Limitations of Bipolar Compact Models for Low Frequency Noise

Application to HICUM

Workshop HICUM 2010 – Dresden – September 24th 2010

Nicolas DERRIER

Internal ref. : dm10.199

Outline

Motivation of this study : application issue of a real case

1/f Noise simulation issue with HICUM on advanced HBTs

Investigations “with the hands” about additional 1/f Noise sources in the HBT

Simulations results when implementing the sources found as preponderant

Conclusion and axes of improvement for HICUM Noise sources

2

Output IF noise level LO and RF "OFF" DCMIX ON

10

100

1000

0.0 0.1 1.0 10.0 100.0 1000.0

Freq (KHz)

IF n

ois

e (

nV

/sq

rt(H

z))

simulation

V2ON C917 W22 D15

V2ON C917 W22 D1

V2 ON C930 W18 D26

V2 ON C917 W22 D34

V2 ON C930 LotHSC W10 D33

x4

1

Application issue (1/2)

LF Noise measurements done on an application designed with ST BiCMOS 0.13 m mmW

technology (confidential)

Good agreements between simulated and measured thermal noise floors

But 1/f noise ~ 4x higher than typical simulation @ Freq=100Hz

3

Application issue (2/2)

1 2 3 4

5 6 7 8 109 11 12

13 14 15 16 1817 19 20

21 22 23 24 2625 27 28

29 30 31 32 3433 35 36

37 38 39 40 4241 43 44

45 46 47 48 5049 51 52

5453 55 56

Confirmation with statistical measurements :

Downmixer V2 Output Noise (IF=100Hz)

10

100

1000

0 10 20 30 40

Die NumberN

ois

e (

nV

/sq

rHz)

B9C930 J936EBR

Confirmation of ST Crolles measurements with LF Noise measurements done at IEMN

(Lille, France) too

4

The investigation begins…

Which device model is guilty in the application ?

Only Resistors and HBTs are preponderant in that application block

Coming from Resistors ?

Not sure, since 1/f Noise characterization is done and modeled for each resistor family, and the

noise level for resistors is much less than for bipolars…

Coming from Vertical HBT SiGe-C NPN High-Speed for mmW (fT=220GHz / fMAX=280GHz) ?

Probably, to confirm with SIB and SIC Noise measurements…

E CC B

Deep Trench Isolation (DTI)

B

As in-situ doped

Emitter

B in-situ doped

SiGe:C Base

B doped

Polybase

Shallow Trench Isolation (STI)

HS implant

Pedestal

oxide

Silicide

Buried Layer + Collector

Epitaxy + HV implant

Collector

Sinker

NPN High-Speed 0.27x3 m2 is used

Multi-fingers : 2 emitters, 4 bases, 3 collectors

5

6

Status on LF Noise sources in SPICE models

• LF Noise sources in HICUM (Level0 shown here, but same for Level2 and other SPICE)

• Only 1 Flicker Noise source

• kF and aF extracted with SIB measurements

B

E

Cx

F

2

r

B

2

r

E

2

r

Cx

2

T T

2

BC jBC AVL

2

CS jCS

2

BE jBE jBE jBEi j

a

F j E

BEp

B

4kTI

r

4kTI

r

4kTI

r

I 2qI

I 2q I I

I 2qI

I 2qI (in HL2, I Ik I

fI )

1/f

7

Issue comes from the HBT ? (1/3)

• Measuring current spectral density SI on the Base, for the HBT High-Speed 0.27x3 m2 2E/4B/3C

(same device used in the application)

• Usual measurements for HBT Flicker noise parameters extraction

• HICUM Level2 v2.23 simulations (Eldo 2010.1) vs SI measurements on the Base OK

VB

VC

EGG AMPLIFIER

HP35670AD.U.T

A

VB

VC

EGG AMPLIFIER

HP35670AD.U.T

A

2 2

measured vre vrb E Ieb E Iec E E IebIec2

d

d B E

b

1SI (f ) S S r r S (f ) r S 2r r r PR S

R

kTr R r r r 1

qImeasured ebSI SI

Mainly due to high r and =1000

Correlation term is supposed negligible

• According to [Benoit2005] :

Shot noise

2qIb

See Annex A for numerical application

8

Issue comes from the HBT ? (2/3)

• Now measuring current spectral density SI on the Collector, for the HBT High-Speed 0.27x3 m2

2E/4B/3C (same device used in the application)

• NOT usual measurements for HBT Flicker noise parameters extraction

EGG AMPLIFIER

HP35670A

A

VC=0.9V

D.U.T A

VB

RL

Cd Cd=22mF at the Base side

RL=1kOhm to be as close as possible to the application

9

Issue comes from the HBT ? Yes. (3/3)

2

2 E B 2

I_ measured vre vrb B E Ieb Iec c2

c

b

r r rS (f ) S S r r S (f ) S g

kTr g

qI r

• HICUM Level2 v2.23 simulations (Eldo 2010.1) vs SI Measurements on the Collector

• Shot noise OK

• Flicker 1/f NOT OK !

