Megasonic Single-Wafer Cleaner

1. Theory

• Definition: A megasonic cleaner couples acoustic energy at about 0.4 to 2 $\mathrm{MHz}$ into the cleaning liquid on the wafer, removing sub-micron particles by acoustic streaming and controlled bubble activity (Kanegsberg; Beck; Vereecke et al.).
• Boundary layer: The viscous layer that shelters particles thins as $\sqrt{2\nu /\omega }$, about 0.6 $\mu \mathrm{m}$ at 1 $\mathrm{MHz}$, so streaming reaches particles near the surface (Settnes and Bruus; Beck).
• Cavitation: Stable oscillating bubbles clean; transient implosions and micro-jets damage fine patterns (Kanegsberg; Wang et al.).
• Trade-off: Power raises particle removal and damage together, and damage rises steeply for lines below 100 $\mathrm{nm}$ (Vereecke et al.).

2. Working Principle

• Transducer: A piezoelectric transducer near the spinning wafer delivers point-of-use acoustic energy (Beck).
• Liquid: Dilute chemistry or water containing dissolved gas flows over the wafer; degassed liquid removes almost nothing (Beck; Vereecke et al.).
• Bubble control: Power is pulsed so bubbles never reach implosion temperature (Wang et al.).
• Rinse and dry: The single-wafer process finishes with rinse and spin dry in under a minute (Beck).

3. Applications

• Post-etch and post-CMP particle removal: Sub-100 $\mathrm{nm}$ particles on patterned wafers (Vereecke et al.; Wang et al.).
• Damage-sensitive structures: Gates and fins at 28 $\mathrm{nm}$ and below with controlled cavitation (Wang et al.).
• Reduced-chemistry cleans: Replacing hot concentrated SC-1 with dilute chemistry (Beck; RCA Wet Bench page).

Acoustics and Removal

Name	Formula	Related Formulae	Variables/Constants
Acoustic boundary layer	$\delta =\sqrt{\frac{2\nu }{\omega }}$	$\delta \left(1\mathrm{MHz}\right)\approx 0.6\mu \mathrm{m}$	$\begin{array}{r}\nu :kinematicviscosity,\omega =2\pi f;Settnes\\ andBruus:thisformulaandabout0.6\mu \mathrm{m}at\\ 1\mathrm{MHz}inwater;Beck:below0.5\mu \mathrm{m}for\\ megasonicsagainstabout2.5\mu \mathrm{m}for\\ ultrasonics(seetheQ/A)\end{array}$ $\begin{array}{r}(assumed:\nu =1\times 10^{-6}{m}^2/s;computed:2.8\mu \mathrm{m}\\ at40\mathrm{kHz},0.94\mu \mathrm{m}at360\mathrm{kHz},0.56\mu \mathrm{m}at1\\ \mathrm{MHz},0.40\mu \mathrm{m}at2\mathrm{MHz})\end{array}$
Frequency bands	$f_{ultra}=20\text{to}350\mathrm{kHz}$	$f_{mega}=360\text{or}500\text{to}2000\mathrm{kHz}$	$\begin{array}{r}Kanegsberg:ultrasonicsabout20to200\\ \mathrm{kHz}andmegasonics360to2000\mathrm{kHz},\\ smallerbubblesathigherf;Beck:20to\\ 350and500to2000\mathrm{kHz};Vereeckeetal.:\\ toolsat0.7to1\mathrm{MHz}\end{array}$ $\begin{array}{r}(myreading:thetwoarticlesplacethe\\ boundarybetween350and500\mathrm{kHz}\\ differently,whichisamatterof\\ definition)\end{array}$
Particle removal efficiency	$PRE=\frac{N_{pre}-N_{post}}{N_{pre}}\times 100\%$	$PRE\gtrsim 80\%(34\text{to}60\mathrm{nm}\text{particles})$	$\begin{array}{r}Wangetal.:definition,above80percent\\ for60\mathrm{nm}PSL;Vereeckeetal.:about80\\ percentormorefor34\mathrm{nm}SiO2,nearzero\\ indegassedsolutionsbelow0.3\mathrm{ppm}{O}_2;\\ Beck:0.10\mu \mathrm{m}particlesremovedbelow5\\ W/c{\mathrm{m}}^2\end{array}$ $\begin{array}{r}(myreading:thecollapseofPREin\\ degassedliquidshowsthatbubbles,not\\ pressurealone,dothecleaning)\end{array}$

