University of California • Berkeley • San Diego • Los Angeles

The Calculus of Clips

IMPACT Seminarintegrated modeling, process, and computation for technology

http://impact.berkeley.edu

Kameshwar PoollaMechanical Engineering

Electrical Engineering & Computer ScienceFebruary 04, 2008

poolla@berkeley.edu510.520.1150

IMPACT • 2

My commute to work

IMPACT • 3

My Office

IMPACT • 4

My commute home

IMPACT • 5

The Industry Team – Thanks!

IMPACT • 6

Deep thanks to our Champions!

Lars LiebmannFrank SchellenbergAndreas Kuelhmann, Zhenhai ZhuRoawen Chen, Albert WuRandhir Thakur, Klaus SchuegrafDave Kyser, Gerry Maloney, Mamoru Miyawaki, Ron Powell,Michael hart, Xin Wu,Victor Moroz, Dipu Pramanik, John Ennals, Bruno LaFontaine,Vivek Singh, Randy Simmons, John Lee, Shankar Sadasivan,Harry Levinson

IMPACT • 7

More Thanks!

Eric Geigerich handling all the research agreementsFarah Pranawahadi budgets, dealing with UC DiscoveryChang-Rui Yin technical support

Andy Neureuther wisdomCostas Spanos patienceTsu-Jae King youth & energy

IMPACT • 8

The Faculty TeamAlon, Elad EECS UCB Integrated System Design, Mixed-signal ICsChang, Jane ChE UCLA Plasma mechanisms, feature modelingCheung, Nathan EECS UCB Plasma modeling, diagnostics, surface intenractionsDornfeld, David ME UCB CMP modeling, mechanics aspects Doyle, Fiona MatSci UCB CMP modeling, chemistry effectsKomvopoulos, K. ME UCB Surface Polishing, NanomechanicsGraves, David ChE UCB Plasma modeling, diagnostics, surface interactionsGupta, Puneet EE UCLA DfM, Optimization, Variability AnalysisHaller, Eugene MatSci UCB Dopant and Self-Diffusion in Si and SiGe AlloysHu, Chenming EECS UCB Device Modeling, Variability AnalysisKahng, Andrew EECS UCSD DFY, DFM, algorithmsKing, Tsu-Jae EECS UCB Novel Electron DevicesLieberman, Michael EECS UCB RF sources and E&M plasma modelingNeureuther, Andrew EECS UCB Litho, Pattern Transfer, Modeling/SimulationPoolla, Kameshwar ME/EE UCB DFM, Modeling, Computation, Control, MetrologySpanos, Costas EECS UCB IC Process Metrology, Diagnosis and ControlTalbot, Jan CE UCSD Chemical-Mechanical Planarization

IMPACT • 9

Facts & Figures

Budget– Total $ 8.9 M– Industry Cash $ 4.1 M– State Matching$ 4.8 M

Major Equipment Donations– Centura 200 epitaxial tool from Applied Materials– EM Suite Simulation Package from Panoramic Technologies– Wafer/Mask processing credits: Spansion, SVTC, Dupont, Photronics

Personnel– 17 Faculty + 23 Graduate Students + 5 Post-docs + 3 Undergraduates

All research agreements in place– Terms are more favorable to our sponsors than under IMPACT

IMPACT • 10

Activities

Workshop – Save the date!– Wednesday April 09, 2008– AMD Commons 10am to 5pm– Plenary by one of industrial experts?– Feedback on how to get feedback

Bi-weekly Seminar – Run by Professor Tsu-Jae King– Begin with the new faculty: Alon, Kahngg, Hu, Gupta– Will also recruit experts from industry

Round-table Groups– To be arranged

IMPACT • 11

Activities …

Student Internships and Hiring– To be arranged

Recruiting more industry partners– Sister program – Micron, TSMC, UMC, Chartered, Qualcomm– Not all are State of California based, so we won’t use UC Discovery– So it may get complicated!– Will ask for approval from IMPACT sponsors first

Website and Wiki– Needs to be updated– Wiki my be an excellent way to share information, papers, ideas

