Etching in microfabrication is the controlled removal of material from a solid surface to create patterns, expose underlying layers, separate structures, or release movable components. It is a central process in manufacturing integrated circuits, where successive films must be shaped into electrical devices and interconnections. It is also used to fabricate optical components, sensors, fluidic channels, and mechanical structures. Unlike uncontrolled corrosion or wear, fabrication etching is designed around specified materials, dimensions, profiles, and stopping conditions.
A typical pattern-transfer sequence begins with photolithography, which defines openings in a protective coating. The etchant then removes material through these openings while the protected areas remain substantially intact. Etching may use liquid chemicals, reactive gases, energetic particles, or combinations of chemical and physical action. Its success depends not only on how quickly material is removed, but also on selectivity, directionality, surface damage, uniformity, and the ability to preserve dimensions throughout a three-dimensional structure.
The field spans relatively simple immersion processes and highly controlled vacuum techniques that remove material in repeated, near-atomic-scale increments. The appropriate method depends on the material system and the intended geometry: an undercut cavity, a vertical transistor feature, a deep memory channel, and a suspended mechanical beam impose very different requirements.
Pattern transfer and the principal measures of etching
An etching process operates on a stack rather than on an isolated material. A patterned resist or hard mask covers the target layer, which may itself lie above an insulating film, a conductor, or the substrate. The mask must survive long enough to transfer the desired shape, while the underlying layer must tolerate whatever exposure follows breakthrough. On a semiconductor wafer, these requirements must be met across many repeated structures and over the usable wafer area. Small systematic differences can produce significant variations in device behavior or manufacturing yield.
The etch rate is commonly expressed as thickness removed per unit time, such as nanometres per minute. For a uniform layer under steady conditions, the nominal removal time is approximately , where is thickness and is etch rate. Actual recipes usually require additional allowance because film thickness, local reaction conditions, and pattern geometry vary. Selectivity compares the removal rates of two materials:
A process with a target-to-mask selectivity of 20 removes twenty units of target thickness for each unit of mask thickness under the specified conditions. Selectivity is not an intrinsic constant of a material pair: it changes with chemistry, temperature, particle energy, and surface condition.
Directionality determines whether the etched profile follows the mask opening. An isotropic process removes material laterally as well as vertically, producing undercut beneath the mask. An anisotropic process removes material at different rates in different directions. One simplified measure is , although practical profile characterization requires more than this single number. Anisotropy can arise from crystal orientation in wet etching or from directed particle bombardment and sidewall protection in dry etching; these mechanisms should not be confused.
Other essential quantities include critical dimension, sidewall angle, depth, roughness, and uniformity. The critical dimension is the feature width or spacing most relevant to function, such as a gate length or contact opening. Etch bias describes its change relative to the patterned mask, with the sign depending on the convention used. Aspect ratio is feature depth divided by width; a 10-micrometre-deep trench 0.5 micrometres wide has an aspect ratio of 20:1. As this ratio increases, reactant delivery and product removal become increasingly difficult. Consequently, a process that works well on an unpatterned film may fail in a narrow trench even when the nominal material and chemistry are unchanged.
Wet chemical etching
Wet etching removes material through reactions with a liquid solution. The wafer may be immersed in a bath or exposed to a controlled flow or spray. Material removal generally involves transport of reactants to the surface, chemical reaction, and transport of dissolved products away from it. Depending on conditions, either the surface reaction or transport through the liquid boundary layer may limit the rate. Temperature, concentration, agitation, dissolved reaction products, and the age of the bath therefore affect reproducibility. Gas bubbles can locally block contact and leave unetched regions.
Silicon dioxide is commonly etched with hydrofluoric-acid-containing solutions. A simplified overall reaction is
Buffered oxide etchants typically combine hydrofluoric acid with ammonium fluoride to control solution chemistry and improve process consistency. Oxide etch rates nevertheless depend strongly on film composition and preparation: thermally grown oxide, deposited oxide, and doped glass need not etch at the same rate. This difference can be useful for selective removal, but it can also produce unexpected recesses in multilayer structures. Silicon nitride is often removed using hot phosphoric acid, with selectivity depending on film properties and bath conditions.
