{
  "id": "finfet",
  "title": "Fin Field-Effect Transistor",
  "lang": "en",
  "source_lang": "en",
  "summary": "A FinFET is a multigate transistor whose fin-shaped semiconductor channel improves electrostatic control, enabling dense, energy-efficient integrated circuits.",
  "category": "computer-science",
  "subcategory": "hardware",
  "type": "concept",
  "aliases": [
    "FinFET",
    "fin field-effect transistor",
    "fin-type field-effect transistor"
  ],
  "generated_at": "2026-11-03",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/finfet/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/finfet/",
      "json": "https://aiwiki.page/zh/finfet.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "A fin field-effect transistor, usually abbreviated FinFET, is a field-effect transistor in which the conducting channel occupies a narrow, upright fin of semiconductor material. A gate electrode extends over two or more surfaces of the fin, allowing it to control the channel more effectively than the single top gate of a conventional planar transistor. Most commercially important FinFETs are silicon devices belonging to the metal–oxide–semiconductor field-effect transistor family.\n\nFinFETs became central to advanced integrated circuits because they reduced the leakage and loss of gate control that increasingly constrained planar transistors as dimensions shrank. Their three-dimensional geometry supported continued increases in transistor density and improved energy efficiency in processors, mobile chips, and other digital systems. Intel introduced high-volume commercial tri-gate technology at its 22-nanometer generation, with processors shipping in 2012; foundries subsequently adopted FinFET processes for a broad range of customers.\n\nThe term describes a device architecture rather than a particular manufacturing generation. FinFETs differ in substrate construction, gate coverage, channel materials, dimensions, and fabrication sequence. Their operation connects semiconductor electrostatics with demanding manufacturing requirements and circuit-design rules. At the most advanced logic generations, the same need for stronger gate control has driven a transition toward transistors whose gates completely surround their channels."
    },
    {
      "heading": "Device geometry and principal variants",
      "text": "A FinFET has the source, channel, and drain regions characteristic of a transistor, but its channel is not confined to a broad, horizontal surface. Instead, a narrow semiconductor ridge projects above the surrounding isolation material. The source and drain lie along the length of this ridge, while the gate crosses it. An insulating gate dielectric separates the gate conductor from the semiconductor, so the gate voltage changes channel conductivity primarily through an electric field rather than through direct current injection.\n\nThree dimensions must be distinguished. Gate length, \\(L_g\\), is the channel dimension along the source-to-drain direction beneath the gate. Fin thickness, \\(T_{\\mathrm{fin}}\\), is the lateral width of the ridge. Fin height, \\(H_{\\mathrm{fin}}\\), is its exposed vertical extent. Narrowing the fin strengthens electrostatic control, while increasing its height provides more conducting surface without a proportionate increase in lateral footprint. These dimensions therefore serve different purposes and cannot be represented adequately by a single manufacturing-node number.\n\nFor a rectangular tri-gate fin, an approximate effective channel width is\n\n\\[W_{\\mathrm{eff}}\\approx 2H_{\\mathrm{fin}}+T_{\\mathrm{fin}}.\\]\n\nThe two sidewalls contribute \\(2H_{\\mathrm{fin}}\\), and the top surface contributes approximately \\(T_{\\mathrm{fin}}\\). A transistor with \\(N_{\\mathrm{fin}}\\) equivalent parallel fins has a nominal effective width of approximately \\(N_{\\mathrm{fin}}(2H_{\\mathrm{fin}}+T_{\\mathrm{fin}})\\). This geometrical estimate is useful for circuit sizing, but it does not imply identical current density everywhere: crystal orientation, corner shape, carrier confinement, and gate-stack properties can make the top and sidewall channels behave differently.\n\nIn a double-gate FinFET, the principal gates act on the two sidewalls, sometimes with a relatively thick insulating cap preventing strong top-surface gating. A tri-gate device also uses the top surface actively. Commercial terminology is not completely uniform, and “FinFET” often serves as an umbrella term for both arrangements. Most production devices electrically connect their gated surfaces to one common gate electrode. Independently controlled side gates have been investigated for threshold adjustment and specialized circuits, but are not the usual configuration in mainstream logic processes.\n\nFinFETs can be fabricated on bulk silicon or on silicon-on-insulator substrates. In the latter, a buried oxide separates the fin from the underlying wafer. In bulk technology, the fin remains connected to the substrate below the active region, making isolation and suppression of unwanted conduction beneath the gate especially important. Despite their vertical cross-section, conventional logic FinFETs normally carry current laterally, parallel to the wafer surface; they are not vertical power transistors."
