a comparison of npn and pnp profile design tradeoffs for complementary sige hbt technology

6
A comparison of npn and pnp profile design tradeos for complementary SiGe HBT Technology Gang Zhang a, * , John D. Cressler a , Guofu Niu a , Angelo Pinto b a Department of Electrical and Computer Engineering, Alabama Microelectronics Science and Technology Center, Auburn University, 200 Broun Hall, Auburn AL 36849, USA b Texas Instruments Deutschland, Haggertystrasse 1, D-85356 Freising, Germany Received 17 June 2000; accepted 27 July 2000 Abstract A comprehensive comparison of npn and pnp profile design tradeos for SiGe HBTs is conducted using calibrated simulations. The parasitic energy band barrier induced by the SiGe/Si transition in the collector–base space charge region produces very dierent design constraints for pnp and npn transistors, and they must be optimized separately. A single SiGe profile is found to give acceptable performance for both npn and pnp devices, while still satisfying film stability constraints. Ó 2000 Published by Elsevier Science Ltd. All rights reserved. 1. Introduction SiGe HBT technology has generated worldwide in- terest for digital, analog, and RF applications, because it combines the transistor performance of III–V technol- ogies with manufacturability, high yield, and low cost associated with conventional Si IC fabrication. In ad- dition, the combination of SiGe HBTs with scaled Si CMOS to form SiGe HBT BiCMOS technology pre- sents exciting possibilities for system-on-a-chip solution for emerging wireless applications [1]. At present, SiGe technology development is exclusively centered on npn SiGe HBTs. For high-speed analog and mixed-signal circuit applications, however, a complementary (npn pnp) bipolar technology oers significant performance advantages over an npn-only technology. Push–pull circuits, for instance, ideally require a high-speed verti- cal pnp transistor with comparable performance to the npn transistor. In addition to the historical bias in favor of npn Si BJTs due to the larger minority electron mo- bility in the base, the valence band oset in SiGe strain layers is generally more conducive to npn SiGe HBT designs, because it translates into an induced conduction band oset and band grading which greatly enhance minority electron transport in the device, thereby sig- nificantly boosting transistor performance over a simi- larly constructed npn Si BJT. For a pnp SiGe HBT, on the other hand, the valence band oset directly re- sults in a valence band barrier, even at low injection, which strongly degrades minority hole transport and thus limits frequency response. Careful optimization to minimize these hole barriers in pnp SiGe HBTs is thus required, and has in fact yielded impressive device per- formance compared to Si pnp BJTs [2,3]. What remains lacking in the literature is a careful analysis of the inherent profile design dierences between npn and pnp SiGe HBTs. Relevant questions in this context include 1. How does SiGe npn and pnp profile design optimiza- tion fundamentally dier? 2. Can a single Ge profile design point be used for both npn and pnp transistors? 3. Is a graded base design preferable to a box profile de- sign for pnp HBTs? 4. How much Ge retrograding is required to obtain ac- ceptable SiGe pnp HBT performance? In this work, we address these issues for the first time, and use calibrated device simulations to shed light on the fundamental SiGe profile design dierences between Solid-State Electronics 44 (2000) 1949–1954 * Corresponding author. Tel.: +1-334-844-1872; fax: +1-334- 844-1888. E-mail address: [email protected] (G. Zhang). 0038-1101/00/$ - see front matter Ó 2000 Published by Elsevier Science Ltd. All rights reserved. PII:S0038-1101(00)00165-9

Upload: gang-zhang

Post on 02-Jul-2016

219 views

Category:

Documents


4 download

TRANSCRIPT

Page 1: A comparison of npn and pnp profile design tradeoffs for complementary SiGe HBT Technology

A comparison of npn and pnp pro®le design tradeo�s forcomplementary SiGe HBT Technology

Gang Zhang a,*, John D. Cressler a, Guofu Niu a, Angelo Pinto b

a Department of Electrical and Computer Engineering, Alabama Microelectronics Science and Technology Center, Auburn University, 200

Broun Hall, Auburn AL 36849, USAb Texas Instruments Deutschland, Haggertystrasse 1, D-85356 Freising, Germany

Received 17 June 2000; accepted 27 July 2000

Abstract

A comprehensive comparison of npn and pnp pro®le design tradeo�s for SiGe HBTs is conducted using calibrated

simulations. The parasitic energy band barrier induced by the SiGe/Si transition in the collector±base space charge

region produces very di�erent design constraints for pnp and npn transistors, and they must be optimized separately. A

single SiGe pro®le is found to give acceptable performance for both npn and pnp devices, while still satisfying ®lm

stability constraints. Ó 2000 Published by Elsevier Science Ltd. All rights reserved.

