Compressed Air for Semiconductor Manufacturing and Cleanroom Applications

In semiconductor factories, CDA falls under facility support. Process integration people do not pay much attention to it. When yield problems come up and the team is chasing root causes, CDA is at the back of the line. Recipe parameters, chamber condition, specialty gas purity, chemical batch, UPW quality, all checked, nothing found, and only then does someone think to go take a look at CDA. By that point the excursion has been running for days, sometimes weeks. Sometimes the yield team has already filed a preliminary 8D pointing at a process drift that was never there, and unwinding that paperwork is its own headache.

Facility Support

Semiconductor CDA Infrastructure

ISO 8573 Class 1 is the highest grade for industrial compressed air. Semiconductor fabs use it as a starting point. Class 1 limits particles in the 0.1 to 0.5μm range to 20,000 per cubic meter. Advanced node fabs control internally to less than 1 particle per cubic foot above 0.01μm. Oil content Class 1 is ≤0.01 mg/m³, fabs control to ≤0.003 mg/m³ or lower. The standard was written for a broad industrial audience. It was not designed with sub-7nm gate-all-around transistors in mind. Treating a Class 1 test report from an air compressor vendor as a sign-off on semiconductor readiness has caused problems on enough new fab projects that it should not still be happening.

Dew point requirement is below -70°C. Adsorption dryers hitting this under steady state is routine.

Dew Point Control

Adsorption Dryer Towers

Adsorption dryers alternate twin towers, one adsorbing one regenerating, then they switch. At switchover the freshly regenerated tower has not reached thermal or adsorption equilibrium. Under steady flow the dew point bump is minor, -70°C to -65°C, recovers in seconds.

Fab CDA demand is not steady. Shift changes can bring fifty or sixty tools calling for CDA within a few minutes. Group re-qual after PM does the same. If this demand spike coincides with a tower switchover, the freshly online tower sees high flow before it has stabilized, residence time in the desiccant bed drops, and the two effects stack. Dew point can hit -40°C. That is 127 ppmv versus the normal 2.5 ppmv.

Here is what happens inside the dryer tower in more detail, because the timing matters. A heatless regen twin-tower dryer doing a 5-minute NEMA cycle spends about 4.5 minutes on adsorption and 30 seconds on repressurization after switchover. During repressurization the tower is pressurizing from near-atmospheric to system pressure using finished CDA as the pressurization source. That finished CDA is clean and dry, so in theory the tower should start producing clean output immediately after repressurization completes. In practice the desiccant near the inlet of the freshly switched tower has been exposed to wet purge exhaust from the regeneration side during the blowdown phase, and even after repressurization, those first few layers of desiccant beads are moisture-loaded. The adsorption front has moved inward. At low flow, there is enough desiccant depth behind the front to still produce -70°C outlet gas. At high flow the gas moves through the bed fast enough that breakthrough happens before the adsorption front has time to re-establish at the inlet. The dew point rises. How much it rises depends on desiccant type, bead size, bed depth, inlet moisture loading, gas velocity, and the temperature profile left over from regeneration.

The pulse of wet air enters the distribution piping. Some moisture adsorbs onto the pipe walls as it travels through, attenuating the peak but loading the pipe walls. That adsorbed moisture then slowly desorbs over the following hours. The original spike lasts seconds, maybe a minute. The desorption tail from the pipe walls lasts hours and keeps the local dew point elevated by maybe 5 to 10°C above normal at points downstream. During those hours, every tool pulling CDA from that section of piping is getting air that is within specification on paper (-70°C at the dryer outlet, monitored continuously, looks fine) but is actually sitting at -63°C or -60°C at the POU because of pipe wall desorption adding back moisture that the dryer already removed. The POU dew point is almost never monitored. The dryer outlet dew point is. The dryer outlet dew point does not represent what the tool is receiving after the gas has traveled through 200 meters of piping with freshly moisture-loaded walls.

