Home Knowledge Base Phosphoric acid etch is the hot, aqueous H₃PO₄ process used to remove silicon nitride selectively from silicon dioxide and silicon.

Hot phosphoric acid etching is a boiling-point-managed selective strip, not merely a heated chemical soak: water activity, acid concentration, thermal recovery, nitride composition, oxide-stop loss, dissolved-silicon loading, isotropic recess, rinse handoff, reflux hardware, and high-temperature containment jointly define the usable process window.

Phosphoric acid etch is the hot, aqueous H₃PO₄ process used to remove silicon nitride selectively from silicon dioxide and silicon. In its classic batch form, concentrated semiconductor-grade acid is heated near its controlled boiling condition, commonly in the neighborhood of 150–170 °C. Water participates in breaking Si–N bonds, while the concentrated acid medium enables soluble reaction products to leave the surface. The useful result is not simply “hot acid”: it is a tightly managed relationship among acid concentration, water content, temperature, nitride film history, dissolved silicon, wafer loading, and exposure time.

The principal integration value is nitride-to-oxide selectivity. Stoichiometric LPCVD silicon nitride is often used as an oxidation mask, polish stop, hard mask, spacer, or protective cap, then stripped while a thin pad oxide or other oxide remains. Hot phosphoric acid can provide strong practical selectivity to dense thermal SiO₂ and crystalline silicon, but the ratio is recipe- and film-specific. Deposited oxides, doped glasses, porous films, oxynitrides, and plasma-damaged surfaces may lose material much faster than dense thermal oxide. A selectivity value from a supplier data sheet is therefore a starting point, not a stack guarantee.

Water is both a reactant and the main concentration-control lever. Heating an approximately 85 wt% incoming acid drives evaporation until the liquid reaches the equipment’s operating concentration and boiling behavior. If water is lost without controlled replacement, acid concentration and boiling temperature shift; if too much water is added, the bath cools and its kinetic state changes. Production tools use combinations of temperature, boiling-point correlation, density or concentration sensing, reflux, vapor management, and metered DI-water replenishment. A temperature reading only represents concentration when pressure, sensor placement, heat input, and boiling state are defined.

Film composition can move the etch rate by multiples. Dense, near-stoichiometric LPCVD Si₃N₄ generally behaves differently from hydrogen-rich PECVD SiNₓ. Silicon-rich, nitrogen-rich, oxynitride, low-temperature, UV-cured, and annealed films each present different bond populations and density. Stress engineering can also correlate with composition and microstructure. Before transferring a time between modules or products, measure the actual deposited film after its full thermal and plasma history—not a nominal “nitride” witness from another flow.

The profile is isotropic, so exposed nitride recedes vertically and laterally. When nitride lies beneath a masking film or beside an oxide feature, the acid enters the opening and produces lateral recess or undercut. This is useful when releasing a nitride feature and dangerous when the residual nitride width, spacer width, cap overlap, or pad-oxide protection is critical. The final dimension depends on starting thickness, vertical removal, lateral access, local wetting, over-etch, and whether a seam or damaged interface creates a fast path.

Common applications all impose different stop-layer budgets. LOCOS and some isolation flows strip the oxidation-mask nitride while preserving pad oxide. STI-related modules may remove a nitride polish stop after CMP while protecting trench-fill oxide and corner geometry. MEMS and sensor processes may clear a structural or sacrificial nitride adjacent to delicate oxide, silicon, metal, or cavity surfaces. Spacer and hard-mask removal can expose complex sidewalls where a nominally selective liquid reaches liners or interfaces that blanket-film tests never exercised.

