CHAPTER 3 EXPERIMENTAL
3.1. The Scanning Electron Microscopes
The SEMs used in this study are currently in common use in modern IC fabs. The
first SEM is a model ES20XP scanning E-beam inspection system manufactured by
KLA-Tencor. The beam conditions used for this SEM were acceleration voltage of
800eV, and a probe current of 66 nanoamperes. The second SEM is a SEMVision CX
manufactured by Applied Materials, using an acceleration voltage of 1000eV, and a
probe current of 58 picoamperes. The third SEM is a model JWS 7515 wafer inspection
system manufactured by JEOL, using an acceleration voltage of 2500eV and probe
current of 10pA.
3.2. The Wafer Cleans
The three wafer cleans used were ALEG310, NE111, and DSP+. ALEG310 is an
alkaline (pH of ~11) amine-based organic solvent blend. The details of the processing
are ~10 minute batch clean , at 85°C, followed by a deionized water (DI) rinse with CO2
bubbler which brings the pH to ~4. The NE111 clean is a buffered pH-stable acidic (pH
of ____) clean, containing ammonium fluoride NH4F. NE111 is used in a single wafer
spin processor at room temperature, with duration less than one minute. This is followed
by a DI rinse with CO2 bubbler which brings the pH to ~4. The DSP+ clean is an acidic
clean, pH = 1, containing less than 10% sulfuric acid, less than 10% hydrogen peroxide,
and HF with a concentration on the order of parts per million. DSP+ is used in a single
wafer spin processor at room temperature, with duration less than one minute. This is
followed by a (DI) rinse with CO2 bubbler which brings the pH to ~4.
3.3. The Electrical Test Structures investigated
For the first analysis, five window-1 stitches were investigated. A stitch is a two-
point test structure with 2000 windows at nominal sizing, connected in series such that
any single open contact will cause the entire structure to fail. There was a 2000 window
stitch for each of the following types of windows:
• Window-1 between metal-1 and 1.5Volt N+-Source/Drain in P-Tub • Window-1 between metal-1 and 1.5Volt P+-Source/Drain in N-Tub • Window-1 between metal-1 and 3.3 Volt N+-Source/Drain in P-Tub • Window-1 between metal-1 and 3.3Volt P+-Source/Drain in N-Tub • Window-1 between metal-1 and 1.5Volt N-doped poly.
The layout of the source/drain stitch is seen in Figure 1. There are 2000 windows
laid out at 0.20µm X 0.20µm. The metal pads are 0.24µm in width with a 0.02µm
extension on either side of the window and an extension beyond window of 0.06µm.
Figure 1. Schematic of 2000-Window Source/Drain Stitch
The layout of the gate stitch is similar to the thinox stitch. There are 2000 windows laid
out at 0.20µm X 0.2µm. The metal-1 pads are 0.24µm in width with a 0.02µm extension
on either side of the window and an extension beyond window of 0.06µm. In addition,
two references structures were electrically tested to determine the parametric values for
the nominal window-1. These structures did not undergo SEM inspection, and should
therefore have been processed correctly. The first structure was a single window Kelvin
cross-bridge, which will yield the ohmic value for a single contact (Contact Resistance
RC.) This is shown in Figure 2.
Figure 2. Schematic of Single Window Kelvin-Cross-Bride That Measures Window-1 to
Source/Drain Contact Resistance
The second structure was a van de Pauw structure, shown in Figure 3. This test
structure gives sheet resistance for the levels tested (in this case source/drain, gate, and
metal-1) in units of ohms/square (Ω/¨).
Figure 3: Four-Point Thinox Van de Pauw Layout for Measuring Sheet Resistance
With the values from the Kelvin and van de Pauw which did not undergo SEM
inspection, along with the layout of the stitch structures, it is possible to calculate an
expected resistance value for a fully functional stitch. This is a validation of the
resistance measurements from the control cell: the stitches that did not receive the SEM
inspection. Analysis of the stitch structures show that for every window there is one half
of a thinox or poly square. Therefore, the expected value of a Source/Drain or gate stitch
is:
(# of Windows) X (RC ) + (# of Windows) X (S/D Sheet Resistance)/2 (23)
. Metal sheet resistance for the window-1 stitches can be ignored since it is orders
of magnitude less than the source/drain or gate sheet resistance.
For the second analysis, window 2 interactions were investigated. In this case
both a window-2 stitch, and a window-2 Kelvin structure underwent SEM inspection.
The layouts for the 2000 window stitch and the single window Kelvin structures are
similar to analysis one. Window-2 is sized at 0.24µm, with both metal-1 and metal-2 laid
out at 0.24µm width so that the via is borderless. The stitch layout can be seen in Figure
4, the Kelvin structure is laid-out similar to the thinox Kelvin of Figure 2.
Figure 4: Window-2 Stitch Layout
3.4. The Procedure - Experiment One:Window-1
Two wafers (wafers 01 and 02) were processed through metal-1, capped and
pulled from the line for electrical testing and analysis. The wafer layout consists of a
total of 76 flash fields laid out in 10 rows by 11 columns. Figure 5 depicts the wafer with
row labels. Row 1 and 10 did not undergo SEM inspection at any time. Rows 2 and 3
underwent ES20 inspection. Rows 4 and 5 underwent JEOL inspection. Rows 6 and 7
underwent SEMVISION inspection. Rows 8 and 9 underwent inspection by all three
SEMs at precisely the same spots on the stitches so that the inspections overlapped.
Figure 5. Wafer Layout Identifying SEM Inspection Row
As stated above the beam conditions for the three SEMs were 800Vacc and 66nA
Ip for the ES20, 1000Vacc and 58pA Ip for the SEMVISION, and 2500Vacc 10pA Ip for
the JEOL. The field of view for the ES20 was 10µm by 10µm, for a total scanned area of
100µm2. The field of view for the SEMVISION was 3µm by 3µm, for a total scanned
area of 9µm2. The field of view for the JEOL was 3.25µm by 2.5µm for a total scanned
area of 8.125µm2. For the SEMVISION and JEOL, the dwell time on each site was 17 ±
1 second. For the ES20, the dwell time on each site was 328 millisecond. Multiplying
the probe current by the scan time and dividing by the scan area yields the total electron
dose per square micron. As such, the calculated doses are for the ES20
___electrons/µm2, for the SEMVISION 6.84E+8 electrons/µm2, and for the JEOL
1.31E+8 electrons/µm2. These details are summarized in Table 1.