Cavitation and Damage

Name	Formula	Related Formulae	Variables/Constants
Cavitation regimes	$\text{stable: oscillating bubbles, streaming}$	$\text{transient: implosion, micro-jets}\Rightarrow \text{damage}$	$\begin{array}{r}Kanegsberg:ultrasonicbubblesimplode,\\ megasoniccavitationstable;Wangetal.:\\ transientcavitationandmicro-jetsdamage\\ patternstoward40\mathrm{nm};Beck:controlled\\ cavitationavoidserosion,threshold\\ fallingwithtemperature(seetheQ/A)\end{array}$ $\begin{array}{r}(myreading:whetherabubbleimplodes\\ dependsonpowerandtime,notonlyon\\ frequency)\end{array}$
Bubble heating and implosion	$T_{2n}=T_0+n\left(\Delta T-\delta T\right)$	${\tau }_i=t_1\left(\frac{T_i-T_0-\Delta T}{\Delta T-\delta T}+1\right)$	$T_0:initialgastemperature$ $\Delta T:risepercompression$ $\delta T:losspercycle$ $T_i:implosiontemperature$ $\begin{array}{r}{t}_1:cycletime;Wangetal.,Eqs.4and5,\\ powercutbefore{\tau }_{i}andrestartedafter\\ cooling\end{array}$ $\begin{array}{r}(myreading:pulsingthepowerkeepsevery\\ bubblebelow{T}_i,socavitationstays\\ stable)\end{array}$
Damage scaling	$\text{damage}\uparrow \text{sharply for}w<100\mathrm{nm}$	$\text{power}\downarrow \Rightarrow \text{damage}\downarrow ,PRE\downarrow$	$\begin{array}{r}Vereeckeetal.:70\mathrm{nm}polylinesabout\\ 170\mathrm{nm}talldamagedas1to2\mu \mathrm{m}pieces,\\ powerthelargestfactor,pressurestripes\\ abovetransducerelements;Wangetal.:\\ damageonlyatthehighest-powermodeon\\ 28\mathrm{nm}lines\end{array}$ $\begin{array}{r}(computed:170/70givesaspectratio2.4\\ forIMEClines;AC{M}^{\prime }slineswere1:5)\end{array}$

Tool Design and Chemistry

Name	Formula	Related Formulae	Variables/Constants
Single-wafer against batch	$t_{batch}=10\text{to}20\min$	$t_{single}<1\min$	$\begin{array}{r}Beck:batchtanks10to20min,\\ single-waferpoint-of-useacousticenergy\\ sub-minute,dilutechemistryeffective\\ withmegasonics;Vereeckeetal.:batch\\ andsingle-wafertoolscompared\end{array}$ $\begin{array}{r}(myreading:single-wafertoolsdeliver\\ energytoonewaferfromanearby\\ transducer,sopoweranduniformitycanbe\\ controlledmoretightly)\end{array}$
Power density	$I<5W/c{\mathrm{m}}^2$	$v_{stream}\propto I,f$	$I:acousticintensity$ $\begin{array}{r}{v}_{stream}:streamingvelocity;Beck:below5\\ W/c{\mathrm{m}}^{2}effectivefor0.1\mu \mathrm{m}particles,\\ streamingvelocityrisinglinearlywith\\ intensityandfrequency;Wangetal.:30\\ to50Wtests\end{array}$ $\begin{array}{r}(myreading:streamingthinstheboundary\\ layerandcarriesreleasedparticlesaway)\end{array}$
Chemistry and etch budget	$\Delta t_{ox}<0.5\text{A˚}$	$\text{dilute SC-1 with megasonics}$	$\begin{array}{r}Vereeckeetal.:thermaloxideetchkept\\ below0.5\mathring{A};Beck:verydilutesolutions\\ cleanwithmegasonicenergy;RCAWetBench\\ pageforSC-1chemistry\end{array}$ $\begin{array}{r}(myreading:acousticenergyreplacessome\\ oftheSC-1undercut,solesssiliconor\\ oxideisconsumed)\end{array}$

Important Questions

Ques: Why do megasonic frequencies remove smaller particles than ultrasonics?

