04/04/2007 FLCC FLCC – Sensors & Control Feature-level Compensation & Control Workshop April 4th, 2007 Sensor and Control A UC Discovery Project
04/04/2007FLCC
FLCC – Sensors & Control
Feature-level Compensation & Control
WorkshopApril 4th, 2007
Sensor and Control
A UC Discovery Project
04/04/2007
2
FLCCFLCC – Sensors & Control
Current Milestones (Year III, 1/27/06 to 1/26/07)
• Complete experimental study for CD non-uniformity reducing across the litho-etch sequence (SENS Y3.2)
– Experimentally verify DI & FI CDU improvement using model based optimal control of PEB with various CD objective functions. DONE
• Using Spatial CD Correlation in IC Design (SENS Y3.3) – Develop test structures and measurement plans for extracting spatial correlation
characteristics. DONE• Aerial Image Metrology (SENS Y3.4)
– Complete the micro-assembly of the commercial CCD with the Si carrier wafer. Integrate the aperture mask and the CCD arrays. IN POGRESS / EMPHASIS SHIFTED TO APPLICATIONS
• Modeling and demonstration of metrology wafer for detection and thin-film roughness monitoring.
– Initiate prototyping of wireless data acquisition/transmission and evaluate performance with measurements made in experimental systems. DONE
04/04/2007
3
FLCCFLCC – Sensors & Control
CD Uniformity ControlCharlie Zhang
• Making each process step spatial uniform is prohibitively expensive
• Our approach: manipulate PEB temperature spatial distribution of multi-zone bake plate (and die-to-die dose) to compensate for other systematic across-wafer CD variation sources
CDU Control Framework
04/04/2007
4
FLCCFLCC – Sensors & Control
Long Term Overall FI CD Improvement ~35%in recently completed experiment at AMD/SDC
across-wafer sigma of 250 1:1 lines, using CDSEM
FI CDV before and after control
0
0.5
1
1.5
2
2.5
Before After
FI C
DV
1 si
gma(
nm) Wfr1
Wfr2
Wfr3
Wfr4
Wfr5
Wfr6
DI CDV before and after control
00.5
11.5
22.5
33.5
44.5
Before After
DI C
DV
1 si
gma
(nm
) Wfr1
Wfr2
Wfr3
Wfr4
Wfr5
Wfr6
DI CD uniformity is sacrificed in order to optimize FI CD uniformity
σ=1.36nm
m=141.9nm
σ=1.21nm
m=142.1nm
σ=1.26nm
m=141.7nm
σ=1.14nm
m=141.5nm
σ=1.41nm
m=141.4nm
σ=1.39nm
m=142.4nm
Qiaolin (Charlie) Zhang, on internship at AMD/ Spansion 2005-06
04/04/2007
5
FLCCFLCC – Sensors & Control
Zero-footprint Optical Metrology Wafer – Prototyping and ModelingData Transmission
Photo/RF Transmitter
Dielectric Layer as Optical Window
Si
Battery Data Acquisition Unit
500µm
Data Transmission
Photo/RF Transmitter
Dielectric Layer as Optical Window
Si
Battery Data Acquisition Unit
500µm
GSR: John Gerling
Faculty: Professor Nathan Cheung
04/04/2007
6
FLCCFLCC – Sensors & Control
Current Milestones (MMC Y4.1)• Simulate and experimentally verify the monitoring of
lateral patterns development (e.g. wet etching high aspect-ratio contact holes in dielectrics).
• Enhance signal-to-noise ratio for better sensitivity with high-brightness LEDs.
• Investigate multi-stage mechanisms of Cu etching near end point.
04/04/2007
7
FLCCFLCC – Sensors & Control
Current Achievements• Simulated monitoring of high aspect-ratio contact holes
in dielectrics.• Enhanced signal-to-noise ratio for better sensitivity with
high brightness LEDs and investigated effects of manufacturing noise on metrology system.
Work In Progress• Investigate Cu etching mechanism.• Experimental verification of lateral pattern etch
monitoring (e.g. wet etching of high aspect-ratio contact holes and trenches in dielectrics).
04/04/2007
8
FLCCFLCC – Sensors & Control
Simulation of Contact Hole Wetting• Goal: Simulate monitoring of wetting in high aspect-ratio contact holes in dielectrics.• Simulation Setup: Thin film stack, simulated wavelengths of 463, 525 and 632 nm.
