This site is an independent educational reference on wafer thickness measurement, drawing on preserved technical literature from the former Philips Advanced Metrology Systems (Philips AMS) of Natick, Massachusetts. The material presented here focuses on the documented principles and published methods associated with non-contact film-thickness metrology, including SurfaceWave™ laser-based techniques, model-based infrared reflectometry, and related approaches for interconnect layers.
The archived content references historical platforms such as the Series 1200, 2300, 3300, and IR3000, and discusses applications in copper/low-k processes, barrier and seed layers, ECD copper, tungsten vias, and CMP for 200mm and 300mm wafers. These excerpts are offered solely for educational review of measurement concepts and industry context.
No current products, services, certifications, or affiliations are claimed. This site is not affiliated with any operating company and provides no commercial or technical support.
Wafer thickness measurement is a foundational metrology step in semiconductor fabrication, affecting lithography focus, etch endpoint, chemical-mechanical planarization (CMP) uniformity, and final device packaging. The measurement is performed non-destructively, meaning the wafer must emerge from the metrology step unchanged in electrical, mechanical, or chemical properties. This requirement eliminates most contact-based gauging methods used in general machining, leaving optical and capacitive techniques as the primary candidates. Understanding the physical mechanisms behind each method, and the distinction between repeatability and accuracy, is essential for choosing the right tool for a given process step.
The most common non-contact method for film and wafer thickness is optical interferometry. A broadband light source (often a tungsten-halogen lamp or a supercontinuum laser) is directed at the wafer surface. The light reflects off the top surface of the film and the underlying substrate interface. These two reflected beams interfere constructively or destructively depending on the optical path difference, which is twice the film thickness multiplied by the refractive index. By analyzing the spectral reflectance across a range of wavelengths, the system extracts the thickness using a model that accounts for the known refractive index dispersion of the film material (e.g., silicon dioxide, silicon nitride, photoresist). For bare silicon wafers, the same principle applies to the native oxide or to the wafer itself if the backside is roughened—but for total wafer thickness, a different approach is used.
For total wafer thickness (typically 500 to 800 micrometers for 200 mm and 300 mm wafers), optical interferometry alone is impractical because the coherence length of broadband light is short relative to the wafer thickness. Instead, manufacturers use either low-coherence interferometry (also called optical coherence tomography) or a chromatic confocal sensor. In low-coherence interferometry, a broadband source is split into a reference arm and a sample arm. The reference arm has a known path length that is scanned mechanically. When the sample arm reflection from the front surface and the back surface match the reference path length within the coherence length, an interference signal appears. The distance between the two interference peaks corresponds to the optical thickness, which is divided by the refractive index of silicon (approximately 3.5 in the near-infrared) to obtain the physical thickness. This method is non-contact, fast, and works on both polished and rough backside surfaces, though rough backside reduces signal amplitude.
Contact methods, such as a stylus profilometer or a capacitance-based probe that physically touches the wafer, are rarely used for production thickness measurement. The primary reason is contamination and damage risk. A stylus can scratch the active device layer, and even a soft contact probe can leave particles or organic residues that later cause yield loss. Additionally, contact methods require a reference surface and are sensitive to wafer bow and warp—if the wafer is not perfectly flat, the probe measures the distance to the chuck, not the true thickness. Non-contact methods eliminate these risks but introduce their own tradeoffs: optical methods require a clean, reflective surface and are sensitive to film roughness, backside texture, and the presence of opaque layers (e.g., metal films) that block the backside reflection. Capacitive sensors, which measure the change in capacitance between a probe and the wafer surface, are non-contact but require a conductive wafer or a conductive film on the surface. They are excellent for measuring total thickness of bare silicon or heavily doped wafers, but they fail on insulating substrates like glass or sapphire.
