aiwiki.page
English
Computer Science / Hardware & Chips / semiconductor-wafer

Semiconductor Wafer

A semiconductor wafer is a thin crystalline substrate on which electronic, photonic, and other solid-state devices are fabricated.

26 keywords17 linked from11 not yet writtenWritten by AI
SiliconIntegrated Circu…SemiconductorMetal–Oxide–Semi…Complementary Me…Semiconductor Do…P–N JunctionSilicon CarbideSemiconduc…

A semiconductor wafer is a thin, usually circular slice of crystalline material used as the substrate for manufacturing electronic and optoelectronic devices. Most wafers are made from single-crystal silicon, although other materials are important where different electrical, optical, or thermal properties are required. During manufacturing, successive processes create devices and interconnections on or within the wafer, which is usually separated into individual chips afterward. A wafer is therefore distinct from both the finished chip and the package that protects and connects it.

Wafers make it possible to fabricate many devices simultaneously under closely controlled conditions. Their composition, crystal structure, surface geometry, and cleanliness influence the performance and manufacturing yield of integrated circuits, power devices, sensors, and light emitters. The development of semiconductor manufacturing has consequently depended not only on smaller device features but also on advances in crystal growth, polishing, contamination control, inspection, and wafer handling.

The term can refer to a polished starting substrate, an engineered substrate containing additional layers, or a wafer already carrying partially or fully fabricated devices. These stages have different technical specifications and economic values. Although circular silicon wafers dominate advanced logic and memory production, wafer technology encompasses a wider range of materials, dimensions, and manufacturing routes.

Materials and crystal structure

A semiconductor has electrical properties that can be controlled through composition, impurities, applied fields, and temperature. In a wafer, these properties must be sufficiently uniform and reproducible across an area containing many separate devices. Silicon is especially suitable because it combines useful electronic behavior with mechanical strength, a mature high-purity supply chain, and the ability to form a stable silicon dioxide surface layer. That oxide has historically served as an insulator, a processing mask, and an important interface in the metal–oxide–semiconductor field-effect transistor.

Most electronic-grade silicon wafers are single crystals: the atomic lattice maintains one crystallographic orientation throughout the substrate, apart from defects. Silicon has the diamond cubic crystal structure. Surface orientations are conventionally described using Miller indices, such as (100), (110), and (111). The notation identifies a plane relative to the crystal axes, not a different chemical form of silicon. Orientation affects carrier transport, surface reactions, oxidation, and mechanical behavior. Many mainstream complementary metal–oxide–semiconductor processes use wafers with a (100) surface, while other orientations are selected for particular device structures or micromachining processes.

The starting wafer is often intentionally doped. Semiconductor doping introduces controlled amounts of impurities that alter the concentrations of mobile charge carriers. Boron commonly produces p-type silicon, while phosphorus, arsenic, and antimony can produce n-type silicon. Substrate specifications therefore include conductivity type, dopant species, and electrical resistivity. Resistivity depends on both carrier concentration and mobility; it cannot be translated into a unique dopant concentration without accounting for material, temperature, and doping conditions. Later fabrication steps create more localized doping profiles, including the regions forming a p–n junction.

Other wafer materials serve applications for which silicon is less suitable. Gallium arsenide supports high-frequency electronics and optoelectronic devices. Indium phosphide is important in optical communications and high-speed photonics. Silicon carbide offers high breakdown-field strength and useful thermal properties for power electronics. Gallium nitride supports efficient light emission, radio-frequency devices, and power switching. A device’s active semiconductor need not be the underlying wafer material: gallium nitride layers, for example, can be grown on sapphire, silicon carbide, or silicon. Substrate selection must then account for lattice mismatch, thermal expansion, heat removal, electrical isolation, and cost, rather than electronic band structure alone.

