In-situ CVD doping introduces a dopant precursor while the host film grows, but the commanded gas ratio is only the first step toward an electrically active profile. Molecules must be delivered and activated; dopant-bearing fragments must adsorb, survive surface competition, incorporate into the growing solid, occupy useful lattice or network sites, avoid segregation, desorption, and clustering, and remain active after every later thermal step. Delivered dose, incorporated dopant, substitutional dopant, and electrically active carriers are different quantities.

The central advantage is profile formation during growth. In-situ doping can create uniformly doped layers, abrupt steps, graded profiles, delta-like regions, and conformally doped three-dimensional films without implantation shadowing or a separate diffusion source. It can reduce implant damage and process steps. The cost is that dopant chemistry becomes inseparable from deposition rate, microstructure, selectivity, composition, stress, and chamber memory.

Choose the doping route from geometry and final electrical function. A thick blanket epilayer may tolerate gradual gas transients but require low compensation and high activation. A source/drain epi structure demands high carrier concentration, strain, facets, selectivity, contact resistivity, and abrupt junction placement. Doped polysilicon cares about grain boundaries and later activation. An ultrathin passivation or contact layer may be dominated by interface damage and dopant carryover.

Dopant precursor concentration is not solid concentration. Gas-manifold dilution, mass-flow accuracy, source purity, line pressure, residence, wall adsorption, decomposition, surface coverage, competing host precursor, and growth rate intervene. A gas ratio can correlate with incorporation inside one qualified window, but it is not a universal transfer function.

Hydride precursors are common and hazardous. Phosphine supplies phosphorus, arsine supplies arsenic, and diborane supplies boron in many silicon-based CVD processes; other hydrides, chlorides, metal-organics, plasma species, or elemental beams serve different materials. Precursor choice changes activation, sticking, hydrogen/halogen chemistry, temperature range, contamination, delivery, and exhaust burden.

Dopant gas changes the host growth rate. PH₃ can inhibit some silicon surface reactions; B₂H₆ can alter nucleation and morphology; carbon sources compete for substitutional incorporation; dopants change surface hydrogen and reconstruction. The direction and strength depend on precursor set, temperature, pressure, phase, and concentration. Correcting thickness by time does not restore the original material.

Growth-rate suppression feeds back into concentration. If dopant flow reduces host deposition rate while dopant arrival remains high, the incorporated atomic fraction can rise nonlinearly. A small gas change can therefore move both numerator and denominator. Always measure growth rate and composition together when building a doping calibration.

Temperature controls incorporation, desorption, segregation, and activation differently. Higher temperature may improve crystal quality and substitutional incorporation but increase dopant surface segregation, evaporation, or diffusion. Lower temperature may confine the profile while trapping dopant in inactive sites or degrading epitaxy. The optimum is a material/process compromise, not simply maximum incorporation.

Surface segregation broadens an intended transition. Dopant can prefer the growth surface over the bulk and ride the advancing interface through subsequent nominally undoped layers. A shutoff at the gas valve then produces a trailing chemical profile. Segregation depends on dopant, host composition, orientation, temperature, growth rate, surfactants, surface reconstruction, and cap strategy.

A locking or encapsulation layer trades movement against crystal quality. Low-temperature or fast initial overgrowth can bury a surface dopant before it segregates, followed by higher-temperature growth or anneal. If the locking layer is too cold or too fast, it creates defects and roughness; if too warm or slow, dopant redistributes. Optimize confinement, epitaxial quality, and electrical activation together.

Desorption creates an upper-temperature or low-coverage loss path. Dopant-bearing species or elemental dopant can leave the surface before incorporation. This makes solid concentration fall despite constant gas or beam flux and can produce strong temperature sensitivity. Source-accounting experiments and SIMS through designed growth sequences separate incorporation, segregation, and evaporation.

Solid solubility is not the same as active-solubility under a real process. Total incorporated concentration can exceed the equilibrium electrically active fraction, particularly in metastable low-temperature growth. Clusters, precipitates, complexes, interstitial sites, grain boundaries, and compensation reduce free carriers. A later anneal may activate dopant, deactivate it through clustering, or redistribute it.

Substitutional fraction matters for both carriers and strain. Phosphorus or boron on appropriate lattice sites can contribute carriers; substitutional carbon in silicon contracts the lattice and creates tensile strain. Interstitial carbon or clustered dopant adds chemical concentration without the intended strain/electrical benefit. XRD/strain, channeling or atomistic methods, and electrical data complement SIMS.