Crystalline silicon provides an important example of directional wet etching. Alkaline solutions such as potassium hydroxide and tetramethylammonium hydroxide attack different crystal planes at different rates. Under commonly used conditions, the planes etch much more slowly than the planes. On a suitably aligned (100) wafer, this produces cavities bounded by inclined faces, approximately to the wafer surface. V-shaped grooves and pyramidal pits therefore follow crystallography rather than simply reproducing vertical mask edges. Wafer orientation, mask alignment, doping, and additives all influence the resulting shape and finish.
Wet methods remain valuable because they can provide high selectivity, relatively inexpensive processing, and low bombardment damage. Their limitations include undercut, liquid-handling constraints, and difficulty maintaining small unsupported structures during drying. In microelectromechanical systems, capillary forces can pull released beams or membranes onto nearby surfaces, causing stiction. Supercritical drying and suitable vapor-phase release processes address this problem through different mechanisms. Wet etching also imposes stringent safety requirements: hydrofluoric acid presents severe exposure hazards, and tetramethylammonium hydroxide is acutely toxic as well as corrosive. Industrial use depends on engineered containment, compatible equipment, monitored handling systems, and controlled waste treatment.
Plasma etching and reactor physics
Plasma etching uses an electrically sustained gas discharge to create reactive and energetic species. Electrons accelerated by an electric field collide with feed-gas molecules, producing radicals, ions, excited species, and photons. Neutral radicals can react with the wafer surface, while positive ions are accelerated through the plasma sheath toward the wafer. The bulk gas can remain much cooler than the electron population, allowing reactive chemistry at substrate temperatures far below those required for comparable thermal reactions. This nonequilibrium behavior is a central advantage of low-pressure plasma processing.
In reactive-ion etching, chemical reactions and ion bombardment act together. Ions can remove a passivating layer, break surface bonds, or activate adsorbed species so that volatile products form more readily. A classic experimental demonstration by John Coburn and Harold Winters in 1979 showed that simultaneous ion bombardment and reactive-gas exposure could remove silicon much faster than either action alone. This synergy explains why an etch rate cannot always be calculated as the sum of an independent chemical rate and a physical sputtering rate.
Capacitively coupled reactors sustain a discharge through radio-frequency voltages applied between electrodes. The wafer commonly sits on a powered electrode and experiences a negative time-averaged bias relative to the plasma. Inductively coupled plasma systems use an external coil to couple energy into the discharge; a separate substrate bias provides additional control over ion impact energy. Source power and bias are not completely independent in practice, but their partial separation expands the operating range. Many anisotropic processes operate at pressures from a few to a few tens of millitorr, although suitable pressures vary widely with reactor design and application.
A complete reactor also includes gas delivery, vacuum pumping, temperature control, chamber-wall management, and exhaust treatment. An electrostatic chuck can hold the wafer while a backside helium layer improves heat transfer. Wafer temperature affects adsorption, reaction rates, product volatility, and protective-film formation, making thermal control part of the chemistry rather than merely an equipment convenience. Chamber walls absorb and release reactive species and can accumulate deposits, so cleaning and seasoning procedures influence repeatability. Directed ion bombardment provides useful anisotropy, but it also introduces trade-offs involving mask erosion, charging, and damage. Reactor design and process development therefore seek an appropriate combination of ion flux, energy distribution, radical supply, and surface temperature rather than simply maximizing plasma density or power.
Material-specific dry-etch chemistries
Dry-etch chemistry is selected partly by whether the target material can form products that leave the surface under the available pressure and temperature. Silicon reacts with fluorine-containing species to form volatile silicon tetrafluoride. Fluorine-rich plasmas can therefore remove silicon rapidly, but without additional control their neutral-radical component also attacks laterally. Chlorine- and bromine-containing chemistries are widely used where controlled silicon profiles and surface passivation are important. The best chemistry is consequently not always the one producing the highest unpatterned etch rate; controllable reaction probability and protection of neighboring materials can be more valuable.
