A Complete Reliability Solution: Reliability Modeling, Applications, and Integration ... · 2021....

Confidential

A Complete Reliability Solution: Reliability Modeling, Applications, and Integration in Analog Design Environment Tianlei Guo, Jushan Xie Cadence Design Systems, Inc. June 19, 2018

2 © 2017 Cadence Design Systems, Inc. Cadence confidential. Internal use only.

Project Vision

•  Make reliability analysis relevant by providing –  Predictive aging models –  Aging analysis accelerated by temperature, process variation, … –  Support mission profiles

Design for Reliability

Failure is NOT an Option


Simulating Device Aging


Terminology for Device Reliability Analysis

•  Age –  Device age is a parameter that represents the device degradation physical

phenomena •  Age (or lifetime or degradation) model

–  Predicts the degradation in device characteristics due to a physical phenomena, such as HCI, NBTI, PBTI, TDDB, …

•  Aged (or degraded) model –  is a device SPICE model that represents the effects of all kind of

degradations at particular future time value •  Fresh Simulation

–  Is a simulation without degradation or a simulation at time = 0 •  Stress Simulation

–  Is a simulation that represents the stress condition •  Aging (or EOL) simulation

–  is a simulation with worst case device degradation due to electrical stress, a simulation at the end of life


HCI (Hot Carrier Injection)

Ben Kaczer, 2016 IEDM tutorial


BTI (Bias Temperature Instability)

NBTI: NMOS BTI PBTI: PMOS BTI strong


TDDB (Time-Dependent Dielectric Breakdown)

Ben Kaczer, 2016 IEDM tutorial


Cadence Aging Reliability Support


Aging Analysis in Spectre


Aging Reliability Analysis HCI and NBTI/PBTI Analysis Flow

Fresh

Aging/EOL

Stress

Circuit Netlist + Reliability Model

Fresh Simulation

Device Degradation, Lifetime

Degraded model/netlist Generation

Aging Simulation

Waveform Compare Circuit degradation Check

Stressing

Aging


•  Reliability effects supported: –  Built-in HCI, NBTI and PBTI –  other mechanisms possible via URI (eg. TDDB, GOI)

•  SPICE Models supported for degradation: –  BSIM3, BSIM3V3, BSIM4, BSIMSOI, MOS9, MOS11, PSP 102, PSP 103,

HVMOS, HISIM2 and HISIM_HV. –  BSIMCMG, BSIMIMG, UTSOI, UTSOI2 –  BJT: VBIC, HiCUM and MEXTRAM –  Resistor: native and R3 –  Diode –  Verilog-A

•  Reliability models: –  ageMOS: Cadence proprietary reliability model for HCI, NBTI and PBTI (with

recovery) and degraded model generation (for legacy nodes). –  ageMOS2: Cadence proprietary reliability model for advance nodes. –  URI: API to allow customer implement own reliability models. –  TMI-aging, TMI self-heating models, TMI TDDB

Reliability Models supported


•  The Virtuoso Unified Reliability Interface (URI) allows you to add your own (custom/proprietary) reliability equations/models and supports user-defined degradation models

Virtuoso Unified Reliability Interface (URI) Using Custom Reliability Models

RelXpert Spectre APS/XPS

C code Custom/Proprietary reliability equations

URI

Reliability Simulation Results

Netlist with your own reliability model


TMI Simulation flow •  TMI Aging model offered three simulation flows with .option tmiAge=1

–  Aging only simulation flow. (tmishe=0) –  Aging stress w/o self-heating effect ( calculate degradation rates under TempE ) –  Measurement simulation w/o self-heating effect.

–  Self-Heating only simulation flow. (tmishe=1) –  w/o aging stress. –  Measurement simulation w/i self-heating effect.

–  Aging simulation with self-heating effect. (tmishe=2) –  Aging stress w/i self-heating effect ( calculate degradation rates under

TempE+dtemperature ). –  Measurement simulation either w/i or w/o self-heating effect.

tmiShe Self-heating aging 0 - Yes default 1 Yes - 2 Yes Yes


Spectre/TMI interface in ADE


RelXpert Spectre Native ageMOS HCI, NBTI, PBTI, TDDB* X X ageMOS2 HCI, NBTI, PBTI, TDDB* X X

Extraction X (verification) TMI HCI, BTI, TDDB X X URI Running X X

Model development, 3rd party spice X Self-heating Built-in (bsimcmg, bsimimg) X X

