SlideShare a Scribd company logo
Study of Inter and Intra Chip
Variations
Rajesh Mekala
Outline
● Motivation
● Chip Manufacturing
○ FEOL
○ BEOL
● PVT variations
○ Metal variations
○ Clocks
● Corners
● Modeling OCV
○ POCV
● Misc. variations
● Margins
● Appendix
Motivation
● Understand sources of variations
● Timing Corners
● Modeling on-chip variation
● Design Margining
● Sign-off criterion
* Embedded hyperlinks for an in-depth study
Manufacturing Stack
Manufacturing Process
● Crystal growth* → Ingot* (Silicon cylinder) → Wafers* → Chips
Manufacturing Stack
● FEOL* (Front End of Line)
○ First portion of IC fabrication
○ CMOS fabrication steps to form transistors:
■ Selecting type of wafer*
■ Cleaning of the wafer
■ Well formation
■ Gate, Drain and Source module formation
○ Manufacturing complexity increases with
FinFets
○ Base Tapeout (longer time to prepare the
masks)
■ Presents the opportunity for metal only
ECOs
Manufacturing Stack
● BEOL* (Back End of Line)
○ Second portion of IC fabrication
○ Devices get interconnected with wiring
○ PMD* module indicates beginning of BEOL
○ BEOL includes
■ Vias
■ Insulating layers (dielectrics)
■ Metal layers
■ Bonding sites (chip2package connections)
○ Double patterning (<16nm)
○ Metal Tapeout
■ Further changes are re-spins
From Sand to Silicon
FEOL variations
Transistor Manufacturing
Main Repetitive Steps:
1. Diffusion
2. Etching
3. Photolithography
4. Deposition
Types of Variations
● Parameters of transistors vary from:
○ Lot to lot (Inter process variation*)
○ Wafer to wafer (Inter process variation)
○ Die to die (Intra process variation)
○ Transistor to transistor (On Chip Variation)
● Systematic Variation
○ Possible to correct (e.g. OPC, fill)
● Random Variation
○ Stochastic* in nature
Aging with time
*RDF = Random Dopant Fluctuation
*LER = Line Edge Roughness
*NBTI = Negative Bias Temperature Instability
*HCI = Hot Carrier Injection
*TDDB = Time Dependent Dielectric Breakdown
FEOL Variations
Random variations in process parameters:
● Wgate
● Lgate
● RDF (Random Dopant Fluctuation - n+ and p+)
● Oxide thicknesses (tox)
● Diffusion Depths
These variations affect performance and power of the design
Process convention
Industry uses two-letters to describe corners:
● 1st letter → NMOS device
● 2nd letter → PMOS device
● 5 classic corners:
○ FF (fast fast)
○ SF (slow fast)
○ SS (slow slow)
○ FS (fast slow)
○ TT (typical typical)
● Process typically tuned to TT
● FF corner → skew both P and N devices to the fast
corner (Process tuning)
● Ring Oscillators to determine the process
BEOL variations
BEOL variations
● Chip area increasing (greater integration)
● Device dimensions scaled down
● Scaled wires are
○ Longer (chip area scaling)
○ Thinner (minimum dimension scaling)
○ Taller (Do not increase resistance)
● Wire delay* limiting performance
● Double patterning
● Via resistance variation
Processes affecting the FEOL and BEOL are
completely orthogonal (uncorrelated parameters)
BEOL variations
● Line Edge Roughness
○ Random variation* in:
■ Lithography
■ Photoresist
■ Etching
● Dishing and Erosion* during CMP
Resistance and Capacitance changes with
BEOL variations → Performance degradation
1. Time independent variations
2. AC-EM cause a shift in performance
over time
Reduce BEOL variations
● Manufacturability:
○ OPC
● Design
○ Metal fill
○ Shielding
○ NDR
Reduce BEOL variations
● Design:
○ Wire length histogram
○ Random Variation:
■ Timing derate
● Remove pessimism of derate by 2-sigma corners
● Modeling:
○ Extraction tool accuracy
○ Higher order Interconnect delay calculation
Corner Lots
Process corner lots
● Definition: Groups of wafers whose process
parameters are carefully tuned to known process
extremes
○ Characterize them at various temperatures
and voltages
● process lots = corner lots = split lots = skewed lots
● If characterization is within accepted limits →
mass manufacturing of chips begins
● Corner lots should be produced after tapeout and
before releasing to production
○ Process Qualification during chip
qualification
Voltage Variations
Voltage Variations
● Voltage Source is non-ideal
● DC static voltage drop
● AC transient variations
● Transistor delays are sensitive to swings in VDD and VSS
Voltage Variations
● VDD → IDS (drive) → tPHL → F
● Delay curve is nonlinear
● SVT devices → more sensitive to voltage change
● IP teams - choose operational voltage and clock
frequencies
○ Multi-mode operation (high perf, low power)
Mitigate Voltage Variations
Voltage
Time
Vsu = Non-ideal power supply voltage
Current Profile
Ptr = Voltage at gate
Leakage current (min/max)
● STA at minimum voltage (capture worst
case)
● Validate library models* for the corners
● Static IR drop < 10% VDD (characterized
library)
○ 10% worst case IR drop from the
power supply to the PG pins of
transistor (can be budgeted)
○ Sense it on-chip to feedback to VR
● DCAP for dynamic IR drop
● Capacitance with low resistive/inductive
paths to the gates
Temperature Variations
Temperature Variations
Cause:
● Work load (application) → Switching activity →
Power dissipation → Heat → Temperature
increase
Mitigation:
● Worst case analysis:
○ Gate dominant ?
