Key Methods in Reducing Pad Crack Risk at Probing Low-k Wafers

Key Methods in Reducing Pad Crack 

Risk at Probing Low-k Wafers 

Taiwan Semiconductor 

Manufacturing Company, Ltd. 

Frank Hwang 

(E-mail:sxhwang@tsmc.com) 

Steve Hsu 

J. H. Chen 

MJC Probe Inc. 

Dean Yang 

(E-mail:dean.yang@mpi.com.tw) 

Wensen Hung 

Jacky Tsai 

1

Agenda 

• Probing Challenge on Low-k Devices 

• Scrub Depth Correlates with Underneath Layers 

• Scrub Depth Model Formulation (SDMF) 

• Review -2005 

• Experiment and Result 

• Constant Parameter Calculation 

• BCF Measurement Limitation 

• Conclusion 

2

Probing Challenge on Low-k Devices 

1 st : Aluminum Tier 

Probing contradictory! 

2 nd : TaN 

Low Probe Force 

VS 

Contact Res. 

4 th : Oxide 

Low-k Structure 

3 rd : Copper 

Low-k experimental wafer – Pad matrix cross section 

Tighter BCF Spec Window 

VS 

Analyser Accuracy 

Contact Force Uniformity 

VS 

Manufacture Deviation 

Multiple Touchdown 

VS 

Pad Crack Risk 

3


What Should Be Concerned on Low-k? 

• Except Pad Void (PV), what risks will be suffered when 

probing Low-k wafer? 

• In comparison with the low-k defects, It’s “lucky” to suffer 

PV, because of the observable defects where after 

aluminium layer removal, copper is physically exposed on 

TaN surface. 

• How about the scrub below? 

It’s OK or NG? In fact, microscope and wafer inspector 

show you “No PV.” 

Al Remove 

4


Hidden Underneath Layer Deformation 

• No PV ≠ Free Damage 

BCF:4gw/mil Tip Dia.=8um 

OD= 45μm Probe:6 times 

Cross 

Section 

Al 

BCF:4gw/mil Tip Dia.=14um 

OD=45μm Probe:6 times 

Deformation 

Cu 

Serious Destruction 

5


Initial Probing Damage 

• After Al was removed, we found micro scratches 

and cracks as below images: 

OD= 65μm 

TD=6 times 

Tip Dia.=8μm 

BCF=4gw/mil 

Scrub 

direction 

Slight Medium Serious 

• Evaluation showed the probability of probing damage: 

TaN Crack > Underlying Deformation > Pad Void 

• Safe probing method is to prevent TaN-crack! 

6

Scrub Depth Correlates with Underneath layers 

Underneath Layer Evaluations 

• Measurements identified underneath layer deformation risk was 

at stake. 

Unacceptable! 

Scrub Depth(um) 

1.8 

1.6 

1.4 

1.2 

1 

Scrub Depth of 4gw/mil Probe 

○PV Found by microscope 

TD 2 2 2 2 4 4 4 4 6 6 6 6 

OD(um) 45 55 65 75 45 55 65 75 45 55 65 75 

Depth(um) 1.2 1.4 1.4 1.3 1.4 1.5 1.4 1.4 1.5 1.6 1.6 1.5 

Parameters 

7


Acceptable Scrub Depth Region 

• Monitor the TaN layers of shallow scrubs. 

Acceptable 

Dangerous 

Smooth Smooth Smooth Wavy 

Scrub Depth 

30% of α 

Scrub Depth 

54% of α 

Scrub Depth 

60% of α 

Scrub Depth 

86% of α 

α= Thickness of Al Layer 

8


Scrub Depth Control is Necessary 

Low-k devices are highly sensitive to probing force issue, more 

severely is the unobservable physical damage inside the pad. 

Hidden damage not seen at wafer inspect after probing, but 

identified at test failure during packaging/final test. 

Experiment and evaluation works indicated the safety band 

probing depth region is fallen below 60% of total Al thick. 

Efficient methodology to control the scrub depth is proposed by 

monitoring the Kyy as the primary dominant factor. 

Continuing the last year presentation on SWTW 2005, the 

proposed SDMF will be demonstrated again as an effective 

backend assessment methodology to prevent the probing 

damage. 

9

Scrub Depth Model Formulation (SDMF) 

PV Case in TSMC 

• Problem description 

• Pad void by 1 st layer needle 

2005 SWTW 

Repeated PV patterns 

10


Primary Factors of Experiment I & II 

By choosing all critical parameters, a two-level L8 orthogonal array 

experiment I has been performed, the influential factors have been 

determined as follow: 

Primary dominant factors tip length, tip diameter 

Secondary dominant factors stiffness K yy , needle diameter 

From TSMC mass production testing, three critical parameters were 

chosen to perform experiment II with a L9 three-level setting. The 

summarized results are: 

Primary dominant factors tip length, stiffness K yy 

Secondary dominant factors tip diameter 

The slight variation in results of these two experiments, it was 

recognized that these experiments still had uncontrolled noise. 

2005 SWTW 

It is concluded that these two experiments indicated that tip length, 

tip diameter, stiffness K yy were the three most influential primary 

parameters. 

11


Theory, Experiment and Verification 

• Constant values B & C were 

found from curve fitting. 

U 

z 

= 

= 

C 

C 

F 

y 

D 

2005 SWTW 

D 

( BK + K ) D 

yx 

yy 

y 

1200 

1000 

Curve Fitting Result 

Unit : 

U 

z 

: nm 

800 

Uz (nm) 

600 

400 

200 

0 

R 2 = 0.835 

K 

D 

yx 

y 

,K 

: mil 

D :μm 

yy 

: gw/mil 

0 0.5 1 1.5 2 

Fy/D (gw/um) 

12


TheoryⅡ, ExperimentⅡ, and VerificationⅡ 

• Experiment background 

• Applicability of the model is evaluated. 

