VOL. 25,
NO. 3,
MARCH 2014
633
INTRODUCTION
634
RELATED WORK
VOL. 25,
NO. 3,
MARCH 2014
SYSTEM MODEL
YILDIRIM AND KANTARCI: EXTERNAL GRADIENT TIME SYNCHRONIZATION IN WIRELESS SENSOR NETWORKS
635
u2N v
u
Lv t Lv t
;
jN v j 1
636
such that t denotes the time just after the update operation.
It can be proven that sensor nodes agree on a common logical
clock speed and clock value by applying these equations.4
If the received synchronization message carries a higher
sequence number (Line 12), this situation indicates that the
reference node has recently flooded its current rate multiplier and time offset into the network. Hence, v updates lref
v ,
ref
,
and
seq
to
the
corresponding
received
values
v
v
(Lines 13-16).
For all times t after the clock speed and clock value
agreement is achieved, it theoretically holds for the reference
node ref and for the node v that href t:lref t hv t:lv t.
Thus, it follows that
href t hv t:
lv t
:
lref t
10
VOL. 25,
NO. 3,
MARCH 2014
YILDIRIM AND KANTARCI: EXTERNAL GRADIENT TIME SYNCHRONIZATION IN WIRELESS SENSOR NETWORKS
637
TABLE 1
Summary of the Measured Skew Values Observed with
FTSP, GTSP, and EGSync during the Experiments
Fig. 2. The placement of sensor nodes to construct the line and ring
topologies for the experiments.
Fig. 3. Global and average global skews measured for FTSP (left), GTSP (middle), and EGSync (right) on the line topology, respectively.
Fig. 4. Local and average local skews measured for FTSP (left), GTSP (middle), and EGSync (right) on the line topology, respectively.
Fig. 5. Global and average global skews measured for FTSP (left), GTSP (middle), and EGSync (right) on the ring topology, respectively.
Fig. 6. Local and average local skews measured for FTSP (left), GTSP (middle), and EGSync (right) on the ring topology, respectively.
638
VOL. 25,
NO. 3,
MARCH 2014
Fig. 7. Maximum skew to the reference node (node with identifier 1) per node measured for FTSP (left column) and EGSync (right column) on the
line topology.
Fig. 8. Maximum skew to the reference node (node with identifier 1) per node measured for FTSP (left column) and EGSync (right column) on the
ring topology.
Fig. 9. The synchronization error between nodes 19 and 20 for FTSP (left), GTSP (middle), and EGSync (right) on the line topology, respectively.
Fig. 10. The maximum local skew per node for FTSP (left), GTSP (middle), and EGSync (right) on the line topology, respectively.
identifier 20 during the experiments. However, the maximum local skews of sensor nodes observed with GTSP and
EGSync are quite close to each other and they do not
increase drastically as the distance to the reference node
increases. This situation can also be observed from the
synchronization error between nodes 19 and 20, which are
at the end of the line topology. In FTSP, since nodes
consider only the clock values flooded by the reference
node and they do not consider the clock of their neighbors,
the synchronization error between nodes 19 and 20 can be
quite large during the execution. However, the maximum
synchronization error between nodes 19 and 20 is quite
small in GTSP and EGSync when compared to FTSP. This is
due to the fact that sensor nodes executing GTSP and
EGSync synchronize to their neighbors. Although EGSync
additionally synchronizes sensor nodes to a reference time
source in contrary to GTSP, the synchronization error
between the neighboring nodes is quite comparable to that
in GTSP.
Figs. 11 and 12 present the synchronization error
between nodes 9 and 10 on the ring topology and the
YILDIRIM AND KANTARCI: EXTERNAL GRADIENT TIME SYNCHRONIZATION IN WIRELESS SENSOR NETWORKS
639
Fig. 11. The synchronization error between nodes 9 and 10 for FTSP (left), GTSP (middle), and EGSync on the ring topology, (right), respectively.
Fig. 12. The maximum local skew per node for FTSP (left), GTSP (middle), and EGSync (right) on the ring topology, respectively.
640
VOL. 25,
NO. 3,
MARCH 2014
Fig. 13. EGSync simulation results for different size of networks with line (left) and ring (middle) topologies. On the right, simulation results of
networks consisting 100 nodes which have different neighbor densities are presented.
5.2
CONCLUSION
YILDIRIM AND KANTARCI: EXTERNAL GRADIENT TIME SYNCHRONIZATION IN WIRELESS SENSOR NETWORKS
ACKNOWLEDGMENTS
Kasim Sinan Yildirim acknowledges the Turkish Scientific
_
BITAK)
and Technical Research Council (TU
for supporting
this work through a domestic PhD scholarship program
(BAYG-2211).
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
641
1232
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO. 3, MARCH 2014
AbstractBy combining a printed reflector-backed one-wavelength bowtie antenna and a printed double loop antenna, a new
wideband high-gain antenna element is demonstrated. Due to its
low-profile structure of
, it is attractive to be used in
an array environment. Experimentally, a series of antenna arrays
operated at 60 GHz are designed and fabricated. They achieve
wide impedance bandwidths covering the 5764 GHz unlicensed
frequency band. Over this band, the measured antenna gains are
ranged from 14.515.5 dBi, 18.320.1 dBi and 22.525.2 dBi for
4-element, 14-element and 50-element antenna arrays, respectively. The measured radiation pattern has low cross-polarization
level and low back radiation. The proposed low-profile antenna
and arrays are low-cost in fabrication as they are simply made on
a single-layer printed circuit board.
Index TermsAntenna array, low-profile antenna, wideband
antenna, 60 GHz radio.
I. INTRODUCTION
0018-926X 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1233
TABLE I
DIMENSIONS FOR THE PROPOSED ANTENNA
Fig. 2. Input impedances of the trapezoidal microstrip patch antenna, reflectorbacked one-wavelength bowtie antenna and loop-loaded reflector-backed onewavelength bowtie antenna.
1234
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO. 3, MARCH 2014
Fig. 5. Top and side views of the proposed antenna arrays. (a) 4-element array,
(b) 14-element array, (c) 50-element array.
C. Radiation Pattern
Fig. 3 shows a simulated radiation pattern of the loop-loaded
dipole antenna at 60 GHz. The maximum gain is about 10 dBi
in the broadside direction. The radiation pattern in the E-plane
is slightly asymmetrical and the cross-polarization level in the
H-plane is high, over 15 dB, which are due to the separation
in one arm of the bowtie to make space for the microstrip-line
to coplanar-waveguide transition. To justify this argument, a
loop-loaded electric dipole antenna excited at its center (without
the MSL to CPW transition) is examined as shown in Fig. 4. It
demonstrates that symmetrical radiation patterns in both principle planes are achieved. Since the cross-polarization levels are
lower than 40 dB in both planes, the
curves cannot be
shown in the figure.