Because the sheltering boundary layer is thinner. Settnes and Bruus give the viscous boundary layer as $\sqrt{2\nu /\omega }$, about 0.6 $\mu \mathrm{m}$ at 1 $\mathrm{MHz}$ in water; with $\nu =10^{-6}$ m${}^2$/s it is 2.8 $\mu \mathrm{m}$ at 40 $\mathrm{kHz}$ and 0.40 $\mu \mathrm{m}$ at 2 $\mathrm{MHz}$ (computed). Beck quotes below 0.5 $\mu \mathrm{m}$ for megasonics against about 2.5 $\mu \mathrm{m}$ for ultrasonics, which the formula reproduces near 2 $\mathrm{MHz}$ and 40 $\mathrm{kHz}$ (computed), so the sources agree once frequency is specified (my reading). Particles inside a thick layer feel little streaming, so megasonics is effective below 0.3 $\mu \mathrm{m}$ and removes 0.10 $\mu \mathrm{m}$ particles (Beck). Kanegsberg adds that bubbles shrink with frequency and that megasonics suits flat surfaces, not blind holes. Reference: M. Settnes and H. Bruus, ‘On the forces acting on a small particle in an acoustical field in a viscous fluid’, arXiv:1110.6037 (DTU): viscous boundary layer delta = sqrt(2 nu/omega), about 0.6 um for 1 MHz in water at room temperature, radiation forces on small particles (fetch-tool summary); M. Beck (ProSys), ‘Megasonic cleaning charts a course to the big time’, Solid State Technology / Semiconductor Digest: megasonics 500 to 2000 kHz against 20 to 350 kHz ultrasonics, boundary layer below 0.5 um against about 2.5 um, streaming velocity rising with intensity and frequency, effective below 0.3 um and removing 0.10 um particles, below 5 W/cm2 effective for 0.1 um particles, batch tanks 10 to 20 min against sub-minute single-wafer point-of-use systems, damage when energy uniformity and intensity are uncontrolled, dilute solutions effective with megasonics, cavitation threshold falling with temperature (fetch-tool summary); B. Kanegsberg and E. Kanegsberg, ‘Ultrasonics vs. Megasonics’, Solid State Technology / Semiconductor Digest: ultrasonics about 20 kHz for metals, 60 to 80 kHz for delicate parts and 100 to 200 kHz newer, megasonics 360 to 2000 kHz, transient implosions in ultrasonics against stable cavitation without implosions at megasonic frequencies, smaller bubbles at higher frequency, megasonics for sub-micron particles on flat surfaces but not blind holes, low damage likelihood, streaming thinning the boundary layer (fetch-tool summary).

Ques: Is megasonic cavitation always stable, or can it damage patterns?

