TECHNICAL WHITE PAPER

Calibrated Leak Technology
White Paper

An Industry Encyclopedia from Principles to Practice
Leak Rate · Flow Regimes · Processes · Metrology · Application Methodology · Solutions

RealMeter Instruments (Shanghai) Co., Ltd.
V2.6 Final Edition · Four Parts, Sixteen Chapters · July 2026

Abstract: From first principles, this paper presents the physics, processes, metrology and application methodology of calibrated leaks — leak-rate language, three flow regimes, the full process spectrum, lifetime physics, uncertainty, leak-test methodology, quantitative semiconductor RGA, system-level air-tightness accuracy, and the Three-Reference helium-reduction paradigm — concluding with solutions and field data across all leak rates, all pressures and all gases.

Chapter 1  What Is a Calibrated Leak: The "Yardstick" of Leak Testing

1.1  A Decision Made Every Second

On production lines for power batteries, automotive air conditioning, semiconductors and medical devices, thousands of such judgments are made every second: does this part leak or not? OK — release; NG — reject. The judgment is made by a leak detector: a helium mass-spectrometer leak detector, a differential-pressure air-tightness tester, or a halogen leak detector. But the detector itself cannot "know" how large a leak counts as rejectable — it is merely an instrument that amplifies a gas signal into a number. What truly answers "how big counts as leaking" is a thumb-sized device: the calibrated leak.

Its role is exactly that of a weight on a balance, or of a standard voltage source when calibrating a multimeter: the instrument gives a reading, the standard tells whether the reading is trustworthy. The leak detector is the scale; the calibrated leak is the weight. If the weight is wrong, the more precise the scale, the more confidently it errs. The two costliest quality accidents on a production line — false rejects (good parts scrapped) and missed rejects (defective parts shipped) — when traced to the end, point in a remarkable proportion of cases to the same root cause: the yardstick itself is untrustworthy, or its working conditions were not used correctly.

1.2  Definition: A Metrologically Characterized "Standard Source"

The definition contains three indispensable keywords — they form the essential difference between a "standard source" and an arbitrary pinhole:

These three elements are also the undercurrent of the whole book: every later chapter — flow-regime physics (Ch. 3), the process spectrum (Ch. 4), lifetime physics (Ch. 5), uncertainty (Ch. 7) — is essentially answering "how to push the three elements to their limits."

1.3  Position in the Measurement System

The calibrated leak sits in the exact middle of a complete chain of trust. Upstream lies metrology; downstream lies product quality:

• Upstream, it is the calibrated party: its leak rate is assigned by a calibration facility (flow-meter method, fixed-quantity gas method, or comparison method) and traced to the national primary standard[1];
• Downstream, it is the assigning party: the detector's sensitivity and the line's reject threshold are anchored by it.

The calibrated leak therefore has a property rarely seen in other measuring instruments: its error is amplified twice — first copied into the detector's systematic error, then copied into every product judgment. A calibrated leak with a +20% rate deviation, unknown to its user, keeps an entire production line running day and night against a wrong threshold. This is why metrology specifications impose hard requirements on the annual stability of calibrated leaks (JJF 1833—2020 requires ≤ 3%/year) and on calibration-facility uncertainty (relative expanded uncertainty better than 10%, k = 2)[1] — and why this book devotes a full chapter each to "decay" and to "uncertainty."

1.4  Leak ≠ Leak Point ≠ Orifice Diameter: The Equivalent Leak Channel

Three concepts are constantly conflated in the field and must be separated:

The key insight: leak testing measures "leak rate," never "orifice diameter." A 50 μm crack and a 5 mm capillary are, for sealing purposes, the same problem as long as they transport gas equally under the same conditions. A calibrated leak provides precisely such a stable, controllable, reproducible equivalent leak channel — used to "impersonate" the critical defect on the workpiece (the OK / NG judgment and helium-reduction systems of Ch. 10–11 are built on this). Every conversion of "what hole diameter does this leak correspond to" is meaningless without specifying gas species, pressure and flow regime (Ch. 3 gives a rigorous treatment).

1.5  The Traceability Chain: From National Primary Standard to Line Judgment

How does the leak-rate value reach your production line, step by step? The figure below shows the complete chain:

Every link has a name: the primary leak-rate standard maintained by the National Institute of Metrology (NIM) → calibration facilities at metrology institutes at various levels (JJF 1833—2020: flow-meter / fixed-quantity gas facilities must have relative expanded uncertainty better than 10%, k = 2; comparison facilities better than 20%)[1] → reference calibrated leaks (master standards used by institutes to establish their standards, each with a national-level certificate) → working calibrated leaks (for daily line checks and detector calibration) → detector readings → product OK / NG judgment.

This chain explains two phenomena familiar in the industry:

• "Certificates with different stated conditions give numbers that don't match": every link is bound to working conditions (gas, pressure, temperature, outlet conditions); change conditions at any link and every downstream number is distorted;
• "Establishing a standard is the institute's business — and also the leak manufacturer's": a reference calibrated leak must itself be an extremely stable product, holding its value between two calibrations (typically 12 months); annual stability, temperature coefficient and shipping robustness directly determine the quality of the established standard. Chapter 15 presents a complete standard-establishing leak set conforming to JJF 1833—2020.

1.6  A Taxonomy of Calibrated Leaks

The calibrated leak is not a single product but a multi-dimensional family. Master the following six classification dimensions and you hold the "map coordinates" of this book:

Reservoir vs passive is the first choice to make in the field, so build the intuition early: a reservoir-type leak is like a "self-inking standard stamp" — open the valve and it works; portable, independent, ideal for detector calibration and metrological traceability. A passive leak is like a "standardized precision capillary" — driven by the local gas supply, it can precisely simulate workpiece leakage at any pressure, and is the foundation of OK / NG judgment systems and helium-reduction schemes (Ch. 10, 11). Neither is better; they simply have different applicable boundaries.

1.7  Why This Industry Is Upgrading

Over the past decade, four industry-level changes have pushed the calibrated leak from "accessory of the leak detector" to "foundation of the quality system":

1.8  Summary

The next chapter settles the most fundamental — and most error-prone — language problem of this industry: the units and magnitudes of leak rate.

Chapter 2  The Language of Leak Rate: Units, Magnitudes and Intuition

2.1  What Leak Rate Measures: pV per Unit Time

Leak rate is neither hole size nor gas volume, but the product of pressure and volume transported through the leak channel per unit time:

Q — leak rate; pV — the pressure × volume product of the transported gas (a measure of "amount of gas"). Under isothermal conditions, pV is proportional to the number of molecules, so leak rate is essentially "how many molecules pass per unit time."

Why not use volume flow directly? Because gas is compressible: 0.01 cm³/s on the high-pressure side may expand to 10 L/s on the low-pressure side. Only pV is invariant along the path (at constant temperature) — this is why vacuum technology worldwide uses "pressure × volume / time" as the base dimension of leak rate. All the units below are merely different pV conventions.

2.2  The Unit Family: One Quantity, Five Dialects

The three most useful relations to memorize:

• 1 Pa·m³/s = 10 mbar·L/s ≈ 9.87 atm·cc/s — the "trio" of the vacuum community, numerically interchangeable within ±1.3% for engineering estimates;
• 1 mbar·L/s ≈ 59.2 sccm ≈ 0.987 std.cc/s — the bridge between vacuum and flow communities;
• 1 atm·cc/s ≈ 1.013 mbar·L/s — semiconductor literature mostly uses atm·cc/s; it can be read as mbar·L/s directly, with the difference absorbed in rounding.

g/y is a special dialect: it is a mass flow, whose conversion to pV flow depends on the molar mass of the medium:

m — molar mass (g/mol); t — temperature (°C); 22.413 L/mol is the standard molar volume; the 273/(273+t) factor corrects the volume from t °C to 0 °C; 1.013 converts mbar to atm. For R134a (m = 102 g/mol): 1 mbar·L/s ≈ 1.3×10⁵ g/y, i.e. 1 g/y ≈ 7.6×10⁻⁶ mbar·L/s — thus the "a few grams per year" regulatory language of the refrigeration industry connects with the "10⁻⁵ ~ 10⁻⁶ class" technical language of the vacuum industry.

2.3  Magnitude Intuition: Anchoring Leak Rates to Time

Numbers alone build no intuition; time does. How long does a leak of a given rate take to leak one standard cubic centimeter (1 std.cc ≈ 1 mL of gas at 0 °C, 1 atm)?

Three intuitions to carry away: ① each order of magnitude is a tenfold world — the difference between 10⁻⁵ and 10⁻⁶ is not 10%, but ten times the time; ② everyday language is misleading — a "slight leak" of 10⁻³ is a hundred billion times a "metrology-class leak" of 10⁻¹³; ③ small leaks do not mean small risk: a 3.5×10⁻⁶ mbar·L/s hydrogen calibrated leak releases only ~0.3 mL of hydrogen per day, far below any combustion-concern threshold — magnitudes must be discussed together with gas species and environment.

2.4  Reading Certificates: Vacuum vs Atmosphere Outlet, Gauge vs Absolute Pressure

Two pairs of conditions on a certificate change the numerical value of the "same" leak; they are the most frequent sources of field disputes.

First pair: vacuum outlet vs atmosphere outlet. The same channel into vacuum (outlet ≈ 0) vs into atmosphere (outlet ≈ 1 atm) gives different rates — in molecular flow the difference is small (rate ∝ Δp: 11 bar vs 10 bar, about 10%); in viscous flow it can be large (rate ∝ p₁² − p₂²: 11² − 1² vs 11² − 0, about 0.8% at 11 bar, but 25% at 2 bar). Certificates must state the outlet condition; between the two ratings, use the flow-regime formulas of Chapter 3 to convert, never "approximately the same."

Second pair: gauge vs absolute pressure. Field pressure gauges read gauge pressure; physical formulas use absolute pressure. A part tested at 20 kPa gauge is at ~121 kPa absolute; a 1% error in absolute pressure terms is not 0.2 kPa but 1.2 kPa. All calculations in this book use absolute pressure; all field settings must add one atmosphere (~101.3 kPa) before use.

2.5  Flow-Rate Units and Temperature Reference: sccm Is Not ccm

The MFC and calibration communities distinguish "standard-state volume flow" (sccm: volume per minute reduced to 0 °C, 1 atm) from "actual volume flow" (ccm: at actual temperature and pressure). They differ by the temperature/pressure correction factor: 1 ccm measured at 23 °C equals ~0.92 sccm. Certificate conversions and calibration-system settings must state the reference temperature — the default in this book is 0 °C (273.15 K), and 23 °C measurements are explicitly marked when used.

2.6  Three Deadly Misconceptions

Misconception 1: treating mbar·L/s as "hole diameter." The same 10⁻⁵ mbar·L/s may correspond to completely different physical channels for different gases, pressures and structures — leak rate is a transport capability, not a geometric size (§1.4).

Misconception 2: converting between gases with one fixed factor. The truth: in molecular flow Q(He) ≈ 2.65 × Q(air) (light molecules leak faster, the 1/√M law of Eq. 3-3), while in viscous flow Q(air) ≈ 1.06 × Q(He) (inverse viscosity ratio, Eq. 3-2) — the two factors, 2.65 and 0.94 (= 1/1.06), differ by nearly a factor of three and point in opposite directions. Gas-conversion factors depend on the flow regime, which depends on pressure and orifice size — see §3.6; this is the most underestimated pit in the industry.

Misconception 3: comparing certificate values without conditions. Two certificates both saying "1×10⁻⁵ mbar·L/s" may differ by tens of percent in actual behavior if one is vacuum-outlet He and the other atmosphere-outlet N₂ — the four conditions (gas, inlet pressure, outlet, temperature) must be aligned before comparison (Rule 2-1).

2.7  Summary

With the language in place, the next chapter opens the leak itself: what determines its output — the three flow regimes, and the temperature effects that quietly rewrite the numbers.

Chapter 3  Physics Inside a Leak: Three Flow Regimes and Temperature Effects

3.1  One Number Divides Three Worlds: The Knudsen Number

Gas flowing through a microchannel exhibits three completely different personalities, depending on the ratio of the molecular mean free path λ (the average distance a molecule travels between collisions) to the characteristic channel size d — the Knudsen number Kn = λ / d:

• Kn ≪ 1 (viscous flow): molecules collide mostly with each other; gas behaves as a continuous fluid, "squeezed" through by pressure, following Poiseuille's law;
• Kn ≫ 1 (molecular flow): molecules collide mostly with the walls; each molecule passes independently, with no collective behavior;
• 0.01 < Kn < 1 (transitional flow): both mechanisms coexist; the hardest to compute and the most error-prone region in engineering.

Concrete feel: for a leak with d = 5 μm at 1 atm inlet, Kn ≈ 0.014 (viscous side); pump the inlet down to 100 Pa and λ grows a thousandfold, Kn ≈ 13 (molecular side). The same physical leak may switch flow regimes as operating pressure changes — this is the physical root of "numbers that don't match" when rating conditions differ (echoing Rule 2-1). For extremely fine channels of 50 nm, the flow is already molecular even near atmospheric pressure.