Now, what the moisture does to ALD specifically. If an ALD chamber is venting its loadlock with CDA when the transient arrives, water molecules enter the loadlock, adsorb on the wafer surface and chamber walls, and do not fully desorb during the pump-down cycle. Desorption activation energy for water on SiO₂ is 0.3 to 0.9 eV depending on coverage and surface hydroxylation state. A room-temperature pump-down cycle lasting a few seconds at a few hundred mTorr base pressure does not supply enough thermal energy to overcome that barrier for the more tightly bound fraction. On a fresh thermal oxide surface with full hydroxyl termination, the first monolayer of water is bound at the high end of that range. Subsequent monolayers are bound more weakly and do come off under vacuum, but the first partial monolayer is stubborn.

So the wafer enters the ALD chamber with an unknown partial coverage of water on its surface. The first precursor pulse, say TMA for an Al₂O₃ process, reacts with available surface sites through self-limiting chemisorption. Sites occupied by water are blocked or produce a different reaction product. TMA can react with adsorbed water to form Al₂O₃ directly, which sounds like it should not matter since Al₂O₃ is the target film anyway, but the reaction between TMA and uncontrolled surface water produces a different local film density and bonding structure than the designed ALD half-reaction of TMA with surface hydroxyl groups. Over one cycle the difference is sub-angstrom and unmeasurable. Over a hundred cycles, ellipsometry sees within-wafer thickness non-uniformity. Over a thousand cycles the film quality shows up in electrical test as dielectric constant variation, leakage current variation, and EOT spread.

With TMA/H₂O chemistry specifically, the oxidation half-cycle already uses water as the oxygen source. Each H₂O pulse is metered to deliver a specific dose, typically calibrated to saturate the TMA-reacted surface and convert the methyl ligands to hydroxyl termination for the next TMA pulse. CDA moisture adds extra water molecules to the loadlock and potentially to the process chamber if the loadlock isolation is not perfect. The extra dose is unquantified and varies from cycle to cycle depending on what the CDA system happened to be doing at the moment of loadlock venting. This adds noise to the oxidation half-cycle. The per-cycle effect may be a fraction of an angstrom. The cumulative effect over a full film deposition shows up as EOT scatter in electrical test. C-V measurements drift. The integration engineer does not know why because the ALD process FDC data looks stable.

Three-tower or four-tower dryer configurations stagger regeneration so at least two towers are always adsorbing. Many older fabs still run twin-tower systems from the original build. Buffer tanks downstream attenuate transients. Oversizing them creates dry down problems after maintenance.

Liquid oil mist removal by coalescer is mature. Gaseous oil vapor relies on activated carbon.

Two points on this. First, the industry shifted from mineral oil to synthetic lubricants in compressors years ago, and carbon bed change-out intervals that were established for mineral oil aromatics may be too long for the lighter fractions in synthetics that have weaker adsorption affinity. Parker Hannifin publishes adsorption isotherm data that shows the difference, buried in technical bulletins.

Compressor

Oil-Free Architecture

Second, oil-free compressors. Oil-free in the compression chamber. Gearboxes and bearings use oil behind shaft seals. Atlas Copco ZR/ZT, Ingersoll Rand Sierra, Kobelco, all the same architecture. After four or five years of 8,000+ hours per year, shaft seals develop micro-leakage. The compressor keeps its oil-free label. The air downstream carries trace oil. A quarterly PID measurement at the outlet, if it happens, may not catch levels below the PID detection threshold but still meaningful for advanced node processes.

Catalytic oxidation modules downstream convert organic vapors to CO₂ and water. Some fabs at 3nm and below have them.

Piping

Electropolished 316L Lines

CDA piping uses electropolished 316L stainless steel, Ra 0.25 to 0.4μm. Electropolishing smooths micro-topography that would otherwise trap water molecules and grows a Cr₂O₃ passivation layer. Swagelok and Valex published surface science data on moisture adsorption versus surface finish back in the 1990s in the context of UHP gas delivery. Electropolished pipe adsorbs three to ten times less moisture per unit area than mechanically finished pipe.

