CMP Machine Endpoint Detection: Optical, Friction-Based & In-Situ Metrology Methods
Knowing precisely when to stop polishing is as critical to CMP success as the polishing process itself. Endpoint detection (EPD) systems integrated into modern CMP machines provide the real-time, in-situ measurement capability that enables wafer-to-wafer repeatable, stop-on-film process control — a level of precision that fixed-time polishing recipes alone cannot achieve. This guide details the three primary endpoint detection methods used in production CMP, their underlying physics, and the application contexts where each is most effective.
This article is part of the JEEZ CMP knowledge base. For the complete equipment overview, see: CMP Machines: The Complete Guide to Chemical Mechanical Planarization Equipment.
Why Endpoint Detection Matters
Over-polishing and under-polishing each carry distinct, significant yield consequences. Over-polishing removes excess material beyond the target depth, thinning dielectric layers, creating dishing in copper interconnect lines, or eroding barrier metals — directly degrading device electrical performance, timing characteristics, and long-term reliability. Under-polishing leaves residual conductive material that can cause inter-layer electrical shorts, or leaves incomplete planarization that propagates as surface topography into subsequent lithography and processing steps, compounding alignment and focus errors at every downstream step.
Fixed-time polishing recipes — simply running each wafer for a predetermined duration — cannot reliably achieve the precision required at advanced process nodes, because incoming film thickness variation, consumable wear state, and process drift all introduce wafer-to-wafer variability that a fixed-time approach cannot compensate for. Real-time endpoint detection hardware directly measures material removal progress as it occurs, allowing the CMP machine to terminate polishing at the precise moment the target condition is reached, regardless of these underlying sources of variability.
Optical Interferometry (In-Situ OES)
Optical interferometry is the most widely deployed endpoint detection method in production CMP, leveraging the same fundamental optical principle used in thin-film thickness metrology throughout semiconductor manufacturing.
Operating Principle
A monochromatic or broadband light source is focused through a transparent window — typically fabricated from sapphire or fused quartz — built directly into the rotating platen surface beneath the polishing pad. As the platen rotates and the window passes beneath the wafer, light reflects from both the top surface of the film being polished and the underlying interface, with these two reflections combining to produce an optical interference pattern. As the film thickness changes during polishing, the optical path length difference between these two reflections changes correspondingly, producing a cyclical interference signal whose oscillation period directly corresponds to the rate of film thickness change.
Strengths and Application Range
Optical EPD provides Angstrom-level sensitivity to film thickness change, enabling real-time, in-situ measurement throughout the polishing process rather than only at a single endpoint trigger moment — a capability that supports advanced process control approaches using the full optical signal trajectory, not just the endpoint trigger. This method is most effective for transparent or semi-transparent films, including oxide, nitride, and polysilicon layers, where light can meaningfully penetrate and reflect from the underlying interface.
Limitations
Optical EPD is significantly less effective for optically opaque metal films such as copper, tungsten, or aluminum, where light cannot penetrate the film to generate the interference signal that the method relies upon. For these applications, optical EPD is typically used during the dielectric or barrier-layer-adjacent portions of a process sequence, supplemented by alternative methods for opaque metal removal monitoring.
Motor Current and Friction-Based EPD
Motor current and friction-based endpoint detection exploits the fact that changes in tribological conditions at the wafer-pad interface produce measurable changes in mechanical friction, detectable through the electrical drive current of the platen or carrier head motors.
Operating Principle
As discussed in our process physics guide, the friction coefficient at the wafer-pad interface depends on the specific material being polished and the lubrication regime in effect. When polishing reaches a film transition point — for example, when bulk copper clears from a dielectric field region, exposing the underlying barrier metal layer — the friction coefficient at the interface changes due to the different mechanical and chemical interaction between the new exposed material and the slurry/pad system. This friction change translates into a detectable shift in the torque required to maintain constant rotational speed, which is measured as a change in the electrical current drawn by the platen or carrier head drive motor.
Strengths and Application Range
Motor current EPD requires no optical access window in the platen, making it mechanically simpler to integrate and effective for any material combination that produces a meaningful friction contrast at the transition point — including transitions between opaque metal films where optical EPD cannot function. This makes motor current sensing a common primary or supplementary endpoint detection method specifically for metal CMP applications, including both copper and tungsten processing steps.
Limitations
The sensitivity of motor current EPD depends on the magnitude of the friction coefficient change at the specific material transition being monitored, which can be relatively subtle for some material combinations, potentially limiting detection sensitivity and timing precision compared to optical or eddy current methods for certain applications.
In-Situ Eddy Current Measurement
In-situ eddy current measurement provides a non-contact, electromagnetic approach to real-time conductive film thickness monitoring, particularly well suited to copper CMP applications where precise endpoint control is especially critical.