○ Wire dominant ?
○ Depending on the technology node
○ Lower VDD, Lower T (Setup)
■ Gate dominant, temperature
inversion
○ Higher VDD, Higher Temp (Hold)
○ Mixed corners to cover skewed affects
● Low Power Design
● Cooling solutions
Summary
Categorization
Variation
Static Dynamic
On-Chip
D2D, W2W
Random Correlated
RDF Leff
Weff Mobility
RSD Rwire
Cwire
Litho CMP
Random Systematic
Misc.
Process
Litho Effects
Systematic
Layout
Effects
Stress
Process
Device
Process
Tem
p
Vdd Jitter SI MIS
Circuit
BTI SER
HCI TDDB
Reliability
History of Focus Areas
● Key physical design focus area with shrinking technology nodes
Clocks
Clock Skew and Jitter
Skew = ‘static’ time difference between any 2 electrical
nodes
Sources: placement distribution (local skew)
Jitter = ‘dynamic’ time difference of a signal with respect to
an ideal signal
Sources: PLL source, clock network jitter, SI
Clock overhead* = setup time, hold time requirements of
sequential cells
L1
L2
Sources of Clock Variation
● Skew:
○ Clock buffers drive strength and channel lengths
○ Static voltage drop
○ Local Hot spots
○ Process variation (Leff)
○ Routes RC mismatch
○ BEOL variations
● Jitter
○ Switching activity → Power supply variations
with time
○ PLL source Jitter
○ Temperature Gradient
○ Clock network jitter (dynamic voltage drop)
○ Wire coupling capacitance
■ Data change in different cycles
Strategies: Reduce Clock Tree OCV
● Minimize clock* divergence
● Choose PLL with least clock source jitter
● Balance cell and net delays
● Inverters - minimize DCD
● Minimize clock tree latency
● LVT cells, same channel length
○ Minimize cross corner delay variation
● NDR, shielding
○ Minimize coupling (spatial + temporal)
● Specific metal layers to minimize BEOL process variations
● Clock tree optimized std_cells in clock path
● Robust PG network
● Global clock tree buffers with embedded DCAP
Timing Corners
Timing Corners
● Corner Definition:
○ A point in the PVT/RC space where cell/net delays have extreme values (all cell delays are the max or
min and all net delays are the max or min independently)
● Timing Sign-off (STA) focuses on Corners [Worst Case Analysis]
Design Space
Process
Voltage
Temperatur
e
Timing Corners
*(Generally Ignored) Vias {max/min/typ cap/resistance}
● All Combinations: 3 * 3 * 5 * 5 = 225
● Chosen Corners: ~50
Voltage
Min
Typ
Max
Tem
p
Min
Typ
Max
FEOL
SS
TT
FF
SF
FS
BEOL
Cw
Cb
RCw
RCb
RCtyp
Corner explosion
Modes of operation:
● func
● test
○ shift
○ capture
*Create additional
corners
Multi Voltage Domains:
● Vmem
● Vlogic
*Create additional
corners
Aging Degradation:
● NBTI
● HCI
● TDDB
*Derated libraries like
EOL and BOL (or)
sign-off derate
Analysis Overhead:
● Temperature Inversion
● DPT corners
● Via variations
● Over drive mode
● Confidence bounds (3-
sigma/2-sigma)
*Create additional corners
Corner Specifications (Example)
Mode Voltage FEOL
process
BEOL
process
Temperature
(Junction)
Variation
Model
Purpose*
func 0.9 * VDDlogic
0.9 * VDDmem
ss cw_ccw -30c aocv Timing Specification (conservative)
func VDDlogic
VDDmem
tt ctyp 25c aocv Timing Specification (Typical/Binning)
func 1.1 * VDDlogic
1.1 * VDDmem
ff cb_ccb 125c aocv Race conditions, Hold timing, Noise
func 0.9 * VDDlogic
0.9 * VDDmem
ss cb_ccb 125c aocv Race of gates against wires
func 1.1 * VDDlogic
1.1 * VDDmem
ff cw_ccw -30c aocv Race of wires against gates
Modeling Variations
Modeling variations
Global OCV (flat-derate):
- Easier with a lot of pessimism
SSTA:
- SSTA libs, STA tool support, large compute time
Advanced OCV:
- Reduce pessimism from GOCV taking
advantage of path depth and spatial correlation
Cell Based AOCV:
- Std_cell specific AOCV tables
Parametric OCV:
- LVF characterization of standard cells
- Representative of SSTA
Parametric OCV
● POCV timing analysis*:
Instead of timing windows (minimum and maximum delays) for each timing arc, calculate the delay as
a function of a Gaussian normal distribution P
delay = nominal_delay + σ * P
nominal_delay, σ are part of library
● POCV characterization is embedded into standard cell library
○ Slew and load dependant variation
● Close to Statistical STA
Modeling and analysis errors
● Error in gate delay models (model-to-hardware)
○ Library to Silicon
● std_cell characterization test bench accuracy
○ SI optimism/pessimism
● Delay calculation errors
○ Delay estimation interpolation and extrapolation
○ max/min trans/cap within lib characterization bounds
● Nonlinear delay scaling across PVT corners
● Extraction tool accuracy
○ Lumped v.s. Distributed RC models
○ 3D modeling
Industry Margining Process
Foundry and EDA tools/vendors
● Extremely important to
monitor different margins
using in the design cycle
● STA sign-off involves
interaction with many teams
for coming to consensus
What is golden/sign-off ?