New piping adsorbs water from ambient air during installation. In Southeast Asia or southern Taiwan during summer, ambient dew point is 28°C or higher. Pipe sits open during fit-up, welding, inspection, leak test. Depending on the construction sequence, that exposure can last weeks.

This is worth walking through in detail because it keeps going wrong on projects and the failure mode is always the same. Mechanical contractor installs CDA piping per the construction schedule. CDA piping is one of dozens of piping systems going in simultaneously: nitrogen, process vacuum, PCW, exhaust duct, bulk chemical lines. The pipe fitters are working on all of them. When they finish a section of CDA piping, the open ends sit there waiting for the next section to be welded on. In a conditioned cleanroom or subfab this is manageable. On a project where the CDA piping is being installed in a subfab that does not yet have its own HVAC running, the ambient conditions are whatever the weather outside is. Southern Taiwan in July. 33°C, 80% relative humidity. Open pipe ends breathing that air for days or weeks. Each weld joint gets a borescope inspection, and the scope goes in through the open end, displacing the stagnant gas column inside the pipe with fresh ambient air. The inspector pulls the scope out, the pipe sits there with 28°C dew point air inside it, moisture adsorbing onto the electropolished surface at a rate that depends on the surface temperature and the residence time of the humid air.

Dry down starts when dry gas flow begins. CDA or nitrogen pushes through the pipe, moisture desorbs from the walls into the gas stream, the endpoint dew point analyzer reads it dropping. -20°C, -30°C, -40°C, -50°C over the first day. Then -55°C. Then -58°C. Then it stalls. -58°C for six hours. -59°C. -58.5°C. The desorption rate decreases as the surface moisture loading decreases, and the last few degrees from -60°C to -70°C take disproportionately long because the remaining water molecules are the ones that were sitting in micro-crevices at weld HAZ locations and in the tightest grain boundary features on the pipe surface, and their desorption activation energy is higher than the loosely bound moisture that came off quickly in the first 24 hours.

Minimum dry down time under favorable conditions: three or four days. Under bad conditions: ten days. Ten days on the critical path. The construction schedule says tool move-in starts on a specific date. Equipment vendors have install crews booked months in advance. They do not wait. They go to another project and come back when a slot opens, adding weeks.

Temporary cross-connect piping, isolation valves, nitrogen supply allocation, all of this has to be designed and procured before construction starts. Trying to set it up after CDA dry down has fallen behind does not work. On TSMC Fab 18 phases and on several Samsung Austin and Pyeongtaek builds, the dry down timeline was reportedly met because nitrogen pre-drying was committed early. On projects where it was not planned from the start, it kept getting proposed as an emergency measure during the crunch and kept being too late to execute because the temporary piping fabrication lead time was longer than the remaining schedule float.

Halogen attack on the Cr₂O₃ passivation layer is a separate pipe integrity issue. Fluorine from degrading compressor seal fluoropolymers, HCl or HF from ambient air at the intake. Passivation breach, exposed substrate, continuous particle and metal ion release. Section has to be cut out. Inlet chemical filtration on the compressor intake side handles it, with filter life depending on local ambient air quality.

Weld HAZ destroys EP finish and passivation locally. Thousands of welds in a fab CDA system. Post-weld passivation on main headers is standard practice. Post-weld treatment on every joint costs a lot. A handful of fabs do it.

Precision

Pressure Stability Control

Lithography air bearings need ±0.5 kPa pressure stability. Two screw compressors at slightly different speeds create a beat frequency in the piping. 50.0 Hz and 49.7 Hz gives a 0.3 Hz beat. SCADA transmitters at one sample per second cannot see it. Air bearing servos can. Shows up in overlay or CDU as a periodic signature.

