — and in automotive, it's legally and safety-bound.
To keep the promise, you must catch bad units before shipping — that's coverage. The fraction you miss (escapes) is your DPPM.
The promise runs backward into your factory: a stricter promise (lower DPPM) forces you to inspect more thoroughly (higher coverage).
But some defects don't match any inspection you've designed (unmodeled) — so a strict promise also makes you add extra checks (SLT, burn-in, on-chip self-test) as insurance beyond your standard inspection.
And the incoming quality of your raw units (defect density) matters: cleaner raw units make the promise easier to keep at the same inspection level.

The whole chapter in one image: a promise (DPPM) you keep by catching defects (coverage) before they escape — bought at an acceptable inspection cost (test time).

5. Working Example — the counter's DPPM

Put representative numbers on the counter's quality, working the chain both ways:

Azvya Education Pvt. Ltd.VLSI Mentor

Snippet

# DPPM — forward chain — REPRESENTATIVE:
  Defect density (defective fraction)  = 1%       (10,000 defective per 1,000,000 dies)  <- process/fab
  Fault coverage                        = 99%      (test detects 99% of the defective dies)
  Escapes                               = 1% of defective = 1% x 10,000 = 100 dies escape per million
  DPPM                                  = 100      (100 defective parts per million shipped)
# Raise coverage 99% -> 99.9%  ->  escapes 100 -> 10  ->  DPPM 100 -> 10.

Azvya Education Pvt. Ltd.VLSI Mentor

Snippet

# DPPM — working BACKWARD from a quality goal — REPRESENTATIVE:
  Automotive goal        = 10 DPPM
  Defect density         = 1% (10,000 defective per million)
  Required escapes       <= 10 per million  ->  required coverage = 1 - 10/10,000 = 99.9%   <- Ch6 target
  If process improves (defect density 1% -> 0.5%): same 10 DPPM needs coverage = 1 - 10/5,000 = 99.8% (a bit easier)
# DPPM depends on BOTH coverage AND defect density.
# AND: even 100% MODELED coverage != 0 DPPM -> UNMODELED defects escape -> add SLT/burn-in/BIST insurance.

The artifact in four questions — the DPPM / quality report

What is it? A DPPM / quality report — the shipped defective-parts-per-million, tied back to coverage, defect density, and field-return data, versus the contractual DPPM goal.
Who generates it? Quality/reliability engineering, combining field returns, test/ATPG coverage, and fab defect-density data.
Who consumes it? Customers (contractual quality), program management (does the part meet the market's goal?), and DFT/design (raise coverage / add insurance to hit the target).
Why should an RTL engineer care? The required coverage that this DPPM goal implies is capped by your RTL's testability — untestable logic → uncovered faults → escapes → DPPM you can't meet. Hitting an aggressive DPPM often starts with an RTL/DFT change (Chapters 4, 6), not a tester setting.

6. Industry Flow — the quality stack

A DPPM goal sits atop a stack: market goal → DPPM target → coverage target → the DFT techniques that deliver it:

Figure 2 — the quality stack (representative), top-down. The MARKET (consumer vs automotive/medical) sets a DPPM GOAL. That goal, given the process DEFECT DENSITY, dictates a required COVERAGE TARGET (Ch6 signoff). Coverage is delivered by the DFT TECHNIQUES at the base -- scan, ATPG, compression, and for zero-defect markets extra insurance (SLT, burn-in, MBIST/LBIST/in-field BIST) for UNMODELED defects. Each layer sets the requirement for the layer below; each lower layer is how the layer above is actually met.

7. Debugging Session — field DPPM above the automotive target

A part meeting its consumer DPPM is designed into an automotive program whose contract demands single-digit DPPM; field DPPM comes in far above target and someone argues the tester is fine so ship it -- but the required coverage set backward from the DPPM goal is much higher, and because 100% modeled coverage still cannot catch unmodeled defects, the fix is raise coverage AND add SLT/burn-in/BIST insurance

DPPM GOAL SETS COVERAGE BACKWARD; 100% COVERAGE ≠ 0 DPPM → ADD INSURANCE

Symptom

A part that meets its consumer DPPM is designed into an automotive program with a single-digit DPPM contract. Field data shows the part's DPPM is far above the automotive target. One voice says 'the tester passes the parts it's told to — the test is fine, ship it.'

