flcc synergistic design- for-manufacturing (dfm)...
TRANSCRIPT
Overview of FLCC DFM Opportunities, August 28, 2006
FLCC Synergistic Design-For-Manufacturing (DFM) Research
Andrew R. NeureutherUniversity of California, Berkeley
08/28/2006
2
UCB-DFM
Feature Level Compensation and Control:Industry/University collaboration
through the U.C. Discovery Program
Test Structures
Sensors
Physical Models
Statistical Models
ProcessesProcesses MetrologyMetrology
SystemSystem
FLCC
Middle Year 3Planning Year 4
Four year research vision for compensating and controlling variability at the feature level
LithographyCMPPlasma Etch
Device/Diff/IntSensors & Ctrl
Novel VehiclesCollaborative
Experiments
Novel Apparatus
08/28/2006
3
UCB-DFM
The Industry Team
08/28/2006
4
UCB-DFM
Thanks to Our Participating Companies and the U.C. Discovery Program
• Very fortunate to have continuity in industry support from 17 companies with UC Discovery matching for 17 students at a time of cut-backs in university research.
• Support for Our University Research is only sustainable long term if it adds value to the Bottom Line of each of our Participating Companies.
• Mutual value starts with dialog between Working Technologists and Our Future Technologists.
• Looking for Guidance on High-Risk Opportunities where University Talent can do Pioneering Research.
08/28/2006
5
UCB-DFM
FLCC Assets for High-Risk Research• Academic Knowledge and Skills on Broad Fronts
– Physical Mechanisms, Process, Device, CAD, Circuits• Flexible Apparatus, Platforms and Software
– Scientific Apparatus and Instrumented Process Apparatus– Source Code for Rapid Prototyping
• Working Structure of FLCC– Novel Test Vehicle: ‘Zero Foot-Print Metrology’– Multi-Student Test Masks
• Collaboration with FLCC Companies– Phase-Shifting Test Masks– Vanilla CMOS Wafer Fabrication
08/28/2006
6
UCB-DFM
Scientific Apparatus: Leverage Univ. Legacywith minor investment (4%)
Sample and sample carrier
Slurry film Polishing padRotating platen
Slurry deliverypressurePotentiostat
samp
le2
Ref.
electr
ode
samp
le1
Polisher
Atomic Layer Deposition Source(Verify Molecular Dynamics Simulation of Clusters in FC)
0 200 400 6001015
1016
1017
1018
Con
cent
ratio
n (a
tom
s/cm
3 )
Depth (nm)
B in Ge as-imp; 32keV, 1013 cm-2
550°C 30 min anneal - simulation 550°C 30 min anneal - SIMS data B implantation - simulation
Annealing Furnace(diffusion Studies)CMP Wet-Bench
LEO SEM and Hitachi CD-SEM(PC and PCI bus upgrades)
08/28/2006
7
UCB-DFM
Process Apparatus: Industry ContributionAMAT Centura (Y1, Y3), laser for 248 nm AMSL (Y2)
Etch ModelingPlasma (JC, DG)
Test StructuresLith (AN)Zero Foot-Print
End PointSensors (NC)
AMAT Centura
Oxi
de
Si T
renc
h
Al • Emission
• Species• Probe
θi
Future MetalsDevice (TK)
• Compatible Gates for high-K• MEMS
CorrosionCMP(FD)
• Passivationalternatives
08/28/2006
8
UCB-DFMAerial Image Sensor
Novel Vehicle: Zero Foot-Print Metrology
θi
Interdisciplinary• Development
• Heterogeneous Assembly• Metrology and Control Interface•Modeling
• Monitoring Applications• Plasma Etch: End- Point Detection• CMP: End-Point Detection• Lithography: Image Monitor
Metrology wafer
Detection Window
CMP
Plasma Etching
08/28/2006
9
UCB-DFM
Multi-Student Process-EDA Test Mask4 Phase Alt-PSM: High-NA, PolFall 04; Fab: PhotronicsTested: AMD, Nikon, => UCB
Att-PSM: Short Loop NMOSMar 05; Fab: DuPontFab: Cypress
193 nm phases+ binary 248
4 Phase Pin-Hole Alt-PSM: High-NA, Pol; Spring 05; Fab: Benchmark & DuPontTest Plan: Nikon, AMD
2006 Mar Att-PSM for Enhanced NMOS, Apr 4 Phase Att-PSM for PI
08/28/2006
10
UCB-DFM
Validation Experiments in Collaboration on FLCC with Cypress Semiconductor: 2006 Tapeout
Cypress Experiments
Device Process
Circuits
60nm 60.5nm 61nm 61.5nm …
Ioff = 2nA 1.5nA 1.2nA 1nA 0.8nA
VT
Shift with different tips implant
80nm
=+
Super low power design
Finding correlation between gates
Circuits built to measure circuit variations
Using device design to measure CD variation
Characterizing your process
Cypress Experiments
Device Process
Circuits
60nm 60.5nm 61nm 61.5nm …
Ioff = 2nA 1.5nA 1.2nA 1nA 0.8nA
60nm 60.5nm 61nm 61.5nm …
Ioff = 2nA 1.5nA 1.2nA 1nA 0.8nA
VT
Shift with different tips implant
80nm
VT
Shift with different tips implant
80nm
=+ =+
Super low power design
Finding correlation between gates
Circuits built to measure circuit variations
Using device design to measure CD variation
Characterizing your process
Made possible by Feature Level Compensation and Control Program
08/28/2006
11
UCB-DFM
DFM Definition (home made)
• Design for Manufacturing (DFM) consists of fully integrating knowledge of the manufacturing process in to the product design process.