Activation is a measured ratio, not a recipe label. Compare chemically measured dopant with carrier concentration or conductivity while correcting for mobility, thickness, compensation, contacts, and dimensional confinement. Hall analysis, electrochemical capacitance–voltage, spreading resistance, four-point probe, and device extraction see different aspects and carry different assumptions.

Mobility falls as impurity scattering rises. Higher active concentration can lower resistivity until mobility degradation, saturation, clustering, or band-structure effects dominate. The lowest sheet resistance does not necessarily correspond to the highest activation fraction or best junction. Interpret carriers, mobility, thickness, and contact resistance separately.

Compensation hides behind total dose. Background or intentional acceptors counter donors and vice versa; carbon, oxygen, hydrogen, defects, and traps alter carrier response. A net carrier concentration can be far below the sum of incorporated dopants. SIMS for all relevant species and temperature-dependent electrical measurements help distinguish compensation from poor activation.

Polycrystalline material adds grain-boundary segregation. Dopant can accumulate at boundaries, change grain growth, reduce or raise boundary barriers, and diffuse rapidly along disordered paths. Doped polysilicon sheet resistance depends on grain size, boundary trapping, activation, and anneal in addition to total dose. The dedicated polysilicon-doping page should own its application-specific window.

Amorphous material adds defect-state and crystallization coupling. Dopant shifts Fermi level and defect occupation, changes plasma/surface chemistry, and may alter later solid-phase crystallization. Anneal moves dopant while grains nucleate and grow. A uniform as-deposited SIMS profile can become a nonuniform electrically active profile after crystallization.

Epitaxial material adds crystal-quality constraints. High dopant flux can roughen surfaces, change step flow, generate stacking faults, alter facets, or reduce selectivity. In SiGe or Si:C, host alloy composition, strain, dopant, and etchant response are coupled. The final useful window is set by defects, activation, shape, and contact performance, not dose alone.

Compound semiconductors have amphoteric and site-occupancy complications. A species may act as donor or acceptor depending on the sublattice or growth condition; hydrogen can passivate dopants; vacancies and compensation defects form to maintain charge balance. V/III ratio, chemical potential, polarity, and anneal influence activation. Do not transfer silicon precursor logic directly.

A delta-doped layer is never an ideal mathematical sheet. Gas/beam rise and fall, chamber residence, surface segregation, diffusion, roughness, SIMS depth resolution, and later anneals broaden the measured profile. Specify integrated sheet dose, peak concentration, full width under a stated deconvolution model, activation, and interface quality.

Abrupt profile transitions start in the delivery system. Valve response, MFC settling, manifold and line volume, pressure controller, injector residence, purge efficiency, and source adsorption create a time-domain step response. At a known growth rate, that response maps into depth. Measure delivered transients rather than assuming an instantaneous command.

The reactor adds another memory reservoir. Dopant adsorbs on chamber walls, showerhead, susceptor, quartz, pump surfaces, and deposited coatings, then desorbs during later nominally intrinsic growth. Memory depends on exposure, temperature, carrier gas, halogen, clean state, wall material, and time. It can persist after gas analysis at the inlet looks clean.

Carryover is especially damaging at low intended background. A trace donor tail after a heavily n-doped layer may dominate an intrinsic spacer or p-type layer. The same absolute contamination can be irrelevant inside a heavily doped cap. Acceptance limits should be based on the most sensitive downstream layer, not a generic chamber baseline.

Recipe order creates asymmetric contamination. A high-dose boron run followed by phosphorus or intrinsic growth need not behave like the reverse order because adsorption and desorption differ. Sequence matrices, dedicated chambers or kits, sacrificial depositions, extended purge, wall cleans, and seasoning are potential controls. Validate the chosen production order.

Purge is a kinetic process, not a fixed waiting period. Dead volumes, valve cavities, showerhead plenums, adsorbed species, pressure, carrier flow, and temperature create multiple time constants. Outlet residual-gas evidence helps with gas-phase clearance, but solid-surface memory still requires wafer-based SIMS or electrical monitors.

A chamber clean can reveal or create memory. Removing host film may expose dopant-rich interfaces or hardware; fluorine/chlorine clean can mobilize residues; overclean changes wall state and particles. Post-clean seasoning should establish both host-film behavior and background dopant before product.