1. Introduction

SiGe HBT technology has generated worldwide in-

terest for digital, analog, and RF applications, because it

combines the transistor performance of III±V technol-

ogies with manufacturability, high yield, and low cost

associated with conventional Si IC fabrication. In ad-

dition, the combination of SiGe HBTs with scaled Si

CMOS to form SiGe HBT BiCMOS technology pre-

sents exciting possibilities for system-on-a-chip solution

for emerging wireless applications [1]. At present, SiGe

technology development is exclusively centered on npn

SiGe HBTs. For high-speed analog and mixed-signal

circuit applications, however, a complementary (npn�pnp) bipolar technology o�ers signi®cant performance

advantages over an npn-only technology. Push±pull

circuits, for instance, ideally require a high-speed verti-

cal pnp transistor with comparable performance to the

npn transistor. In addition to the historical bias in favor

of npn Si BJTs due to the larger minority electron mo-

bility in the base, the valence band o�set in SiGe strain

layers is generally more conducive to npn SiGe HBT

designs, because it translates into an induced conduction

band o�set and band grading which greatly enhance

minority electron transport in the device, thereby sig-

ni®cantly boosting transistor performance over a simi-

larly constructed npn Si BJT. For a pnp SiGe HBT,

on the other hand, the valence band o�set directly re-

sults in a valence band barrier, even at low injection,

which strongly degrades minority hole transport and

thus limits frequency response. Careful optimization to

minimize these hole barriers in pnp SiGe HBTs is thus

required, and has in fact yielded impressive device per-

formance compared to Si pnp BJTs [2,3].

What remains lacking in the literature is a careful

analysis of the inherent pro®le design di�erences between

npn and pnp SiGe HBTs. Relevant questions in this

context include

1. How does SiGe npn and pnp pro®le design optimiza-

tion fundamentally di�er?

2. Can a single Ge pro®le design point be used for both

npn and pnp transistors?

3. Is a graded base design preferable to a box pro®le de-

sign for pnp HBTs?

4. How much Ge retrograding is required to obtain ac-

ceptable SiGe pnp HBT performance?

In this work, we address these issues for the ®rst time,

and use calibrated device simulations to shed light on

the fundamental SiGe pro®le design di�erences between

Solid-State Electronics 44 (2000) 1949±1954

* Corresponding author. Tel.: +1-334-844-1872; fax: +1-334-

844-1888.

E-mail address: [email protected] (G. Zhang).

0038-1101/00/$ - see front matter Ó 2000 Published by Elsevier Science Ltd. All rights reserved.

PII: S0 03 8 -1 10 1 (00 )0 0 16 5 -9

Page 2: A comparison of npn and pnp profile design tradeoffs for complementary SiGe HBT Technology

npn and pnp SiGe HBTs that might be encountered, for

instance, in developing a viable complementary SiGe

HBT technology.

2. Simulation approach

The drift-di�usion simulator, MEDICI [4] was used

in this work. For simplicity, 1D simulations were used,

since we are only interested here in the intrinsic pro®le

design. In addition, we chose to focus on hypothetical

``toy'' npn and pnp SiGe pro®les (Fig. 1) with constant

emitter, base, and collector doping, and Ge content not

subject to stability constraints, such that pro®le design

di�erences between npn and pnp devices could be more

easily discriminated and not masked by doping gradient

e�ects (stability issues are addressed in Section 4). The

base doping was chosen to give a pinched base sheet

resistance in the range of 8±12 kX/h. This arti®cial as-

sumption on constant doping yields performance num-

bers (e.g. fT), which are lower than what would be

expected for a real complementary SiGe technology, but

relative comparisons between npn and pnp devices

should nonetheless be valid.