It can damage, despite claims of stability. Kanegsberg states that at megasonic frequencies the fluid motion gives stable cavitation without implosions, and Beck that controlled cavitation avoids erosion. Wang et al., however, show transient cavitation and micro-jets knocking off 28 $\mathrm{nm}$ poly lines at their highest power, and Vereecke et al. found damage to 70 $\mathrm{nm}$ lines in all five tools, rising sharply below 100 $\mathrm{nm}$ and driven mainly by power. The measured damage is more convincing, so stability depends on power and time rather than frequency alone (my reading). Wang et al. therefore cut power before bubble gas reaches its implosion temperature, ${\tau }_i$, and restart after cooling, and found no damage in their lower modes. Reference: B. Kanegsberg and E. Kanegsberg, ‘Ultrasonics vs. Megasonics’, Solid State Technology / Semiconductor Digest: ultrasonics about 20 kHz for metals, 60 to 80 kHz for delicate parts and 100 to 200 kHz newer, megasonics 360 to 2000 kHz, transient implosions in ultrasonics against stable cavitation without implosions at megasonic frequencies, smaller bubbles at higher frequency, megasonics for sub-micron particles on flat surfaces but not blind holes, low damage likelihood, streaming thinning the boundary layer (fetch-tool summary); M. Beck (ProSys), ‘Megasonic cleaning charts a course to the big time’, Solid State Technology / Semiconductor Digest: megasonics 500 to 2000 kHz against 20 to 350 kHz ultrasonics, boundary layer below 0.5 um against about 2.5 um, streaming velocity rising with intensity and frequency, effective below 0.3 um and removing 0.10 um particles, below 5 W/cm2 effective for 0.1 um particles, batch tanks 10 to 20 min against sub-minute single-wafer point-of-use systems, damage when energy uniformity and intensity are uncontrolled, dilute solutions effective with megasonics, cavitation threshold falling with temperature (fetch-tool summary); D. H. Wang et al. (ACM Research and Shanghai Huali), ‘Damage-Free Cleaning of Advanced Structure Using Timely Energized Bubble Oscillation Megasonic Technology’, Solid State Phenomena 282, 64 (2018): transient cavitation and micro-jets damaging patterns toward 40 nm and below, power switched off before the bubble gas reaches implosion temperature and restarted after cooling, Delta T per compression and T_2n = T_0 + n(Delta T - delta T), implosion time tau_i formula, 30 to 50 W tests, PRE above 80 per cent for 60 nm PSL with PRE = (pre - post)/pre x 100, no damage except at the highest power mode, 28 nm poly lines at aspect ratio 1:5 (fetch-tool summary); G. Vereecke, F. Holsteyns, J. Veltens, M. Lux, S. Arnauts, K. Kenis, R. Vos, P. W. Mertens and M. M. Heyns (IMEC), ‘Evaluation of Megasonic Cleaning Systems for Particle Removal Efficiency and Damaging’, Sensor Products white paper: five batch and single-wafer tools at 0.7 to 1 MHz, 34 nm SiO2 test particles, PRE about 80 per cent or more, near zero in degassed solutions below 0.3 ppm O2, 70 nm poly lines about 170 nm tall damaged as removed or bent 1 to 2 um pieces, damage rising sharply below 100 nm width, power the largest factor with lower power cutting both damage and PRE, pressure-film stripes above transducer elements, oxide etch below 0.5 A, killers above about 45 nm at the 90 nm node (fetch-tool summary).

Ques: How are particle removal and damage traded in a megasonic tool?

Through acoustic power and its uniformity. Vereecke et al. evaluated five tools at 0.7 to 1 $\mathrm{MHz}$ with 34 $\mathrm{nm}$ silica particles and 70 $\mathrm{nm}$ poly lines: all reached about 80 per cent removal, none combined high removal with no damage, and lowering power cut damage and removal together. Pressure-sensitive film showed stripes above transducer elements, matching higher damage. Wang et al. define removal as $(N_{pre}-N_{post})/N_{pre}$ and report above 80 per cent for 60 $\mathrm{nm}$ PSL with no damage except their highest-power mode on 28 $\mathrm{nm}$ lines with aspect ratio 1:5. Beck finds below 5 $W/c{\mathrm{m}}^2$ effective for 0.1 $\mu \mathrm{m}$ particles. The IMEC lines had aspect ratio 2.4 (computed), so taller lines are more fragile (my reading). Reference: G. Vereecke, F. Holsteyns, J. Veltens, M. Lux, S. Arnauts, K. Kenis, R. Vos, P. W. Mertens and M. M. Heyns (IMEC), ‘Evaluation of Megasonic Cleaning Systems for Particle Removal Efficiency and Damaging’, Sensor Products white paper: five batch and single-wafer tools at 0.7 to 1 MHz, 34 nm SiO2 test particles, PRE about 80 per cent or more, near zero in degassed solutions below 0.3 ppm O2, 70 nm poly lines about 170 nm tall damaged as removed or bent 1 to 2 um pieces, damage rising sharply below 100 nm width, power the largest factor with lower power cutting both damage and PRE, pressure-film stripes above transducer elements, oxide etch below 0.5 A, killers above about 45 nm at the 90 nm node (fetch-tool summary); D. H. Wang et al. (ACM Research and Shanghai Huali), ‘Damage-Free Cleaning of Advanced Structure Using Timely Energized Bubble Oscillation Megasonic Technology’, Solid State Phenomena 282, 64 (2018): transient cavitation and micro-jets damaging patterns toward 40 nm and below, power switched off before the bubble gas reaches implosion temperature and restarted after cooling, Delta T per compression and T_2n = T_0 + n(Delta T - delta T), implosion time tau_i formula, 30 to 50 W tests, PRE above 80 per cent for 60 nm PSL with PRE = (pre - post)/pre x 100, no damage except at the highest power mode, 28 nm poly lines at aspect ratio 1:5 (fetch-tool summary); M. Beck (ProSys), ‘Megasonic cleaning charts a course to the big time’, Solid State Technology / Semiconductor Digest: megasonics 500 to 2000 kHz against 20 to 350 kHz ultrasonics, boundary layer below 0.5 um against about 2.5 um, streaming velocity rising with intensity and frequency, effective below 0.3 um and removing 0.10 um particles, below 5 W/cm2 effective for 0.1 um particles, batch tanks 10 to 20 min against sub-minute single-wafer point-of-use systems, damage when energy uniformity and intensity are uncontrolled, dilute solutions effective with megasonics, cavitation threshold falling with temperature (fetch-tool summary).