Repeatability refers to the tool’s ability to produce the same reading on the same wafer under identical conditions over time. It is typically expressed as a standard deviation (1-sigma) of repeated measurements. Accuracy, on the other hand, is how close the measured value is to the true physical thickness, which is usually established by a reference standard traceable to a national metrology institute. In semiconductor fabs, repeatability is often more important than accuracy for process control, because the goal is to detect small changes in thickness from wafer to wafer or across a single wafer. A tool with excellent repeatability but poor accuracy can still be used for run-to-run control, as long as the offset is stable. However, for tool-to-tool matching or for transferring a process from one fab to another, accuracy becomes critical. Optical interferometry is inherently accurate if the refractive index model is correct, but the model can be wrong for films with unknown stoichiometry (e.g., silicon oxynitride). In such cases, the tool may show excellent repeatability but a systematic error of several percent. Calibration against a known standard, such as a certified silicon wafer with a measured thickness, is required to correct this offset.
A single-point thickness measurement is rarely sufficient. Modern fabs require thickness maps across the entire wafer to detect CMP non-uniformity, etch rate variations, or deposition center-to-edge profiles. This introduces a tradeoff between spot size, measurement speed, and spatial resolution. A small spot size (e.g., tens of micrometers) allows fine spatial resolution but requires a longer scan time to cover the wafer. A large spot size (e.g., several millimeters) averages over local roughness and is faster but misses small-scale defects. For production, a typical map uses 49 to 121 points, with a measurement time of a few seconds per point. The choice of spot size also affects the measurement on patterned wafers: a large spot averages over device structures, which may be desirable for film thickness control, but it cannot resolve thickness variations within a single die. For in-line metrology on patterned wafers, the tool must be able to align to specific measurement sites, which requires an optical microscope and pattern recognition. This adds complexity and reduces throughput.
For bare silicon wafer thickness measurement at incoming inspection, a capacitive sensor or a low-coherence interferometer is typical. Capacitive sensors are fast and robust to backside roughness, but they require a conductive wafer. For doped wafers, this is fine; for high-resistivity wafers, the capacitance signal becomes weak, and optical methods are preferred. For film thickness measurement (e.g., oxide, nitride, photoresist), spectral reflectance interferometry is the workhorse because it is fast, non-destructive, and works on transparent films. For opaque films like metals, optical methods fail because the light does not penetrate to the substrate; in that case, a four-point probe measures sheet resistance, which is correlated to thickness but is not a direct thickness measurement. For very thin films (below a few nanometers), ellipsometry is chosen because it measures the change in polarization state of reflected light, which is more sensitive to sub-nanometer thickness changes than simple reflectance. For total wafer thickness on patterned wafers, low-coherence interferometry is preferred because it can measure through the front-side films as long as they are transparent in the near-infrared; however, if the front side has thick metal layers, the backside reflection is blocked, and the measurement fails. In that case, a capacitive sensor on the backside, combined with a front-side optical sensor, can measure total thickness by subtracting the front-side film thickness from the total optical path.
All optical thickness measurements are sensitive to temperature because the refractive index of silicon and most dielectrics changes with temperature. A temperature drift of a few degrees can shift the measured thickness by a small but measurable amount, so production tools often include temperature compensation or are installed in temperature-controlled environments. Additionally, wafer bow and warp affect the measurement because they change the angle of incidence of the light beam. Most tools use a telecentric lens to ensure normal incidence, but extreme bow can still cause errors. Finally, the measurement is only as good as the optical model. For unknown films, a multi-angle or multi-wavelength measurement is required to simultaneously solve for thickness and refractive index, but this increases measurement time and complexity. In practice, process engineers rely on a combination of techniques: a fast, repeatable optical tool for run-to-run control, and a slower, more accurate reference tool (e.g., a spectroscopic ellipsometer) for calibration and model verification. The choice is never about finding the single best tool, but about balancing speed, precision, and robustness against the specific material stack and process requirements.
This independent educational reference summarizes general technical concepts. Verify current standards, dimensions, and manufacturer specifications before making a procurement or engineering decision.