Integration surfaceRelative behavior in hot H₃PO₄Main qualification questionTypical role
Stoichiometric LPCVD Si₃N₄intended removal film; comparatively controlledrate after the real thermal historyoxidation mask, CMP stop, hard mask
PECVD SiNₓ / Si-rich nitriderate can differ substantiallycomposition, hydrogen, stress, and damagecap, passivation, spacer, liner
Dense thermal SiO₂usually a strong stop relative to nitrideoxide-loss budget through over-etchpad oxide or protected dielectric
Deposited or doped oxideoften less resistant than thermal oxidedensification and dopant dependenceSTI fill, ILD, sacrificial oxide
Crystalline silicongenerally retained in the qualified windowsurface condition and exposed junction risksubstrate or device surface
Metals and barrier filmscompatibility is material-specificcorrosion, galvanic coupling, adhesioncontacts, heaters, sensors, routing

Rate control starts with a real thermal state. The wafer and cassette must enter a bath whose temperature and concentration have stabilized, and timing must be referenced to a defined immersion event. A large cold load temporarily changes bath temperature. Heat loss differs between one wafer and a full cassette, and wafer spacing changes convection and product removal. Ramp recovery, lot size, dummy-wafer policy, cassette material, agitation, and immersion orientation should all be fixed in the process specification.

Dissolved silicon creates both aging and particle risk. Nitride removal loads silicon-containing reaction products into the bath. As concentration rises, the chemistry can drift and dissolved material may nucleate or precipitate, especially during local cooling, idle periods, plumbing transitions, or uncontrolled dilution. Feed-and-bleed, lot-count limits, silicon-load accounting, filtration, controlled standby, and complete bath replacement are different strategies for managing the same state. Filter pressure drop alone cannot prove that the liquid remains chemically healthy.

Wetting determines whether every patterned region starts together. Hydrophobic organic residue, trapped air, dense hole arrays, deep cavities, backside films, and wafer-to-cassette contact can delay liquid access. A qualified pre-wet, controlled entry angle, gentle cassette motion, or single-wafer dispense can reduce bubbles and boundary-layer variation. Surfactants should not be introduced casually because they can alter wetting, contamination, downstream rinse behavior, and exhaust loading. Fragile structures also constrain agitation and wafer motion.

Endpoint is normally a calibrated thickness-and-time decision. Blanket monitor wafers establish nitride and oxide rates using ellipsometry, reflectometry, step height, or other film metrology. Patterned cross sections establish lateral recess and residual stop-layer thickness. The production time combines nominal nitride clearing with an over-etch sized for incoming thickness, rate variation, loading, and worst-case pattern access. A visually clear surface or fixed legacy time is not an adequate endpoint when a few nanometers of pad oxide or spacer width matter.

Rinse and cool-down stop the process. A wafer leaving hot acid carries a reactive liquid film, so lift speed, drain time, transfer delay, and first-rinse flow contribute to total exposure. The rinse must dilute acid and remove soluble products without creating a thermal shock, watermark, particle redeposition, or trapped residue. Multiple overflow or quick-dump stages may be required for batch loads. Drying must match the structure: spin or displacement drying can be appropriate for robust wafers, while capillary-sensitive MEMS features may need a specialized low-stiction sequence.

Failure signatures should be separated by mechanism. Residual nitride islands point toward poor wetting, insufficient time, dense film, low temperature, or local loading. Excess oxide loss can indicate the wrong oxide type, an over-concentrated or contaminated bath, excessive time, plasma-damaged stop film, or an unrecognized exposed edge. Across-cassette gradients implicate thermal recovery, circulation, spacing, or replenishment. Particles implicate silicon saturation, cold spots, filters, plumbing, tank films, or rinse redeposition. Lateral CD loss is an isotropic geometry problem, not automatically a chemistry-rate problem.

Equipment must be designed as a hot-acid system, not a generic wet tank. Wetted vessels, heaters, temperature probes, level sensors, pumps, filters, valves, plumbing, lids, and cassettes require material and lifetime qualification for concentrated H₃PO₄ at operating temperature. The design must avoid localized overheating and dry-heater exposure, control condensate and vapor, and keep replenishment water from flashing or splashing. Local exhaust, interlocked heat and level control, secondary containment, leak detection, compatible drains, and site-specific chemical response procedures are part of the process boundary.

