confidential and proprietary - nxp · pdf filetm freescale, the freescale logo, altivec, c-5,...

TM

Freescale, the Freescale logo, AltiVec, C-5, CodeTEST, CodeWarrior, ColdFire, C-Ware, t

he Energy Efficient Solutions logo, mobileGT, PowerQUICC, QorIQ, StarCore and Symphony

are trademarks of Freescale Semiconductor, Inc., Reg. U.S. Pat. & Tm. Off. BeeKit, BeeStack,

ColdFire+, CoreNet, Flexis, Kinetis, MXC, Platform in a Package, Processor Expert, QorIQ

Qonverge, Qorivva, QUICC Engine, SMARTMOS, TurboLink, VortiQa and Xtrinsic are trademarks

of Freescale Semiconductor, Inc. All other product or service names are the property

of their respective owners. © 2011 Freescale Semiconductor, Inc.

Confidential and Proprietary

TM

Freescale, the Freescale logo, AltiVec, C-5, CodeTEST, CodeWarrior, ColdFire, C-Ware, the Energy Efficient Solutions logo, mobileGT, PowerQUICC, QorIQ, StarCore

and Symphony are trademarks of Freescale Semiconductor, Inc., Reg. U.S. Pat. & Tm. Off. BeeKit, BeeStack, ColdFire+, CoreNet, Flexis, Kinetis, MXC, Platform in a

Package, Processor Expert, QorIQ Qonverge, Qorivva, QUICC Engine, SMARTMOS, TurboLink, VortiQa and Xtrinsic are trademarks of Freescale Semiconductor, Inc.

All other product or service names are the property of their respective owners. © 2011 Freescale Semiconductor, Inc.

2


• eSys (130nm Hip7NVM) products have been in production for

many years. As the 0.13um Hip7NVM process becomes more

and more mature and defectivity (D0) gets lower and lower, it

enables us to reduce the Burn-In durations via stress

optimization to further improve manufacturing throughput without

impacting the baseline PPM performance.

• Although current production Burn-In is stated to be 14 hours, the

actual stress duration is only about 10 – 11 hours while the rest is

buffer time for slowest parts in NVM W/E cycling.

• The key data-driven optimizations to reduce the Burn-In duration to 5 hours: •NVM write/erase cycle reduction from 1500 cycles to 100 cycles (~2.5 hours saving)

•NVM bulk read stress to replace HTOL flash reads (~2 hours saving)

•Scan and BIST duration reduced (~1.5 hour savings)

• Qualification plan: • Qual vehicle / driver product:

•100K production reject analysis to confirm no risk from the key updates

•10K validation flow to validate the Burn-In implementation and no additional fallout from additional Burn-In

•Derivative products:

•Qual-By-Similarity (QBS) to the driver product, with IP gap analysis

TM





3


Qual Vehicle

• MPC5565

Derivative Product

• MPC5554, MPC5566, MPC5553, MPC5533, MPC5534, MPC5567, MPC551x

TM





4


Tests in Burn-In (TIBI)

RAM BIST

Current: 60 minutes; Reduce to 20 minutes

Non-NVM Module Exercise

NVM High-Voltage Stresses

Post-Stress TIBI

SCAN Stress

Current: 60 minutes; Reduce to 20 minutes

NVM Write/Erase Cycling

Current: 1500 cycles (3 hours); Reduce to 100 cycles (30 minutes)

NVM Read Disturb

Current: 2 hours; Reduce to 12 minutes (duty cycle improvement)

TM





5


• All Hip7NVM devices run 1500 Write/Erase(W/E) cycles

• With high-voltage (HV) erase stress prior to W/E cycling, most

cycling related defects are screened by HV stress.

− Reducing R/W cycling to 100 cycles on all blocks.

• Validated by

− reviewing data logs of Burn-In W/E cycling failures from 6 months

production data of MPC5565 (~895k units)

All units with valid Burn-In cycling fail code failed at 1st cycle

− collecting and verifying Burn-In W/E cycling failures from 100k

MPC5565 (eSys) devices.

No W/E cycling failures beyond 100 cycles – all failed within 30 cycles.