● Abstraction reduces accuracy
○ Floorplanning doesn’t need sign-off timing
● Models (.libs, etc.) → SPICE → Silicon* (Hardware)
● Margins feedback from synthesis to bring-up and process data
● Test chips to study lithography issues, transistor model correlation, statistical properties of
the process
● Robust clock tree design (Cell selection, NDRs, shielding, Mesh)
● Exhaustive checks across timing corners
● Robust IR drop analysis
OCV margins
Phase of the design Setup Hold max_trans/max_
cap
Synthesis flat uncertainty number to account for:
1. Clock: skew + jitter + OCV + SI
2. Data: OCV, SI
3. Tools miscorrelation (including delay
calculation, analysis differences)
4. Virtual to Actual route differences
5. Model mismatches (CCS, NDM etc.)
6. Blanket setup margin (Other margins
and RTL feedback)
7. Top to Block miscorrelation
8. Library guardband
flat margin (no optimization in synthesis) flat number
Implementation Placement: all the margins in synthesis
CTS: all except clock skew, clock SI
Route: all except clock skew, clock and data SI,
miscorrelation
flat uncertainty number accounting for:
1. Clock: uncertainty
2. Top to Block miscorrelation for
IPs
3. Route, model differences, extra
guardband
Margin w.r.t. each
scenario and clock
frequency
Sign-off Sign-off timing tool (PBA) with accurate margins
(no pessimism) - ignoring guardband margins.
(Dependant on clock frequency, VT-type, corner)
Sign-off timing tool with accurate
margins ignoring guardband margins
(Dependant on clock frequency, VT-
type, corner)
Sign-off criterion w.r.t.
Each scenario and clock
frequency
Appendix
Reading material
Good reads for in-depth analysis:
● Variability Modeling and Statistical Parameter Extraction for CMOS Devices (**)
● History of Timing Closure (*)
● Layout variation effects (*)
● Corner-based Timing Signoff and What Is Next (***)
● Clocking for high-performance circuits (**)
● Methodology to close timing with hundreds of MCMM scenarios (***)
● How to deal with large process variations (*)
● Parameter variation tolerance (*)
● Multi Input Switching (**)
● CMOS VLSI Design (**)
● Process Technology Variation (***)
Sources of Variations
Random distribution of identically drawn devices is
caused by variations in process parameters such as:
● Impurity concentration densities
● Oxide thicknesses
● Diffusion Depths
● Many other physical aspects of a transistor
All of these variations affect cell delay, threshold current
=> impacts performance and power
Changes in these parameters cause electrical
parameters to vary such as threshold voltage, Weff,
Leff, etc.
Parameters affecting the FEOL and BEOL are
completely orthogonal (uncorrelated)
Variations in the dimensions of the devices: Limited
resolution of the photolithographic process which
causes W/L variations in MOS transistors.
BEOL variations
What can we model deterministically?
● Optical and etching environment
● CMP etching as function of location on a wafer
● Feature size as a function of position in optical field
BEOL variations
● With double patterning in metal layer masks
beyond 20nm technologies, interconnect
modeling corners increased
Study of inter and intra chip variations
The saturation current of a cell depends on the power supply. The
delay of a cell is dependent on the saturation current. In this way,
the power supply inflects the propagation delay of a cell.
Throughout a chip, the power supply is not constant and hence the
propagation delay varies in a chip. The voltage drop is due to
nonzero resistance in the supply wires. A higher voltage makes a
cell faster and hence the propagation delay is reduced. The
decrease is exponential for a wide voltage range.
