© IMEC 2010 / CONFIDENTIAL Optimizing high frequency ultrasound cleaning in the semiconductor industry Steven Brems

© IMEC 2010 / CONFIDENTIAL Outline ▸ Introduction to particle removal ▸ Improving state-of-the-art megasonic cleaning -Acoustic pulsing -Oversaturated liquids -Traveling waves ▸ Future of particle removal with liquid motion in the semiconductor industry ▸ Conclusions 2

© IMEC 2010 / CONFIDENTIAL Introduction: Particle cleaning ▸ Nanoparticle removal with pure chemical cleaning is only effective if >2 nm material is removed. ▸ A combination of physical and chemical cleaning methods will become more important Particle attached to wafer surface Lift-off from surface: repulsive forces (electrostatic: z) Breaking of the Van der Waals forces (under)etching Transport away from surface: diffusion, convection Mechanism of particle removal by pure chemical cleaning v F nm 20 nm

© IMEC 2010 / CONFIDENTIAL Outline ▸ Introduction to particle removal ▸ Improving state-of-the-art megasonic cleaning -Acoustic pulsing -Oversaturated liquids -Traveling waves ▸ Future of particle removal with liquid motion in the semiconductor industry ▸ Conclusions 4

© IMEC 2010 / CONFIDENTIAL Towards a control of bubble size: Pulsing ▸ At sufficiently high gas concentration and acoustic pressures, bubbles can grow by rectified diffusion and bubble coalescence ▸ Microbubbles (< 4  m) will always shrink when ultrasound is turned off and dissolved gas saturation is below 130% -Bubbles could kept around resonance radius by turning the acoustic field on (bubbles grow) and off (bubbles dissolve) J. Lee et al., JACS 127, (2005) Pulse on timePulse off time 5

© IMEC 2010 / CONFIDENTIAL In-situ measuring micro-bubble activity oscilloscope amplifier Hydrophone Wafer Transducer Example of cavitation noise spectra ▸ Bubble oscillation -Frequency distribution of the oscillating bubble motion can contain harmonics, subharmonics and ultraharmonics  The components arise from the nonlinear motion of a bubble  acoustic emission ▸ Non-integer harmonics (5f 0 /2, 7f 0 /2, 9f 0 /2…) : -Particular characteristic of non-linear (stable) bubble motion  Can be used as an indicator for bubble activity ▸ Strong (transient) cavitation produces white noise (increase of background signal) -Instable cavitation = damaging cavitation 6

© IMEC 2010 / CONFIDENTIAL Cavitation noise spectra: Influence of pulses ▸ Experimental details -Oxygen concentration: 120 %, applied power: 640 mW/cm 2 -Duty Cycle is varied ▸ Optimal pulse off time (indicated with ) is independent of duty cycle variation ▸ Bubble activity decreases with increasing duty cycle ▸ However, a lower DC also means a lower effective cleaning time! 7 -8dB=40% DC 10% DC 25% DC 50%

© IMEC 2010 / CONFIDENTIAL Understanding of optimal pulse off time Dissolved oxygen concentration 120% ~ resonant bubble size Dissolution time resonant bubble The dissolution time of a resonant bubble lies very close to the optimal experimental determined pulse off time Bubble size ‘reservoir’ Lost bubbles Dissolution during pulse-off time Growing to active size during pulse-on time Production of new bubbles (transient collapse, shape instabilities) Bubble size distribution centered around resonance radius Inactive bubbles that continue to grow or active bubbles that grow out of resonance 8

© IMEC 2010 / CONFIDENTIAL Cavitation Activity: Role of On-Time Cavitation noise data 9 Pulse on times ▸ A simple bubble model based on bubble growth, bubble loss and bubble creation mechanisms can model the pulse on time variation. -A maximum bubble activity is reached with a pulse on time of ~50 ms  grow = 8.6 ms  eff =1.1 s Pulse on time variation at constant pulse off time (150 ms) and 105 % dissolved gas Bubble size Reservoir Lost bubbles

© IMEC 2010 / CONFIDENTIAL Continuous125 ms150 ms175 ms 0.42 W/cm W/cm 2 Influence of pulse off time 10 PRE maps for variable pulse off times, a fixed pulse on time (50 ms) and a dissolved oxygen concentration of 105% Acoustic field 145 mm from transducer surface ▸ Non-uniform acoustic field is a near-field (interference) effect caused by the transducer size. ▸ Non-uniform fields result in localized cleaning. Experiment Simulation Acoustic pulsing noticeably improves particle removal without changing acoustic power densities PRE (%)