III. ANTENNA ARRAYS
A. Array Geometries
,
Fig. 5(a) shows the 4-element antenna array (
) which adopts the parallel feeding method. This
method suffers from an asymmetrical radiation pattern with a
slightly high cross-polarization level because of the radiation
loss from the feed lines and the element asymmetrical geometry
caused by the MSL to CPW transition.
1235
Fig. 7. Simulated and measured gains of the 4-element, 14 element and 50-element antenna arrays.
B. Array Performances
Fig. 6. Simulated and measured SWRs. (a) 4-element array, (b) 14-element
array, (c) 50-element array.
GHz in free space). The Line 2 and Line 3 are responsible for
feeding the middle 28 elements. And the Line 1 and Line 4 feed
the upper and lower 22 elements. Notice that the Line 1, 2, 3 &
4 do not function as impedance transformers because of
lines between opposite neighboring elements [25]. The Line 5
and Line 6 with
are responsible for connecting the Line 1
& 2 and Line 3 & 4, respectively. The junctions in the middle
of the Line 2 and Line 3 are shifted upwards and downwards
for the Line 5 & 6, so
respectively for keeping a length of
that all elements can be excited in phase. The junctions of the
50 impedance
Line 2 & 3 joined to the feed point through
transformers
and tapered lines for impedance
transformation. The feed point is positioned at the lower side of
the substrate for the sake of confirming 180 phase difference
at the junctions of the Line 2 & 3. It can be observed that most
of the elements are connected together at the edges of the loops
. The connection
for reducing the element spacing to
between adjacent elements has no influence on the element
operation. The reason is that the main electromagnetic radiation
is from the radiating slots of the bowtie arms. Although some
part of the radiation is from the loops, especially at higher
frequencies, the adjacent elements are in phase and therefore
the connection between them will not affect the element radiation. The unconnected elements are not joined to their adjacent
elements for leaving enough space occupied by the vertical
lines (such as Line 5 & 6). This 50-element array, printed on the
single-layer Duroid 5880 substrate, is mounted on an aluminum
by using four metal
antenna fixture
screws (M1). A V-band connector (SHF: KPC185F302) and a
glass bead (SHF: GB185) are assembled together and launched
at the back of the antenna fixture. The inner conductor of the
connector is connected to the array feed network at the feed
point, as shown in Fig. 5. In practice, the three proposed arrays
may be integrated into a 60-GHz MMIC or silicon CMOS
system directly so that the antenna fixture is not a compulsory
part.
All simulations were implemented by EM simulation software Ansoft HFSS. In simulation, the setups for the substrate
and metal are referred to [25] (
,
,
. And the measurements on
impedance bandwidth, gain and radiation patterns were accomplished by a millimeter wave band Agilent E8361A Network
Analyzer and an in-house far-field millimeter-wave antenna
measurement system which was also employed in [14]. It is
noted that this system has been updated so as to broaden the
. The details on the
measurable elevation angle range to
measurement setup have been introduced in [14].
Fig. 6 shows the simulated and measured SWRs of the three
proposed arrays. The measured impedance bandwidths of the
are
4-element, 14-element and 50-element arrays
24.2% (51.8 to 66.1 GHz), 24% (55 to 70 GHz) and 16.7%
(56.7 to 67 GHz), respectively, which all cover 5764 GHz
band. Fig. 7 shows the simulated and measured broadside gains
of the proposed arrays. The measured 3-dB gain bandwidths
of the 4-element, 14-element and 50-element arrays are 20.9%
(54.3 to 67 GHz), 18.9% (56.4 to 68.2 GHz) and 12.5% (57
to 64.6 GHz), respectively. For the sake of the series/parallel
feeding method, the gain bandwidth tends to decrease, as the
number of antenna elements increases [27]. The measured peak
gains for the three arrays are 15.5 dBi at 60.2 GHz, 20.1 dBi
at 61 GHz and 25.2 dBi at 58.4 GHz. The overlapped 3 dB
gain bandwidth and the impedance bandwidth for the 4-element,
14-element and 50-element arrays is 19.6% (54.3 to 66.1 GHz),
18.9% (56.4 to 68.2 GHz) and 12.5% (57 to 64.6 GHz), respectively. Simulations are in relatively good agreement with results obtained from experiments. Fig. 8 shows the simulated and
measured radiation patterns of the 50-element antenna array.
Over the whole operating frequency band, the measured radiation pattern exhibits a cross-polarization level less than 30
dB, front-to-back ratio larger than 31 dB and the first side-lobe
level less than 9 dB. Both the measured and simulated results
agree very well. By applying the Tai and Pereira equation [28],
the antenna directivity can be evaluated. The approximated directivity at 60 GHz is 26.5 dBi, and the corresponding radiation
and aperture efficiencies are 63.7% and 72.5%, respectively.
1236
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 62, NO. 3, MARCH 2014
TABLE II
KEY DATA OF ANTENNA ARRAYS FOR 60 GHz APPLICATIONS
The aperture efficiency and radiation efficiency are the measured values at the operation frequency of each antenna (60 GHz for our works).
Aperture efficiency was calculated from simulated 90% radiation efficiency.
Fig. 9. Photos of the antenna array prototypes. (a) Top view of all proposed arrays, (b) backside view of the fabricated PCB for 50-element array, (c) backside
view of the 50-element array with antenna fixture, (d) side view of the 50-element array with antenna fixture.
achieve wide impedance bandwidth, high gain, and high radiation and aperture efficiencies.
IV. CONCLUSION
ACKNOWLEDGMENT
The authors would like to thank K. B. Ng for helping with the
antenna measurements.
REFERENCES
[1] T. Sehm, A. Lehto, and A. V. Raisanen, A large planar 39-GHz antenna array of waveguide-fed horns, IEEE Trans. Antennas Propag.,
vol. 46, no. 8, pp. 11891193, Aug. 1998.
[2] P. Smulders, Exploiting the 60 GHz band for local wireless multimedia access: Prospects and future directions, IEEE Commun. Mag.,
vol. 40, no. 1, pp. 140147, Jan. 2002.
[3] B. G. Porter, L. L. Rauth, J. R. Mura, and S. S. Gearhart, Dual-polarized slot-coupled patch antennas on Duroid with Teflon lenses for
76.5-GHz automotive radar systems, IEEE Trans. Antennas Propag.,
vol. 47, no. 12, pp. 18361842, Dec. 1999.
[4] G. P. Gauthier, J.-P. Raskin, L. P. B. Katehi, and G. M. Rebeiz, A
94-GHz aperture-coupled micromachined microstrip antenna, IEEE
Trans. Antennas Propag., vol. 47, no. 12, pp. 17611766, Dec. 1999.
[5] S.-S. Hsu, K.-C. Wei, C.-Y. Hsu, and H. Ru-Chuang, A 60-GHz millimeter-wave CPW-fed Yagi antenna fabricated by using 0.18CMOS technology, IEEE Electron Device Letters, vol. 29, no. 6, pp.