3.2  Viscous Flow: Q ∝ (p₁² − p₂²), Sensitive to the Fourth Power of Diameter

d — channel diameter; L — length; η — gas viscosity; p₁, p₂ — inlet / outlet absolute pressures.

Three engineering corollaries: ① d⁴ sensitivity — a 1% change in diameter shifts the rate by 4%; this is the fundamental difficulty of making stable leaks, and the reason metal-crushed and drawn-glass processes struggle with consistency; ② rate depends on the pressure difference of squares — doubling inlet pressure nearly quadruples the rate, which is why rating pressure must be pinned down; ③ rate is inversely proportional to viscosity η — for the same channel, gases with higher viscosity (air 18.5 μPa·s) leak slower than lower-viscosity hydrogen (8.9 μPa·s), and helium (19.6 μPa·s) leaks slowest of the three.

3.3  Molecular Flow: Q ∝ Δp / √M, Light Molecules Run Faster

v̄ — mean molecular thermal speed; M — molar mass; T — absolute temperature.

Key corollary: in molecular flow the rate is inversely proportional to √M — for the same channel, the smaller the molecular mass, the faster the leak. He (M = 4) vs N₂ (M = 28): the molecular-flow rate ratio is √(28/4) ≈ 2.65. This is the physical reason helium is the premier tracer gas: for the same defect, the helium signal is 2.65 times the air signal (and its background in air is only 5.2 ppm, so the signal-to-noise ratio is overwhelming). Note the d³ scaling here (vs d⁴ in viscous flow): geometry matters differently across regimes — the basis of the "process yardsticks" in Chapter 4.

3.4  Transitional Flow: Add the Two, Never Use Just One

Z(Kn) — a transition correction factor (empirical form, e.g. the Knudsen interpolation); in engineering it is often sufficient to add the two components directly and validate against measurement.

The danger of the transitional region: the apparent "pressure exponent" drifts between 1 and 2, and gas-conversion factors lie between the viscous and molecular limits without a universal value. Whenever you see "rate vs pressure is neither linear nor quadratic," you are in the transitional region — at that point, only a full-range calibrated leak, or a computational model validated by full-range data, can give trustworthy ratings. RealMeter's RMI-Cal™ conductance-simulation engine is built on precisely this class of models (Ch. 13).

3.5  Temperature Effects: Three Mechanisms, Three Personalities

Temperature is the invisible hand that rewrites leak rates daily. The three structural families behave completely differently:

Eq. (3-6) is the theoretical lower bound for viscous flow; with structural corrections, real metal-channel leaks measure about −0.6 %/°C. Eq. (3-7) is the Arrhenius form for permeation: Ea — activation energy.

The engineering conclusion writes itself: temperature coefficient is the sharpest single indicator separating leak technologies. A production floor without climate control swings ±5 °C day and night: a permeation leak drifts ±20%, while a metal-channel leak drifts less than ±1% (viscous) or ±0.85% (molecular) — whether a daily OK / NG judgment needs temperature compensation depends entirely on this number. This is why the selection chapter (Ch. 12) makes "tempco ≤ 0.2 %/°C" a hard threshold for line duty.

3.6  Gas Conversion: One Table, Two Regimes

The rigorous answer to Misconception 2 of §2.6 — rate ratios between gases for the same channel (using He→N₂ and He→H₂ as examples):

3.7  Liquid-Medium Leaks: Evaporation–Transport, Not Flow

Water, dodecane, DMC and other liquid-media "leaks" do not flow — the liquid evaporates at the channel inlet, vapor is transported along the channel, and condenses at the outlet. The rate depends on saturated vapor pressure and channel conductance, and is far more temperature-sensitive (the vapor-pressure curve is steep): dodecane runs ~12 %/°C, so a ±0.5 °C temperature-control error alone means ~6% rate drift. Such leaks are mainly used for liquid-cooling-plate and refrigerant-circuit simulation; Ch. 6 (§6.5) and Ch. 9 (§9.5) discuss their rating and use.

3.8  Decay of Reservoir-Type Leaks: An Exactly Computable Loss

A reservoir-type (self-contained) leak is a small gas tank leaking through a calibrated channel: as gas leaves, pressure falls, and the rate falls with it — this is "decay," not a fault but an inherent property. With initial rate Q₀, reservoir pressure p and volume V:

Reading the figure: ① 11 MPa / 300 cc — only 22.3% decay in ten years; ② 11 MPa / 160 cc — 35.0% in ten years; ③ 1 MPa / 160 cc (ordinary charge) — 70.3% gone in just four years. Decay is not destiny; it is design (the day-by-day iteration behind this figure is equivalent to the closed-form solutions of Eqs. 3-9 / 3-10): high-pressure charging and enlarged volume push annual decay below 3%, satisfying the JJF 1833—2020 stability requirement; low-pressure small-volume products are disqualified at birth. When choosing a reservoir leak, ask for the decay-curve calculation before the certificate (Ch. 12).

3.9  Summary

With physics in hand, Part II uses this yardstick to measure every process route in the industry: who can make stable channels, and who cannot.

Chapter 4  The Full Process Spectrum of Calibrated Leaks

4.1  Four Physical Yardsticks for Judging a Process

Chapter 3 told us that a leak's macroscopic behavior is governed by four physical quantities. To judge any process, simply ask how well it controls these four:

The orifice item deserves quantification. From Q ∝ d⁴, the relation between rate tolerance and orifice tolerance:

For a 5 μm orifice, ±2.4% means diameter control within ±120 nm. Leak manufacturing is, at its core, a contest of sub-micron geometric control — every process difference below ultimately lands on this point.

4.2  Drawn-Glass Capillary: The First-Generation Mainstream

Principle: a glass tube is fused with a metal wire (or glass rod) and drawn thin; the micro-gap formed by the differential thermal expansion of glass and metal on cooling serves as the leak channel. Capability: typical rates 10⁻⁴ ~ 10⁻⁷ mbar·L/s, pressure generally < 1 MPa. Failure modes: first, the gap is an annular taper rather than a circular tube — orifice varies along the axis, forward and reverse conductance differ, i.e. the rate is directional (§4.9, §5.4); second, glass brittleness means it fears drops, freezing and pressure shocks, with short field life; third, the gap geometry is not designable — target rates are achieved by "draw a batch, measure a batch, pick a few," with consistency left to luck.

4.3  Crushed-Metal: Simple but Uncontrollable

Principle: a fine metal tube (or metal foil sandwiching a wire) is repeatedly crushed, measured, crushed again, until the rate falls into the target band. Capability: typical rates 10⁻¹ ~ 10⁻⁶ mbar·L/s, pressure generally < 2 MPa. Failure modes: the crushed channel is a crescent-section product of trial and error — directional as well; metal creeps slowly under stress, so the rate drifts over time; die wear translates directly into batch variation. Its merits are low cost and large rates, so it survives in coarse testing — but "adjustable, computable, reproducible" is out of the question.

4.4  Quartz Permeation Membrane: The Old King of Tiny Leaks

Principle: small molecules such as He undergo a "dissolve–diffuse–desorb" permeation through quartz glass; the rate is set by membrane area and thickness. Capability: can reach 10⁻⁹ ~ 10⁻¹² class — long the mainstream choice for built-in calibration leaks of helium detectors. Failure modes: first, tempco as high as ~ +4 %/°C (the Arrhenius mechanism of Eq. 3-7): a 5 °C workshop swing means ~20% drift, so a temperature-correction table is mandatory; second, the differential pressure across the membrane is limited (the membrane is the weak point) and glass is brittle; third, the medium is locked — only small molecules like He and H₂ permeate; N₂, Ar, refrigerants and mixtures are all excluded; fourth, quartz permeation of He is itself a ~10⁻¹⁰ atm·cc/s "background" — going smaller means drowning in your own background (§5.2).

4.5  Powder-Sintered and Porous Media

Principle: metal or ceramic powder is sintered into a porous body; conductance comes from the statistical average of a vast number of parallel micro-pores. Capability: suits larger rates of 10⁻¹ ~ 10⁻⁴, decent pressure tolerance. Failure modes: wide pore-size distribution — consistency is statistical, not geometric; the huge internal surface area makes adsorption / desorption significant (§5.3), so humidity and contamination slowly drift the rate; cannot scale down across orders of magnitude.

4.6  Critical-Flow Nozzle: A Special Solution for Large Flows

Principle: when the downstream/upstream pressure ratio falls below the critical ratio (~0.528 for air), the throat reaches sonic speed; flow depends only on upstream pressure and throat area, independent of downstream. Capability: the correct solution for large-flow (sccm to L/s) gas-flow standards, widely used in MFC calibration. Failure modes: requires a sufficient pressure ratio to stay "critical" — fails at low pressure differences; cannot be made small; sensitive to upstream cleanliness. It is a flow-metrology device rather than a leak-testing standard — a distant cousin of this book's protagonist.

4.7  MEMS Etching: A Semiconductor-Mindset Attempt

Principle: microchannels are photolithographically etched into a silicon wafer and bonded into a package. Capability: geometrically controllable and batch-replicable in principle — a hot direction in academia and startups. Failure modes: silicon brittleness and the sealing background of anodic glass bonding limit pressure rating and minimum rate; adhesives and solders in packaging introduce outgassing and permeation; batch-to-batch consistency and long-term stability data remain insufficient. A direction worth watching, but the road "from wafer to production-line yardstick" is not yet finished.

4.8  Metal-Based Microchannel: Turning Leaks from "Craft" into "Design"

The six processes above share a common fate: the rate is a result "made, then measured," not a design "computed, then built." The metal-based microchannel process (RealMeter RMI-MTC™) inverts the logic: geometrically defined circular microchannels are machined directly into a metal body (orifice continuously controllable from 50 nm to 200 μm); target diameter and length are computed from the conductance formulas (Eq. 3-2, 3-3, 3-4) first, then built to the drawing. Measured against the four yardsticks:

1. Orifice and consistency: geometry is a design value; rate tolerance starts at ±10%, available at ±5% or tighter on request (the ±2.4% orifice tolerance of Eq. 4-1 is routine for CNC machining) — batch consistency no longer depends on selection;
2. Channel regularity: near-ideal circular tube ⇒ no directionality, clog-resistant (no throttling throat in a uniform bore), computable across all regimes;
3. Material background: metal body + all-metal sealing (permeation < 10⁻¹⁵ atm·cc/s, §5.2) — no organic materials introduced;
4. Mechanical strength: monolithic metal structure ⇒ passive leaks rated to 40 MPa, passing a 1.2 m × 5-cycle drop test, operating temperature −20 ~ +65 °C[3].

The result is a combination no traditional process can achieve simultaneously: all leak rates (100 mL/s ~ 10⁻¹³ Pa·m³/s) × all pressures (0.1 Pa ~ 40 MPa) × all media (any non-corrosive gas / mixture / liquid) × rate and specified pressure customized together. This is not leadership in any single metric — it is a generational change across all four yardsticks.

4.9  Comparison of the Seven Processes

How to read the map: the further left (small tempco) and the further up (high pressure rating), the stronger the process. The seven processes stratify naturally — this is not any vendor's marketing conclusion but a direct corollary of Chapter 3 physics: tempco is set by flow regime and material mechanism; pressure rating is set by structure and sealing.

4.10  Summary

Process determines a leak's "nature"; the next chapter discusses its "nurture": clogging, background, adsorption, creep — every enemy a leak faces over its lifetime.

Chapter 5  Lifetime Physics of Leaks

5.1  The Threat List: Five Ways a Leak Dies

In reading this list, one classification axis matters more than the count of five — classify by where the cause of death resides. On one side is external intrusion: particles and liquid films breaching the filter element from outside — triggered at the usage end, the only death the user can actively defend against. On the other side is endogenous failure: causes written into the genes of material and geometry, which no amount of careful usage can rewrite. Endogenous failure falls into two families — endogenous clogging (the material "grows" its own blockage: progressive oxide narrowing, surface poisoning of the permeation interface, pore creep, spalling aging products) and structural destruction (fracture, relaxation-springback, thermal instability — rewriting the rate directly or through distortion). In field statistics, about 90% ~ 99% of conventional-process failures belong to these two endogenous families (the §5.6 matrix unfolds along this axis); external intrusion is only the small remainder. Recognizing this axis tells you, in every section that follows, whether a "defense strategy" belongs at the design end or the usage end.

5.2  Seal Background: Who Drowned 10⁻¹²

Making a channel-type leak nominally rated 10⁻¹² atm·cc/s: the real difficulty is not "making such a small channel" but "proving that the measured 10⁻¹² comes entirely from the channel itself." A leak's total output is the channel rate plus system background:

Engineering rule (same SNR logic as Ch. 3): the sum of all background terms should be below 1/10 of the nominal rate. Self-permeation magnitudes of common sealing schemes:

This figure explains an industry fact: why channel-type He leaks of 10⁻¹¹ ~ 10⁻¹³ Pa·m³/s (10⁻¹⁰ ~ 10⁻¹² mbar·L/s) were long available only as permeation membranes — not because channels couldn't be made, but because conventional seals (10⁻⁹ ~ 10⁻¹⁰) sit exactly in that range and drown the channel signal. Every order of magnitude the seal background is pushed down unlocks one more order of magnitude in the leak's lower limit. That is why the figure "all-metal sealing < 10⁻¹⁵ atm·cc/s" (5~6 orders of magnitude ahead of conventional sealing[3]) is the entry ticket to high-end leaks — and the premise on which the 10⁻¹¹/10⁻¹²/10⁻¹³ series of Chapter 13 can exist.