08/28/2006
12
UCB-DFM
Main Point of This Talk
Call to Arms to do more work on DFM in FLCC!
• Change in Mind Set: Linking our science among FLCC disciplines to form a whole is as important as individual new discoveries.
• Change in Investment: More of our effort should go into linking between disciplines and prototyping and applying new collaborative views of the whole.
08/28/2006
13
UCB-DFM
Disclaimer• The views expressed here are only those of the author
and not the FLCC.• In assembling inputs from industry observations
made in conversation have been used – without permission or proper credit and– with serious paraphrasing (filtering).
• The audience is welcome to express their views or identify and correct the phrasing of their observations.
08/28/2006
14
UCB-DFM
Outline
• Background on FLCC• Manufacturing Inabilities are a major Major Concern to
our Industry– Manufacturers, Fab-less Designers, Equipment and Software
Vendors, and 60 Start-Ups
• Key Questions for FLCC• Great FLCC/SRC/DARPA starting position• Key Opportunities and Actions for FLCC• Conclusion
08/28/2006
15
UCB-DFM
Inabilities Experienced by Industry• The inability to even identify the critical path due to lack of predictability
of interconnect effects has resulted in catastrophic time to market delays.• Twenty fold range of variation in standby leakage power when the
maximum can only be 1/3 of the maximum normal operating power.• Fab-Less designers have experience a disastrous variation in yields among
foundries and much of this is attributed to the lack of predictability of the foundry calibrated OPC parameters on the geometries that have passed design rule check in the chip layout.
• Without knowledge of process condition in a foundry it is nearlyimpossible and costly in terms of product delay for a circuit designer to initiate any actions to combat device and interconnect variations.
• New processes and materials such as resists require extensive calibration or model parameters that must be made with limited wafers, a small set of test patterns and fuzzy SEM measurements.
These issues are getting worse in going from 130, 90,65 and 45 nm generation.
08/28/2006
16
UCB-DFM
Inabilities Experienced by Industry (Cont.)• In the FEOL litho combined with etch results in significant variations in
gate length with variation components from feature to wafer length scales.• In the BEOL variations in CMP thickness and etch linewidth transfer result
in nonlinear increases in resistance in interconnect.• The low k1 factors in optical lithography today result in lower image
slopes, more cross-talk among neighbors, higher sensitivity to parameters in resolution enhancement and in feature compensation (not aberrations).
• Double patterning with k1 of 0.2 is a back-up to EUV lithography below 45 nm and there are major layout fracturing, overlay, and hard mask issues.
• Line edge roughness in resist development and its change over correlation lengths of 100 nm may impose a fundamental lithography limit.
• Both the statistical variations in doping and geometry will result in variations among devices.
• Enhancement of transistor performance by strained materials will likely result in pattern dependent stress variations and pattern dependent variation in diffusivity of dopants due to defects created by stress relaxation.
08/28/2006
17
UCB-DFM
Advice on An ApproachThe personal opinion of a technologist in one of our FLLC
companies is that there are two ways to do small feature control:
1) Do it with big R&D, by actually putting in large pilot lines withcutting edge (read expensive) equipment and create integration
practices that result in good control. This I clearly have opinions on how to do, but doesn't really apply to FLCC project.