Table 1: SEM Scanning Conditions for the Five Stitches in Experiment One
SEM Vacc Probe Current
Field of View
Scan Area Scan Time Total Electron Dose electrons/µm^2
ES20XP 800eV 66nA 10µm X 10µm
100 µm^2 328 msec
SEMVision CX 1000eV 58pA 3µm X 3µm 9 µm^2 17 sec ~6.84E+08
Jeol JWS7515 2500eV 10pA 3.25µm X 2.5µm
8.125 µm^2 17 sec ~1.31E+08
The battery of SEM scans were performed at four points in the processing. These scan
points were:
Scan #1. On the oxide of dielectric-1, after window-1 pattern and plasma etch and clean, prior to window-1 IMP Ti/TiN liner deposit
Scan #2. After window-1 IMP Ti/TiN liner deposition prior to CVD tungsten deposition
Scan #3. After window-1 tungsten CMP and clean prior to Metal-1 stack deposition Scan #4. After metal-1 patterning and etch
The four areas of the stitches scanned by the SEMs are indicated in Figure 6.
Figure 6. SEM Scan Locations after (1) Window Etch, (2) Window Liner Deposition, (3) Tungsten CMP, (4) Metal-1 Etch, With Accompanying SEM Image
It should be noted here that metal-1 was intentionally misaligned in both the X and Y
direction by ~80µm. This amount of misalign is the upper limit of the specification for
overlay tolerance, but it is within spec. After the two wafers completed the fourth run of
SEM inspections, they were split. One wafer received an ALEG310 clean, and then
dielectric caps were deposited. The other wafer did not receive the ALEG310 clean, but
went directly to caps deposition after SEM scan #4. Either process falls within the
normal flow of a product wafer.
3.5. The Procedure: Experiment 2: Window-2
The same basic stitch and Kelvin layouts used in experiment 1 are used again in
experiment 2. In this case, the window-2 module was investigated. Four wafers were
processed, according to the normal process flow through metal-2 deposition. Metal-2
was then misaligned by ~80µm in both the X and Y direction, exposing the underlying
tungsten plugs. The window-2/metal-2 surface was then scanned by the SEMs. The
ES20 was not available for this experiment so only the SEMVision and JEOL were
utilized in the experiment. The flash-fields scanned remained as detailed in Figure 5 with
the exception of the ES20, but in this run both the 2000 window stitch and the single
window Kelvin cross-bridge structures were scanned by the SEMs. The scan areas
remained the same as in experiment 1, 3µm by 3µm for SEMVISION, and 3.25µm by
2.5µm for JEOL. The beam conditions for the SEMs also remained identical to
experiment 1, but scan time was increased to 25 ±1 seconds. These SEM conditions for
experiment 2 are detailed in Table 2.
Table 2: SEM Scanning Conditions for the Window-2 Stitch and Window-2 Kelvin in Experiment Two
SEM Vacc Probe Current
Field of View
Scan Area Scan Time Total Electron Dose electrons/µm^2
SEMVision CX 1000eV 58pA 3µm X 3µm 9 µm^2 25 sec 1.01E+09
Jeol JWS7515 2500V 10pA 3.25µm X 2.5µm
8.125 µm^2 25 sec 1.92E+08
After the SEM scanning, the wafers were split. One wafer each went through an
ALEG310 clean, an NE111 clean, a DSP clean, and one wafer did not receive any clean.
The details of the cleans split by wafer are given in Table 3
Table 3: Specification of Cleans Split by Wafer
Wafer Clean
03 Skip
04 ALEG310
05 NE111
07 DSP+
06 Control
08 Control
09 Control
The four wafers, in addition to three control wafers were then capped and pulled from the
line for electrical testing and physical analysis. The three control wafers (wafers 06, 08,
and 09) were processed through metal-2 according to the normal process flow of the
production routing. The data for these three wafers were used as the control data for both
experiment 1 and experiment 2.
CHAPTER 4 RESULTS AND DISCUSSIONS
4.1 Experiment 1 – Window-1 Interactions
4.1.1 Electrical Testing Results
The electrical resistance of the 2000 window-1 stitches for each of the five
window-1 types is presented in the following five figures and tables
N+_
CH
AIN
_1V
_Rt (
kOhm
s)
100
200
300
400
500
.Con
trol
.Non
e
ES
20
JEO
L
SE
MV
isio
n
SE
MV
isio
n+E
S20
+JE
OL
SEM
.NoneES20JEOLSEMVision
.01 .05 .10 .25 .50 .75 .90 .95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
Figure 7: Oneway Analysis of Win-1 to 1.5V NSD Stitch Resistance By SEM Inspection,
With Quartile Plot
Table 4: Total Resistance Mean, Standard Deviation, and Quartiles of The Win-1 to 1.5V NSD Stitch Resistance By SEM Inspection (Units of Kilohms)