Feedback: poolla@berkeley.edu

IMPACT • 12

Our Research

Five Inter-connected Research Themes:

Novel Technologies

Tsu-Jae King

CMP

Dave Dornfeld

Etch

Jane Chang

Lithography

Andy Neureuther

Design-Manufacturing Interface

Puneet Gupta

IMPACT • 13

Research Theme A – Litho

Objective Invent a range of approximate-but-fast models based on first principles for assessing manufacturing realities upstream in the design flow

Key Projects

Simulation of electromagnetic effects of mask edgesCompact models for through-focus modelingProcess parameter specific electrical devices and circuitsLitho-aware decomposition for double patterning

IMPACT • 14

Litho – People

Faculty: Neureuther, Spanos, Poolla

Students: 7 Graduate students (some on partial support)

Juliet (Holwill)Rubinstein

Lynn Wang Eric ChinMarshal Miller

Interconnect Delay

Lateral Int. RO Circuits

M ask EM Effects

DP & E-Test

Dan Ceperley

IMPACT • 15

Litho: Through-Focus fast-CAD

Imy CEIImy CEI

RealReal

0.5um Separation

PMF: 0.257

Zernike.txt

IFT

PatternMatcher

SPLAT

MaskLayout

Pattern(coma)

MatchLocation(s)

Aerial Image Simulator

Zernike.txt

IFT

PatternMatcher

SPLAT

MaskLayout

Pattern(coma)

MatchLocation(s)

Aerial Image Simulator

Defocus Spherical HO Spherical

(λ/NA) (λ/NA)(λ/NA)

(λ/NA) (λ/NA)

Coma HO Coma

Mask phases• yellow = 0°• green = 90°• red = 180°

Aberrations Polarization

Line End Shortening, with 2 line surround

0

20

40

60

80

100

120

140

-0.6 -0.4 -0.2 0 0.2 0.4 0.6

Focus, waves RMS

LES

(nm

) Dark Trim

90 Degree Trim

270 Degree Trim

Clear Trim

Att-mask 90o

edge effects

Pattern matching

Compact model though focus

Standard Cell Interactions

2D Lateral Weight DRC

Lateral Influence Functions

IMPACT • 16

Litho: Electrical Test Patterns & Circuits

CD shift with Defocus (OPC)

-1

-0.5

0

0.5

1

1.5

2

2.5

3

3.5

4

0 200 400 600 800 1000

pitch (nm)

CD

shi

ft fr

om n

omin

al (n

m)

20nm

40nm

60nm

80nm

Focus Sensitive Pitch

Focus Insensitive Pitch

Collaborative Platformfor DFM

Collaborative Platformfor DFM

Circuit Simulation

Transistor Modeling Process Simulation

Circuit Simulation

Transistor Modeling Process Simulation

65nm Testchips

Simulation

Parametric Yield Simulator

Solutions across disciplines rather than within disciplines

Cypress Poly Block

Cypress DDLI Block

Metal Active Contact

Corner Poly

Center Poly

Cypress Poly Block

Cypress DDLI Block

Metrology

Quasar OPC Poly

Annular OPC Poly

Cypress Poly Block

Cypress DDLI Block

Metal Active Contact

Corner Poly

Center Poly

Cypress Poly Block

Cypress DDLI Block

Metrology

Quasar OPC Poly

Annular OPC PolyDuPont

Mask

Cypress Wafer Fab. ON

ONONON

ON

Drain 1 Drain 2

Source 1 Source 2

Collaborative Platform for DfM

Screening for Focus Sensitive Candidates Focus Test Patterns

Multi-Student Test MasksNovel Leakage Testing

=+

Contact Pad Thin line of conductive material

Open circuit created when aberration present

Defocus = 0.02 Defocus = 0.2Defocus = 0.0 Defocus = 0.02 Defocus = 0.2Defocus = 0.0

On-Line Database Sim/Exp

IMPACT • 17

Research Theme B – DMI

Objective Increase design predictability, decrease manufacturing cost and yield ramp time

