Optical Maskless Lithography - OML

Transcription

1 Optical Maskless Lithography - OML Kevin Cummings 1, Arno Bleeker 1, Jorge Freyer 2, Jason Hintersteiner 1, Karel van der Mast 1, Tor Sandstrom 2 and Kars Troost slide 1

2 Outline Why should you consider Optical Maskless Lithography (OML)? What are we planning to build? How will OML perform? Summary slide 2

3 Why would anybody want to use Maskless Lithography? A maskless scanner could save $50k / design for 0.13 µm technology $1M / design for 90 nm technology $4.5M / design for 65 nm technology This implies the ROI of a maskless scanner can be very high 50% for 0.13 µm technology 85% for 90 nm technology Time to market for designs can be greatly reduced Multiple products on a reticle (Shuttle) can also reduce some of the costs from masks however there is often a penalty in TAT. There are a significant number of interesting things that might occur given a robust maskless lithography tool Lithography foundry, IC customization, multiple & unique designs slide 3

4 Why Optical Maskless Lithography? It will be Fab Transparent The tool will look and behave like a Step and Scan system Product can start on OML & be ported to the conventional scanner slide 4

5 OML Platform is Based on TWINSCAN Leverage Existing Litho Platform Capabilities Wafer stage Wafer alignment & focus Tool control, environmental facilities & support Fab interfaces Fab processes Key Changes Projection Optics: Higher de-magnification Illuminator: New illumination area & dose control Reticle Stage / Reticle Handler: Replaced with SLM unit and supporting data-path electronics Synchronization: Mirrors and laser are synchronized to the stage movement. Projection Optics Illumination Optics TWINSCAN XT MASKLESS slide 5

6 Significant Number of Modules Are Ready for OML Tool Field Pattern Mgmt. Illumination Projection Optics System Metrology Wafer Mgmt. System Control Platform SLM Device Laser Optics Wavefront Control (ILIAS/TIS) Dual Wafer Stage Synch. Algorithms Isolation Systems (frames) Multi-SLM Array Dose & Uniformity Control Calbiration Optics SLM Calibration Wafer Handling System Wafer Material Scheduler Nitrogen Purge Pattern Processor Illumination Optics PO Control Alg. Wafer Alignment Field Pattern Scheduler Environmental Control SLM Electronics / Data Path Illumination Control Alg. Wafer Leveling Exposure Manager Utility Interface Field Pattern Control Alg. Human & Fab Interfaces Tool Enclosure Legend (Changes Relative to existing Scanners) New / Unique Systems Significant Redesign Required No / Minor Changes slide 6

7 Wafer Fab Philosophy Regular scanners Verified high volume designs Few reticles most wafers High volume wafers Most designs few wafers Low volume and design prototype wafers New and low volume designs OML slide 7

8 Why Optical Maskless Lithography? It will be Fab Transparent The tool will look and behave like a Step and Scan system Product can start on OML & be ported to the conventional scanner Follows the mask-based lithography roadmap for λ and NA slide 8

9 Follows the mask-based lithography roadmap for λ and NA Advances in fab processes are directly leveraged with OML Baseline architecture allows for multiple technology options 65-nm node 193 > 0.9 NA 45-nm node 193i > 1 NA 157 > 0.9 NA Or whatever your favorite roadmap might be after this forum 32-nm node 157i > 1 NA EUV > 0.2 NA slide 9

10 Why Optical Maskless Lithography? It will be Fab Transparent The tool will look and behave like a Step and Scan system Product can start on OML & be ported to the conventional scanner Follows the mask-based lithography roadmap for λ and NA Leverages existing optical knowledge k 1 reduction (OPC, illumination, PSM...) can be applied to OML slide 10

11 Pixel Modeling Example of Optical Proximity Correction w/ Gray Scaling No Optimization Mirror Tilt [mrad] Example shows lines on mirror grid. Off-grid lines achieved with different tilt angles Optimized Mirror Tilts slide 11

12 Why Optical Maskless Lithography? It will be Fab Transparent The tool will look and behave like a Step and Scan system Product can start on OML & be ported to the conventional scanner Follows the mask-based lithography roadmap for λ and NA Leverages existing optical knowledge k 1 reduction (OPC, illumination, PSM...) can be applied to OML No additional infrastructure to create Resists, vacuum handling, charge reduction, alignment No fundamental limitations on throughput Photons don t interact (at this level anyway) slide 12