• Missing a 1/f component in our HICUM simulation…

• According to [Benoit2005] :

• Svre=Svre(f) ? Svrb=Svrb(f) ? SIec=SIec(f) ?

measured ecSI SI at low current

No simplification at medium and high current

See Annex A for

numerical application

10

Other LF Noise sources ? (1/2)

• Models with more complex Noise sources exist ([Ziel86&87], [Haaren98], [Tartarin99]],

[Borgarino99], [Kirtania96], [Kleinpenning92&94&95], [Benoit05], [Nunez-Perez07])

• Some of them include 1/f component for each Noise source

• Logical at least for the series resistances, since any passive resistor shows usually a

Flicker Noise component, and our passives models include it…

[Haaren98]

[Tartarin99]

[Borgarino99]

11

Other LF Noise sources ? (2/2)

• The expressions of the noise sources shown in the previous small signal scheme are

classically :

re rb

Sieb Siec

2 2

e e b bre e rb b

e e

b ceb b ec c

e e

2

2 b en re rb b e eb ec

e

Af Af

re rb

Af Af

Sieb Siec

cn eb 2

r I r ISv 4kTr Sv 4kTr

A .f A .f

.I .ISi 2qI Si 2qI

A .f A .f

r r rSv Sv Sv r r S

Kf Kf

Kf Kf

i Si

SiSi Si

Correlation Sv b en n b e eb ec2r r r

i (r r )Si Si

Red variables = parameters

• 1/f source parameters KfSieb and AfSieb of BE junction (Base current) are extracted as

usual with SIB measurements

• For NPN HS in ST BiCMOS 0.13 m mmW technology : KfSieb=4,3.10-10 m2 and AfSieb=2

12

1/f Noise in Siec (CE junction) (1/2)

• Process DOE : comparing the LF noise of NPN High-Speed with the NPN Medium-

Voltage (only the intrinsic collector doping level differs)

• Only difference between both NPN flavours : RCI0(NPN MV) = 10 x RCI0(NPN HS)

1.1018cm-31.1017cm-3

13

1/f Noise in Siec (CE junction) (2/2)

NPN MVNPN HS

Almost no difference

• 1/f component in Siec seems negligible

• KfSiec=0 in SiecA

cec

f

ecc

e

SiKf .ISi 2qIA .f

14

1/f Noise of Svrb (Base series resistance)

• Using our Base resistance DC test structures (emitter ring structures) to measure LF

Noise :

y = 2,85E-12x1,98E+00

R2 = 9,79E-01

1,E-18

1,E-17

1,E-16

1,E-03 1,E-02

Irb (A)

SIr

b (

A²)

at

1 H

z

NRBA

L= 15 µm

Rb noise level after correction of Rs impact

rb2

b b

Af

rrb

rb

11 2

r

b b

b

e

r ISv 4kT

Kf

Af 2

Kf 1,5.1

r

m

A .f

0

• 1/f Noise exists in the Base resistance :

• But influence of this 1/f noise neglected due

to the low value of the base current, compared

to the collector current or the emitter current

( =IC/IB=1000 in this technology)

1,00E-23

1,00E-22

1,00E-21

1,00E-20

1,00E-19

1,00E-18

1,00E-17

1,00E-16

1,00E-15

10 100 1000 10000 100000

Frequency (Hz)

SIr

b (

A²/

Hz) 5,00E-03

2,50E-03

1,00E-03

NRBA

Rs= 220 Ohm

L= 15 µm

I=

IB=

15

1/f Noise of Svre (Emitter series resistance) (1/2)

• No existing DC test structures to measure directly the Emitter resistance

• So, using a process split to “play” with the Emitter resistance (in reality the poly-emitter

doping level, the depth of the crystallization, so the interface emitter poly-mono), and

measure the consequence on the HBT 1/f noise

• Poly-emitter Arsenic doping level from 280cc to 90cc (/3) (so RE x 3, confirmed by PT )