● Best case: fast process,
highest voltage and lowest
temperature
● Worst case: slow process,
lowest voltage and highest
temperature
Summary of Variations
● Manufacturing variations
○ Leff: random and systematic components
○ Metal thickness variations due to chemical-mechanical
polishing (CMP)
○ Random dopant effects which cause random Vt variations
○ Line-edge roughness effects
● Voltage variations
○ IR drop across surface of chip, in different voltage islands
○ IR drop from cycle to cycle
● Temperature variations
○ Across surface of chip
○ From cycle to cycle
● Electrical reasons
○ Coupling noise which may or may not occur in any given
cycle
○ Pre-charging of internal nodes
○ History effect in SOI circuits
○ Simultaneous switching
● Aging at different rates
○ Hot electrons, NBTI/PBTI, electromigration
Interpolated voltage variation model
Solving BEOL variations
Litho Etching CMP Location
in field
Location
on wafer
Fab Random
variations
Compensat
e later
OPC Metal fill
Reduce
uncertainty
Shielding Metal fill
Reduce
impact
OPC Clock
Mesh
Statistical
analysis
Worst case
analysis
Margins at different stages
PHASE TOOL SETUP Comment HOLD Comment MAX_TRANS
synthesis DC ? blanket setup margin for all blocks, 1 corner (primary setup) ? ? 100pS
DCT ? blanket setup margin for all blocks, 1 corner (primary setup) ? ? 100pS
clock_tree_margin skew + noise + Jsrc + Jmn + Jcts
placement Innovus ? blanket setup margin for all blocks, 1 corner (primary setup) ? ? ?
CTS Innovus 0% ? ?
post-CTS Innovus Jsrc + Jmn + Jcts + (skew = 0) + (noise = 0)
route Innovus 0% ? ?
post-route Innovus
signoff PT 0% ? ?
minSA ? ? global speed up
POCV On Chip Variations
MIS for hold POCV On Chip Variations

More Related Content

PDF
Vlsi best notes google docs
DOCX
Vlsi interview questions compilation
PPTX
Flip Chip technology
PDF
Clock Tree Timing 101
PPT
Asic backend design
PPTX
Library Characterization Flow
PDF
VLSI Physical Design Physical Design Concepts
PPT
Low Power Design Techniques for ASIC / SOC Design
Vlsi best notes google docs
Vlsi interview questions compilation
Flip Chip technology
Clock Tree Timing 101
Asic backend design
Library Characterization Flow
VLSI Physical Design Physical Design Concepts
Low Power Design Techniques for ASIC / SOC Design

What's hot (20)

PPTX
PPTX
Floor plan & Power Plan
PDF
VLSI-Physical Design- Tool Terminalogy
PDF
Physical design-complete
PPTX
Multi mode multi corner (mmmc)
PPTX
Low power in vlsi with upf basics part 2
PPTX
Low power in vlsi with upf basics part 1
PDF
Physical Verification.pdf
PDF
Sta by usha_mehta
PPT
Design challenges in physical design
PDF
Implementing Useful Clock Skew Using Skew Groups
PDF
Digital VLSI Design : Combinational Circuit
PPTX
Powerplanning
DOCX
Pd flow i
PDF
Digital VLSI Design : Introduction
PPTX
ASIC DESIGN : PLACEMENT
PPTX
Define Width and Height of Core and Die (http://www.vlsisystemdesign.com/PD-F...
PPTX
ASIC DESIGN FLOW
PPT
Timing Analysis
PDF
Understanding cts log_messages
Floor plan & Power Plan
VLSI-Physical Design- Tool Terminalogy
Physical design-complete
Multi mode multi corner (mmmc)
Low power in vlsi with upf basics part 2
Low power in vlsi with upf basics part 1
Physical Verification.pdf
Sta by usha_mehta
Design challenges in physical design
Implementing Useful Clock Skew Using Skew Groups
Digital VLSI Design : Combinational Circuit
Powerplanning
Pd flow i
Digital VLSI Design : Introduction
ASIC DESIGN : PLACEMENT
Define Width and Height of Core and Die (http://www.vlsisystemdesign.com/PD-F...
ASIC DESIGN FLOW
Timing Analysis
Understanding cts log_messages
Ad

Similar to Study of inter and intra chip variations (20)

PDF
VLSI Static Timing Analysis Timing Checks Part 5 - On Chip Variation
PDF
[Back2School] STA Methodology- Chapter 7pdf
PPTX
Physical design
PDF
P358387
PDF
HC24.29.625-IA-23-Wide-Ruhl-Intel_2012_NTV_iA
PPTX
Complete ASIC design flow - VLSI UNIVERSE
PPTX
PDF