    },
    {
      "heading": "Electrostatic control and short-channel behavior",
      "text": "The main reason for using a fin is improved control of the channel’s electrical potential. In a long-channel planar MOSFET, the gate largely determines whether a conducting inversion layer connects source and drain. As the channel becomes shorter, the source and drain electric fields penetrate a larger fraction of the channel. The drain can then lower the source-to-channel energy barrier even when the gate is intended to keep the transistor off. This is one of the principal short-channel effects that constrain transistor scaling.\n\nAn important manifestation is drain-induced barrier lowering, or DIBL. It is commonly characterized by the decrease in threshold voltage per increase in drain voltage, with values often expressed in millivolts per volt. Strong DIBL increases off-state current and makes switching behavior more dependent on the voltage at the output terminal. Related effects include threshold-voltage roll-off as gate length decreases and, in poorly controlled structures, punch-through between source and drain.\n\nA narrow fin improves control because the gate acts from opposing surfaces. If the semiconductor body is sufficiently thin, the electric potential through its thickness is strongly constrained by those gates rather than by the drain. The body can be fully depleted of mobile carriers in the off state, without leaving a substantial interior region poorly controlled by the gate. Electrostatic scaling-length models capture this advantage: the characteristic distance over which terminal fields disturb the channel decreases as the body becomes thinner and the gate dielectric becomes electrically thinner. Exact expressions depend on geometry and material assumptions.\n\nThe sharpness of switching below threshold is described by the subthreshold swing,\n\n\\[S=\\frac{dV_G}{d\\log_{10}I_D},\\]\n\nwhere \\(V_G\\) is gate voltage and \\(I_D\\) is drain current. For a conventional thermionic MOSFET in equilibrium near room temperature, the ideal limit is approximately \\(60\\ \\mathrm{mV}\\) per decade of current. A FinFET can approach this limit more closely than a poorly controlled short-channel planar transistor, but its geometry does not remove the underlying thermal limit. Interface traps, parasitic coupling, and other nonidealities increase the measured swing.\n\nBetter electrostatics therefore permit lower leakage at a given threshold voltage, or a lower threshold at a comparable leakage target. They do not eliminate every off-state current path. Gate-dielectric tunneling, junction leakage, gate-induced drain leakage, and direct source-to-drain tunneling remain relevant under suitable conditions. The practical advantage is a better balance among channel length, operating voltage, switching current, and leakage—not an intrinsically perfect off state."
    },
    {
      "heading": "Carrier transport and electrical performance",
      "text": "When the gate voltage exceeds the threshold appropriate to the device polarity, mobile carriers populate the gated fin surfaces and connect source to drain. An n-channel FinFET conducts mainly through electrons; a p-channel device conducts mainly through holes. Together they implement complementary metal–oxide–semiconductor circuits. The gate controls carrier density, but the resulting current also depends on transport through the channel and resistance in the source, drain, and contacts.\n\nLong-channel MOSFET equations provide useful intuition but become incomplete for aggressively scaled FinFETs. Carrier velocity does not continue increasing linearly with electric field, and transport can become quasi-ballistic when the channel length is comparable to relevant scattering lengths. Models must account for velocity saturation, injection from the source, quantum confinement, and the voltage lost in access regions. A transistor with excellent gate control can still deliver disappointing current if contact or source/drain resistance is excessive.\n\nThe crystalline surfaces of a fin are not electrically interchangeable. Depending on wafer orientation and fin direction, the sidewalls and top expose different silicon crystal planes, affecting carrier mobility and scattering. The fin’s small dimensions also confine carriers, modifying their allowed energy states and the distribution of charge within the body. At sufficiently small thicknesses, a purely classical picture of an infinitesimally thin surface channel is inadequate. Charge may extend substantially through the fin, while confinement changes threshold voltage and transport.