13 What is the problem that needs solved? Mask set costs - extending the current trend to 22nm Impossible, the IC industry is going to break 100 Mask set ($M) 10 1 $6.9M-$11.5M = = $3.9M - $7.5M = $1.9M - $4.5M = $9M - $14.4M Technology Nodes (nm) KDC* ISMT * Based on the author s view of the world slide 13

14 Risk of Design Introduction Keeps Increasing Std Cell ASIC Design and Delivery Cycle 130 nm Design Start Design is $5-10M Reticle Manufacturing ($ k) Design to mfg Initial Prototype does not Meet Specifications >18 months 3 Backend Reticle Re-Spin (~$250k) Re-optimize Design Purchase New Reticles Final Prototype available Respin reasons Currently design release relies on simulation 35 Simulation is increasingly complex with new technology 30 nodes 25 Design defects can occur at sort resulting in costly 20 re-spins Need to offer cost effective silicon verification Several design iterations running concurrently Percent Functional logic Minimal cycle time impact with no additional reticle cost. Signal integrity Reliability Power use To fast Firmware error Aart de Geus, SNUG slide 14

15 Mask usage is becoming increasingly inefficient The majority of masks are used on a small number of wafers ASIC-like fab data Saving $ here will allow more products and risk OML Sematech Data slide 15

16 Example: 50% of mask sets print < 1% of the wafers Data scaled to mm This is why even a slow tool can make good economic sense Cumulated number of mask sets 1% 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Cumulated number of wafers slide 16

17 Why does OML save so much money? Maskless Practically Eliminates the Fixed Cost of a Mask Cost per Run Maskless more economical Maskless Scanner Mask Scanner Fixed cost of masks Small Data Prep Cost Breakeven #Wafers Mask more economical # Wafers slide 17

18 Run cost ($M) Example: Maskless economic model - Run costs layer/mask 4 layer/mask Maskless Tech node (nm) layer/mask = standard lithography 4 layers/mask = similar levels on mask Mask prices are projected to increase with technology node 0.18 µm = $12k 0.13 µm = $32k 0.09 µm = $95k µm = $285k Maskless scanner Cost ~ $25M, 5 WPH & 50% utilization Device / Wafer Data 300mm wafers, Die = 8.5 mm, 20 wafers/design, 70 OML/year & 30% are re-spun Model includes depreciation, cost of capital, return on capital, taxes & clean room Start with 10 OML layers at 0.18 µm 16 OML layers at µm. slide 18

19 Why should you consider Optical Maskless Lithography? Conclusions For 65 nm the OML breakeven point is > 1700 wafers/reticle That = 6% of the wafers in a foundry-like fab A maskless scanner could save $1M / design for 90 nm technology $4.5M / design for 65 nm technology OML allows the flexibility and ability to decrease the time from design to product. OML will be Fab Transparent OML will follow the mask-based lithography roadmap for λ and NA OML does not require additional infrastructure OML does not have the limitations on throughput from charged particle interactions in the lithography system slide 19

20 Outline Why should you consider Optical Maskless Lithography (OML)? What are we planning to build? Lens Spatial Light Modulator (SLM) How does it work? Gray Scaling Throughput and the data path How will OML perform? Summary slide 20

21 OML Platform Based on TWINSCAN TWINSCAN XT MASKLESS Plus... slide 21

22 Enabling Technology: Spatial Light Modular Micronic SLM chip 8 mm x 33mm active area (512 x 2048 pixels) MPixels each mirror 16 µm x 16 µm Lithography pattern generation is controlled by reflecting laser light from individual pixels towards or away from the wafer slide 22

23 Equals an Optical Maskless Lithography tool > 2.5 years ago ASML & Micronic started an OML study Wavelength = 193 nm Illumination = Conventional, Annular, Dipole, Quasar, Quad... Throughput five 300-mm wph (or ten 200-mm wph) Technology node 65 nm Includes track interface, on-board metrology and factory interface We are ready for commitment from potential customers now Subject to this commitment we will develop the tool Available in 3 years (after start) Price ~ same as conventional scanners New Proj. Opt New Illum Opt TWINSCAN XT MASKLESS slide 23

24 Optical Column Design (Preliminary) Catadioptric Projection Lens w/ Dual Diffractive Array Illuminator Multi-SLM Array Optimal Configuration Wide Spectral Bandwidth Extensible Least Complex Least # of Elements Lowest Cost of Goods Illumination System Wafer Plane Laser & Beam Delivery System omitted from illustration.