5.3  Adsorption and Outgassing: The Invisible "Phantom Leak"

Every solid surface exposed to gas carries an adsorbed molecular layer. An order-of-magnitude estimate: one monolayer on 1 cm² holds ~10¹⁵ molecules, i.e. a gas quantity of:

These molecules desorb slowly under vacuum, acting as a "phantom leak": for a 10 L vacuum system targeting a 10⁻⁶ Pa background, full desorption of just 1 cm² of monolayer raises the pressure by 4×10⁻⁴ Pa — an overshoot of ~400× (more than two orders of magnitude). This is the daily reality of RGA / UHV work: the biggest "leak" in the system is often not a leak at all, but a surface. The defense strategy follows directly:

Defense therefore follows directly:

• Materials: no adhesives, no polymer seals; all-metal construction plus electropolishing to cut specific surface area;
• Structure: valve and leak integrated into one body, eliminating assembly interfaces (every interface is a pocket that traps gas);
• Usage: bake out UHV systems; pre-evacuate a calibrated leak before connection so the atmospheric adsorbed layer is not carried in.

Adsorption has a subtler incarnation: stratification and memory in gas mixtures — heavy components adsorb into and release from organic seals and dead corners at different rates than light ones, so a mixture leak's output composition drifts over time. This is the same root as the "memory effect" of dead volume in §6.2.

5.4  Directionality: An Overlooked Authenticity Test

The conductance of an ideal circular tube depends only on geometry and flow regime, never on flow direction — mounted forward or reversed, the rate is strictly equal. Tapered gaps and crushed channels, being geometrically asymmetric (sudden contraction at the inlet, gradual expansion at the outlet), can differ by a visible ratio, in extreme cases by a factor of two. Directionality causes two practical harms:

• if calibration and use occur in different orientations, the rate is simply wrong;
• if a leak is accidentally reverse-mounted in the field, readings distort without any warning.

Directionality is therefore both a usage trap and a process touchstone (Rule 4-1): equal bidirectional rates are the most direct and hardest-to-fake evidence of a "near-ideal circular tube." Add this one action to acceptance inspection (measure forward, measure reversed): five minutes of cost, enormous information.

5.5  Mechanical and Thermal Damage: The Daily Violence of the Factory Floor

The difference between a laboratory leak and a production-line leak is that the latter lives amid vibration, drops, temperature swings and oil mist. The physical essence of damage is geometric mutation: channel cracks, loosened seal interfaces, displaced capillaries — each directly rewrites d or the seal background, and the rate jumps irreversibly. Mechanical tolerance must therefore be a design specification, not a usage caution:

• Monolithic metal structure (no adhesives, no glass) eliminates brittle failure modes at the root;
• Quantified verification: e.g. rate unchanged after a 1.2 m × 5-cycle drop test, unchanged after −20 °C snap freezing (Ch. 15 presents measured data and video references[3]);
• Wide-temperature design: operating −20 ~ +65 °C, tolerating −30 °C freezing, so the leak can follow the workpiece into real conditions rather than a greenhouse.

One account is often left out of the ledger: the most dangerous outcome of structural destruction is not death but distortion. A shattered glass leak is discovered and replaced the same day — an explicit failure costs one leak; but a channel leak that "works wounded" after geometric relaxation or micro-deformation does not stop breathing — it only quietly changes its rate, handing the line a wrong reference while looking healthy for months. Covert distortion corrupts every judgment the line makes (this is the microscopic mechanism of the "blind judging" in Ch. 10). In rating mechanical endurance, therefore, "survives the drop" is only the passing grade; "rate unchanged on re-measurement after the drop" is the full score — which is exactly why quantified verification (the previous item) must demand re-measurement data.

5.6  The Failure Matrix: A "Map of Deaths" for Four Process Routes

Overlaying Chapter 4's process routes with this chapter's failure mechanisms yields a complete failure matrix. Three reading conventions first:

• Filter elements are a constant, not a variable: precision filtration (against dust and liquid) is standard across all processes and broadly equal in protective capability — it creates no differentiation. The differences lie only in "whether the channel itself is vulnerable" and "what other endogenous failures remain." Filter placement follows the contamination path: passive leaks are guarded at both ends; reservoir leaks, whose sealed reservoir leaves the outlet as the only ingress path, are guarded at the outlet only;
• The denominator must be stated: in field statistics, roughly 90% ~ 99% of conventional-process failures belong to the two endogenous families — endogenous clogging and structural destruction (including rate distortion and secondary clogging caused by structural destruction); external intrusion is the smaller share. Once microchannel construction structurally deletes the two endogenous families, the residual cause of death is almost solely "external contamination breaching the filter element." The two statements do not contradict — their denominators differ: one is the failure composition of conventional processes, the other the residual composition after the microchannel reset;
• This matrix disparages no process: every route has its application window and its historical contribution. The matrix's value is making the full failure spectrum visible so that selection becomes an informed decision — the very premise of process improvement;
• "Permeation leaks don't clog" is only half true: a permeation membrane has no through-channel — helium dissolves into the glass and diffuses through the bulk — so particle clogging simply has no site to occur (matrix row 1); that has been the scheme's founding merit for decades. But backstreaming pump oil and hydrocarbon vapors film the membrane surface and rewrite helium solubility and the interfacial gradient — "surface poisoning," its equivalent way to die (row 3). Meanwhile, the "channel-type clogs easily" reputation belongs to the old throat-bearing processes (tapered gaps and crushed crescents carry a natural sharp throat where a particle self-locks) — a uniform circular bore has no throat, particles either pass or stop at the inlet face, and with filter elements, intrusive clogging has been downgraded to a low-probability, preventable event. In short: channel types fear "particle clogging"; permeation types fear "surface poisoning" — both sides carry contamination sensitivity, only in different forms.

The defense of intrusive clogging has two ends. Design end: a uniform circular bore has no throat — a particle either passes through or is stopped at the inlet face (where a protective filter can sit); this is the geometric essence of "clog resistance." Dust/water protection at the inlet (e.g. a dust-and-water cap for vacuum duty) keeps interception outside the channel. Liquid-phase condensation obeys the same geometry — oil mist and water vapor condense capillarily inside micro-pores (Kelvin effect: the smaller the pore, the lower the saturation vapor pressure, the easier a liquid bridge forms), and a throat-free, bakeable structure makes it reversible. Usage end: filter and dry the gas supply (oil content < 0.01 mg/m³ class), avoid liquid carryover, seal ports during storage. A rule of thumb: any leak that demands "handle with care, store at constant temperature, dedicated gas supply" is a leak whose unfinished design homework has been dumped on the user.

The last column reveals a structural fact: the microchannel leak's failure modes converge. The whole family of endogenous failures — brittleness, thermal drift, hysteresis, seal aging — is physically deleted by the monolithic metal structure, uniform geometry and all-metal sealing, leaving almost a single residual cause of death: external contamination breaching the filter. The nature of death thereby changes: conventional leaks mostly die of "illness" (failures written into their structural genes), while the microchannel leak can only be "killed" (contamination imported from outside) — the weapon sits in the user's plumbing, not in the leak. High-volume after-sales data agree with this model: in production-line supply to leading battery-makers, the microchannel leak's field failure rate is 1 ~ 2 orders of magnitude below conventional processes, with service life several times longer — "long life" thus turns from a claim into a decomposable mechanism. One closing sentence: a leak's leading cause of death is not "clogging" but "breaking" — endogenous clogging and structural destruction cannot be written into a usage procedure; they can only be written into the design drawings.

5.7  Summary

The next chapter widens the view from the leak body to its "accessory": the valve. A dead volume the size of a fingernail is enough to overturn everything this chapter has carefully engineered.

Chapter 6  Valves, Dead Volume and System Integration

6.1  Dead Volume: The "Hidden Gas Cell" Inside the Valve

Dead volumes of conventional calibrated-leak valves are typically 0.01 ~ 0.1 cc (10 ~ 100 μL). That sounds negligible — the following three calculations will change your mind.

6.2  The Three Costs of Dead Volume (Quantified)

Cost one: cycle loss. Every time the valve opens, the high-pressure gas in the dead volume is vented:

With $V_{\text{dead}}$ = 0.1 cc and 10 MPa (100 atm) charge: loss per cycle = 0.1 × 100 = 10 atm·cc ≈ 10 mL of standard gas. A reservoir leak cycled 100 times wastes 1 L of standard gas for nothing — money for ordinary He, an unacceptable waste for Xe, Kr or ³He isotopes. Compress the dead volume to 0.098 μL and the same conditions cost 0.0098 atm·cc per cycle — about 1 mL per 100 cycles, negligible (Fig. 6-1).

Cost two: vacuum shock. The high-pressure gas trapped in the dead volume rushes into the vacuum system the instant the valve opens. Estimate: 0.1 cc × 100 atm = 10 atm·cc ≈ 10³ Pa·L, dumped into a 100 L chamber at 10⁻³ Pa background — an instantaneous pressure jump of +10 Pa, i.e. 10⁴ times the background: at best the signal is buried for minutes, at worst a protective shutdown. This is why "an ordinary leak must be pre-evacuated before connecting to an RGA," and the physical basis of the zero-volume valve's claim "no shock no matter how long it has been closed": 0.098 μL × 100 atm ≈ 0.01 atm·cc ≈ 1 Pa·L — an instantaneous increment of only ~10⁻² Pa in the same scenario, three orders of magnitude below the conventional valve. This theoretical ledger is backed by public measurements: with an RGA system at a base vacuum of 4×10⁻⁶ hPa, opening a PSOZV™ zero-volume-valve leak whose reservoir holds helium at 100 bar raised the system pressure only to 3×10⁻⁵ hPa — an increment of about +2.6×10⁻³ Pa, less than one order of magnitude — with no disturbance to the turbopump or filament; a conventional-valve leak under the same conditions makes the pressure jump by several orders of magnitude and can even trip the turbopump into protective restart[15]. The measured increment sits in the same 10⁻³ ~ 10⁻² Pa range as the estimate — theory and measurement agree.

Cost three: memory effect. Dead volume is a "slow gas cell": gas from the previous use (especially heavy mixture components and water vapor) is adsorbed and trapped inside, then slowly released in the next measurement, forming false backgrounds and composition drift. It directly lengthens the system's background recovery time — on a high-takt line, every second of recovery time is throughput. Shrink the dead volume by three orders of magnitude and the memory gas shrinks in proportion — that is the quantitative explanation of "fast background recovery."

6.3  The Zero-Volume Valve: From 100 μL Down to 0.098 μL

The zero-volume valve aims to push dead volume an order of magnitude below the physical norm. Taking RMI-PSOZV™/MDZV™ as an example, the dead space is Φ0.5 mm × 0.5 mm:

That is 2 ~ 3 orders of magnitude below the 10 ~ 100 μL of conventional valves, and all three calculations of §6.2 shrink in proportion[3]. The two form factors divide the work:

• MDZV™ manual zero-volume valve: metrology and laboratory scenarios — human-operated, minimal and reliable;
• PSOZV™ normally-closed pneumatic zero-volume valve: automation scenarios. The value of normally-closed design is active fool-proofing — forgetting to close the valve is the number-one human cause of gas loss in reservoir leaks; a normally-closed valve deletes "will someone forget" from the risk list and, combined with automation takt, achieves near-zero natural decay — especially significant for expensive isotope gases (Xe, Kr, ³He).

6.4  The Physics of Partial Rates in Gas-Mixture Leaks

RGA calibration and semiconductor process monitoring make heavy use of gas-mixture calibrated leaks (e.g. 1% each of H₂/He/N₂/Kr/Xe balance Ar, or 1 ppm Kr/Xe balance N₂). There is a widely misunderstood physical point here: in molecular flow, components transport independently, and the partial-rate ratio does not equal the mole-fraction ratio. From Eq. (3-3), the partial rate of component $i$:

An example: 1% He + 99% Ar. In molecular flow the He share of the total rate is

Helium at 1% concentration contributes 3.1% of the rate — the light component is "amplified." Three engineering corollaries:

• A mixture leak's "rate" must state its basis: the partial rate of a component (P basis, e.g. 95P = major-component partial rate) or the total rate (T basis, 95T). Get the basis wrong and the whole calibration is wrong[7];
• In viscous flow there is no √M amplification (components are "bound" together by viscosity) — the same mixture leak outputs different compositions in the two regimes: flow regime decides everything, again;
• Light components escape faster from the reservoir, so the stored composition slowly "gets heavier" over long storage — another reason to watch the decay basis of reservoir-type mixture leaks.