E CC B

Deep Trench Isolation (DTI)

B

As in-situ doped

Emitter

B in-situ doped

SiGe:C Base

B doped

Polybase

Shallow Trench Isolation (STI)

HS implant

Pedestal

oxide

Silicide

Buried Layer + Collector

Epitaxy + HV implant

Collector

Sinker

280cc 90cc

16

1/f Noise of Svre (Emitter series resistance) (2/2)

• Measuring the same device on both 280cc and 90cc wafers shows a strong effect of

the emitter poly-mono interface on the 1/f Noise :

1/f

shot

17

Spice library modification

• Decision to add an 1/f source on SvRE in the HICUM model

• Yes, but ST deliveries not in VA code, so no access to the inside of HICUM…

• Workaround : “Removing” RE native resistance by setting RE parameter = 1mΩ

• Adding an additional external RE in our sub-circuit, with an 1/f noise source

E1

rErEAf

2 ErE

E

rE

E

I 4kTI

A .

f

f r

K

18

Results on the HBT transistors

• Eldo simulations (lines) vs measurements (dots) for the noise current spectral density

measured on the collector, for different HBTs geometries :

KfRE=3,0.10-14 m2 and AfRE=0.8

Results on the application

Updated Noise HBT HICUM model shows very good correlation on the 1/f noise from

T=-40°C to 120°C :

Output noise level LotHSC W10 D27 / LO&RF OFF

10

100

0.1 1.0 10.0 100.0 1000.0

Freq (KHz)

IF n

ois

e (

nV

/sq

rt(H

z)) Meas -40°C

Meas 27°CBS Sire -40°CBS Sire 27°CSimu 27°CMeas 120°CBS Sire 120°C

No 1/f in SvRE

19

Same 1/f noise issue for other ST bipolar technologies ?

Issue seems historically present for all ST BiCMOS technologies

Simulations (lines) vs measurements (dots) always ok for SIB measurements, but not for

SIC ones

Yes, but not an application issue, since circuits usually > [1kHz-10kHz]

Example of HBTs Noise SIC measurements for 3 other technologies :

SiGeC HBT fT=60GHz in 250nm

technology

SiGeC low-cost HBT fT=50GHz

in 130nm technology

SiGeC advanced HBT

fT=160GHz in 130nm technology

20

Conclusion (1/2)

An issue concerning Flicker Noise Spice simulations vs measurements was

shown, on a ST application real case (no more details for confidentiality

reasons)

This application issue comes from the miscorrelation between HICUM

simulations and measurements on SiGeC vertical NPN in a ST BiCMOS

0.13 m mmW technology

It is visible only with Noise measurements on the Collector (SIC)

According to many papers, 1/f noise sources exist at different locations in the

HBT, especially in the series resistances

It was demonstrated thanks to DOE process splits and measurements, that our

problem mainly comes from an 1/f source existing in the Emitter series

resistance, which does not exist in the HICUM model

As a workaround, adding this 1/f source in an external RE (not in native one,

because no VA code delivery) enables to correlate perfectly simulations and

measurements, for several HBTs geometries (law proportional to Emitter area)

21

Conclusion (2/2)

As a consequence, ST would appreciate that TUD improves the 1/f Noise sources model

in HICUM…

For example, 1/f source available for each Noise source (“who can do more can do

less”)?

FrB

rB

FrE

rE

FrC

rC

FCE

CE

FBC

BC

FCS

FBE

B

CS

B

E

Cx

E

2

r

B

2

r

E

2

r

Cx

2

T T

2

BC jBC AVL

2

CS j

a

F B

a

F E

a

F C

a

F T

a

F jBC AVL

a

F jCS

CS

2

BE jBE jBE jBEi jBEp

a

F jBE

4kTI

r

4kTI

r

4kTI

r

I 2qI

I 2q I I

k I

f

k I

f

k I

f

k I

f

I 2qI

I 2qI (in HL2, I

k I

f

I I

k I I

f

k I

f

)

1/f

1/f

1/f

1/f

1/f

1/f ?

1/f ?

Not in HICUM

In HICUM

22

THANKS FOR YOUR ATTENTION !

