Breakthrough in Quality management

Dan Glotter

CEO & Founder OptimalTest

Devices’ Quality drive profitability

Big electronics players changed the game rules

Consumer mass production products like cell phones,

Pads, laptops use dual strategy ingredients

Samsung S3 smart phone has 2 design wins for its cpu –

nVidia’s Tegra & Qualcomm’s Snapdragon

Their BoM price is similar – what differentiates sellable

quantities is Quality levels (RMA#s)

Excelling in quality became a big deal and a bottom line

revenue differentiator

P#2

Trends driving quality attention Wafer level packaging

(WLCSP | WCSP | WLP | WLBGA)

For the last few years new Wafer Level Packaging technology is being

used for many products like mobile phones, PDAs, laptop PCs, disk

drives, digital cameras, MP3 players, GPS etc

After the FAB process the wafer goes through those steps:

wafer bumping, wafer level test, back grind, dicing, and

packing in tape & reel to support a full turn-key WLCSP solution

That means that THERE IS NO MORE FT/BI/SLT OPERATION !!!

The tape is going directly to the customer after wafer sort for board

level mounting… (i.e to Apple, Samsung, dell Sony etc)

Thus, Quality becomes critical since there is no other gate keeper

P#3

The need

Handling potential escapes & outliers, requires a comprehensive

system which covers the end-to-end supply chain:

Analysis and simulation tools to evaluate potential escapes and

outlier algorithms on historical data

Rule generation and publication processes to deploy escape

prevention and outlier rules at test houses

Execution of the escape prevention and outlier rules on

OptimalTest's servers once testing is completed anywhere in the

supply-chain

Fully integrated and automated modification of inkless bin maps for

assembly or in Final-Test anywhere in the supply-chain

Monitoring and feedback tools to track the actual performance of the

escape prevention and outlier detection

P#4

Tight quality “safety net”

The solution should be an Escape Prevention Solution (EPS)

enables a tight safety net that becomes the “escape gate-

keeper” in any of your testing operations

The solution should offer Fabless or IDM Business-Units the

ability to create & activate rules vis-à-vis their Foundry, OSAT

or IDM Factory -The rules should be executed through a

integrated supply chain infrastructure to provide full Quality &

health control

P#5

– Strategic supplier of top Fabless, IDM & OSAT

– Installed across whole world-wide Fabless/Foundry/OSAT supply chain at

-

Some facts… about OptimalTest

#1 Fabless #2 Fabless #3 Fabless

#5 IDM #3 OSAT

2013: ~3,300 testers - over 25M units run on OT per day… ~10B per year…. and growing

#4 Fabless #5 Fabless

P#6

Escape Prevention Solution

OptimalTest’s Escape Prevention Solution consist of the

following elements:

RMA database for thorough management of the escapes

3 families of Outlier Detection capabilities for Wafer Sort &

Final Test: Parametric , Geographical & Cross-Operational

OT-Detect: an excursion prevention system that

automatically tracks after ALL your products for ANY

changes in BASELINE production (HB, SB, Params)

Dozens of unique algorithms that were “created with blood”

following many escapes and thorough RMAs analysis

OptimalTest is the only Outlier Detection provider that has an

infrastructure embedded into the Foundries & OSATs operation

(like TSMC & ASE)

P#7

WS - Parametric Outlier Detection

These algorithms detect outliers based on the behavior

of specific parametric tests.

DPAT: "Dynamic Part Average Testing" is a standard industry

algorithm for outlier detection which captures every die with a

parametric characteristic falling outside of a statistically calculated

boundary.

NNR: "Nearest Neighbor Residual" is the best algorithm to use for

avoiding yield overkill caused by Fab-related geographical differences

(Center – Edge & Reticle locations) and Influence of intrinsic Test site

differences.

The data is "smoothed" to eliminate peaks and then NNR automatically

applies different coefficients to reticle locations and test sites if the

algorithm reveals a significant difference between groups.

• It can also use "bivariate" tests - virtual tests created as a

regression of the two real parametric tests.

