On-chip Detection of Process Shift and Process …mahfuz/9C_2.pdfFF FS 0.6 1 1.4 1.8 0.2 0.6 1 1.4...

On-chip Detection of Process Shift andProcess Spread for Silicon Debugging and

Model-Hardware Correlation

Islam A.K.M. Mahfuzul and Hidetoshi Onodera

Department of Communications and Computer EngineeringGraduate School of Informatics

Kyoto [email protected]

ATS, 2012/11/22

Mahfuz (Kyoto University) Monitor Circuits ATS, 2012/11/22 1 / 23

Outline

1 Introduction

2 Proposed Monitor Circuits

3 Measurements from Corner Chips

4 Parameter Extraction Results

5 Conclusion


Introduction Background

Testing of Chip is a Huge Task

Testing is a must to ensure correct operation.Test effort > Design effortNeed correct strategy.

Fault TypesManufacturing fault.Parametric fault.

Test TypesFunctional test.Delay test.

This research proposes process-sensitive monitor circuits forparametric fault based delay defects.



Delay Defects

Manufacturing defect or parametric defect?If parametric defect, how to debug the defect?

What caused the delay defect? Threshold voltagefluctuation? Gate length deviation? etc.The model was not correct? Failed to predict the values.Need to correlate the model to hardware. Continuousfeedback required.

Le

aka

ge

Maximum Frequency



Process Variation

Variation in performanceLeakage

Maximum Frequency

Process variation

Process spread

Process shift

nMOSFET Threshold Voltage

pM

OS

FE

T T

hre

shold

Voltage

FF

SS

SF

FS

TT

Design optionDesign for worst cases.Design for typical case and bin the products.

Maximum operating frequency?

Amount of spread for statistical design approach?Mahfuz (Kyoto University) Monitor Circuits ATS, 2012/11/22 5 / 23


Process Control ModuleConventional

PCM

How helpful are PCM data?

Difference in layout.Location in the wafer.Not product-representative. DC vs. AC.Often different sizes due to probing.

Product-representative monitor circuits 1 ⇒ On-chip Detection1Gattiker et al., ITC’2006.


Introduction Motivation and Contribution

Motivation and Contribution

Fabrication

Transistor model

Variation model

Circuit

System

Extraction technique

(1)

(2)

Monitoring process-characteristics from product chips.Efficient model-to-hardware correlation methodology.Distinguish manufacturing fault and parametric fault.


Proposed Monitor Circuits On-chip Monitor Circuits

Outline

1 Introduction

2 Proposed Monitor CircuitsOn-chip Monitor CircuitsModel-Hardware Correlation



5 Conclusion



Proposed Monitor Circuits

(a) Standard (b) N-sensitive (c) P-sensitive

0

1

2

3

4

5

0 1 2 3 4 5 6 7

Se

nsitiv

ity t

o ∆

Vth

p

Sensitivity to ∆Vthn

Standard inverter nMOSFET monitor

pMOSFET monitor

Embed monitor circuits.Process-sensitive monitorstructures.



Parameter-sensitive Monitor Structure

Conventional

0

1

2

3

-40 0 40

Norm

aliz

ed F

requency

Threshold Voltage Change [mV]

VDD-Vthn

"L"Ip = 0

InnMOSFET

pMOSFET

Sensitive to bothnMOSFET and pMOSFETNot-suitable for on-chipdetection.

Proposed

1

2

3

-40 0 40

Norm

aliz

ed F

requency

Threshold Voltage Change [mV]

"L"Ip = 0

In

VDD

Sensitive to eithernMOSFET or pMOSFET.Suitable for on-chipdetection.



Corner Detection Capability

0.7

0.8

0.9

1

1.1

1.2

1.3

0.7 0.8 0.9 1 1.1 1.2 1.3

pM

OS

FE

T o

n c

urr

en

t [N

orm

aliz

ed

]

nMOSFET on current [Normalized]

TT

SF

SS

FF

FS

0.6

1

1.4

1.8

0.2 0.6 1 1.4 1.8 2.2

pM

OS

FE

T m

on

ito

r fr

eq

ue

ncy

nMOSFET monitor frequency

TT

SF

FS

SS

FF

Process corners are distinguishable using the proposedmonitor circuit outputs.

Quick detection of process-characteristics.Reduce debugging time for delay defects.