Level Number Mean Std Dev Min 10% 25% Median 75% 90% Max.Control 218 190.0 123.5 168.7 171.8 174.7 179.5 187.8 194.7 2000.0.None 12 199.5 4.6 191.3 192.2 195.7 199.2 204.2 205.5 205.7ES20 32 189.8 11.0 172.5 175.5 181.0 189.5 196.4 208.2 213.3JEOL 36 178.2 7.9 166.8 168.6 170.6 177.2 184.1 190.9 194.5
SEMVision 36 178.1 7.9 168.6 169.2 172.3 176.4 183.4 191.3 197.0SEMVision+ES20+JEOL 30 15734.5 84841.4 185.9 187.8 196.2 225.1 274.5 370.6 464940.0
Quantiles
Wafer-1 in Red Wafer-2 in Black
N+_
CH
AIN
_3V
_Rt (
kOhm
s)
100
1000
.Con
trol
.Non
e
ES
20
JEO
L
SE
MV
isio
n
SE
MV
isio
n+E
S20
+JE
OL
SEM
.01 .05 .10 .25 .50 .75 .90 .95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
Oneway Analysis of N+_CHAIN_3V_Rt (kOhms) By SEM
Figure 8: Oneway Analysis of Win-1 to 3.3V NSD Stitch Resistance By SEM Inspection,
With Quartile Plot
Table 5: Total Resistance Mean, Standard Deviation, and Quartiles of The Win-1 to 3.3V NSD Stitch Resistance By SEM Inspection (Units of Kilohms)
Level Number Mean Std Dev Min 10% 25% Median 75% 90% Max.Control 218 9.18E+07 1.35E+09 168.1 171.0 174.1 179.6 188.5 196.6 2.00E+10.None 12 201.9 4.9 195.9 196.0 198.0 201.0 204.6 211.1 213.6ES20 32 189.6 11.7 171.3 174.2 180.3 189.0 196.9 207.5 216.8JEOL 36 178.0 8.4 166.5 167.9 170.4 176.4 185.3 191.0 195.2
SEMVision 36 178.1 8.9 167.4 168.4 171.8 175.4 184.1 193.9 199.5SEMVision+ES20+JEOL
30 1.67E+05 6.48E+05 186.2 188.68 202.28 228.4 326.3 3635 3.00E+06
Quantiles
Wafer-1 in Red Wafer-2 in Black
P+_
CH
AIN
_1V
_Rt (
kOhm
s)
1000
10000
.Con
trol
.Non
e
ES
20
JEO
L
SE
MV
isio
n
SE
MV
isio
n+E
S20
+JE
OL
SEM
.01 .05 .10 .25 .50 .75 .90 .95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
Quantiles
Oneway Analysis of P+_CHAIN_1V_Rt (kOhms) By SEM
Figure 9: Oneway Analysis of Win-1 to 1.5V PSD Stitch Resistance By SEM Inspection,
With Quartile Plot
Table 6: Total Resistance Mean, Standard Deviation, and Quartiles of The Win-1 to 1.5V PSD Stitch Resistance By SEM Inspection (Units of Kilohms)
Level Number Mean Std Dev Min 10% 25% Median 75% 90% Max.Control 218 1.38E+06 2.03E+07 437.5 459.8 476.3 500.8 562.7 634.0 3.00E+08.None 12 606 57.8 520.6 526.1 556.7 607.9 664.2 683.1 686.8ES20 32 571 39.8 493.8 520.8 533.6 571.4 607.8 628.1 641.0JEOL 36 546 54.3 475.2 480.0 500.7 537.9 589.7 622.6 683.3
SEMVision 36 572 80.0 478.2 487.9 500.6 555.0 630.2 663.6 856.0SEMVision+E
S20+JEOL 30 3442 7808.2 536.4 565.4 641.1 901.4 1484.5 9572.1 34254
Quantiles
Wafer-1 in Red Wafer-2 in Black
P+_
CH
AIN
_3V
_Rt (
kOhm
s)
1000
10000
.Con
trol
.Non
e
ES
20
JEO
L
SE
MV
isio
n
SE
MV
isio
n+E
S20
+JE
OL
SEM
.01 .05 .10 .25 .50 .75 .90 .95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
Oneway Analysis of P+_CHAIN_3V_Rt (kOhms) By SEM
Figure 10: Oneway Analysis of Win-1 to 3.3V PSD Stitch Resistance By SEM
Inspection, With Quartile Plot
Table 7: Total Resistance Mean, Standard Deviation, and Quartiles of The Win-1 to 3.3V PSD Stitch Resistance By SEM Inspection (Units of Kilohms)
Level Number Mean Std Dev Min 10% 25% Median 75% 90% Max.Control 219 539 108 449.1 466.7 484.2 509.4 561.2 636.0 1572.None 12 980 1260 532.2 538.3 562.1 642.3 661.2 3689 4979ES20 32 582 40 501.5 531.7 545.6 582.0 614.7 638.0 653.7JEOL 36 568 103 479.5 493.4 508.6 551.2 593.9 649.0 1090
SEMVision 36 630 326 483.8 496.7 515.5 575.2 640.7 663.7 2498SEMVision+E
S20+JEOL 30 271795 1141549 560.2 601.0 643.4 961.9 3276 70857 6000000
Quantiles
Wafer-1 in Red Wafer-2 in Black
Pol
y_C
HA
IN_R
t (kO
hms)
35
40
45
50
55
.Con
trol
.Non
e
ES
20
JEO
L
SE
MV
isio
n
SE
MV
isio
n+E
S20
+JE
OL
SEM
.Control
.None
ES20JEOLSEMVision
.01 .05 .10 .25 .50 .75 .90 .95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
Oneway Analysis of Poly_CHAIN_Rt (kOhms) By SEM
Figure 11: Oneway Analysis of Win-1 to Poly-Silicon Gate Stack Stitch Resistance By
SEM Inspection, With Quartile Plot
Table 8: Total Resistance Mean, Standard Deviation, and Quartiles of The Win-1 Poly-Silicon Gate Stack Stitch Resistance By SEM Inspection (Units of Kilohms)
Level Number Mean Std Dev Min 10% 25% Median 75% 90% Max.Control 219 41.93 2.02 36.69 38.91 40.25 42.30 43.72 44.36 45.19.None 12 43.38 3.55 39.17 39.25 40.63 42.40 46.84 48.96 49.07ES20 32 42.48 2.38 37.90 38.46 40.61 42.96 44.31 45.64 45.97JEOL 36 43.11 1.67 40.06 40.68 41.96 43.34 44.37 45.21 46.45
SEMVision 36 43.23 2.32 38.27 39.78 41.05 43.85 45.03 45.91 47.18SEMVision+E
S20+JEOL 30 1245 6561.5 41.53 43.34 45.80 47.33 48.61 50.44 35986
Quantiles
It is evident from quartile plots for the four thinox stitches, that the sites that
received SEM inspection by all three SEMs have considerable disturbances in resistance.