Ambient layer [Vacuum], z=semi-infiniteSi3N4 layer (1), z=650.00 nm
4X [SiO2/Vacuum] multilayer, N=20, d=6.25 nm, Gamma=0.700SiO2 layer (2), z=4.37 nmH2O or Vacuum layer (3), z=1.88 nm
Substrate layer [H2O], z=semi-infinite
Effective Medium Theory: Effective refractive index of layer used in simulation is equal to volume fraction of A*(na+ika) + volume fraction of B*(nb+ikb)
Justification: Thickness of layer and lateral dimensions of features < wavelength of photon
0% 25% 50% 75% 100%
04/04/2007
9
FLCCFLCC – Sensors & Control
nm525=λ
nm463=λ
Simulation of Contact Hole Wetting Cont.
Reflectance as a function of angle for 463 nm and 525 nm incident light for 0%, 25%, 50% , 75%, and 100% wetting conditions.
04/04/2007
10
FLCCFLCC – Sensors & Control
• Sweet spots for the angular reflection between 40-60 degrees for substrate thicknesses varying from 500 nm to 2000 nm.
nm632=λ
Simulation of Contact Hole Wetting Cont.
04/04/2007
11
FLCCFLCC – Sensors & Control
HB LED Apparatus
• SemiLEDs Peak Wavelength: 463±15 nm• Advanced Photonix Inc. Photodiode: PDB-V106 • Sample: Spin-coated Shipley S1818 photo resist • on Pyrex/Glass 520-540 µm substrate• Etchant: Photoresist Stripper PRS3000• Data Collection: HP4145
04/04/2007
12
FLCCFLCC – Sensors & Control
04/04/2007
13
FLCCFLCC – Sensors & Control
Signal-to-noise ratio enhancement with HB LED• PR etch monitoring with room light only, then room light plus LED.
• Significantly less dependence on ambient light levels than with the low power LED used in previosu prototypes
0 10 20 306.875
7.000
7.125
7.250
7.375
7.500
Room light ON Room light OFF
Sign
al (V
)
Etching time (s)
Experimental Condition: Sputtering deposition, etching solution: Cyantek CR-7 (Perchoric based), Glass window thickness 500µm, LED peak wavelength 463±15nm, Cu thickness 60-70nm. I = LED current. Light on/off indicates the external room light.
HB LED 2.7 um PR Etch with 0.1 s resolution
6.50E-05
7.00E-05
7.50E-05
8.00E-05
8.50E-05
9.00E-05
0 10 20 30 40 50
Time [s]
Cur
rent
[A]
HB LED + Room Lights
Low Brightness LED
04/04/2007
14
FLCCFLCC – Sensors & Control
Effect of Specular Reflections• Silicon wafer mirror to reflect ambient and LED light back to photodiode.
Conducted under various lighting conditions.
• Specular reflection from strong point sources can add significant noise to system. Specular reflection from room lights is almost negligible. This is a benefit of the HB LED.
Specular Reflection
0.0E+00
2.0E-05
4.0E-05
6.0E-05
8.0E-05
1.0E-04
1.2E-04
0 5 10 15 20 25 30
Time [s]
Cur
rent
[A] Room Light
room Light + LED
Room Light + LED +Flashlight
04/04/2007
15
FLCCFLCC – Sensors & Control
• Conducted with room light and LED on for sample thickness ranging from 1.5 um to 2.7 um. Drop of PRS3000, monitoring done with HP4145B at 0.05 second resolution.
• 1) transient effect from PRS3000 drop, • 2) linear or slight parabolic decrease in signal as etch proceeds • 3) flat response when etch is complete.
Photoresist Wet Etch Monitoring
PR Etch 50 sec with 0.05 sec resolution
6.0E-05
6.5E-05
7.0E-05
7.5E-05
8.0E-05
8.5E-05
9.0E-05
9.5E-05
0 10 20 30 40 50
Time [s]
Cur
rent
[A] 1.5 um
1.8 um2.1 um2.7 um
04/04/2007
16
FLCCFLCC – Sensors & Control
Future Milestones
• Experimental verification of lateral pattern etch monitoring (e.g. wet etching of high aspect-ratio contact holes and trenches in dielectrics).
• Investigate Cu/PR etching mechanism.• Prototyping the self-contained metrology unit
with wireless I/O.