Root Cause

The DPPM goal sets a much higher required coverage than the consumer design ever needed — and even maxing coverage won't reach zero because of unmodeled defects. Work the chain backward: a single-digit DPPM target, given the process defect density, implies a required fault coverage far higher than the consumer part carried (e.g. 10 DPPM at 1% defect density needs ~99.9% coverage — much tighter than a consumer 100+ DPPM part). So the current coverage produces an escape rate whose DPPM exceeds the automotive clause — this is a coverage-vs-goal gap, not a tester malfunction ('the test is fine' is true and irrelevant: the test simply isn't detecting enough faults for this market). But there's a second cause that pure coverage can't fix: 100% modeled coverage still isn't 0 DPPM, because unmodeled defects — real physical failures that no fault model represents — can escape even a perfect modeled-coverage test. Zero-defect markets are precisely where those unmodeled escapes matter, so coverage alone, even maximized, won't guarantee the automotive DPPM.

Fix

Raise coverage to the target the DPPM goal demands, and add insurance for the defects coverage can't model. First, set the coverage target backward from the DPPM goal and defect density (Chapter 6 signoff), then close coverage to it — improve RTL testability (Chapter 4/6), tighten fault models (Chapter 2, e.g. add transition/at-speed for timing defects), and let ATPG target the newly-reachable faults (Chapter 5). Second, because unmodeled defects escape even at 100% modeled coverage, add the insurance layers the zero-defect market expects: system-level test and burn-in (1.3) to catch system-only and infant-mortality failures, and on-chip test — MBIST for memories (Chapter 8), LBIST / in-field BIST for safety self-test in the running system (Chapter 9). The chapter-closing principle: DPPM (defective parts per million) is the customer-facing quality metric, produced by the chain coverage → escape rate → DPPM; a market-set DPPM goal works backward to set the required coverage target given defect density; and because DPPM depends on both coverage and defect density and even 100% modeled coverage cannot catch unmodeled defects, hitting an aggressive DPPM takes higher coverage plus extra test insurance (SLT, burn-in, BIST) — all delivered at an acceptable test cost (1.4). That sentence is the entire manufacturing-test mindset: catch defects, minimize escapes, hit the DPPM goal affordably.

8. Common Mistakes

Treating DPPM as fixed. It's market-set and contractual — consumer and automotive live in different worlds.
Assuming 100% coverage = 0 DPPM. Unmodeled defects escape; zero-defect needs insurance (SLT/burn-in/BIST).
Ignoring defect density. DPPM depends on coverage and the process — a noisy process needs higher coverage.
Setting coverage without a DPPM goal. Coverage targets should be derived backward from the DPPM the market demands.
'The tester is fine, so quality is fine.' A passing tester at low coverage still ships escapes → high DPPM.

9. Industry Best Practices

Derive the coverage target from the DPPM goal and the defect density (Chapter 6 signoff).
Match effort to market — consumer vs automotive/medical demand very different DPPM (and coverage).
Add insurance for unmodeled defects in zero-defect markets — SLT, burn-in, MBIST/LBIST/in-field BIST.
Track field DPPM against the contract and feed it back to coverage targets.
Balance DPPM against test cost (1.4) — hit the goal at an acceptable cost per die.

10. Senior Engineer Thinking

Beginner: "We have high coverage, so DPPM must be basically zero."
Senior: "DPPM is set backward from the market's goal, and it depends on coverage and defect density. High modeled coverage still leaves unmodeled defects — so for automotive I hit the coverage target the DPPM demands and add SLT/burn-in/BIST insurance. The chain is coverage → escapes → DPPM; I engineer the whole chain to the contractual number, at an acceptable test cost."

The senior treats DPPM as a market-set contract and engineers the whole chain to it — coverage plus insurance.

