Instantaneous Fluid Film Imaging in Chemical Mechanical Planarization Daniel Aponea, Caprice Graya, Chris Rogersa, Vincent P Mannoa, Chris Barnsb, Mansour Moinpourb, Sriram Anjurc, Ara Philipossiand a Tufts University, Department of Mechanical Engineering, Medford, MA, USA b Intel Corporation, Santa Clara, CA, USA c Cabot Microelectronics Corporation, Aurora, IL, USA d University of Arizona, Department of Chemical Engineering, Tucson, AZ, USA ABSTRACT Dual Emission Laser Induced Fluorescence (DELIF) is employed to attempt to experimentally determine the nature of the lubrication regime in Chemical Mechanical Planarization Our DELIF setup provides images of the polishing slurry between the wafer and pad Static images were taken to provide a baseline for determining Analyzing these images shows that the wafer only contacts the pad in a small number of places around the wafer, mainly due to the pad’s topography INTRODUCTION For many years the semi conductor industry has employed a polishing technique known as Chemical Mechanical Planarization Ultra flat surfaces are needed during the production of integrated circuits, both on a local (nearly molecular level) and a global (12 inch silicon wafers) scale In this process a silicon wafer is pressed against a polyurethane polishing pad Each is rotated, with a slurry lubricating the interface There are many variables in this process, such as pad type, rotation speeds, applied pressure, pad conditioning, as well as slurry composition and flow rate Industry and the research community have long teamed to characterize this complex process, yet many questions still remain This research aims to optically measure the interaction of pad and wafer to determine which lubrication regime the CMP process lies in There has been much debate as to whether or not the pad is supporting the wafer, or if the two are fully separated by a layer of slurry Most likely it is a mixed polishing regime, with the pad's asperities partly supporting the wafer, and the fluid layer supporting load too Modeling work indicates that it is a delicate balance between hydrodynamic lubrication, mixed solid-liquid contact and direct solid-solid contact [1] Modeling efforts have been employed to predict the film thickness and removal rate based on the contact regime Higher removal rates can be obtained with more solid-solid contact, but with a higher incidence of wafer scratching Changing experimental parameters to increase the film thickness will result in less scratching of the wafer, but also a lower removal rate [2] Measuring of film thicknesses is a difficult task, since their scale (50 microns) renders traditional measurement techniques useless Light Induced Fluorescence (LIF) has been used to measure thicknesses, with a camera capturing the fluorescent image The brightness in the image is directly related to the thickness of the film The more fluid there is, the more dye is present, therefore a brighter image is captured The problem is that the initial excitation intensity may not be even across the image This would result in inaccurate measurements The solution is to use a two-dye system Initially one dye is excited, and the light emitted from it excites the second dye Using two cameras to capture the image (each camera fitted with a filter for one of the dyes) and then taking a ratio of the two camera images allows for true fluid film measurements [3] Switching to a temperature sensitive set of dyes allows for the optical measurement of fluid temperature [4] SETUP This research builds upon work conducted at Tufts University since 1995 Key to this work is a technique known as Dual Emission Light Induced Fluorescence, which employs an ultraviolet light source that excites a fluorescent dye Only a brief overview is presented here; Copetta and Rogers reported in detail [5] This dye's emission then excites another fluorescent dye By photographing the two dye's emissions, we can determine the amount of dye that was present Basically, the more dye, the brighter the image We mix the dye into the polishing slurry and can then determine the thickness of the fluid layer Previous DELIF work at Tufts University used an ultraviolet lamp to induce fluorescence Camera shutters would have to be held open for seconds to capture enough light New to this research is the use of an ultraviolet laser This imparts much more energy to the slurry than the UV lamp did, so we are essentially able to take instantaneous images of the dynamic polishing process We use a Struers RotoPol 31 as our table top polisher It sits on a AMTI MC-12 force table, capable of measuring forces in and moments about all three axes It allows real time measurements of the forces of the wafer and pad