Wafer-1 in Red Wafer-2 in Black
The mean, median and standard deviation of the stitch resistance for the combined cell is
significantly greater than any of the other cells of the experiment as well as the control
cell. While the upper and lower quartiles for all but the combined SEM NSD stitches fall
between approximately 170 and 200 kO for both the 1.5V and 3.3V, the NSD quartiles
for the combined SEM cell fall between ~200 and 325kO. While the upper and lower
quartiles for all but the combined SEM PSD stitches fall between approximately 475 and
650 kO for both the 1.5V and 3.3V, the PSD quartiles for the combined SEM cell fall
between ~640 and 3300 kO. In addition, for the four source/drain stitches approximately
10% of the combined SEM structures have measured resistances in the megohms up to
and including electrical opens. For the other five cells the few opens that occurred were
obvious outliers and the distributions were much tighter. Scanning electron microscope
imaging has caused the electrical failure of window-1 to source/drain contacts.
For the window-1 stitch to gate, there is also a statistically significant increase in
measured resistance, however, the gate stitches did not show the high number of
catastrophic failures that the source/drain stitches showed. Figure 12 replots the window-
1 to poly stitch excluding a single outlier data point from the combined SEM cell.
Pol
y_C
HA
IN_R
t (kO
hms)
36
38
40
42
44
46
48
50
.Con
trol
.Non
e
ES
20
JEO
L
SE
MV
isio
n
SE
MV
isio
n+E
S20
+JE
OL
SEM
Each PairStudent's t 0.05
.Control
.NoneES20JEOLSEMVision
SEMVision+ES20+JEOL
Level
36.69
39.17 37.9 40.06 38.27
41.53
Minimum
38.91
39.248 38.462 40.675 39.777
43.14
10%
40.25
40.6325 40.61
41.9575 41.045
45.77
25%
42.3
42.395 42.96 43.34 43.845
47.27
Median
43.72
46.835 44.3125 44.3725 45.0325
48.485
75%
44.36
48.962 45.637 45.212 45.914
50.39
90%
45.19
49.07 45.97 46.45 47.18
50.63
Maximum
Quantiles
.Control
.NoneES20JEOL
SEMVisionSEMVision+ES20+JEOL
Level
219 12 32 36
36 29
Number
41.9282 43.3758 42.4794 43.1092
43.2256 47.0376
Mean
2.01582 3.55013 2.38251 1.67499
2.32092 2.21079
Std Dev
0.1362 1.0248 0.4212 0.2792
0.3868 0.4105
Std Err Mean
41.660 41.120 41.620 42.542
42.440 46.197
Lower 95%
42.197 45.631 43.338 43.676
44.011 47.879
Upper 95%
Means and Std Deviations
Oneway Analysis of Poly_CHAIN_Rt (kOhms) By SEM
Figure 12: Oneway Analysis of Win-1 to Poly-Silicon Gate Stack Stitch Resistance By SEM Inspection, Excluding a Single Outlier Point From Combined SEM Cell
In this case, the mean of the combined SEM cell increased from ~43kO to ~47kO
as compared to the other experimental cells, which is an increase of only ~9%. In
addition, the standard deviation remained similar between the cells with no increased
spread in the distribution of the resistances for the combined SEM cell. This percentage
increase may seem inconsequential, but if it can be assumed that only one of the four
scan points was responsible for the increase (not necessarily a safe assumption) then that
4kO resistance was caused by a scan area of only a few square microns, encompassing
twenty windows or less. This is, potentially, an average increase of 200O per contact, for
a gate contact resistance that normally measures ~10 O per contact. It is also interesting
that the average measured resistances for every experimental cell is approximately 1-1.5
kO greater than the average measured resistance for the control cell. The most likely
explanation for this is that the experimental wafers (wafers 01 and 02) experienced a
significant amount of time lag between processing steps, whereas the control wafers
(wafers 06, 08, 09) flowed through the process relatively unencumbered. It is a known
phenomenon that significant time delays between certain processing steps in the backend
will lead to increased contact resistances. This should not be a concern, in that on the
experimental wafers, the data from rows 1 and 10 (labeled as the None cell) which did
not receive SEM scans can be considered as a control reference. If this is assumed, then
Figure 12 would indicate that the poly stitches that only received a single SEM inspection
are not significantly different from poly stitches that did not receive any SEM inspection.
The electrical test data also included measurements of contact resistance and
source/drain sheet resistance. These Kelvin and van de Pauw structures did not undergo
SEM inspection. From these measurement values, it is possible to calculate the expected
stitch value for the corresponding window flavor, on a point by point basis. As stated
above, for a 2000 window stitch with one half of a source/drain thinox square per
window, the calculated stitch resistance would be (2000 X W1 Rc) + (1000 X Rsheet).
This is not an exact method since some assumptions have been made, but it is a valid
approximation to the first order. The ratio of the measured stitch resistance to the
calculated theoretical stitch resistance should approach unity if there is no SEM
interaction with the sample that affects the integrity of the tungsten plugs. The following
four figures plot, by SEM split, the ratio of measured stitch resistance to calculated
theoretical stitch resistance, for the four types of window-1 to source/ drain stitch.