1. Introduction to PnR.pdf
PPT
Statistical Framework for Technology-Model-Product Co-Design and Convergence
DOC
Saturation Throughput and Delay Aware Asynchronous Noc Using Fuzzy Logic
DOC
Saturation Throughput and Delay Aware Asynchronous Noc Using Fuzzy Logic
PDF
VLSI testing and analysis
PDF
Bt31482484
PDF
PosterFormatRNYF(1)
PDF
Design for Test [DFT]-1 (1).pdf DESIGN DFT
PPTX
EEL71090_Lecture12_13_BKumar_IIITJodh.pptx
DOCX
Full IC Flow.docx
PDF
EMT529-VLSI-Design-wk1.pdf
PDF
Implementation of Low Power Test Pattern Generator Using LFSR
PPTX
IO Circuit_1.pptx
VLSI Static Timing Analysis Timing Checks Part 5 - On Chip Variation
[Back2School] STA Methodology- Chapter 7pdf
Physical design
P358387
HC24.29.625-IA-23-Wide-Ruhl-Intel_2012_NTV_iA
Complete ASIC design flow - VLSI UNIVERSE
1. Introduction to PnR.pdf
Statistical Framework for Technology-Model-Product Co-Design and Convergence
Saturation Throughput and Delay Aware Asynchronous Noc Using Fuzzy Logic
Saturation Throughput and Delay Aware Asynchronous Noc Using Fuzzy Logic
VLSI testing and analysis
Bt31482484
PosterFormatRNYF(1)
Design for Test [DFT]-1 (1).pdf DESIGN DFT
EEL71090_Lecture12_13_BKumar_IIITJodh.pptx
Full IC Flow.docx
EMT529-VLSI-Design-wk1.pdf
Implementation of Low Power Test Pattern Generator Using LFSR
IO Circuit_1.pptx
Ad

More from Rajesh M (11)

PDF
Daily Habits.pdf
PDF
Clock relationships
PDF
Node Scaling Objectives
PDF
Technology scaling introduction
PDF
Problems between Synthesis and preCTS
PDF
Setup fixing
PPTX
#50 ethics
DOCX
Power Reduction Techniques
PDF
680report final
PPTX
676.v3
PDF
Clock mesh sizing slides
Daily Habits.pdf
Clock relationships
Node Scaling Objectives
Technology scaling introduction
Problems between Synthesis and preCTS
Setup fixing
#50 ethics
Power Reduction Techniques
680report final
676.v3
Clock mesh sizing slides

Recently uploaded (20)

PDF
Reach Out and Touch Someone: Haptics and Empathic Computing
PDF
Peak of Data & AI Encore- AI for Metadata and Smarter Workflows
PDF
Architecting across the Boundaries of two Complex Domains - Healthcare & Tech...
PDF
Dropbox Q2 2025 Financial Results & Investor Presentation
PPTX
Digital-Transformation-Roadmap-for-Companies.pptx
PDF
cuic standard and advanced reporting.pdf
PPTX
Cloud computing and distributed systems.
PDF
Advanced methodologies resolving dimensionality complications for autism neur...
PPT
Teaching material agriculture food technology
PPTX
MYSQL Presentation for SQL database connectivity
PPTX
Programs and apps: productivity, graphics, security and other tools
PDF
7 ChatGPT Prompts to Help You Define Your Ideal Customer Profile.pdf
PPTX
Effective Security Operations Center (SOC) A Modern, Strategic, and Threat-In...
PDF
Review of recent advances in non-invasive hemoglobin estimation
PDF
Agricultural_Statistics_at_a_Glance_2022_0.pdf
PDF
NewMind AI Weekly Chronicles - August'25 Week I
PDF
MIND Revenue Release Quarter 2 2025 Press Release
PDF
Machine learning based COVID-19 study performance prediction
PDF
Electronic commerce courselecture one. Pdf
PPTX
ACSFv1EN-58255 AWS Academy Cloud Security Foundations.pptx
Reach Out and Touch Someone: Haptics and Empathic Computing
Peak of Data & AI Encore- AI for Metadata and Smarter Workflows
Architecting across the Boundaries of two Complex Domains - Healthcare & Tech...
Dropbox Q2 2025 Financial Results & Investor Presentation
Digital-Transformation-Roadmap-for-Companies.pptx
cuic standard and advanced reporting.pdf
Cloud computing and distributed systems.
Advanced methodologies resolving dimensionality complications for autism neur...
Teaching material agriculture food technology
MYSQL Presentation for SQL database connectivity
Programs and apps: productivity, graphics, security and other tools
7 ChatGPT Prompts to Help You Define Your Ideal Customer Profile.pdf
Effective Security Operations Center (SOC) A Modern, Strategic, and Threat-In...