04/04/2007
17
FLCCFLCC – Sensors & Control
Integrated Aerial Image Sensor
Mask Pattern
Black: ideal, diffraction limited opticsRed: optics with typical aberrations
Students: Jing Xue, Chaohao Wang, Yu Ben
Faculty: Prof. Costas J. Spanos• Students: Jing Xue, Chaohao Wang, Yu Ben
• Faculty: Prof. Costas J. Spanos• Title: Integrated Aerial Image Sensor (IAIS)
04/04/2007
18
FLCCFLCC – Sensors & Control
2007 Main Objective• Complete IAIS prototype and test within the UC Berkeley Microlab• Interact with key company collaborators for testing beyond the UC
Berkeley Microlab.• Explore fully integrated design with special purpose UV detectors
• Complete the Aberration Monitor, specified on the defocus and coma target
• Complete the Process Window Monitor (combine with ODP test pattern)
• Explore the Aerial Image Distortion & Variation, OPC and Post-OPC properties, combining with DFM work
• Investigate the Polarization Monitor properties
04/04/2007
19
FLCCFLCC – Sensors & Control
IAIS System Description
Litho.
λ, ΝΑ, σ, mask, ab., …
Aerial Image(AI)
IAISAperture Mask
IAISPhoto-detector
Detector Image(DI)
Photo-current(PC)
Φ1 Φ2 Φ3 Φ1 Φ2 Φ3
Substrate
Photo- detectorAperture mask
Mask aperture
IAIS AMF: x -> ξAI DI
x ξ Noise thresholder PC
yOutputInput
04/04/2007
20
FLCCFLCC – Sensors & Control
IAIS Metrology Capabilities
*292nm 1:1 line/space pattern, 193nm, 0.75NA
Nonab: slope= 2.65 /(λ/NA); CD = 145 (nm)Df .06: slope= 2.29 /(λ/NA); CD = 145 (nm)Df .08: slope= 1.81 /(λ/NA); CD = 143 (nm)
a. Image intensity &intensity variation
b. Image contrast & contrast variation
Nonab: slope= 12.13/(λ/NA); CD = 145 (nm)Df .06: slope= 11.89 /(λ/NA); CD = 145 (nm)Df .08: slope= 8.28 /(λ/NA); CD = 143 (nm)
d. Image slope & slope variation
c. CD & CD variation e. Image shift & its variation
04/04/2007
21
FLCCFLCC – Sensors & Control
IAIS Aperture Mask PatterningFolded Spatial shift Matrix Measurement Results:
Shift 300.6nm
Shift 299.4nm
Shift 303.8nm
Distance 7.249 µm
Distance 7.249 µm
Distance 7.252 µm
)(550 nm±Aperture Groups spatial shift
04/04/2007
22
FLCCFLCC – Sensors & Control
IAIS Microlab Prototype (I-line)Aperture mask pattern
Circuit connection: BCAM v.2 CCD
Assembly measurement results by Profilometer:
CCD dummy chip is co-planar to the wafer carrier with maximum height difference of 1.17µm, and tilt angle 0.0014o (along x)
04/04/2007
23
FLCCFLCC – Sensors & Control
IAIS Application - Aberration Monitoring
p
d
wb
wn d
90o/180o -90o/-180o0o 0o
wb
0o
Top
mas
k de
sign
0.49 0.72Distance between 0o and the phase shift grating determines the sensitivity of this aberration and the orthogonality to the other aberrations
Defocus highest sensitivity occurs at 0.556 (λ/NA); highest orthogonality to spherical occurs at 0.53 (λ/NA)
* A. R. Neureuther, etal., Proc. SPIE, 2001; G. Robins, PhD 2005
Def
ocus
Tar
get
04/04/2007
24
FLCCFLCC – Sensors & Control
IAIS Application – Aberration Monitoring
%5.82=∆I %5.82=∆I
Spherical TargetDefocus Target
%2.18/;%3.48/ 21 =∆=∆ nonnon IIII
Defocus:
Spherical:
%8.165/;%5.82/ 21 =∆=∆ nonnon IIII Defocus: %63/;%5.59/ 21 =∆=∆ nonnon IIII
Spherical: %95/;%9.216/ 21 =∆=∆ nonnon IIII
04/04/2007
25
FLCCFLCC – Sensors & Control
Aberration Analysis System• Characterizing the aberrations of an
exposure system is crucial for design verification as design rules shrink– Valuable information for performing mask
correction– Needed for OPC generation and OPC
verification
• Conventional methods– Time consuming– Indirectly done through OPC calibration
patterns– Cannot obtain a complete aberration map– Cannot be used to compare individual optical
columns
• A fast, in-situ aberration measurement is desired
Students: Yu Ben, Jing Xue, Chaohao Wang
Faculty: Costas J. Spanos
04/04/2007
26
FLCCFLCC – Sensors & Control
2007 Main Objective• Refine the Zernike coefficients extraction scheme to
achieve better robustness
• Adjust the grating and pellicle design with partial coherence taken into account
• Design new feature adaptable to Optical Digital Profiling (ODP) analysis
04/04/2007
27
FLCCFLCC – Sensors & Control
The Problem – Partial Coherence “blurs” Aberrations
Mask Substrate
Mask Pattern
Diffraction Pattern
Mask Substrate
Mask Pattern
Diffraction Pattern
Partial coherence prevents the probing ray from sampling a well-defined point on the pupil plane, yielding an averaged value of phase distortion.