1.5V
NS
D (m
srd/
calc
'd)
1
10
.Con
trol
.Non
e
ES
20
JEO
L
SE
MV
isio
n
SE
MV
isio
n+E
S20
+JE
OL
SEM
.Control
SEMVision+ES20+JEOL
.01 .05 .10 .25 .50 .75 .90 .95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
.Control
.NoneES20JEOLSEMVisionSEMVision+ES20+JEOL
Level
0.8950310.899902 0.863940.8354360.8752090.889767
Minimum
0.9380190.9024420.8863080.8701410.8925070.919474
10%
0.9561990.9153440.9075550.9111220.9082480.993259
25%
0.9829740.9589760.9206420.9327180.9399491.093473
Median
1.005860.9808240.9504520.9527040.9535661.269629
75%
1.031180.9877070.9775370.9735770.973433 1.76216
90%
9.5634290.9900631.0248341.0053851.0107532133.633
Maximum
Quantiles
.Control
.NoneES20JEOLSEMVisionSEMVision+ES20+JEOL
Level
218 12 32 36 36 30
Number
1.0228 0.9489 0.9291 0.9280 0.9340 72.2693
Mean
0.582 0.035 0.037 0.037 0.033
389.330
Std Dev
0.039 0.010 0.007 0.006 0.005 71.082
Std Err Mean
0.95 0.93 0.92 0.92 0.92
-73.11
Lower 95%
1.10 0.97 0.94 0.94 0.95
217.65
Upper 95%
Means and Std Deviations
Oneway Analysis of 1.5V NSD (msrd stitch Rt/ calc'd stitch Rt) By SEM Split
Figure 13: Ratio of 1.5V NSD Measured Stitch Resistance to Calculated Theoretical Stitch Resistance Value by SEM Split
Wafer-1 in Red Wafer-2 in Black
3.3V
NS
D (m
srd/
calc
'd)
1
10
.Con
trol
.Non
e
ES
20
JEO
L
SE
MV
isio
n
SE
MV
isio
n+E
S20
+JE
OL
SEM
.Control
SEMVision+ES20+JEOL
.01 .05 .10 .25 .50 .75 .90 .95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
.Control
.NoneES20JEOLSEMVisionSEMVision+ES20+JEOL
Level0.9054590.8770020.9228660.867225 0.873310.977866
Minimum0.9503950.8845750.9320830.9047520.9020830.997798
10%0.9697040.9235940.9413770.9178530.9188911.027294
25%0.9911970.9626240.9620160.9481310.9469521.186114
Median1.0276660.9850960.977416 0.977270.9670741.600372
75%1.0545191.0198360.9896520.9861570.99080516.54986
90%971817301.0235951.0119961.030394
1.025613638.84
Maximum
Quantiles
.Control
.NoneES20JEOLSEMVisionSEMVision+ES20+JEOL
Level 218 12 32 36 36 30
Number 445836
1 1 1 1
807
Mean 6581974
0 0 0 0
3092
Std Dev 445788
0 0 0 0
564
Std Err Mean-432792
1 1 1 1
-347
Lower 95%1324463.80.98294980.96894810.95923650.95698521961.8246
Upper 95%
Means and Std Deviations
Oneway Analysis of 3.3V NSD (msrd stitch Rt/ calc'd stitch Rt) By SEM Split
Figure 14: Ratio of 3.3V NSD Measured Stitch Resistance to Calculated Theoretical Stitch Resistance Value by SEM Split
Wafer-1 in Red Wafer-2 in Black
1.5V
PS
D (m
srd/
calc
'd)
1
10
.Con
trol
.Non
e
ES
20
JEO
L
SE
MV
isio
n
SE
MV
isio
n+E
S20
+JE
OL
SEM
.Control
SEMVision+ES20+JEOL
.01 .05 .10 .25 .50 .75 .90 .95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
.Control
.NoneES20JEOLSEMVisionSEMVision+ES20+JEOL
Level
0.8371720.8744750.8066950.8736050.813122 0.84812
Minimum
0.9164250.879747 0.83011
0.905562 0.87168
0.930398
10%
0.953542 0.919980.8673890.9343540.936893 1.02401
25%
0.9938080.9708290.921047 1.024141.0131181.276079
Median
1.1369561.0208680.9869341.0883071.1492012.449486
75%
1.2952361.1963751.0676781.2116461.22561511.99895
90%
570342.21.2532561.4215581.2789861.54764140.88565
Maximum
Quantiles
.Control
.NoneES20JEOLSEMVisionSEMVision+ES20+JEOL
Level
218 12 32 36 36 30
Number
2617.51 0.99 0.94 1.03 1.04 4.53
Mean
38628.4 0.1 0.1 0.1 0.1 9.6
Std Dev
2616.2 0.0 0.0 0.0 0.0 1.8
Std Err Mean
-2539 1 1 1 1 1
Lower 95%
7774.0 1.0 1.0 1.1 1.1 8.1
Upper 95%
Means and Std Deviations
Oneway Analysis of 1.5V PSD (msrd stitch Rt/ calc'd stitch Rt) By SEM Split
Figure 15: Ratio of 1.5V PSD Measured Stitch Resistance to Calculated Theoretical Stitch Resistance Value by SEM Split
Wafer-1 in Red Wafer-2 in Black
3.3V
PS
D (
msr
d/ca
lc'd
)
1
10
.Con
trol
.Non
e
ES
20
JEO
L
SE
MV
isio
n
SE
MV
isio
n+E
S20
+JE
OL
SEM
.None
SEMVision+ES20+JEOL
.01 .05 .10 .25 .50 .75 .90 .95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
.Control
.NoneES20JEOLSEMVisionSEMVision+ES20+JEOL
Level
0.7816580.8929530.8008490.8560040.918126 0.86678
Minimum
0.9205690.8946620.8616960.9159510.9404290.998452
10%
0.9693760.9230980.8813520.9858210.967093 1.22921
25%
1.0281091.030149 0.944131.0597221.1137851.479641
Median
1.1636281.2213461.015173 1.213171.2191645.051506
75%
1.2765926.3742911.2604931.3136411.40435681.17646
90%
3.1036538.5329911.3193312.0047825.0060127742.935
Maximum
Quantiles
.Control
.NoneES20JEOLSEMVisionSEMVision+ES20+JEOL
Level
218 12 32 36 36 30
Number
1.090 1.675 0.979 1.114 1.223
355.191
Mean
0.23 2.16 0.14 0.21 0.67
1480.49
Std Dev
0.02 0.62 0.02 0.04 0.11
270.30
Std Err Mean
1.1 0.3 0.9 1.0 1.0
-197.6
Lower 95%
1.12 3.05 1.03 1.19 1.45
908.02
Upper 95%
Means and Std Deviations
Oneway Analysis of 3.3V PSD (msrd stitch Rt/ calc'd stitch Rt) By SEM Split
Figure 16: Ratio of 3.3V PSD Measured Stitch Resistance to Calculated Theoretical Stitch Resistance Value by SEM Split
Kelvin and van de Pauw window-1-to-gate data was not available for this
analysis. In considering the above four figures it becomes clear that for the stitches
which underwent only a single SEM inspection no disturbance in the electrical integrity
Wafer-1 in Red Wafer-2 in Black
of the windows took place. In general, the mean and median values for the ratio is
centered at unity plus or minus ~10%, and in most cases the 90th percentile is within 10%
of the expected value. In fact, for a majority of these stitches, the theoretical value
overestimates the actual measured resistance, and the ratio is less than one.
