Achieving 10⁻¹² Pa·m³/s Leak Rates: Semiconductor Leak Rate Metal Seal Design for Fluorine and Ammonia Environments

A 10⁻¹² Pa·m³/s helium leak-rate target in fluorine or ammonia semiconductor service is a joint specification for a semiconductor leak rate metal seal, not a catalog rating. That number sits near the practical detection floor of modern helium mass-spectrometer leak detectors, and it is only meaningful when the test gas, pressure differential, method, and acceptance location are stated. Meeting it in halogen or ammonia chemistry requires an all-metal seal architecture, residual contact stress after bake-out, and alloys that do not corrode or outgas into the process.

What Semiconductor Leak Rate Metal Seal Target Should Engineers Specify?

The leak rate a semiconductor leak rate metal seal must hold is the rejection leak rate on the drawing, converted to helium standard conditions and tied to a named test method. Leybold calls a high-vacuum system “very tight” below 10⁻⁶ mbar·l/s of air (10⁻⁷ Pa·m³/s). A 10⁻¹² Pa·m³/s helium target is four decades tighter and belongs on UHV or UHP gas-delivery specifications, not generic industrial ratings.

Leybold’s vacuum-fundamentals guidance is explicit on two points that semiconductor teams often collapse into one sentence. First, “no detectable leaks” is not an acceptance criterion; every experienced engineer writes a finite leak rate under defined conditions. Second, a leak rate of 1.0·10⁻¹² mbar·l/s corresponds to a hole on the order of 1 Å and is described as the lower detection limit for modern helium leak detectors. Converted to SI units, that detector floor is 10⁻¹³ Pa·m³/s. The 10⁻¹² Pa·m³/s figure used on many UHV and semiconductor drawings is therefore one decade above the quoted detector floor — still an extremely small pV-throughput, and still close enough to instrument noise that background helium, fixture leaks, and calibration drift can dominate the reading. See Leybold’s definition and measurement of vacuum leaks for the unit language, the high-vacuum tightness bands, and the helium-standard-condition definition referenced to DIN EN 1330-8.

For high-purity gas piping and components — the fluorine and ammonia delivery path into the tool — SEMI F1-0521 is the current specification that “defines the leak testing requirements and leakage rates for high-purity gas piping systems and components used in semiconductor manufacturing.” SEMI F1 applies to complete systems, subsystems, and individual components, and it states requirements for both the user and the manufacturer. This article does not quote a numeric SEMI F1 limit that is not in the public abstract; the correct engineering move is to buy the current revision and put that limit, plus the test method, on the RFQ. SEMI F74 then supplies a test method specifically for evaluating metal seal designs used in those gas-delivery systems, covering both conventional metal face seals and surface-mount gas systems.

Process-chamber closures, lid seals, and transfer interfaces are not always written to SEMI F1. Those joints often inherit a helium spray or hood-test number from the OEM vacuum specification. The same physics still apply: the seal is only one element in a compressed joint. Groove depth, flange finish, bolt load, plating, and thermal movement decide whether the specified semiconductor leak rate is reachable after the first bake-out, not only at the incoming inspection stand. Sonkit’s semiconductor metal seal applications page is the product-family entry point for that joint-level discussion.

Why Do Fluorine and Ammonia Attack Conventional Seals First?

Fluorine-based etch chemistry and anhydrous ammonia destroy elastomer seals by chemical attack, swelling, compression set, and particle shedding before a well-designed metal seal loses residual contact stress. NIST documents fluorine chemistry as the workhorse of dielectric etch. NIOSH lists anhydrous ammonia as a corrosive nitride-process gas with a 25 ppm TWA and 35 ppm STEL, and notes that ammonia attacks copper and galvanized surfaces.

NIST’s publication on the role of fluorine chemistry in anisotropic etching of dielectric materials is the right starting point for the process-gas side of the seal problem. Fluorine-containing compounds have been used for years to etch silicon oxide. The etch depends on a balance between material removal at the trench bottom and a protective sidewall film. For the seal, that chemistry means halogen-rich plasma, chamber-clean residues, and hydrofluoric species that can reach the sealing land during process, clean, or maintenance. An elastomer that swells, cracks, or sheds filler in that environment creates both a leak path and a contamination source.