625627, Jun. 2008.
[6] O. Kramer, T. Djerafi, and K. Wu, Very small footprint 60 GHz
stacked yagi antenna array, IEEE Trans. Antennas Propag., vol. 59,
no. 9, pp. 32043210, Sep. 2011.
[7] K. Gong, Z. N. Chen, X. Qing, P. Chen, and W. Hong, Substrate integrated waveguide cavity-backed wide slot antenna for 60-GHz bands,
IEEE Trans. Antennas Propag., vol. 60, no. 12, pp. 60236026, Dec.
2012.
[8] T. Tomura, Y. Miura, M. Zhang, J. Hirokawa, and M. Ando, A 45
linearly polarized hollow-waveguide corporate-feed slot array antenna
in the 60-GHz band, IEEE Trans. Antennas Propag., vol. 60, no. 8,
pp. 36403646, Aug. 2012.
[9] X.-P. Chen, K. Wu, L. Han, and F. He, Low-cost high gain planar
antenna array for 60-GHz band applications, IEEE Trans. Antennas
Propag., vol. 58, no. 6, pp. 21262129, Jun. 2010.
[10] T. Sehm, A. Lehto, and A. V. Risnen, A high-gain 58 GHz box-horn
array antenna with suppressed grating lobes, IEEE Trans. Antennas
Propag., vol. 47, no. 7, pp. 11251130, Jul. 1999.
[11] B. Zhang and Y. P. Zhang, Grid array antennas with subarrays and
multiple feeds for 60-GHz radios, IEEE Trans. Antennas Propag., vol.
60, no. 5, pp. 22702275, May 2012.
[12] J. Xu, Z. N. Chen, X. Qing, and W. Hong, Bandwidth enhancement
for a 60 GHz substrate integrated waveguide fed cavity array antenna
on LTCC, IEEE Trans. Antennas Propag., vol. 59, no. 3, pp. 826832,
Mar. 2011.
[13] B. Pan, Y. Li, G. E. Ponchak, J. Papapolymerou, and M. M. Tentzeris,
A 60-GHz CPW-fed high-gain and broadband integrated horn antenna, IEEE Trans. Antennas Propag., vol. 57, no. 4, pp. 10501056,
Apr. 2009.
[14] K. B. Ng, H. Wong, K. K. So, C. H. Chan, and K. M. Luk, 60 GHz
plated through hole printed magneto-electric dipole antenna, IEEE
Trans. Antennas Propag., vol. 60, no. 7, pp. 31293136, Apr. 2012.
[15] L. Pazin and Y. Leviatan, A compact 60-GHz tapered slot antenna
printed on LCP substrate for WPAN applications, IEEE Antennas
Wireless Propag. Lett., vol. 9, pp. 272275, 2010.
[16] S. J. Franson and R. W. Ziolkowski, Gigabit per second data transfer
in high-gain metamaterial structures at 60 GHz, IEEE Trans. Antennas
Propag., vol. 57, no. 10, pp. 29132925, Oct. 2009.
[17] A. E. I. Lamminen, A. R. Vimpari, and J. Sily, UC-EBG on LTCC for
60-GHz frequency band antenna applications, IEEE Trans. Antennas
Propag., vol. 57, no. 10, pp. 29042912, Oct. 2009.
[18] H. Vettikalladi, O. Lafond, and M. Himdi, High-efficient and highgain superstrate antenna for 60-GHz indoor communication, IEEE
Antennas Wireless Propag. Lett., vol. 8, pp. 14221425, 2009.
[19] B. Biglarbegian, M. Fakharzadeh, D. Busuioc, M.-R. N. Ahmadi, and
S. S. Naeini, Optimized microstrip antenna arrays for emerging millimeter-wave wireless applications, IEEE Trans. Antennas Propag.,
vol. 59, no. 5, pp. 17421747, May 2011.
[20] L. Wang, Y.-X. Guo, and W.-X. Sheng, Wideband high-gain 60-GHz
LTCC L-probe patch antenna array with a soft surface, IEEE Trans.
Antennas Propag., vol. 61, no. 4, pp. 18021809, Apr. 2013.
1237
[21] T. Seki, N. Honma, K. Nishikawa, and K. Tsunekawa, Millimeterwave high-efficiency multilayer parasitic microstrip antenna array on
Teflon substrate, IEEE Trans. Antennas Propag., vol. 53, no. 6, pp.
21012106, Jun. 2005.
[22] H. Sun, Y.-X. Guo, and Z. Wang, 60-GHz circularly polarized U-slot
patch antenna array on LTCC, IEEE Trans. Antennas Propag., vol.
61, no. 1, pp. 430435, Jan. 2013.
[23] D. Liu, J. A. G. Akkermans, H.-C. Chen, and B. Floyd, Packages with
integrated 60-GHz aperture-coupled patch antennas, IEEE Trans. Antennas Propag., vol. 59, no. 10, pp. 36073616, Oct. 2011.
[24] E. Levine, G. Malamud, S. Shtrikman, and D. Treves, A study of microstrip array antennas with the feed network, IEEE Trans. Antennas
Propag., vol. 37, no. 4, pp. 426434, Apr. 1989.
[25] D. Liu, B. Gaucher, U. Pfeiffer, and J. Grzyb, Advanced Millimeterwave Technologies, Antennas, Packaging and Circuits. New York,
NY, USA: Wiley, 2009.
[26] S. Ahmed and W. Menzel, A novel planar four-quad antenna, in
Proc. Antennas and Propagation (EUCAP), Prague, Mar. 2012, pp.
19461949.
[27] S. Demir, Efficiency calculation of feed structures and optimum
number of antenna elements in a subarray for highest G/T, IEEE
Trans. Antennas Propag., vol. 52, no. 4, pp. 12041029, Apr. 2004.
[28] C. A. Balanis, Antenna Theory: Analysis and Design. New York, NY,
USA: Wiley, 2005.
Mingjian Li (S10) received the B.Sc. (Eng.) degrees
in electronic and communication engineering from
City University of Hong Kong in 2010, where he is
currently working toward the Ph.D. degree.
His recent research interests include wideband
antennas, millimeter-wave antennas and arrays, base
station antennas, circularly-polarized antennas and
small antennas.
Mr. Li received the Honorable Mention at the student contest of 2011 IEEE APS-URSI Conference
and Exhibition held in Spokane, US. He was awarded
the Best Student Paper Award (Second Prize) in the 2012 IEEE International
Workshop on Electromagnetics (IEEE iWEM2012) held in Chengdu, China.
Kwai-Man Luk (M79SM94F03) was born
and educated in Hong Kong. He received the B.Sc.
(Eng.) and Ph.D. degrees in electrical engineering
from The University of Hong Kong, in 1981 and
1985, respectively.
