high frequency ac electromigration lifetime...

Symposia on VLSI Technology and Circuits

High Frequency AC Electromigration Lifetime Measurements from a 32nm

Test Chip

Chen Zhou1, Xiaofei Wang2, Rita Fung3, Shi-Jie Wen4, Rick Wong4 and Chris H. Kim1

1University of Minnesota, Minneapolis, MN2Intel Corporation, Hillsboro, OR, USA

3Cisco Systems, Inc., Hong Kong4Cisco Systems, Inc., San Jose, CA, USA

Outline

• Motivation• Proposed circuit based electromigration

characterization technique• AC and DC electromigration lifetime

measurements from 32nm test chip• Summary

Slide 1

Outline




Slide 2

void

M2

M1

e-

Atom

e-

A.S. Oates, et al., TDMR, 2009

Abrupt failure Progressive failure

void

M2

M1

e-

Atom

e-

Wire Failure due to Electromigration

• EM lifetime affected by current density/direction, temperature, and possibly switching frequency

Previous EM Test Structure

• Limitations of traditional probing method– Frequency (~5MHz) lower than actual chip clock freq.– Unable to generate realistic AC current– Large test area due to pads, long test time

Slide 4

0t

I

0t

IDC Pulsed DC Square AC

DUTI

<5MHz<5MHz

0 t

I

Test instrument

Proposed Circuit based Approach

• Advantages of proposed circuit based approach– High operating frequency (~GHz)– Realistic AC current– Small test area due to shared pads, short test time– BTI and HCI effects in driver captured closer to reality

Slide 5

0 t

I

0 t

IDC Pulsed DC Square AC

DUTI

Real AC

~GHz~GHz

0t

I

~GHz

0t

I

On-chip clock source

Outline




Slide 6

32nm Test Chip Overview

Slide 7

Heater1 Heater2 Heater3

60 DUTs

220µm

350µ

m

Metal gate heaters

200µm

50nmM1

M2

EN_B

Driver A LoadDriver B

IN_BIN_AEN_A

Driv

ers

& s

witc

hes D

rivers & sw

itches

Metal gate heaters

V

EM Stress Modes Supported

• Supports four different EM modes by configuring driver inputs

Slide 8

1

1

1

0

1

1

1

1

Local Metal Gate Heaters

• Local heaters can raise DUT temp. to >350ºC • Stress temperature is set by adjusting heater

voltage until target resistance is reachedSlide 9

0 50 100 150 200 250 300 3500.9

1

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8


60 DUTs

Metal gate heaters

Heater Thermal Simulation Results

Slide 10

• More uniform temperature can be obtained using multiple heater voltages (e.g. 28V, 20V, 28V)

0V0V 0V

28V 20V 28V

300267233

200

60 DUTs

ºC

Air

Ceramic

Die

Cross-section view

HeaterLow-k Dielectric

Silicon Sub

Top view

6mm

Die

2mm

Heater

220μm

350μ

m

BOX


Automated EM Testing Flowchart

• Script program with automatic temperature control enables accurate and efficient data collection

Slide 11

EM stress/measurement loop

Heater calibration

Configure stress mode

Set temp. target to 325ºC

Stress for 10 minutes

Set temp. target to 100ºC

Measure res. of DUTs

Measure heater res.

Translate res. to temp.

Hold for 1 second

Temperature control loop

Adjust heater power

Target reached?Yes

No

Wait until target reached

Wait until target reached0

50

100

150

200

250

300

350

400

Tem

pera

ture

(ºC

)

Time (minutes)

3 heaters Stress

Measurement

Temperature reduced to 100˚C for accurate

measurements

0 5 10 15 20

32nm Test Chip Die PhotoProcess 32nm HKMG

# of DUT 60

Stress modes DC, pulsed DC,square AC, real AC

Stress current source

On-chip VCO and driver (>3GHz @ RT)