Spectre-SHE X X TMI-SHE X X

MC-aging ageMOS, ageMOS2 X TMI X

Reliability models support matrix

TDDB*: prototype


New Aging Model


Model HCI Degradation Saturation Effect

1.E-13

1.E-11

1.E-09

1.E-07

1.E-05

1.E-03

1.E-01

1.E-08 1.E-06 1.E-04 1.E-02 1.E+00 1.E+02 1.E+04 1.E+06 1.E+08

HCI(Vgs=1V)

Vds=0.8V 1.2V 1.6V

Degradationsaturation

V. Huard, IEDM, 2007

“New Generation Reliability Model”, S.-Y. Liao, C. Huang, T. Guo, A. Chen, Jushan Xie, Cadence Design Systems, Inc. S. Guo, R. Wang, Z. Yu, P. Hao, P. Ren, Y. Wang, R. Huang, Peking University, MOS-AK December 2016

http://www.mos-ak.org/berkeley_2016/publications/T11_Xie_MOS-AK_Berkeley_2016.pdf


1E-71E-51E-3 0.1 10 1000 1E-71E-51E-3 0.1 10 1000(a)

FinFET silicon dataNBTI stress

ΔVT

H(a.

u.)

Time(s)

FinFET silicon dataNBTI stress

Conventional model A (Power law) (b)

Conventional model B (Log law)

ΔVT

H(a.

u.)

Time(s)

1E-7 1E-5 1E-3 0.1 10 1000 1E-7 1E-5 1E-3 0.1 10 1000(c)

FinFET silicon data NBTI recvoery

Convertional model A (Power law)

Norm

aliz

ed Δ

VTH

(a.u

.)

Time(s)

FinFET silicon data NBTI recovery

Convertional model B(Log law) (d)

Norm

aliz

ed Δ

VTH

(a.u

.)

Time(s)1E-7 1E-5 1E-3 0.1 10 1000 1E-7 1E-5 1E-3 0.1 10 1000

(c)

FinFET silicon data NBTI recvoery

Convertional model A (Power law)

Norm

aliz

ed Δ

VTH

(a.u

.)

Time(s)

FinFET silicon data NBTI recovery

Convertional model B(Log law) (d)

Norm

aliz

ed Δ

VTH

(a.u

.)

Time(s)

Model limitation in FinFET BTI effect (log-log nonlinear)

TD model RD model

RD model TD model


Model result demonstration – New BTI models New model equations for FinFET: DC verification

•  Curve slope changed against time (in log-log, or log-lin scale) •  Gate bias dependency changed

1E-5 1E-3 0.1 10 1000 1E-5 1E-3 0.1 10 1000

(d)(c)

(a) Stage 2

Stage 1

Stage 2 Stage 2Stage 1

Stage 2

Stage 1

Vg=0V, -0.2V, -0.4V, -0.6V

Vg=-1.3V, -1.4V, -1.5V, -1.6V

NBTI stress

NBTI stress

Stage 1

(b)

T=50°C, 75°C,

100°C, 125°C

T=50°C, 75°C, 100°C, 125°C

NBTI recoveryNBTI Recovery

ΔVT

H (a

.u.)

ΔVT

H (a

.u.)

Time (s)

lines: new modelsymbols: FinFET silicon data



Time (s)


1E-5 1E-3 0.1 10 1000 1E-5 1E-3 0.1 10 1000

(d)(c)

(a) Stage 2

Stage 1


Stage 2

Stage 1

Vg=0V, -0.2V, -0.4V, -0.6V

Vg=-1.3V, -1.4V, -1.5V, -1.6V

NBTI stress

NBTI stress

Stage 1

(b)

T=50°C, 75°C,

100°C, 125°C

T=50°C, 75°C, 100°C, 125°C


ΔVT

H (a

.u.)

ΔVT

H (a

.u.)

Time (s)




Time (s)


1E-5 1E-3 0.1 10 1000 1E-5 1E-3 0.1 10 1000

(d)(c)

(a) Stage 2

Stage 1


Stage 2

Stage 1

Vg=0V, -0.2V, -0.4V, -0.6V

Vg=-1.3V, -1.4V, -1.5V, -1.6V

NBTI stress

NBTI stress

Stage 1

(b)

T=50°C, 75°C,

100°C, 125°C

T=50°C, 75°C, 100°C, 125°C


ΔVT

H (a

.u.)

ΔVT

H (a

.u.)