interacting, whether in the Z (applied pressure) or the X and Y directions (friction due to polishing) The polisher's platen is fitted with 12 inch diameter Freudenberg FX9 pads Since a silicon wafer is opaque, we need a substitute as we are looking to image the region of fluid under the wafer A inch BK-7 glass wafer is used instead It is 1/2 inch thick and has good transmission qualities at the wavelengths needed (300nm600nm) The wafer's applied pressure and rotation are controlled by an aluminum shaft pressing the wafer into the pad The shaft is held in place by two linear bearings that allow the shaft to rotate as well as slide to apply pressure A 1/2 hp Dayton motor rotates the shaft, and a series of weights sit atop the shaft to increase the applied pressure The shaft and motor are held above the polisher by a custom 80/20 extruded aluminum support structure Motor Wafer Platen RotoPol-31 Steel Table Force Table Figure Experimental setup; laser and cameras are not pictured The slurry used is a 9:1 dilution of Cabot's Cab-O-Sperse silicon polishing slurry It is diluted to minimize polishing effects on our glass wafers The slurry is constantly stirred with a magnetic stir bar to discourage aggregation of the silica nano-particles in the slurry It is pumped onto the pad by a peristaltic pump at a rate of 50 cc per minute The slurry also contains Calcein, a fluorescent dye, at a concentration of g/L A Quantel Brilliant B Nd-YAG laser discharges a rated 2.5 W at 355 nm The laser light passes through a beam expander, then off a mirror onto the pad The light passes through the BK-7 glass wafer to the fluid/pad interface underneath The pad has a natural fluorescence that is used to excite a single dye We use this single dye variant of the original DELIF process for simplicity and decreased cost The light from these emissions is captured by two cameras with filters, one to view the pad's emission, one to view the Calcein's emission The absorption and emission spectra, as well as the filter regions for both cameras can be seen in Figure Each pixel in the Calcein image is divided by the corresponding pixel in the pad image This division removes any variations in the original light intensity For example, if the left hand side of an image were brighter, you wouldn’t know if the fluid were thicker, or if the original light were brighter in that section The cameras are 12 bit linear Evolution VF Peltier cooled models They provide images at a resolution of 1392 x 1040, however we run the camera’s at a binning setting of to attain brighter images This results in images that are 696 x 520 These are connected to the lab control computer via a firewire interface A custom LabView interface controls the image acquisition, laser pulsing, platen rotation, and force table measurements Figure Emission and absorption spectra for the pad and dye as well as the filtered regions for both cameras Figure Bubbles in an oncoming slurry wave on the left, striations from the conditioner on the right The small dots in the right image are probably shadows of dried slurry on top of the wafer RESULTS Figure is sample images obtained by the cameras These are air bubbles contained within an incoming slurry wave on the left They appear darker because they lack the Calcein that emits the wavelengths the cameras view The image on the right has visible grooves we attribute to our conditioner These striations are made when the conditioner "combs" the pads asperities, much the way a lawnmower leaves patterns in cut grass These images are taken with a lens setup that yields a resolution of about 50 microns per pixel The total viewing area of the cameras is about inch across Research was conducted at this zoom level to verify previous results The instantaneous images and measurements were in line with previous findings; typical slurry film thicknesses were 15 to 30 microns The optical setup was adjusted to provide a resolution of about microns per pixel Figure is a typical image taken at this zoomed in resolution, a total area about mm across This was necessary to resolve smaller features on the pad, which has a textured surface consisting of many peaks and valleys These asperities are protrusions that aid in the distribution of slurry and removal of waste These asperities range in size, with 10-15 microns being a common height They range in diameter too, many being around 40 microns in diameter We believe the asperities behavior will indicate when and to what degree wafer/pad contact is occurring The pad's profile is essentially a Gaussian distribution of points There are outliers on either end such as very tall peaks, or very deep holes into the pad The bulk of the asperities lie clustered around some mean