6.5  Liquid-Medium Leaks: The Evaporation–Transport Model

Semiconductor outgassing analysis, EUV carbon-contamination monitoring and aerospace mass-based acceptance all need liquid-medium standard sources (water, n-dodecane C₁₂H₂₆, DMC, PFTBA, etc.). Their physics differs fundamentally from gas leaks: the liquid evaporates first; the vapor is then transported through the channel — the rate is dominated by vapor pressure:

Vapor pressure depends on temperature exponentially (Clausius–Clapeyron), so liquid leaks carry tempcos of several percent per degree — for n-dodecane, ~12 %/°C (exponential)[8]. Three usage rules follow:

• Temperature control is a necessity: a ±0.5 °C control error alone means ~6% rate drift; liquid leaks must be used within ±0.5 °C of the certificate temperature, or corrected strictly by the exponential curve;
• A large tempco is not a defect but a physical property: it is set by the medium's enthalpy of vaporization — no vendor can "improve" it; what can be improved is measuring the correction curve accurately and making the zero-volume valve normally closed (to avoid pointless vapor loss);
• The value of long-chain alkanes: dodecane (M = 170), tetradecane/hexadecane and other high-mass media fill the gap in RGA high-m/z calibration and EUV heavy-hydrocarbon contamination quantification — a category no gas leak can in principle replace.

6.6  Summary

With the physics and hardware of leaks complete, the next chapter enters metrology: how to grade a leak's "accuracy" in the language of uncertainty, and how China's leak-rate values travel from the national primary standard into your hand.

Chapter 7  Uncertainty and Metrological Traceability

Before uncertainty proper, three concepts that are constantly conflated must be separated — they describe three different qualities of a leak, and no axis can substitute for another:

The axes are orthogonal but mutually supporting: however high the machining precision, without low-uncertainty calibration it cannot be proven to a third party; however precise the calibration, poor stability voids the value within months. The most frequent customer-side confusion is taking the manufacturing tolerance for the uncertainty — a leak made to 5×10⁻⁶ mbar·L/s ±10% and then calibrated carries a certificate reading "measured 5.12×10⁻⁶, U = 8% (k = 2)": the ±10% describes manufacturing capability (the scatter of hits around the bull's-eye), the 8% describes how precisely the hit's position is known. The two numbers mean entirely different things and must never be interchanged in contracts or acceptance documents.

7.1  A Leak Rate Without Uncertainty Is Not a Quantity

"This leak is 1.0×10⁻⁶ Pa·m³/s" — metrologically, that sentence is incomplete. It lacks uncertainty: by how much the value might deviate, with what confidence. The correct expression is "Q = 1.0×10⁻⁶ Pa·m³/s, U = 10% (k = 2)," meaning: with ~95% confidence, the true value lies within [0.90, 1.10]×10⁻⁶. The difference between the two sentences is the difference between a "number" and a "quantity value." Uncertainty is not a cover for error but a score for measurement quality — the smaller it is, the stronger the measurement system behind the leak.

7.2  The GUM Framework for Uncertainty Evaluation

The internationally accepted method is specified by the GUM (Guide to the Expression of Uncertainty in Measurement, adopted in China as JJF 1059.1—2012[11]), in three steps:

1. Build the measurement model: output Y = f(X₁, X₂, …). A typical model for leak calibration: calibrated rate = standard value × signal ratio × temperature correction × pressure correction × flow-regime correction;
2. Evaluate the standard uncertainty of each input: Type A (statistics of repeated measurements, $u = s/\sqrt{n}$) and Type B (certificates, instrument specifications, assumed distributions — rectangular: $u = a/\sqrt{3}$; normal: $u = a/2$);
3. Combine and expand:

cᵢ — sensitivity coefficient (the factor by which the output changes per unit change of the input). The sensitivity coefficient of the temperature-correction term is the temperature coefficient itself — this is why 0.1 %/°C and 4 %/°C are worlds apart in a budget table.

7.3  An Uncertainty Budget for Leak Calibration (Example)

Take the calibration of a 10⁻⁶-class He calibrated leak by the comparison method; a typical budget is shown in Table 7-1 (values are illustrative; actual evaluation rests with the laboratory):

How to read a budget matters more than how to compute one: find the dominant components first, then decide what to compress. In this example, "upper-level standard value" and "intermediate stability" contribute nearly 90% of the variance — meaning: refining the detector readout is useless; the only two upgrade paths are a better upper-level standard (move up the traceability chain) and a more stable leak (compress the "intermediate stability" term by design). Swap in a permeation tempco of 4 %/°C and the temperature term alone (u = 4×2.5/√3 ≈ 5.8%) lifts u_c to nearly 8% — tempco is squared in uncertainty combination: the engineering echo of §3.5.

7.4  Map of China's Leak-Rate Metrology Specifications

The industry's most confused point: specifications for "leaks" and for "leak detectors" are two separate systems. One table sorts them out:

The strategic meaning of this map: the detector specifications (1986, 1964) both designate calibrated leaks as the measurement standard — the leak is the intersection of the two systems; if leaks are wrong, the calibration of both detector classes collapses. Meanwhile the leak specifications (1833, 1627, Mil. 165) write "channel/permeation," "annual stability" and "uncertainty" into mandatory language — precisely the projection of this book's Chapters 3~5 physics onto the regulatory level.

7.5  Standard Establishment and Traceability on the User Side

To establish a vacuum-helium-leak calibration facility, a metrology institute needs a reference calibrated-leak set covering all orders of magnitude, each piece individually traceable. A typical JJF 1833—2020-compliant configuration (full measured list in §15.3): six leaks covering (3×10⁻¹⁰ ~ 3×10⁻⁵) Pa·m³/s, one per decade; annual stability 0.1% ~ 1.0% (specification requires ≤ 3%); tempco 0.1 %/°C; each piece traceable to national-level institutes such as NIM; recalibration cycle 12 months[3]. Air-tightness and refrigerant media have their corresponding standard-establishing sets (JJF 1627 / JJF 1964 systems).

The enterprise-side logic follows: working leaks follow reference leaks; reference leaks follow the national primary standard. Recommended asset-management actions:

• Line working leaks: per-shift check (Q₀ signal drift monitoring), annual external calibration;
• Reference leaks: fixed purpose, fixed conditions, dedicated custodian, 12-month recalibration;
• Certificate archive: four elements verifiable — rate value, uncertainty, calibration conditions, traceability institute.

7.6  How to Read a Certificate: The Five-Element Checklist

For any calibrated-leak certificate, check the following in order — 30 seconds tells you whether it is usable:

Reading a real extreme-class certificate against these five elements (a 10⁻¹² Pa·m³/s-class He leak, National Primary Vacuum Metrology Station of the defense industry, July 2026): rate value, U_r = 22% (k = 2), outlet into vacuum, helium medium, upper-level standard with validity date — every element present. The full example is Table 15-1 in §15.1.

7.7  Summary

Part II is complete: process (Ch. 4), lifetime (Ch. 5), system (Ch. 6), metrology (Ch. 7) — four pieces forming a complete knowledge system for judging any leak. Part III takes this system to the field: leak-test methodology, quantitative RGA, system-level air-tightness accuracy, the helium-reduction paradigm and selection methodology.

Chapter 8  Overview of Leak-Test Methodology

8.1  One Map: Sensitivity × Cost

Choosing a leak-test method is essentially finding the intersection between "minimum detectable leak rate" and "cost / takt per part":

Three principles for reading the map: first, sensitivity needs margin — the method's detectable floor should sit 1~2 orders of magnitude below the reject threshold (for a 10⁻⁵ threshold, the method should reach 10⁻⁶~10⁻⁷); second, cost jumps in steps — crossing from pressure decay to helium testing lifts equipment and per-part cost by a tier, so "can pressure decay solve this?" is always the first question; third, no method does everything — high-end lines normally combine: pressure decay to screen gross leaks + helium for fine leaks.

8.2  Helium Mass-Spectrometer Leak Testing: Vacuum / Sniffer / Bombing

A helium leak detector uses a small magnetic-sector mass spectrometer to exclusively "count" ions of mass number 4. The reasons for choosing helium as tracer are textbook material themselves: extremely low natural abundance (only 5.2 ppm in air — clean background), small and penetrating molecules, inert and safe. The three basic modes impose very different demands on leaks:

The sniffer method is the most used — and most misused — mode on production lines: readings depend strongly on probe-to-leak distance, scan speed and ambient He background — it is a "relative comparator," not an "absolute meter." That is exactly why a sniffer line's judgment system must be built with calibrated-leak simulation pieces under identical conditions (the Four-Piece Kit of Ch. 10 and the Three-Reference Verification of Ch. 11 both grow from here).

8.3  Differential-Pressure and Pressure-Decay Air-Tightness: ΔP = Q·Δt / V

Air-tightness testing needs no tracer gas: the test volume is charged with air / N₂ to test pressure, stabilized, and the pressure change is monitored. From the definition of leak rate (Eq. 2-1) and the gas law, the pressure drop caused by leakage is:

V — test volume including plumbing (m³); Δt — test time (s); ΔP — pressure drop (Pa). Example: V = 2 L, 10 Pa drop in 10 s → Q = 2×10⁻³ Pa·m³/s = 2×10⁻² mbar·L/s ≈ 1.2 sccm.

Eq. (8-1) is the "Newton's second law" of the air-tightness industry, and it reveals the method's three inherent properties:

1. Sensitivity ∝ time / volume: for the same threshold, a tenfold larger volume means either tenfold waiting time or tenfold worse sensitivity — large-volume parts are the method's natural difficulty;
2. It measures a volume-normalized rate: the same part on a different fixture (changed volume) changes the rate corresponding to the same ΔP;
3. Everything that moves pressure is an interferer: temperature (Eq. 10-1), adiabatic effects, fixture elastic deformation — the entry point to all of Chapter 10.

The differential-pressure configuration (a differential sensor between the part under test and a reference volume) suppresses temperature and charge-pressure fluctuations by "dual-channel common-mode rejection" and is the mainstream machine type; the direct-decay configuration is simpler but more sensitive to supply and environment. JJF 1986—2022 regulates the calibration of differential-pressure air leak testers (test pressure ≤ 700 kPa, leak rate 0~100 mL/min), one of whose measurement standards is the calibrated leak[9].

8.4  Hydrogen, Halogen and Other Methods

Hydrogen testing: uses 5% H₂ + 95% N₂ (non-flammable mixture) as tracer with a semiconductor-sensor sniffer; sensitivity 10⁻⁵ ~ 10⁻⁷ class, far lower cost than helium, no helium-resource dependency — an important alternative on the helium-reduction path; its leak requirements mirror He sniffer testing (H₂/N₂ simulation leaks at same pressure and medium). Halogen / refrigerant testing: uses the refrigerant itself (R134a, R1234yf, etc.) as tracer — the refrigeration industry's regulatory-compliance tool (JJF 1964—2022, detectable 10⁻⁶ ~ 10⁻⁵ Pa·m³/s[10]); since the medium is the product's working fluid, leaks must be calibrated with the real refrigerant and certified on the g/y basis (Eq. 2-2). Water-bubble and ultrasonic: coarse screening above the 10⁻² ~ 10⁻³ class — intuitive and cheap, commonly used as "gross-leak interception" before helium testing.

8.5  RGA Quantification and Micro-Flow Calibration: Two "Instrument-Level" Scenarios (Map First; Ch. 9 Expands)

Two scenarios do not test products directly but are the high-end battlefield of calibrated leaks: first, residual gas analyzer (RGA) quantification — semiconductor and UHV systems use RGAs to monitor process atmospheres and outgassing composition; quantitative calibration needs standard sources of known partial pressure / partial rate: single gases, mixtures (component partial rates, §6.4), isotopes (³He, D₂, ¹⁸O₂, etc.), even liquid-medium vapors (§6.5). Such scenarios are extremely sensitive to valve dead volume and background (the zero-volume valve is a hard threshold, Rule 6-1). Second, mass-flow-controller (MFC) calibration: calibrated leak rates of sccm ~ L/min class serve as flow references, positive-pressure outlet; the soap-film flowmeter method of JJF 1627—2017 is one traceability path[6]. A positioning note is worthwhile here: calibrated leaks and MFCs are complementary, not competing — an MFC is an active control element (regulating flow in real time), while a calibrated leak is a passive reference (a known, stable, traceable flow truth). Use the former to run the process and the latter to calibrate the former — the same division of labor as a variable-frequency drive and a standard resistor.

8.6  Method-Selection Quick Reference

8.7  Summary

The next chapter enters the instrument-level main battlefield, and the chapter of heaviest strategic significance: quantitative RGA calibration for semiconductors — from qualitative to quantitative, equipping China's high-end manufacturing with "eyes that can measure."

Chapter 9  From Qualitative to Quantitative: Semiconductor RGA Calibration

9.1  Why Semiconductors Deserve a Chapter of Their Own

The core processes of advanced nodes — PVD, CVD, ALD, plasma etch, ion implantation, EUV lithography — all run in vacuum. Each node generation tightens the requirement on vacuum-background purity by roughly an order of magnitude:

The meaning of this table: at the deepest level, the competition of advanced manufacturing is a competition of trace-contamination metrology capability. ppm-level contamination can scrap an entire wafer lot, and the only instrument that can simultaneously resolve multiple gas species and their partial pressures in vacuum is the RGA (quadrupole residual gas analyzer). The RGA is the "eye" of high-end equipment; and all the argumentation of this book's first eight chapters condenses into one sentence: eyes need a yardstick, and the yardstick must be traceable. Vacuum acceptance of semiconductor tools, contamination monitoring of process chambers, incoming inspection of high-purity materials, root-cause tracing in failure analysis — these links have long depended on imported calibration means. Making the RGA calibration system self-reliant fills in the lowest-level piece of the "metrology foundation" for high-end manufacturing. That is why this chapter stands alone — and why it is the largest growth market of the calibrated-leak industry over the next decade.