Silicon dioxide presents a different balance. Fluorocarbon gases such as carbon tetrafluoride, trifluoromethane, and octafluorocyclobutane can generate both reactive fluorine species and carbon-rich surface films. Ion bombardment helps remove or modify these films at the feature bottom, while sidewalls retain more protection. Oxygen released from an oxide surface changes the local carbon balance, allowing different behavior on oxide, silicon, and nitride. Selective oxide etching therefore relies on a coupled competition between deposition and removal, not simply on the presence of fluorine. Gas composition, bias, pressure, and temperature determine which surface remains protected and which continues to etch.
Metal etching illustrates the importance of product volatility. Aluminum can be patterned using chlorine-based plasmas, often with boron trichloride contributing to removal of oxygen-containing surface material. Tungsten can form volatile tungsten hexafluoride in fluorine-containing processes. Copper is much more difficult to pattern by conventional reactive plasma etching because its common halide products have low volatility under many practical process conditions. This limitation helped favor interconnect schemes in which trenches and holes are etched in a dielectric, filled with copper, and then planarized, rather than directly etching a blanket copper film. Photoresist and other organic films, by contrast, are commonly removed with oxygen-containing plasmas.
Purely physical removal provides another option. In ion-beam etching, a directed ion beam, often using argon, ejects surface atoms through momentum transfer. It can pattern materials for which convenient volatile reaction products are unavailable, including some magnetic and optical films. Its drawbacks include comparatively low material selectivity, mask erosion, redeposition, and damage. Material ejected from the feature bottom may accumulate on sidewalls and form fences or bridges. Reactive ion-beam variants combine the directed beam with chemical assistance, but still require careful control of surface and angular effects. Across all these methods, a suitable chemistry must be evaluated against the entire stack: an aggressive reaction that removes the target efficiently may also consume a thin stop layer, damage an exposed conductor, or leave residues incompatible with the next process step.
Deep etching and high-aspect-ratio structures
Deep reactive-ion etching enables trenches, holes, and cavities whose depth greatly exceeds their width. It is important in mechanical sensors, through-silicon connections, and three-dimensional memory manufacturing. Silicon structures tens to hundreds of micrometres deep can be produced with specialized processes, while other applications require extremely narrow openings through thick dielectric stacks. These tasks share transport and profile-control problems, but they do not use a single universal recipe. Deep silicon etching and high-aspect-ratio dielectric etching involve different surface chemistries and protection mechanisms.
The Bosch process alternates silicon removal and sidewall protection. A fluorine-producing step, commonly based on sulfur hexafluoride, etches exposed silicon. A subsequent fluorocarbon step, commonly using octafluorocyclobutane, deposits a protective film. In the next etching interval, directional ion bombardment preferentially clears protection from the bottom, allowing downward removal while the walls remain protected. Repetition generates deep, nearly vertical structures. Because the process is cyclic, the sidewalls typically show scallops. Shorter cycles can reduce scallop size, but they also change throughput, gas-transition requirements, and the balance between protection and removal.
Cryogenic silicon etching instead combines fluorine chemistry, often sulfur hexafluoride with oxygen, with a cooled substrate, commonly near . Under suitable conditions, an oxygen-containing silicon surface layer protects the sidewalls, while ion-assisted reactions sustain etching at the bottom. This can produce smooth walls without the characteristic Bosch scallops. Its performance depends sensitively on actual wafer temperature, oxygen supply, mask behavior, and thermal contact. Local heating or nonuniform cooling can alter passivation sufficiently to change the profile, so temperature uniformity is a critical process variable.