Confirming it requires high-bandwidth dynamic pressure sensors at hundreds of Hz on the piping cross-correlated with stage position data. ASML has issued technical notes on CDA supply requirements covering this. POU regulators with buffer tanks fix it. Regulator Cv has to match the actual flow range. Oversized Cv at low flow puts the plug in a high-gain region that amplifies upstream disturbances.

Straightforward to fix once diagnosed. The diagnosis is the hard part, requiring specialized instrumentation and frequency-domain analysis that is not part of routine CDA monitoring.

CDA does not enter the EUV exposure path. Organic AMC from CDA can contaminate the reticle transfer environment. At 13.5nm wavelength, 92 eV per photon, adsorbed hydrocarbons crack and deposit carbon on reticle surfaces. Some early EUV fabs found their CDA organic removal fell short. Chemical filter modules added at EUV POU.

Distribution

Clean Delivery

Filtration

POU Purification Modules

Treatment

AMC Control

The reticle carbon deposition problem is well documented by ASML and by IMEC through EUV consortia work. The CDA contribution to organic AMC loading is one source among several including outgassing from polymeric components in the tool, cleanroom AMC, and FOUP materials.

CDA moisture before HMDS treatment means water competes with HMDS for surface hydroxyl groups. Incomplete hydrophobic conversion, resist lifting at pattern edges. Gets investigated as an HMDS recipe or chemical issue. Oven temperature, vapor concentration, chemical batch all get checked first. All look fine. The queue time between wet clean and HMDS oven, and the CDA moisture level in the mini-environment during that queue time, are the variables that matter but are rarely examined because they belong to the facility system, not the track tool.

Traditionally nitrogen. Some fabs testing CDA for copper interconnect wafers, reasoning that 21% oxygen provides controlled passivation that a fully oxygen-free nitrogen purge does not. Copper in ultra-dry nitrogen does not form a stable native oxide, and the rapid uncontrolled oxidation when the wafer later exits into ambient can produce a worse surface condition than gradual controlled oxidation. Sensible metallurgy for certain integration schemes. CDA brings contamination risk into a small enclosed volume. Purification requirements for FOUP-grade CDA would need to be very tight. Open question whether the benefit justifies the cost and complexity.

Condensate drains, tank bottoms, filter housings. Stagnant water, trace organics, microbial growth. MVOCs pass particle filters in gas phase. Organic acids cause MIC pitting on pipe internals downstream of all main purification equipment.

Maintenance: drain condensate, replace filter elements, check drain valve function. Drain valves in the subfab are low-priority equipment that can fail in the closed position and sit unnoticed for months, accumulating water that seeds biofilm colonies.

These three topics get one combined section because the problems and fixes are well known. What distinguishes a well-run fab from a poorly-run one in these areas is maintenance discipline, not engineering sophistication.

Offline quarterly sampling catches one point in time. Transients between readings are invisible. Online dew point analyzers and particle counters at critical POU locations feeding FDC data are the minimum for cross-correlation with process excursions.

Online oil vapor monitoring is expensive, drifts, and rarely installed. Most fabs use adsorbent tube sampling with a one-week turnaround from the lab.

Some fabs are attaching POU CDA parameter snapshots to wafer MES records. Straightforward concept, large data infrastructure project, rolling out tool group by tool group.

CDA is 3 to 5 percent of total fab electrical load. Process tools and HVAC dwarf it.

CDA energy efficiency matters for capacity margin during ramp. More efficient system, more usable CDA from the same installed compressors, deferred expansion. Adding compressors in a running fab means subfab construction, vibration risk, electrical upgrades, six or more months procurement lead time.

1 bar pressure reduction saves 6 to 8 percent of compressor power. Heatless regen dryers waste 15 to 20 percent of throughput as purge. HOC dryers recover compression heat and need little or no purge. Oil-free compressor specific power is higher than oil-injected because the chamber runs hotter without oil. Fabs choose oil-free for air quality.