high frequency ac electromigration lifetime...
TRANSCRIPT
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
• Motivation• Proposed circuit based electromigration
characterization technique• AC and DC electromigration lifetime
measurements from 32nm test chip• Summary
Slide 2
Slide 3
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
• Motivation• Proposed circuit based electromigration
characterization technique• AC and DC electromigration lifetime
measurements from 32nm test chip• Summary
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
Heater1 Heater2 Heater3
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
Heater1 Heater2 Heater3
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
• Motivation• Proposed circuit based electromigration
characterization technique• AC and DC electromigration lifetime
measurements from 32nm test chip• Summary
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
Stress time (hours)100 101 102
Measure: 100ºCStress: 325ºC, 1.5V
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
Slide 19
Stress time (hours)100 101 102
Failu
re ra
te (%
)
DCDC after 200MHz real ACDC after 900MHz real AC
Measure: 100ºCStress: 325ºC, 1.5V
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