04/04/2007
28
FLCCFLCC – Sensors & Control
Pupil Plane Sampling• Assuming a perfectly coherent,
point light source, pupil plane sees the Fourier Transform of the reticle.
• Points at pupil plane correspond to features with different spatial frequencies
• Gratings with different periods can diffract light onto different points on the pupil plane
• An opaque pattern on pellicle can be used to select the desired probing rays1
mask
pellicle
1Nigel R. Farrar”, Adlai L. Smithb et al, Proc. SPIE, Vol 4000 (2000)
04/04/2007
29
FLCCFLCC – Sensors & Control
Pupil Sampling With Pellicle Obscuration
•Radius ~ 844.6 µm on mask, 168.9 µm on wafer
04/04/2007
30
FLCCFLCC – Sensors & Control
Aberration Measurement via Image Shift Detection• Wave aberration is equivalent to image shift
– Lateral shift amount = – For a system with Strehl ratio larger than 97%, total variance can be
estimated to be less than 0.03 following Maréchal formula
• The corresponding aberration is in the order of a few hundredths of a wavelength
• The lateral shift is in the order of a few nanometers• Two approaches
– Integrated Arial Image Sensor
– Optical Digital Profilometry (ODP)
( )22 21 φσ−≈S
0 5 10 15 20 25 30 35 40 45 500.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Displacement (nm)
Nor
mal
ized
Inte
nsity
TETM
TEMPEST simulation result
( ) ( )fwfw ⋅− λ/)0()(
04/04/2007
31
FLCCFLCC – Sensors & Control
Effect of Partial Coherence• Partial coherence causes the probing ray to sample a widened area
• The measurements reflect an averaged pupil map corresponding to various illumination schemes
Averaged pupil map with incoherent (σ=0.3) illumination
Averaged pupil map with quadrupole illumination
Original Pupil Map
Zernike coefficients can be extracted by least mean square fitting
Pupil map can be reconstructed
Reconstructed pupil map at the presence of quadrupole illumination
04/04/2007
32
FLCCFLCC – Sensors & Control
Zernike Coefficient Extraction with Partial Coherence
∑=j
jjZaw ),(),( θρθρ
∑ ⋅=j
nmjjnm Zaw )},({)},({ θρθρ OO
An over-determined linear system which can be solved by least mean square fitting
O{ ⋅ } represents the averaging operation
Extraction method:
In this simulation:72 sampling points (8 in radial direction, 9 in angular direction, evenly distributed in both directions)Coefficients are randomly chosen to generate the original aberration function, which corresponds to Strehl ratio of 97%A constant background ( ~ 1% of total variation) is introduced as noise
04/04/2007
33
FLCCFLCC – Sensors & Control
Coherence Enhancement• Substrate back surface can be patterned1 to enhance
the coherence• Partial coherence factor can be reduced to ~ 0.1 with
T = 6 mm, and Lx = 100 µm
Mask Substrate
Pattern
Substrate Back surface
1Shinroku Maejima et al, Proc. SPIE, Vol 6283, 62833A-1 (2006)
mm)(6=T
167.0mm6
µm100==≈
TLxα
16.058.0NAsin max ===
Mθ
10.0max
≈=θασ
04/04/2007
34
FLCCFLCC – Sensors & Control
Future Goals• Complete system simulation confirming the
pellicle pattern functionality • Design patterns compatible with ODP
measurements• Build system prototype and perform ODP
analysis to extract aberration information
04/04/2007
35
FLCCFLCC – Sensors & Control
Modeling Spatial Gate Length Variation in the 0.2µm to 1.15mm Separation Range
• Manufacturing-induced variation in device parameters leads to variability in circuit performance
• Two approaches to address this concern:– Tailor IC design to minimize
sensitivity to parameter variation– Use process control to reduce
manufacturing variation• Investigate these approaches
through Monte Carlo analysis of canonical circuits
• For accurate, useful predictions, Monte Carlo framework must model reality very well
Specific focus of this work: detailed spatial variation of gate length
Student(s): Paul Friedberg, Qian Ying Tang, George Cheng, Willy Cheung, Kun QianFaculty: Costas J. Spanos
04/04/2007
36
FLCCFLCC – Sensors & Control
Short(µm)-Range CD Test Structures• Dense ELM base case test structure (90nm tech.):
• To increase measurable range, insert dummy lines:
• Maximum 255 dummy lines: 1.15mm range, 71µm pitch
Base test structure: Range = 4.76µmPitch = 280nm
Range
Range = 9.52µm Pitch = 560nm
04/04/2007
37
FLCCFLCC – Sensors & Control
Full Chip View2.
