Post on 07-Sep-2020
On-Chip Silicon Odometers and their Potential Use inand their Potential Use in
Medical ElectronicsJohn Keane1 and Chris H. Kim2
1. Intel Corporation, Hillsboro, OR, USA 2. University of Minnesota, Minneapolis, MN, USA
Purpose• Shrinking feature sizes and voltage marginsShrinking feature sizes and voltage margins,
along with quickly evolving process architecture– Designs are susceptible to agingg p g g
• Expensive to characterize all aging mechanismsMajor consideration in fields such as aerospace– Major consideration in fields such as aerospace and medical
• On-chip sensors characterize or trigger compensation schemes for aging mechanisms
High frequency shift & timing resolution– High frequency shift & timing resolution– Automated tests with simple interfaces
Test many devices in parallel
2/18
– Test many devices in parallel– No expensive probing equipment
OutlineIntroduction to Circuit Aging• Introduction to Circuit Aging
• Overview of Aging Measurement Circuits• Relevance to Medical Electronics• Beat Frequency Silicon OdometerBeat Frequency Silicon Odometer– Separating BTI and HCI– Statistical Odometer
• Array-based TDDB Measurement SystemSummary• Summary
3/18
Combined Stress in Digital Circuits
4/18
Aging Measurement Circuits
• Digital systems for characterizationg y– ROSC edge capture (ISSCC ’10)– Critical path replica ROSC (VLSI Symp’10)– Starved ROSCs (ISSCC ‘08 and ‘10)
• Transistor arrays
5/18
– Smart NBTI array (IEDM ‘11)
Silicon Odometers in Medical Electronics• Want higher performance and smaller form• Want higher performance and smaller form
factors...but reliability is paramount– 10+ year lifetime requirements10+ year lifetime requirements
• Require efficient methods to characterize reliability mechanismsreliability mechanisms
• Focused on low power and analog space – Soft gate breakdown leads to increased powerSoft gate breakdown leads to increased power
consumption– Vth shift due to NBTI impacts sensing circuits
• Require precise measurements at low-stress conditions
6/18
• High volume to observe distribution tails
Beat Frequency Silicon Odometer
freffstress
7/18(Low frequencies used for illustration; Fig. in part from D. Schneider, IEEE Spectrum)T. Kim, R. Persaud, and C.H. Kim, "Silicon Odometer: An On-Chip Reliability Monitor for Measuring Frequency Degradation of Digital Circuits", JSSC, Apr. 2008
Digital Silicon Odometer Outputs1.E+03 1.E+01)
1.E+02ut C
ount
1.E+00
y Shi
ft (%
)“Dead Zone”: Counter maxes out or measurement takes too long Degradation
with stress
1 E+01
1.E 02
eter
Out
pu
1.E-01
Freq
uenc
y
1 E 00
1.E+01
Odo
m
1.E-02
Mea
sure
d
REF_ROSC slower than STR_ROSC
REF_ROSC faster than STR_ROSC
Higher 1st count = Higher initial resolution
Count decrease w/ stress1.E+00
-15 -10 -5 0 5 10 15
[REF_ROSC / STR_ROSC] Freq. Diff. (%)
1.E-03150 130 110 90 70 50 30 10
MOdometer Output Count
• With the reference ROSC faster than stressed ROSC, count decreases with stressR l i i hi h l i
8/18
• Resolution is highest early in stress– Use trimming caps to make fref and fstress close
Odometer Measurement Error1.0020) 1.0020)
1 0005
1.0010
1.0015
rmal
ized
)
1 0005
1.0010
1.0015
rmal
ized
)
0 9990
0.9995
1.0000
1.0005
ency
(No
0 9990
0.9995
1.0000
1.0005
ency
(Nor
0.9980
0.9985
0.9990
0 20 40 60
Freq
ue
0.9980
0.9985
0.9990
0 20 40 60
Freq
ue• Measured no-stress (i.e., 0V) characteristics
0 20 40 60Measurement Time (min)
