Analysis of solder joint edge effect in acoustic micro imaging of microelectronic packages General Engineering Research Institute Electronic and Ultrasonic Engineering Chean Lee Dr. Guangming Zhang Prof. Dave Harvey 27 February 2013

Outline Acoustic Microscopy of Microelectronic Packages Objectives Finite Element Modelling Results Post Processing Results and Discussion Conclusion

Acoustic Microscopy of Microelectronic Packages Non-Destructive technique Sensitive to voids, delaminations and cracks Detects sub-micron discontinuities within material and interconnects Technique characterized by spot size, focal depth, focal length & frequency Study microstructures of specimen X-Ray AMI Unreflowed Solder Bump, AMI presents better contrast of defect

Acoustic Microscopy of Microelectronic Packages Edge Effect occurring at solder bump and die edge Manifest as dark circles around the bump Acoustic information loss, obscure minor cracks or voids. Characterization of bond quality is inexplicit Edge Effect seen around the flip chip solder bumps and the edges of the flip chips

Objectives The physical mechanism of edge effect is not clear. Scatter theory is widely accepted The acoustic energy is scattered by the physical geometry of the interface This study aims to test scatter theory Utilizing ANSYS APDL 2D Cross-Section Acoustic Simulation Propose characterization methodology

Finite Element Modelling : Basics Two things to understand about finite element modelling. Nodes and elements. The nodes make up a web of mesh that contain the structural properties. Elements are formed when the web establishes a relationship with neighbouring nodes via vectors. Basic Equation The stiffness matrix defines the geometric and material properties between nodes. In its simplest form is derived from Fx = kxyuy Fx = Force in direction X kxy = Stiffness coefficient of x-y uy = Displacement in Y www.sv.vt.edu In short, Finite Elements can be said to be giant molecules that make an object. The finer the molecules the more accurate the results but require more computing power. This is where number of Element Per Wavelength (EPW) becomes important to determine how fine the mesh should be. (Usually 5 EPW for draft simulations and 20 EPW for fine detailed simulations.) The real explanation is much more complicated but this is brief rough description. F=KU only represent the relationship between nodes for one direction. To adequately describe it, a matrix is build using this equation in all directions. Hence the “stiffness matrix”. This is the very basic equation for Finite Element Analysis.

Finite Element Modelling : Acoustic Wave Equation Fluid Element propagation equation C = speed of sound = ρo= mean fluid density K = bulk modulus of fluid P = acoustic pressure t = time This equation neglects viscous dissipation. Therefore represents a lossless wave equation for sound in fluids. For Fluid-Structure Interactions , the transient dynamic equilibrium equation below is considered simultaneously with the above acoustic wave equation 1. Wave propagation equation 2. Infomation (energy) transfer equation (when the simulation data transfer from fluid to solid medium) [M] = Structural Mass Matrix [C] = Structural Damping Matrix [K ] = Structural Stiffness matrix {ϋ} = nodal acceleration vector {ύ} = nodal velocity vector {U} = nodal displacement vector {Fa} = applied load vector The equation is employed using the generalized-α method.

Finite Element Modelling : General Simulation Model Transducer is represented as a curve to produce a focused effect. Transducer parameters used in this experiment (Sonoscan) Diameter of transducer = 4750µm Focal Length = 9500µm Frequency = 230MHz Wavelength in water = 7.5µm If 20 Element per Wavelength, then total elements = 120million elements (4750µm X 9500µm) Hence, computational load is impractical. The rule of thumb is that 15 to 20 element per wavelength to accurately represent the waveform. Anymore and it is a waste of resources, any less and it will compromise the fidelity of the simulation. 5 is good for draft simulations.

Finite Element Modelling: Simulation Model Top half of the solder bump Water medium Virtual transducer Absorbing boundaries Under bump metallization Solder bump 140µm diameter SiO2 thickness 33µm Overall width 187µm 230MHz acoustic pulse Reducing the model is normal practice but the tricky part was the create an accurate scaled down virtual transducer. Virtual transducer is placed very close to structure Ansys guideline: Distance of absorbing boundary = 0.2Lambda from any structure. Thickness of SiO2 layer is calibrated to account for compressional focus Compressed focal point incidents on the Under Bump Metalization

Finite Element Modelling: Virtual Transducer 2 steps for rescaling of the transducer to a virtual microscale transducer. Step 1, decide on the radius of the microscale transducer Snell’s Law has to be considered and is given as V1 = Acoustic Velocity of Material 1 V2 = Acoustic Velocity of Material 2 θ1 = Incident Angle θ2 = Refraction Angle The results show the focal point that occurs after the magnifying effect (compressed focus) Simulation is carried out to measure performance of the virtual transducer inside SiO2. Results show the focal point at 38µm and focal depth of 22µm inside SiO2 Thickness of the SiO2 layer of the model is then adjusted accordingly.