P#8

WS - Geographic Outlier Detection

These algorithms detect outliers by analyzing the location

of the die on the wafer and the die's neighbors:

Z-PAT: "Z-Axis Part Average Testing" Looks at dice in the same X,Y

coordinates across multiple wafers in a lot to "kill" dice in locations that

fail too frequently

GDBN: "Good Die in Bad Neighborhood" calculates the yield of the

neighboring die for each good die; the die surrounded by a cluster of

failing neighbors is removed based on a weighting algorithmic recipe

Bad Reticule Detection: "Bad Reticule Detection" captures

specific reticule X, Y locations which have low yield in the current lot

Zonal: "Bad Zonal Detection" captures a specific zone with low yield

in the current lot

P#9

Final Test - Parametric Outlier Detection

OptimalTest’s Outlier Detection for Final Test is based on

2 type of algorithms:

1) Post Final-Test operation and Based on Die-ID

(ULT/OTP/ECID)

a) Option a: Next Operation execution (i.e SLT or WH)

b) Option b: FT-PAT operation (Short TP that reads only Die ID)

2) In real-time at Final-Test operation without Die-ID the downside of this method is the outlier baseline statistical size

P#10

Cross Operational Outlier Detection

– Cross operational Quality based on Die-ID

– Contributing operations

• ETEST/PCM/WAT

• Wafer Sort

• Final-Test

• Burn-In

• System Level Test

• Example: E-Test based bin switching post WS The ability to identify potential bad devices based on E-test

data geographical analysis – The bin switching post wafer sort

- Requires data feed forward within the supply chain .

P#11

RMA Database

The new RMA Database will provide detailed information about parts returned from customers.

– Data Entry: Users can identify parts by ECID and mark them as returned in the database, together with categorization data.

– Data Retrieval: The RMA database is searchable and is summarized in standard summary tables so that information about RMA’s can be analyzed in OT-Portal.

– Historical Analysis: Lots containing parts which are returned are flagged in the database as “unpurgable”. It impacts all operations in which the part or wafer was tested. Cross operation reports can be used to analyze the cause of the failure.

P#12

• Probe mark tracking

The algorithm tracks probe marks per each die at wafer sort and compares with

a spec value. The rule takes into account restests & multiple operations as well

as “hidden” probe marks in parallel testing when dice are touched but not tested.

Example of Escape Prevention Rules

P#13

• Good die/device with “out of spec” test results

The algorithm catches parametric tests which are marked as “pass” despite having a result

which is out of spec limits.

• Failing tests in good parts

Except for some specific cases, tests should not fail in good units. This rule checks that no

failing tests matching a specified signature exist in a good part.

• PRR validation (Part Results Record)

For each good die/device, the rule checks that the number of tests reporting in the PRR

records for good parts is above spec. This rule can use a baseline to calculate the limit.

• ULT validation

For products identified as ULT enabled, the rule will check each good die/device for ULT

value (ULT = OTP/ECID)

• Freeze detection

The rule can monitor selected critical tests and detect freeze cases by comparing

parametric test values across multiple devices.

• Parametric trend

The rule can monitor selected critical tests and detect statistical trend patterns in the test

results that may indicate process quality issues .

• Process capability (CPk)

The rule can monitor selected critical tests and detect CPk related abnormal behavior in

the test results that may indicate process quality issues.

Example of Escape Prevention Rules

P#14

• OT-Detect automatically tracks after ALL

your products for ANY changes in

BASELINE production

• Description: These statistical rules allows the

user to detect issues through monitoring

baselines of “too Bad” or “too Good”

performance that are statistically suspicious

e.g. Yields, Bin & S-Bins occurrences (SBLs) &

Parametric Tests (STLs) etc…

• Once triggered it provides a step-by-step

ROOT-CAUSE ANALYSIS. This means that

any extreme change in the products'

manufacturing, test or assembly processes will

be tracked, captured and assessed.

Leveraging massive data: OT-Detect

Lot level

Analysis

Prod level

Analysis

Bin level

Analysis

Param level

Analysis

Equp level

Analysis

Facility level

Analysis

Note: The ability to do so many baselines on the fly is a huge technological breakthrough since it requires optimized algorithms to enable super fast computations

What is a “Baseline”?

Preforming a “baseline” means that the system automatically identify the incoming product, scan the last 20-40 lots of that product and determine if the current lot signature significantly exceeds the value of the historical baseline that was created on the fly - Either in Real-Time or in Off-line

15

P#15

OptimalTest Escape Prevention

Its time to check

if “good”

is really good ?

Thank you !