25 The micromirror Spatial Light Modulator (SLM) chip Mirror array 16 µm Torsional 16 x 16 micron mirrors Actuated by electrostatic force Matrix-addressed CMOS chip similar to a TFT screen below the mirrors M Matrix addressing of analog array A1 A2 A3 A4 Lithography pattern generation is controlled by reflecting laser light from individual pixels towards or away from the wafer C T µm E G2 G3 G1 slide 25

26 SLM Details slide 26

27 Imaging with a Optical Maskless Lithography system Use Mirror Tilt Phase Effects Mirror - Phase change 5ϕ 4ϕ 3ϕ 2ϕ 1ϕ 1ϕ 2ϕ 3ϕ 4ϕ 5ϕ + Phase change Intensity I e iφ dφ slide 27

28 Imaging with a Optical Maskless Lithography system Use Mirror Tilt Phase Effects Mirror with tilt Im Re Amplitude: Intensity: 4.47% So negative amplitude can be created on the wafer giving additional optimization parameter slide 28

29 How do SLMs image? Diffraction is used to create contrast When the SLM surface is flat all light is reflected specularly A very small surface perturbation causes the light to be diffracted and the specular reflection is attenuated or extinguished Variable attenuation by variable non-flatness enables analog addressing for Gray Scale slide 29

30 OML imaging - the need for Gray Scaling If 2 N (64) steps are needed for edge placement: Binary mirrors need 2 N (64) bit Gray scale mirrors need N (6) bit There is a large reduction of the data flow to the SLM and the number of mirrors However there is some reduction of line edge fidelity edge placement Tilt angle slide 30

31 Gray Scale vs. Tilting mirror aerial image comparison. Tilting mirrors Grayscale pixels slide 31

32 How does OML reproduce OPC & RET? Masks have sub resolution patterns SLM grid Sub-grid address Sub resolution patterns Scatterbars Serifs slide 32

33 OML uses Gray Scaling SLM SLM grid Sub-grid address Gray-scale patterns Scatterbar Serifs slide 33

34 Contact hole with corner serifs Sigma results slide 34

35 SLM requirements - Throughput What pixel size will be needed for the 65 nm node? At wafer ~ 30 nm SLM mirror size will be 8 µm so magnification = ~266x Or 1mm field on wafer = 266 mm field in SLM plane SLM will have ~ 10 MPixel / SLM Multiple SLMs are needed and must fit into reasonable sized lens slide 35

36 Throughput -- How s it done? Multiple SLM Array Scan direction Use Multiple SLMs in parallel Simulated Step and Flash Short laser pulse duration (< 10 ns) synchronized with continuous scan speed Effectively, the wafer steps by the active area width of one SLM per laser pulse Mult-pass averaging Each pixel on the wafer sees multiple shots from multiple SLMs Leading & trailing SLMs could print in one pass Some SLMs are always leading (i.e. first pulse), and others are always trailing (i.e. subsequent pulses) As we reverse direction, the SLMs reverse their leading or trailing state slide 36

37 CONFIDENTIAL

38 Optical Maskless Lithography exposure Multi-SLM Scanning Laser Pulse: Steady-State slide 38

39 Data-Path: Data Rate Impact With: 300mm Wafer size 5 WPH 30nm Pixel size Dual pass 8 bits/pixel You have: Gbyte/s 70 mph Printing 360,000 pages with text and images per second or filling bookshelves with SPIE Proceedings at 70 miles per hour! slide 39

40 Preliminary Data Flow Requirements Pattern is pre-processed and fractured off-line OPC and system-constants applied Image fractured into pieces corresponding to SLM layout Rasterization and adjustment Done in near real-time Application of SLM calibration Filtering for off-grid effects SLM Interface Between 200 to 400 GBytes / sec peak data transfer rates (function of pixel size at wafer and dose control strategy) Massive real-time parallel computing architecture High-speed gigabit ethernet, RapidIO bus architecture, and multiple FPGAs for real-time control of SLM pixels slide 40

41 Outline Why should you consider Optical Maskless Lithography (OML)? What are we planning to build? How will OML perform? Imaging simulations Overlay simulations Summary slide 41