Finite Element Modelling: Virtual Transducer Step 2, calculate the chord to accurately apply excitation loads given as Transducer C0 Where R0 Chord = Chord of the micro transducer curve R = Radius of microscale transducer C0 = lens diameter of sonoscan transducer R0 = focal lenght of sonoscan transducer The curve of the transducer that is modelled is actually longer then the excitation arc, so the chord had to be determined to get an accurate representation. Microscale Transducer Chord R

Finite Element Modelling: Execution Acoustic Pulse generated from 1nm displacement vectored perpendicular to the transducer curve. Length of excitation arc is inferred from the chord length. Each iteration: Transducer position moved 0.5µm Scan area: centre to edge of solder bump (70µm, half of model) The chord is determined as specified in the last slide. Transducer moves from centre to the edge, 70um is the position of the edge assuming the centre axis is 0um Start of Scan X=0µm End of Scan X=70µm

Results : Video Acoustic pulse at 35µm off-central axis of solder ball This is the result when the transducer is positioned 35um off the central axis, a.k.a iteration no 70 Acoustic pulse at 35µm off-central axis of solder ball Each iteration produces an A-Scan information. This is combined to create a B-Scan

Results : B-Scan Image Each iteration produces an A-Scan information. This is combined to create a B-Scan Main Bang (Initial Acoustic Pulse) Water to SiO2 interface This is only half of the model cross-section, from the centre of the solder bump to the edge. Since the geometry is symmetrical, this data can be mirrored to make a complete cross-section. SiO2 to Under Bump Metallization Interface A-Scan B-Scan

Post Processing: C-Line Novel method: C-Line, to characterize edge effect Obtained by finding the peak of the gated data at every x-axis position. Edge Effect Occurring Between 30µm to 50µm Intensity begins to decline around 25µm

Post Processing: C-Scan Reconstruct Rotate the C-Line data around a central axis A C-Scan image of that layer can be constructed. This C-Scan would represent an symmetrical solder bump. Comparison between measured C-Scan image and simulated version of the same solder bump

Post Processing: C-Scan Reconstruct Measured (Averaged) Measured Simulated

Results and Discussion: Scatter Theory Portions of the acoustic energy is loss as the scan progresses closer to the edge. Results show that the edge effect occurs long before the acoustic pulse nears the beginning of the edge. The pulse a.k.a spot size of the setup is 5µm wide Results shown in slide: “Post Processing: C-Line” The beginning of the edge is filled with water, which has a higher impedance mismatch. Reflection should be stronger.

Result and Discussion: Measured vs Simulated C-Line Results shown in slide: “Post Processing: C-Line”

Result and Discussion: Measured vs Simulated C-Line Hypothesis Side lobe effect Phase Scatter Modal change Results shown in slide: “Post Processing: C-Line”

Results and Discussion: Initial Evaluation Side Lobe Side lobes are secondary acoustic emissions. Could contribute measurable effect on overall average intensity detected by receiving transducer Offset at 35µm The vector of side lobe should allow it to reflect strongly from the water interface and into the transducer causing a strong signal. But overall intensity shows a decline. Results shown in slide: “Post Processing: C-Line” X-Axis = 35µm, where the intensity has already declined significantly Phase of the side lobe doesn’t show. Decline of intensity starting too early.

Conclusion Scatter theory does not have straight forward relationship with Edge effect Phenomena Novel method of characterizing Edge Effect proposed Results show that 60% of structural information is obscured. Work raises more questions about edge effect mechanism

Further Work Employing C-Line to evaluate defect detect mechanism Crack propagation Voids Minimum detectable defect Hypothesis testing Side Lobe Mode Changes Etc Predict effective spot size inside medium