A production control plan connects bath state to wafer evidence. Record chemistry lot, make-up and replenishment volumes, temperature trajectory, concentration proxy, boil-state time, wafer count, exposed nitride area, dissolved-silicon estimate, filter condition, and bath age. Correlate those signals with nitride rate, oxide loss, selectivity, within-wafer and wafer-to-wafer uniformity, recess, particles, metals, residue, and downstream electrical or mechanical performance. Control charts should distinguish a bath-state excursion from a deposition-film excursion because both can present as an etch-rate shift.

The transferable recipe is a removal distribution, not a temperature and timer. It defines the exact nitride and stop films, patterned access, bath-management method, load size, thermal recovery rule, immersion and withdrawal events, over-etch basis, rinse handoff, and metrology sampling. With those boundaries, hot phosphoric acid becomes a precise selective strip. Without them, its apparent simplicity hides concentration feedback, film dependence, lateral loss, and accumulating bath history.

Hot Phosphoric Acid — Selectivity Depends on Bath State Water balance controls temperature and rate while Si₃N₄ recedes isotropically above a thin oxide stop BATH-STATE FEEDBACK H₃PO₄ + H₂O hot selective nitride strip evaporation DI replenish temperature water ratio Si loading ISOTROPIC NITRIDE REMOVAL hot acid reaches the top surface and exposed sidewalls thin SiO₂ stop layer silicon substrate retained Si₃N₄Si₃N₄ lateralrecess clear nitride while budgeting oxide loss blanket rate alone does not predict patterned undercut CONTROL CHAIN FILM HISTORYLPCVD · PECVD · stress THERMAL STATEconcentration · recovery LOAD + AGEarea · dissolved silicon TIME + RINSEover-etch · reaction stop measure both films QUALIFIED OUTPUT = NITRIDE CLEAR + OXIDE LOSS + LATERAL RECESS + PARTICLE STATE ellipsometrynitride + oxide rate cross-section CDundercut + residue bath historywater + Si load particles + metalscontamination evidence post-rinse surfaceresidue + stop integrity A temperature and timer do not define the process; film state, water balance, loading, and metrology do.

Following hot phosphoric etch from water balance and nitride bond structure through isotropic recess, silicon loading, rinse timing, and stop-layer metrology is the kind of chemistry-to-integration connection Chip Foundry Services makes explicit—turning a familiar wet-bench step into a process window that equipment, integration, and yield teams can share.

Ready=>start: Stabilized hot-phosphoric module
State=>condition: Temperature, boil state, water balance, loading, and exhaust pass?
Load=>operation: Prewet and immerse qualified cassette
Recover=>operation: Recover thermal state; start defined exposure clock
Etch=>operation: Strip nitride with controlled reflux and replenishment
Budget=>condition: Nitride clear within oxide, recess, and bath limits?
Transfer=>operation: Controlled withdrawal and hot-to-cool handoff
Rinse=>operation: Staged compatible rinse to residue endpoint
Verify=>condition: Nitride, oxide loss, particles, residue, and CD pass?
Release=>end: Release lot; update silicon-load model
Hold=>end: Hold lot; contain and investigate
Ready->State
State(yes)->Load->Recover->Etch->Budget
State(no)->Hold
Budget(yes)->Transfer->Rinse->Verify
Budget(no)->Hold
Verify(yes)->Release
Verify(no)->Hold

Read hot phosphoric acid etching through a water-activity, film-selectivity, silicon-loading, and thermal-module control lens rather than a boiling acid nitride-strip timer lens.


Water Activity, Boiling Point, and Reflux Control

Water is chemically and operationally central. Incoming semiconductor-grade phosphoric acid is often near 85 wt%, but heating drives water loss and raises concentration until the tool reaches its operating boil state. Water participates in hydrolysis of Si–N bonds, so evaporation without replacement changes both thermal behavior and reaction chemistry. Adding too much water cools the bath and changes concentration; adding it too quickly can flash, spatter, or create a local excursion.