42 Pixel Modeling Tilted Mirror Pixel Representation Pixel representation in Prolith for tilted mirrors: 30 or 40 nm 30 or 40 nm Each pixel is divided into 30x1 nm segments To mimic a mirror tilted by angle α, a phase range of Φ=4παL/λ is linearly applied across the pixel with a zero average phase. (L is the mirror width (reticle scale) and λ is the wavelength) Pattern Parameters (L/S, CH, generic.) Rasterization Algorithm (Matlab) Illumination Table -2παL/λ 0 2παL/λ Illumination Paramaters (mode, p.c., etc.) Illumination Table Generator (Matlab & Prolith) Set of "Maskless" Mask Files Aerial Image / Resist Analysis (Prolith) Output Data slide 42

43 Pixel Modeling - A reminder of what we are doing Example of Optical Proximity Correction w/ Gray Scaling No Optimization Mirror Tilt [mrad] Example shows lines on mirror grid. Off-grid lines achieved with different tilt angles Optimized Mirror Tilts slide 43

44 Imaging 70 nm with an example OML system 40 nm pixels First guess Layout Optimized Layout Phase ranges: Phi1=0 deg Phi2=133 deg Phi3=257 deg Phi4=360 deg Phase ranges: Phi1=0 deg Phi2=275 deg Phi3=475 deg Phi4=550 deg slide 44

45 Pixel Modeling Effect of negative amplitude vs. binary mask No Negative Amplitude Binary (chrome) slide 45

46 Pixel Modeling Effect of negative amplitude vs. binary mask Blacker-than-Black provides better performance than binary, and acts more like an Attenuating Phase-Shift Mask (Att-PSM)! Larger Process Latitude! 2.8% negative amplitude Binary (chrome) slide 46

47 60 nm Isolated Line Behavior Thru-Focus vs. Att-PSM 6% Isolated 60 nm lines (420 nm pitch), Double Exposure Average CD through Focus (CD measured through pitch) OML shows equivalent performance to the ideal 6% Att-PSM in CD Improvements in NILS expected with enhancements to the optimization algorithm Average CD (nm) Maskless Defocus (nm) Ideal 6% Att-PSM Isolated 60 nm lines (420 nm pitch), Double Exposure Average NILS through Focus (CD measured through pitch) 1.5 Case Description 60 nm lines, 40 nm bias, 420 nm pitch 30 nm pixels phase-tilt mirrors Two-pass, 0.6/0.9 annular Average NILS NA, 193 nm wavelength Vector model, unpolarized light Defocus (nm) Model outputs are: 1D aerial images thru-focus Maskless Ideal 6% Att-PSM slide 47

48 60 nm 1:1 Dense Lines Behavior Thru-Focus vs. Att-PSM 6% Optical Maskless matches the Ideal 6% Att-PSM in CD Improvements in NILS & Contrast for Optical Maskless expected with enhancements to the optimization. Average CD (nm) Dense 60 nm lines (1:1), Double Exposure Average CD through Focus (CD measured through pitch) Maskless Defocus (nm) Ideal 6% Att-PSM Dense 60 nm lines (1:1), Double Exposure Average NILS through Focus (CD measured through pitch) Dense 60 nm lines (1:1), Double Exposure Average Contrast through Focus (CD measured through pitch) Average NILS Average Contrast Defocus (nm) Defocus (nm) Maskless Ideal 6% Att-PSM Maskless Ideal 6% Att-PSM slide 48

49 Pixel Modeling Rasterization Optimization Optimized rasterization can drastically improve behavior of CD, placement, and NILS through grid NILS can be optimized BETTER than the ideal Att-PSM case Task is to make this optimization real-time capable Placement 70 nm iso line placement error, (nm) Placement error vs shift. Const image intensity threshold Current algorithm Newly optimized --- Current Micronic Algorithm optimization with 4 active pixels --- Optimization w/ 4 active pixels NILS NILS NILS vs shift Current algorithm Newly optimized Ideal 6% Att:PSM Micronic optimization with 4 active pixels 6% AttPSM mask with no bias shift, nm Feature shift wrt grid (nm) shift, nm Feature shift wrt grid (nm) slide 49

50 80 nm Half-Pitch Contact Holes Shift = 0 nm 160 nm pitch contacts (80 nm half-pitch) 30 nm pixels phase-tilt mirrors Two- pass Diagonal Quasar (0.97/0.8, 15 o blade angle) 0.93 NA, 193 nm Vector model, unpolarized light, current rasterization Model outputs: 2D aerial images thru-focus slide 50