9.2  RGA Quantification Principle and Three Evolution Stages

The quantitative basis of an RGA is the linear relation between ion current and partial pressure:

I_g — ion current of gas g's characteristic peak (A); P_g — partial pressure (mbar); S_g — sensitivity factor (A/mbar), jointly determined by electron energy, ion-source geometry, quadrupole transmission function and detector gain, and drifting with filament aging and contamination.

Eq. (9-1) is structurally identical to Eq. (8-1) of air-tightness testing: instruments output "signals"; between signal and "quantity value" stands a coefficient that must be calibrated periodically. The three stages of RGA application are essentially the history of calibrating that coefficient:

9.3  The Four Calibration Methods

First, the sensitivity-factor method. Connect a pure-gas calibrated leak of known rate to the system, record the ion current at steady flow, and solve directly for $S_g = I_g / P_g$. Two engineering points: the zero-volume valve lets calibration run in-situ on production tools without shocking the vacuum (Ch. 6); annual decay below the 0.2% class makes calibration results portable and reusable.

Second, the calibrated-leak comparison method. For target sources that cannot be measured directly, compare signals against a known leak at the same pumping speed:

For extremely low rates (< 10⁻¹¹ Pa·m³/s), use the static accumulation method: isolate the pumps and measure the rate of pressure rise in volume $V$:

Note that Eq. (9-3) is the same formula as Eq. (8-1) of air-tightness testing — vacuum accumulation and positive-pressure decay are physical mirror images: the recurring theme of this book that "leak rate has only one physics."

Third, fragmentation-pattern analysis. In mixtures, every mass peak is a linear superposition of multiple components:

α_Mg — fragmentation factor of gas g at mass number M. The main peaks of CO / N₂ / C₂H₄ overlap completely at m/z 28; separation must rely on fragment differences (CO at m/z 12, C₂H₄ at m/z 27) — and calibrating the fragmentation-factor database requires mixture calibrated leaks of precisely known composition as input.

Fourth, the liquid-medium quantification breakthrough. Water vapor, photoresist solvents (DMC), pump-oil tracers (PFTBA) and n-alkanes (organic-contamination fingerprints) have long been the RGA's "organic-vapor blind zone": vapor pressure varies exponentially with temperature, gas-phase standards are lacking, and the high-m/z region has no response reference. Microchannel controlled-release liquid leaks (liquid–vapor equilibrium + microchannel flow-resistance release), combined with known temperature coefficients (9~12 %/°C, §6.5), turn the blind zone into a quantifiable one (±10% class) — PFTBA fragments spanning m/z 69~671 are the recognized gold standard for high-mass calibration[13].

9.4  Trace Calibration: Two-Stage Dilution and the Natural Mass Ruler

Advanced processes need the ability to verify "ppb-level impurities in the process background." The two-stage dilution architecture of ppm-class mixture leaks provides the answer: Kr/Xe at 1 ppm in the source gas (first-stage dilution), with the ratio of total leak rate to system pumping speed further suppressing the partial pressure (second stage):

Q_total — total leak rate; C_Kr — Kr mole fraction in the source; S_eff — effective pumping speed. What the RGA "sees" of Kr is set by the ratio of Kr partial rate to pumping speed.

Note: "relative level" is the Kr partial pressure as a fraction of a typical 10⁻⁷ mbar system background (e.g. 10⁻¹¹/10⁻⁷ = 100 ppm).

One table spanning six orders of magnitude — the "input end" of trace verification thus becomes traceable. The other half of the problem is mass discrimination: quadrupole transmission efficiency varies with mass number, and the response at m/z 132 can differ by orders of magnitude from m/z 28. The natural multi-isotope abundances of Kr (78~86 amu) and Xe (124~136 amu) are "mass rulers written into physical constants": comparing measured isotope ratios against theoretical values yields the transmission correction factor at each mass, building a cross-mass sensitivity response curve (Fig. 9-1).

9.5  Full-Chain Scenario Map and Coverage

Calibration coverage assessed by stage: tool vacuum acceptance > 95%, PVD > 95%, lithography / clean verification > 90%, packaging & test > 95%, material incoming > 90%; CVD / ALD / etch / implantation, which involve reactive specialty gases, about 65% ~ 70% — reasons in the next section[13].

9.6  Technical Boundaries: An Honest List and the Complementary Architecture

A responsible white paper must state its boundaries. Reactive / corrosive / highly toxic specialty gases — SiH₄, AsH₃, PH₃, WF₆, TMA, Cl₂, HBr, NF₃, F₂ — cannot currently be made into stable calibrated leaks — the reason lies in gas physics itself (decomposition inside microchannels, lethal safety risks, channel corrosion, strong adsorption), not in process capability; even in the markets with the most advanced calibrated-leak technology, such leaks are equally scarce or nonexistent. "Surrogate-gas calibration" serves trend judgment only within a narrow window (single mass number, no overlapping peaks, no memory effect) and introduces 20% ~ 100% systematic error — tactically usable, strategically undependable.

The industry's mature solution is a cross-sensor complementary architecture: direct measurement of specialty gases goes to FTIR, OES and electrochemical sensors; the RGA returns to its "environmental watchdog" role — process-fingerprint baselines, cross-contamination monitoring, clean verification, and indirect inference via correlated parameters (e.g. for WF₆, watch not m/z 298 but the by-product SiF₄ at m/z 85). In this architecture, the calibrated leak's task is to keep the RGA's long-term response stable — precisely the value zone of low-decay, zero-dead-volume leaks. Best practices for consistency: sensitivity factors recalibrated every 3~6 months, in-situ first, mixtures for cross-checking the high-mass region, liquid media calibrated specifically every 6 months[13].

9.7  Strategic Significance: Completing the Metrology Foundation for High-End Manufacturing

Putting this chapter back in the big picture: self-reliance in semiconductor equipment ultimately comes down to "every tool's vacuum performance being credibly verified at shipment, and every wafer's process environment being credibly monitored." The credibility of verification and monitoring, in turn, ultimately comes down to whether the calibration system is self-reliant and traceable. For a long time, high-end calibration leaks and RGA quantification methods were bundled inside imported supply systems; once the microchannel process made "all media (single gases / mixtures / isotopes / liquids), all leak rates (10⁻³ ~ 10⁻¹⁰ mbar·L/s), in-situ zero-shock calibration" a routine domestic supply, the last mile of upgrading RGA from a "reference tool" to a "metrological instrument" — the calibration link — was opened. As fabs expand and nodes advance (below 2 nm pushing tolerances toward ppt), this market grows not linearly but multiplicatively with "verification stations × calibration frequency." The directions are equally clear: lower concentrations (ppb → ppt mixtures), wider mass range (> 500 amu precursors), materials-science breakthroughs for reactive-gas calibration, and automated periodic calibration integrated with RGA software.

9.8  Summary

The next chapter returns to the main battlefield of production-line application: system-level accuracy of air-tightness testing.

Chapter 10  System-Level Accuracy of Air-Tightness Testing

10.1  Why an "Allowable ΔP" Cannot Be the Criterion

From Eq. (8-1), judging leakage = judging pressure drop. The problem: pressure is not changed by leakage alone. For a sealed volume, the gas law gives the direct effect of temperature on pressure:

Insert typical line conditions: test pressure 20 kPa gauge (~121 kPa absolute), ambient 27 °C (300 K), a 1 °C drift of part or environment within the test cycle:

A 1 °C temperature drift ≈ a 400 Pa pressure drift. And what is a typical decision ΔP in air-tightness testing? Tens of Pa. In other words, temperature noise is an order of magnitude larger than the leak signal — any practice of "set an allowable ΔP and judge OK/NG" is physically equivalent to measuring microns with a thermally expanding ruler. The differential-pressure architecture (common-mode comparison between reference and test part) cancels slowly varying common-mode temperature fields, but not two residuals: the temperature difference between test and reference parts (uneven incoming temperatures, hot parts fresh off the line), and the asymmetry of adiabatic effects (gas heats on charge compression, then loses heat to the walls, pressure settling exponentially with a time constant of seconds — Fig. 10-1).

Conclusion: ΔP-based judgment is essentially "substituting physical assumptions for verification" — assuming uniform temperature, assuming known volume, assuming completed thermal equilibrium. All three assumptions hold in the laboratory and are doubtful on the line. The way out is not a pricier sensor but making "judgment capability" a daily-verifiable system: the Golden-Part Four-Piece Kit.

The same cognitive traps exist on the helium side, more insidiously. First, what a leak detector displays is not a leak rate but an ion current — filament aging can make the same leak rate produce signals differing severalfold; the reading only acquires physical meaning through a calibration factor (CF), and an uncalibrated reading is a number without dimensions. Second, single-point calibration cannot cover the full range — ion-source contamination drifts the capture efficiency differently at each magnitude, and system conductance changes with flow regime (viscous for large leaks, molecular for small ones, Ch. 3), so extrapolated "linearity" across decades is a dangerous assumption. Third, a 100% He leak cannot be rescaled to a low-concentration condition — partial pressure is not total pressure, and a mixture transports through micro-orifices differently from pure helium (§6.4); low-concentration testing must be verified with same-concentration simulation leaks under real conditions. All three point to one rule: calibrate near the threshold, verify under real working conditions — the Three-Reference Verification of the next chapter is exactly this rule, systematized.

10.2  From Instrument Accuracy to System Accuracy: A Shift of View

The datasheet's "accuracy 0.1% FS" describes a sensor; what the line needs is judgment. The gap between the two is opened by four links: tooling, environment, part condition, operation. System-level thinking splits leak judgment into three layers, each verified with a different standard piece:

• Machine level: is the tester itself still accurate? — verified daily with an external / built-in calibrated leak (Q₀ check piece);
• Fixture level: tester + fixture + plumbing + program parameters — can the assembled system still resolve a borderline leak? — verified with OK / NG simulation dummy parts;
• System level: is judgment stable across continuous production? — verified by periodic re-testing of boundary pieces and GR&R statistics.

One common counter-example of the three-layer architecture must be named: a line that relies solely on the tester's "self-check" is essentially a black box. The self-check calibrates the instrument against a built-in leak, yet the built-in leak's own drift is verified by no one — using an unverified ruler to verify the ruler is a circular argument; and a built-in leak usually provides a single rate point, so single-point fitting extrapolated to other magnitudes is unreliable (cognitive trap no. 2 at the end of §10.1). The credible practice is a dual evidence chain: a reservoir-type leak verifies the instrument itself (machine level — "is the gauge right"), while passive simulation leaks verify the system under real pressure, real medium and real takt (fixture / system level — "is the job right"). If the former fails, the instrument is flying blind; if the latter is missing, the whole line is judging blind. Neither is dispensable.

10.3  The Golden-Part Four-Piece Kit: Building the Judgment System into the Fixture

The Four-Piece Kit is the minimal complete configuration for system-level verification[7]:

The logical loop of the kit: the OK piece guards the false-reject line, the NG piece guards the missed-reject line, Q₀ guards instrument drift, and the golden part provides a trustworthy carrier. Run all four pieces at the start of each day, and the "today's health" of the judgment system is immediately known — turning the physical dilemma of §10.1 from "assumptions" into "daily measurement." Two engineering details: first, the OK/NG dummy leaks should be built into or side-connected inside the golden part, so that leak path, volume and temperature response match the real part (five identical conditions of simulation: same volume, pressure, medium, temperature, takt[7]); second, the stability of the leaks themselves is the premise of the kit's credibility — if a dummy's leak drifts with temperature (permeation +4 %/°C) or decays off, the referee's ruler changes by itself; this is why the kit prefers channel-type low-tempco leaks (Table 3-2).

Terminology note: the Q₀ + OK + NG pieces form the core verification logic; the golden part is their realistic carrier in pressure-decay scenarios (making leak path, volume and temperature response identical to real parts). In tracer-gas (helium) testing, the same verification system is called the Three-Reference Verification — with the same decision criteria: GR&R ≤ 15% + three-tier separation (Ch. 11). OK / NG dummy rates are (1/3~1/2) Q₀ and (2~3) Q₀ throughout this book.

10.4  The Connection-Method Accuracy Ladder and the Bypass SOP

How a calibrated leak is connected to the test system directly decides the credibility of the verification value. Four connection methods form a ladder from low to high fidelity:

Bypass-method SOP (system-level accuracy verification and equivalent-volume calibration):

1. Clamp a real part; charge and stabilize per the normal program;
2. After entering the test segment, open the bypass valve to inject the calibrated leak of known rate Q₀ in parallel;
3. Read the incremental pressure change ΔP₀, and solve for the system's equivalent volume from Eq. (8-1): V = Q₀·Δt / ΔP₀;
4. Close the bypass valve to restore normal operation. This V contains all real plumbing, fixture and part — far better than any drawing estimate;
5. Repeat periodically: drift of V is an early signal of system state changes.