\n\nStrain engineering improves performance by modifying semiconductor band structure. Tensile strain is commonly useful for electron transport, while compressive strain can enhance hole transport in suitable orientations. Embedded silicon–germanium source/drain regions are a prominent means of introducing compressive stress in p-channel devices. Such regions can also help reduce resistance when appropriately doped and contacted. The resulting benefit depends on the complete structure: fin dimensions, crystal direction, source/drain shape, thermal history, and gate construction all affect how stress reaches the channel.\n\nCircuit speed depends on the relationship between current and capacitance, not on drive current alone. A useful first-order delay estimate is\n\n\\[t_d\\sim\\frac{C_{\\mathrm{load}}V_{\\mathrm{DD}}}{I_{\\mathrm{on}}},\\]\n\nwhere \\(C_{\\mathrm{load}}\\) is the switched load capacitance, \\(V_{\\mathrm{DD}}\\) is supply voltage, and \\(I_{\\mathrm{on}}\\) is a representative on-state current. Taller or more numerous fins generally increase available current, but also increase gate charge and some parasitic capacitances. Gate-to-source/drain fringe fields, contact geometry, and interconnect loading can therefore offset part of a transistor-level improvement. Performance comparisons require consistent voltage, leakage, loading, and area assumptions rather than a simple comparison of fin height or nominal node labels."
    },
    {
      "heading": "Fabrication and process integration",
      "text": "FinFET fabrication begins with a semiconductor wafer and the definition of narrow, regular semiconductor fins. In a common bulk-silicon flow, a hard mask defines the fin pattern, anisotropic etching creates the ridges, and insulating material fills the spaces between them. The isolation dielectric is subsequently recessed to expose the required fin height. Fin thickness, sidewall angle, surface roughness, and height uniformity all influence the eventual transistor, so this early patterning sequence has electrical consequences throughout the process.\n\nThe fin pitch can be smaller than the spacing directly printable with a particular exposure system. Photolithography has therefore been combined with spacer-based multiple-patterning techniques. These use sidewall spacers formed around an initial pattern to create more closely spaced features. Cutting or blocking steps then remove unwanted segments. Extreme-ultraviolet lithography has replaced some multiple-patterning sequences in advanced processes, but it does not remove the need for precise etching, alignment, defect control, or additional pattern-definition steps.\n\nThe gate stack typically combines a high-permittivity dielectric with a metal electrode, collectively described as high-k/metal-gate technology. Hafnium-based dielectrics provide a relatively large physical thickness while retaining a small equivalent oxide thickness, reducing direct gate tunneling compared with an equally effective ultrathin silicon-dioxide layer. Conformal deposition is essential because the dielectric and electrode must cover the fin sidewalls and top. Atomic layer deposition is particularly useful for forming controlled thin films on such three-dimensional surfaces.\n\nMany advanced processes employ a replacement-metal-gate, or gate-last, sequence. A temporary gate defines the channel region while spacers, source/drain structures, and selected high-temperature steps are completed. The temporary material is then removed and replaced by the final gate dielectric and metal stack. Work-function engineering establishes different threshold voltages for n-channel and p-channel devices and for different leakage–performance options. The detailed sequence varies between manufacturers, and integration choices are closely coupled to the available materials and thermal budget.\n\nSource and drain regions are often enlarged by selective epitaxial growth. Raised regions provide more volume for doping and contacting than the narrow fin alone, and adjacent source/drain fins may merge into a common epitaxial structure. Contacts and multiple levels of metal wiring then connect the devices into circuits. Throughout this sequence, doping remains important even when the channel itself is lightly doped: heavily doped source/drain regions, wells, and isolation-related implants serve distinct functions. Yield depends on controlling not only nominal dimensions but also rare defects, gate-fill voids, residues, contact failures, and local pattern-dependent variations."
