At ppt Level, You Are Usually Measuring the Sample System
Start with a sense of scale. One part per trillion is one in 10¹² — divide a 400-metre running track into a trillion parts and one of them is 0.4 micrometres, roughly the length of a bacterium. To declare a gas "in spec" at that scale, the measurement chain itself has to be several orders of magnitude cleaner than what it measures.
That leads to the most counter-intuitive fact about XCDA acceptance: most failing results describe the sampling system, not the gas. An under-purged sample line, one polymer fitting, one unnecessary dead leg — any of these is enough to condemn a perfectly healthy purifier on paper. And in the other direction, the wrong instrument choice lets a system that never met spec pass comfortably.
So acceptance is two jobs in sequence: first prove the measurement chain is credible, then prove the gas is in spec. Reverse the order and the data means nothing.
Five Instrument Classes and How Low Each One Sees
Fig. 2: How Low Can Each Instrument Actually See?
Solid = trustworthy range; dashed = the limit only reachable with large sample volumes — data in the dashed zone should not stand alone on an acceptance sheet
The same instrument can differ by two orders of magnitude between species: CRDS reaches hundreds of ppt on H₂O but often only ppb on CO₂. Always ask which species and under which sampling conditions a quoted detection limit applies — a spec sheet with a single number is meaningless.
| Instrument | Principle in one line | Main targets | Online / offline | Practical role |
|---|---|---|---|---|
| Dew point meter (chilled mirror / Al₂O₃) | Condensation temperature or dielectric change | H₂O | Online | Good for monitoring the dryer stage, unfit for XCDA acceptance |
| CRDS cavity ring-down | Laser bounces in a high-reflectivity cavity, decay rate is measured | H₂O, CO₂, CH₄ | Online, continuous | The most useful box on site: fast response, long-term trending |
| APIMS atmospheric-pressure ionisation MS | Ionisation at ambient pressure, then mass separation of trace species | H₂O, O₂, CO, CO₂, CH₄, N₂ | Online (usually outsourced) | Highest credibility for acceptance, highest cost and skill barrier |
| TD-GC/MS thermal desorption | Sorbent tube sampling, thermal desorption, chromatographic separation | Organics, refractory compounds (HMDSO) | Offline | The only practical ppt-level route for organics and siloxanes |
| IC ion chromatography (with impinger) | Gas passes through absorbing solution, ions analysed | Volatile acids, volatile bases | Offline | The standard method for acid and base specs, long sampling time |
Three points that are routinely missed:
- ▸One instrument can vary by two orders of magnitude across species. CRDS reaches hundreds of ppt on H₂O but often only ppb on CO₂. A bare "detection limit 0.1 ppb" on a quotation always deserves the follow-up: for which species, under what sampling conditions?
- ▸The limit of an offline method depends on sample volume. TD-GC/MS and IC have no fixed detection limit — the more you draw and the longer you sample, the lower it goes. Any report must therefore state sample volume and duration; a report with concentration alone cannot be audited.
- ▸Online and offline must cross-check each other. Online instruments show trends and disturbances; offline methods give absolute values and species detail. Use only one and you have a blind spot.
The Sample Line Fools You More Often Than the Analyser
This is the most practically useful section of the article. Each of the five problems below is on its own enough to invalidate the data.
1. Material and permeation
People running positive-pressure systems often assume nothing can get in. That holds for leakage; it does not hold for permeation, which is driven by partial pressure difference. Inside the tube, H₂O partial pressure is pushed to ppt level; outside it is ppm to percent. Across that gap, water molecules diffuse steadily inward through any polymer component — O-rings, hoses, plastic fittings are all permeation paths.
| Material | Suitable for | Not suitable for | Notes |
|---|---|---|---|
| EP 316L electropolished stainless | Everything, H₂O especially | — | The standard for ppt systems; the passivated inner layer suppresses adsorption |
| PFA | Acid/base sampling, corrosive service | H₂O measurement | Noticeable moisture permeation and memory effect |
| PTFE hose | Temporary, non-critical measurement | Any ppt-level parameter | Porous structure both adsorbs and releases; strong memory effect |
| Ordinary stainless / copper tubing | Facility CDA | XCDA | Rough inner wall, high adsorption capacity, very slow dry-down |
2. Dry-down time
A freshly connected sample line holds adsorbed moisture on its inner wall and takes a long time to give it up. That period is called dry-down, and its length depends on material, internal surface area, temperature and purge flow. Empirically, an EP 316L line under continuous purge takes hours to days to fall from ppm to ppt on H₂O; a PTFE hose may never get there.