Leverage unexplored interactions between design and manufacturing

Build design-usable models of process and use them to analyze/optimize design

Key Projects

Variation modeling in BSIM compact modelsLeakage modeling, monitoring and optimization in

presence of variabilityImpact of variations on power of mixed-signal circuitsModeling and optimizing for pattern-dependent variations

in standard cell designsDesign-aware mask inspectionComprehensive chip-scale variability modeling

IMPACT • 18

DMI – People

Students– UCB: Kedar Patel, Yu Ben, Kun Qian, John Crossley + 1 TBD– UCLA: 2 TBD– UCSD: 1 TBD

Faculty– Puneet Gupta (UCLA)– Costas Spanos (UCB)– Elad Alon (UCB)– Chenming Hu (UCB)– Andrew Kahng (UCSD)– Kameshwar Poolla (UCB)

IMPACT • 19

DMI:Non-Rectangular Gate ModelingThree components– Poly gate imperfections: well-studied (SPIE’05)– Active rounding: first publication (ASPDAC’08)– Line-end shortening: first publication (DAC’07)– Line-end tapering: under progress (PMJ’08)

Key elements– Equivalent length/width models– Separate modeling for Ion and Ioff– Takes narrow-width effect into account– Modeling for overlay distributions

nominal w/ diffusion rounding delta (%)

leakage (nW) 138.69 83.49 39.8

fall 70.57 68.54 2.9clk→q(ps) rise 76.07 74.07 2.6

fall 20.43 18.08 11.5setup time (ps) rise 42.71 35.01 18.0

Case Study: DFF

IMPACT • 20

Research Theme C – Plasma

Objective Couple models at various scales to understand plasma-surface interaction and predict profile evolution

Key Projects

Develop fast algorithms to determine energy and angular distributions of all plasma species

Develop fundamental models for plasma-surface interactions

Develop predictable profile simulator for etch and deposition processes

IMPACT • 21

Plasma – People

Students/Post-docs – Alan Wu– Monica Titus– John Hoang– Postdocs: TBD

Faculty– Jane Chang– David Lieberman– David Graves Sponsors

Lieberman(Theory, PIC-MCC)

Graves(“Beam” and MD)Surface-scale experiments

Chang(Fluid and DSMC)Feature-scale experiments

Electron energy

deposition

Plasma species energy and angular

distributions

Ion and neutral fluxes

Molecular dynamics

Feature level profile evolution and control

IMPACT • 22

Plasma: Surface & Feature Scale ModelsParticle-in-cell, Monte Carlo collision (PIC-MCC)

Molecular dynamics (MD) simulations and beam experiments

Monte Carlo feature scale model coupling with reactor model

Energy and angle of all species

Fundamental surface reactions

Origin of surface evolution

Couple models at various scales to understand plasma-surface interaction and predict profile evolution

100 nm

100 nm

100 nm

Low DCLow Wb

High DCLow Wb

Low DCHigh Wb2 nm hole in Si etched with 200 eV CF2

+

RF

ESC

ZResist etched by 150 eV Ar+; VUV; at 100C

~ 1000 nm

Energy (eV) Energy (eV)Energy (eV)

10 mTorr 500 mTorr80 mTorr

Neutral energy distributions

Ion and hot electron density

IMPACT • 23

Plasma: Research Integration

X. Hua, et. al, J. vac. Sci. Technol. B 24, 1850 (2006)

193nm PR

248nm PR

LER issue for 193nm PR

Ion Angular and EnergyDistribution (Particle in Cell modeling: M. Lieberman)

Species Distribution (Reactor-Scale Modeling: (D. Graves and J. Chang)

Photoresist Reaction Mechanism (Beam Experiments: D. Graves)

Profile Evolution (Feature Scale Modeling: J. Chang)

• Define testbeds for research integration (LER, Gate Stack Etch, PVD ….. etc.)