Review of recent advances in non-invasive hemoglobin estimation
Agricultural_Statistics_at_a_Glance_2022_0.pdf
NewMind AI Weekly Chronicles - August'25 Week I
MIND Revenue Release Quarter 2 2025 Press Release
Machine learning based COVID-19 study performance prediction
Electronic commerce courselecture one. Pdf
ACSFv1EN-58255 AWS Academy Cloud Security Foundations.pptx

Study of inter and intra chip variations

  • 1. Study of Inter and Intra Chip Variations Rajesh Mekala
  • 2. Outline ● Motivation ● Chip Manufacturing ○ FEOL ○ BEOL ● PVT variations ○ Metal variations ○ Clocks ● Corners ● Modeling OCV ○ POCV ● Misc. variations ● Margins ● Appendix
  • 3. Motivation ● Understand sources of variations ● Timing Corners ● Modeling on-chip variation ● Design Margining ● Sign-off criterion * Embedded hyperlinks for an in-depth study
  • 5. Manufacturing Process ● Crystal growth* → Ingot* (Silicon cylinder) → Wafers* → Chips
  • 6. Manufacturing Stack ● FEOL* (Front End of Line) ○ First portion of IC fabrication ○ CMOS fabrication steps to form transistors: ■ Selecting type of wafer* ■ Cleaning of the wafer ■ Well formation ■ Gate, Drain and Source module formation ○ Manufacturing complexity increases with FinFets ○ Base Tapeout (longer time to prepare the masks) ■ Presents the opportunity for metal only ECOs
  • 7. Manufacturing Stack ● BEOL* (Back End of Line) ○ Second portion of IC fabrication ○ Devices get interconnected with wiring ○ PMD* module indicates beginning of BEOL ○ BEOL includes ■ Vias ■ Insulating layers (dielectrics) ■ Metal layers ■ Bonding sites (chip2package connections) ○ Double patterning (<16nm) ○ Metal Tapeout ■ Further changes are re-spins
  • 8. From Sand to Silicon
  • 10. Transistor Manufacturing Main Repetitive Steps: 1. Diffusion 2. Etching 3. Photolithography 4. Deposition
  • 11. Types of Variations ● Parameters of transistors vary from: ○ Lot to lot (Inter process variation*) ○ Wafer to wafer (Inter process variation) ○ Die to die (Intra process variation) ○ Transistor to transistor (On Chip Variation) ● Systematic Variation ○ Possible to correct (e.g. OPC, fill) ● Random Variation ○ Stochastic* in nature Aging with time *RDF = Random Dopant Fluctuation *LER = Line Edge Roughness *NBTI = Negative Bias Temperature Instability *HCI = Hot Carrier Injection *TDDB = Time Dependent Dielectric Breakdown
  • 12. FEOL Variations Random variations in process parameters: ● Wgate ● Lgate ● RDF (Random Dopant Fluctuation - n+ and p+) ● Oxide thicknesses (tox) ● Diffusion Depths These variations affect performance and power of the design
  • 13. Process convention Industry uses two-letters to describe corners: ● 1st letter → NMOS device ● 2nd letter → PMOS device ● 5 classic corners: ○ FF (fast fast) ○ SF (slow fast) ○ SS (slow slow) ○ FS (fast slow) ○ TT (typical typical) ● Process typically tuned to TT ● FF corner → skew both P and N devices to the fast corner (Process tuning) ● Ring Oscillators to determine the process
  • 15. BEOL variations ● Chip area increasing (greater integration) ● Device dimensions scaled down ● Scaled wires are ○ Longer (chip area scaling) ○ Thinner (minimum dimension scaling) ○ Taller (Do not increase resistance) ● Wire delay* limiting performance ● Double patterning ● Via resistance variation Processes affecting the FEOL and BEOL are completely orthogonal (uncorrelated parameters)
  • 16. BEOL variations ● Line Edge Roughness ○ Random variation* in: ■ Lithography ■ Photoresist ■ Etching ● Dishing and Erosion* during CMP Resistance and Capacitance changes with BEOL variations → Performance degradation 1. Time independent variations 2. AC-EM cause a shift in performance over time
  • 17. Reduce BEOL variations ● Manufacturability: ○ OPC ● Design ○ Metal fill ○ Shielding ○ NDR
  • 18. Reduce BEOL variations ● Design: ○ Wire length histogram ○ Random Variation: ■ Timing derate ● Remove pessimism of derate by 2-sigma corners ● Modeling: ○ Extraction tool accuracy ○ Higher order Interconnect delay calculation
  • 20. Process corner lots ● Definition: Groups of wafers whose process parameters are carefully tuned to known process extremes ○ Characterize them at various temperatures and voltages ● process lots = corner lots = split lots = skewed lots ● If characterization is within accepted limits → mass manufacturing of chips begins ● Corner lots should be produced after tapeout and before releasing to production ○ Process Qualification during chip qualification
  • 22. Voltage Variations ● Voltage Source is non-ideal ● DC static voltage drop ● AC transient variations ● Transistor delays are sensitive to swings in VDD and VSS
  • 23. Voltage Variations ● VDD → IDS (drive) → tPHL → F ● Delay curve is nonlinear ● SVT devices → more sensitive to voltage change ● IP teams - choose operational voltage and clock frequencies ○ Multi-mode operation (high perf, low power)
  • 24. Mitigate Voltage Variations Voltage Time Vsu = Non-ideal power supply voltage Current Profile Ptr = Voltage at gate Leakage current (min/max) ● STA at minimum voltage (capture worst case) ● Validate library models* for the corners ● Static IR drop < 10% VDD (characterized library) ○ 10% worst case IR drop from the power supply to the PG pins of transistor (can be budgeted) ○ Sense it on-chip to feedback to VR ● DCAP for dynamic IR drop ● Capacitance with low resistive/inductive paths to the gates