10.5  Closing with GR&R: Statistical Proof of the Judgment System

Once the system is built, prove it metrologically: take samples near the boundary (OK dummy, NG dummy and borderline parts), run repeated measurements by multiple operators, and perform a GR&R (gage repeatability & reproducibility) analysis; for attribute judgment (OK/NG), run an attribute agreement analysis (Kappa). Common industry thresholds: variable GR&R ≤ 10% acceptable, 10~30% risk-dependent, > 30% unusable; for OK/NG judgment systems, the more direct engineering criterion is GR&R ≤ 15% + three-tier separation (OK, Q₀, NG readings clearly tiered without overlap, §11.3). A GR&R report is also the language most accepted by OEMs and auditors — the Four-Piece Kit is the daily thermometer; GR&R is the periodic health check.

10.6  Summary

This chapter solved system accuracy for "air testing"; the next chapter solves the cost dilemma of "helium testing": in the era of helium at 300+ RMB/m³, how to cut helium usage by one to two orders of magnitude without losing judgment capability.

Chapter 11  The Helium-Reduction Paradigm: Equivalent-Orifice Relative Comparison and Three-Reference Verification

11.1  The Helium Crisis and Three Mindsets

In March 2026, escalating geopolitical conflict in the Middle East severely damaged facilities at Ras Laffan Industrial City in Qatar — the world's second-largest helium production base — with repairs expected to take 3~5 years; together with the United States, Qatar forms the two pillars of global helium supply (30%~39% of the world total). Chinese landed prices for industrial-grade helium soared from 50~70 RMB/m³ in 2025 to 300~400 RMB/m³ in April 2026, a 5~10-fold increase. China's helium consumption relies on imports for about 90%, of which 54%~66% comes from Qatar — even those willing to pay the premium may not secure a stable quota[12]. "Reducing helium consumption per test" has upgraded from a cost-optimization option to an industrial-survival option.

Under the crisis, the industry shows three mindsets:

The key cognitive shift: from "concentration determines reliability" to "verification determines reliability." "80% He is less reliable than 100% He" is an untested assumption — the correct comparison is "80% He + three-reference verification = reliability confirmed" versus "100% He + no verification = reliability unknown." Concentration anxiety is unsolvable (helium prices are beyond your control); verification anxiety is solvable (three references, a few minutes per shift).

11.2  The Physics First Principle: The Four-Layer Framework of Equivalent-Orifice Comparison

All air-tightness and leak testing does exactly one thing: compare the size of the equivalent leak channel on the workpiece (§1.4). Split this judgment into four layers and the industry's misconceptions become visible at a glance:

Whether you charge 100% helium or 1% helium, the physical orifice on the workpiece does not change by a single micron — leak rate Q is a function of gas species, pressure difference, temperature and flow regime (Eq. 3-1 ~ 3-4), not an intrinsic property of the leak channel; switching concentration only changes the mixture's mean molar mass and effective partial pressure. The industry's core misconception is mistaking the fourth layer (the reading) for the first layer (physical reality), and mistaking one option in the third layer (100% He) for the only option.

A long-overlooked fact: air leak testers proved long ago that non-helium gases can reliably compare orifices. A differential-pressure air tester uses compressed air (containing only ~5 ppm helium) as medium, has no mass spectrometer, and still judges Pass/Fail stably on production lines (Ch. 10). From air testing → low-concentration helium testing → pure-helium testing is a continuous spectrum, not a black-and-white boundary — if testing's essence were "absolute accuracy of the instrument reading," air testers could not work at all; they do work, which means the essence is "sufficient discrimination." The ultimate output of line testing is a binary Pass/Fail; the core need is judgment consistency, not absolute accuracy.

Under ideal conditions, this judgment can be written as clean mathematics. The helium-test signal (sniffer or vacuum method) for a leak Q is:

S — instrument signal (rate reading); k — instrument sensitivity coefficient (transmission, ionization, amplification); c — helium concentration in the tracer gas; Q — total leak rate of the workpiece's leak point.

Measure the workpiece and the calibrated leak under the same concentration of tracer gas (same source), and take the signal ratio:

In the ratio $R$, $k$ and $c$ cancel approximately — this is the ideal model of judgment. But note: it is only a zero-order approximation; the deviation sources and engineering countermeasures follow in §11.3.

11.3  From Mathematical Cancellation to Engineering Comparison: Three-Reference Verification

Early versions of the helium-reduction argument (including this book up to V1.2) stated the cancellation in Eq. (11-2) as "exact." Deep engineering practice has shown this statement over-idealized; it needs correction. Real systems exhibit three classes of deviation:

• Flow-regime nonlinearity: when He concentration drops from 100% to 1%, the mixture's mean molar mass changes from 4 to about 28. In molecular flow (Kn ≫ 1) conductance scales as 1/√M and the signal scales roughly linearly with concentration; but in transitional flow (0.01 < Kn < 1) the viscous component is not negligible, and mixture viscosity follows complex nonlinear mixing rules — the signal may fall by 80× or 120× instead of exactly 100×, and workpieces of different orifice sizes may even behave inconsistently as they cross regimes;
• System non-idealities: filament aging, electronics temperature drift and vacuum-background fluctuation make $k$ not strictly constant;
• Concentration non-constancy: dynamic consumption from the mixing tank and line adsorption/desorption make $c$ not strictly constant.

The methodology must therefore complete three turns: from "mathematical cancellation" to "engineering comparison" — no longer pursuing a proof that the concentration term cancels exactly; three-reference measured verification replaces mathematical cancellation as the basis of judgment. From "single-piece Q₀ calibration" to "Three-Reference Verification" — Q₀ + OK + NG verified together, ensuring the system works reliably in its actual state of the day. From "theoretically determined concentration" to "iterating to the MWC" — not presuming 1% must work; use the stepwise concentration-reduction method to find each scenario's optimal concentration by measurement (§11.4). This correction does not weaken the scheme's reliability — it strengthens its honesty and operability.

Three-Reference Verification (Q₀ + OK + NG) is the gold standard of leak testing — whether at 100% He or 1% He, whether or not helium is reduced:

Run all three references at the start of each day / shift — this is the correct way to open a leak-testing line. Only two conditions decide system usability: ① system GR&R ≤ 15% (the common measurement-system-analysis standard); ② three-tier separation (OK, Q₀, NG readings clearly tiered, no overlap). When both hold, the system works — whether on 100% He, 1% He, or air. The discrimination ratio (measured reading ratio) is a natural consequence of the two conditions rather than an independent constraint: with the conventional choices OK = (1/3~1/2)Q₀ and NG = (2~3)Q₀, even at SNR as low as 3 the discrimination ratio typically stays between 2~5; GR&R ≤ 15% keeps within-tier fluctuation far below the inter-tier gap, leaving no "gray zone" between tiers (Fig. 11-1).

This is the dividend of the paradigm shift: the traditional "absolute metrology" paradigm demands SNR ≥ 10 (the rate value must be known precisely); the "size comparison" paradigm needs only SNR ≥ 3 ~ 5, provided GR&R ≤ 15% — from "precise measurement" to "sufficient discrimination," the threshold drops dramatically.

There is also a frequently inverted positioning of Three-Reference Verification: it is not an accessory of helium reduction — it is the best practice even under pure helium. 100% He solves only the single dimension of "signal strength," while test reliability is a system engineering of instrument, environment, operation and standards — instrument sensitivity drift (filament aging, electronics drift — concentration-independent), environmental background spikes after gross leaks (pure He makes gross leaks dump more helium and pollute worse), calibrated-leak self-drift (permeation leaks decay 1%~5%/year — concentration-independent), fixture and operation differences, vacuum-system micro-leaks: none of these five problem classes disappears automatically under pure helium. Thought experiment: two lines both on 100% He — line A, which never verifies, may "fly blind" for months after 30% filament aging, shipping batches of missed rejects unnoticed; line B, verifying daily, would spot an 18% drift of the Q₀ reading on day three and act immediately. The problem has never been concentration; it is whether you verify. Helium reduction is an extension of Three-Reference Verification — since you should verify daily anyway, finding the lowest workable concentration along the way is pure cost saving: what helium reduction cuts is cost, not quality.

11.4  Stepwise Concentration Reduction: Finding the Minimum Working Concentration (MWC)

The Minimum Working Concentration (MWC) is not computed — it is measured, through three-reference verification. The procedure: first establish baseline readings of the three references at 100% He and record the discrimination ratio (Step 0); then step down through 50% → 20% → 10% → 5% → 2% → 1%, re-running the full three-reference verification at each level — if the three tiers stay clearly separated, keep descending; if tiers overlap or GR&R > 15%, stop, and MWC = the previous level. Every part–equipment combination finds its own optimum this way — the methodological upgrade from "guessing 1%" to "scientifically finding the optimum." Two execution details: first, the iteration must run at the target takt — verify with the same integration time as the line's actual takt (minimum detectable rate depends on integration time; a 3~5 s/part ultra-fast line has different margin than a tens-of-seconds conventional line, though even the fastest lines retain large margin in industrial helium testing); second, record environmental parameters such as temperature and humidity alongside.

Reference starting concentrations by threshold magnitude (iteration starting points, not conclusions):

Low concentration carries a counter-intuitive dividend: the active noise-reduction effect. On pure-He lines, test exhaust (about 60%~70% of total emissions), joint and seal micro-leaks, part-surface desorption and calibration exhaust can push the workshop's local helium background from the natural 5 ppm to hundreds or even thousands of ppm; switching to a 1% mixture cuts the absolute helium emitted per test by ~100×, total workshop He emissions by ~99%, and the background returns to natural levels. The "memory effect" after gross-leak parts (helium adsorbed on mass-spectrometer chamber walls releasing slowly, conventionally requiring minutes to tens of minutes of clean-up waiting) shrinks to seconds under the mixture scheme, since the absolute helium flushed in drops by two orders of magnitude — no shutdown needed. In SNR = signal / noise: the numerator (signal) falls with concentration, but the denominator (ambient He interference) falls even more — at sites with severe ambient helium pollution, the effective SNR may actually rise rather than fall. This is the scheme's core insight: "counter-intuitive" yet "physically rigorous."

A simple flow-regime guide: molecular flow (Kn ≫ 1) — signal-vs-concentration is approximately linear, but three-reference verification is still mandatory; transitional flow (0.01 < Kn < 1) — possibly nonlinear, verification absolutely required; viscous flow (Kn < 0.01) — significantly nonlinear, helium reduction not recommended. All flow-regime effects are absorbed by three-reference comparison — every step of concentration iteration trusts measurement, not theory.

11.5  The Cost Ledger: Starting from 300 RMB/m³

Helium cost ≈ usage × concentration (at fixed recovery rate, Fig. 11-2):

A concrete ledger (battery-pack helium-testing line, 5~6 million parts/year, industry operating-data caliber[12]): the pure-He scheme consumes ~300 cylinders/year (50 L × 20 MPa, ~3,000 RMB/cylinder), ~900,000 RMB/year of helium; at 10% concentration, consumption falls to ~30 cylinders, ~90,000 RMB — annual savings of about 800,000 RMB — while the one-time investment for Three-Reference Verification is only 10,000~30,000 RMB (Q₀ calibrated leak + OK / NG dummy fabrication), and mixing equipment costs far less than recovery systems. The hidden gains are equally substantial: daily three-reference verification eliminates batch misjudgment from systematic drift (false-alarm rates can drop by 50%+); post-gross-leak clearing time falls from minutes to seconds, lifting effective takt by 3%~8%; per-shift three-reference calibration records give quality audits evidence (satisfying IATF 16949 and similar systems); and low consumption is itself a supply-chain risk hedge.

The ledger has a second dimension — the total cost of ownership (TCO) of the standard itself. The hidden expenses of conventional permeation / glass-process leaks are spread over three items: transport losses and accidental scrap from fragility, frequent recalibration trips with line-down waiting, and batch misjudgment caused by drift. Leaks with low decay (0.1% ~ 2.5%/year, §16.3), proven drop/freeze performance (§15.2 measurements), and wide-temperature, no-special-storage operation stretch the recalibration cycle reliably to 12 months (the §15.3 apparatus example) and remove "handle with kid gloves" from daily work. These savings are not denominated in helium, but they reach the income statement just the same. At selection time, replacing "unit price" with "TCO = purchase price + recalibration fees × years + scrap rate × replacement cost + misjudgment risk cost" often reverses the conclusion.

Helium reduction is not a point technology but a three-factor product of "concentration × recovery × judgment method" — and the judgment-method factor (Three-Reference Verification) happens to be the cheapest and purest cognitive lever of the three. The bigger picture: when industrial helium testing cuts helium usage to 10% of before, the same import volume supports more than 10× the testing demand, domestic limited capacity becomes "enough," and the helium crisis is dissolved on the demand side — a "demand-side revolution" that also gives China the chance to become the definer of low-concentration helium-testing standards[12].