At fixed ambient pressure, boiling temperature can serve as a concentration proxy only after heater power, reflux return, vapor loss, bath level, sensor placement, and dissolved load are defined. A probe near a heater may read 165 °C while cassette channels remain cooler. Reflux condensers return some evaporated water; metered DI makeup replaces net loss; lids and exhaust determine vapor capture. Pressure and exhaust changes can shift boiling behavior without any deliberate recipe change.

Hot H₃PO₄ water balance: evaporation, reflux, and makeup close the loopTemperature is a useful proxy only while pressure, heat input, level, and vapor return are controlled.HOT ACID BATHCONDENSERreflux water returnmetered DI makeupnet vapor to exhaustWater balance = makeup + reflux − net evaporation − drag-out + drag-inVerify with temperature trajectory, concentration proxy, level, and wafer rate. An Arrhenius relation $R=Ae^{-E_a/k_BT}$ describes kinetic sensitivity over a bounded region, but boiling chemistry couples $T$ and concentration, so a temperature split alone may not isolate activation energy. Production characterization varies water makeup or analytical concentration independently where safe, maps rate after full thermal stabilization, and measures cold-load recovery for minimum and maximum cassette sizes. The recipe must define startup concentration, warm-up and equilibration, stable-boil criteria, makeup logic, reflux cooling, allowable temperature band, bath level, exhaust state, idle behavior, and restart after interruption. “165 °C for 30 min” omits nearly every state variable that controls whether those 30 min are reproducible. ## Nitride Film State and Nitride-to-Oxide Selectivity Stoichiometric LPCVD $Si_3N_4$ is not interchangeable with PECVD $SiN_x$. Silicon-rich, nitrogen-rich, hydrogen-rich, low-temperature, UV-cured, plasma-damaged, implanted, and annealed films differ in density, bond population, stress, and water access. Rate may shift by multiples even when ellipsometric thickness and refractive index appear acceptable. Product qualification uses the film after its complete upstream history. The stop is equally specific. Dense thermal oxide often survives far better than deposited or doped oxide. TEOS, PECVD oxide, PSG, BPSG, flowable oxide, porous low-k, oxynitride, and plasma-damaged interfaces require separate rates. Practical selectivity is $S_{N:O}=R_N/R_O$, but integration cares about absolute oxide loss during clear and overetch. A 100:1 ratio still loses 5 nm while removing 500 nm of nitride before margins. Selectivity matrix: both the nitride and oxide identities matterQualify absolute removal on the actual films; generic ratios conceal stop-layer risk.dense thermal oxidedeposited oxidedamaged / doped oxideLPCVD Si₃N₄PECVD SiNₓSi-rich / damagedstrongest windowmeasure oxide losshigh riskfilm-specificfilm-specifichigh riskrate can shiftnarrow windowavoid assumptionsRelease matrix = nitride rate + absolute oxide loss + surface conditionRepeat after deposition, anneal, CMP, implant, or plasma changes.

The matrix also includes crystalline silicon, metals, liners, exposed junctions, backside films, and bevels. Hot acid can reach seams and interfaces that blanket monitors never expose. Cross-section TEM or SEM confirms stop integrity and lateral recess; ellipsometry measures blanket film loss; FTIR and refractive index help correlate nitride composition; wafer curvature connects stress to film state.

Named applications impose different constraints. LOCOS nitride strip preserves thin pad oxide. STI polish-stop removal protects trench-fill oxide and corners after CMP. Spacer or hard-mask strip protects sidewalls and junctions. MEMS processing may expose cavities, metals, and fragile structures. One qualified time cannot be copied among them without recalculating every material budget.

Isotropic Recess and Endpoint-by-Film Budget

Hot phosphoric acid removes accessible nitride laterally as well as vertically. For an approximately isotropic front, lateral recess $U$ is on the order of $R_{lat}t$ at each exposed edge. Clearing 200 nm of nitride with 25 percent overetch may create roughly 250 nm lateral recession per side if lateral and vertical rates are comparable. Fast interfaces, seams, stress gradients, or transport confinement can make patterned behavior depart from blanket rate.