35m
m
1.8mm
04/04/2007
38
FLCCFLCC – Sensors & Control
Raw Data (Averaged Across Chips)• 25 total pre-diced chips (wafer location unknown)
• Two instances per chip (horiz, vert), 9 DUT’s per chip, 17 measurement positions within each DUT
3825 total measurements for each orientation
1
4
7
10
1316
123456789
(bla
nk)
130.5
131.5
132.5
133.5
134.5
135.5
136.5
Avg
. CD
(nm
)
DUT
position
• Vertical orientation:• Horizontal orientation:
Avg
. CD
(nm
)
DUT
position1
4
7
10
1316
123456789
(bla
nk)
129
130
131
132
133
134
135
04/04/2007
39
FLCCFLCC – Sensors & Control
Pattern Density-Dependent Etch Model• Etch microloading can explain the deterministic pattern
observable in measurements from DUT’s 1-4
• Model contains two density metrics1) “Local” density: intended to capture shorter-range etch effects, we count
the total polysilicon area within a moving window that is centered on the line in question:
2) “Global” density: intended to capture longer-range etch effects, we calculate the total area of polysilicon within a given DUT
04/04/2007
40
FLCCFLCC – Sensors & Control
0
0.2
0.4
0.6
horiz vert
Varia
nce
(nm
^2)
pattern dependentrandom
Breakdown of µm-Scale Variance
• Etch micro-loading model accounts for roughly 1/3rd of the µm-range spatial variation
• Does autocorrelation exist in the remaining residuals?
04/04/2007
41
FLCCFLCC – Sensors & Control
Autocorrelation in Residual Component• Autocorrelation analysis for model residuals:
• Very little correlation (for the most part, not different from zero correlation within statistical significance)• Consistent positive correlation due to slight incompleteness in modeling of deterministic variation.
Horizontal Orientation Vertical Orientation
95% confidence interval around zero correlation
04/04/2007
42
FLCCFLCC – Sensors & Control
M1,n-1M1,n-2 M1,n
Gate
Pulse
Clk1
Clk2Pulse
Source force
Source sense
Drain force
Drain sense
MOS Test Structures• NMOS array equipped to allow full 4-pt probing:
M1,nM1,n-1M1,n-2
1
0 01000…
04/04/2007
43
FLCCFLCC – Sensors & Control
Future Goals• Milestone M45: Spat. CD Correlation in IC
Design• May 1, 2007: Submit novel ELM and MOS test
structures to foundry.• Spring, 2007: Incorporate CD measurements
into Monte Carlo framework.
• This work is also funded by SRC grant 1324.001.
04/04/2007
44
FLCCFLCC – Sensors & Control
Impact of Line Edge Roughness on sub-42nm Device Variability
• Intrinsic parameter fluctuations greatly impact circuit performance and yield. The three primary sources of variability that have peaked interest in the scientific community are:– Line edge roughness (LER)– Gate oxide thickness variability (GOV), and– Random dopant fluctuations (RDF)
• The formation of LER is stochastic event and its origins lie in the resist patterning process. Polymer aggregates contained in resist films have been identified as the primary cause of LER.
• Novel double-gate and triple-gate device (FinFET like) structure have been suggested as replacements for traditional CMOS.
• We are interested in understanding the impact of LER on the device performance of such structures.– In particular, we would like to model device
parameters such as Vt, Ioff, and Idsat using LER descriptors (α, ξ, and σ )
Student: Kedar Patel
Faculty: Costas Spanos
04/04/2007
45
FLCCFLCC – Sensors & Control
Objectives• Gate LER Investigation
– Generate a model to produce LER with α, ξ, and σ as parameters
• Done. Used MATLAB.
– Generate a basic structure in Sentaurus for the device
• Done.