We began our simulations by calibrating MEDICI

for both npn and pnp Si BJTs to measured comple-

mentary Si BJT hardware [5], using actual device layouts

and measured SIMS data. We found that the default

hole mobility modeling capability of MEDICI was de-

®cient and tuning was required to obtain reasonable

agreement between data and simulation, particularly

under high-level injection. The SiGe parameters from

our extensive prior work on simulating high-speed SiGe

npn HBTs was used [6], and assumed to be the same for

both npn and pnp devices.

3. Results and discussion

A comparison of the conduction and valence band

edges for both npn and pnp devices without any Ge

retrograding into the collector is shown in Figs. 2 and 3

for a hypothetical 25% Ge triangle pro®le at equilib-

rium. Observe that, while there is no visible conduction

band barrier present in the npn HBT, there is an obvious

valence band barrier in the pnp HBT. This is consistent

with the valence band o�set in strained SiGe on Si, and

clearly indicates that pnp SiGe HBT design is inherently

more di�cult than npn SiGe HBT design. In addition,

due to the intrinsic mobility di�erences between elec-

trons and holes, it is also clear that npn devices will

consistently outperform pnp devices, everything else

being equal. Unlike a well-designed npn HBT (i.e. Ge

Fig. 1. Doping and Ge pro®les of the pnp and npn SiGe HBTs.

Fig. 2. Conduction band edge of npn SiGe HBTs for di�erent

values of peak Ge content.

Fig. 3. Valence band edge of pnp SiGe HBTs for di�erent

values of peak Ge content.

1950 G. Zhang et al. / Solid-State Electronics 44 (2000) 1949±1954

Page 3: A comparison of npn and pnp profile design tradeoffs for complementary SiGe HBT Technology

outside the neutral base edges), where conduction band

barrier e�ects are uncovered only at high JC under Kirk

e�ect [7], the valence band barrier in pnp HBTs is op-

erative at even low injection, and acts to block minority

holes transiting the base. This pile-up of accumulated

holes produces a retarding electric ®eld in the base,

which compensates the Ge-grading-induced drift ®eld,

dramatically decreasing JC, b and fT�. This e�ect wors-

ens as the current density increases, since more hole

charge is stored in the base (Figs. 4 and 5). In this case,

the fT of the pnp HBT is signi®cantly lower than that

of the pnp Si BJT. As expected, however, retrograding

of the Ge edge into the collector can ``smooth'' this va-

lence band o�set in the pnp HBT, and thus improve this

situation dramatically, although at the expense of ®lm

stability. For an increase of the Ge retrograde from 0 to

40 nm, the pnp HBT performance is dramatically im-

proved, giving roughly twice in peak fT over the pnp Si

BJT performance at equal doping (Fig. 5).

Figs. 6 and 7 show the variation in peaks fT and b as

a function of peak Ge content for both npn and pnp

HBTs, for both 0 and 100 nm Ge retrogrades. At 100 nm

retrograde, the performance of the pnp HBT mono-

tonically improves as the Ge content rises, while the

maximum useful Ge content is limited to about 10%

without retrograding. Fig. 8 indicates that 40±50 nm of

Ge retrograding in the pnp HBT is su�cient to

``smooth'' the valence band barrier, and this is re¯ected

in Fig. 9, which explicitly shows the dependence of pnp

peak fT on Ge retrograde distance, for both triangular

and box Ge retrograde shapes. Observe that the box Ge

retrograde is not e�ective in improving the pnp HBT

Fig. 4. Gummel characteristics of pnp HBTs for di�erent Ge

retrograde pro®les.

Fig. 5. Cuto� frequency vs. collector current density for pnp

and npn SiGe HBTs.

Fig. 6. Cuto� frequency vs. peak Ge content for di�erent Ge

retrograde pro®les.

Fig. 7. Current gain vs. peak Ge content for di�erent Ge ret-

rograde pro®les.

