Megasonic cleaning couples high-frequency acoustic energy, commonly in the high-kilohertz to low-megahertz regime, into a liquid to remove particulate contamination from wafer surfaces. Compared with lower-frequency ultrasonic cleaning, megasonic operation changes bubble dynamics, acoustic streaming, and the balance between removal and structural damage; it does not eliminate cavitation or guarantee gentle cleaning. The technique can appear after CMP, before oxidation or deposition, and after etch or strip steps, but its suitability depends on the incoming residue and exposed structures. As killer-defect dimensions shrink, a useful recipe must remove the relevant particle population without exceeding the damage limit of the wafer, making megasonic cleaning a configurable acoustic-mechanical process rather than a universal solution.

The acoustic field generated by a megasonic transducer produces at least two distinct mechanisms relevant to particle removal — acoustic streaming and cavitation — and their relative contributions depend on frequency, power density, dissolved-gas concentration, liquid properties, and the geometry of both the transducer and the wafer surface. Acoustic streaming is the steady, time-averaged flow induced in the liquid by attenuation of the sound wave. Near a boundary such as the wafer surface, this streaming creates a velocity gradient that exerts drag on adhered particles. Cavitation involves the nucleation, oscillation, and possible collapse of gas or vapor bubbles in the acoustic field. At megasonic frequencies, the cavitation threshold is higher and bubble dynamics differ from lower-frequency ultrasonic cleaning, but cavitation is not absent: dissolved-gas content, local pressure variations, and nucleation sites on the wafer or in the liquid influence whether bubbles form and how they behave. The drag force from acoustic streaming and the pressure impulses or microstreaming from cavitation events both contribute to overcoming the adhesion forces that bind particles to the surface, but separating their individual contributions experimentally is difficult because changing one parameter — such as power — affects both.

Particle adhesion to a wafer surface involves van der Waals attraction, electrostatic interactions, capillary forces when a liquid meniscus forms, and in some cases chemical bonding, and the removal force must exceed the net adhesion for a particle of a given size in a given chemical environment. In a simplified contact model, the van der Waals adhesion force between a spherical particle of radius $R_p$ and a flat surface scales as

where $A_H$ is the Hamaker constant for the particle-liquid-surface system and $d_0$ is the assumed separation. This ideal smooth sphere-plane result illustrates a size trend under fixed conditions; it is neither a lower bound nor a direct prediction for a processed wafer. Roughness, contact deformation, adsorbed films, chemical bonding, and hydrodynamic geometry can change both adhesion and removal forces, so performance must be measured by particle size on representative surfaces.

Frequency selection in megasonic cleaning involves coupled tradeoffs rather than a single optimum, and a frequency that works for one particle, feature, tool, and chemistry may not transfer to another. Frequency changes resonant bubble behavior, attenuation, field uniformity, and near-wall flow; observed damage can decrease in one setup yet remain dominated by local transducer and bubble conditions in another. The Stokes oscillatory boundary-layer thickness $\delta$ provides one useful scaling,

where $\eta$ is dynamic viscosity, $\rho$ is density, and $f$ is frequency; numerical prefactors depend on the convention used. A thinner oscillatory layer can change the velocity gradient near a particle, but delivered acoustic pressure and the actual three-dimensional flow still control the result. Frequency alone therefore cannot define a safe operating window.

Dissolved gas is a process-critical variable because gas species and concentration influence cavitation inception, bubble population, acoustic propagation, particle removal, and damage. Degassing often raises the cavitation threshold, while gas addition can increase bubble nuclei, but neither direction guarantees a better outcome. Experiments show that removal and pattern damage can rise together as dissolved gas changes, and surfactants can alter that relationship. The useful range is therefore tool-, gas-, chemistry-, and structure-specific and must be measured rather than described by a universal monotonic rule.

The chemistry of the cleaning liquid — its pH, ionic strength, surfactant content, and the presence of complexing or passivating agents — interacts with the acoustic field in ways that affect both particle removal and damage. Chemistry changes the zeta potential of particle and wafer surfaces, altering electrostatic adhesion and the energy barrier to redeposition after removal. A first-order view based on DLVO theory suggests that adjusting pH to make particle and surface like-charged creates electrostatic repulsion that may reduce adhesion and prevent reattachment, but DLVO is an idealized framework and real surfaces involve roughness, chemical heterogeneity, and non-DLVO forces, so measured adhesion can differ from the prediction. Surfactants can modify wetting, change interfacial tension, and alter bubble dynamics — all of which feed back into both removal and damage mechanisms. Alkaline chemistries (dilute NH₄OH or TMAH with or without peroxide) and acidic chemistries (dilute HCl, citric acid, or dilute HF blends) each target different adhesion and contamination chemistries, and the acoustic parameters that work with one chemistry may not transfer to another because the liquid properties, surface charges, and bubble nucleation conditions all change.

Damage to wafer features during megasonic cleaning — pattern collapse, line fracture, or delamination — sets an upper bound on usable acoustic exposure. The forces that remove particles also load patterned structures, and survival depends on material properties, geometry, aspect ratio, anchoring, wetting, and the local acoustic field. There is no universal threshold expressed by generator power or frequency alone because delivered pressure and bubble activity vary spatially and with the liquid and tool. Qualification must therefore include damage inspection on representative vulnerable structures, not only blanket wafers.

Process integration of megasonic cleaning requires matching the cleaning step to the preceding process, the incoming contamination, and the requirements of the next film or patterning level, so the cleaning recipe cannot be developed in isolation from the integration context. After CMP, the incoming contamination includes slurry abrasive particles (silica, alumina, or ceria), dissolved metal species, organic inhibitor residues, and corrosion products — each with different adhesion characteristics and chemical removal pathways. After etch, polymer residues and redeposited material may require different chemistry and acoustic conditions. The DI water quality, including resistivity, dissolved oxygen, total organic carbon, and particle counts, sets a floor on the cleanliness achievable regardless of the acoustic parameters. Megasonic cleaning is typically one step in a multi-step cleaning train — it may be followed or preceded by brush scrubbing, chemical immersion, or spray steps — and its performance must be evaluated as part of the complete sequence rather than as a standalone operation.

Define the incoming contamination from the preceding process step and the cleanliness specification for the next level → Select candidate frequency and transducer configuration based on feature geometry and damage sensitivity of the most vulnerable structures → Establish dissolved-gas control range through measurement rather than assuming ambient conditions → Select and qualify the liquid chemistry for particle-surface charge, metal complexation, and compatibility with exposed materials → Map the power-density window by measuring particle removal efficiency by size and damage inspection on patterned structures across the power range → Qualify the recipe across consumable variation, liquid aging, transducer aging, and incoming-wafer variation → Integrate the megasonic step into the complete cleaning sequence and verify that earlier cleaning gains survive subsequent steps → Monitor particle removal by size, damage counts, and acoustic-field uniformity as part of ongoing process control → Requalify when the device layout, film stack, or preceding process changes

Read megasonic cleaning through an acoustic-energy-budget lens: the transducer delivers acoustic energy into the liquid, that energy is partitioned among streaming, bubble dynamics, heating, and attenuation depending on frequency, delivered pressure, gas content, chemistry, and geometry, and the useful fraction must meet the particle specification while feature loading stays below the qualified damage limit—a budget that can tighten as structures become smaller or mechanically less robust.