Ammonia is a different chemical attack with the same commercial consequence. The NIOSH Pocket Guide entry for ammonia describes a colorless gas with a pungent odor, boiling point −28 °F (−33 °C), flammable range 15–28% in air, and incompatibilities that include strong oxidizers, acids, and halogens. NIOSH also states that ammonia is corrosive to copper and galvanized surfaces. In the fab, anhydrous ammonia is a standard precursor for silicon-nitride and related films. A polymer seal in that path ages by chemical attack and by permeation. A copper-rich plating or a copper alloy land in the same path is a materials error even if the leak test passed on helium at incoming inspection.

Contamination control is the third failure mode, and it is not the same problem as leak rate. NASA’s long-running vacuum outgassing program exists because material release in vacuum is an engineering variable: total mass loss and collected volatile condensable material change system cleanliness. In a semiconductor chamber, those condensables become defectivity. A seal that meets a helium number on day one and then outgasses or sheds particles after plasma exposure has failed the application even if the leak detector still reads “tight.” All-metal sealing removes the polymer reservoir. It does not remove the need to control machining oils, packaging films, and handling particles on the metal itself.

How Should Engineers Convert and Specify a 10⁻¹² Leak Rate?

Specify a 10⁻¹²-class leak rate as a helium pV-throughput under named conditions, not as a unitless slogan. The SI unit is Pa·m³/s; the vacuum unit is mbar·l/s; UHP fittings often use atm·cm³/s. They convert by 1 Pa·m³/s = 10 mbar·l/s ≈ 9.87 atm·cm³/s. A drawing that says “10⁻¹²” without unit, gas, and method is not a specification.

ISO 20485:2017 describes the techniques for detecting a leak using a tracer gas and a tracer-gas-specific detector. That is the method-language standard semiconductor teams should cite when they write “helium leak test” on a metal-seal drawing. The public abstract is on the ISO 20485:2017 tracer gas method page. ASTM E499/E499M is the companion practice for mass-spectrometer leak detectors used in detector-probe (sniffer) mode; its scope covers locating leaks at 1 × 10⁻⁷ Pa·m³/s (1 × 10⁻⁸ std cm³/s) or greater. That probe-mode floor is five decades coarser than a 10⁻¹² Pa·m³/s vacuum-method target. Citing ASTM E499 as the acceptance method for a 10⁻¹² Pa·m³/s chamber lid is a category error. Use the ASTM E499/E499M practice for sniffer work on pressurized gas sticks; use a vacuum spray or hood method, calibrated against a reference leak, for UHV chamber joints.

Leybold’s conversion rules matter at these magnitudes. At leak rates above about 10⁻⁴ mbar·l/s the flow is laminar-viscous; below about 10⁻⁶ mbar·l/s it is molecular. A 10⁻¹² Pa·m³/s helium leak is deep in molecular flow. In molecular flow, leak rate scales with the square root of molar mass, so a helium reading is not numerically identical to an air, fluorine, or ammonia leak through the same geometry. The purchase order should state the tracer gas (almost always helium), the high-side and low-side pressures, the temperature, and whether the number is a local spray reading or an integral hood reading. ISO 20486:2017, which covers calibration of reference leaks, notes that comparison calibration against known reference leaks is readily applicable for helium leaks below 10⁻⁷ Pa·m³/s — which is exactly the decade where semiconductor metal-seal work lives.

Two more specification rules prevent most incoming-inspection fights. First, state whether the limit applies to the seal coupon, the assembled groove, or the installed tool joint. A seal that is tight in a laboratory flange can leak in a thin aluminum chamber lid that rotates under bolt load. Second, state the bake-out and thermal-cycle sequence before the acceptance reading. A metal seal that passes at room temperature and then loses contact after a 150 °C or 250 °C bake has failed springback — the elastic recovery that keeps residual load on the sealing land after the joint moves. Sonkit’s article on metal sealing methods for ultra-high vacuum applications is the companion discussion of UHV method and groove practice.

Which Metal Seal Architecture Holds Residual Load After Bake-Out and Gas Exposure?