Ques: Why does dissolved gas matter in megasonic cleaning?

Because bubbles do much of the work. Vereecke et al. found particle removal close to zero in degassed solutions below 0.3 $\mathrm{ppm}$ oxygen, showing that cavitation bubbles, not pressure waves alone, release particles (my reading). Kanegsberg and Beck describe the useful regime as stable cavitation, where bubbles oscillate and drive streaming, which Beck says rises with intensity and frequency and thins the boundary layer. Beck adds that the cavitation threshold falls as temperature rises, so cooler liquid limits violent cavitation. Wang et al. treat each bubble as a gas that heats a little per compression, $T_{2n}=T_0+n(\Delta T-\delta T)$, and switch power off before it reaches implosion temperature. Gas content and temperature are therefore process variables alongside power (my reading). Reference: G. Vereecke, F. Holsteyns, J. Veltens, M. Lux, S. Arnauts, K. Kenis, R. Vos, P. W. Mertens and M. M. Heyns (IMEC), ‘Evaluation of Megasonic Cleaning Systems for Particle Removal Efficiency and Damaging’, Sensor Products white paper: five batch and single-wafer tools at 0.7 to 1 MHz, 34 nm SiO2 test particles, PRE about 80 per cent or more, near zero in degassed solutions below 0.3 ppm O2, 70 nm poly lines about 170 nm tall damaged as removed or bent 1 to 2 um pieces, damage rising sharply below 100 nm width, power the largest factor with lower power cutting both damage and PRE, pressure-film stripes above transducer elements, oxide etch below 0.5 A, killers above about 45 nm at the 90 nm node (fetch-tool summary); B. Kanegsberg and E. Kanegsberg, ‘Ultrasonics vs. Megasonics’, Solid State Technology / Semiconductor Digest: ultrasonics about 20 kHz for metals, 60 to 80 kHz for delicate parts and 100 to 200 kHz newer, megasonics 360 to 2000 kHz, transient implosions in ultrasonics against stable cavitation without implosions at megasonic frequencies, smaller bubbles at higher frequency, megasonics for sub-micron particles on flat surfaces but not blind holes, low damage likelihood, streaming thinning the boundary layer (fetch-tool summary); M. Beck (ProSys), ‘Megasonic cleaning charts a course to the big time’, Solid State Technology / Semiconductor Digest: megasonics 500 to 2000 kHz against 20 to 350 kHz ultrasonics, boundary layer below 0.5 um against about 2.5 um, streaming velocity rising with intensity and frequency, effective below 0.3 um and removing 0.10 um particles, below 5 W/cm2 effective for 0.1 um particles, batch tanks 10 to 20 min against sub-minute single-wafer point-of-use systems, damage when energy uniformity and intensity are uncontrolled, dilute solutions effective with megasonics, cavitation threshold falling with temperature (fetch-tool summary); D. H. Wang et al. (ACM Research and Shanghai Huali), ‘Damage-Free Cleaning of Advanced Structure Using Timely Energized Bubble Oscillation Megasonic Technology’, Solid State Phenomena 282, 64 (2018): transient cavitation and micro-jets damaging patterns toward 40 nm and below, power switched off before the bubble gas reaches implosion temperature and restarted after cooling, Delta T per compression and T_2n = T_0 + n(Delta T - delta T), implosion time tau_i formula, 30 to 50 W tests, PRE above 80 per cent for 60 nm PSL with PRE = (pre - post)/pre x 100, no damage except at the highest power mode, 28 nm poly lines at aspect ratio 1:5 (fetch-tool summary).