As a feature deepens, fewer reactive particles reach its bottom and products have a longer path to escape. Aspect-ratio-dependent etching describes the resulting dependence of rate or profile on feature geometry. Narrow openings often etch more slowly, although inverse behavior can occur under particular conditions. Ion angular spread, neutral sticking, charge accumulation, and sidewall interactions all contribute. In stacked flash memory, deep channel holes must pass through many alternating layers without unacceptable taper or bending. In through-silicon vias, profile and roughness affect subsequent insulation, barrier deposition, and conductor filling. Increasing depth therefore requires simultaneous control of transport, charging, mask lifetime, and sidewall protection. Simply extending the etch time may enlarge the opening, erode the mask, or damage the bottom before the desired structure is obtained.
Atomic-layer and highly selective etching
Atomic-layer etching divides removal into separate, controlled surface-reaction steps. In an idealized cycle, one exposure modifies only a limited surface region, and a second removes the modified material without substantially attacking the unmodified substrate. Purging or temporal separation prevents the reactants from continuously interacting. A widely studied example involves chlorinating a silicon surface and then applying low-energy argon ions that preferentially remove the chlorinated region. If the ion energy is too low, removal is incomplete; if it is too high, ordinary sputtering undermines the intended control.
The term “atomic-layer” describes a mode of control rather than a guarantee that exactly one crystallographic plane is removed per cycle. The etch-per-cycle value depends on surface composition, orientation, reaction completeness, and the thickness of the modified region. Evidence for self-limiting behavior includes saturation with reactant dose and, where applicable, a useful energy range in which removal remains selective to the modified surface. A periodically pulsed plasma process is not automatically atomic-layer etching: it must exhibit the relevant surface-limited behavior. Uniform removal over complex structures also depends on whether each exposure reaches all intended surfaces.
Thermal atomic-layer etching uses sequential chemical reactions without requiring directional ion bombardment. For some metal oxides, one step converts the surface to a fluoride-containing layer and another uses ligand-exchange chemistry to form removable products. Because thermal reactions need not rely on a vertically directed ion flux, they can support conformal removal from three-dimensional surfaces. The counterpart is that they do not inherently provide a vertical pattern-transfer profile. Plasma-assisted and thermal approaches therefore address overlapping but different needs, with trade-offs in directionality, damage, temperature, materials compatibility, and throughput.
Highly selective etching is a broader category and need not proceed in discrete self-limiting cycles. A prominent application is removing silicon–germanium sacrificial layers while preserving silicon channels in a gate-all-around transistor. The process must reach between stacked sheets, clear the sacrificial material, and avoid thinning or roughening the channels. Selectivity measured on blanket films may not predict performance in these confined spaces because composition, strain, native surface layers, and transport change the local reaction. Similar demands arise when exposing thin dielectrics or removing a temporary material adjacent to a sensitive conductor. Atomic-scale dimensional control is therefore a stack-level property: precise chemistry must be combined with reproducible starting surfaces, sufficiently uniform reactant delivery, and reliable termination of the process.
Masks, etch stops, and integration into device fabrication
The etch mask translates a lithographic image into a durable pattern. Photoresist is convenient because exposure and development directly define its openings, but it may erode, deform, or change chemically during plasma processing. Deep etches and demanding material systems often use a hard mask made from silicon dioxide, silicon nitride, carbon, or a metal-containing film. The resist first patterns this intermediate layer; the hard mask then protects the target during the main etch. Adding a hard mask improves durability but also adds deposition, pattern-transfer, and removal steps, each with its own dimensional errors and contamination risks.
A simple mask-thickness estimate follows from selectivity. If a target depth must be removed with target-to-mask selectivity , the nominal mask consumption is . Practical designs require additional margin for nonuniformity, process transitions, corner erosion, and deliberate overetch. Mask shape matters as much as initial thickness. Faceting of the mask edge can change the angular distribution of incoming particles and widen the transferred feature. Roughness in the patterned resist may also propagate into the target, affecting electrical or optical behavior even when the mean feature width remains acceptable.