G. Zhang et al. / Solid-State Electronics 44 (2000) 1949±1954 1951

Page 4: A comparison of npn and pnp profile design tradeoffs for complementary SiGe HBT Technology

performance, since it does not smooth the Ge barrier,

but rather only pushes it deeper into the collector, where

it is still felt at the high JC needed to reach peak fT. This

box Ge retrograde is undesirable from the stability

standpoint as well. The e�ects of Ge retrograding on the

npn HBT performance, on the other hand, are minor,

while the ®lm stability is signi®cantly worse due to the

additional Ge content. This suggests that using one Ge

pro®le design for both npn and pnp HBTs is not opti-

mum for high peak Ge values. (Note that, while the peak

fT is unchanged with Ge retrograding in the npn HBT,

the fT response above peak fT does not roll o� as rapidly

due to the high-injection-induced barrier, consistent

with the results in [7].)

Fig. 10 compares the frequency response of the npn

and pnp HBTs as a function of front-side Ge pro®le

shape (triangle vs. box), and peak Ge content. Observe

that for the npn HBT, the base transit time reduction

from the Ge-grading-induced drift ®eld of the triangle

Ge pro®le gives a signi®cant advantage above 10% peak

Ge, indicating that the npn HBT is base transit time

limited. Interestingly, for the pnp HBT, however, the

di�erences between the box and triangle Ge pro®les are

less pronounced. The box Ge pro®le gives a slight ad-

vantage at low Ge content due to the low b, and hence

importance of the emitter transit time (sE / 1=b), but

once b is su�ciently high, the triangle Ge pro®le dom-

inates at higher peak Ge content, where the base transit

time limits the overall response. In both npn and pnp

devices, the triangle Ge pro®le has better performance

and better stability, and thus can be considered an op-

timum shape. This is even more apparent if we examine

the early voltage of the device (Fig. 11), a key ®gure-

of-merit in complementary analog technology. Here, the

triangle Ge pro®le has as clear advantage due to its

Fig. 10. Cuto� frequency vs. peak Ge content for triangle and

box Ge pro®les.

Fig. 11. Early voltage vs. peak Ge content for triangle and box

Ge pro®les.

Fig. 9. Cuto� frequency vs. Ge retrograde pro®le shape and

depth.

Fig. 8. Valence band edge of pnp SiGe HBTs for di�erent Ge

retrograde distances.

1952 G. Zhang et al. / Solid-State Electronics 44 (2000) 1949±1954

Page 5: A comparison of npn and pnp profile design tradeoffs for complementary SiGe HBT Technology

graded band gap [1], and both npn and pnp show sig-

ni®cant improvements in VA with increasing Ge content.

We have also examined the e�ects of front-side

(emitter) Ge misalignment on both npn and pnp per-

formance. Fig. 12 shows b as a function of emitter±base

o�set into the emitter. As expected, b increases with Ge

movement into the emitter for both npn and pnp SiGe

HBTs, since the e�ective Ge content at the emitter±base

junction increases, and b depends exponentially on the

Ge o�set at that point. As the Ge misalignment in the

emitter increases, however, a slight degradation in peak

fT results, since the e�ective grading across the neutral

base is reduced.

4. Stability issues

The total amount of Ge that can be put into a SiGe

HBT is limited by the thermodynamic stability criterion.

Above the critical thickness, the strain in the SiGe ®lm

relaxes, generating defects. The empirical critical thick-

ness of SiGe multi-layer is approximately 1.65 times the

theoretical stability result of Matthews and Blakeslee

[8,9]. In general, varying peak Ge content or retrograde

distance (i.e. ®lm thickness) moves along di�erent con-

tours in stability space (Fig. 13). Under the SiGe sta-

bility constraint, the peak Ge content must be traded for

the Ge retrograde distance in the collector±base junc-

tion. Fig. 14 shows that a 11% peak Ge pro®le with 25

nm retrograde gives the highest fT for pnp. Fig. 15

shows that the 12.5% peak Ge pro®le with 9 nm retro-

grade gives the highest fT for npn. Note, however, that

the npn performance is not sensitive to the SiGe pro®le,

and hence, without a signi®cant loss of performance, the

same Ge pro®le may be used for both pnp and npn. This

may be advantageous from a fabrication viewpoint.

These results are valid for current bipolar processes with

about 100 nm base width. If the base width is further

reduced with technology scaling, the peak Ge can be

obviously increased, while maintaining ®lm stability.