++

++

n

nn

n

IMPACT • 24

Research Theme D – CMP

•

Objective Identify key influences of chemical and mechanical activity including the coupling” of CMP/polishing

Develop an integrated model of CMP material removal Verify model thru simulation and test, as a platform for model

based process optimizationKey Projects

Determine fundamental mechanics of the electro-chemical removal of material

Comprehensive model of CMP material removal (including pattern dependency, prior deposition processes, material inducedvariations etc)

Establish mechanical elements of CMP material removal via FEM (incl: pad, abrasive/slurry/device/surface interaction) Understand effects of slurry chemistry on abrasive agglomeration/dispersion and material nano-hardness Investigate material removal at nanoscale

IMPACT • 25

CMP – People

Students– Shantanu Tripathi, UCB (ME/MSE)– Seungchoun Choi, UCB (ME/MSE)– Huaming Xu, UCB (ME)– Moneer Helu, UCB, (ME, NSF support)– Adrien Monvoison, UCB (ME)– Robin Ihnfeldt, UCSD, (Chem Eng)

Faculty– Dave Dornfeld – Fiona Doyle– Jan Talbot– Kyriakos Komvopoulos

IMPACT • 26

Head

Platen

Pad

dow

n-

forc

e

slurry supply

rotation of wafer

head

Wafer 4-12”

Copper

Feature

pad asperity

abrasive particles

100nm-10µm

~1µm

1-10µmPad asperity

Abrasive

Pad/Wafer

Die

Feature/Asperity

Abrasive Contact

Let’s “Google” CMP…effects at different scales

IMPACT • 27

Pad asperity

Abrasive

Abrasive ContactLet’s zoom in…

Slurry agglomerationCharging issues in lapping

IMPACT • 28

Summary of Current Progress

Pad/Wafer (~m)

Die (~cm)

Asperity (~µm)

Feature (45nm-10µm)

Abrasive contact (10nm)

Data structure for capturing multiscale behavior: tree based

multi-resolution meshes

Integrated chemo-mechanical modeling of material removal

Pattern Evolution Model for HDPCVD STI

Pattern density

Chip Layout

Evolution

IMPACT • 29

Research Theme E – Novel Technologies

Objective Investigate advanced device designs, materials, and processes to enhance and reduce variability in bulk MOSFET performance

Develop prediction and abatement methods for systematic variations due to lithography, CMP, and etch processes

Key Projects

Advanced bulk MOSFET design (King Liu) 3-D strain engineering (Cheung)Diffusion in hetero-structures (Haller)Scatterometry-based parameter extraction for calibration

of OPC, CMP, and etch processes (Spanos)Optical metrology for in-situ process monitoring (Spanos)

IMPACT • 30

Novel Technologies – People

Students– X– Y– Z

Faculty– Tsu-Jae King – Eugene Haller– Nathan Cheung– Costas Spanos– Chenming Hu

31

IMPACT • 31

Segmented Bulk MOSFET for Reduced Variability

A multi-gate structure provides for improved control of short- and narrow-channel effects, and reduces STI-induced stress effects.Steep retrograde channel doping reduces VT variation due to SDF.The segmented bulk MOSFET combines these features to reduce variability in performance, while retaining compatibility with strained-Si, high-k/metal gate & active body biasing technologies.

•Continuum doping ID-VG

DR

AIN

CU

RR

ENT

[A/2

0nm

]

30

20

10

σVT = 27.1 mV

•Continuum doping ID-VG

σVT = 10.1 mVSegmented Bulk MOSFETPlanar Bulk MOSFET

30

20

10D

RA

IN C

UR

REN

T [A

/ 20n

m]

DR

AIN

CU

RR

ENT

[µA

/20n

m]

DR

AIN

CU

RR

ENT

[µA

/20n

m]

C. Shin et al. (UC Berkeley), to be published

100 atomisticsimulations

LG = 20nm

EOT = 0.9nm

VDS = 1V

IMPACT • 32

The Calculus of Clips

Speculative seminarWant: reaction, opinion, outrage opinions, and even outrage

Basic Assertion:Working with clips is more efficient and natural than with distances or rules

Pixel size 10nm Mask 20mm x 20mm 2e6 x 2e6 pixelsClip < 2µ x 2µ 2e2 x2e2 pixels

clip

mask

IMPACT • 33

The Opportunity

Industry is moving to regularization– Restricted design rules– Re-use of standard cells, design blocks– BRIX