    },
    {
      "heading": "Research origins and commercial adoption",
      "text": "FinFET technology emerged from decades of research into multigate and fully depleted transistors rather than from a single isolated invention. As planar MOSFETs became smaller, researchers investigated structures in which more than one gate surface controlled a thin semiconductor body. These approaches sought to preserve switching behavior without relying solely on increasingly heavy channel doping, which introduces mobility penalties and statistical variability. Thin-body devices and silicon-on-insulator technology supplied important experimental foundations.\n\nA notable predecessor was the DELTA device, reported by Digh Hisamoto and colleagues in 1989. Its name referred to a fully depleted lean-channel transistor, and its raised semiconductor body anticipated important features of later fin-based structures. During the 1990s, research clarified how narrow bodies and gates on opposing sides could suppress short-channel effects. A University of California, Berkeley program led by Chenming Hu, with collaborators including Hisamoto, Tsu-Jae King, and Jeffrey Bokor, developed structures that helped establish the FinFET architecture.\n\nThe term “FinFET” appeared in late-1990s research, and the 2000 paper “FinFET—A Self-Aligned Double-Gate MOSFET Scalable to 20 nm” became a widely cited account of the concept. Its significance lay not simply in showing a transistor with a three-dimensional shape, but in demonstrating a promising route to strong electrostatic control with fabrication methods related to existing silicon processing. Subsequent research addressed manufacturability, parasitic resistance, strain, gate materials, and integration into complete CMOS technologies.\n\nIntel announced its 22-nanometer tri-gate technology in 2011 and shipped Ivy Bridge processors using it in 2012. This marked a major commercial transition from planar transistors to a three-dimensional gate structure in high-volume logic. In 2015, Samsung’s 14-nanometer and TSMC’s 16-nanometer FinFET technologies entered prominent commercial products. The architecture subsequently spread across mobile processors, high-performance computing devices, programmable logic, and custom digital chips.\n\nThese milestones should not be interpreted as a ranking based only on the numbers attached to each process. Manufacturing-node names became technology-generation labels rather than direct measurements of gate length or fin thickness. Processes carrying similar labels can differ substantially in density, metal pitch, power, and performance. FinFET adoption also did not make planar devices obsolete everywhere: mature planar CMOS remained important for cost-sensitive products, analog functions, high-voltage circuits, and technologies whose requirements do not favor the added complexity of narrow fins."
    },
    {
      "heading": "Digital circuits, sizing, and design automation",
      "text": "FinFETs changed the relationship between transistor geometry and circuit design. In a planar process, transistor width can often be adjusted over a relatively continuous range subject to layout rules. In a FinFET process, useful width is largely quantized by the number of fins under a gate. A one-fin, two-fin, or three-fin device represents a discrete sizing choice. Fin height is normally fixed by the manufacturing process rather than selected independently for every transistor.\n\nThis quantization affects standard-cell libraries, which supply predesigned logic gates, storage elements, and other building blocks. Designers select fin counts to balance delay, leakage, input capacitance, and area. The n-channel and p-channel devices in an inverter may require different fin counts to balance rising and falling transitions, but the available ratios are discrete. Cell height, power-rail placement, gate pitch, and contact-access rules constrain which choices are practical. Electrically attractive transistor dimensions are therefore not always the most area-efficient layout solution.\n\nThe energy used in switching remains strongly dependent on supply voltage:\n\n\\[P_{\\mathrm{dynamic}}\\approx\\alpha C_{\\mathrm{sw}}V_{\\mathrm{DD}}^2f,\\]\n\nwhere \\(\\alpha\\) is switching activity, \\(C_{\\mathrm{sw}}\\) is switched capacitance, and \\(f\\) is operating frequency. Improved FinFET electrostatics can support reduced voltage at useful performance, while lower off-state leakage reduces static power. Nevertheless, voltage cannot be reduced without limit: transistor drive weakens, delay becomes more sensitive to variation, and memory stability deteriorates. Different threshold-voltage options allow a chip to mix fast, leakier cells with slower, lower-leakage cells.\n\nStatic random-access memory presents particularly demanding trade-offs. A conventional six-transistor SRAM cell must preserve its state during a read while remaining writable through its access transistors. Fin quantization restricts the strength ratios among pull-up, pull-down, and access devices. Designers consequently coordinate transistor choices with cell layout and read/write-assist techniques. SRAM density also depends on contact placement and wiring rules, so improvements in logic density do not automatically translate into identical improvements in memory bit-cell area.\n\nElectronic design automation incorporates these constraints through process design kits, compact models, extraction rules, and characterized libraries. A compact model must reproduce current, terminal charge, temperature dependence, and relevant geometry effects over a range of bias conditions. BSIM-CMG is a widely used modeling framework for common-gate multigate transistors. At chip scale, designers combine such models with extracted interconnect resistance and capacitance, timing analysis, and statistical variation analysis. The resulting design process is tightly coupled to the manufacturing technology: a circuit schematic alone does not specify a physically valid or performant FinFET implementation."