© IMEC 2010 / CONFIDENTIAL Outline ▸ Introduction to particle removal ▸ Improving state-of-the-art megasonic cleaning -Acoustic pulsing -Oversaturated liquids -Traveling waves ▸ Future of particle removal with liquid motion in the semiconductor industry ▸ Conclusions 11

© IMEC 2010 / CONFIDENTIAL Maximazing bubble formation 12 Bubble formation is limiting the megasonic cleaning efficiency. ▸ An increased dissolved gas concentration facilitates the nucleation of bubbles 90%100%110%120%125%130% Impossible to nucleate bubbles Bubbles do not dissolve anymore PRE as function of dissolved oxygen concentration Duty cycle is 10%, pulse off time is optimized for dissolved gas concentrations and applied power is 420 mW/cm 2. The optimal dissolved gas concentration facilitates bubble formation ( ≥ 100%) and enables bubble dissolution ( < 130%) PRE (%)

© IMEC 2010 / CONFIDENTIAL Bubble dissolution or growth in the absence of an acoustic field is given by Dissolution Upper limit dissolved gas concentration 13 Bubble resonance size This term determines bubble growth or dissolution Growth

© IMEC 2010 / CONFIDENTIAL Outline ▸ Introduction to particle removal ▸ Improving state-of-the-art megasonic cleaning -Acoustic pulsing -Oversaturated liquids -Traveling waves ▸ Benchmarking of physical cleaning techniques ▸ Future of particle removal with liquid motion in the semiconductor industry ▸ Conclusions 14

© IMEC 2010 / CONFIDENTIAL ▸ Standing wave field -Bubbles experience an acoustic radiation force (Bjerkness force):  At moderate acoustic powers, bubbles smaller (larger) than resonance size will travel up (down) a pressure gradient. So small bubbles go to pressure antinodes and large bubbles go to pressure nodes. ▸ Traveling wave -To simulate bubble motion in a traveling wave, acoustic radiation force, added mass force (inertia) and viscous drag force need to be taken into account. As a result, radial and translational equations are coupled. Increasing PRE: transport of bubbles towards the wafer surface 15 Simulation of a 2.7  m sized bubble (radius) in an acoustic field of 0.73 W/cm 2. The average bubble velocities is in the order of m/s.

© IMEC 2010 / CONFIDENTIAL Influence of a traveling wave on particle removal efficiency Wafer Damping material Transducer ▸ A silicon wafer is transparent for acoustic waves at a specific angle ▸ With the combination of damping material, a traveling wave can be formed -Bubbles are transported towards the wafer surface and improve particle removal 16

© IMEC 2010 / CONFIDENTIAL Outline ▸ Introduction to particle removal ▸ Improving state-of-the-art megasonic cleaning -Acoustic pulsing -Oversaturated liquids -Traveling waves ▸ Future of particle removal with liquid motion in the semiconductor industry ▸ Conclusions 17

© IMEC 2010 / CONFIDENTIAL Large particles 18 Small particles 200 nm 100 nm Boundary layer thickness >> 100 nm Although the removal force increases for larger particles, it gets easier to remove large particles because drag force scales with radius and velocity 30 nm 100 nm A structure with a high aspect ratio gets problematic, due to a strong increase in drag force on that structure Physical cleaning techniques based on a fluid flow are ideally suited to remove ‘larger’ particles. Particle cleaning with liquid motion

© IMEC 2010 / CONFIDENTIAL Conclusions ▸ System optimization -Experimental megasonic system is optimized  Controlling average bubble size with acoustic pulsing  Facilitating bubble nucleation with slightly oversaturated liquid  Transporting bubbles towards wafer surface with traveling waves ▸ Challenges -Megasonic cleaning uniformity needs to be solved -Cleaning of 30 nm and smaller silica particles with low damage levels is not yet achieved  Boundary layer and aspect ratio of structures makes current techniques not suitable for continued scaling 19

© IMEC 2010 / CONFIDENTIAL Acknowledgements Thanks to ▸ Marc Hauptmann, Elisabeth Camerotto, Antoine Pacco, Geert Doumen, Stefan De Gendt, Marc Heyns, Geert Doumen and Tae-Gon Kim (Imec) ▸ Christ Glorieux (KULeuven) ▸ Aaldert Zijlstra (University of Twente) 20 ANTOINE PACCO