Backside and bevel sources matter. Heavily doped substrate, backside films, prior implants, or bevel deposits can outgas or transfer dopant at temperature. They also change emissivity and wafer temperature. Back-seal layers, edge exclusion, wafer handling, and susceptor contact should be part of contamination control.

Autodoping differs from intentional co-flow. Dopant evaporated from the substrate or buried layers enters the boundary layer and reincorporates, often producing radial, front/back, or time-dependent profiles. Lower temperature, pressure/flow changes, backside sealing, reactor design, and encapsulation can reduce it. Gas-line calibration cannot diagnose autodoping alone.

Pattern loading changes dopant concentration. Selective or feature-scale growth consumes host precursor differently across open and dense layouts. If dopant supply or incorporation scales differently, solid concentration and activation vary with pattern density. Map composition, SIMS where feasible, strain, Rs/contact resistance, and geometry across pattern contexts.

Conformal deposition does not guarantee conformal active doping. In high-aspect-ratio features, host and dopant precursors have different sticking, diffusion, and depletion. A constant film thickness can contain a depth-dependent dopant fraction; later activation and grain structure add more variation. Use depth-resolved chemical and electrical proxies on representative structures.

Plasma activation changes dose and damage together. In PECVD or plasma doping, power, frequency, bias, pressure, and gas ratio determine radicals and ions. More dissociation can improve incorporation but increase substrate damage, charging, sputtering, hydrogen, or defect density. Delivered ion energy and radical flux should be related to functional activation.

Hydrogen can passivate dopants and defects. During PECVD or later hydrogenation, dopant–hydrogen complexes reduce electrical activity even as interface passivation improves. Annealing may dissociate complexes and reactivate carriers, but also drives diffusion and defect evolution. FTIR, effusion, SIMS, and electrical recovery together clarify the trade.

Oxygen and carbon contamination alter activation. Moisture, leaks, source impurities, wall memory, and substrate outgassing introduce species that form complexes or precipitates. Carbon may be an intentional strain dopant in one process and an unwanted electrically inactive contaminant in another. Specify relevant impurities and functional limits.

SIMS measures chemistry with convolution. Primary-ion species and energy, matrix effects, ion yield, crater roughness, knock-on, sputter mixing, calibration standard, and detection limit shape a profile. Abrupt interfaces and high concentrations are especially prone to artifacts. Report raw conditions and use deconvolution cautiously.

SIMS concentration should close a mass balance. Compare integrated dopant dose in the solid with delivered exposure, expected incorporation, surface reservoir, desorption, and exhaust where possible. An unexplained missing fraction may indicate wall deposition, evaporation, or calibration error. Mass accounting helps separate chemistry from metrology.

Hall measurements separate carrier density and mobility under assumptions. Parallel conduction, multilayers, degenerate statistics, contact geometry, magnetic field, temperature, and nonuniform profiles complicate extraction. A sheet carrier density compared with integrated SIMS dose gives an activation estimate only if the same conducting volume is sampled.

Four-point-probe sheet resistance is fast but ambiguous. Thickness, carrier density, mobility, surface depletion, parallel substrate conduction, and edge effects all contribute. Pair Rs with thickness and periodic carrier/profile data. A time correction that recovers Rs can mask simultaneous thickness and activation drift.

Spreading resistance and ECV provide depth-sensitive electrical profiles with limits. Contact mechanics, calibration, carrier mobility, depletion, surface preparation, and resolution affect the result. Compare against SIMS and device junction behavior rather than treating any one profile as absolute truth.

XRD and Raman can reveal dopant/alloy strain, not dose directly. Substitutional dopants and alloy composition alter lattice spacing; relaxation, stress, temperature, and defects also shift peaks. Use these methods to constrain substitutional fraction or strain when anchored by composition and elasticity models.

TEM and atom probe answer local questions. They can locate clusters, defects, interfaces, and three-dimensional composition at high resolution, but sample tiny volumes and face preparation/reconstruction artifacts. Use them to identify mechanisms behind wafer-scale SIMS, XRD, defect, and electrical trends.

Activation must be checked after the full thermal history. Implant anneal, oxidation, silicidation, contact firing, dielectric cure, and packaging can diffuse, cluster, deactivate, or reactivate dopant. Oxidizing interfaces can drive segregation. Qualify final profiles and carriers after representative downstream steps, not just as grown.