On the other hand, the window-1 to source/drain stitches that underwent SEM
inspection by all three SEMs had ratios that averaged in the 10’s to 100’s, which is of
course skewed by the high number of megohm resistances. Nevertheless, even the
median values ranged from +10% to +47%, the 75th percentile ranged in value from
+27% to 400%, and the 90th percentile ranged in value from +76% to 8000%. The above
four plots also reveal that the PSD stitches are more adversely affected by SEM
inspection than the NSD stitches. There does not appear to be any difference in
resistance distributions between the 1.5V and 3.3V PSD stitches or NSD stitches
The next four figures display the stitch resistance by the post-metal-1 treatment.
Wafer 01 received no post-metal etch clean prior to capping. Wafer 02 received an
ALEG310 clean prior to capping. The three control wafers, which did not receive any
SEM inspection, did not receive a post-metal-etch clean prior to capping.
N+_
CH
AIN
_1V
_Rt (
kOhm
s)
100
1000
.Control Aleg310 Skip
Cleans
.01 .05 .10 .25 .50 .75 .90 .95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
.ControlAleg310Skip
Level
171.5 185.9 187.5
Minimum
173.26 186.68 189.84
10%
177.35 196.4 195.7
25%
182.2 213.7 236.4
Median
188.95 242.9 291.8
75%
198.14 399.18
186199.4
90%
203.4 580.8
464940
Maximum
Quantiles
.ControlAleg310Skip
Level
45 15 15
Number
184.1 241.2
31227.9
Mean
8 97
119983
Std Dev
1 25
30979
Std Err Mean
182 187
-35216
Lower 95%
187 295
97672
Upper 95%
Means and Std Deviations
NSD_CHAIN_1V_Rt (kOhms) By Cleans, (Rows 8 and 9 only)
Figure 17: Resistance of Window-1 to 1.5 NSD stitch by Cleans Split (Data From Rows 8
and 9 only)
N+_
CH
AIN
_3V
_Rt (
kOhm
s)
100
1000
.Control Aleg310 Skip
Cleans
.01 .05.10 .25 .50 .75 .90.95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
.ControlAleg310Skip
Level 170.9 186.2 188.5
Minimum 173.16 187.16 189.88
10% 177.5 210.4 201.3
25% 183.3 223.5 233.2
Median 192.55 271.2 830.2
75% 202.86 1804.28 2400000
90% 2000000
3869 3000000
Maximum
Quantiles
.ControlAleg310Skip
Level 45 15 15
Number 44625 484
333665
Mean 298115
938 899604
Std Dev 44440 242
232277
Std Err Mean -44938
-36-164519
Lower 95% 134189 1003
831849
Upper 95%
Means and Std Deviations
NSD CHAIN_3V_Rt (kOhms) By Cleans, (Rows 8 and 9 only)
Figure 18: Resistance of Window-1 to 3.3V NSD stitch by Cleans Split (Data From Rows
8 and 9 only)
P+
_CH
AIN
_1V
_Rt (
kOhm
s)
1000
10000
.Control Aleg310 Skip
Cleans
.01 .05.10 .25 .50 .75 .90.95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
.ControlAleg310Skip
Level
468.6 536.4 578.2
Minimum
487.64 541.2 580.3
10%
492.7 614.1 908.6
25%
519.8 700.8 1241
Median
618.65 894.1 5487
75%
690.06 1807.4 30538.8
90%
785.9 2840 34254
Maximum
Quantiles
.ControlAleg310Skip
Level
45 15 15
Number
556.54 877.87 6006.23
Mean
84.5 566.7
10577.4
Std Dev
12.6 146.3 2731.1
Std Err Mean
531.14 564.01 148.67
Lower 95%
582 1192 11864
Upper 95%
Means and Std Deviations
PSD CHAIN_1V_Rt (kOhms) By Cleans, (Rows 8 and 9 only)
Figure 19: Resistance of Window-1 to 1.5 PSD stitch by Cleans Split (Data From Rows 8
and 9 only)
P+
_CH
AIN
_3V
_Rt (
kOhm
s)
1000
10000
.Control Aleg310 Skip
Cleans
.01 .05.10 .25 .50 .75 .90.95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
.ControlAleg310Skip
Level
466.7 560.2 582.8
Minimum
496.18 584.56 599.48
10%
510.65 607.8 796.8
25%
546.4 757.2 1002
Median
612.4 2353 5708
75%
675.72 844796 2421978
90%
1177 2000000 6000000
Maximum
Quantiles
.ControlAleg310Skip
Level
45 15 15
Number
573 139289 404301
Mean
111 515099 1548030
Std Dev
17 132998 399700
Std Err Mean
540-145963-452970
Lower 95%
606.83057424541.081261571.4
Upper 95%
Means and Std Deviations
PSD CHAIN_3V_Rt (kOhms) By Cleans, (Rows 8 and 9 only)
Figure 20: Resistance of Window-1 to 3.3V PSD Stitch by Cleans Split (Data From Rows
8 and 9 only)
VIA
1_C
HA
IN_R
t (kO
hms)
35
40
45
50
55
.Control Aleg310 Skip
Cleans
.Control
Aleg310
.01 .05.10 .25 .50 .75 .90.95 .99
-3 -2 -1 0 1 2 3
Normal Quantile
.ControlAleg310Skip
Level 38.56 41.53 43.14
Minimum 39.582 42.226 44.754
10% 42.11 45.32 47.38
25% 43.79 46.4 48.57
Median 44.24 47.27 50.39
75% 44.608 47.694
14424.78
90% 45.19 47.7
35986
Maximum
Quantiles
.ControlAleg310Skip
Level 45 15 15
Number 43.01 45.96
2444.05
Mean 1.80 1.80
9279.10
Std Dev 0.3 0.5
2395.9
Std Err Mean 42 45
-2695
Lower 95% 43.5 47.0
7582.6
Upper 95%
Means and Std Deviations
WN1-to-POLY CHAIN_Rt (kOhms) By Cleans, (Rows 8 and 9 only)
Figure 21: Resistance of Window-1 Poly Stitch by Cleans Split (Data From Rows 8 and 9
only)
The cleans split did not present a clear signal. There is some indication that
catastrophic fails were reduced by the ALEG bath. In general, the resistance were less
for the ALEG split than the skip clean split – this is especially the case for the 1.5V PSD
stitch (Figure 19) where the ALEG data showed a ~400kO reduction in resistance. The
data does not conclusively determine if the post-metal-etch cleans interacts with an SEM
damaged stitch to alleviate or exacerbate the damage. It certainly, however, provides a
signal that is worth investigating further.