11.6  Applicability Boundaries and Honest Disclosure

An honest white paper must state its boundaries clearly. Applicability by test method:

• Dual-track system: industries that mandate absolute-rate reporting (aerospace GJB, medical devices, etc.): use low-concentration Three-Reference Verification for daily production, and retain pure-He capability for type tests or customer acceptance reports;
• Magnitude boundaries: 10⁻⁴ ~ 10⁻⁵ mbar·L/s (vacuum / accumulation methods) — mature and reliable; 10⁻⁶ class — feasible after careful verification; 10⁻⁷ and below — outside the scope claimed by this scheme;
• Temperature effects: low-tempco leaks (0.1 %/°C class) are negligible in general industrial environments (±2 °C), but the workpiece's own rate may vary with temperature (polymer seal thermal expansion, etc.); environments swinging > 10 °C warrant a temperature-sensitivity test;
• Concentration-fluctuation control: size the mixing tank > 200× single-cycle consumption, install online concentration monitoring, zone the supply for multi-station lines, qualify 2+ gas-mixture suppliers;
• Response time: signal buildup lengthens at low concentration — leave stabilization margin in takt design;
• Three hard specs for simulation leaks: rate and specified pressure customized together (threshold = target), low tempco (no climate control on lines), shock and drop resistance (line environment); and the leak must match the workpiece in concentration, pressure and medium — mixture simulation leaks are customized to the site's actual concentration and certified on the total-rate basis (§6.4 P/T basis); a "fake same-source" pair (pure-He leak vs mixture-charged workpiece) voids the ratio. Methodology ultimately lands back on device physics (Ch. 3~5).

11.7  Summary

The final chapter of Part III returns to the most everyday question: given a requirement, how to choose the right leak step by step — and the pitfalls we have stepped into over the years.

Chapter 12  Selection Methodology and Common Pitfalls

12.1  The Five-Step Selection Method

Selection is essentially fixing the "four working-condition elements + form" of Chapter 2. Converge in the following five steps, each mapping to earlier chapters:

After the five steps, what you get is not a model number but a line of "working-condition language" — e.g. "N₂, 2.0×10⁻⁴ Pa·m³/s, inlet 6 bar absolute, outlet atmosphere, passive, VCR 1/4." Any serious leak manufacturer should quote price and lead time from that line. A requirement that cannot be stated this way will repay, with interest at acceptance time, every cent saved on the quotation.

12.2  Reservoir or Passive: A Decision Table

12.3  Decoding the Ordering Logic: Five Promises in One Code

A proper ordering code should decode segment by segment (illustrative structure): medium – leak rate – inlet pressure – fitting – certificate level. At acceptance, check each segment against the certificate's five elements (Table 7-3): if the code says "6 bar absolute" while the certificate says "6 bar gauge," the one-atmosphere gap is an order-of-magnitude error on the low-pressure side (§2.4). Code, nameplate and certificate telling one consistent story is the first formality of incoming acceptance.

12.4  Customization Flow and Lead-Time Expectations

One meaning of the microchannel process is "customization as the norm": specified pressure + specified rate + specified medium, with geometry computed first by conductance simulation (Eq. 3-4), then machined, then actually calibrated and certified. Reasonable expectations: routine customization delivers in ~3~5 working days; extreme specs (below 10⁻¹², 40 MPa, special media) need technical review. The touchstone for whether a manufacturer truly owns the process: can they accept "pressure and rate customized together" — instead of only letting you pick from a shelf list.

12.5  Top-Fifteen FAQ

1. How long does a leak last? Passive channel-type: no natural lifetime limit (periodic recalibration suffices). Reservoir: watch Q/(pV) (§3.8); high-pressure large-volume products can hold annual decay to 0.1%~2%.
2. Can an inaccurate leak be repaired? For microchannel leaks, most cases are clogging or contamination — clean, unclog and recalibrate; geometric damage is irreparable. Glass pieces are essentially irreparable.
3. Dropped it once — still usable? Glass / quartz pieces: write off by default. For drop-verified monolithic metal leaks (e.g. 1.2 m × 5 cycles), simply re-measure the rate to confirm.
4. Certificate expired — still usable? The value loses traceability: usable only as a "reference piece," not a "standard"; recalibration restores status.
5. Why don't the new leak's readings match the old one's? Check the five elements first (Table 7-3): medium, pressure basis, outlet condition, temperature, uncertainty — eight times out of ten it is a basis mismatch, not a broken leak.
6. Are lower workshop readings in winter normal? Check the tempco sign: viscous-dominated air-tightness leaks gain rate when colder; molecular-flow-dominated helium leaks lose rate (Table 3-2), ~0.1~0.6 %/°C in magnitude — correct first, suspect the instrument later.
7. Why does a permeation leak need a temperature-correction table? The +3~4 %/°C Arrhenius mechanism (§3.7): without correction you carry a built-in drift of a dozen-plus percent per month.
8. Can a calibrated leak serve as a product leak point? Not in reverse: the leak is the "standard source," the workpiece leak point is the "object under test"; simulate the critical defect with OK/NG dummies (§10.3).
9. How bad is mixing up sccm and mbar·L/s? A factor of 59.2 (Eq. 2-3). Write units in full when specifying; convert first at acceptance.
10. What if someone forgets to close a reservoir leak's valve? Continuous discharge shortens life — choose normally-closed valves or build check habits (§6.3).
11. Are hydrogen calibrated leaks dangerous? At micro rates they release ~0.3 mL/day, far below any safety-concern limit (§2.3 magnitude intuition); ordinary ventilation suffices.
12. Why are field readings at 5 MPa wrong with a 10 MPa certificate? In viscous flow Q ∝ p² (Eq. 3-2): halving pressure quarters the rate. Order for the use pressure; never redeploy across pressures.
13. Which "rate" does a mixture leak's rate mean? Always ask: P basis (a component's partial rate) or T basis (total rate) — they differ by √M weighting (§6.4).
14. What should a metrology institute buy to establish its standard? A reference-leak set: covering 5~6 orders of magnitude, each piece traceable to national level, annual stability and uncertainty meeting JJF 1833—2020 (§7.5).
15. How to quickly judge a leak manufacturer's real level? Four questions: What is the tempco (and can they explain it by flow regime)? Are bidirectional rates equal (Rule 4-1)? Can they customize pressure and rate together? Do they dare show drop / freeze–thaw re-test data?

12.6  Summary

Part III is complete. The reader now owns the full equipment from physics to methodology. Part IV shows what a product matrix and technology system look like when one factory fully masters and engineers this knowledge system.

Chapter 13  Four Core Technology Platforms: Answers to the Physics of the Previous Twelve Chapters

13.1  The Logic of Answering: A Problem List

13.2  RMI-MTC™ Microchannel Capillary Process: From Craft to Design

Geometrically defined microchannels are machined directly into a metal body, with orifice continuously controllable from 50 nm to 200 μm — computed first from the conductance formulas (Eq. 3-2, 3-3, 3-4), then machined[3]. Against the four yardsticks of Chapter 4:

• Orifice as design value: pressure and rate customized together (e.g. "2.3×10⁻⁷ Pa·m³/s at 34 bar gauge"); tolerance from ±10%, available at ±5% or tighter — the ±2.4% orifice tolerance of Eq. (4-1) is routine for CNC micro-machining;
• Ideal circular tube: non-directional rate (direct evidence for Rule 4-1); uniform bore without a throat, hence clog-resistant (§5.6); all-regime behavior fully described by Eq. (3-4);
• Mechanical strength: monolithic metal structure — passive leaks rated 40 MPa, enduring repeated 400-bar pressure shocks, passing a 1.2 m × 5-cycle drop test, operating temperature −20 ~ +65 °C (withstanding −30 °C freezing)[3];
• Tempco delivers theory: reservoir (molecular-flow dominated) +0.1 %/°C; passive (viscous-flow dominated) −0.6 %/°C — exactly the magnitudes predicted by Eqs. (3-5)/(3-6), an order of magnitude below permeation membranes (Table 3-2).

13.3  RMI-Metal™ All-Metal Sealing: Pressing the Background Down to 10⁻¹⁵

Seal background < 1×10⁻¹⁵ atm·cc/s — 5~6 orders of magnitude ahead of conventional sealing processes (10⁻⁹ class)[3]. Re-read this number through §5.2's 10:1 rule: it means background is no longer a constraint on any foreseeable rate spec — the lower limit of channel-type leaks advances to the 10⁻¹² atm·cc/s (10⁻¹³ Pa·m³/s) class. Meanwhile, the adhesive-free, polymer-free, integrated structure eliminates adsorption/outgassing paths (§5.3) — the entry condition for semiconductor RGA, UHV and isotope scenarios.

13.4  RMI-PSOZV™ / MDZV™ Zero-Volume Valves: 0.098 μL of System Engineering

Dead space Φ0.5 mm × 0.5 mm = 0.098 μL, rewriting all three ledgers of Chapter 6[3]:

• Cycle loss: conventional valve 0.1 cc × 100 atm = 10 cc/cycle; zero-volume valve ~0.01 cc/cycle — about 1 cc lost per 100 cycles, negligible (Eq. 6-1, Fig. 6-1);
• Vacuum shock: no matter how long it has been closed, the instantaneous release at opening does not shock HV / RGA systems — in-situ calibration on production tools becomes possible (§9.3);
• Minimized decay: gas is lost only while open; combined with the normally-closed pneumatic design (PSOZV™), the largest human risk — "forgot to close" — is actively eliminated; for expensive isotopes (Xe, Kr, ³He), near-zero natural decay is achieved. Measured examples: MDZV™ H₂ leak (160 cc, He 1.60×10⁻⁶ Pa·m³/s @ 10.8 bar, multi-pressure calibration points), annual decay < 2.5% (at 60 uses × 60 min per year); PSOZV™ mixture leak (Xe-132 / Kr-84 isotope set), annual decay < 1%[3].

13.5  RMI-Cal™ Digital Conductance Engine: From "Trial and Error" to "Computed First"

RMI-Cal™ is a gas-properties library plus a conductance-simulation algorithm (the engineering implementation of the superposition model of Eq. 3-4): inputs are medium (pure gas / mixture / liquid), inlet pressure and target rate; outputs are orifice and length machining parameters; mixture partial rates are computed per Eq. (6-3) with automatic P/T basis conversion; liquid media follow the vapor-pressure model (Eq. 6-5). It turns customization from "repeated crushing trials" into "simulate first, machine second, then actually calibrate," compressing delivery from traditional weeks to 3~5 working days (CNAS certificate +1 week)[3]. To the user, its most tangible form is the ordering-code matrix — medium, rate, pressure, fitting, valve, certificate encoded segment by segment (§16.1): one line of code is a complete working-condition contract. It should be said that RMI-Cal™'s theoretical foundation is no mystery — it is precisely the open physics of Chapter 3: the flow-regime criterion (Kn), the viscous / molecular-flow formulas and the 1/√M conversion form the computational core, and the gas-properties library supplies viscosity and molecular-weight data. Its value lies in turning this physics into executable, reviewable, batch-capable engineering computation, not in any black-box secret.

13.6  The Closed Loop: How the Four Technologies Enable Each Other

Individually, each technology solves one class of physical problem; together they form a closed loop: RMI-Cal™ makes it "computed right" (design layer) → RMI-MTC™ makes it "built right" (manufacturing layer) → RMI-Metal™ makes it "sealed tight" (background layer) → RMI-PSOZV™/MDZV™ makes it "lasting in use" (system layer). Missing any link, the extreme specs collapse: without metal sealing, microchannels cannot reach 10⁻¹³; without zero-volume valves, the high-pressure reservoir's decay advantage drains away in use; without the simulation engine, all-media customization falls back to trial and error. This is the structural reason "all leak rates × all pressures × all media" coverage can hold simultaneously.

Chapter 14  The Full-Scenario Product Matrix

14.1  Matrix Overview: Four Dimensions of Full Coverage

The physical significance of full coverage: the three axes of the figure correspond to Chapter 3's three variables (rate ← flow regime, pressure ← driving force, medium ← molecular properties). "Full coverage" means there are no physical blind zones in the product space — every scenario from Ch. 8 to Ch. 12 has a landing point in the matrix.

14.2  Extreme-Spec Products: Answers to the Physical Frontiers

10⁻¹¹ ~ 10⁻¹³ Pa·m³/s microchannel He calibrated leaks (the product form of Chapter 5's physics): what pressed the channel-type lower limit down by three orders of magnitude was precisely the all-metal sealing background (< 10⁻¹⁵ atm·cc/s, §13.3). The value of this spec: reference calibration of the most sensitive helium detectors, bombing-method hermetic-package testing, metrology institute standard establishment[5]. 40 MPa passive simulation leaks: hydrogen-energy and high-pressure vessel scenarios — pressure no longer an excuse for judgment failure. Mixture / isotope leaks: 1 ppm Kr/Xe balance N₂ (two-stage dilution source, §9.4); multi-component H₂/He/N₂/Kr/Xe balance Ar (mass-discrimination correction, §9.4); ³He, D₂, ¹⁸O₂ isotopes (scientific and nuclear applications); all certified on explicit P/T bases (§6.4). Liquid-medium leaks: n-dodecane, DMC, PFTBA (§6.5, §9.3), filling the RGA high-m/z calibration gap.

14.3  Coverage Answers by Scenario

14.4  Summary

Products are what specifications say; evidence is what data shows. The next chapter: field data.