Oxidation can redistribute dopant strongly. Moving Si/SiO₂ or SiGe/oxide interfaces reject or collect dopant according to segregation and diffusion behavior. Pileup changes interface electric fields and contact/passivation response. A stable profile in inert anneal does not guarantee stability during oxidation.

Silicidation and contact formation consume the doped layer. Dopant redistributes between silicide, semiconductor, and interface; high concentration changes phase formation and contact resistance. Epi shape, surface oxide, defects, and active concentration control the result. Contact resistivity is a system output, not a dopant concentration proxy.

Junction abruptness is chemical and electrical. A sharp SIMS step can still have a broad depletion or active-carrier transition because of compensation, band tails, defects, and activation. Conversely, SIMS mixing can make a sharp junction look broad. Use electrical structures, capacitance, leakage, or device response alongside chemistry.

Variability grows as doped volume shrinks. In nanoscale structures, discrete dopant number, placement, clustering, surfaces, and quantum confinement create device-to-device variation. Average concentration loses meaning when only a small number of active atoms contribute. Statistical electrical data and atomistic-informed models become necessary.

A calibration curve must state the process state. Relate solid concentration and active carriers to dopant precursor flow or partial pressure at fixed host flow, temperature, pressure, growth rate, wafer loading, pattern density, chamber age, and thermal history. If any of these changes, recalibrate or demonstrate invariance.

Designed experiments should target interactions. Sweep dopant/host ratio with temperature and growth rate; dopant steps with purge and wall state; concentration with anneal; pattern density with composition and Rs; alloy fraction with dopant incorporation; and selectivity/shape with dopant phase. These interactions are the mechanism, not experimental nuisance.

Deliberate transition structures are efficient monitors. Grow repeated dopant on/off steps, dose ladders, intrinsic spacers, reversed p/n order, and thickness markers in one wafer. SIMS and electrical analysis then separate valve response, residence, segregation, diffusion, and memory. Include fresh, seasoned, and post-high-dose chamber states.

Tool matching compares incorporation surfaces, not flow labels. Match growth rate, total and active dopant, transition widths/tails, background memory, alloy/strain, morphology, defects, selectivity, particles, and device metrics across dose, temperature, loading, and chamber age. Identical MFC set points do not ensure identical delivered or incorporated dose.

Production monitoring combines leading and lagging signals. Leading inputs include source concentration/purity, MFC and valve response, pressure, carrier and host flow, wafer temperature, recipe order, wall exposure, clean/season state, exhaust conductance, and pattern mix. Lagging outputs include rate/thickness, SIMS monitors, Rs/Hall, composition/strain, particles, defects, and contact/device data.

Safety is inseparable from dopant delivery. Phosphine, arsine, diborane, germane, silane, hydrogen, metal-organics, halogens, and cleaning gases can be acutely toxic, pyrophoric, corrosive, or flammable. Gas cabinets, compatible delivery, detection, purge, ventilation, abatement, interlocks, cylinder-change controls, emergency planning, and current SDS/site procedures are mandatory.

Abatement and maintenance contain dopant inventory. Walls, forelines, pumps, traps, and scrubbers accumulate toxic/reactive arsenic-, phosphorus-, boron-, silicon-, germanium-, and halogen-containing residues. Conductance changes affect the process while maintenance disturbs deposits. Track exposure/mass and use residue-specific safe cleanout.

The honest specification separates delivered, incorporated, substitutional, and active dopant. State film/material phase, total chemical profile and integrated dose, active-carrier profile, mobility or resistivity, compensation, transition tails, background memory, microstructure/crystal defects, pattern dependence, and post-thermal stability. One “concentration” cannot stand for all of them.

Production-worthy in-situ doping creates the intended active profile without losing the host film. It controls gas delivery and chamber memory, preserves growth rate, phase, composition, selectivity, morphology and defects, bounds segregation and carryover, survives downstream thermal/contact processing, and produces repeatable device function across wafer, pattern, chamber age, and recipe sequence.

In silicon epitaxy, the useful upper concentration is set by a coupled metastability window. Nonequilibrium growth can place phosphorus above equilibrium expectations and generate tensile strain through substitutional incorporation, but vacancies, clusters, defects, and later redistribution can reduce carriers. Post-growth annealing may improve electrical behavior by reducing vacancy-related deactivation without eliminating the strain state, or it may drive clustering and diffusion under a different condition. A record-breaking SIMS peak is not automatically a manufacturable source/drain layer; contact resistivity, defectivity, strain relaxation, and downstream stability close the qualification.