An etch-stop layer limits removal by etching much more slowly than the target. It is not an absolute barrier: finite selectivity means that extended exposure eventually consumes it. For a contact opening, the process must remove all insulating material at the bottom while preserving the underlying conductor or semiconductor. A breakthrough step may first remove surface oxide or residues, a main step removes most of the film, and a lower-damage or more selective step completes clearing. Overetch accommodates variation in thickness and local rate, but excessive overetch can deepen exposed regions, increase charging damage, or enlarge the opening.
Integration requirements differ across device generations and geometries. In a fin field-effect transistor, etching defines narrow vertical silicon bodies whose width, height, and sidewall condition affect electrostatic control. Spacer fabrication uses conformal deposition followed by directional etching to leave material along feature sidewalls, enabling dimensions and alignments not determined by a single lithographic exposure. In CMOS fabrication, these operations coexist with strict contamination and thermal-budget constraints. After etching, residues must be removed without undoing the desired selectivity or damaging newly exposed surfaces. The correct unit of optimization is therefore the complete process module—including deposition, patterning, etching, cleaning, and subsequent film formation—rather than the etch recipe in isolation.
Endpoint detection, metrology, and process control
An etch endpoint is the condition at which the intended removal has been achieved. Time-based processing is adequate when starting thickness and rate are sufficiently stable, but many applications require direct or indirect observation of the process. Optical emission spectroscopy monitors light emitted by excited plasma species. When a target film clears, the concentration of a reaction product or reactant may change, producing an observable spectral transition. The signal depends on exposed area and plasma conditions, so a small population of contact holes may provide a much weaker endpoint than a broadly exposed film.
Optical reflectometry and interferometry observe changes in light reflected from the wafer. Film-thickness evolution changes optical phase and intensity, enabling removal rate or remaining thickness to be inferred using an appropriate model. The familiar appearance of interference oscillations does not imply a universal thickness increment: wavelength, refractive index, incidence angle, and stack structure matter. Exhaust monitoring by mass spectrometry can provide complementary information about reaction products. None of these measurements directly proves that every opening across a patterned wafer has cleared; endpoint interpretation must account for the relationship between the measured signal and the slowest or most vulnerable structures.
Post-etch metrology measures the actual result. Scanning electron microscopy can characterize top-down dimensions and, with suitable sample preparation, cross-sectional profiles. Transmission electron microscopy resolves thin interfaces and local damage at finer scales, but examines a limited region and requires more demanding preparation. Atomic force microscopy measures accessible surface topography, while ellipsometry is useful for remaining film thickness and optical properties. Deep trenches and overhanging structures complicate measurement because the most important surfaces may be inaccessible to the probe or hidden from the viewing direction. Reporting only the top opening can miss taper, bowing, or bottom narrowing.
Manufacturing control combines these measurements with equipment signals and statistical monitoring. Gas flows, pressure, forward and reflected power, chuck temperature, and endpoint traces can reveal chamber drift or abnormal runs. Test structures help distinguish across-wafer nonuniformity from layout-dependent effects. Chamber matching requires more than copying set points because wall condition, pumping behavior, electrode geometry, and sensor calibration affect the realized process. Sampling plans must also reflect measurement uncertainty and the cost of destructive inspection. Increasingly sophisticated data analysis can identify correlations, but a correlation between a sensor trace and a dimensional error does not by itself establish the physical cause; process changes still require validation against structural and device measurements.
Damage, defects, and pattern-dependent behavior
Etching can modify material that is not visibly removed. Energetic ions may displace atoms, amorphize a shallow surface region, or implant species. Plasma-generated photons can alter dielectric bonds, while reactive species may change surface termination or penetrate vulnerable films. In porous low-permittivity interconnect dielectrics, processing can remove carbon-containing groups and increase moisture uptake, changing electrical properties. The importance of damage depends on the device: a modification tolerable in a thick mechanical component may be unacceptable near a thin gate insulator, a semiconductor interface, or an optically active surface.