The working rule is simple: while the reading is still falling monotonically, no data counts. Only after the value has been stable for several hours should acceptance recording begin.
3. Dead legs
Sample valves, pressure gauges and spare ports create branches. If a branch is too long, the gas inside it is not swept by the main flow and becomes a small reservoir continuously bleeding contamination back into the header. The convention in semiconductor ultra-high-purity systems is to keep branch length within three times the tube diameter (L ≤ 3D) and to prefer zero-dead-leg valves.
Dead legs have a recognisable signature: very slow stabilisation after start-up, readings that jump when flow changes, and data that degrades after a stop-restart.
4. Flow conditions during sampling
Breakthrough behaviour in an adsorption bed is tied directly to superficial velocity. Passing at low flow says nothing about passing at full process flow — higher velocity means shorter residence time, and weakly held species break through first. Acceptance sampling must be done at design flow, and it is worth adding a flow-step test (for example a sudden 50% to 100% step).
5. Background and blank testing
Before connecting the gas under test, run the whole chain on a known zero gas (or a verified high-purity source) to establish the background of sample system plus instrument. If that background already sits near the specification threshold, this measurement chain is not qualified to perform the acceptance. The step is frequently skipped, yet it underpins the credibility of everything that follows.
| Sampling error | What the data looks like | Commonly misdiagnosed as |
|---|---|---|
| Insufficient purge | Reading falls monotonically, stabilises only after days | Inadequate purifier performance |
| Permeation through polymer parts | H₂O stuck at a floor value, everything else normal | Saturated adsorption bed |
| Dead leg | Readings jump with flow changes, worse after stop-restart | Valve sequencing fault |
| Sample point downstream of a regenerating vessel | Periodic spikes synchronised with switchover | Insufficient regeneration gas |
| Background never established | Every parameter slightly higher than expected | Overall system underperformance |
Why a Dew Point Meter Cannot Sign Off XCDA
Three reasons:
- 1Range. The XCDA limit of H₂O <100 ppt corresponds to roughly −122°C atmospheric dew point, far below the credible range of a chilled-mirror instrument.
- 2Response. Below −90°C a chilled mirror takes hours to equilibrate; aluminium oxide and silicon sensors show marked hysteresis and drift in the ultra-dry region and need frequent recalibration.
- 3One species only. Even if H₂O were measured accurately, CO₂, TOC, acids, bases and refractory compounds remain completely unaddressed.
A dew point meter does have a place in an XCDA system — it is excellent for watching the health of the stage-2 dryer, which lives in the ppm-to-ppb band. Using it to sign an acceptance certificate is applying a tool where it physically cannot reach.
Sequencing FAT and SAT
| Phase | Purpose | What is measured | Pass logic |
|---|---|---|---|
| FAT (factory test) | Verify the equipment and control logic | Outlet specs at full and partial load, a complete regeneration cycle, alarms and interlocks, data logging integrity | Must cover at least one full switchover cycle, not steady state alone |
| Line purge and dry-down | Remove sampling and piping interference | Online H₂O trend until flat | Complete only when the reading stops falling monotonically |
| SAT stage 1: background | Prove the measurement chain is credible | Zero gas background, instrument blank, sample line blank | Background should sit well below the threshold — conventionally an order of magnitude |
| SAT stage 2: steady state | Prove the gas is in spec | All parameters, continuously recorded at design flow | 72 hours continuous is the usual baseline, covering at least two full regeneration switchovers |
| SAT stage 3: disturbance | Prove it holds under real operation | Flow steps, switchover moments, stop and restart | The maximum during disturbance must also be within spec |
The key concept is that the criterion applies to the whole period, not to a single point. "We measured 8 ppt one afternoon" and "every data point over three continuous days was below 10 ppt" are evidence of entirely different strength. A contract that says only "outlet <10 ppt", without stating the measurement period and the decision rule, has effectively said nothing.
Five Failure Modes and Their Data Signatures
| Symptom | Most likely cause | How to confirm |
|---|---|---|
| Regular spikes synchronised with switchover | Insufficient regeneration flow or temperature, valve sequence timing error | Overlay the switchover signal on the concentration trace and check alignment |
| TOC degrades alone, everything else normal | Breakthrough in the oil removal stage, sorbent poisoned by oil | Check pre-treatment differential pressure and drain records, add an inlet TOC measurement |
| Target not met for days after start-up, then normal | Dry-down incomplete, or unsuitable line material | Extend purge and re-test; compare against a verified EP 316L sample line |
| All species step up together | Bed breakthrough, or actual flow exceeding design | Review flow records and cumulative bed hours against the regeneration cycle setting |
| One sample point abnormal, others fine | Dead leg, fitting permeation or insufficient purge at that point | Move a verified sample line to the same port and compare |
One counter-intuitive warning: be suspicious when data suddenly looks beautiful. The most common false pass at ppt level comes from a sample line that is not properly connected, a valve left closed, or an instrument auto-zeroing the real signal away as baseline. Genuine ppt-level data carries visible noise and small fluctuations; a perfectly flat line usually means the instrument is not looking at the gas.