51 80 nm Half-Pitch Contact Holes Shift = 5 nm 160 nm pitch contacts (80 nm half-pitch) 30 nm pixels phase-tilt mirrors Two- pass Diagonal Quasar (0.97/0.8, 15 o blade angle) 0.93 NA, 193 nm Vector model, unpolarized light, current rasterization Model outputs: 2D aerial images thru-focus slide 51

52 80 nm Half-Pitch Contact Holes Shift = 10 nm 160 nm pitch contacts (80 nm half-pitch) 30 nm pixels phase-tilt mirrors Two- pass Diagonal Quasar (0.97/0.8, 15 o blade angle) 0.93 NA, 193 nm Vector model, unpolarized light, current rasterization Model outputs: 2D aerial images thru-focus slide 52

53 80 nm Half-Pitch Contact Holes Shift = 15 nm 160 nm pitch contacts (80 nm half-pitch) 30 nm pixels phase-tilt mirrors Two- pass Diagonal Quasar (0.97/0.8, 15 o blade angle) 0.93 NA, 193 nm Vector model, unpolarized light, current rasterization Model outputs: 2D aerial images thru-focus slide 53

54 80 nm Half-Pitch Contact Holes Shift = 20 nm 160 nm pitch contacts (80 nm half-pitch) 30 nm pixels phase-tilt mirrors Two- pass Diagonal Quasar (0.97/0.8, 15 o blade angle) 0.93 NA, 193 nm Vector model, unpolarized light, current rasterization Model outputs: 2D aerial images thru-focus slide 54

55 80 nm Half-Pitch Contact Holes Behavior Thru-Focus vs. Att-PSM 6% Ideal 6% Att-PSM shows slightly better performance in CD Improvements in CD and NILS expected with enhancements to the optimization algorithm Average CD Dense 80 nm contact holes (1:1), Double Exposure Average CD through Focus (CD measured through pitch) Defocus (nm) Maskless Ideal 6% Att-PSM Dense 80 nm contact holes (1:1), Double Exposure Average NILS through Focus (CD measured through pitch) Dense 80 nm contact holes (1:1), Double Exposure Average Contrast through Focus (CD measured through pitch) Average NILS Average Contrast Defocus (nm) Defocus (nm) Maskless Ideal 6% Att-PSM Maskless Ideal 6% Att-PSM slide 55

56 Stitching - An example An isolated horizontal dark line on a bright background The line is ~ 2000 nm long ~ 9 λ / NA. Coherent illumination, scalar model 0.85 NA, λ = 193 nm, pixel size = 40nm The line is formed by two adjacent rows of mirrors No stitching & no effect on the line shape Considerable effect on the line shape slide 56

57 Stitching: Butting No Compensation Exposure 2 Exposure line width vs X no stitching scenario 1 (simple stitching) line width, nm X, nm Exposure 1 D scenario1 = D1 + D2 + D3 Line width corresponds to 70 nm line width image intensity threshold D1 = D2 = D3 = slide 57

58 Stitching: 400nm Overlap Linear Attenuation in the Object Plane Exposure 2 Exposure line width vs X no stitching scenario 2 (400nm overlap with passive linear attenuation in object plane) line width, nm Exposure 1 D scenario2 = D1 + D2 + D X, nm D1 = D2 = D3 = slide 58

59 Outline Why should you consider Optical Maskless Lithography (OML)? What are we planning to build? How will OML perform? Summary slide 59

60 Conclusions Optical Maskless Lithography OML is the most economic low-volume production solution and significantly reduces Mask cost Time-to market Risk of design introduction OML follows the mask-based lithography roadmap for Wavelength reduction (193 -> 157 -> 13 nm) k 1 reduction (OPC, off-axis illumination, phase shifting) NA increase (> 1 with immersion) OML complements existing optical lithography equipment OML leverages existing optical knowledge and infrastructure at the customer and within ASML OML does not have the charged particle throughput limitations slide 60

61 OML Summary Using ASML s TWINSCAN platform, optics expertise and Micronic s pattern generation knowledge, we are ready for 193 nm 0.9 NA system in 2007 Resolution 65 nm node with 5 wph (300 mm) The system will be immersion enabled slide 61

62 End slide 62