– Introduce LER in the Sentaurus structure– Propose a model for device behavior using LER
descriptors (α, ξ, and σ )
04/04/2007
46
FLCCFLCC – Sensors & Control
What is Line Edge Roughness (LER)?1
2
1
1
N
ii
LWR
L L
Nσ =
⎡ ⎤−⎢ ⎥
⎢ ⎥=−⎢ ⎥
⎢ ⎥⎣ ⎦
∑
( )
2 2 2
2 2
2
2 1
LWR L R L R
L R LER
LWR LER
σ σ σ ρσ σσ σ σ
σ ρ σ
= + += ≡
= +
LWR
LER
W
<L>
LWR
LER
W
<L>
LWR
LER
W
<L>
Line edge roughness (LER) and line width roughness (LWR) are oftenused synonymously. Mathematically, they are related but different…
04/04/2007
47
FLCCFLCC – Sensors & Control
Transfer of LER
Schematic of a typical gate stack
Resist
Poly-SiGate Dielectric
Substrate
Hard MaskBARC
Pawloski (SPIE 2006)
04/04/2007
48
FLCCFLCC – Sensors & Control
Illustration of LER in FinFET
+ =Fin
Gate
FT
B
h
t
L
+ =Fin
Gate
FT
B
FT
B
h
t
h
t
LLT
Gate LER
t
F and B
Fin LER
h
04/04/2007
49
FLCCFLCC – Sensors & Control
Unfolded FinFET
FG
Body
BG
S D
h
h
T
B
F
t
<L>
Etch Progression
Average Gate Length
Tapered Etch Profile(Curtain Effect)
Li
X X’
Y Y’
Gate
Fin DS
Top View
Distribution of determines the FG andBG placement and CD
04/04/2007
50
FLCCFLCC – Sensors & Control
LER Descriptors• 3σ variation is not a sufficient
descriptor of LER
• Complete description of LER is achieved by a power spectral density function, a first-order autoregressive process or height-height correlationfunction and σ
• Extraction of these descriptors starts with DSP of the SEM image
04/04/2007
51
FLCCFLCC – Sensors & Control
Future Goals• Near Term
– Create statistical variability model with LER and long distance large scale variability of CDs.
– Assess the impact of Ioff and Ion for sub-42nm nodes• Long Term
– Fin LER Investigation• Study mobility degradation
– LER in contact/via structures• Contact edge roughness can greatly change the resistance
04/04/2007
52
FLCCFLCC – Sensors & Control
Circuit Size Optimization with Multiple Sources of Variation and Position Dependant Correlation
• Digital Circuit Sizing Optimization Problem– Goal: size the gates in a combinational logic circuit– Minimize the effects of individual gate delay variations and
spatial correlations on the overall circuit delay• Previous Work: Geometric Programming approach
– Objective:– where: Di = nominal delay for gate I
k = a constant ~ 2derived from Pelgrom’s
Modelwith model parameter γ– Constraints: Fixed maximum total circuit area
)]}([min{max }{ DkD ipi
ipathsallcircuitp σ+∑∈
∈
iii DxD )()( 2/1−= γσ
† S. Boyd, S.-J. Kim, D. Patil, and M. Horowitz , “A Heuristic Method for Statistical Digital Circuit Sizing ,” 31st SPIE International Symposium on Microlithography, February 2006.†† M. Pelgrom, A. Duinmaijer and A. Welbers, “Matching Properties of MOS Transistors,”, IEEE J. Solid-state Circuits, Vol.24, No. 5, pp.1433-1439, Oct. 1989.
†
††
Students: Qian Ying Tang and Paul Friedberg Faculty: Costas Spanos
04/04/2007
53
FLCCFLCC – Sensors & Control
Spatial Correlation based Modeling
)]}([min{max }{ DkD ipi
ipathsallcircuitp σ+∑∈
∈
• Adding delay variation dependence on Leff in the objective function:
iiLeffiithi DxDxD )()()( 2/12/1 −− += γγσ
]}[)]([min{max,,
22}{ ∑∑
≠∈∈∈ ++
jipjijiiji
piipathsallcircuitp kDkD σσρσ
iiLeffiithi DxDxD )()()( 2/12/1 −− += γγσρij ≡ spatial correlation between gate i and j with separation dij
where:
• Adding variation dependence on Leff and Spatial Correlation
where:
XL characteristic correlation lengthρB characteristic correlation baseline
Large scale model ⎩
⎨⎧
≥≤−−
=LijB
LijBLij
XdXdXd
ρρ )1(/1
04/04/2007
54
FLCCFLCC – Sensors & Control
• A model for calculating the parameter mismatch ∆P between two rectangular devices (WxL) separated by a distance Dx.
• Pelgrom’s Model suggests a spatial correlation structure implicitly– devices situate close together have higher degree of correlation/similarity than
devices separated by further apart
• Pelgrom’s model can be used to model the variance of the gate delay. • The major difference between this approach and the spatial correlation based
model: – Both terms are used vs. only the first term is used for deriving the variance of the
gate delay.