2) Have a group of smart, experienced scientists like yourself cometogether and define how they can innovate without creating the
more industrial and exhaustive environment described in #1 above.
08/28/2006
18
UCB-DFM
Key Questions for FLCC
• Is there academic content to DFM?
• How do students work with our manufacturing, equipment, and software supporters and 60 start-ups on an industry wide win-win-win-win-win basis?
• How do we deal in an open university with the proprietary nature of sensitive design, process, compensation, and yield data?
08/28/2006
19
UCB-DFM
Key Answers for FLCC• Let’s find the academic content in DFM?
• Lets find a win-win-win-win-win approach for students to work on an industry wide basis with our manufacturing, equipment, and software supporters, and 60 start-ups.
• Let’s make one of the research goals to prototype new alternatives to deal with the proprietary nature of sensitive design, process, compensation, and yield data.
08/28/2006
20
UCB-DFM
Great FLCC/SRC/DARPA starting position• Feature Level Compensation and Control is already a subset of
DFM• The breadth of FLCC covers most of the processes• Our models provide understanding and efficient
characterization • Our test-pattern and validation experiments are of interest
industry wide• The “Collaborative platform for DFM” pioneered by Wojtek
Poppe is recognized as a first in the DFM field.• Our 2D maximal impact test patterns go where design rules
don’t reach• Our Pattern-Matcher software supports unique fast-CAD ideas
08/28/2006
21
UCB-DFM
Collaborative Platformfor DFM
Collaborative Platformfor DFM
Circuit Simulation
Collaborative Platform for DFM
Transistor Modeling Process Simulation
Validation Experiments
Manufacturing
Design
WojtekPoppe
SRC/DARPA
08/28/2006
22
UCB-DFM
Collaborative Platform for DFM: Philosophy
Communication is the Issue• Many domains of expertise and viewpoints
– Process/Metrology/Device/Circuit/CAD
• Moving information between domains is as important as the depth of understanding within one domain
• The university atmosphere is a great venue for creative ideas on DFM– Process/Metrology/Device/Circuit/CAD
08/28/2006
23
UCB-DFM
Parametric Yield Simulator (PYS)
Module 1Processing
Module 3Circuit
Module 2Device
BSIM transistor
model
Circuit Simulation across characterized process window
Non-rectangular transistors
08/28/2006
24
UCB-DFM
Verification Opportunity at Cypress SemiconductorControl of LithoControl of Mask and OPC
Different Illumination (Annular, Quasar, NA, Sigma)
Metal Active Contact
Corner Poly
Center Poly
Quasar OPC
Annular OPC
Metal Active Contact
Corner Poly
Center Poly
Metal Active Contact
Corner Poly
Center Poly
Quasar OPC
Annular OPC
Programmed defocus
08/28/2006
25
UCB-DFM
Characterizing All Sources of Process Variations
5BCAM
Leff Spatial Variation Decomposition
= +
Average Wafer Scaled Mask Errors Across-Field Variation
Across-Wafer Variation
+ + +
Die-to-Die Variation “Random” Variation
Paul Friedberg 2005
Bias features appropriately in Calibre.
Characterize your process
Characterized CD distribution can stem from any process variation. Multiple simulations can be done for various dies across the wafer.
08/28/2006
26
UCB-DFM
Liang Teck Pang’s Ring Oscillator Test Chip
Quantify the effects of layout, Vdd and body biasing on variations by measuring ring oscillator frequencies and leakage current.
1a 2a 3a 4a 5a 6a
M1
1b 2b 3b 4b 5b 6b
Dummy polySi
90nm CMOS16 columns and 10 rows(160 tiles) of each layout configurationDie size around 1.5mm x 1.2mm (incl. I/O pads)Ring Oscillator frequency and leakage current measurement results for 1 chip at Vdd=1V and 0.81V
Die photo of testchip
Previous experiments using ST’s 90nm flow showed various types of systematic variation. Cypress experiments to identifying the major causes.
