Design, Optimization and Implementation of Fresnel Domain Computer Generated Holograms (CGHs) • Diffraction optical elements: reconstruct semi-arbitrary 2D or 3D optical fields • Numerical design: flexible encoding strategy high diffraction efficiency and uniformity • Avoid complications from conventional optical recording process • History: Detour (Brown, 1966), Kinoform (Lesem, 1969) • Applications: beam shaping, optical trapping, communications, 3D television, optical testing - Pure phase: binary*, multi-level - Fabrication method: electron-beam lithography 1 Motivation • Increasing demand for smaller sized features, large working area semiconductor devices (e.g. LCD manufacture) need novel lithographic methods • CGHs promising candidates for replacing conventional 2D or 3D lithographic techniques • Key advantages: - Non-contact - Depth of focus control - Robust design - Parallel exposure - Standard fabrication - High resolution - Large working area - Simple optical setup - 2D or 3D patterning - Cost effective In-line CGH Lithography Final Device Processing 2 Problem Definition • Performance of CGHs depends primarily on optimization algorithm and fabrication method • Previous work: X-ray (Jacobsen, 1992), UV (Wyrowski, 2001), EUV (Isoyan, 2006) • Local search methods: inefficient, sensitive to initial point, get trapped at local minima • Current multi-search schemes: optimize small size CGHs CGH Plane x Encoding Reconstruction Plane F1 I des ei1 Reconstruction Plane H size CGH Plane Back-propagation Free parameter Desired Pattern 1 Fresnel O F2 F2 ei2 x' Osize Osize H size pix F2 1 2 H H eiH y y' ' z pix ' pix pix d F2 F2 ei2 F1 F1 ei1 Inverse Problem 3 Problem Definition • Performance of CGHs depends primarily on optimization algorithm and fabrication method • Previous work: X-ray (Jacobsen, 1992), UV (Wyrowski, 2001), EUV (Isoyan, 2006) • Local search methods: inefficient, sensitive to initial point, get trapped at local minima • Current multi-search schemes: optimize small size CGHs CGH Plane x Decoding CGH Plane H eiH Reconstruction Plane H size Reconstruction Plane Forward-propagation OFresnel R R eiR | | x' Osize Osize H size pix 2 y y' z pix ' pix pix I est ' d Photoresist Exposure Final Pattern F2 F2 ei2 F1 F1 ei1 Inverse Problem 4 Reduced Complexity Hybrid Optimization Algorithm (RCHOA) • Efficient optimization of Fresnel binary and multi-level phase CGHs • Reduce problem complexity by introducing: Local Diffuser Phase Elements (LDPE) and Local Negative Power Elliptical Phase Elements (LNPEPE) masks • Optimize reduced subset of variables • Key features: - Multi-point parallel search - Robust: insensitive to initial points - Flexible choice of encoding signal - Reduced complexity - Optical efficient results - Computationally efficient: GPU implementation 5 System Geometries In-Line Geometry* Off-Axis Geometry TIR Geometry 6 Local Diffuser Phase Elements Mask • Maximize information transfer: amplitude (reconstruction plane) to phase (CGH plane) • Step 1: decompose desired pattern into Nbp binary patterns Mask Decomposition • Step 2: assign local diffuser phase element to each pattern (q) (q) (q) , Ffactor • Diffusivity of qth element controlled by: Dfactor and shift • LDPE mask: (q) 2 Ffactor (q) (q) (q) (q) exp i arg exp i 2 Dfactor R ishift A Jinc q 1 ev Nbp PLDPE Binary function A( q ) Random matrix (q) Ffactor ev CGH Plane Reconstruction Plane 2 • Reduced number of DOF: DOFLDPE 3Nbp x Fresnel Back-Propagation Each element has different diffusivity 7 Local Negative Power Elliptical Phase Elements Mask • Maximize information transfer: amplitude (reconstruction plane) to phase (CGH plane) • Step 1: decompose desired pattern into Nbp binary patterns • Step 2: apply LNPEPE to each pattern Binary pattern center coordinates • Controlled parameters: f1(q) , f2(q) , x(q) , y(q) • LNPEPE mask: 2 (q) (q) x 'sin x y 'sin y exp i (q)(q) exp i q 1 Nbp PLNPEPE Binary function Truncation window x ' x ( q ) 2 y ' y ( q ) 2 c c (q) (q) f1 f2 