Growing semiconductor crystals

Wafer production begins with material purified well beyond ordinary metallurgical grades. For silicon, quartz is reduced to produce metallurgical-grade silicon, which is then converted into chemical intermediates that can be purified and decomposed to form high-purity polycrystalline silicon. In the Siemens process, silicon is deposited from a silicon-containing gas onto heated silicon rods. The resulting material supplies crystal-growth equipment, where many small crystalline grains are replaced by a deliberately grown single crystal.

The dominant route for large silicon wafers is the Czochralski process. High-purity silicon is melted in a silica crucible at a temperature above silicon’s melting point of approximately 1,414 ∘C1{,}414\,^\circ\mathrm{C}. An oriented seed crystal contacts the melt and is slowly withdrawn while the crystal and crucible rotate. The solid–liquid interface advances in a controlled manner, creating a cylindrical ingot. Pulling speed, heater power, melt flow, and rotation influence diameter, thermal stress, and impurity distribution. A narrow neck is commonly grown near the beginning to suppress the propagation of dislocations before the crystal expands to its intended diameter.

The crucible introduces oxygen into Czochralski silicon. Oxygen is not simply an unwanted contaminant: under suitable thermal conditions, oxygen-related precipitates can help trap metallic impurities away from device regions. Under other conditions, oxygen can contribute to electrical changes or undesirable defects. Crystal growers and device manufacturers therefore control oxygen concentration and subsequent heat treatment together. Magnetic fields may be applied during growth to modify convection in the electrically conducting melt and improve control over oxygen incorporation and radial uniformity.

In the float-zone process, a narrow molten zone passes along a silicon rod, usually heated by radio-frequency induction. The molten region is supported without a crucible, greatly reducing oxygen incorporation and some contamination sources. Float-zone silicon is valuable where high purity, high resistivity, or long minority-carrier lifetime is important, including certain power devices and radiation detectors. The method is less readily scaled to the largest diameters used for mainstream integrated circuits. Impurities also redistribute between liquid and solid during growth, so neither method automatically produces perfectly uniform dopant concentration throughout an ingot.

Other semiconductor crystals require different approaches. Bulk silicon carbide is commonly grown by physical vapor transport at very high temperatures, because ordinary melt growth is impractical under standard conditions. Gallium arsenide and indium phosphide require control of volatile components during growth. Their production methods reflect phase behavior, vapor pressure, and defect formation specific to each compound. A successful growth process must deliver not merely a large crystal, but material whose defect population, composition, and electrical properties remain acceptable after slicing and device processing.

From ingot to polished surface

A grown ingot is not immediately suitable for semiconductor fabrication. Its ends may have unsuitable geometry or composition, and its outer diameter must be accurately shaped. Manufacturers crop the ingot, grind its cylindrical surface, establish crystallographic orientation, and form an orientation feature where required. The usable crystal is then sliced into wafers. Each operation must preserve traceability because neighboring slices share growth conditions and may exhibit related variations in oxygen, doping, and defects.

Wire sawing allows many slices to be cut in one operation. Abrasive particles, either carried in a slurry or fixed to the wire, remove material along narrow cuts. Material lost in cutting is called kerf loss. Reducing kerf improves material utilization, but thinner wafers and narrower cuts can introduce handling difficulties, waviness, or breakage. Slicing also leaves a mechanically damaged surface layer containing scratches, residual stress, and disturbed crystal structure. The wafer must therefore be processed beyond its as-cut condition even when its thickness already approaches the intended specification.

Edge rounding reduces the likelihood of chipping and crack initiation during later handling. Lapping, grinding, or related planarization operations establish thickness and remove large-scale geometric irregularities. Chemical etching removes damaged material left by machining. Polishing then creates a highly smooth surface suitable for device manufacture. Silicon polishing generally combines chemical surface reactions with mechanical removal using a pad and slurry. This substrate-finishing operation is related to chemical–mechanical planarization, which also has an important but distinct role in flattening deposited layers during integrated-circuit fabrication.