Electrical charging creates another class of defects. Electrons and ions have different transport properties and angular distributions, particularly within deep insulating features. Unequal arrival rates can create local potentials that deflect ions, modify bottom profiles, or stress connected devices. Notching near an insulating stop layer is one possible manifestation. Large conducting patterns connected to sensitive structures can collect charge during plasma exposure, producing the so-called antenna effect. Pulsed operation, adjusted plasma conditions, and layout-level protection can reduce particular risks, but their effectiveness depends on the process and the electrical connectivity of the structure.
Pattern density affects chemical supply and surface reactions. Microloading describes differences associated with local patterned surroundings: a dense region of openings may consume reactants or accumulate products differently from an isolated opening. This is distinct from aspect-ratio dependence, although the two often occur together. Macroloading concerns changes associated with the total exposed area or broader wafer-scale demand. Consequently, a recipe optimized on sparse test patterns may behave differently on a product wafer. Dummy features and layout rules can help manage such differences, but they must be compatible with the electrical, mechanical, and downstream fabrication requirements.
Profile defects have identifiable, though sometimes overlapping, mechanisms. Bowing is excessive lateral removal at an intermediate depth. Microtrenching is enhanced removal near the bottom corners, often associated with ion trajectories and reflection from sidewalls. Mask erosion can widen features, while excessive passivation can constrict an opening or stop removal entirely. Particles and locally resistant residues can act as unintended masks, leaving pillars or “grass.” Redeposition is especially important when removed species do not readily escape. Diagnosis therefore uses the spatial location, material composition, and geometry of the defect rather than its appearance alone. Cleaning, lower-energy finishing steps, and modified passivation balances can address some problems, but each intervention may introduce a new trade-off in selectivity, throughput, or dimensional control.
Historical development and the pressures shaping modern etching
Early semiconductor manufacturing relied heavily on wet chemical processing, which could selectively remove oxides, metals, and semiconductor material using comparatively simple equipment. The growth of planar integrated circuits made repeated pattern transfer increasingly important. As lateral dimensions decreased, isotropic undercut became harder to tolerate, while multilayer stacks demanded improved selectivity. Dry plasma methods expanded during the 1960s and 1970s, initially serving applications such as organic-film removal and subsequently becoming central to fine-pattern transfer. Their adoption was driven by the need for controlled profiles rather than by a universal superiority over wet chemistry.
Research in the 1970s clarified the joint role of chemical reaction and energetic bombardment. The 1979 Coburn–Winters experiments became a foundational demonstration of ion-assisted etching synergy. Subsequent reactor development improved control over plasma density, substrate bias, and wafer temperature. High-density plasma sources became especially useful as manufacturers needed substantial reactive flux without relying exclusively on high substrate bombardment energy. At the same time, larger wafers and tighter dimensional tolerances made uniformity, equipment matching, and chamber conditioning increasingly important engineering disciplines.
Deep silicon etching developed into a major enabling technology for micromechanical fabrication. Cryogenic silicon etching was reported in the late 1980s, and Bosch developed its alternating etch-and-passivation method in the early 1990s. These approaches made complex, high-aspect-ratio silicon structures more practical and supported applications beyond conventional transistor manufacture. Later demand from stacked memory, advanced interconnects, and three-dimensional transistors shifted attention toward different combinations of depth, profile precision, and material selectivity. The transition from planar devices to fins and stacked channels increased the number of surfaces whose dimensions and chemistry had to be controlled simultaneously.
Current development is also constrained by environmental impact and resource use. Several fluorinated gases used in etching are potent greenhouse gases, and their climate significance depends on both consumption and the fraction destroyed or captured. Gas substitution cannot be evaluated solely by the feed gas’s properties because plasma decomposition and abatement can generate other products. Exhaust systems must also handle corrosive or hazardous reaction products, while vacuum pumping, cooling, and chamber cleaning consume energy and materials. More efficient gas utilization, validated abatement, selective removal at lower damage levels, and longer stable chamber operation are therefore connected process goals. Research must determine whether gains demonstrated on small test structures persist across full wafers, realistic pattern densities, complex stacks, and the repeated operating cycles required in manufacturing.