He joined the Department of Electronic Engineering, City University of Hong Kong, in 1985
as a Lecturer. Two years later, he moved to the
Department of Electronic Engineering, Chinese
University of Hong Kong, where he spent four years.
He returned to the City University of Hong Kong in
1992, and is currently Chair Professor of Electronic Engineering and Director
of State Key Laboratory in Millimeter waves (Hong Kong). He is the author
of three books, nine research book chapters, over 290 journal papers and 220
conference papers. He has received five US and more than 10 PRC patents. His
recent research interests include design of patch, planar and dielectric resonator
antennas, and microwave measurements.
Prof. Luk is a Fellow of the Chinese Institute of Electronics, PRC, a
Fellow of the Institution of Engineering and Technology, UK, a Fellow of the
Institute of Electrical and Electronics Engineers, USA and a Fellow of the
Electromagnetics Academy, USA. He is Deputy Editor-in-Chief of PIERS
journals. He was a Chief Guest Editor for a special issue on Antennas in
Wireless Communications published in the PROCEEDINGS OF THE IEEE in
July 2012. He was Technical Program Chairperson of the 1997 Progress in
Electromagnetics Research Symposium (PIERS), General Vice-Chairperson
of the 1997 and 2008 Asia-Pacific Microwave Conference (APMC), General
Chairman of the 2006 IEEE Region Ten Conference (TENCON), Technical
Program Co-chairperson of 2008 International Symposium on Antennas and
Propagation (ISAP), and General Co-chairperson of 2011 IEEE International
Workshop on Antenna Technology (IWAT). He received the Japan Microwave
Prize at the 1994 Asia Pacific Microwave Conference held in Chiba in December 1994, and the Best Paper Award at the 2008 International Symposium
on Antennas and Propagation held in Taipei in October 2008. He was awarded
the very competitive 2000 Croucher Foundation Senior Research Fellow in
Hong Kong and the 2011 State Technological Invention Award (2nd Honor)
of China.
220?M
1QL`9X3M@gk9`?g[[x`Q>9tQg`s @h9kt[@`t
4- 9k>@Xg`93@>M
@[9QX H9`>@s>
t9kk@sxh>
@?x
/@}@ks@!`LS`@@kS`L#kgxh
%)*)1%
@[9QX s9X}9?gk
MQ?9XLg>sQ>
@s
+-)- 5 0*,5 & &* &5 ,5 .5 (*',,5 .'5 ř &5
&.*.5 */ .5 .'5 '. &5
$.* #,5 #' 5 */ .*35 /&. '&# .35 (*'*$&5 &5 '.*5
*#0&.5 ./*,5 5
&,(. '&5
&5
, ,5
*'*5
.#5
$5
(*',, &5 ,5 (*,&.5 ,5 5 */ ./#5 #5 '*5 /.'$. '&5 &5 . ,5
((*5 5 ,3,.$5 '*5 .5 */ .*35 2.*. '&5 , ,5 &5
(*,&.. '&5 ,5 ,* 5 .5 ,5 0 5 &5 .*5 #'",5 2 $5
# &5 ' 5
.,5 '# 4. '&5 2 '& . '&5 &5
*' (5
0 .'*5 5 ((*5 (*,&.,5 &5 '0*0 15 '5 .5 '$(#.5
,3,.$5 &5 ,5 $ # ,5 '&5 .5 ,* (. '&5 '5 .5 $5
(*',, &5 #'* .$,5 ..5 *5 ((# 5 .'5 .5 *&.5 #'",5
,/5 ,5
*&,'*$5 5 &5 #' 5 .5 #'# 4. '&5 2 *'& . '&5
0-.2 )%+$5 ,. . &5 $', &5 '!.5 *'& . '&5
5 *0*,5 & &* &5
%*'&+*&%2
+`XtM@X9tt@kQstk@9t@?R`tMQsh9h@k
3M@[9S`Q?@9gB 0!Qstg9`9X@9AZ9Xhkg?x>ttgg=t9S`
RcBgk[9tQg` >g`>@pS`L Qts ?@sQL` tM@ [9t@kQ9Xs 9`?
>g[hg`@`ts tM9t >g`stQtxt@s Qt J`>tQg`9XQt h@kBgk[9a>@ 9`?
gtM@k k@X@}9`t B@9txk@s
>>gk?R`L tg tM@ st9t@gBtM@9kv R`
%`t@Lk9t@?Rk>xQt/@}@ks@!`LR`@@kS`L7
8 tMQshk9>tQ>@>9`=@
Lkgxh@? R` s@}@k9X sh@>Q9XQ9tQg` 9k@9s hkg?x>t t@9k?g~`s
sst@[ X@}@X 9`9XsQs hkg>@ss 9`9XsQs 9`? >Rk>xQt @vk9>tQg`
+`XtM@X9tt@kQstk@9t@?R`tMQsh9h@k
Rk>xQt@tk9>tQg`QstM@%k@}@ks@@`LS`@@kS`Lh9kvBg>xs@?
g` tM@ [Q>kg>MQh s>M@[9tQ> 9a? `@tXQst k@tkQ@}9X
!ss@`tQ9XX
tM@9`9XsQs hkg>@ssS`}gX}@s9 ?@>9hsxX9tQg` gBtM@ [Q>kg>MQh
tg 9hhX 9?@X9@kS`L[@tMg? tM9t >g`sQsts S` stkQhhS`L gBB tM@
[Q>kg>MQhk@L9k?R`LQts[@t9XX9@ks
"gk@9>MX9@k9`S[9LS`L
3M@k@sxXts9m@hk@s@`t@?S`1@>tQg`%%% 9`?AZ9XX1@>tQg`%5
>g`t9R`stM@>g`>XxsQg`s9`?Jtxk@~gkV
(M $>:BD<F;@M%9E8D<CF<BAM
%%
1613!) !10'-3%+*
3M@st9L@sgB %[9L@3QXRcLsst@[9k@XQst@?=@Xg~
g[hxt@tM@tk9`sX9tQg`}@>tgks=@t~@@`S[9L@sxsS`L
9` 9XLgkQtM[ Bgk 9xtg[9tQ> B@9txk@ ?@t@>tQg` 9`?
[9t>MR`L =@t~@@` `@QLM=gkR`L S[9L@s
3M@ 9XLgkQtO]
Qs9>xstg[S@?}@ksQg`gB1%"37 8
Xxst@k tM@tk9`sX9tQg` }@>tgks Bgk @9>M h9Rk gB S[9L@s
xsR`L 9 MQ@k9k>MQ>9X tk@@ =9s@? g` tM@ !x>XQ?@9`
?Qst9`>@
1>gk@tM@[9t>MS`L k@XQ9=QXQt =9s@? g`tM@h@k>@`t9L@
gBtk9`sX9tQg`}@>tgksHg[tM@X9kL@st>Xxst@k
hhX 9 k@LQg` Lkg~R`L 9XLgkQtM[ tg k@>g`stkx>t tM@
[gs9Q> ~M@k@tM@[gk@k@XQ9=X@ [9t>MR`Ls9k@[@kL@?