Proposed Monitor Circuits Model-Hardware Correlation

Model-hardware CorrelationExtraction of Parameters

Performance 1

Performance 2

Parameter 1

Parameter 2

FF

SSFS

SF

Sensitivity

matrix

MeasurementEstimation

Reference Reference

Extract parameters from monitor circuit outputs2.Need parameter-sensitized monitor circuits.Need to characterize monitor circuits by simulation.

2Mahfuzul et al., ICMTS’2011



Model-Hardware CorrelationProcess shift

Process Shift Amount of shifts in process parameters. Globalvariation.

Parameter ExtractionConsider RO frequency as a functionof ∆Vthp,∆Vthn and ∆L.

F = f (∆Vthp,∆Vthn,∆L)

= F0 + kp∆Vthp + kn∆Vthn + kl∆L

3 equations from 3 ROs.Solve the equations and deriveunknown ∆Vthp,∆Vthn and ∆L.a

aMahfuzul, et al., ICMTS 2011

#1 #2 #3 #4 #5 #6 #7

(a) INV-STD (b) INV-NPASS-I (c) INV-PPASS-I

0

1

2

3

4

5

0 1 2 3 4 5 6 7

Sensitiv

ity to ∆

Vth

p

Sensitivity to ∆Vthn

Standard inverter nMOSFET monitor

pMOSFET monitor



Model-Hardware CorrelationProcess spread

Process Spread Amount of deviation in process parameterswithin a chip. Local variation.

Parameter ExtractionAssume linear sensitivity of each variability source.

F = F0 +∑

i

(KVthpi ∆Vthpi + KVthni ∆Vthni + KLi∆Li

).

(σF

µF

)2

=∑

i

(k2

Vthpiσ2

Vthp+ k2

Vthniσ2

Vthn+ k2

Liσ2L

).

3 equations from 3 ROs.Solve the equations and derive σVthp

,σVthn, and σL.a

aFujimoto, et al., ICMTS 2012


Measurements from Corner Chips

Outline

1 Introduction


3 Measurements from Corner ChipsTest Chip DesignMeasurement Results


5 Conclusion


Measurements from Corner Chips Test Chip Design

Test Chip

ROs

Contr

oller

ROs

Enable

14x21

65 nm triple well process.Array based structure to capture local variation.14× 21 = 294 instances of the same RO type.TT, SS, FF, FS, SF corner chips.


Measurements from Corner Chips Measurement Results

Monitor Frequencies

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

2.2

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2pM

OS

FE

T m

on

ito

r fr

eq

. [n

orm

aliz

ed

]

nMOSFET monitor freq. [normalized]

Model

SS

FS

SF

TT

FF

Mismatch between corner model and measurement.Amount of process shift?

Amounts of shifts in key process parameters?Amount of spread?


Measurements from Corner Chips Measurement Results

WID Variation

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

TT FF FS SFSS TT FF FS SFSS TT FF FS SFSS

% V

ariation

nMOSFET monitor

pMOSFET monitor

Standard inverter monitor

nMOSFET variability is larger than pMOSFET variability⇒Important for statistical design.


Parameter Extraction Results

Outline

1 Introduction




5 Conclusion



Shifts in Parameters

0.6

0.7

0.8

0.9

1

1.1

1.2

1.3

0.7 0.8 0.9 1 1.1 1.2 1.3

pM

OS

FE

T thre

shold

[norm

aliz

ed]

nMOSFET threshold [normalized]

EstimationModel

Corner in model

Estimated corner

SS

SF

FS

FF

TT

Amount of shifts are extracted.Prediction of performances.Feedback to model.



Spreads in Parameters

Table: Extracted standard deviation of MOSFET threshold voltagesand gate length from RO frequency measurements.

Corner σVthn[mV] σVthp

[mV] σL [nm]TT 16.6 11.9 0.89SS 18.3 14.5 0.53FF 20.9 16.6 1.14FS 18.2 13.3 0.99SF 18.2 13.6 0.99

FF corner has larger variation.SS corner has smaller ∆L variation.


Conclusion

Outline

1 Introduction




5 Conclusion


Conclusion

Conclusion

SummaryUse of monitor circuits for detection of process corner andprocess spread.Product-representative monitor circuits suitable fordetection is proposed.Model-to-hardware correlation methodology is proposed.Test chip has been fabricated in a 65 nm process.Experimental results from corner chips show the validity ofthe proposed circuits.


Date post:	09-Aug-2020
Category:	Documents
Upload:	others
View:	0 times
Download:	0 times

On-chip Detection of Process Shift and Process …mahfuz/9C_2.pdfFF FS 0.6 1 1.4 1.8 0.2 0.6 1 1.4...

Documents