To restate, it is evident that in- line scanning electron microscope imaging of
source/ drain contacts has modified the electrical characteristics of the window-1 contact
resistance, and caused the electrical failure of a significant number of those windows. In
addition, in- line scanning electron microscope imaging of gate contacts has modified the
electrical characteristics of the window-1 contact resistance, although it did not cause
catastrophic electrical failure those stitches. It remains to be seen however, which of the
four SEM scan points caused the source/drain stitch failure and what physical mechanism
was responsible for it.
4.1.2 Failure Mode Analysis
Voltage Contrast SEM imaging was attempted in order to isolate some of the
failing windows. This method has been used with great success in locating failing
windows between metal levels as well as windows to gate. The method used is as
follows. One of the two bond pads from a failing stitch is laser-blasted. This shorts the
metal-1 pad to the substrate, grounding one side of the stitch. The capping layer is then
removed using a CHF3 plasma etch, exposing the metal-1 pads. The die is then imaged in
the SEM using a low-voltage high-current beam. The grounded side of the stitch will
image bright up to the point of fail. Beyond the point of fail, the floating side of the
stitch will image dark (see Error! Reference source not found..) The identified failing
via is then typically cross-sectioned by focused ion beam (FIB), for SEM or TEM
imaging. Unfortunately, no voltage contrast imaging was detectable from the
source/drain stitches. It is speculated that the charge build up that should occur on the
floating side of the stitch is dissipated by the junction leakage of the N+ to Ptub or P+ to
Ntub junctions underneath the window-1 contacts.
Multiple random FIB cross-sections were then taken from failing stitches. The
location of the cuts were the four scan areas as detailed in Figure 6. A representative
STEM for each of the four scan areas on a failing PSD stitch can be seen in Figure 22
Figure 22: Scanning TEM Cross-Sectional Micrographs of the Four SEM Scan Areas From Electrically Open Failing Stitches
The random cross-sections did not identify either the failure locations or the failure
mechanism. In fact, cross-sections indicate well formed and filled windows, with no
apparent foreign interfaces, between silicon and liner, between liner and tungsten, or
between tungsten and Ti/TiN barrier.
Multiple failing stitches were then deprocessed by CHF3 plasma etch of the caps.
The metal-1 aluminum was then etched away in an HCl bath, and the Ti/TiN barrier was
polished off via CMP. Top-down SEM images of these failed stit ches were taken.
Figure 23: Top-Down SEM of the Four Scanned Areas On a Failing 3.3V PSD Stitch
After the Removal of Metal-1 – Tungsten Coring Evident In Scan Area #1
Figure 24: Top-Down SEM of the Four Scanned Areas On a Failing 1.5V PSD Stitch
After the Removal of Metal-1 – Tungsten Coring Evident In Scan Area #1
Figure 25: Top-Down SEM of the Four Scanned Areas On a Failing 3.3V NSD Stitch
After the Removal of Metal-1 – Tungsten Coring Evident In Scan Area #1
Figure 26: Top-Down SEM of the Four Scanned Areas On a Failing 3.3V NSD Stitch After the Removal of Metal-1 – Tungsten Coring Evident In Scan Area #1
Top-down SEM imaging of the failing source/drain stitches gives an indication
that the source of the electrical variation may be in SEM scanning of the oxide surface
(scan #1) prior to the window Ti/TiN liner. High magnification images of these windows
clearly indicate incomplete tungsten fill.
Figure 27: High Magnification SEM Images of Incompletely Filled Window-1 in Scan
Area #1.
In some instances, the area that received SEM scans after liner deposition (scan #2) also
showed some degree of incomplete fill. Scan areas #3 and #4 did not show any coring,
rather the windows appeared to be correctly filled. Because the precise location of the
failing window was not pinpointed, the incompletely filled windows must be considered
circumstantial, but they are at the very least an indication
As stated in the introduction, the physical failure mechanism will most likely fall
into one of three categories
1. The SEM deposits a thin hydrocarbon interfacial film that disrupts the electrical integrity of the thin films
2. SEM interaction disrupts the normal deposition of the thin films or the normal structural integrity of the thin films
3. The SEM injects a stored charge on the sample surface that upon subsequent processing, drives electrochemical/galvanic corrosion of one of the metal thin films.
The fact that source/drain stitches showed catastrophic fails and a wide
distribution of resistance whereas the poly stitches showed neither catastrophic fails nor a
wide distribution in resistances eliminates the first premise. A purely physical deposition
of a carbon ‘burn’ mark would not preferentially choose a patterned oxide over thinox
versus a patterned oxide over gate stack. Likewise a hydrocarbon interface between
Ti/TiN liner and tungsten or between tungsten and metal barrier, would not be expected
to preferentially deposit over a thinox stitch as compared to a poly stitch.
As to the second premise, thin film perturbations, several mechanism are
conceivable that could account for the observations. Source/ drain contact resistance
relies on the silicidation of the deposited Ti/TiN window liner to form an ohmic contact.
For the window to gate contact, the tungsten plug is landing on a deposited metal silicide
in the gate stack so the Ti-silicidation is not a requirement for good ohmic contact. A
hydrocarbon interface between silicon and liner could retard or prevent the silicidation
reaction of the source/drain reaction. This mechanism would not explain the evident
coring of the plugs in the Scan #1 and Scan #2 areas, since the tungsten nucleation and
deposition would not be disturbed. However, since the tungsten coring is circumstantial
the aforementioned mechanism can not be eliminated. Alternatively, a charge field
injected into the oxide by the SEM prior to Ti/TiN liner deposition could impact the
coherence of that liner. The impinging titanium is ionized, and therefore would be
subject to electrical fields associated with the surface. The sidewall thickness of the
window is on the order of only tens of angstroms, so even minor alterations of the liner
deposition could lead to discontinuous coverage over the sidewall. An incoherent liner
would impact the nucleation of tungsten, which would explain the cored plugs.