The same optimization methodology employed here can

be used in that case to determine the best SiGe pro®le

for both devices.

5. Summary

A comprehensive comparison of npn and pnp pro®le

design tradeo�s for SiGe HBTs is conducted using cal-

ibrated simulations. Based on our simulation results, a

triangular Ge pro®le is the optimum Ge pro®le shape for

Fig. 13. SiGe stability diagram. Ge pro®les (1), (2), (3) and (4)

on the stability curve are used in Figs. 14 and 15.

Fig. 14. Cuto� frequency vs. collector current density of pnp

SiGe HBTs for di�erent Ge pro®les on the stability diagram.

Fig. 12. Current gain vs. emitter±base Ge o�set.

G. Zhang et al. / Solid-State Electronics 44 (2000) 1949±1954 1953

Page 6: A comparison of npn and pnp profile design tradeoffs for complementary SiGe HBT Technology

both npn and pnp SiGe HBTs. Signi®cant Ge retro-

grading in the collector is required for optimum SiGe

pnp performance, however, in order to smooth the va-

lence band barrier in the collector±base junction. For a

100 nm SiGe ®lm thickness, a single SiGe pro®le is

found to provide adequate performance for both npn

and pnp devices.

Acknowledgements

This work was supported by Texas Instruments, Inc.

The authors would like to thank B. El-Kareh, G.

Howard, J. Babcock, D. Tatman, and J. Erdeljac for

their contributions to this work.

References

[1] Cressler JD. SiGe HBT technology: a new contender for Si-

based RF and microwave circuit applications. IEEE Trans

Microwave Theory Tech 1998;46(5):572±89.

[2] Harame DL, Stork JMC, Meyerson BS, Crabb�e EF,

Patton GL, Scilla GJ, de Fr�esart E, Bright AA, Stanis C,

Megdanis AC, Manny MP, Petrillo EJ, Dimeo M, Mclntosh

RC, Chan KK. SiGe-base PNP transistors fabrication

with n-type UHV/CVD LTE in a ``NO DT'' process.

Technical Digest of the 1990 Symposium on VLSI Tech-

nology, 1990. p. 47±8.

[3] Harame DL, Meyerson BS, Crabb�e EF, Cotte JM, Stork

JMC, Megdanis AC, Patton GL, Sti�er SR, Johnson JB,

Warnock JD, Comfort JH, Sun JYC. 55 GHz polysilicon-

emitter graded SiGe-base PNP transisitors. Technical Di-

gest of the 1991 Symposium on VLSI Technology, 1991.

p. 71±2.

[4] MEDICI,, 2-D semiconductor device simulator, version 4.0.

Avant! Inc, 1997.

[5] Patel R, Milam W, Cooley G, Corsi M, Erdeljac J, Hutter L.

A 30 V complementary bipolar technology on SOI for high

speed precision analogy circuits. Proceedings of the 1997

IEEE Bipolar/BiCOMS Circuits and Technology Meeting,

1997. p. 48±50.

[6] Niu G, Ansley WE, Zhang S, Cressler JD, Webster CS,

Grove RA. Noise parameter optimization of UHV/CVD

SiGe HBTs for RF and microwave application. IEEE Trans

Electron Dev 1999;46:1589±98.

[7] Joseph AJ, Cressler JD, Richey DM, Niu G. Optimization

of SiGe HBTs for operation at high current densities. IEEE

Trans Electron Dev 1999;46:1347±54.

[8] Sti�er SR, Comfort JH, Harame CL, de Fr�esart E,

Meyerson BS. The thermal stability of SiGe ®lms deposited

by ultrahigh-vacuum chemical vapor deposition. J Appl

Phys 1991;70(3):1416±20.

[9] Fischer A, Osten H-J, Richter H. An equilibrium model for

buried SiGe strained layers. Solid-State Electron 2000;

44:869±73.

Fig. 15. Cuto� frequency vs. collector current density of npn

SiGe HBTs for di�erent Ge pro®les on the stability diagram.

1954 G. Zhang et al. / Solid-State Electronics 44 (2000) 1949±1954