Every chip that powers a modern processor, memory module or image sensor depends on a sequence of patterning steps that transfer circuit geometries from a photomask onto a silicon wafer. At the heart of each of those steps lies a deceptively simple question: when the etch process removes material from one layer, how well does it leave the layers beneath it intact? The answer is captured in a single metric called etch selectivity, and as circuit dimensions continue to shrink toward the single-digit nanometre scale, that metric has become one of the most consequential numbers in semiconductor manufacturing.
This article builds understanding of etch selectivity from the ground up. It starts with a precise definition, then traces how selectivity shapes the quality of patterned features, why the challenge intensifies at advanced nodes, which material properties determine etch resistance, and how hard masks address the problem in practice. The final section covers how process engineers diagnose and resolve the selectivity failures that arise in high-volume manufacturing.
What is etch selectivity in semiconductor manufacturing?
Etch selectivity is the ratio of the etch rate of a target material to the etch rate of a reference material under identical process conditions. In formal terms it is expressed as S = Rtarget / Rreference, where R represents the etch rate, typically measured in nanometres per minute.
In practice, selectivity describes how much faster the process removes the layer you want to open compared with the layer you want to protect. A selectivity of 10:1 means the target material etches ten times faster than the underlying or adjacent material. A selectivity of 1:1 means both materials erode at the same rate, which is almost always unacceptable in a precision patterning context.
For example, consider a simple two-layer stack in which a patterned photoresist sits on top of a silicon dioxide film. The etch process must open the oxide where the resist has been exposed and developed, while leaving the silicon substrate below the oxide essentially untouched. If the oxide-to-silicon selectivity is high, the process stops cleanly at the substrate surface. If selectivity is low, the silicon is attacked before the oxide is fully cleared, and the resulting features are distorted or damaged. Selectivity is therefore not an abstract figure of merit but a direct determinant of whether a patterning step produces usable devices.
How etch selectivity shapes pattern fidelity
Pattern fidelity refers to how accurately the final etched feature replicates the intended geometry defined by the lithography step. Etch selectivity influences fidelity through several interconnected mechanisms.
Vertical profile control
When selectivity between the mask layer and the film being etched is high, the etch process can sustain a vertical sidewall profile throughout the full depth of the feature. If the mask erodes too quickly relative to the film, the mask edge recedes laterally and the sidewall angle tilts, broadening the feature at the top and narrowing it at the bottom. This profile distortion translates directly into critical dimension (CD) variation across the wafer.
Loading and microloading effects
Selectivity also interacts with the local pattern density on the wafer. In dense arrays, reactive species in the plasma are consumed more rapidly than in isolated features, which means local etch rates differ depending on neighbourhood geometry. Materials with inherently higher etch resistance are less sensitive to these loading effects, effectively buffering the process against pattern-density-dependent CD variation.
Endpoint detection reliability
Many etch tools rely on optical emission spectroscopy or interferometry to detect when a target layer has been fully cleared. The signal change at the endpoint is strongest when the underlying layer has a clearly different etch rate from the layer being removed. Poor selectivity blurs the endpoint signal, making it difficult to stop the etch at the right moment and increasing the risk of over-etching into underlying structures.
Why selectivity becomes critical at advanced nodes
Building on the pattern fidelity principles established above, it becomes clear why selectivity demands grow sharply as node dimensions shrink. At older process nodes, feature heights were large relative to their widths, which meant mask layers had enough physical thickness to survive the etch without being consumed before the underlying film was cleared. At advanced nodes, that geometric margin disappears.
At the 7 nm node and below, the aspect ratios of features increase while the absolute thickness of patterning layers must decrease to maintain resolution. A photoresist film that is 30 nm thick carries very little etch budget. If the resist erodes at even a modest fraction of the rate of the film beneath it, the resist will be consumed before the etch is complete, and the pattern collapses or is transferred incorrectly.
EUV patterning introduces an additional complication. EUV photoresists are formulated to absorb 13.5 nm wavelength radiation efficiently, and the chemistries that maximise photon absorption do not always produce the highest etch resistance. This creates a direct tension between lithographic sensitivity and pattern-transfer durability, a tension that grows more acute with each successive node. The industry response to this tension is the multilayer patterning stack, in which the resist is decoupled from the etch burden by interposing one or more intermediate layers specifically engineered for high etch resistance.
Material properties that determine etch resistance
Etch resistance in plasma or dry etching environments is not a single property but the outcome of several interacting material characteristics. Understanding these properties helps engineers select or design materials that will survive aggressive etch conditions.
- Elemental composition: Materials rich in silicon, carbon or metal atoms tend to form involatile reaction products when exposed to fluorine- or oxygen-based plasmas. These products accumulate on the surface and act as a self-limiting passivation layer that slows further erosion. In contrast, organic polymers composed primarily of carbon and hydrogen are readily converted to volatile species and etched away quickly.