Endpoint is normally timed from measured distributions. Required nitride removal is $h_N(1+N)(1+O)$, where $N$ covers incoming and process nonuniformity and $O$ is overetch. Oxide loss is that effective removal divided by selectivity, with additional exposure after local nitride clears. Mask width, spacer CD, cap overlap, pad oxide, corner geometry, and exposed-surface state must all remain positive at the fast site.

Clear-to-stop budget: slow nitride versus fast-site collateral lossTime must clear the thick, slow location without consuming oxide or lateral CD elsewhere.Si₃N₄ target filmthin oxide stopsilicon / integrated stacklateral recess Uslow site clearsoxide + CD lossEndpoint is the overlap of clear, oxide, CD, and surface budgets. Patterned test structures include isolated and dense features, multiple widths, long interfaces, corners, and all exposed stop materials. Blanket monitors alone cannot see seam attack or lateral recession. Cross sections anchor layout bias. Statistical release uses worst-site residual nitride, maximum oxide loss, maximum recess, and post-rinse residue—not mean etch rate. ## Dissolved Silicon, Bath Age, and Particle Formation Every stripped nitride wafer adds silicon-containing products. Bath state changes with total exposed nitride area and thickness, not lot count alone. A 25-wafer blanket cassette can load far more silicon than many patterned lots. Feed-and-bleed replaces active chemistry but only partially controls product concentration; filtration captures particles but cannot remove dissolved silicon. Precipitation risk rises when chemistry crosses solubility limits or sees a local cold spot, abrupt dilution, stagnant plumbing, idle cooling, or incompatible carryover. Particles may nucleate in the bath, filter housing, return line, lid condensate, cassette, or rinse. Differential pressure is evidence of captured load, not proof of chemical health. A mass balance and wafer particle data are both required. Silicon-loading lifecycle: dissolved product can become a particle excursionTrack material removed, thermal history, dilution events, and non-replenishable bath limits.NITRIDE REMOVALarea × thicknessDISSOLVED Siproduct inventorySUPERSATURATIONcold / dilute / idlePARTICLESwafer addersControllableexposed-area accountingfeed / bleed and full dumptemperature + idle managementcompatible dilution sequencefilter at qualified flowEvidencebath silicon analysisfilter ΔP at fixed T and flowliquid particle counterKLA wafer addersSEM/EDS residue identityDump limits protect chemistry that replenishment and filtration cannot restore.

Useful controls include estimated grams of nitride removed, analytical dissolved silicon, bath hours at temperature, idle cycles, replenishment volume, bleed volume, filter differential pressure at normalized viscosity and flow, liquid particle count, and wafer adders. A bath can pass temperature and rate while failing particles; a clean-looking bath can exceed its dissolved-load limit.

Root cause follows spatial evidence. Random adders implicate suspended particles or rinse redeposition. Slot trends implicate flow and cold-load recovery. Edge rings implicate condensate, bevel, or handling. Chemistry found on dried wafers implicates rinse and dry. SEM/EDS, ion chromatography, ICP-MS, and tank inspections distinguish silicon-rich precipitation from incoming metal or polymer debris.

Thermal Hardware, Withdrawal, Rinse, and Dry

A hot-acid tool must avoid heater dry-fire, local overheating, sensor bias, uncontrolled condensate, and water flashing. Wetted vessels, heaters, probes, level sensors, filters, valves, plumbing, lids, cassettes, seals, and drains require lifetime qualification at temperature. Quartz may be used in some architectures; fluoropolymers and ceramics have temperature and mechanical constraints. Every material must be assessed in fresh and silicon-loaded acid.

Withdrawal begins the stopping sequence. A hot reactive film remains on the wafer while it drains and moves toward rinse. Lift speed affects drag-out and exposure. Directly shocking a hot cassette with cold water can stress wafers or hardware and disturb precipitate. A staged, compatible handoff dilutes acid, controls temperature, prevents redeposition, and reaches an ionic residue endpoint.