Fkst
$M %M'@7:9M3<><A:M
3MQs =Xg>V ?@9Xs ~QtM tM@ stQt>MR`L gB tM@ S[9L@s g=t9R`@?
Hg[ tM@ R]9LRcL hkg>@ss
%ts hxqgs@ Qs tg k@>g`stk|>t 9
[gs9Q>Bgk@9>MX9@kgB tM@[Q>kg>MQh
9
%tQsR]hgkv9`ttg`gt@tM9ttM@k@h@9t9=QXQtRcsx>Ms>@`9kQg
[xst =@ Rc>k@9s@? ?x@ tg tM@ MQLM sQ[QX9kQt9`? Mg[gL@`@Qt
gB tM@S[9L@s
=
"QLxk@ /
2 %[9L@s 9 9`? = k@hk@s@`t 9 [9t>MR`L @9[hX@
tM9t >g`t9S`s Mg[gL@`@Qt 9k@9s
!2 Qs S[hgkv9`t tg `gt@ tM@
`g`x`QBgk[ hgXQsMS`L k@hk@s@`t@? = tM@ ~MQt@ shgt Hg[ tM@
X@Ik@LQg`gBtM@R]9L@9
2 -B:<8M&7F9EM-B87><H7F<BAM 2 198B:A<F<BAM
"QLxk@
%[9L@s>9`?? sMg~s9[9t>MR`L@9[hX@~QtM
9sQL`QE>9`tsS[QX9kQt
$>:BD<F;@M%9E8D=F<BAM
3M@ stk9t@L 9hhXQ@? tg sgX}@ tMQs hkg=X@[ Qs ?Q}Q?@? Q`
Xg>9XQ9tQg`2 k@>gL`QtQg`
"QLxk@
"kg[ X@I tg kQLMt 9 XgLQ> L9t@ h9tt@p S[9L@
MQLMXQLMtS`L9>g`t9>thgTu9`?Qts[9sV
2 .<8DB8;<CM/7G<:7FBDM
3M@)Q>kg>MQh*9}QL9tgkQshk@s@`t@?9s9hkgLk9[9=X@tg
[9`9L@ X9kL@ S[9L@s k@hk@s@`t tM@ [@t9R`Bgk[9tQg` gB tM@
[Q>kg>MQh >g[hg`@`ts 9`? Qt Qs 9 >g[hX@t@ Lk9hMQ>9X xs@k
Sbu@kB9>@tg?@Fc@tM@`g?@stgTu@k>gbc@>ttMgs@>g[hg`@`ts
$>:BD<F;@M%9E8D<CF<BAM
"QLxk@
@XXX9@kk@LQg`~QtMXgLQ>L9t@s9`?Qts@`@kL
*9}QL9tQg`?@9Xs~QtMtM@[9`9L@[@`tgBX9kL@S[9L@s tM@
k@hk@s@`t9tQg` gB tM@ [@t9R`Bgk[9tQg` gB tM@ hk@}QgxsX
Q?@`tQE@? >g[hg`@`ts s@@ "QL 9`? tM@ S[9L@ 9XQLd[@`t
9[g`LtM@>xkk@`t9`?tM@9?U9>@`t [Q>kg>MQh X9@kS[9L@s
%`
gk?@k tg sgX}@ tM@ >g[hxt9tQg`9X XQ[Qt9tQg`s tg M9`?X@ X9kL@s
S[9L@s tM@ hkghgs@? sgXxtQg` Qs sS[QX9m tg #ggLX@ )9hs tM@
[gs9Q>s 9k@ ?Q}Q?@? S` tQX@s gB hQ@Xs 9`? g`X tMgs@
tQX@s k@jx@st@? = tM@ xs@k >g`>@pS`L MQsM@k Xg>9tQg` ~QtMRc
tM@ X9@k 9k@ Xg9?@? S` k@9X tR]@
3M@ `x[=@k gB tQX@s tg =@
Xg9?@? ?@h@`?s g` tM@ >g[hxt@k s>k@@` k@sgXxtQg` 9`? sS@
3M@k@>gL`QtQg`st9L@QsBg>xs@?tgF`?S`st9`>@sgBs@}@k9X
XgLQ> L9t@ h9tt@ps tOlgxLM tM@ hk@}Qgxs ?@AZ@? k@LQg`s
s
tM@k@9m@`g >M9`L@s R` s>9X@ 9`? kgt9tQg`S` tMQs s>@`9kQg tM@
-@9ksg` >gkk@X9tQg` Qs 9 sxQt9=X@ [@9sxk@ tg >g[hxt@ tM@
sS[QX9kQt =@t~@@` 9 h9tt@k` 9`? tM@Rk S`st9`>@s
$g~@}@k tM@
>g[hX@Qt k@sxXts Hg[ tM@ Rkk@LxX9k hgXQsMRcL tM9t 9Xt@ks tM@
R]9L@ Rct@`sQtQ@s 9`? >g`s@jx@`tX tM@ Fkst 9`? s@>g`?
st9tQstQ>9X [g[@`ts >g[hxt@? = tM@ -@9ksg` >gr@X9tQg`
3g
k@?x>@ tMQs S`>g`}@`Q@`>@ 9 =S`9k [9sV Qs 9hhXQ@? tg @9>M
>gr@X9tQg` ~QtM tM@ 9S[ tg MQLMXQLMt tM@ [gk@ k@XQ9=X@ XgLQ>
L9t@ h9kvs
3Mgs@ h9kvs xsx9XX =@Xg`L tg tM@ sQXQ>g` k@LQg` gB
tM@ XgLQ> L9t@s 9s >9` =@ g=s@k}@? R` "QL k@hk@s@`t@? = tM@
~MQt@L9hs
3M@`g?@?@AZQtQg`[g?xX@xs@sMQ@k9k>MQ>9Xtk@@stg?@AZ@
tM@Tu@k>gbc@>tQg`9[g`L [Q>kg>MQh >g[hg`@`ts
3M@s@tk@@s
9k@ Bgk[@? = Bgxk ?QBB@k@`t th@s gB `g?@ hgR`ts =@LSd`S`L
>g\@k ?Q}@kL@`t 9`? t@k[Rc9X
@LSd`S`L hgSbus xs@ tg =@ 9`
gxthxt >g`t9>t hgR`t gB 9 XgLQ> L9t@ >gp@ks 9k@ xs@? tgUgRc
t~g tk9>Vs ?Q}@kL@`t 9k@ XRV@ >g\@ks =xt Qt k@X9t@s [gk@tM9`
t~g tk9>Vs 9`? t@k[S`9X Qs ~M@k@ tM@ `g?@ FcQsM@? ~MQ>M
xsx9XX=@Xg`LstgtM@S`hxt>g`t9>thgRctgB9`gtM@kXgLQ>L9t@
+`>@9XXtM@S`t@k>gbc@>tQg`sM9}@=@@`?@AZ@? tM@`@tXQst
>9` =@9xtg[9tQ>9XX L@`@k9t@?S`9hkgB@ssQg`9X Bgk[9t 9sBgk
Rcst9`>@5@kQXgL
3M@>g[hxt9tQg`9X>gst?@h@`?sgBtM@`x[=@k gB?QBB@k@`t
XgLQ>L9t@sh9tt@psgBtM@@}9Xx9t@?[Q>kg>MQh9`?tM@`x[=@k
gBS`st9`>@stg=@Xg>9t@?