- Bond energy and density: Stronger and more densely packed chemical bonds require more energy to break. Inorganic and hybrid inorganic-organic materials, such as siloxane networks, typically have higher bond energies than purely organic resists, which contributes to their superior etch resistance.
- Cross-link density: Highly cross-linked polymer networks are physically more resistant to ion bombardment and chemical attack. Thermal or UV curing steps that increase cross-link density after deposition can meaningfully improve etch performance.
- Film density: Denser films present fewer pathways for reactive plasma species to penetrate below the surface. Spin-on materials that are formulated and cured to achieve high film density therefore outperform their porous or loosely packed counterparts under identical plasma conditions.
- Thermal stability: Plasma processes generate localised heat at the wafer surface. Materials that maintain their structural integrity at elevated temperatures resist the softening and reflow that can distort feature profiles during etching.
How hard masks solve the selectivity challenge
With the material properties above in mind, the role of the hard mask in advanced patterning becomes straightforward to understand. A hard mask is a thin film inserted between the photoresist and the layer to be etched. Its function is to receive the pattern from the resist in a first etch step and then act as the primary etch mask in a second, more aggressive etch step that opens the target film.
This two-stage transfer decouples the lithographic requirements of the resist from the etch-resistance requirements of the mask. The resist only needs to survive a relatively mild etch that opens the hard mask, which is a far less demanding task than surviving the full etch of the underlying film. The hard mask, meanwhile, is engineered specifically for etch resistance rather than for photon absorption or dissolution contrast.
Spin-on hard masks are particularly well suited to advanced-node manufacturing because they can be deposited uniformly over complex topography without the line-of-sight limitations of physical vapour deposition. Silicon-rich spin-on hard masks, for example, leverage the involatile silicon oxide passivation mechanism described in the previous section to achieve high selectivity against both oxide and nitride films in fluorine-based plasmas. Carbon-rich spin-on hard masks offer complementary selectivity in oxygen-based plasma environments. In multilayer stacks, silicon and carbon hard mask layers are often combined, with each layer optimised for a specific etch chemistry in the process sequence.
PiBond’s patterning materials, which include silicon-based anti-reflective coatings and spin-on hard masks produced from proprietary siloxane monomers and resins, are designed to address precisely this selectivity challenge in high-volume manufacturing environments.
Diagnosing and resolving common etch selectivity failures
Even well-designed patterning stacks encounter selectivity failures in production. Recognising the signatures of different failure modes is the first step toward an effective resolution.
Mask erosion and CD loss
When the mask layer erodes faster than expected, the most visible symptom is a systematic increase in the critical dimension of etched features, often accompanied by rounded top corners and tapered sidewalls. The root cause is typically either a process drift in plasma power or chemistry that increases the mask etch rate, or a change in the mask film itself, such as a reduction in cure temperature or a shift in film composition. Resolution involves characterising the mask etch rate independently, identifying whether the drift is in the process or the material, and adjusting accordingly.
Etch stop failures
An etch stop failure occurs when the process fails to halt at the intended interface and continues into the underlying layer. This is the direct consequence of insufficient selectivity between the target film and the stop layer. Engineers address this by reviewing the selectivity ratio under the specific plasma chemistry in use, increasing the concentration of passivating species in the plasma, or switching to a stop layer material with inherently higher resistance to the etch chemistry.
Pattern collapse in high-aspect-ratio features
As aspect ratios increase, narrow features become mechanically fragile. If the etch process undercuts the base of a feature due to isotropic etching components in the plasma, the feature loses its mechanical support and collapses. Increasing the directionality of the plasma by raising bias power or adjusting the ion energy distribution reduces the isotropic component and improves sidewall integrity.
Microloading-driven uniformity failures
Non-uniform CD across a die or wafer that correlates with local pattern density is the hallmark of microloading. Addressing this requires either adjusting the plasma chemistry to reduce the sensitivity of the etch rate to reactive species depletion, modifying the patterning stack to use a material with lower intrinsic etch rate variability, or applying computational etch correction techniques at the design stage.
Etch selectivity is ultimately a systems problem. The resist, the hard mask, the etch chemistry and the underlying film stack must all be co-optimised to achieve the pattern fidelity that advanced-node devices require. Contact PiBond to discuss your material requirements and explore how purpose-engineered patterning materials can address the selectivity challenges in your specific process integration context.
Related Articles
- Can dielectric material choice reduce electromigration risk in copper interconnects?
- CMOS image sensor materials: what engineers need from a development partner
- Spin-on hard mask materials: supplier capabilities and qualification considerations
This content was generated with the help of AI and it may contain mistakes