08/28/2006
27
UCB-DFM
Collaborative Platform for DFM: StrategyLeverage what is Known• Much of the Device and Circuit Variation can be
attributed to a Few Physical Causes– Takes only a few key parameters
• Local effects are not only Correlated Laterally but they are even Interdependent Laterally– There is action where design rules don’t reach
Leverage Software Know How• Perturbational Pattern-Matching methods are Fast and
accurate– Real-time assessment during design likely possible
08/28/2006
28
UCB-DFM
Going Where Design Rules Don’t Reach
Example: proximity effect influence function for coma. • Function is about 5 feature sizes in diameter and easily reaches across
cell or compaction boundaries. • By computing influence functions for diffraction limited proximity Z1,
Defocus Z4 and Coma Z7 it is possible to quickly assess image changes through the process window and along a scanner slit.
• Similar patterns exist at a mm scale for assessing flare and CMP.
08/28/2006
29
UCB-DFM
Leverage Prior SRC IP from 2004: Pattern Matching Methods for Linking TCAD and EDA
LAYOUTPATTERN
SPLAT
AB-CAD
.gds
ExtractedMatch Region
• Prototype system with extraction to TCAD for aberrations, flare, CMP, reflective notching, etc.
• Validation with perfect SPLAT correlation
• New data structures and algorithms give 400X OPC speed
Frank GennariSRC PhD 2004
3mmx3mm abacus IC
Delta E vs. Match Factor
y = 0.2855x + 0.0037R2 = 0.9831
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6
Match Factor
SPLA
T de
lta E
(Sim
ulat
ed)5.6GB hierarchical layout with OPC,
>100M rectangles/polygons, two matching test patterns, at every corner:17 min on a 2.8 GHz desktop (with file reading) and 1.3 GB RAM
Gennari SPIE 04
08/28/2006
30
UCB-DFM
Process Aware EDA Toolkit• New SRC Grant supporting 2 students for 3 years• Goes the next level in communication from
Collaborative Platform for DFM to Real-Time Assessment and Feedback During Design
• Strategy– Use only a few key physical parameters– Use lateral influence functions to go where design rules
don’t reach– Utilize the Speed of Pattern Matching and Perturbational
Methods
08/28/2006
31
UCB-DFM
UCB DFM-CAD TEAM
Robustness Metric and
Drag and Drop Hot-Spot Fixer
Lynn WangWojtek Poppe
Juliet HolwillEric Chin
CrosstalkJae-Seok Yang
Interconnect Delay
Lateral Image Interactions and Placement
08/28/2006
32
UCB-DFM
Leverage Current FLCC OpportunitiesTarget more in the DFM area with the resources we
have to attract additional long term future funding.• Understand relative contributions to device variations
from doping variations and feature geometry variations.
• Link CMP variations and pattern transfer feature size change including roughness contributions non-linear resistance changes in interconnect.
• Global non-uniformity in plasma etching possibly due to chamber flow, pattern global and micro loading.
• Industry-wide (universal) test patterns for calibration of OPC and PPC models.
08/28/2006
33
UCB-DFM
Recommended Action for Higher Leverage
Form a new operational structure that enhances and demonstrates the inter-area contributions among all projects to our theme of feature level compensation and control.
• This might be a student task force that summarizes likely physical causes and novel test patterns for distinguishing and quantifying them.
• We would include these patterns on test masks for experiments in the university and at Cypress and then interpret the results.
• We would also make the layouts and analysis descriptions available on the web to participating companies.
08/28/2006
34
UCB-DFM
Possible New Year 4 Proposal Ideas• No-fault assurance: The ability to prescreen complete chip layouts for
potential effects visa via the richness of a test pattern set. It is likely possible that this testing could be carried out by with SRC software by a third party or by a “virtual third party PC.” Alternatively we could find mathematically complete and physically reasonable test vectors and play them against both the test and chip layouts.
• System Information View of Characterization: We could develop aninformation systems approach to assess predictability of feature printability given the limited number and fuzziness of SEM measurements in parameter calibration. SEM to database comparisons might be possible by collaborating with AMD and or vendors on comparison of SEMs of printed patterns and layout.
• Understand the generation of defects from patterning and thermaltreatments of strained silicon layers and their role in dopant and Gediffusion.
08/28/2006
35
UCB-DFM
Conclusion
• DFM is a communications issue and moving information across boundaries is as important as the depth of understanding within the domains.
• We have expertise in all aspects of the DFM problem including process, device, circuits, CAD and technology CAD.
• Our science and instrumented apparatus in FLCC give us much more telling view into the physics than is available on production equipment.
• We have an important first that our students and their peers have their arms around the entire problem.
• Can we create strong win-win-win-win-win ideas that industry ‘Cannot afford not to fund”?