Chapter 15  Field Data: Turning Specifications into Evidence

15.1  Metrological Evidence

• Participation in drafting JJG (Mil.) 165—2024, Verification Regulation for Vacuum Calibrated Leaks (effective 2025-01-01)[2];
• Reference leaks procured by multiple metrology institutes/stations as master standards for establishing their own standards: He calibrated leaks of 10⁻¹⁰, 10⁻¹¹, 10⁻¹² mbar·L/s classes and H₂ calibrated leaks, with calibration certificates from the National Primary Vacuum Metrology Station of the defense industry — the deepest endorsement of a leak manufacturer is to have its products used as someone else's yardstick[3];
• Long-term stability data: tracking batches of reference leaks over multiple recalibration cycles, with year-to-year changes within the certificates' stated uncertainty (data continuously archived, available for review on-site).

A concrete certificate from July 2026[16]:

This certificate is also Chapter 7's uncertainty language instantiated at the extreme class: the 22% relative uncertainty is not process sloppiness — 10⁻¹² Pa·m³/s already approaches the boundary of the national metrology system, whose own upper-level apparatus carries U_r = 14% ~ 18%, and the calibrated value cannot beat its source. Read the other way: that a certificate with uncertainty and a complete traceability chain can be issued at this magnitude at all means the domestic value-transfer system has closed the loop for 10⁻¹² Pa·m³/s-class microchannel leaks.

A common customer question deserves a note in passing: the same leak receives different uncertainties from different institutes — the value is governed chiefly by the capability of the measuring apparatus (the upper-level-standard component of the Chapter 7 budget). The National Primary Vacuum Metrology Station of the defense industry can assign 3% ~ 5%; the National Institute of Metrology routinely assigns 10%; regional institutes typically 15%. RealMeter's in-house certified values (5.5% at the 10⁻⁵ mbar·L/s class, ±8% over 10⁻⁶ ~ 10⁻⁹ mbar·L/s) sit in between, resting on repeated comparisons against the company's reference-leak sets and on process priors. All of this presupposes a good leak in the first place — short-term instability feeds directly into the repeatability component: with a poor leak, no apparatus, however capable, can assign a small uncertainty.

15.2  Extreme-Condition Re-measurement Data

The quantified form of §5.5 "mechanical and thermal damage" — public demonstration records[14]:

• Passive leak (4.75×10⁻⁵ mbar·L/s): after one night at −20 °C + 1.2 m drop, re-measured retention 101.1% ~ 103.2% (within the certificate uncertainty);
• Reservoir leak (1.16×10⁻⁷ mbar·L/s): after one week at −20 °C + drop test, re-measured 1.10×10⁻⁷ mbar·L/s (retention ~95%); the ~5% change is explainable by tempco (+0.1 %/°C × 40 °C ≈ −4%) under the ~40 °C non-climate-controlled re-test conditions.

The point of such data is not "dropping is fun" but the quantifiable verification logic of Chapter 5: a leak's robustness should not be a usage caution but a design specification with re-measured evidence. A further piece of hardware-endurance evidence is the vacuum-shock measurement — opening a zero-volume-valve leak disturbs an RGA system's pressure by less than one order of magnitude; the data are in §6.2[15].

15.3  Third-Party Reports and Field Cases

Cases across key industries (desensitized):

• Metrology: a provincial institute's reference-leak set (6 pieces covering 3×10⁻¹⁰ ~ 3×10⁻⁵ Pa·m³/s), passing JJF 1833—2020 standard-establishment assessment, in service beyond two recalibration cycles;
• Power batteries: a head battery-maker's helium-reduction retrofit (100% → 10% He), Three-Reference Verification passing GR&R ≤ 15%, annual helium cost reduced by ~90% (Ch. 11 ledger);
• Semiconductors: a PVD tool maker's RGA in-situ calibration scheme (Ar + H₂O + ppm Kr/Xe combination), reducing vacuum-acceptance time per tool by ~40%;
• Refrigeration: a compressor maker's R134a leak replacement of permeation references — temperature-compensation table abolished, winter/summer judgment difference disappearing (Table 3-2 physics delivered).

The metrology case above is now documented in full public detail (a municipal institute of metrology and testing, 2026)[18]:

Common specifications: 100% He; channel-type (microchannel); KF25 outlet; −20 ~ +65 °C; tempco 0.1 %/°C; uncertainty ±10% (k = 2); 12-month recalibration; all six calibrated and traced by the National Institute of Metrology (NIM), China.

Against JJF 1833—2020 §6.2.1 (the apparatus shall carry multiple calibrated leaks covering every decade of its range, with stability better than 3%/year): this set covers 3×10⁻¹⁰ ~ 3×10⁻⁵ Pa·m³/s — six decades, one leak per decade — with annual decay of 0.1% ~ 1.0%, an order of magnitude better than required. The piece-to-piece decay differences are a textbook display of the decay law Q/(pV) of Eq. (3-10): higher-rate models carry larger reservoirs (300 ~ 1000 cc for TLS-5P / 6P; 100 cc for TLS-9P / 10P), keeping even the high-rate end below 1%/year. Establishing a standard is not buying six leaks — it is buying a value echelon that has every decade covered and every piece surviving annual recalibration.

Outgoing-repeatability evidence, also public: a 1.0 mL/min @ 20 kPa air-tightness Air calibrated leak underwent 32 consecutive repeatability runs before shipment, with flow fluctuation within the rated specification; the full test video and data record are openly accessible[17]. Such "N-run consecutive re-tests" should become a routine acceptance action for air-tightness leaks — it maps directly onto Chapter 10's system-level logic: a single-point calibrated value is only the entry ticket; consecutive repeatability is the evidence of production-line credibility.

15.4  Summary

The final chapter: the service system — how to make this knowledge and evidence reproducible for every customer.

Chapter 16  Service and Quality Assurance System

16.1  Customization Service: One Code Is One Contract

The ordering-code matrix (medium – rate – pressure – fitting – valve – certificate) is the engineering interface of the RMI-Cal™ engine (§13.5): the customer states working conditions, simulation outputs parameters, machining builds to drawing, actual calibration issues the certificate — code, nameplate and certificate tell one story (§12.3). Routine customization: 3~5 working days; CNAS / institute certificates: +1 week[3].

16.2  Technical Support: Beyond Selling Devices

• Application-method design: Four-Piece Kit configuration, bypass-method SOP, helium-reduction scheme design (Three-Reference + stepwise concentration iteration, Ch. 10, 11);
• Metrology support: standard-establishing configuration consulting (reference-leak sets), uncertainty-budget templates (§7.3), recalibration management;
• RGA quantification support: calibration-scheme design, mixture / liquid selection, sensitivity-factor workflows (Ch. 9);
• Knowledge output: this book, plus the RealMeter Leak Research Institute's public technical content (white papers, cases, videos[7][14]).

16.3  Quality Commitments and After-Sales

• Factory test certificate with every unit; CNAS / institute calibration certificates on request;
• Clogging / contamination repair + recalibration service (Chapter 5's "maintainable design" delivered);
• Reservoir-leak decay tracking and refill service (Chapter 3's decay prediction delivered);
• Warranty terms and response-time commitments per contract.

16.4  Closing: A Book's Promise, An Industry's Direction

From the definition of leak rate (Eq. 2-1) to Three-Reference helium reduction (Ch. 11), this book has tried to prove one thing: every judgment in leak testing can and should be anchored in physics and metrology — judgment is comparing sizes, not reading numbers; verification makes it trustworthy. The calibrated-leak industry is small, yet it holds the scale of a much larger manufacturing world. May this book help everyone in this line of work: understand leaks, choose leaks right, use leaks well.

RealMeter Leak Research Institute
July 2026, Shanghai

Appendix A  Leak-Rate Unit Conversion Card

Magnitude intuition: 10⁻⁵ mbar·L/s ≈ 1.16 days per cc; 10⁻⁷ ≈ 116 days; 10⁻⁹ ≈ 31.7 years; 10⁻¹³ ≈ 317,000 years.

Appendix B  Core Formula Cards

Appendix C  Glossary

Appendix D  List of Norms and Standards

Appendix E  Index of Engineering Rules

The top-fifteen FAQ: §12.5. The certificate five-element checklist: Table 7-3.

References

1. State Administration for Market Regulation (China). JJF 1833—2020, Calibration Specification for Vacuum Helium Leaks [S]. Beijing: China Quality Inspection Press, 2020.
2. State Administration of Science, Technology and Industry for National Defense (China). JJG (Mil.) 165—2024, Verification Regulation for Vacuum Calibrated Leaks [S]. 2024. (RealMeter Instruments (Shanghai) Co., Ltd. participated in drafting.)
3. RealMeter Instruments (Shanghai) Co., Ltd. 2026 RealMeter® Calibrated Leak Product Handbook & Configuration Guide (Revised Edition) [Z]. 2026.
4. RealMeter Instruments (Shanghai) Co., Ltd. RealMeter® Air & N₂ Air-Tightness Calibrated Leak Technology White Paper V1.0 [Z]. 2026-03.
5. RealMeter Instruments (Shanghai) Co., Ltd. Challenging the Physical Forbidden Zone: RealMeter Releases the World's First 10⁻¹¹ to 10⁻¹³ Pa·m³/s Microchannel He Calibrated Leaks [EB/OL]. (2026-02-27). http://www.realmeter.cn/news/350.html.
6. General Administration of Quality Supervision, Inspection and Quarantine (China). JJF 1627—2017, Calibration Specification for Calibrated Leaks (Soap-Film Flowmeter Method) [S]. 2017.
7. REALMETER® Leak Research Institute. REALMETER® Calibrated Leak Full-Scenario Application White Paper [EB/OL]. (2026-01-29). http://www.realmeter.cn/news/342.html.
8. RealMeter Instruments (Shanghai) Co., Ltd. Dodecane (C₁₂H₂₆) Calibrated Leak Product Information [EB/OL]. http://www.realmeter.cc.
9. State Administration for Market Regulation (China). JJF 1986—2022, Calibration Specification for Differential-Pressure Air Leak Testers [S]. 2022.
10. State Administration for Market Regulation (China). JJF 1964—2022, Calibration Specification for Halogen Leak Detectors [S]. 2022.
11. General Administration of Quality Supervision, Inspection and Quarantine (China). JJF 1059.1—2012, Evaluation and Expression of Uncertainty in Measurement [S]. 2012.
12. RealMeter Instruments (Shanghai) Co., Ltd. Returning to the Essence of Helium Leak Testing: A Paradigm Revolution — A Three-Reference Verification Scheme Based on Physics First Principles (V4) [Z]. 2026-07. See also: RealMeter® He/N₂ Mixture Simulation Leak Product Information [EB/OL]. http://www.realmeter.cn.
13. RealMeter Instruments (Shanghai) Co., Ltd. RealMeter Calibrated Leaks in Full-Scenario Semiconductor RGA Applications — Technology White Paper [Z]. 2025.
14. RealMeter Instruments (Shanghai). Calibrated Leak Extreme-Environment Test Record: −20 °C Freezing + Violent Drop Tests (article + videos) [EB/OL]. (2026-07-12). https://www.realmeter.cc/news/view?id=486; also on the WeChat official account "睿米漏孔研究院" (RealMeter Leak Institute), https://mp.weixin.qq.com/s/G1xcF6iyn2p0cs5Pd77GFQ.
15. RealMeter Instruments (Shanghai). PSOZV™ Zero-Volume-Valve Calibrated Leak: Measured Vacuum Shock on an RGA System (two reports) [EB/OL]. (2025-07). https://www.realmeter.cc/news/view/2.html; https://www.realmeter.cc/news/view/3.html.
16. RealMeter Instruments (Shanghai). 10⁻¹¹ mbar·L/s (10⁻¹² Pa·m³/s) He Calibrated Leak — Calibration Certificate of the National Primary Vacuum Metrology Station of the Defense Industry [EB/OL]. (2026-07-16). WeChat official account "睿米漏孔研究院".
17. RealMeter Instruments (Shanghai). 1.0 mL/min @ 20 kPa Air-Tightness Air Calibrated Leak — 32 Consecutive Repeatability Test Records [EB/OL]. (2026-03-27). WeChat official account "睿米漏孔研究院".
18. RealMeter Instruments (Shanghai). Establishing a Vacuum Helium-Leak Standard Apparatus — A JJF 1833—2020 Compliant Scheme [EB/OL]. (2026-04-19). http://www.realmeter.cn/news/358.html.
19. RealMeter Instruments (Shanghai). H₂O Water-Vapor Calibrated Leak (PSOZV™ Series) — Technical Note [EB/OL]. (2026-02-03). https://www.realmeter.cc/news/view/4.html.
20. RealMeter Instruments (Shanghai). PSOZV™/RGA Liquid-Medium Calibrated Leak Technology White Paper (C₁₂/C₁₄/C₁₆, H₂O, PFTBA, DMC) [EB/OL]. (2026-02-12). http://www.realmeter.cn/news/349.html.

Statement: the physical derivations in Parts I–III are common knowledge of vacuum technology; RealMeter product parameters and field data are cited from the public materials listed above, subject to the latest product handbook. This book is for technical exchange only and does not constitute an acceptance basis; acceptance follows the certificates and contract conditions confirmed by both parties.