)],(),,([2)],([)],([
)],(),([)(
2211222
112
221122
yxPyxPCorryxPyxP
yxPyxPP
−+=
−=∆
σσ
σσ
WLAp
2 22xp DS+
variance of the parameter “spatial correlation”
Pelgrom’s Model and Implications
222
2 )( xpp DS
WLA
P +=∆σ
04/04/2007
55
FLCCFLCC – Sensors & Control
Sources of Variation• Original Pelgrom’s Model accounts for systematic across-wafer
variation only.
• A modified Model that accounts for all sources of variation is required- Wafer-level variation
• Slowly varying parabolic shaped variation across entire wafer• A consequence of loading effects in etching or deposition• Contributes a random component to the parameter value for each device in
a die.- Within-field variation
• Varies systematically and deterministically across a single die, but is identical for all dies in a wafer
• Mainly a consequence of mask errors and systematic variations in the exposure tool
• Pelgrom’s model no longer gives an accurate predication of the relationship between variance of mismatch and device separation due to this type of variation
- Layout or density dependant variation• Introduced during etching, lithography or CMP processing steps.
04/04/2007
56
FLCCFLCC – Sensors & Control
Monte-Carlo Simulation ResultsHistograms of the circuit delay obtained from 5000 Monte-Carlo Samples
Design E(D) (nsec)
Std Dev (nsec)
Upper 95% quantile (nsec)
Percentage improvement in
quantile (%)Sp. Corr. 20.2 0.32 20.7
deterministic 20.6 0.52 21.64.2
Frequency
Delay(nsec)
Spatial correlation based design
Deterministic Design
(a)
Delay (nsec)
Modified Pelgrom’s Model based design
Deterministic Design
(b)
Design E(D) (nsec)
Std Dev (nsec)
Upper 95% quantile (nsec)
Percentage improvement in
quantile (%)
Modif. Pelg. 12.2 1.77 15.4
deterministic 23.0 3.48 29.247.3
(a) Comparison of the spatial correlation based design and the corresponding deterministic design; (b) Comparison of the modified Pelgrom’s model design and the corresponding deterministic design.
Frequency
04/04/2007
57
FLCCFLCC – Sensors & Control
Modeling IC Robustness to Process Variability• Variability of IC performance induced by
manufacturing variations emerges as a major challenge.
• While the control of the fabrication process is improving, it is also necessary to make circuits more robust against process variations.
• The robustness of circuit components that constitute critical delay paths, in particular, must be modeled and understood.
• Basic Model of Process Variability– Systematic: Across Lot, Across Wafer,
Across Field (including field position and pattern depended effects).
– Random: Lot to Lot, Wafer to Wafer, Field to Field, Line Edge Roughness Effects, Random Dopant Fluctuation.
• The proper robustness model must comprehend both random and systematic variability.
– In this work we focus on systematic Across Filed and Across Wafer variability, and we treat random variability as uncorrelated white noise.
Student: Kun Qian
Faculty: Costas J. Spanos
04/04/2007
58
FLCCFLCC – Sensors & Control
2007 Main Objective• Establish variability robustness models
– Ring oscillators– Sequences of inverter chains– 1-D and 2-D floor placement of the above
• Investigate the interaction between the process induced device parameter variations and the delay (speed) of sample circuits– Random CD variations ( White Noise )– Systematic CD variations
04/04/2007
59
FLCCFLCC – Sensors & Control
Ring Oscillator Variability ModelDx Dx
• 3-stage Ring Oscillator• 1-D placement
– Field size T– stage separation DX
– Placement is randomlychosen within [-T+Dx, T-Dx]
• Uniform random CD variation
• Quadratic CD systematic variation along x- direction
L
xT
04/04/2007
60
FLCCFLCC – Sensors & Control
Simplified Device ModelDelay for stage-I of ring oscillator circuit:
IDSAT ,i =W
L + δ Li
⋅ COX µ ⋅ VDD − VTh0 + VDDe− L +δ Li( )/l −VDSAT ,i
2⎛⎝⎜
⎞⎠⎟
VDSAT ,i ⋅ 1+ λVDD( )
≈W
L + δ Li
⋅COX µ ⋅ VDD − VTh0 + VDDe− L /l ⋅ 1 −δ Li
l⎛⎝⎜
⎞⎠⎟
−VDSAT ,i
2⎡
⎣⎢⎤
⎦⎥VDSAT ,i ⋅ 1 + λVDD( )
tP,i =COXVDD W L + δ Li( )⎡⎣ ⎤⎦
IDSAT ,i
⋅ 1 +L + δ Li+1
L + δ Li
⎛⎝⎜
⎞⎠⎟
=COXVDD W 2L + δ Li + δ Li+1( )⎡⎣ ⎤⎦
IDSAT ,i
Dx Dx
i-1 i i+1
• Variability can be obtained through closed form solutions.