Additionally, hydrocarbon deposition on the surface of the liner could also disrupt the
nucleation of tungsten, which could explain the cored tungsten. The disruption of
tungsten nucleation, would not however, be expected to occur preferentially over thinox
as compared to poly.
Finally, the third premise appears quite feasible. As discussed in the literature
review, stored charge driven electrochemical corrosion has been reported for both the
tungsten plugs and the titanium metal barrier. In fact the observations made by Lee et
alError! Bookmark not defined. agree with the electrical findings of this study: plugs
electrically connected to the source/ drains failed whereas plugs terminating at gate did
not. The argument against this premise is that the findings of these reports generally
found completely voided windows, whereas the deprocessed stitches in this study showed
only slight perturbations if any at all. Even so, the third premise seems most likely in
light of the Lee finding.
4.2 Experiment 2 – Window-2 Interactions
The electrical resistance of the 2000 window-2 stitches and the single window
metal-2 to metal-1 are presented in the following two figures. In this experiment, wafers
were fully processed through metal-2, with a metal-2 misalignment. The SEM scans
were then only performed on the metal/exposed tungsten plugs, which corresponded to
SEM Scan #4 in experiment 1. One wafer each was run through an ALEG310 clean, an
NE111 clean, and a DSP+ clean, while one wafer skipped the cleans. The three control
cells did not receive any SEM inspection or cleans. The resultant data is displayed as a
variability chart with the SEM split within the metal cleans split.
82~V
IA2_
CH
AIN
_Rt (
kOhm
s)
9
10
11
12
13
14
15
16
17
18
.Con
trol
.Non
e
JEO
L
SE
MV
isio
n
SE
MV
isio
n+JE
OL
.Non
e
JEO
L
SE
MV
isio
n
SE
MV
isio
n+JE
OL
.Non
e
JEO
L
SE
MV
isio
n
SE
MV
isio
n+JE
OL
.Non
e
JEO
L
SE
MV
isio
n
SE
MV
isio
n+JE
OL
.Control ALEG310 DSP+ NE111 Skip
SEM within Cleans Figure 28: Variability Chart for Window-2 Stitch Resistance by SEM Split Within Cleans
Split
142~
RC
_MT
2/M
T1_
Rc
4.5
5.0
5.5
6.0
6.5
7.0
7.5
.Con
trol
.Non
e
JEO
L
SE
MV
isio
n
SE
MV
isio
n+JE
OL
.Non
e
JEO
L
SE
MV
isio
n
SE
MV
isio
n+JE
OL
.Non
e
JEO
L
SE
MV
isio
n
SE
MV
isio
n+JE
OL
.Non
e
JEO
L
SE
MV
isio
n
SE
MV
isio
n+JE
OL
.Control ALEG310 DSP+ NE111 Skip
SEM within Cleans Figure 29: Variability Chart for Window-2 Kelvin Contact Resistance by SEM Split
Within Cleans Split
Although the distribution is very tight and there is a large degree of overlap
between the datasets, the data for the 2000 window stitches does indicate that the
combined SEM split has the highest mean resistance within every clean cell. This trend
does not occur in the single window Kelvin measurements. This data however, does not
implicate the SEM inspection. The distribution for the control cell is greater than any of
the individual distributions. In addition, if the stitch resistances for only the control
wafers is plotted by row it is evident that what appears to an increased resistance of the
combined SEM cell is merely a processed induced phenomenon, exacerbated by the
small sample size of the data set from single wafers. Figure 30 shows that the average of
the stitch resistances increases with increasing row number in a standard processed wafer.
82~V
IA2_
CH
AIN
_Rt (
kohm
s)
9
10
11
12
13
14
15
16
17
18
1 2 3 4 5 6 7 8 9 10
.Control
.Control
Figure 30: Metal-2 to Metal-1 2000 Window Stitch Resistance By Wafer Row for the
Three Control Wafers Only
Recall that for Figure 28 and Figure 29, the control cell data is taken from all rows of the
control wafers, the None cell data is taken from rows 1,2,3,and 10, the JEOL cell data is
taken from rows 4 and 5, the SEMVision cell data is taken from rows 6 and 7, and the
combined SEM cell data is taken from rows 8 and 9.
While the various cleans did shift the average resistances of the stitches and
Kelvins, there was no SEM scan interaction with these cleans. Evidently then, SEM
scans by the two SEMs did not induce a shift electrical characteristics in the window-2
stitches or window-2 Kelvin regardless of the post-metal-etch processing scheme
employed.
CHAPTER 5 SUMMARY AND CONCLUSIONS
The interactions of SEM inspection in an IC processing environment were
investigated. A new failure mechanism by scanning electron microscope induced
electrical failure of a window-1 contact was reported. It was found that in the event of
multiple SEM inspections of a source/ drain contact at various points in the wafer
processing, the electrical resistance of the inspected structures increased dramatically - up
to and including electrical opens. Furthermore, there was some indication that a post-
metal-etch clean served to decrease the overall resistance of the SEM scanned stitches,
however the resistance did not return the level of the control data. The experimental
procedure used obeyed the normal production processing guidelines of a 0.16µm process
technology. This indicates a potential source of defectivity in high volume integrated
circuit manufacturing, as the use of SEM defect detection and defect analysis steadily
increases. In addition, it was found that multiple SEM inspections of a gate contact did
not cause catastrophic electrical fail but did cause an increase in the resistance of those
gate contacts, possibly by as much as 200O per contact.
Failure analysis by cross-sectional STEM and top-down SEM of the high
resistance source/drain stitches did not conclusively determine the failure mode. Some
indication of the source of the disturbance was indicated, however, by the presence of
partially cored out tungsten plugs in the areas that received SEM scans on patterned oxide
and on window liner. Some speculations on potential physical mechanisms for the
failures were given.
Finally, SEM scanning of exposed window-2 tungsten plugs underneath
intentionally misaligned metal-2, was found to not cause a change in the electrical
characteristics of those structures investigated. This conclusion holds true for wafers that
received post-metal-etch cleans after SEM scans that were acidic, alkaline and neutral, as
well as structures that received no clean at all.
LIST OF REFERENCES