Operating Principle
A sensor coil embedded within the platen generates a time-varying electromagnetic field that induces circulating eddy currents within any conductive film present on the wafer surface as it passes over the sensor location. The magnitude of these induced eddy currents — and the corresponding impedance change measured by the sensor coil — depends on the thickness and conductivity of the film, providing a continuous, real-time film thickness measurement as polishing proceeds, with spatial resolution that can be mapped across the wafer radius as the wafer rotates relative to the fixed sensor position.
Strengths and Application Range
Eddy current EPD is particularly effective for copper CMP, where it enables closed-loop control of copper removal to extremely tight endpoint targets — typically within ±0.5 to 1nm equivalent thickness — across the full wafer population. Because the method directly measures conductive film thickness rather than relying on a transition-point signal change, it can support real-time, continuous removal rate feedback throughout the copper removal process, not just a single endpoint trigger.
Limitations
Eddy current measurement is inherently specific to conductive films and provides no useful signal for dielectric material removal monitoring, restricting its application to metal CMP process steps where a conductive film is present throughout the measurement.
Method Comparison and Selection Guidance
| EPD Method | Best Application Fit | Key Limitation |
|---|---|---|
| Optical Interferometry | Transparent/semi-transparent films (oxide, nitride, poly-Si) | Ineffective for opaque metal films |
| Motor Current / Friction | Metal CMP transitions (copper, tungsten); no optical window required | Sensitivity depends on friction contrast magnitude |
| Eddy Current | Copper CMP requiring tight endpoint precision | Requires a conductive film; not applicable to dielectrics |
Method selection should be driven primarily by the optical and electrical properties of the film being removed at each specific process step, with most modern multi-step CMP processes employing different EPD methods — or combinations of methods — across the sequential platens of a single tool, matched to the specific material transition occurring at each step.
Combined and Redundant EPD Approaches
Leading-edge production CMP tools from Applied Materials and Ebara typically integrate two or more endpoint detection methods simultaneously on advanced process steps, providing both redundant confirmation of the endpoint trigger and the ability to cross-validate signal interpretation when ambiguous transitions occur. This combined approach is particularly valuable for the most yield-critical CMP steps — such as copper barrier metal clearing, where the consequences of endpoint timing error are especially significant for device reliability — where the additional engineering cost of integrating multiple EPD methods is well justified by the resulting improvement in endpoint accuracy and confidence.
JEEZ slurry formulations are characterized for optical and chemical consistency to support stable, predictable endpoint detection signal behavior across production lots, helping minimize endpoint-related process variability that can otherwise complicate root cause analysis when EPD timing shifts are observed.
JEEZ slurries are formulated with consistent optical and chemical properties to support reliable endpoint detection across all major EPD methods and production CMP platforms.
From process physics and tool components to consumable selection and endpoint detection, JEEZ provides the technical depth and direct application engineering support semiconductor manufacturers need to run stable, high-yield CMP operations.
Contact the JEEZ Technical Team →よくある質問
Endpoint detection (EPD) refers to real-time, in-situ measurement systems integrated into a CMP machine that monitor material removal as it occurs, signaling the precise moment to stop polishing. It is essential because both over-polishing (causing dishing, erosion, or film thinning) and under-polishing (causing electrical shorts or incomplete planarization) directly degrade device yield, and fixed-time recipes alone cannot reliably compensate for wafer-to-wafer process variability.
Optical interferometry endpoint detection relies on light reflecting from both the surface and underlying interface of a film to generate an interference signal, which requires the film to be transparent or semi-transparent. Copper and other metal films are optically opaque, preventing light from penetrating to generate the necessary interference signal, which is why copper CMP typically relies on motor current/friction-based or eddy current endpoint detection methods instead.
Eddy current EPD uses an embedded sensor coil that generates an electromagnetic field, inducing eddy currents in the conductive copper film whose magnitude depends on film thickness. This provides continuous, real-time film thickness measurement throughout the polishing process, enabling closed-loop control of copper removal to endpoint targets typically within ±0.5 to 1nm equivalent thickness.
Leading-edge production CMP tools from manufacturers like Applied Materials and Ebara typically combine two or more endpoint detection methods on advanced process steps, providing redundant confirmation and cross-validation of the endpoint trigger. This is particularly common for the most yield-critical steps, such as copper barrier metal clearing, where endpoint timing accuracy has significant reliability implications.
Yes. Optical endpoint detection accuracy can be affected by slurry optical properties, while motor current/friction-based endpoint detection depends on consistent friction characteristics at the wafer-pad interface. Lot-to-lot slurry variation in these properties can introduce endpoint timing drift independent of any CMP tool hardware issue, making slurry consistency an important factor in stable endpoint detection performance.