0 20 40 60Measurement Time (min)
( , )• Singled ended ROSC varies with environmental
conditions
9/18
• Differential odometer more immune to variations, and takes fast measurements
HCI + BTI Silicon Odometer
• Stress Mode (ROSC loops opened)– BTI_ROSC gated off from supply– DRIVE_ROSC drives transitions; I/P driven by VCO
• Measurement Mode (ROSC loops closed)
10/18
– Both ROSCs connected to the power supply @ VCC – Switches between them are opened
Measurement Interface• All measurements• All measurements
done with LabVIEW & National Instruments data acquisition board
• Each reading ac ead gtriggered with a single pulse from tester
• Low speed digital interface– Internal circuits used
to minimize BTI recovery time
11/18
recovery time
HCI & BTI Measured Results1.E+00 1.E+01
hift
(%)
1.E+00hift
(%)
1.E-01
eque
ncy
S
1.E-01
eque
ncy
S
1.E-021.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05
Fre
1.E-021.E+00 1.E+02 1.E+04 1.E+06
Fre
• HCI slightly reduced with higher temperature
Stress Time (s) Stress Time (s)
– Reduced drain current • Both mechanisms degrade with stress voltage
12/18
– Point when HCI begins to dominate pushed out in time by >1 order of magnitude at 1.8V vs. 2.4V
Statistical ROSC Aging OdometerN d t d &• Need stressed & reference ROSC frequencies to be rip
h
qclose
• Difficult, costly to t h t d
Row
Pe
tune each stressed ROSC
• Use multiple Ref Ref ROSC 1Use multiple Ref. ROSCs with different frequencies Ref ROSC 2
• Cover the frequency distribution of the stressed array SCANOUT
RESULTS
Ref ROSC 3
RECOVERRESET
13/18
stressed array RESULTSMEASSTRESSRECOVER
Stressed ROSC Design
ctrl_loop
full_loopfull_loop ctrl_loop
• Only a section of each ROSC is stressed• Other control devices are 2.5V thick oxideOther control devices are 2.5V thick oxide• First measure the period of the ctrl loop• During full loop measurement cancel out this
14/18
• During full loop measurement, cancel out this portion
Measured DC Stress Results4.5) 1010
3 0
3.5
4.0
Shift
(%) Frequenft
μ(%
)
2.0
2.5
3.0
eque
ncy
11
ncy Shiftency
Shi
f
0 5
1.0
1.5
DU
T Fr
e
0 10 1
t σ(%
)Freq
ue
0.5250 260 270 280Fresh DUT Frequency (MHz)
0.10.11E+1 1E+2 1E+3 1E+4 1E+5
Stress Time (s)
• No significant correlation of the frequency shift with fresh frequency
15/18
• μ and σ of ∆f increase w/ power law behavior
Automated TDDB Characterization Array
col<m>row<n>
VSTRESS
GStressed Device
G
• Gate currents measured with A/D current monitor and hi t l l i
16/18
on-chip control logic• 16b results scanned out after each measurement
TDDB Array Stress Measurements2 o
0
1
F))
oB-3
-2
-1
ln(-l
n(1-
FB
D
-5
-4
-3l
-1 1 3 5 7 9 11 13TBD or TFAIL (ln(s))BD FAIL
• Array-based design allows us to extract Weibull slope factors with a single stress test
17/18
• Convenient method to investigate spatial correlation
SummaryOn chip monitors facilitate:• On-chip monitors facilitate:– High frequency measurement resolution (< 0.1% with
the Silicon Odometer)the Silicon Odometer)– High timing resolution (sub-μs measurements)– Simple, automated measurementsp ,– Parallel stressing for shortened experiment times
• Data can be used to create/validate modelsData can be used to create/validate models
• Concepts could be extended in to in-situ space, enabling real time system adjustmentsenabling real time system adjustments
• Valuable part of medical technology reliability i ’ t lkit
18/18
engineer’s toolkit