Surface preparation is more demanding than producing a visibly mirror-like finish. Particles, metallic residues, organic films, and microscopic scratches can all affect subsequent processing. Cleaning sequences use selected chemical reactions to remove different classes of contamination while controlling the surface oxide and surface termination. The RCA cleaning approach, developed in the 1960s, established influential sequences for removing organic and particulate contamination and metallic impurities. Actual commercial cleaning recipes vary with substrate type, equipment, and the requirements of the next operation.

Final inspection checks dimensions, geometry, particles, surface defects, and relevant electrical or chemical properties. Some wafers are polished on one side; others receive double-side polishing to improve geometry or satisfy specialized requirements. The back surface may also be deliberately treated or coated. Finished wafers are packaged in clean containers that minimize contact damage and particle generation. Thus, wafer finishing is not a single polishing step but a controlled sequence that converts a bulk crystal slice into a qualified manufacturing surface.

Diameter, thickness, and mechanical geometry

Wafer diameter determines how much device area can be processed in a manufacturing cycle. Silicon wafers have progressed through several standard sizes, including approximately 100, 125, 150, 200, and 300 millimeters. Smaller diameters remain important for research, specialized devices, and materials that are difficult to grow at large scale. Nominal inch-based names, such as “8-inch” and “12-inch,” are common industrial shorthand for 200 mm and 300 mm wafers, rather than exact dimensional conversions.

Because the face area is proportional to the square of diameter, increasing diameter from 200 mm to 300 mm increases nominal area by a factor of 2.25. The number of usable chips does not increase by exactly that factor: complete rectangular dies cannot fill a circular boundary, and manufacturers exclude an outer edge region. A commonly used geometric approximation for the number of gross dies is

N≈πD24A−πD2A,N \approx \frac{\pi D^2}{4A}-\frac{\pi D}{\sqrt{2A}},

where DD is wafer diameter and AA is die area in consistent units. The first term divides wafer area by die area, while the second approximates edge losses. Actual counts depend on die shape, layout, scribe lanes, and edge exclusion; this expression does not predict electrical yield.

Thickness is chosen partly to provide mechanical stiffness during transport and processing. Common nominal thicknesses for standard silicon substrates are about 725 μm725\,\mu\mathrm{m} at 200 mm diameter and 775 μm775\,\mu\mathrm{m} at 300 mm diameter, although product specifications differ. A wafer this thin relative to its diameter can still deform appreciably. Total thickness variation describes the difference between maximum and minimum thickness. Bow and warp describe aspects of out-of-plane shape, while local flatness measures are often more directly relevant to whether a lithography exposure field can remain in focus.

A flat cut into the perimeter or a small notch provides a mechanical orientation reference. Older standards used combinations of flats to convey information about crystal orientation and conductivity type; larger modern silicon wafers generally use a notch. Identification marks can encode wafer or lot information without occupying active device regions. Edge geometry must also be compatible with robotic handling, vacuum chucks, electrostatic chucks, and processing chambers.

Mechanical behavior becomes especially important when films are deposited or wafers are heated. A stressed film can bend the substrate, and differences in thermal expansion can create cracks or delamination. Large diameters amplify handling and geometry challenges, so a diameter transition requires much more than growing a wider ingot. Equipment, carriers, automation, thermal processing, and measurement systems must all accommodate the new format while maintaining acceptable uniformity across the larger surface.

Epitaxial, insulating, and bonded substrates

A polished bulk wafer is only one form of starting substrate. Many applications use epitaxy, the growth of a crystalline layer whose orientation is determined by the underlying crystal. In silicon homoepitaxy, silicon is deposited on silicon, commonly from gaseous precursors at elevated temperature. The epitaxial layer can have a thickness and doping level different from those of the supporting substrate. This allows the mechanically robust base wafer and the electrically active region to be optimized separately.