Time (s)




Time (s)


Slope1

Vgsdependence1

Slope2

Vgsdep.2

Slope1

Vgsdep.1Slope2

Vgsdep.2

Stress phase Recovery phase


Model result demonstration – New BTI models History effect: sequential simulation step

Time

Time

ΔDeg

0

Vgsstress

t1

Vgs1

Stressphase

t2

Vgs2

Recoveryphase

t3

Vgs3

Stressphase


BTI model with recovery effect -New Model

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18

Degrad

ation[a.u

.)

Time[s]

Vgs

0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18

Degrad

ation[a.u

.)

Time[s]

Stressonly

Simu.Result

-1V

-1.2V

-1.4V

-1.6V

0V


BTI model prediction with recovery effect

103 104 105 106 107

Degradation results@3years

Type B

ΔV

TH (a

.u.)

Frequency(Hz)

Type A

(b)10-3 100 103 106 109

ΔV

TH (a

.u.)

Time(s)

w/ recovery wo/ recovery

Iterated resultsIteration step:N=10 (a)

Over-estimationin3years

Vgs

Time

Duty=50%

Aperiodicsquarewaveformisappliedonaninvertorduringafewmilliseconds.Thenpredictionfor3years.


Comparison of AgeMOS and AgeMOS2

AgeMOS

§  Developed around 1995 §  Based on Lucky-Electron model §  Degradation log-log linear §  Many enhancements §  Using skew parameters

−  Vth, U0, Ua, Ub, Nfactor, Vsat, … §  Simple recovery model

−  linear extrapolation

§  Current CMC Approach

AgeMOS2

§  Developed around 2015 §  Based on trapping/de-trapping

model §  Degradation log-log nonlinear §  Degradation saturation/history

effect §  Using degradation directly

−  Vth and Ids degradation

§  Recovery with step-wise model −  More accurate −  Frequency dependency


AgeMos2 Model Extraction (Standalone tool)

Grouped reliability characteristics, such as degradation vs. time curves with varying Vds for a

fixed temperature.

Loading the data, and treating it for

extraction preparation, such as sorting, regrouping, data checking, etc.

Specified reliability model parameter

extraction

Data complete

Output EDA readable reliability modelcard

Fitting error report compared to the

input data, simulation results in text files,

graphically, etc.

No

Yes


Reliability Monte Carlo


Reliability Variation Components

Variation Aging variation

(AV)

Process Variation

(PV)

Correlation


Simulation Diagram for Flow I (1+N flow) (Aged model + Process Variation))

Mean Aging calculations

Aged spice model generations

Mean reliability model

Mean spice model Mean transistor-level simulation

Fresh

PV transistor-level simulation

Stress

MC Aging

Mean fresh

Degradation variation

MC RUNs: 1+N

N

•  Can perform Aging MC with current foundry aging + MC model •  Not expensive


Simulation Diagram for Flow II (N+N flow) (Age Variation + Process Variation)

AV aging calculations

Aged spice model generations

AV reliability model

PV spice model PV transistor-

level simulation

MC Fresh

PV transistor-level simulation

MC Stress

MC Aging

Fresh variation

Degradation variation

MC RUNs: N+N

N

N

N


VTH Comparison between Flow Zero, I and II

0

20

40

60

80

100

-0.52 -0.5 -0.48 -0.46 -0.44 -0.42

Flow I

Flow II

Flow Zero

VTH (V)

Flow Zero & Flow I

Fresh Aged

µ=-0.487 σ=0.0120

µ=-0.487 σ=0.0044

µ=-0.437 σ=0.0045

Ø Variation distribution (Y-axis) Ø Variation of aged VTH of flow I (only PV) is smaller than flow II (PV + AV)


Process Variation and Aging Variation Correlation

Ø Aged VTH and IDS variation become smaller after considering correlation between PV/AV

Spectre syntax for correlation analysis


Scatter Points for Operation Outputs

Ø The scatter points between aged VTH and aged IDS, GM, GDS, ROUT, IGS, and VDSAT Ø The correlation between PV/AV makes the outputs correlation more complicated


13-Stage Ring Oscillator Waveform Variation

Ø Flow II shows the fresh and aged RO waveform variation. The voltage degradation variation includes the PV-aware age calculations

A Complete Reliability Solution: Reliability Modeling, Applications, and Integration ... · 2021....

Documents

Transcript of A Complete Reliability Solution: Reliability Modeling, Applications, and Integration ... · 2021....