The dose-response curve should be plotted on several vertical axes. Dopant precursor partial pressure or molar ratio belongs on the input axis; growth rate, film thickness, total chemical concentration, substitutional fraction, carrier concentration, mobility, sheet resistance, lattice strain, roughness, and defect density belong on outputs. Saturation or rollover in one output while another continues to rise is mechanistic evidence. For example, rising SIMS with flat carrier concentration indicates deactivation or compensation, whereas flat SIMS with falling growth rate indicates surface poisoning or host-rate suppression.

Abruptness has at least four kernels. The delivery kernel contains valve, MFC, manifold, injector, residence, and purge response. The surface kernel contains adsorption, segregation, desorption, and incorporation. The thermal kernel contains diffusion and clustering during growth and later steps. The metrology kernel contains sputter mixing, roughness, matrix effects, and instrumental resolution. If each is approximated as a normalized response, the measured profile is their convolution. Deconvolution is credible only when independent constraints exist; fitting one broad SIMS transition cannot uniquely assign the broadening.

Activation analysis should preserve units and dimensionality. SIMS commonly reports volume concentration versus depth, Hall yields sheet carrier density and mobility for a conducting stack, ECV derives electrically active concentration through depletion assumptions, and four-point probe reports sheet resistance. For a uniform film, $R_s=1/(q\mu n t)$ offers a closure check, but nonuniform profiles require integration of $q\mu(z)n(z)$ and parallel channels. Degenerate statistics, thickness-dependent mobility, substrate conduction, and contacts can break the simple relation. Disagreement is a prompt to inspect assumptions, not an invitation to select the preferred instrument.

At very high concentration, carrier density and mobility cannot be optimized independently. Ionized-impurity scattering lowers mobility as active dopants rise; bandgap narrowing, incomplete ionization assumptions, carrier degeneracy, clustering, and compensation complicate extraction. Contact resistivity may continue to improve after sheet resistance saturates because tunneling width at the metal-semiconductor interface changes. Conversely, a chemically high surface concentration with inactive clusters may not deliver the intended tunneling. Device-relevant contact chains and transfer-length measurements must accompany blanket Hall or Rs.

Substitutional strain is useful evidence but not a standalone dose meter. In Si:P, substitutional phosphorus and vacancy populations both influence lattice response; in Si:C, substitutional carbon supplies tensile strain while interstitial or clustered carbon does not; in SiGe:B, alloy composition, relaxation, and boron all affect diffraction. Reciprocal-space mapping separates coherent strain and relaxation better than one peak position. Combining XRD with composition, TEM defects, and electrical activation constrains the atomic-state model.

Recipe sequencing can turn memory from random noise into a predictable state variable. A high phosphorus exposure may condition walls differently from high boron exposure; alternating them can produce compensation tails not seen in single-species qualifications. Product mix changes cumulative inventory. A run-to-run model should include integrated dopant exposure, temperature-weighted residence, cleans, seasoning, idle time, and preceding material. Dedicated chambers are justified when the allowed intrinsic or counter-doped background is below the reproducible desorption floor.

The incoming wafer can itself be the reservoir. Heavily doped substrates release dopant from the backside or bevel during high-temperature epitaxy, and boundary-layer transport can redeposit it on the front. Autodoping may show center-edge, front-back, or run-position patterns distinct from gas-line memory. Backside oxide or nitride seals, edge design, susceptor contact, pressure, and flow alter it. A blank intrinsic monitor on an undoped substrate and a heavily doped substrate split help separate chamber carryover from wafer-origin dopant.

Safety controls must be treated as process controls because delivery protection changes the physical dose path. Gas cabinet regulators, restricted-flow orifices, excess-flow devices, double containment, purge panels, valve timing, detectors, and abatement pressure all affect available conductance and transient response. Maintenance or cylinder changes can introduce moisture or modify line adsorption. No process optimization should bypass site gas-safety review; PH$_3$, AsH$_3$, and B$_2$H$_6$ hazards require engineered containment, monitoring, interlocks, verified purge, compatible materials, and trained response.