The Arithmetic Behind the Particle Specification
XCDA particle requirements are often written as ≤1 pcs/m³ @0.003 μm. Verifying that one line costs far more than most people expect.
At the typical condensation particle counter sampling rate of 2.83 LPM (0.00283 m³/min), accumulating one cubic metre takes 353 minutes — nearly six hours of continuous counting. And that only fills 1 m³ once; establishing statistical confidence that the average is ≤1 particle requires longer accumulation or repeated runs.
Also mind the 0.003 μm (3 nm) size. The D50 of a general-purpose condensation particle counter is usually 10 nm or 20 nm; counting reliably at 3 nm requires an ultrafine unit (UCPC). When reviewing an acceptance report, check the instrument's D50 and its counting efficiency at 3 nm — otherwise "zero counts" may simply mean the instrument cannot see them. For a broader comparison of particle and AMC monitoring methods, see AMC monitoring technology compared.
Six Questions to Ask a Supplier Before Purchase
These are not traps — they move future disputes to the front of the project:
- 1What is the calibration basis of each specification? Acids as SO₂, bases as NH₃, TOC as C₃H₈ or something else? Numbers on different bases cannot be compared directly.
- 2Under which inlet conditions and flow does the guarantee hold? Will seasonal swings in ambient air — monsoon humidity, emissions from a neighbouring plant — void it?
- 3Does the guarantee cover regeneration switchover? If it only covers steady state, who owns quality during the disturbance?
- 4Which instruments will be used for acceptance, and who issues the report? Self-check, third party, or a metrology institute? Does the report include raw data and sampling conditions?
- 5How are sample points configured? Is there a qualified sample valve at the outlet, is it zero-dead-leg, and what material is it? Leaving this out of the design stage is painful to retrofit.
- 6Is the guaranteed value a typical value or a period maximum? This single question defines the acceptance criterion and is the most common source of contractual dispute.
For selection logic and flow calculation see the gas purifier selection guide; for the physical meaning and failure mechanism behind each specification, see the companion article What is XCDA? From clean dry air to parts-per-trillion purity.
FAQ
Q: Does acceptance require APIMS?
Not necessarily — it depends on what you need to prove. APIMS carries the highest credibility and suits first acceptance of a new system or arbitration in a dispute; routine monitoring with CRDS plus periodic offline TD-GC/MS and IC is usually sufficient. The pragmatic pattern is to outsource a full analysis for the initial acceptance, then hold the trend with online instruments and re-verify offline quarterly or half-yearly.
Q: If H₂O passes, does that mean everything passes?
No, and this is the most common shortcut. H₂O, organics and acid/base species travel through different bed layers and fail by different mechanisms. There are real cases where H₂O was perfectly normal while TOC had already broken through, because the oil-removal stage failed first while the water bed was still fine. Cover at least four families: H₂O, TOC, acids and bases.
Q: How long does continuous measurement need to run?
The principle is that it must cover at least two complete regeneration switchover cycles, since a single switchover may happen to behave. Most systems cycle somewhere between a few hours and a day, which makes 72 continuous hours a common and reasonable baseline. Systems with unusually long cycles need proportionally more.
Q: What should we check in a third-party test report?
Beyond the concentration numbers, confirm four things: sample point location, sample volume and duration, instrument model and detection limit, and the background (blank) value. Without a background figure there is no way to tell whether the number came from the gas or from the sampling system. Also check that the report date lines up with the actual operating state — sampling at low flow or right after a regeneration naturally produces flattering data.
Q: What should routine monitoring watch?
Online, prioritise H₂O (fastest and most sensitive, a leading indicator of system health) and differential pressure (the warning sign for pre-treatment and bed blockage). Adding online TOC is worthwhile if budget allows, because it is the first signal of oil breakthrough. Acids, bases and refractory compounds are offline only, so manage them with scheduled sampling and trend tracking.
Q: The data fails right after start-up — is the equipment faulty?
Usually not. The whole system including the piping needs to dry down, over hours or even days, and a steadily falling reading during that window is normal. The distinction to make is between "still falling" and "stuck". The first means not yet stable; only the second is a problem. Judge it from the slope of the trend curve, not from the number at any one moment.