  • 26. Temperature Variations Cause: ● Work load (application) → Switching activity → Power dissipation → Heat → Temperature increase Mitigation: ● Worst case analysis: ○ Gate dominant ? ○ Wire dominant ? ○ Depending on the technology node ○ Lower VDD, Lower T (Setup) ■ Gate dominant, temperature inversion ○ Higher VDD, Higher Temp (Hold) ○ Mixed corners to cover skewed affects ● Low Power Design ● Cooling solutions
  • 28. Categorization Variation Static Dynamic On-Chip D2D, W2W Random Correlated RDF Leff Weff Mobility RSD Rwire Cwire Litho CMP Random Systematic Misc. Process Litho Effects Systematic Layout Effects Stress Process Device Process Tem p Vdd Jitter SI MIS Circuit BTI SER HCI TDDB Reliability
  • 29. History of Focus Areas ● Key physical design focus area with shrinking technology nodes
  • 31. Clock Skew and Jitter Skew = ‘static’ time difference between any 2 electrical nodes Sources: placement distribution (local skew) Jitter = ‘dynamic’ time difference of a signal with respect to an ideal signal Sources: PLL source, clock network jitter, SI Clock overhead* = setup time, hold time requirements of sequential cells L1 L2
  • 32. Sources of Clock Variation ● Skew: ○ Clock buffers drive strength and channel lengths ○ Static voltage drop ○ Local Hot spots ○ Process variation (Leff) ○ Routes RC mismatch ○ BEOL variations ● Jitter ○ Switching activity → Power supply variations with time ○ PLL source Jitter ○ Temperature Gradient ○ Clock network jitter (dynamic voltage drop) ○ Wire coupling capacitance ■ Data change in different cycles
  • 33. Strategies: Reduce Clock Tree OCV ● Minimize clock* divergence ● Choose PLL with least clock source jitter ● Balance cell and net delays ● Inverters - minimize DCD ● Minimize clock tree latency ● LVT cells, same channel length ○ Minimize cross corner delay variation ● NDR, shielding ○ Minimize coupling (spatial + temporal) ● Specific metal layers to minimize BEOL process variations ● Clock tree optimized std_cells in clock path ● Robust PG network ● Global clock tree buffers with embedded DCAP
  • 35. Timing Corners ● Corner Definition: ○ A point in the PVT/RC space where cell/net delays have extreme values (all cell delays are the max or min and all net delays are the max or min independently) ● Timing Sign-off (STA) focuses on Corners [Worst Case Analysis] Design Space Process Voltage Temperatur e
  • 36. Timing Corners *(Generally Ignored) Vias {max/min/typ cap/resistance} ● All Combinations: 3 * 3 * 5 * 5 = 225 ● Chosen Corners: ~50 Voltage Min Typ Max Tem p Min Typ Max FEOL SS TT FF SF FS BEOL Cw Cb RCw RCb RCtyp
  • 37. Corner explosion Modes of operation: ● func ● test ○ shift ○ capture *Create additional corners Multi Voltage Domains: ● Vmem ● Vlogic *Create additional corners Aging Degradation: ● NBTI ● HCI ● TDDB *Derated libraries like EOL and BOL (or) sign-off derate Analysis Overhead: ● Temperature Inversion ● DPT corners ● Via variations ● Over drive mode ● Confidence bounds (3- sigma/2-sigma) *Create additional corners
  • 38. Corner Specifications (Example) Mode Voltage FEOL process BEOL process Temperature (Junction) Variation Model Purpose* func 0.9 * VDDlogic 0.9 * VDDmem ss cw_ccw -30c aocv Timing Specification (conservative) func VDDlogic VDDmem tt ctyp 25c aocv Timing Specification (Typical/Binning) func 1.1 * VDDlogic 1.1 * VDDmem ff cb_ccb 125c aocv Race conditions, Hold timing, Noise func 0.9 * VDDlogic 0.9 * VDDmem ss cb_ccb 125c aocv Race of gates against wires func 1.1 * VDDlogic 1.1 * VDDmem ff cw_ccw -30c aocv Race of wires against gates
  • 40. Modeling variations Global OCV (flat-derate): - Easier with a lot of pessimism SSTA: - SSTA libs, STA tool support, large compute time Advanced OCV: - Reduce pessimism from GOCV taking advantage of path depth and spatial correlation Cell Based AOCV: - Std_cell specific AOCV tables Parametric OCV: - LVF characterization of standard cells - Representative of SSTA
  • 41. Parametric OCV ● POCV timing analysis*: Instead of timing windows (minimum and maximum delays) for each timing arc, calculate the delay as a function of a Gaussian normal distribution P delay = nominal_delay + σ * P nominal_delay, σ are part of library ● POCV characterization is embedded into standard cell library ○ Slew and load dependant variation ● Close to Statistical STA
  • 42. Modeling and analysis errors ● Error in gate delay models (model-to-hardware) ○ Library to Silicon ● std_cell characterization test bench accuracy ○ SI optimism/pessimism ● Delay calculation errors ○ Delay estimation interpolation and extrapolation ○ max/min trans/cap within lib characterization bounds ● Nonlinear delay scaling across PVT corners ● Extraction tool accuracy ○ Lumped v.s. Distributed RC models ○ 3D modeling
  • 43. Industry Margining Process Foundry and EDA tools/vendors ● Extremely important to monitor different margins using in the design cycle ● STA sign-off involves interaction with many teams for coming to consensus
  • 44. What is golden/sign-off ? ● Abstraction reduces accuracy ○ Floorplanning doesn’t need sign-off timing ● Models (.libs, etc.) → SPICE → Silicon* (Hardware) ● Margins feedback from synthesis to bring-up and process data ● Test chips to study lithography issues, transistor model correlation, statistical properties of the process ● Robust clock tree design (Cell selection, NDRs, shielding, Mesh) ● Exhaustive checks across timing corners ● Robust IR drop analysis