Potential opportunities for clip calculus– OPC re-use– Faster DRC– Faster hot-spot detection and repair– Faster printability analysis– Fast mask fragmentation for multiple-patterning– Faster RCX

IMPACT • 34

Definitions

Clips– Could be non-rectangular– Standard cells, macros, etc …

– Core “Central” part of a clip– Context “Outer” part of a clip– Library Collection of clips

Core & Context depend on target application– Ex: DRC– Ex: OPC– Ex: Printability analysis

context

core

IMPACT • 35

Ex: DRC

Current Practice– DRC brick is 2,000 pages and exploding– Conventional geometric rule-based DRC at 22nm will be unmanageable

Alternatives: Work with clips not distances– Leverage the speed of pattern match– Produce library of good, bad, or graded patterns– Use library to detect and correct new design layouts

Open problems– Clip-based DRC– Hybrid Rule-Clip based DRC– Correction!– Redundancy removal in rule bricks

IMPACT • 36

Ex: DRC …

Core and Context– Core is the region that is DRC clean given the fixed Context– Cannot assure that Context is DRC clean because that depends on

Context(Context)

Use Case– Library of known DRC clean (in core) clips– In a mask M, use PatternMatch against library L– Can eliminate the core of every matched clip – Will have to do DRC on remaining areas

DRC Clean core

Context

IMPACT • 37

Ex: OPC re-use

Current Practice– OPC is a serious computational bottleneck– Computed many times in design flow– Changes upstream in design flow require re-doing OPC– Printed pattern is then simulated for timing, power, printability, etc

Alternatives: Work with clips– Leverage the speed of pattern match– Produce library of clips and corresponding OPC corrected cores– Use library to reduce OPC computations

Open problems– Respecting and exploiting hierarchy– Library generation – Library extension (for a core, under what other contexts should we use

the same OPC correction)

IMPACT • 38

Ex: OPC re-use …

Core and Context– Core is the region for which OPC corrections are known given the fixed

Context– Do not know OPC corrections in Context because that depends on

Context(Context)

Use Case– Library of clips with corresponding OPC corrections in core– In a mask M, use PatternMatch against library L– Can eliminate the core of every matched clip – Will have to do OPC on remaining areas

Context

known OPC in core

IMPACT • 39

Clip Metrics

What is a good metric on the space of clips?Difficult problemMust also extend to Alternating PSM, Attenuating PSMMetric must also be computable in the language of rectangles

Standard pixel based metrics fail:

Treats each pixel independentlyDoes not respect proximity

‖ci− cj‖2 = # ofdisageeing pixels

IMPACT • 40

Clip Metric depends on application …

Ex: Printability AnalysisWhen are two clips similar?– If the images in Silicon of the core of both clips are similar– Litho model– Clips– Projection onto core

Suggests that we need application dependent weightings– Exposure & Dose sensitivity analysis will have different weights

‖c1 − c2‖ = ‖ΠL(c1 − c2)‖

c1 c2L

Π

IMPACT • 41

Some Computational Problems

mask M , clip c, library

1. ExactMatch: Find all instances of c in M[done & commercialized]

2. RoughMatch: Find all sub-patterns p in M with

3. ExactTile: Tile M with core of clips drawn from library i.e. choose tiling to maximize Area(core union)

4. RoughTile: Tile M with core of clips drawn from library such that context of mask sub-pattern p andlibrary clip c are close:

‖p− c‖≤ ε

L = {ck}

‖p− c‖≤ ε

IMPACT • 42

Some computational problems …

There are other very interesting problems– Net-list covering by clips– Library generation

These are all problems in CS, with a twist– Must compute based on rectangles– Must respect hierarchy– Must work– Must work on huge problems

Algorithm Classes– Randomized– Adaptive

IMPACT • 43

Conclusions

Computation will be a new bottleneck– OPC will get even more computationally expensive– Design rules will become extremely complex – Interactions across features, between layers, among processes

Clip-based paradigms have good potential– OPC re-use– Faster DRC, RC extraction, printability analysis– Faster hot-spot detection and repair– Fast mask fragmentation for multiple-patterning

Many challenging problems– Metrics– Use cases– Fast Algorithms