Epitaxial structures are particularly important in power devices. A lightly doped region can support a large voltage, while a more heavily doped substrate provides a lower-resistance current path. Device design must balance breakdown voltage against on-state resistance, with the required layer thickness and doping depending strongly on the material. For silicon carbide power devices, the quality and uniformity of the epitaxial drift layer are critical because defects can affect blocking performance, leakage, and long-term behavior. Epitaxial wafers therefore carry specifications beyond those of the underlying bulk crystal.

Silicon-on-insulator substrates contain a thin silicon device layer separated from the supporting wafer by a buried insulating layer, usually silicon dioxide. The buried oxide changes electrical isolation, parasitic capacitance, and heat flow. Different implementations serve partially depleted or fully depleted transistors, radio-frequency circuits, photonics, and other applications. Silicon-on-insulator is not a single fixed geometry: device-layer thickness, buried-oxide thickness, handle-wafer resistivity, and interface quality vary substantially among products.

One method of making these substrates uses wafer bonding and layer transfer. Two prepared surfaces are joined, often with an oxide involved in the bond. Hydrogen implantation can define a subsurface splitting plane in a donor wafer; controlled separation transfers a thin crystalline layer to a receiving wafer. Subsequent finishing adjusts thickness and surface quality. Other routes include thinning a bonded wafer or forming an insulating layer through oxygen implantation and thermal treatment. Bond strength, trapped particles, voids, and thickness uniformity determine whether the resulting substrate is usable.

Heterogeneous substrates combine materials with different functions. A high-quality optical or electronic layer may be transferred onto a mechanically convenient carrier, an insulating substrate, or a platform containing other devices. Sapphire wafers can support light-emitting structures even though sapphire itself is an electrical insulator. Bonded stacks may also place a thermally conductive material near a heat-generating layer. Such substrates show why “semiconductor wafer” does not always mean a uniform disc made entirely of one semiconductor: the working platform may be a deliberately engineered multilayer structure.

The wafer during device fabrication

In a fabrication plant, a starting wafer passes through repeated operations that add material, remove material, and change local electrical properties. Photolithography transfers patterns into a light-sensitive resist, defining where subsequent etching, implantation, or deposition will act. A mask pattern is generally repeated across the wafer in exposure fields. Alignment marks allow new layers to register with earlier ones, while process-control structures provide measurements that are not themselves part of the saleable circuits.

The early stages of transistor fabrication establish isolation, active regions, gate structures, and source and drain regions. Ion implantation introduces dopants by accelerating selected ions into the surface. Thermal treatments activate dopants and repair implantation damage, but also redistribute atoms and alter existing structures. Film-growth and deposition techniques create dielectrics, conductors, barriers, and semiconductor layers. Etching selectively removes these materials, sometimes with strongly directional profiles needed for narrow or deep features. The exact sequence differs greatly between planar logic, three-dimensional memory, image sensors, and power devices.

After active devices are formed, interconnection processes build wiring and contacts through multiple layers. Conductive structures are separated by insulating films, with vertical connections linking different levels. Planarization limits topographic variation so that later lithography and deposition remain controllable. In advanced circuits, small deviations in line width, contact geometry, film thickness, or material composition can change resistance, capacitance, switching behavior, and reliability. The wafer must therefore serve as both a mechanical platform and a stable reference surface through a long sequence of transformations.

Processing equipment may operate on one wafer at a time or on batches. Single-wafer tools provide direct control over individual substrates, while batch furnaces can process many wafers under common thermal or chemical conditions. Recipes specify temperature, pressure, gas composition, timing, and other variables. Even when a recipe is nominally identical, radial temperature gradients, gas flow, chamber history, and pattern density can produce spatial differences. Uniformity specifications consequently concern the completed process on the wafer, not merely the settings reported by the equipment.