A useful failure analysis starts with the mismatched layer of the accounting chain. If SIMS and active carriers both fall, test delivery, adsorption, and incorporation. If SIMS rises while carriers saturate, test substitutionality, defects, compensation, and activation. If concentration is correct but thickness or morphology changes, test dopant-induced host kinetics. If a nominally intrinsic spacer is contaminated, distinguish gas clearing, surface segregation, wall memory, and diffusion by designed markers. If blanket material passes while contact resistance fails, inspect surface concentration, oxide, silicide consumption, geometry, and interface activation.

problem=>start: In-situ-doped layer fails function
chem=>condition: Is total chemical profile correct by calibrated SIMS or equivalent?
delivery=>operation: Check source, MFC/valve transient, residence, adsorption, host-rate coupling and incorporation
active=>condition: Do active carriers close against chemical dose?
state=>operation: Test substitutional fraction, clusters, vacancies, compensation, hydrogen and anneal response
host=>condition: Are thickness, composition, strain, morphology and defects correct?
growth=>operation: Map dopant-induced kinetics, temperature, facets, selectivity and pattern loading
tail=>condition: Is an edge or intrinsic spacer broadened?
memory=>operation: Separate delivery purge, surface segregation, wall carryover, autodoping, diffusion and SIMS convolution
contact=>condition: Does blanket pass but contact or device fail?
integration=>operation: Inspect surface active dose, oxide, silicide consumption, junction electrostatics and geometry
close=>end: Change one physical mechanism and repeat matched chemistry, structure and electrical monitors
problem->chem
chem(no)->delivery->active
chem(yes)->active
active(no)->state->host
active(yes)->host
host(no)->growth->tail
host(yes)->tail
tail(yes)->memory->contact
tail(no)->contact
contact(yes)->integration->close
contact(no)->close

Qualification should use matched states rather than convenient measurement timing. The as-grown film, post-cap film, post-activation film, post-oxidation film, post-silicide/contact film, and final device may all have different profiles and active fractions. Save split wafers at those states, but recognize that a split lacking subsequent overburden or pattern stress is not identical to product. Thermal budgets should be expressed as full time-temperature-ambient sequences, including ramps and cools, because short high-temperature exposure can dominate diffusion or clustering.

Statistical monitoring must target the sensitive tail. Mean SIMS dose may be stable while rare defects create junction leakage; average Rs may pass while a few high-resistance contacts dominate yield; chamber-average background may hide the first wafer after a high-dose lot. Sampling by wafer position, lot sequence, pattern class, and chamber state is essential. Control charts should track growth rate, Rs, carrier density, mobility, strain, transition tail, and background markers separately so compensating drifts cannot cancel in one composite metric.

The minimum golden data package pairs input delivery traces with output evidence. Preserve dopant source lot and concentration, MFC calibration, valve timing, line pressure, host flows, chamber pressure, actual wafer temperature, deposition rate, recipe order, cumulative wall exposure, clean and seasoning state, substrate/backside, pattern density, and full thermal history. Pair them with calibrated chemical profiles, integrated dose, active-carrier profile, mobility, thickness, host composition, strain and relaxation, defects, morphology, contact resistance, leakage, and device response.

Tool matching should compare response surfaces rather than single setpoints. Two reactors can deliver the same center-wafer SIMS concentration at one condition while differing in growth-rate suppression, temperature sensitivity, transition tail, pattern dependence, memory after a high-dose run, or activation after anneal. Match low, nominal, and high dopant exposures across temperature and loading, then compare normalized maps and transition structures. If a different gas ratio is required to reproduce the same physical outcomes, that is a valid tool-specific calibration; forcing identical MFC labels can create a false match.

Model uncertainty should be carried into the specification. SIMS quantification has matrix and standard uncertainty; Hall extraction has parallel-channel and geometry uncertainty; thickness and composition affect both; contact resistivity depends on current-crowding models; diffusion fits correlate initial profile, diffusivity, and boundary conditions. Report confidence intervals or guard bands where practical. A narrow numeric limit unsupported by measurement capability produces false excursions and false confidence. Gauge repeatability, reproducibility, detection limit, spatial sampling, and deconvolution sensitivity before assigning process capability.

The response to an excursion should preserve evidence. When active carriers fall, do not immediately raise dopant flow before saving gas traces, growth rate, thickness, SIMS, Hall mobility, strain, defects, preceding-lot sequence, and chamber history. Raising flow could restore Rs by increasing chemical dose while worsening growth suppression, compensation, memory, and junction tails. A containment split can test the narrow hypothesis, but the original wafer and chamber state provide the best chance to distinguish delivery drift from activation or host-film drift.