%`tM@@}9Xx9t@?[Q>kg>MQhtM@k@9k@
9kgx`?
?QBB@k@`t XgLQ>L9t@h9tt@k`s9`? [gk@gkX@ss
XgLQ> S`st9a>@s
2 9}@k9L@tS[@ gB [S`xt@s Bgk @9>M s@t gB
XgLQ> L9t@s Qs @stS[9t@? Bgk tM@ k@>gL`QtQg` hkg>@ss Rc 9
X9htgh
3MQs tS[@ S`>Xx?@s tM@ hkg>@ssS`L tR]@ 9s ~@XX 9s tM@
xs@kTu@k9>tQg`
%%%
0!14(31
3M@ @`tRk@ sst@[ M9s =@@` t@st@? sx>>@ssJXX g}@k 9
[Q>kg>MQh Bgk[@? = s@}@k9X X9@ks ~QtM 9 tgt9X `x[=@k gB
hQ@Xs h@k [gs9Q> sxh@kQgk tg ~M@k@9s tM@
`x[=@k gB XgLQ> L9t@s k@9>M@s tM@ gk?@k gB t@`s gB tMgxs9a?s
%5
+*(41%+*1
%` tMQs h9h@k 9 sst@[ tg 9xtg[9t@ tM@ [gs9Q>
k@>g`stkx>tQg` tM@ XgLQ> L9t@s Xg>9XQ9tQg` 2 k@>gL`QtQg` 9`?
tM@ ?@AZQtQg` gB tM@ [Q>kg>MQh >g[hg`@`ts R`t@k>gbc@>tQg` Qs
hk@s@`t@?
3M@ sst@[ Qs ?Q}Q?@? S` tOl@@ =Xg>Vs %[9L@
3QXS`L (gLQ>#9t@s(g>9XS9tQg` 2 /@>gL`QtQg`9`?)Q>kg>MQh
*9}QL9tgk
"Rkst =Xg>V xs@s gxk 1%"3 }@ksQg` tg stQt>M [gs9Q>
S[9L@s
3M@ Q?@`tQE>9tQg` gB tM@ XgLQ> L9t@s Qs 9>MQ@}@? =
[@9`sgBtM@-@9ksg`>gr@X9tQg`9ssS[QX9kQt[@9sxk@J`>tQg`
S` tM@ s@>g`? =Xg>V
3g k@?x>@ tM@ >g[hxt9tQg`9X >g[hX@Qt
tM@>gr@X9tQg` Qs g`X 9hhXQ@? tg9 }@k s[9XX s@t gB>9`?Q?9t@
hgSbus
"S`9XX tM@)Q>kg>MQh*9}QL9tgkQs9kQ>Mxs@kSbu@kB9>@
9=X@ tg k@hk@s@`t 9`? 9XQL` [Q>kg>MQh [gs9Q>s 9s ~@XX 9s tg
?@F`@Qts>g[hg`@`tsTu@k>gbc@>tQg`
"xtxk@k@s@9k>MXS`@s9k@Bg>xs@?g`tM@9xtg[9tQ> tk9>VR`L
gB tM@ `g?@s tM9t >gbc@>t tM@ [Q>kg>MQh >g[hg`@`ts
$M %M'@7:9M3<><A:M
Fkst S^iX@[@`t9tQg` ~9s t@st@? xsS`L -@9ksg`
>gr@X9tQg` =xtQts>g[hxt9tQg`9X>gst9`?@BE>Q@`>X@9?xstg
?@}@Xgh tM@ >xstg[S@? 1%"3 R]hX@[@`t9tQg`
/@L9k?R`L tM@
>g[hxt9tQg`9X >gst tMQs Q[hX@[@`t9tQg` gxth@kBgk[s tM@
-@9ksg` >gkk@X9tQg` }@ksQg` = 9 B9>tgk gB ~M@k@9s tM@
?@s>kQhtgk kg=xst`@ss h@`9XQ@s tM@ >g[hxt9tQg`9X >gst = 9
B9>tgk
>g`>@pR`L tM@ L@`@kQ> 1%"3 }@ksQg`
"gk R`st9`>@
tM@ 9}@k9L@ tR]@
k@jxRk@? Bgk tM@ >xstg[Q@? 1%"3
R]hX@[@`t9tQg` Bgk S[9L@s gB 2,, hR@Xs Qs 9kgx`?
s@>g`?s ~M@k@9s tM@ L@`@kQ> }@ksQg` Qs [gk@ gk X@ss ,
2
s@>g`?s
3M@[9t>MS`Lh@kBgk[9`>@Qs>9X>xX9t@?Bgk9sx=k@LQg`gB
g`@ gB tM@ [@t9X X9@ks Bgk[@? =
S[9L@s
3M@ >gkk@>t
[9t>MR`L h@k>@`t9L@ R` tM@ MgkSg`t9X ?Rk@>tQg` Qs 9=gxt
~M@k@9sR`tM@}@ktQ>9X?Rk@>tQg`Qs
$g~@}@k
tM@ =@st >g[=S`9tQg` gB =gtM ?Rk@>tQg`s ?@F`@? = tM@ k@LQg`
Lkg~R`L9XLgkQtN[g=t9R`s9
"%&,#$%*2
3M@9xtMgksgB tMQsh9h@ks~9`ttgtM9bWtM@>g`tkQ=xtQg`gB
Lk9?x9t@stx?@`tX=@kt)9kvG
('%)2
2 -B:<8M&7F9EM-B87><H7F<BAM 2 198B:A<F<BAM
k@>9XX gB
:a?9 hk@>QsQg`
Qs g=t9R`@? Bgk
tM@ >g`t9>t hgTus ?@t@>tQg` 9XLgkQtO] xsS`L 9 Lkgx`? tkxtM gB
>g`t9>t hgR`ts
%` tMQs >9s@ tM@ k@>9XX Qs hk@B@k9=X@ k9tM@k
tM9a hk@>QsQg`
Xg~ k@>9XX 9C@>ts tM@ XgLQ> L9t@ Xg>9tQg` 2
? :0?
?