  • 45. OCV margins Phase of the design Setup Hold max_trans/max_ cap Synthesis flat uncertainty number to account for: 1. Clock: skew + jitter + OCV + SI 2. Data: OCV, SI 3. Tools miscorrelation (including delay calculation, analysis differences) 4. Virtual to Actual route differences 5. Model mismatches (CCS, NDM etc.) 6. Blanket setup margin (Other margins and RTL feedback) 7. Top to Block miscorrelation 8. Library guardband flat margin (no optimization in synthesis) flat number Implementation Placement: all the margins in synthesis CTS: all except clock skew, clock SI Route: all except clock skew, clock and data SI, miscorrelation flat uncertainty number accounting for: 1. Clock: uncertainty 2. Top to Block miscorrelation for IPs 3. Route, model differences, extra guardband Margin w.r.t. each scenario and clock frequency Sign-off Sign-off timing tool (PBA) with accurate margins (no pessimism) - ignoring guardband margins. (Dependant on clock frequency, VT-type, corner) Sign-off timing tool with accurate margins ignoring guardband margins (Dependant on clock frequency, VT- type, corner) Sign-off criterion w.r.t. Each scenario and clock frequency
  • 47. Reading material Good reads for in-depth analysis: ● Variability Modeling and Statistical Parameter Extraction for CMOS Devices (**) ● History of Timing Closure (*) ● Layout variation effects (*) ● Corner-based Timing Signoff and What Is Next (***) ● Clocking for high-performance circuits (**) ● Methodology to close timing with hundreds of MCMM scenarios (***) ● How to deal with large process variations (*) ● Parameter variation tolerance (*) ● Multi Input Switching (**) ● CMOS VLSI Design (**) ● Process Technology Variation (***)
  • 48. Sources of Variations Random distribution of identically drawn devices is caused by variations in process parameters such as: ● Impurity concentration densities ● Oxide thicknesses ● Diffusion Depths ● Many other physical aspects of a transistor All of these variations affect cell delay, threshold current => impacts performance and power Changes in these parameters cause electrical parameters to vary such as threshold voltage, Weff, Leff, etc. Parameters affecting the FEOL and BEOL are completely orthogonal (uncorrelated) Variations in the dimensions of the devices: Limited resolution of the photolithographic process which causes W/L variations in MOS transistors.
  • 49. BEOL variations What can we model deterministically? ● Optical and etching environment ● CMP etching as function of location on a wafer ● Feature size as a function of position in optical field
  • 50. BEOL variations ● With double patterning in metal layer masks beyond 20nm technologies, interconnect modeling corners increased
  • 52. The saturation current of a cell depends on the power supply. The delay of a cell is dependent on the saturation current. In this way, the power supply inflects the propagation delay of a cell. Throughout a chip, the power supply is not constant and hence the propagation delay varies in a chip. The voltage drop is due to nonzero resistance in the supply wires. A higher voltage makes a cell faster and hence the propagation delay is reduced. The decrease is exponential for a wide voltage range. ● Best case: fast process, highest voltage and lowest temperature ● Worst case: slow process, lowest voltage and highest temperature
  • 53. Summary of Variations ● Manufacturing variations ○ Leff: random and systematic components ○ Metal thickness variations due to chemical-mechanical polishing (CMP) ○ Random dopant effects which cause random Vt variations ○ Line-edge roughness effects ● Voltage variations ○ IR drop across surface of chip, in different voltage islands ○ IR drop from cycle to cycle ● Temperature variations ○ Across surface of chip ○ From cycle to cycle ● Electrical reasons ○ Coupling noise which may or may not occur in any given cycle ○ Pre-charging of internal nodes ○ History effect in SOI circuits ○ Simultaneous switching ● Aging at different rates ○ Hot electrons, NBTI/PBTI, electromigration
  • 55. Solving BEOL variations Litho Etching CMP Location in field Location on wafer Fab Random variations Compensat e later OPC Metal fill Reduce uncertainty Shielding Metal fill Reduce impact OPC Clock Mesh Statistical analysis Worst case analysis
  • 56. Margins at different stages PHASE TOOL SETUP Comment HOLD Comment MAX_TRANS synthesis DC ? blanket setup margin for all blocks, 1 corner (primary setup) ? ? 100pS DCT ? blanket setup margin for all blocks, 1 corner (primary setup) ? ? 100pS clock_tree_margin skew + noise + Jsrc + Jmn + Jcts placement Innovus ? blanket setup margin for all blocks, 1 corner (primary setup) ? ? ? CTS Innovus 0% ? ? post-CTS Innovus Jsrc + Jmn + Jcts + (skew = 0) + (noise = 0) route Innovus 0% ? ? post-route Innovus signoff PT 0% ? ? minSA ? ? global speed up POCV On Chip Variations MIS for hold POCV On Chip Variations

Editor's Notes

  • #4: This presentation covers a lot of breadth, but the hyperlinks are provided to learn more about particular topics
  • #12: nMOS and pMOS variation