A Comparison of Wafer Retest Methods: In vs. Post sort ...
Transcript of A Comparison of Wafer Retest Methods: In vs. Post sort ...
A Comparison of Wafer Retest Methods: In‐Situ vs. Post‐sort SelectiveMark Banke, Altera Corp, [email protected]
1Silicon Valley Test Conference 2010
Agenda
• What is Wafer Retest, and Why do We do It?• What are the Types of Wafer Retest?• Wafer Retest Attributes• Software Issues Involving Retest• Results from Altera Wafer Retest Data• Cost Savings Potential• Retest Yield Distribution• Rule Based Retest• Summary
2Silicon Valley Test Conference 2010
What is Wafer Retest, and Why do We do it?• The semiconductor business is driven by yield:
• Testing doesn’t always yield the expected amount on the first pass.
• To reclaim possible false failures, wafers may be retested.– False Failure: Dice which fail on the first test, but pass on subsequent retest(s). Can be caused by test hardware, test program, wafer / probe contamination, etc.
• The result is money is reclaimed from the scrap bin.
3
Yield $
Silicon Valley Test Conference 2010
What are the Types of Wafer Retest?
4
•Whole Wafer Retest
•In-situ and Selective Wafer Retest
Whole Wafer: Retest every die after complete lot has finished test
In-situ: Retest only specific dice immediately after whole wafer test
Selective: Retest only specific dice after lot has finished test
Silicon Valley Test Conference 2010
Wafer Retest Attributes
• Whole Wafer:– Verifies all dice. May eliminate false passes and false fails due to test program
issues, setup error, hardware problems
– No additional test software – just retest the wafer and nullify past results
– Increases chance for bond pad damage
– Increases sort cycle time
5Silicon Valley Test Conference 2010
Wafer Retest Attributes (continued)
• In‐situ:– Only retests failing dice (or specified dice based on whole wafer binning)
immediately upon whole wafer completion
– Saves on setup time, reduces setup error frequency, catches most of the false failures
– May not catch gross failures due to initial setup errors.
– May subject passing dice in multi‐site probing setup to more probe damage
– Requires more complicated software
6
These passing dice get extra probe
contact
Silicon Valley Test Conference 2010
Wafer Retest Attributes (continued)• Selective:
– More likely to catch gross failures due to setup.– Single site probe card less likely to cause damage to adjacent dice
– Effective for testing dice at non‐yielding steps– Increases number of setups, may increase setup errors– May increase test cost– Increase in test time – tester controlled prober indexing
– Possible increase in cycle time– Requires more complicated test software
7Silicon Valley Test Conference 2010
Software Issues Involving Retest
• In‐Situ: Matching summaries. Total die qty passing on retest needs to be added to original passing die total, but total tested shouldn’t be incremented.
8
W1F W1I W2F W2I W3F W3I Grand TotalTotal
Tested XXX 65 XXX 70 XXX 48 XXXX
TotalPassed
YYY 3 YYY 2 YYY 0 YYYY
TotalFailed
65 62 70 68 48 48 361
Grand Total Tested = only sum of “F” Totals
Grand Total Passed= Sum of all Passing
Why? Need to keep accurate yield data!
Yield based on Total Passed / Total Tested (Full Wafer Dice)
Silicon Valley Test Conference 2010
Software Issues Involving Retest (continued)
• Selective: Summary data must be downloaded from server to tester so only specified failure dice are re‐tested
9
W1 FS
W2 FS
W3 FS
Grand Total
TotalTested
XXX XXX XXX XXX
Total Passed YYY YYY YYY YYYTotalFailed
46 63 44 153
W1 SR
W2 SR
W3SR
Grand Total
TotalTested
46 63 44 153
TotalPassed
5 25 10 40
TotalFailed
41 38 34 113
Selective Retest Summary
Full Sort Summary
Final Yield = (Total Passed Full + Total Passed Selective) / Total Tested Full
Silicon Valley Test Conference 2010
Software Issues Involving Retest (continued)
Wafer OCR / Barcode reader must be used ‐ prevent wrong wafer associated with Retest data
Tester must gather summary data from network so it can direct prober to re‐test die locations.
10
TesterProber
Network Server
Silicon Valley Test Conference 2010
Which retest method provides the best compromise between efficiency and yield?
11
In-Situ Vs SelectiveIn‐Situ Selective
Silicon Valley Test Conference 2010
Results from Altera Wafer Sort Data
Total Wafers
Avg Retest Recovery Yield
Total Unique Wafers: 9428 ‐# In‐situ Recovery
"Winners" 5439 9.0%# Selective Recovery
"Winners"
1091 4.6%
No Die Difference 2898 ‐
12
Over 6 Million Dice Tested
Silicon Valley Test Conference 2010
In‐Situ vs. Selective Retest :Difference of Die Yield Reclaimed
0.0%5.0%
10.0%15.0%20.0%25.0%30.0%35.0%40.0%
1.0%2.7% 4.1% 5.3%
7.9% 8.4%
27.4%
20.7%
35.6%
21.6%
13.3%
6.8%
0.4%
% o
f Waf
ers
Test
ed
% Reclaim Yield Difference by Retest Method
13Silicon Valley Test Conference 2010
In‐Situ vs. Selective Retest :Difference of Die Yield Reclaimed
14
0%
5%
10%
15%
20%
25%
30%
35%
1% 1% 0% 1% 2%
7%
31% 31%
13%
6%
2% 2% 3%% o
f Waf
ersT
este
d
Difference in Dice Reclaimed by Retest MethodSilicon Valley Test Conference 2010
In‐Situ vs. Selective Retest :• So, looks like the winner is In‐situ…
• Upon “further review…”– It’s not necessarily the Retest method recovery difference, it’s the amount of dice actually recovered because of the Retest…
15
?