pad height The dark areas of the image reflect a lack of fluid, meaning those are the high peaks that reach nearly up to the wafer The bright regions are deeper holes in the pad where a thick layer of fluid sits Contact points are determined by taking a high pressure static image and finding the darkest pixels Images are taken at multiple applied pressures to make sure that the intensity of contact points doesn’t change as the pressure is increase As you can see in Figure 4, there is a single peak of an asperity that is the first to contact the wafer at this instant, although some images don’t seem to have a clear contact point This would indicate that we need to look at a larger section of pad at one time Figure This is a static image taken at 10 psi Those darkest pixels on the image represent the thinnest fluid layer, and therefore the contact region The high pressure causes so much compression of the pad, that the image gets slightly out of focus It turns out that the pad's surface not only has these small asperities that we can see in Figure 4, but there are larger, lower frequency undulations as well This was first noticed when a test pad was repeatedly used until glazing started to show up The glazing is a combination of the silica particles in the slurry and wafer waste that settles down into the pores of the pad Conditioning the pad with an abrasive disc prevents pad glazing This test pad showed an odd pattern of glazing, where there were heavy sections of glazing and other sections that didn’t show any glazing Figure is a profilometer scan across the pad which confirmed the hypothesis that the glazed sections were high points and the unglazed sections were valleys There turns out to be nearly a 100 micron height difference between the tops of these larger scale ridges and the bottoms of the large scale valleys Our region of interrogation (ROI) is only millimeters wide, meaning that some images are taken as a ridge top is passing beneath the ROI, and some when a valley is below This is significant because at the hilltops the wafer seems to be compressing the asperity layer, and in the valleys the asperities have a thick layer of fluid above them This was further confirmed by taking static images (no pad or wafer rotation) at low downforce (0.4 psi) and then again in the same spot at high downforce (10 psi) at many different locations around the pad If the pad were perfectly flat, the difference between the high and low pressure images would be fairly constant That isn’t the case at all, however Some images show thin fluid layers in both the low and high pressure cases These are being taken when a high point is under our ROI The valley images show a very thick fluid layer in the low pressure case, and then a thin film in the high pressure case meaning the fluid has been squeezed out by the increase in pressure Figure Profilometer trace of the FX9 pad The zoomed in region represents the size of our camera’s viewing area CONCLUSION Pad/wafer contact can be observed using the DELIF process True contact points seem to be few and far between, as if the wafer only sits on a few high peaks As this is ongoing research, more study is needed to make confident assertions The experimental setup will have to be changed such that cameras view a larger area (to gain a better picture of the pad’s topology) yet still can resolve asperity behavior REFERENCES S R Runnels and L M Eyman Tribology analysis of chemical-mechanical polishing Journal of the Electrochemical Society, 141(6):1698–1701, 1994 D G Thakurta, C L Borst, D W Schwendeman, R J Gutmann, and W N Gill Pad porosity, compressibility and slurry delivery effects in chemical mechanical planarization: Modeling and experiments Thin Solid Films, 336:181–190, 2000 C H Hidrovo and D P Hart Emission reabsorption laser induced fluorescence (erlif) film thickness measurement Measurement Science and Technology, 12:467–477, 2001 J Sakakibara and R J Adrian Whole field measurement of temperature in water using two-color laser induced fluorescence Experiments in Fluids, 26:7–15, 1999 J Copetta, C Rogers, Experiments in Fluids, 25, 1, (1998) ... effects in chemical mechanical planarization: Modeling and experiments Thin Solid Films, 336:181–190, 2000 C H Hidrovo and D P Hart Emission reabsorption laser induced fluorescence (erlif) film. .. case, and then a thin film in the high pressure case meaning the fluid has been squeezed out by the increase in pressure Figure Profilometer trace of the FX9 pad The zoomed in region represents... Some images show thin fluid layers in both the low and high pressure cases These are being taken when a high point is under our ROI The valley images show a very thick fluid layer in the low pressure