    },
    {
      "heading": "Analog, radio-frequency, and computing applications",
      "text": "FinFETs became widely used in central processing units, graphics processing units, and systems on a chip. Their combination of useful drive current and reduced leakage suits products containing large numbers of digital switching devices. In mobile systems, leakage matters both during active operation and while blocks are idle. In high-performance processors, higher transistor density and energy efficiency help accommodate more execution resources, larger caches, and specialized accelerators, although interconnect and cooling remain major constraints.\n\nThe benefits are not confined to digital logic. Strong channel control can reduce output conductance and improve intrinsic transistor gain under suitable bias conditions. A common small-signal measure is\n\n\\[A_{\\mathrm{intrinsic}}=\\frac{g_m}{g_{ds}},\\]\n\nwhere \\(g_m\\) is transconductance and \\(g_{ds}\\) is output conductance. Reduced drain influence on the channel can help this ratio. However, analog performance must also account for noise, matching, capacitance, voltage headroom, and linearity. An improvement in electrostatics does not guarantee that every analog circuit becomes easier to implement.\n\nAdvanced logic processes generally operate at relatively low core supply voltages. This reduces the signal swing available to amplifiers and makes stacked transistor arrangements more difficult. Fin-count quantization limits fine adjustment of device aspect ratios, while restrictive layout rules can complicate carefully balanced analog structures. Designers often use longer-than-minimum gate lengths, multiple parallel devices, or specialized process options where available. Thick-oxide or higher-voltage devices may coexist with core transistors, but their availability and permissible operating conditions are process-specific.\n\nFor radio-frequency circuits, relevant device metrics include current-gain cutoff frequency, \\(f_T\\), and maximum oscillation frequency, \\(f_{\\max}\\). High transconductance and short channels can support high-frequency operation, but gate resistance, source/drain access resistance, and parasitic capacitance become decisive. The three-dimensional gate improves channel control while also creating complex fringe-field coupling. Device layout, contact placement, and interconnect design can therefore materially alter RF performance even when nominal fin count and gate length are unchanged.\n\nFinFET logic also supports machine-learning accelerators and other specialized processors, but it does not determine their computational architecture. Arithmetic precision, memory organization, data movement, and parallel execution remain separate design choices. Similarly, the use of FinFETs in a processor does not mean that every transistor in its package, memory subsystem, or associated power circuitry uses the same structure. Modern systems combine multiple processes and device types, often through semiconductor packaging and chip integration, to meet different requirements for density, bandwidth, voltage, and cost."
    },
    {
      "heading": "Variability, thermal effects, and reliability",
      "text": "FinFETs reduce some sources of variation while exposing others. Lightly doped channels can lessen random dopant fluctuation, a problem that becomes severe when the active volume contains only a small number of dopant atoms. Yet small changes in fin thickness, gate length, or sidewall shape can alter threshold voltage, capacitance, and drive current. Line-edge roughness introduced during patterning can become fin-width roughness after etching, making geometric control an important source of device-to-device variation.\n\nMetal gates introduce additional microscopic variability. Different crystalline grains can have different effective work functions, so a small gate may sample only a limited number of grains. Trapped charges in or near the gate dielectric can also shift the channel potential. In very small devices, the capture or emission of a single carrier can produce observable random telegraph noise. Such effects matter especially for low-current analog circuits, small memory cells, and operation near the minimum supply voltage.\n\nSelf-heating is another important consequence of geometry and materials. Heat generated in the channel must flow through the fin, source/drain regions, gate surroundings, and substrate. Narrow semiconductor bodies and surrounding dielectrics can impede this flow, particularly where buried insulating layers are present. Local temperature rise changes mobility, threshold voltage, leakage, and aging rates. Measurements also depend on timescale: a short electrical pulse may probe the device before it reaches the temperature associated with steady-state operation.\n\nEstablished MOS reliability mechanisms remain relevant. Bias-temperature instability changes threshold voltage and other characteristics through trapping and defect processes under sustained gate bias and elevated temperature. Hot-carrier degradation results from energetic carriers interacting with the semiconductor–dielectric system, particularly under high electric fields. Time-dependent dielectric breakdown concerns the progressive formation of damaging conduction paths through insulating layers. Their relative importance depends on the material stack, voltage, temperature, waveform, and operating history rather than on the FinFET label alone.\n\nThree-dimensional structures make reliability assessment spatially complex. Fin corners, sidewalls, and top surfaces can experience different fields and defect populations. Circuit-level consequences likewise depend on use: a small threshold shift may modestly slow a logic path but disturb the operating margin of a tightly balanced memory cell. Qualification therefore combines accelerated stress tests, statistical analysis, physical characterization, and circuit-aware aging models. Separately, the wires and contacts connecting FinFETs face resistance growth, electromigration, and dielectric reliability limits; robust transistor behavior does not by itself ensure reliable operation of the entire integrated circuit."