• No need to run intensive Monte Carlo simulations.
04/04/2007
61
FLCCFLCC – Sensors & Control
CD White noise Variation vs. Lgate• ∆L ~ Uniform Distribution, [-0.025um, 0.025um]• Leff=L- ∆L
0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.260
10
20
Rel
ativ
e V
aria
nce
(Per
cent
) σ( τ
)/E( τ
)
L(µm)
Whitenoise
0.08 0.1 0.12 0.14 0.16 0.18 0.2 0.22 0.24 0.260
10
20
Mea
n de
lay
(ps)
Parameter VariableLeff RandomDX
Dx Dx
• No correlation between gate delay and gate separation• Larger gate length reduces the relative delay variance
σ/E(tP), but increases the average delay.
04/04/2007
62
FLCCFLCC – Sensors & Control
Deterministic Variation vs. Dx
• Larger separation generates smaller relative delay variance, since the possible position on the die is limited by Dx.
• Dx T/2 : no variation at all.
Dx Dx
0 1000 2000 3000 4000 50000
2
4
6
Rel
ativ
e V
aria
nce
(Per
cent
)σ
( τ)/E
( τ)
Dx(µm)
Deterministic Variation
0 1000 2000 3000 4000 500016
17
18
19
Mea
n de
lay
(ps)
Parameter Variable
LGate Quadratic
DX
04/04/2007
63
FLCCFLCC – Sensors & Control
Arbitrary Orientation
• 3-stage Ring Oscillator• 2-D placement• The placement of devices is not
parallel to either X or Y-axis.• Position is random within
[-T+Dx, T-Dx]*[-T+Dy, T-Dy]Dy
Dy
Dx Dx
Parameter Variable
LGate Quadratic
DX
04/04/2007
64
FLCCFLCC – Sensors & Control
Relative Variance ContourVariance changes more rapidly with orientation at small separation, while tend to be insensitive when separation is comparable to field size.
0.5 1 1.5 2 2.5 3 3.5 4 4.5
x 10-3
1
2
3
4
x 10-3
0
0.01
0.02
0.03
0.04
0.05
0.06
Dx(m)
Dy(m)
Rel
ativ
e V
aria
nce(
m)
• Define the appropriate method to integrate into the design flow• Applying the same method to more complex circuit structures.
– Evaluate different units from standard cell libraries.– Investigate the impact of different circuit topologies.
Future Goals
04/04/2007
65
FLCCFLCC – Sensors & Control
New Milestones (Year IV, 1/27/07 to 1/26/08)• Zero-footprint Metrology Wafer (continuing) (MMC Y4.1) Investigate multi-stage mechanisms
of Cu etching near end point. Simulate and experimentally verify the monitoring of lateral patterns development (e.g. wet etching high aspect-ratio contact holes in dielectrics). Enhance signal-to-noise ratio for better sensitivity with high brightness LEDs.
• Complete Aerial image Sensor Testing Phase. (continuing) (MMC Y4.2) Complete prototype and test within the UC Berkeley Microlab. Interact with key company collaborators for testing beyond the UC Berkeley Microlab. Explore fully integrated design with special purpose UV detectors.
• Incorporate LER into the process variability problem (new). (MMC Y4.3) Create statistical variability model with LER and long distance large scale variability of CDs. Assess impact on the saturation and off-regimes using device simulation for sub 42nm nodes.
• Develop engineering models for OPC Calibration (new). (MMC Y4.4) Use model based simulation, pattern matching, test-patterns, typical layouts and experiments to assess the completeness and stability of current approaches and to assess the potential improvements of novel representations, sampling, and systems optimization.
• Complete feasibility study of sensitivity model based edge updating for OPC (new). (MMC Y4.5) Calculate sensitivity model for test binary mask. Develop algorithm to calculating edge corrections globally using sensitivity model and distributed optimization. Assess iteration stability and compare with standard roster-based edge updating.
• Compare ODP and electrical metrology measurements of long-range correlation (new) (MMC 4.6) Layout identical Optical Profilometry 2D test patterns that have a variety of 2D patterns and compare the across chip and across wafer spatial correlations in electrical and scatterometry results.
• Develop suite of 2D ODP test patterns for monitoring manufacturing issues. (MMC 4.7 may appear in a supplemental proposal) Apply learning from initial measurements of Optical Digital Profilometry on test patterns with simulation of process non-idealities to design, layout, and measure a second generation of parameter sensitive and parameter specific manufacturing monitors.