'(? 6=-? ? 19:15+:1<-? 4)/-? -):;8-9? .864? #+)2-5<)81)5:?
->7615:9? 5:-85):165)2? 6;85)2? 6.? 647;:-8? &19165? <62?
?
56 77
?
'
(? )>? 2 $;>:-2))89? $? )5,? &)5? 662? ? #;8.? #7--,-,? ;7?
86*;9:? .-):;8-9? ;867-)5? 65.-8-5+-? 65? 647;:-8? &19165?
?
'(?
!
1686
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 62, NO. 8, AUGUST 2014
I. INTRODUCTION
A. Millimeter-Wave (mmWave) CMOS Power
Generation Challenges
Manuscript received November 09, 2013; revised February 09, 2014 and
March 13, 2014; accepted May 11, 2014. Date of publication June 12, 2014;
date of current version August 04, 2014.
The authors are with the Department of Electrical Engineering, Columbia
University, New York, NY 10027 USA (e-mail: ac3215@columbia.edu;
harish@ee.columbia.edu).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TMTT.2014.2327919
Fig. 1. (a)
of deep-submicrometer CMOS technology nodes from the
literature. (b) Supply voltage of deep-submicrometer CMOS technology nodes.
(c) Survey of the saturated output power achieved by reported RF and mmWave
CMOS PAs. (d) PAE achieved by reported RF and mmWave CMOS PAs.
0018-9480 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
CHAKRABARTI AND KRISHNASWAMY: HIGH-POWER HIGH-EFFICIENCY CLASS-E-LIKE STACKED mmWAVE PAs IN SOI AND BULK CMOS
1687
Fig. 2. (a) Stacked CMOS Class-E-like PA concept with voltage swings annotated in volts and (b) loss-aware Class-E design methodology for stacked CMOS
PAs.
1688
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 62, NO. 8, AUGUST 2014
CHAKRABARTI AND KRISHNASWAMY: HIGH-POWER HIGH-EFFICIENCY CLASS-E-LIKE STACKED mmWAVE PAs IN SOI AND BULK CMOS
1689
Fig. 3. (a) Theoretical and simulated (post-layout) output power and PAE and
(b) device size and theoretical device stress for the optimal design as a function
of number of devices stacked based on the loss-aware Class-E design methodology at 45 GHz in 45-nm SOI CMOS. Loss in dc-feed inductance is included
for theoretical results. Output power and PAE for a switch capacitor-based
model for the four-stacked configuration are also annotated.
1690
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 62, NO. 8, AUGUST 2014
TABLE I
NORMALIZED DEVICE PARAMETERS USED IN LOSS-AWARE CLASS-E ANALYSIS
Note:
,
,
,
, and
correspond to ON-resistance, gatesource, gatedrain, and drainsource capacitances, and
input power (to switch a device between the triode and cutoff regions), respectively, for a device of width
(including layout parasitics).
is the operating
frequency.
and
refer, respectively, to the high and low amplitude levels of a 45-GHz square-wave input signal.
Estimated in triode region of operation.
Estimated in cutoff region of operation.
Estimated as average of capacitance values in cutoff and triode regions of operation.
of the loss-aware Class-E design methodology. An excellent description of the characteristics of switching PAs can be found in
[29]. We have
(1)
where
and
are input power to the PA and the dc power
consumption, respectively. The loss in the PA
is given
by
(2)
(3)
(4)
where is the number of devices stacked in series,
is
the ON-resistance of each device in the -stacked PA,
is
the width of each device/switch, and
is the root mean
square (rms) value of the current flowing through the stack of
switching devices (excluding the output capacitance).
is the switching loss associated with the output capacitance of
the PA and is dependent on the topmost drain voltage value at the
switching instant. In general, at mmWave frequencies, the capacitive discharge loss is negligible compared to the conduction
loss in the switching device(s). This is evident from Table II,
where the conduction loss in the switch and the capacitive discharge loss have been tabulated for the optimal designs at different levels of stacking described in Fig. 3. Indeed, this reinforces our earlier assertion that the conventional ZVS/ZdVSbased Class-E design methodology is not applicable at mmWave
frequencies. Therefore, for the purpose of simplifying our analysis, we shall ignore the contribution of the term
to the
overall loss. In a switching PA, the average current drawn from
the supply is always proportional to
. The proportionality
constant depends on the tuning of the load network [29]. Tuning
of a Class-E load network is determined by the dc-feed inductance
and the load impedance
in relation to the device output capacitance. Since we are in a regime where conduction loss is significant, the proportionality constant will also depend on the value of the total switch ON-resistance
relative to the output capacitance
. Since
is a technology constant, specifying ,
, , and
completely characterizes the tuning of the stacked Class-E PA.
CHAKRABARTI AND KRISHNASWAMY: HIGH-POWER HIGH-EFFICIENCY CLASS-E-LIKE STACKED mmWAVE PAs IN SOI AND BULK CMOS
1691
TABLE II
CONDUCTION LOSS AND CAPACITIVE DISCHARGE LOSS FOR THE OPTIMAL DESIGNS AT DIFFERENT LEVELS OF STACKING DESCRIBED IN FIG. 3.
VALUES FOR THE WAVEFORM FIGURES OF MERIT
AND
FOR LOSS-AWARE AND ZVS-BASED DESIGNS ARE ALSO TABULATED
(5)
where
is a waveform figure of merit (FOM)
defined in [29] and
is the average supply current with
devices stacked.
For a stack of devices, the supply voltage scales linearly
with . On the other hand, in a switching PA, average supply
current is proportional to the product of output capacitance,
supply voltage, and the operating frequency
, the constant
of proportionality being dependent on the tuning of the circuit. The linear dependence on output capacitance is simply
an artifact of circuit scaling properties, while the linear scaling
with supply voltage arises from the fact that switching PAs are
linear with respect to excitations at the drain node (e.g., supply
voltage) [29]. Denoting the impedance of the device output capacitance at the fundamental frequency by
is defined in [29] as
(6)
is the
in (6), we get
(7)
Consequently,
(8)
(9)
where is a technology- and frequency-dependent constant
of proportionality that results from the input power functions
shown in Table I.
Table II lists the values of the waveform figures of merit for
designs based on the loss-aware and ZVS methodologies for
different levels of stacking. While the waveform metric
is
comparable for both, the loss-aware methodology shapes the
waveforms to minimize
, thereby yielding optimal designs
with highest possible PAE.
The foregoing expression captures the variation in PAE in
terms of technology constants and number of devices stacked.
The only design-related variables in this expression are the waveform figures of merit. The third term captures the PAE benefit of
stacking since output power is increased quadratically, but input
power is only provided to the bottom device in the stack. This explains why PAE improves when one goes from a single-device
Class-E-like PA to a two-stacked Class-E-like PA in Fig. 3(a).