Stress driver VDD <1.5V

Stress/measure temperature 325/100 ºC

Stress frequency <900MHz @ 325ºC

Heat source Metal gate heaters

Test approach 4 terminal Kelvin measurement

Slide 12

2.0mm

Driv

er

Driv

er

VCOHeater Feed

Heater Feed

DUT core

Outline




Slide 13

EM Lifetime under DC

• Results consistent with previously reported data• Abrupt failure has smaller mean and larger

varianceSlide 14

Stress time (hours)

500

1000

1500

2000

2500

3000

3500

4000

4500

5000

ΔR=10%

Res

ista

nce

(Ω)

0 5 10 15 20 25 30 35 40 45 50

Abrupt failure

Progressive failure

Stress: DC, 325ºC, 1.5V Measure: 100ºC

Stress time (hours)10 10 100 1 2

Fa

ilure

rate

(%)

Total = abrupt + progressive

Criteria: ΔR>10% Meas.: 100ºCStress: DC 325ºC 1.5V

void

Progressive failure

M2

M1

e-

void

Abrupt failure

M2

M1

e-

1

510

25

50

75

9095

99

EM Lifetime under Pulsed DC

• Possible reasons for ratio>2:– Lower Iaverage less Joule heating lower temperature– EM self-recovery during off periods– BTI aging during off period reduces stress current

Slide 15

Stress time (hours)100 101 102

Failu

re ra

te (%

)

×2.7

Measure: 100ºCStress: 325ºC, 1.5V

DCPulsed DC (200MHz)

1

510

25

50

75

9095

99

0t

I

0t

I

DC

Pulsed DC200MHz, 50%

duty cycle

EM Lifetime under Square/Real AC

• Negligible resistance change under both square AC and real AC current from 200 to 900MHz

Slide 16

Stress time (hours)

Res

ista

nce

(Ω)

0 5 10 15 20 25 30 35 40 45 50

60 wires

500

1000

1500

2000

2500

3000

Stress time (hours)0 5 10 15 20 25 30 35 40 45 50

No changeNo change

Stressed @ 325ºC, 1.5V, measured @ 100ºC

60 wires I

t

f=900MHz

00t

I f=900MHz

3% 3%

Alternative Testing Method: Two Phase Stress

• Apply AC stress first and then switch to DC stress*

• DC EM lifetime can reveal AC EM stress impact

Slide 17

Real AC + DC

Square AC + DC

0 t

IPure DC

I

t

0 t

I

*R. Shaviv, et al., IRPS, 2011

EM Lifetime under Square AC + DC

• No apparent difference between pure DC and square AC + DC

• Weak dependence on frequency Slide 18



Failu

re ra

te (%

)

DCDC after 200MHz square ACDC after 900MHz square AC

1

510

25

50

75

9095

99

0 t

I

0

I

t

I

0

200MHz

900MHz

52.7hrs

Pure DC

Square AC + DC

Square AC + DC

t


Failu

re ra

te (%

)

DCDC after 200MHz real ACDC after 900MHz real AC


1

510

25

50

75

9095

99

• Real AC pre-stress results in 64-83% longer DC EM lifetime

• Weak dependence on frequency

EM Lifetime under Real AC + DC

0 t

I

0 t

I

t

I

0

200MHz

900MHz

52.7hrs

MTF=14.2hrs

MTF=23.3hrs

MTF=26.0hrs

Pure DC

Real AC + DC

Real AC + DC

• Real AC stress may actually make wires more robust• Additional time may be required for DC EM

vacancies to nucleate and evolve

Slide 20

t

I

0

Fewervacancies

After real AC stress

Initialvacancies

Fresh

Void

After DC stress

Possible Explanation for Longer EM Lifetime under Real AC

Another Explanation for Longer EM Lifetime under Real AC

• BTI in driver lower stress current longer DC EM lifetime

Slide 21

Real AC + DC

Square AC + DC

t

IPure DC

I

t

t

I

Due to BTI, HCI

Pure DCDue to HCI

With square AC

Pure DC

With real AC

Due to BTI

Summary• EM lifetime measured up to 900MHz from a 32nm

test chip • Square AC did not change DC EM lifetime• Real AC increased DC EM lifetime

– Real AC could actually make wires more robust– Front end BTI aging may reduce EM stress current

Slide 22

high frequency ac electromigration lifetime...

Documents