YOU CAN MAKE 1/f NOISE :

OR

23

References

[Nunez-Perez07] : Jose Cruz NUNEZ PEREZ, “Contribution à la Conception de Système de RadioCommunication : de la

Modélisation de Transistors Bipolaires à l’Evaluation des Performances d’un Système d’Emission-Reception”, Thèse de

Doctorat soutenue le 03 Décembre 2007, INSA Lyon

[Benoit05] : Patrice BENOIT, “Influence de Parametres Technologiques sur le Bruit Basse Frequence des Transistors

Bipolaires a Heterojonction Si/SiGe:C”, Thèse de Doctorat soutenue le 09 Décembre 2005, Université de Montpellier II

[Tartarin99] : Tartarin J.G. et al., “Noise properties in SiGe BiCMOS devices”, in High Performance Electron Devices for

Microwave and Optoelectronic Applications, 1999.

[Borgarino99] : Borgarino M., Kovacic S., Lafontaine H., “Low Noise considerations in SiGe BiCMOS Technology for RF

Apllications”, European Wireless’99, october1999, Munich, Allemagne

[Haaren98] : Haaren B.V. et al., “Noise properties of SiGe heterojunction bipolar transistors”, in Silicon Monolithic Integrated

Circuits in RF Systems, 1998.

[Kirtania96] : Kirtania A.K., Das M.B., Chandrasekhar S. et al., “Measurement and Comparison of 1/f Noise and G-R Noise

in Silicon Homojunction and III-V Heterojunction Bipolar Transistors”, IEEE Transactions on Electron Devices, 1996

[Kleinpenning95] : KleinPenning and Markus, “Low-Frequency Noise in Polysilicon Emitter Bipolar Transistors”, IEEE

Transactions on Electron Devices, 1995

[Kleinpenning94] : KleinPenning, “Low-Frequency Noise in Modern Bipolar Transistors : Impact of Intrinsic Transistor and

Parasitic Series Resistances”, IEEE Transactions on Electron Devices, 1994

[Kleinpenning92] : KleinPenning, “Location of Low-Frequency Noise Sources in Submicrometer Bipolar Transistors”, IEEE

Transactions on Electron Devices, 1992

[Ziel87] : Van Der Ziel A., Pawlikiewicz H., “Location of 1/f Noise sources in BJTs and HBTs – II. Practice”, IEEE

Transactions on Electron Devices, 1987

[Ziel86] : Van Der Ziel A., Zhang X., Pawlikiewicz H., “Location of 1/f Noise sources in BJTs and HBTs – I. Theory”, IEEE

Transactions on Electron Devices, 1986

24

Annex A : numerical application (1/2)

2

measured E E IebIec2

2

E Ievre E Ieb2

d

vr2

d d

cb2

d

2

d

1r r S

1S

R

1SI (f ) 2r r(

1S f )

Rr PR Sr

RS

RR

1

Shot noise

2qIb

LF noise measurements on the Base (see slide #7) :

Numerical application for the a NPN HS CBEBCBEBC

0.27x3.0 m2

=2,52 10-08 =4,3 10-20

@VBE=0.86V

f>10kHz=2,52 10-08

=4,2 10-19 =0,99 =1,7 10-24 =1,7 10-07=1 10-21

=8,2 10-04

=1,0 10-27A2/Hz =1,0 10-26A2/Hz =1,7 10-24A2/Hz =1,7 10-28A2/Hz

=7,4 10-14 =4,3 10-20

@VBE=0.68V

f>1kHz=7,4 10-14

=4,2 10-19 =0,99 =2,2 10-27 =5 10-13=2,5 10-24

=8,2 10-04

=3,2 10-33A2/Hz =3,1 10-32A2/Hz =2,2 10-27A2/Hz =1,3 10-36A2/Hz

Simeasured = SIeb

Simeasured = SIeb

25

Annex A : numerical application (2/2)

LF noise measurements on the Collector (see slides #8-9) :

Numerical application for the a NPN HS CBEBCBEBC

0.27x3.0 m2

@VBE=0.87V

f>10kHz

@VBE=0.74V

f>1kHzSimeasured = SIec

222

c vr

2

E B22

c c IecvI_ mea c B E Ieb 2rs e e bur dg rg S r S

rS (f

r rg S(f) )g S

=0.015 =4,3 10-20

=0.015

=4,2 10-19 =12 =1,7 10-24 =1 =1 10-21

=6,5 10-22A2/Hz =6,3 10-21A2/Hz =2 10-23A2/Hz =1 10-21A2/Hz

=2,5 10-6 =4,3 10-20

=2,5 10-6

=4,2 10-19 =2 10-03 =1,2 10-26 =1 =1,3 10-23

=1 10-25A2/Hz =1 10-24A2/Hz =2,4 10-29A2/Hz =1,3 10-23A2/Hz

Simeasured = no

simplification

26