• Two more parameters included: 

(1) Three most commonly used prober machines 

(2) Different needle diameter (4 mils) 

• Prober set-up based on TSMC production used methods. 

• Results 

Prober 

Probe 

Type 

OD 

(μm) 

Tip Dia. 

(μm) 

Tier 

Stiffness-Kyx 

(gw/mil) 

Stiffness-Kyy 

(gw/mil) 

UF200 

UF3000 

TEL P12 

40 

60 

75 

8 

13 

1-3 

1st: 7.28 

2nd: 5.23 

3rd: 4.19 

2 

Additional set up notes:. undershoot at UF prober is 25 um, while on TEL P12, double 

touchdown function is activated. 

13



• R-square was in high agreement for UF200 

◆ Actual Measurement data 

— Fitting Curve 

14



• UF3000 also showed appreciable agreement 



15



• R-square showed lower fitting agreements. 

• Prober chuck movement mechanism was attributed as major 

factor in the result variation. 

• Initial guess is TEL P12 having deeper probing height than 

UF families. 



16



• SDMF – Constant parameter calculation 

【UF200】 

Two normal TDs scrubbed 

2~22% deeper than single 

TD. (see blue vs. red line) 

25um undershoot TDs has 

15~30% deeper scrub 

mark than the nonundershooting 

one. 

(see green vs. red line) 

17




【UF3000】 

Two common TDs 

generated 25~45% deeper 

scrub than with only 1 TD. 

(see blue vs. red lines) 

Activating undershoot 

25um, scrubs became 

0~15% deeper than non 

activated one. 

(see green vs. red lines) 

18




【TEL P12】 

“TEL double touchdown” 

function physically differs with 

UF’s “undershoot function.” 

Two normal TDs’ scrub marks 

were 4~9% deeper than the 

one by single TD. 

(blue vs. red) 

After activating “double 

touchdown”, scrub is 

increasingly 8~17% deeper 

than non-activated one. 

(green vs. red) 

19


Summary 

SDMF results again showed the high level of quantitative 

prediction agreement with the corresponding experimental 

measurement data. 

“Linear Scrubbing / Slope Scrubbing” based assumption of SDMF 

is theoretically and experimentally proven to be capable of 

predicting the scrub depth the complex scrubbing action. 

Constant parameter modification factors of prober including setup 

and multiple TDs functions, still need further statistic sampling 

data to obtain accurate results. 

Measurement errors existed in the experimental data is still 

acceptably tolerable as to be used in engineering level application. 

Production data feedback is always an on-going process for 

better modification results of certain constant values of the model. 

20

BCF Measurement Limitation 

Analyzer, Statistic, Experiment 

• BCF control is unavoidable for low-k probing 

• How to address a common BCF definition by suppliers 

and vendors ? 

• Currently the industry available BCF measurement 

tooling (ex. Equipment A) showed spec as below: 

Simply states: 

With measured value 1gw/mil, it will have confidence intervals 99.7% 

that the deviation range should fall from 0.75~1.25 gw, also denoted as (1gw ± 25%) 

21


Analyzer, Statistic, Experiment 

12% 

(1)BCF Spec:4gw/mil ±20% 

(2)Accuracy:±0.25gw @3σ 

(1)BCF Spec:1gw/mil ±20% 

(2)Accuracy:±0.25gw @3σ 

10%-offset BCF Measure 

100% confidence within Spec. 

10%-offset BCF Measure 

88% confidence within Spec. 

As metrology accuracy has been pushed to its limit, allowable manufacture deviation suffers 

more and more tighter tolerance. New BCF metrology platform is urgently needed. 

22


Analyzer, Statistic, ExperimentⅠ 

• At low BCF values, measurement accuracy is degraded to 

marginal range 

Average Standard Deviation: 

σ Ave 

-TSMC is 0.065 

σ Ave 

-MPI is 0.075 

BCF under 3σ confidence interval: 

TSMC≒μ±0.195 4 gw/mil ±5% 

MPI ≒μ±0.225 4 gw/mil ±5% 

Average Standard Deviation: 

σ Ave 

-TSMC is 0.033 

σ Ave 

-MPI is 0.025 

BCF under 3σ confidence interval: 

TSMC≒μ±0.1 1 gw/mil ±10% 

MPI ≒μ±0.075 1 gw/mil ±7.5% 

Accuracy is 

getting worse! 

23


Analyzer, Statistic, ExperimentⅡ 

Procure a qualified 

low-k probe card. 

Only 85% could 

satisfy the spec- 

1gw/mil±20% 

Carried out 5 repetitive 

BCF measurements 

from selected 50 pins. 

For required spec 

1gw/mil ±20%, sigma 

must at

Conclusion 

• High density of mechanically weak structural layers of IC pad 

introduced in low-k wafers will increase the pad crack possibilities. 

• Now TSMC keeps practicing the SDMF as the standard guideline 

for monitoring and controlling the probing scrub marks in 

achieving the robust wafer sort. 

• SDMF results can also be further implemented into probe card 

design in order to obtain acceptable probe depth. 

• SDMF was again validated experimentally under consideration of 

more complete practical probing parameters. However, prober 

set-up (particularly chuck movement) still considered as important 

factor. 

• Available BCF metrology is recently at its bottleneck limit, for 

low-k card, new enhancement tooling to obtaining accurate 

measurement is under request. 

25

Q&A 

26

Key Methods in Reducing Pad Crack Risk at Probing Low-k Wafers

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?