A spring-energized Metal C-Ring holds residual contact after bake-out better than a hollow Metal O-Ring driven past its elastic range, because the C-profile and internal spring store elastic energy. The architecture choice is a load-and-recovery decision, not a brand preference. Fluorine and ammonia add a second filter: jacket and plating must survive the chemistry without particles or a corrosion channel on the sealing line.

A hollow Metal O-Ring seal is a robust static closer when the flange is stiff, the bolt load is high, and the joint is rarely opened. The tube wall yields to create conformity. That is an advantage on a thick stainless gas-box flange. It is a liability on a large, thin chamber lid that cannot deliver crush load without bowing, and it is a liability after thermal cycling if the O-ring has taken a permanent set. Pressure-energized (vented) O-rings recover some load when system pressure is present; they do not help a vacuum-side joint that sees atmosphere-to-vacuum differential in the opposite direction from the vent.

A Metal C-Ring seal changes the mechanics. The open C-section is pressure-assisted when the pressure is on the open side, and the remaining wall acts as a spring. Seating load is typically lower than an equivalent hollow O-ring, which protects aluminum and coated semiconductor flanges. Springback — the seal’s elastic recovery after compression — is the reason C-rings are the default conversation for joints that see bake-out, lid cycling, or modest flange rotation. A spring-energized Metal C-Ring adds an internal helical or ribbon spring so residual load does not depend only on the jacket alloy’s own elasticity. That is the architecture to evaluate when the leak-rate target is in the 10⁻⁹ to 10⁻¹² Pa·m³/s conversation and the flange cannot be treated as infinitely stiff.

None of these profiles will hold a detector-adjacent leak rate if the groove is wrong. The compression window must include plated seal height, actual machined groove depth, flange flatness, and plating thickness. A groove that is too shallow over-compresses the jacket, collapses the spring, and removes the recovery the design was bought for. A groove that is too deep never develops the contact stress needed to close machining valleys. Radial scratches across the sealing land are leak channels at helium-detector sensitivity even when the Ra number looks acceptable. The semiconductor leak rate metal seal is therefore specified as a system: profile, jacket, spring, plating, groove, and load path. Sonkit groups those options under metal ring seal products so the profile discussion stays attached to the application inputs rather than to a single catalog page.

What Jacket Alloy and Plating Survive Fluorine and Ammonia Chemistry?

Jacket alloy and plating are selected for residual spring load and for compatibility with fluorine residues and ammonia, not for a generic nickel-alloy label. INCONEL alloy 718 is the high-strength spring and jacket candidate when temperature and recovery dominate. HASTELLOY C-276 is the jacket candidate when halide or hydrofluoric residues dominate. Copper-rich surfaces are a documented ammonia incompatibility.

Special Metals’ INCONEL alloy 718 technical bulletin states that alloy 718 (UNS N07718 / W.Nr. 2.4668) is a high-strength, corrosion-resistant nickel-chromium material used at −423 °F to 1300 °F. In the AMS 5662 / 5663 bar, forging, and ring condition, the bulletin lists a minimum room-temperature 0.2% yield strength of 150 ksi (about 1034 MPa) and 125 ksi at 1200 °F (649 °C). The same bulletin’s spring section states that the alloy’s relaxation resistance is a factor in its successful use for springs at temperatures up to 1100 °F (593 °C). Those are the numbers that justify 718 as a spring-energized metal-seal spring alloy and as a jacket alloy when the joint sees bake-out and must keep residual load. They are not a license to treat 718 as immune to every halogen.

Haynes International’s HASTELLOY C-276 alloy page describes UNS N10276 as a nickel-chromium-molybdenum alloy (nominal 16% Cr, 16% Mo, 4% W) with outstanding resistance to pitting and crevice attack in chlorides and other halides. Haynes publishes hydrofluoric-acid corrosion rates for C-276: 0.34 mm/y in 5% HF at 52 °C, 0.41 mm/y in 10% HF at 52 °C, and 0.48 mm/y in 20% HF at 52 °C, and warns that hydrofluoric acid can also induce internal attack of nickel alloys. Those figures are laboratory rates in reagent HF, not a seal-life prediction in a plasma chamber, but they are the right order-of-magnitude evidence for choosing a C-276 jacket when fluorine-clean residues or HF vapor can reach the sealing land. C-276 sheet in the mill-annealed condition shows a room-temperature 0.2% yield near 356 MPa (51.6 ksi) on Haynes’ data — much lower than aged 718 — so C-276 is typically the corrosion jacket, not the high-load spring.