Ques: How does a single-wafer megasonic cleaner differ from a batch bath?

In process time and control. Beck reports 10 to 20 min per tank for batch systems against sub-minute processing for single-wafer tools with point-of-use acoustic energy, and notes that very dilute solutions clean effectively when megasonics is added. Vereecke et al. compared batch and single-wafer tools and kept thermal oxide loss below 0.5 angstrom, far less than a hot SC-1 bath removes (RCA Wet Bench page; my reading). A transducer close to a single spinning wafer lets power and uniformity be set for that wafer, which matters because non-uniform energy, as in the pressure stripes Vereecke et al. measured, causes local damage (my reading). Beck also warns that concentrated sources such as nozzles have raised damage concerns. Reference: M. Beck (ProSys), ‘Megasonic cleaning charts a course to the big time’, Solid State Technology / Semiconductor Digest: megasonics 500 to 2000 kHz against 20 to 350 kHz ultrasonics, boundary layer below 0.5 um against about 2.5 um, streaming velocity rising with intensity and frequency, effective below 0.3 um and removing 0.10 um particles, below 5 W/cm2 effective for 0.1 um particles, batch tanks 10 to 20 min against sub-minute single-wafer point-of-use systems, damage when energy uniformity and intensity are uncontrolled, dilute solutions effective with megasonics, cavitation threshold falling with temperature (fetch-tool summary); G. Vereecke, F. Holsteyns, J. Veltens, M. Lux, S. Arnauts, K. Kenis, R. Vos, P. W. Mertens and M. M. Heyns (IMEC), ‘Evaluation of Megasonic Cleaning Systems for Particle Removal Efficiency and Damaging’, Sensor Products white paper: five batch and single-wafer tools at 0.7 to 1 MHz, 34 nm SiO2 test particles, PRE about 80 per cent or more, near zero in degassed solutions below 0.3 ppm O2, 70 nm poly lines about 170 nm tall damaged as removed or bent 1 to 2 um pieces, damage rising sharply below 100 nm width, power the largest factor with lower power cutting both damage and PRE, pressure-film stripes above transducer elements, oxide etch below 0.5 A, killers above about 45 nm at the 90 nm node (fetch-tool summary); RCA Wet Bench page of this vault.

References

1. B. Kanegsberg and E. Kanegsberg, “Ultrasonics vs. megasonics,” Solid State Technology, Semiconductor Digest. Available: https://sst.semiconductor-digest.com/?p=44787
2. M. Beck, “Megasonic cleaning charts a course to the big time,” Solid State Technology, Semiconductor Digest. Available: https://sst.semiconductor-digest.com/?p=44798
3. M. Settnes and H. Bruus, “On the forces acting on a small particle in an acoustical field in a viscous fluid,” arXiv:1110.6037. Available: https://arxiv.org/pdf/1110.6037
4. G. Vereecke, F. Holsteyns, J. Veltens, M. Lux, S. Arnauts, K. Kenis, R. Vos, P. W. Mertens and M. M. Heyns, “Evaluation of megasonic cleaning systems for particle removal efficiency and damaging,” IMEC, Sensor Products white paper. Available: https://old.sensorprod.com/news/white-papers/emc/wp_emc.pdf
5. D. H. Wang et al., “Damage-free cleaning of advanced structure using timely energized bubble oscillation megasonic technology,” Solid State Phenomena, vol. 282, pp. 64-72, 2018. Available: https://www.acmr.com/wp-content/uploads/ACM-Huali-Paper-UCPSS-2018.pdf
6. RCA Wet Bench page of this vault.

Related pages in this vault: RCA Wet Bench, Defect Sources and Classification, Laser-Scattering Particle Inspection, CMP Polisher, Patterned-Wafer Optical Inspection, Spin Rinse Dryer and IPA Vapor Dryer, Vapor HF and Plasma Dry Clean Tool.