Silicon Valley Test Conference 2010
In‐Situ vs. Selective Retest :
Total Wafers
Avg Retest Recovery Yield
Additional Recovery Dice
Total Unique Wafers: 9428 ‐ ‐# In‐situ Recovery
"Winners" 5439 9.0% 50,014# Selective Recovery
"Winners"
1091 4.6% 21,199
No Die Difference 2898 ‐ ‐
16Silicon Valley Test Conference 2010
In‐Situ vs. Selective Retest :Difference of Die Yield Reclaimed
0%
5%
10%
15%
20%
25%
30%
35%
1% 1% 0% 1% 2%
7%
31% 31%
13%
6%
2% 2% 3%% o
f Waf
ersT
este
d
Difference in Dice Reclaimed by Retest Method
What about these?
17Silicon Valley Test Conference 2010
In‐Situ vs. Selective Retest : Difference of Die Yield Reclaimed – by Product Family
0%
10%
20%
30%
40%
50%
60%
1% 0% 0% 0% 1%4%
55%
30%
7%
2% 0% 0% 0%
1% 1% 1% 2% 3%
13%
19%
42%
12%
4% 2% 0% 0%% o
f Waf
ers
Test
ed
Difference in Dice Reclaimed by Retest Method
Low Cost Families Medium Cost Families High Cost Families
18Silicon Valley Test Conference 2010
Cost Savings Potential
• Possible savings may accrue if the test cost model charges extra for selective retest setup time.
19Silicon Valley Test Conference 2010
Cost of Setup Time vs # of Setups with 10 minute Average Setup Time
$0
$5,000
$10,000
$15,000
$20,000
$25,000
10 20 50 75 100 200 500 1000
Cos
t of S
elec
tive
Ret
est S
etup
Ti
me
# of Setups
$50.00 $75.00 $100.00 $125.00
20
Test Cost $ / Hr
Silicon Valley Test Conference 2010
Cost of Setup Time vs # of Setups with 10 minute Average Setup Time
$-
$2,000
$4,000
$6,000
$8,000
$10,000
$12,000
$14,000
$16,000
$18,000
$20,000
1 Unit 50 Units
100 Units
200 Units
300 Units
400 Units
500 Units
750 Units
1000 Units
Pote
ntia
l $ V
alue
of U
nits
not
R
ecov
ered
Number of Units Not Recovered
Number of Units vs $ Cost of Non-Recovered Part
$1 $5 $10 $35 $50 $100 $500
Selective Retest Setup Costs @ 500 Setups
21Silicon Valley Test Conference 2010
Cost Savings Potential with In‐Situ Retest
# of Retest Units
22
Setup Cost
Unit Cost
IF….
Silicon Valley Test Conference 2010
Cost Savings Potential with In‐Situ Retest
• The results favored in‐situ Retest on average when average yield from first initial test was greater than certain yield thresholds, but selective sort proved necessary when problematic test setup conditions occurred.
23Silicon Valley Test Conference 2010
In‐Situ Recovery Rate vs Increase over Yield Threshold
24
2% 1% 1% 3% 3%
2017
23
25
6
0
5
10
15
20
25
30
0%10%20%30%40%50%60%70%80%90%
100%
A B C D E
# of Dice R
ecoverd by Selective Retest
% o
f Waf
ers
Test
ed
PRODUCT
% Wafers with No Selective Resort Dice Recovered% Wafers with Selective Resort Dice Recovered% > Yield Threshold# of Lost Dice if Selective Resort Not Used
Silicon Valley Test Conference 2010
Retest Yield Distribution
• Insitu Retest does in fact, on average, increase yield vs selective Retest. But the data suggest that one pass Retest may not always be enough to capture all available yield.
25Silicon Valley Test Conference 2010
Distribution of Additional Dice Recovered
0%
10%
20%
30%
40%
50%
60%
70%
0 < 5 5 to 9 >= 10
26% 25%
8% 9%
64%
24%
5% 8%
% o
f Tot
al W
afer
s Te
sted
Additional Dice Recovered
1st In-situ Resort AttemptAll Selective Resort Attempts
26Silicon Valley Test Conference 2010
Distribution of Additional Dice Recovered Per Retest Attempt
0%5%
10%15%20%25%30%35%40%45%50%
0 < 5 5 to 9 >= 10
26% 25%
8% 9%
48%
16%
2% 2%
16%
7%3%
5%
% of Total W
afers Tested
Additional Dice Recovered
1st In‐situ Resort Attempt
1st Selective Resort Attempt
2nd Selective Resort Attempt
3rd Selective Resort Attempt
27Silicon Valley Test Conference 2010
Rule Based Retest• In‐Situ Retest will recover the majority of false failures. But there are cases where Selective Retest should be used.– Rule Based Retest
28
Type of Part Condition Action
Any 1st Pass Yield << Std(Bad Setup)
Selective Retest
Low Cost Die 1st Pass Low Yield In‐Situ Retest
Long Test Time Particular Failure Type Selective Retest on ParticularFailure Dice
High Cost Die 1st Pass In‐Situ Retest, thenSelective Retest on Sample Wafers. If yield improvement, Selective Retest on all wafers
Silicon Valley Test Conference 2010
Summary
• Both In‐Situ and Selective Retest Methods are Viable
• In‐Situ will likely Recover most invalid failures, usually on the first attempt
• Rule‐based Retest will likely provide the most cost effective solution
29Silicon Valley Test Conference 2010
Thank you!
• Questions?
30Silicon Valley Test Conference 2010