Plating is the third material decision. A soft overlay (silver or gold is the usual vacuum conversation) exists to fill machining texture at a seating load the flange can actually deliver. Thickness is an engineering variable: too thin and the valleys stay open; too thick and the overlay smears, changes groove fit, or relaxes after bake-out. Ammonia adds a hard constraint that NIOSH already stated: ammonia is corrosive to copper and galvanized surfaces. A copper-plated seal or a copper sealing land in an ammonia line is the wrong interface even if helium leak testing is excellent. Gold is chemically quieter in both halogen and ammonia service and is the usual UHV cleanliness choice; silver remains common where seating load is limited and the gas chemistry has been reviewed. The plating specification should name the metal, the thickness range, the adhesion test, and the cleaning process, not just a color.

How to Select a Semiconductor Leak Rate Metal Seal for Fluorine and Ammonia Duty?

Select the semiconductor leak rate metal seal by available seating load, required recovery after thermal cycling, and chemical exposure — in that order — then confirm that the helium method can resolve the written leak rate. An elastomer is wrong for process-facing fluorine or ammonia joints. A crush-heavy hollow Metal O-Ring is wrong for a thin, frequently opened chamber lid.

The table is a screening tool, not a substitution chart. A spring-energized C-ring in the wrong groove will leak. A hollow O-ring on a stiff 316L face-seal gland, with a C-276 or 718 jacket and a chemistry-reviewed plating, can be the correct gas-stick answer. The decision record should show the available bolt load at the real flange, the bake-out temperature, the process and clean gases, the written helium limit with method, and whether the joint is single-use or make-break. If those five inputs are missing, the seal conversation is not ready for a drawing release.

What Checklist Should the Design Team Complete Before Freezing the Groove?

Freeze the groove only after the leak-rate method, media list, load budget, and alloy/plating pair are written down. A semiconductor metal-seal groove copied from a previous tool and then asked to hold a 10⁻¹² Pa·m³/s helium number in fluorine or ammonia service is the usual path to a redesign. The checklist below is the minimum package Sonkit application engineering needs.

1. Write the rejection leak rate with units, gas, and method. Example language: “Helium, vacuum spray method, ΔP = 1 bar to <1 mbar, ≤ 1 × 10⁻¹² Pa·m³/s at the assembled joint after bake-out,” or the SEMI F1 limit that applies to the gas-delivery component. Do not write “zero leak” or “helium tight.”
2. Name the test standard. ISO 20485 for tracer-gas method language; SEMI F1 / SEMI F74 for high-purity gas-delivery hardware; ASTM E499 only if the test is actually a sniffer/probe test on a pressurized stick.
3. List every gas that can reach the sealing land. Process gas (F-containing etch or NH₃), chamber-clean chemistry, purge gas, and maintenance solvents. Include whether HF vapor or aqueous HF is used on the tool.
4. State the thermal cycle. Bake-out temperature and duration, process temperature, number of lid or valve cycles, and whether the acceptance leak test is before or after the first bake.
5. Measure the real load path. Flange thickness, material (aluminum, 316L, nickel alloy, coated land), bolt size and pattern, and maximum allowable flange stress. Torque is not seating load.
6. Record finished groove and land data. Depth, width, corner radius, flatness, roughness, and lay direction — as-machined, not as-modeled. Prohibit radial scratches across the sealing line.
7. Select jacket, spring, and plating as a set. 718 when residual load and temperature dominate; C-276 when halide/HF residues dominate; no copper-rich plating on ammonia joints; gold or reviewed silver for vacuum cleanliness.
8. Define cleanliness and handling. Packaging, glove protocol, and a cleaning method that does not leave film on the plated land. Treat the plated surface as a sealing interface, not as a handling surface.
9. Decide make-break life. Single-use crush versus limited reuse. A detector-adjacent leak rate and a damaged land after the first service event is a failed maintenance model.
10. Plan the witness-mark and helium map. After the first article, inspect contact continuity and localize any helium indication before changing alloy or plating.