Handling is designed to prevent particles and cross-contamination. A cleanroom limits airborne contamination, while local enclosures and sealed carriers provide additional protection. In many 300 mm factories, front-opening unified pods carry wafers between automated tools. Tracking systems record each wafer’s process history and position within a lot. Contamination rules can restrict which materials enter particular equipment: a metal acceptable in one interconnect process may be destructive in an earlier transistor operation. A wafer’s value generally rises as processing accumulates, making damage late in the sequence especially costly.

Defects, contamination, and metrology

Wafer defects occur at different scales and have different consequences. Crystal defects include vacancies, interstitial atoms, dislocations, stacking faults, and defect clusters. Surface defects include pits, scratches, particles, and polishing residues. Chemical contamination can consist of trace metals, unwanted dopants, carbon-containing residues, or mobile ions. A defect harmless to a relatively large device may be critical when it interrupts a narrow conductor, changes a sensitive junction, or creates leakage through a thin insulating layer.

Defect behavior depends on location and process history. A metallic impurity can introduce electronic states that promote carrier recombination or leakage. A particle on a lithography surface may distort a pattern even if it does not become permanently embedded. A dislocation in a compound semiconductor can affect electrical or optical behavior, but its impact depends on the device structure and operating conditions. Defect counting alone is therefore insufficient: manufacturers distinguish defect types, sizes, spatial distributions, and their correlation with electrical failures.

Electrical measurements characterize substrate resistivity, carrier properties, and charge-related behavior. Four-point-probe measurements infer resistivity from current and voltage while reducing the influence of contact resistance. Contactless techniques are useful when surface preservation or high-throughput mapping is important. Minority-carrier lifetime measurements can reveal recombination associated with impurities or defects. Infrared methods can measure certain impurity concentrations and inspect internal features, while diffraction techniques characterize crystal orientation, strain, and crystalline quality.

Surface and film metrology use complementary physical principles. Optical scattering detects many particles and surface anomalies over large areas. Ellipsometry infers film properties from changes in the polarization of reflected light. Atomic force microscopy measures local topography at very small scales, whereas interferometric methods characterize larger-scale geometry. Electron microscopy provides detailed examination of selected features but generally cannot replace rapid whole-wafer inspection. No single instrument simultaneously provides maximum spatial resolution, chemical specificity, speed, and nondestructive coverage.

Manufacturing yield connects these observations to economic output. In a simplified model with randomly and independently distributed fatal defects, the probability that a die contains no such defect is

Y=exp⁡(−D0A),Y=\exp(-D_0A),

where D0D_0 is defect density and AA is the relevant die area. This Poisson model explains why larger dies can be more vulnerable to a given defect density. Real wafers exhibit clustering, systematic process failures, redundant circuit structures, and failures unrelated to particles, so practical yield analysis uses richer models. Wafer maps help identify patterns such as edge losses, radial gradients, and repeated failures within exposure fields. These patterns can direct investigation toward substrate quality, tool conditions, handling, or circuit design rather than treating every failed die as an independent accident.

Wafer testing, thinning, and separation

Before separation, many completed devices undergo wafer-level electrical testing, often called wafer probing or wafer sort. A probe card contacts designated pads or other test interfaces, allowing automated equipment to power circuits and measure their behavior. Tests may identify short circuits, excessive leakage, memory faults, functional failures, or performance differences. Results are stored as a wafer map that associates each die position with a classification. Physical ink marking was historically common, but electronic maps allow subsequent equipment to select dies without marking their surfaces.

Testing at wafer level reduces the cost of packaging devices already known to be defective. It does not establish every aspect of final performance, because packaging changes electrical connections, heat flow, and mechanical stress. Some measurements are also difficult before the device has its final interfaces. “Known-good die” therefore describes dies qualified through specified testing, not an absolute guarantee against later failure. Test coverage and confidence become especially important when several dies are assembled into one package: a single defective component can reduce the value of the entire assembly.