11. Silicon Impact

DPPM is where the manufacturing-test mindset meets the balance sheet and the safety case. It's the number in the customer contract, and in automotive/medical it's bound to functional-safety standards (ISO 26262 and kin) — miss it and you face penalties, line-down at the customer, recalls, or safety liability. This lesson closes Chapter 1 by making the causal chain explicit: coverage → escape rate → DPPM, run backward from a market-set goal to a required coverage target (Chapter 6). The two nuances are what separate a naive engineer from a real one: DPPM depends on defect density too (so process maturity shares the load with coverage), and 100% modeled coverage is not 0 DPPM (so unmodeled defects demand insurance — SLT/burn-in in this chapter, MBIST/LBIST/in-field BIST later). For the RTL/DFT engineer, the punchline is that the DPPM a product can promise is ultimately bounded by the testability you build into the RTL — untestable logic caps coverage, caps DPPM, and caps which markets the part can even enter. Every remaining chapter — fault models (2), scan (3), scan insertion (4), ATPG (5), coverage closure (6), compression (7), memory/logic BIST (8–9) — exists to move a point on this chain: more coverage per test-second, fewer escapes, lower DPPM, at acceptable cost. That is the whole game, and Chapter 1 has now given you its scoreboard.

12. Engineering Checklist

Stated the DPPM goal from the market/contract (consumer vs automotive/medical).
Derived the required coverage target backward from DPPM and defect density (Ch6).
Remembered 100% modeled coverage ≠ 0 DPPM — planned insurance (SLT/burn-in/BIST) for zero-defect.
Tracked field DPPM against the contract and fed it back to coverage.
Balanced DPPM against test cost (1.4) — met the goal affordably.

13. Try Yourself

For a lot with 1% defect density, compute DPPM at 99% and 99.9% coverage (100 → 10 DPPM).
Work backward: for a 10 DPPM automotive goal at 1% defect density, find the required coverage (99.9%).
Now improve the process to 0.5% defect density — recompute the required coverage for 10 DPPM (99.8%). Note both levers.
Argue why 100% modeled coverage still isn't 0 DPPM, and name the insurance layers (SLT, burn-in, MBIST/LBIST).
Summarize the chain in one sentence — coverage → escapes → DPPM → customer — and where the goal enters (backward).

The arithmetic is tool-independent — a spreadsheet suffices. Coverage numbers come from the ATPG flow (Chapters 5–6); the mindset is what this chapter delivers. No paid tool required.

14. Interview Perspective

Weak: "DPPM is how many bad parts there are."
Good: "DPPM is defective parts per million shipped — the escape rate normalized to a million."
Senior: "DPPM = defective parts per million shipped — the customer-facing quality metric, produced by the chain coverage → escape rate → DPPM. It's market-set: consumer tolerates higher DPPM, automotive/medical demand single-digit, zero-defect (ISO 26262). The goal runs backward — a DPPM target sets the required coverage given defect density. And two things engineers miss: DPPM depends on coverage and defect density, and 100% modeled coverage isn't 0 DPPM because of unmodeled defects — so zero-defect parts add SLT, burn-in, and on-chip BIST as insurance. The mindset: catch defects, minimize escapes, hit the DPPM goal affordably."

15. Interview / Review Questions

16. Key Takeaways

DPPM (Defective Parts Per Million) is the customer-facing quality metric — the escape rate normalized per million shipped.
The chain this chapter built: fault coverage → escape rate → DPPM → customer quality — raise coverage, lower DPPM.
A quality goal is a market-set DPPM target — consumer tolerates higher DPPM; automotive/medical/aerospace demand single-digit, zero-defect (ISO 26262) — and it works backward to set the required coverage target (Chapter 6) given the process defect density.
DPPM depends on both coverage and defect density, and 100% modeled coverage ≠ 0 DPPM because unmodeled defects escape — so zero-defect markets add insurance (SLT, burn-in, MBIST/LBIST/in-field BIST).
The manufacturing-test mindset of Chapter 1 in one line: catch defects, minimize escapes, and hit the DPPM goal at an acceptable test cost. Next: Chapter 2 — Fault Models (how we mathematically model the defects this chapter has been counting).

17. Quick Revision

DPPM — the quality scoreboard (Ch1 closer). DPPM = defective parts per million shipped = escapes normalized. Chain: COVERAGE → ESCAPE RATE → DPPM → CUSTOMER. Quality goal = a market-set DPPM target (consumer higher; automotive/medical single-digit, zero-defect, ISO 26262) → works BACKWARD to set the required COVERAGE target (Ch6) given DEFECT DENSITY. Nuances: DPPM depends on coverage AND defect density; 100% modeled coverage ≠ 0 DPPM (unmodeled defects) → add SLT/burn-in/BIST insurance. Mindset: catch defects, minimize escapes, hit DPPM affordably. Next: Chapter 2 — Fault Models.