Frequently Asked Questions

Can a metal seal actually achieve 10⁻¹² Pa·m³/s in fluorine or ammonia service?

A metal seal can be part of a joint that meets a 10⁻¹² Pa·m³/s helium reading when the groove, load, finish, alloy, and test method are designed together; it is not a standalone catalog rating. Leybold places the modern helium-detector floor at 10⁻¹² mbar·l/s (10⁻¹³ Pa·m³/s), so a 10⁻¹² Pa·m³/s target is detector-adjacent. Background helium, fixture leaks, and an unbaked joint will hide or fake that number. Specify the method and the baked, assembled joint — then choose a spring-energized or C-ring architecture that still has residual load after the thermal cycle.

What is the difference between 10⁻¹² Pa·m³/s and 10⁻¹² mbar·l/s?

They differ by a factor of ten: 1 Pa·m³/s = 10 mbar·l/s, so 10⁻¹² Pa·m³/s = 10⁻¹¹ mbar·l/s, while 10⁻¹² mbar·l/s = 10⁻¹³ Pa·m³/s. Mixing the units on a drawing is one of the most expensive specification errors in vacuum work. Always write the unit, the tracer gas, and the pressure differential. Leybold’s helium-standard-condition language (DIN EN 1330-8) is the conversion reference when the factory test is not identical to the service condition.

Why do elastomer seals fail in fluorine etch and ammonia nitride tools?

Elastomer seals fail in those tools because the chemistry attacks the polymer, the vacuum extracts condensable species, and maintenance generates particles. NIST documents fluorine chemistry as the basis of dielectric etch; NIOSH documents ammonia as a corrosive, copper-attacking process gas with a 25 ppm REL. Even when the elastomer still “seals” on a pressure test, outgassing and shed particles have already failed the contamination-control requirement. An all-metal seal removes the polymer reservoir; it does not remove the need for clean handling of the metal.

Should the jacket be INCONEL 718 or HASTELLOY C-276?

Use INCONEL 718 when residual spring load and bake-out temperature dominate; use HASTELLOY C-276 when halide or hydrofluoric residues dominate the sealing land. Special Metals rates 718 from −253 °C to 704 °C and cites spring use to 593 °C, with a 150 ksi minimum room-temperature yield in the AMS 5662/5663 condition. Haynes rates C-276 for halide pitting resistance and publishes HF corrosion rates on the order of 0.3–0.5 mm/y at 52 °C in dilute reagent HF. Many joints use 718 for the spring and a more corrosion-resistant jacket, or C-276 as the jacket with a 718 spring. The pair is an application decision, not a single-alloy slogan.

Is a sniffer test enough to accept a 10⁻¹² Pa·m³/s chamber seal?

No. ASTM E499 probe-mode practice is scoped to leaks of 1 × 10⁻⁷ Pa·m³/s or greater, which is five decades coarser than a 10⁻¹² Pa·m³/s vacuum-method target. Use a vacuum spray or hood method with a calibrated reference leak for chamber and UHV joints. Reserve sniffer testing for locating leaks on pressurized gas-delivery hardware, and write SEMI F1 / SEMI F74 when the part is a high-purity gas-system component rather than a chamber lid.

When should Sonkit be brought into the groove design?

Bring Sonkit in before the groove is frozen whenever the helium target is 10⁻⁹ Pa·m³/s or tighter, the media include fluorine-clean chemistry or ammonia, the flange is aluminum or coated, or the joint will be opened for maintenance. Early review of compression window, plating, and load path is faster than changing the chamber lid after the first failed helium map. Send the flange drawing, groove dimensions, media list, bake-out profile, available bolt load, and the written leak-rate method.

What Should Engineering Teams Do Next?

Treat the 10⁻¹² Pa·m³/s semiconductor leak rate as a joint-level helium specification in fluorine or ammonia service, then build the metal seal to hold residual contact after bake-out without contaminating the chamber. Write the number with units and method. Choose a C-ring or spring-energized architecture when the flange cannot deliver crush load. If those decisions are still open, the next step is an application review, not a part number.