Reconstruction Plane CGH Plane • Reduced number of DOF: DOFLNPEPE 4Nbp x Fresnel Back-Propagation Negative power elliptical phase 8 Genetic Algorithms Block • Multi-point optimization scheme • Inspired in biological evolution: “survival of the fittest” • Reduced complexity allow optimizing large populations • Individual: (1) (1) (1) xk Dfactor , Ffactor , shift , (N bp ) (Nbp ) (N ) , Dfactor , Ffactor , shiftbp or xk f1(1) , f 2(1) , x(1) , y(1) , ( Nbp ) , f1 ( Nbp ) , f2 , x ( Nbp ) , y ( Nbp ) Global minimum The MathWorksTM 9 MER Block • Local search, iterative optimization method • Refine solution: fast convergence • Compare results with: diffracted field (DF) and simulated optically recorded hologram (SORH) encoding strategies 10 Error Metrics Photoresist Contrast Curve • Four considered error metrics • Choice of error metric is application dependent - Mean square error: bias estimator ( and 2 ) MSEbefore MSEafter 1 2 N 1 2 N N N Iest Ides 2 x '1 y '1 N N R I des , x '1 y '1 2 dose D 1 R D0 0 inside pattern otherwise Amplitude Constraint - Diffraction efficiency eff : Effective efficiency D D0 ub 4 fill(0) f d(0) (d ) fd 2 Osize H size f d(0) 2 2 f H size Signal Power Inside Hsize Signal Power (d ) 2 d Osize 2 fill(0) Input power 2 H size (G. Zhou, et al., 2000) - Additional metrics: L1 (bias) and normalized cross-correlation (similarity), hybrid 11 Optimization Results Main Parameters: • Optimization example: - binary phase CGH: resolution target - LDPE encoding strategy Wavelength 532nm Elite Children 5 Working Distance 150μm Crossover Fraction 0.6 Pixel Size 200nm Generations 100 CGH Size 300μm Population Size 100 Object Window 180μm Iterations 400 Intensity Intensity Convergence GA Block Reconstruction Reconstruction fromfrom Multi-Level Binary CGH CGH at at Photoresist Photoresist Plane Plane (Before (Before Exposure) Exposure) Desired Pattern Phase Map: Optimized Binary CGH Optimized LDPE Mask Convergence MER Block MSEbefore 27.18 147.31 MSE before Intensity Intensity effeff 79.46% 35% 12 Optimization Results • Optimization example: - binary phase CGH: resolution target - LDPE encoding strategy Comparison of Encoding Strategies After GAs Block: Multi-Level CGH • Sensitivity Analysis: problem parameters (e.g. cross-over fraction, population size, etc.) • Parallel implementation on graphic processing unit: speedup >180X - GPU computational time: 4.47 hours - CPU estimated time: 16.48 days! 13 Extending the Depth of Focus Multiple Plane Constraint • Extend DOF: tolerate potential axial misalignments during exposure process • Modify RCHOA to impose constraints at multiple planes •Regular DOF: z 2 2NAeff Error Comparison: Binary CGH Extended DOF CGH z 266nm Extended: 2 z 14 CGH Fabrication Fabrication Process • Fabricated using electron-beam lithography • Binary phase CGH • Resist: Hydrogen Silsesquioxane (HSQ) Fused Silica Aluminum HSQ E-beam Patterning Remove Aluminum & Develop HSQ Scanning Electron Microscope Image of Fabricated Sample 50μm 15 Characterization of Fabricated CGHs • Implemented methods: evaluation algorithm*, optical characterization*, exposure test • Evaluation algorithm: analyze fabricated CGH 2D error map (correct over/under dose) Block Diagram of Evaluation Algorithm Stitched Binarized Fabricated CGH 2D Error Map 16 Characterization of Fabricated CGHs • Implemented methods: evaluation algorithm*, optical characterization*, exposure test • Optical characterization: measure reconstructed intensity Optical Setup: Coherent Illumination Measured Reconstructed Intensities -Fabricated CGHs not fully optimized Binary CGH: DF Encoding Strategy Binary CGH: Diffuser Encoding Strategy - Eliminate speckle using partial coherence illumination 100μm 17 Sensitivity Analysis • Estimate and assist in the correction of potential fabrication errors • Considered errors: e-beam over/under dose, proximity effect, uniform/nonuniform phase, stitching and positional errors Stitching Error Analysis MSE Dilation Test: Over Dose Error Offset Distance 18