    },
    {
      "heading": "Scaling limits and the transition to gate-all-around devices",
      "text": "FinFET scaling involves several dimensions that cannot be reduced independently without consequences. Shorter gates increase the need for a thinner body to preserve electrostatic control. Thinner fins, however, intensify confinement and surface sensitivity, complicate fabrication, and can increase resistance. Taller fins offer more effective width per unit footprint, but very high aspect ratios make etching, mechanical stability, gate coverage, and contact integration more difficult. Reducing the number of fins per transistor eventually reaches the indivisible limit of a single fin.\n\nParasitic elements increasingly determine the benefit of further scaling. As the conducting cross-section shrinks, contact and access resistance can consume a larger fraction of the available voltage. Closer contacts and gates increase capacitive coupling, while narrower metal wires tend to have higher resistance. Improvements in transistor density associated with Moore’s law therefore do not automatically produce proportional improvements in clock frequency or energy per operation. Design restrictions and wiring demand can also limit how much nominal device density is usable.\n\nThe principal architectural successor is the gate-all-around transistor, in which the gate completely surrounds the channel cross-section. Nanowire and nanosheet implementations extend the multigate principle beyond the accessible top and sides of a fin. Stronger enclosure improves electrostatic control, allowing useful channel geometries at dimensions where a conventional fin becomes increasingly difficult to optimize. This is an evolution within field-effect transistor technology rather than a replacement of CMOS logic with a fundamentally different switching mechanism.\n\nStacked nanosheet devices offer an additional sizing variable. Several horizontal sheets can be placed above one another, and sheet width can be varied within process constraints. This provides more flexibility in effective channel width than fixed-height fins with integer fin counts. A common fabrication approach grows alternating silicon and silicon–germanium layers, patterns the stack, and selectively removes sacrificial material to release the channels before completing the surrounding gate. Inner spacers, selective etching, sheet stability, and gate filling introduce substantial integration challenges.\n\nThe transition is not uniform across manufacturers or product categories. Samsung announced initial production using a 3-nanometer gate-all-around process in 2022, while other 3-nanometer-class technologies continued to use FinFETs. This coexistence illustrates why a node label does not uniquely identify transistor architecture. FinFET processes remain useful where their performance, manufacturing maturity, available design infrastructure, and cost meet product requirements. Further development also extends beyond the channel itself, including new contact materials, backside power delivery, improved interconnects, and three-dimensional integration, each addressing constraints that stronger gate control alone cannot remove."
    }
  ],
  "links": [
    "field-effect-transistor",
    "semiconductor",
    "mosfet",
    "integrated-circuit",
    "transistor",
    "cmos",
    "semiconductor-wafer",
    "photolithography",
    "extreme-ultraviolet-lithography",
    "semiconductor-doping",
    "logic-gate",
    "static-random-access-memory",
    "electronic-design-automation",
    "central-processing-unit",
    "graphics-processing-unit",
    "system-on-a-chip",
    "machine-learning",
    "semiconductor-packaging",
    "moores-law",
    "gate-all-around-transistor"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}