TM





6


-0.02%

0.00%

0.02%

0.04%

0.06%

0.08%

0.10%

Burn

in 1

0H

V W

E (

cycle

1)

Burn

in 1

0H

V W

E (

cycle

2)

Burn

in 1

0H

V W

E (

cycle

3)

Burn

in 1

0H

V W

E (

cycle

4)

Burn

in 1

0H

V W

E (

cycle

5)

Burn

in 1

0H

V W

E (

cycle

6)

Burn

in 1

0H

V W

E (

cycle

7)

Burn

in 1

0H

V W

E (

cycle

8)

Burn

in 1

0H

V W

E (

cycle

9)

Burn

in 1

0H

V W

E (

cycle

10)

Burn

in 1

KW

E (

cycle

1)

Burn

in 1

KW

E (

cycle

s2

-10)

Burn

in 1

KW

E (

cycle

s11

-50)

Burn

in 1

KW

E (

cycle

s51

-99)

Burn

in 1

KW

E (

cycle

s100

-…

Burn

in 1

KW

E (

cycle

s200

-…

Burn

in 1

KW

E (

cycle

s300

-…

Burn

in 1

KW

E (

cycle

s400

-…

Burn

in 1

KW

E (

cycle

s500

-…

Burn

in 1

KW

E (

cycle

s600

-…

Burn

in 1

KW

E (

cycle

s700

-…

Burn

in 1

KW

E (

cycle

s800

-…

Burn

in 1

KW

E (

cycle

s900

-…

Burn

in 1

KW

E (

cycle

1000)

Burn

in 1

KW

E (

invalid

…

602 603 604 605 606 607 608 609 610 611 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654

• The below chart shows the W/E failure by cycle count

• All valid Burn-In W/E cycling failures occur at cycle 1 after HV erase stress.

Fail code corresponding to failing W/E cycle # Fail codes for HV stress cycles

TM





7


• Collected and verified all Burn-In NVM W/E cycling fails

• All NVM W/E cycling fails failed within 30 cycles

0.000%

0.001%

0.002%

0.003%

0.004%

0.005%

0.006%

0.007%

0.008%

W/E cycle count

TM





8


• Replace the NVM continuous reads with the bulk read stress

− Current continuous reads: read one 32-Byte page at a time. Loop through all pages in entire flash array for 2 hours. The total stress time for each bit on 2MB flash array is: (1/65536)*120 minutes = 110 mS.

− New bulk read stress: all bits in the flash array are stressed in parallel. Total stress time for each bit is 2 minutes.

− Much improved stress duty cycle via bulk read vs old continuous reads during HTOL

− Elevated stress gate voltage in bulk read stress (5v vs. 4.2v). 5v is below the normal program verify voltage of 6v and safe to NVM.

• MPC5565 100k production Burn-In reject analysis showed all HTOL flash continuous read failures can be caught by the new bulk read stress.

• The new bulk read stress has already been used in C90 production Burn-In to replace HTOL continuous read stress, and no quality issue from millions for C90 units shipped afterwards.

TM





9


• Scan and BIST reduced to 20 minutes each from 1 hour each

− Current Hip7NVM probe program already contains HVST (high-

voltage stress test) to cover SRAM and logic.

• MPC5565 6-month production data (895k) showed 0 RAM

BIST fail.

• MPC5565 100K production reject analysis showed 0 fail from

Scan and BIST at Burn-In.

TM





10


5Hr BI

(125C)

Hot

test

Room

test

Assy

Data

Review

Fail collection + analysis

Fail collection + analysis

Cold

test

Hot

test 5Hr BI

(125)

Room

test 4Hr BI

(125C)

Fail collection

+ analysis

Cold

test

Room

test

Burn-In

Segment

A

Burn-In

Segment

B

Burn-In

Segment

C

Wafer

probe

Result :

No genuine failure observed after Burn-In Segment A (target 5 hour Burn-In)

TM





11


Conclusion:

• New 5 hour Burn-In flow provides equivalent test coverage to

the current 14 hour Burn-In flow.

confidential and proprietary - nxp · pdf filetm freescale, the freescale logo, altivec, c-5,...

Documents