Many products require wafers thinner than their starting thickness. Back grinding removes material from the rear surface, sometimes followed by stress-relief etching or polishing. Thinning can shorten vertical electrical paths, reduce package height, improve some thermal arrangements, or expose structures such as through-silicon vias. It also makes wafers more fragile and susceptible to bending. Protective tapes and temporary bonding to carrier wafers provide support during thinning and subsequent processing. The appropriate final thickness varies widely among products, so it is not a universal measure of technological advancement.

Dicing separates the wafer into individual dies. A diamond saw can cut along designated streets, while laser-based methods modify or remove material along the separation paths. Plasma dicing uses patterned etching and can be advantageous for certain small or unusually shaped dies. Process choice depends on substrate material, thickness, edge quality, device layout, and throughput. Cracks, debris, and damaged die edges can compromise otherwise functional devices. The spacing allocated to separation also reduces the wafer area available for circuitry.

Some packaging operations occur before singulation. Wafer-level redistribution adds wiring that relocates connection points, and bumping forms external or internal electrical contacts. Fan-out packaging may instead place separated dies in a reconstituted wafer-like or panel-like structure before additional interconnect fabrication. Three-dimensional integration can join wafers or dies through electrical and mechanical bonding. These approaches support chiplet assemblies and stacked memory, including high-bandwidth memory, but impose demanding requirements on alignment, surface cleanliness, testing, and thermal management. The wafer is thus increasingly a platform for assembly as well as device fabrication.

Industrial development, economics, and resource use

The industrial history of wafers is closely connected to the development of the transistor and integrated circuit. Early semiconductor electronics used germanium and relatively small crystals. Silicon became dominant as purification, single-crystal growth, diffusion, and oxidation processes matured. The planar process developed by Jean Hoerni in 1959 made a protective silicon dioxide surface central to practical device manufacture. Robert Noyce’s work on planar integrated circuits helped establish the technological basis for fabricating and interconnecting multiple devices on one substrate.

Successive diameter increases made it possible to process more device area per manufacturing cycle. The transition to 200 mm became important during the 1990s, and high-volume 300 mm production expanded in the early 2000s. Larger diameter complemented the device scaling associated with Moore’s law, but it was a separate engineering development. A smaller transistor feature does not require a larger wafer, and a larger wafer does not by itself improve an individual transistor. Their economic effects interact through equipment throughput, process cost, die size, and yield.

Proposals for 450 mm manufacturing received substantial industrial attention, but the format did not replace 300 mm as the mainstream platform for advanced high-volume silicon integrated circuits. The obstacle was not simply whether a large crystal could be grown. An entire ecosystem of process tools, automation, inspection systems, and facilities would need costly development and qualification. Investment decisions also had to compete with other routes to improved economics, including process advances and new device architectures. Meanwhile, established smaller-diameter production remained useful for analog, power, sensing, and other products with different cost structures.

The supply chain distinguishes wafer manufacturers, which produce substrates, from fabrication plants that build devices on them. Major silicon wafer suppliers include Shin-Etsu Handotai, SUMCO, GlobalWafers, Siltronic, and SK Siltron. Semiconductor foundries manufacture devices for external designers, while integrated device manufacturers combine design and fabrication within their businesses. A substrate is qualified against a specific process rather than accepted solely because it meets a nominal diameter and resistivity. Changing suppliers or substrate specifications can therefore require extensive evaluation.

Resource use extends across purification, crystal growth, finishing, and device manufacture. Crystal growth consumes energy at high temperature; slicing and polishing consume material and generate waste streams; cleaning requires high-purity chemicals and ultrapure water. Device fabrication adds further demands for electricity, gases, and chemical treatment. Monitor and test wafers can sometimes be reclaimed by removing deposited layers and restoring the surface, although reclamation is limited by remaining thickness, contamination, and intended use. Manufacturing efficiency consequently depends on material utilization and yield as well as reduced consumption per process step: losing a wafer late in fabrication discards the resources invested throughout its entire processing history.