The process ends after controlled cool, rinse, and dryWithdrawal time and thermal handoff contribute to total nitride and oxide exposure.HOT ETCHdefined clearWITHDRAWcarryover filmSTAGED COOLno thermal shockRINSEionic endpointDRYwafer temperaturereactive carryover concentrationQualify total exposure from immersion through first effective quench. Rinse monitoring may include time, flow, outlet conductivity, temperature, ion chromatography, particle adders, and wafer residue. Robust wafers may use spin or IPA-assisted dry; capillary-sensitive MEMS can need a low-stiction route. A visually dry wafer is not proof of phosphate or silicon-product removal. Hardware fault tests include loss of level, heater over-temperature, sensor disagreement, reflux cooling loss, exhaust trip, makeup valve stuck open, circulation loss, drain blockage, robot interruption with wafers immersed, power failure, and restart after an unknown idle. The safe response must isolate heat and delivery, contain chemistry, define wafer disposition, and prevent automated restart into an unknown bath state. ## Metrology, SPC, Safety, and Production Release The control plan separates deposition-film drift from bath drift. Co-process known nitride and oxide references; map multiple sites and cassette slots; correlate rate to temperature trajectory, makeup, silicon load, and exposed area. Ellipsometry or reflectometry measures blanket films, profilometry checks steps, cross-section SEM/TEM measures recess and stop integrity, KLA maps particles, TXRF or ICP-MS checks metals, and ion chromatography identifies rinse residue. An illustrative window might stabilize at 160 °C within ±0.3 °C, remove LPCVD nitride at 5 nm/min, limit thermal-oxide loss to 0.1 nm/min, demonstrate at least 50:1 selectivity, clear a 200 nm film with a 40 nm overetch allowance, retain at least 15 nm of pad oxide, keep recess error within 25 nm, recover a full cassette within 3 min, transfer within 10 s, use a 0.2 µm-rated filter, alarm on a 2 °C transient, and hold particle adders below 0.05 cm⁻². These are examples, not universal recipes. Production release: intersect four independent gatesA correct nitride rate cannot compensate for oxide loss, particles, or unsafe hot-acid containment.FILM / PROFILEnitride clear + uniformityoxide loss + lateral recesssurface and downstream resultPASS: integration budget closesBATH STATEwater + temperature trajectorysilicon load + bath agereflux + replenishmentPASS: chemistry state closesCONTAMINATIONparticles + bath countsmetals + ionic residuefilter + hardware conditionPASS: purity closesEHS / CONTAINMENTheat + level + exhaust interlockscompatible containment and drainmaintenance + emergency responsePASS: module may operateRelease only where all four gates overlap.

Concentrated hot phosphoric acid causes severe thermal and chemical burns. Operation requires chemistry-specific PPE, local exhaust, covered and interlocked equipment, leak detection, secondary containment, compatible drains, controlled water addition, trained maintenance, and site-approved emergency procedures. Water must never be added ad hoc to a hot bath. Only trained personnel following facility procedures may operate or service the module.

Equipment from SCREEN, Tokyo Electron, Lam Research, and Applied Materials can differ in vessel, reflux, circulation, dosing, and wafer-handling architecture. Transfer by matching a displayed temperature and time is unsafe and technically incomplete. Intel, TSMC, Samsung, SK hynix, and Micron may use proprietary windows, but all must close water balance, thermal recovery, film-specific selectivity, silicon loading, contamination, rinse, and containment.

The transferable recipe is the entire state machine: exact films and layouts, analytical chemistry, stable-boil criteria, water makeup and reflux, load size, exposed area, thermal recovery, immersion clock, overetch calculation, bath-life limit, withdrawal, rinse, dry, metrology, SPC, fault response, maintenance, and EHS approval. That is what turns hot phosphoric acid into a controlled selective nitride strip.

phosphoric acid etchetchhot phosphoric acidphosphoric nitride stripsilicon nitride wet etchh3po4 etchhot acid nitride etch

Explore 500+ Semiconductor & AI Topics

From EUV lithography to CUDA optimization — search the full knowledge base or chat with our AI assistant.