However, as stacking is increased beyond two devices, the benefits from the third term wear off and the second term causes a
reduction in PAE due to a reduction in drain efficiency. It is well
known that
is the technology constant (which we
shall refer to as the switch time constant) that determines the
drain efficiency of a Class-E PA [29] for a given operating frequency and this constant degrades linearly for an -stacked device since the ON-resistances add in series while the output capacitance remains that of a single device. However, the loss-aware
Class-E design methodology optimizes the output network tuning
to ensure that the PAE degradation is gradual as stacking is increased by minimizing the
product. The benefits of
the loss-aware Class-E tuning methodology over the ZVS/ZdVSbased tuning methodology can be appreciated in Fig. 4, where the
product for the loss-aware design technique can be
observed to be lower than that corresponding to the ZVS design
methodology by a factor of 23 depending on the number of devices stacked.
The preceding analysis also highlights the importance of the
switch time constant as a technology metric that determines the
efficiency of switching PAs. For linear-type PAs,
is a sufficient metric to gauge the PA efficiency. For switching-type PAs,
determines the input power requirements (via the technology- and frequency-dependent constant ) while the switch
time constant determines the drain efficiency.
As the levels of stacking are increased, the switch time
constant becomes more significant than
. As can be seen in
1692
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 62, NO. 8, AUGUST 2014
However, interwinding and self-resonant capacitances introduce asymmetry in transformer power combiners, degrade
efficiency, and cause stability problems, usually permitting a
maximum of two transformer sections to be combined in series
[4]. Ignoring the effect of parasitic capacitances, a two-section
series transformer combiner is used to power-combine two
two-stacked PAs. The secondary inductance is chosen for
maximum efficiency subject to a 50- load, and the PAs are appropriately scaled to drive the load impedance presented by the
primary of the transformer. As shown in Fig. 5(a), transformer
power combining utilizing two-stacked PAs can yield results
similar to stacking only under ideal conditions and is fundamentally limited to two-way combining. The corresponding
results for Wilkinson and transformer-based power combining
using one-stacked (single device) Class-E-like PAs obtained
from the loss-aware design methodology are also included to
emphasize the inefficacy of the traditional design technique of
using single-device PAs for high-power amplification.
E. Stacking Versus Impedance Transformation
The efficiency of the alternative technique of impedance
transformation is dependent on the steepness of transformation as well as the topology of the impedance transformation
network. The two- and one-stacked Class-E-like PAs in 45-nm
SOI obtained from the loss-aware Class-E design methodology
at 45 GHz are again employed for the purpose of comparison.
In order to achieve output power comparable to those obtained
CHAKRABARTI AND KRISHNASWAMY: HIGH-POWER HIGH-EFFICIENCY CLASS-E-LIKE STACKED mmWAVE PAs IN SOI AND BULK CMOS
from device stacking, the Class-E-like PAs are scaled appropriately while an impedance transformation network is used
to transform the 50- load to the corresponding lower load
impedance for the scaled PAs. A two-element LC impedance
transformation network is designed and used in each case.
The quality factors of the inductor and capacitor are assumed
to be 15 and 10, respectively, at 45 GHz, based on measured
characterizations of inductors, capacitors, and transmission
lines in the 45-nm SOI CMOS technology [20]. A comparison
of the PAEs of impedance transformation and device stacking
is summarized in Fig. 5(b). Device stacking results in designs
with 1030% higher efficiency for the same output power
compared to impedance transformation.
Once device stacking is exploited to the limit as dictated by
secondary breakdown mechanisms (e.g., that of the BOX in
the SOI), it is interesting to consider the combination of device
stacking with impedance transformation and/or power combining to achieve watt-class output power levels at mmWave
frequencies.
1693
Fig. 6. Close-up of: (a) devices and (b) devices with connections to gate capacitors in four-stacked PA layout implemented in 45-nm SOI CMOS.
The measured
and
of the test structures of the
individual power devices used in the implemented 45-nm
SOI two-stacked and four-stacked PAs (distinct from the
custom stacked layout as discussed earlier) are shown in
Fig. 8(a) and (b), respectively. Device measurements were
conducted up to 65 GHz using a pair of coaxial 1.85-mm
(dc65 GHz) groundsignalground (GSG) probes, calibrated
at the probe tip planes. The industry-standard open-short
de-embedding was performed to a reference plane at the top of
the gate and drain vias. The measured
and
were obtained by extrapolating the measured Masons unilateral power
gain
and
at 20 dB/decade. The measured is observed
to have 20-dB/decade slope up to 65 GHz and the modeled
exhibits the same slope up to
. Peak
of 180 GHz
and 190 GHz are achieved for these power devices.
It is difficult to achieve
for power devices that is similar to that of smaller devices due to layout challenges [33]. For
reference, our measurements reveal that a 1 m 10 40 nm
device achieves an
of 250 GHz in this technology [30].
The use of a compact device layout with a continuous array of
large number of gate fingers with large finger width reduces the
parasitic capacitance and causes the 204- m device to have a
higher
compared to the 115- m device However, the layout
1694
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 62, NO. 8, AUGUST 2014
CHAKRABARTI AND KRISHNASWAMY: HIGH-POWER HIGH-EFFICIENCY CLASS-E-LIKE STACKED mmWAVE PAs IN SOI AND BULK CMOS
1695
Fig. 10. Simulated voltage profiles for two-stacked Class-E-like PA: (a)
without tuning inductor and (b) with tuning inductor. (c) Close-up of voltage
profiles with (bottom) and without (top) tuning inductor.
Fig. 11. Schematic of the two-stage 45-nm SOI CMOS -band Class-E-like
PA with a two-stacked driver stage and a four-stacked main PA.
unchanged at
until the end of the OFF half-cycle,
when the drain voltage of the top device reduces to
and the top and bottom node voltages roll-off in tandem thereafter. Introducing an inductor at the intermediate node results in
a Class-E-like voltage profile [see Fig. 10(b)], which causes the
top device to turn back on earlier during the latter part of the
OFF half-cycle, as shown in Fig. 10(b). This leads to additional
power loss in the top device. Consequently, no tuning inductor
is used in designing the two-stacked PA. For the four-stacked
PA, a tuning inductor at
is seen to provide benefit. Intuitively, a four-stacked configuration can be viewed as a stack of
two two-stacked PAs with the inductor serving as an inter-stage
tuning element. Fig. 9(d) shows the drain waveforms for the
four-stacked PA. As is evident, drainsource voltage swings
are almost equally shared across all four devices. The lack of
a tuning inductor at
results in a relatively flat-topped waveform. This is to be expected in view of the foregoing discussion
for the two-stacked PA. The situation is somewhat different for
node
. Despite the absence of a tuning inductor, we can observe a Class-E-like waveform even when device
is off. This
is a consequence of capacitive coupling through
and
of
(in conjunction with capacitive voltage division due to presence of the 80-fF gate capacitor), which induces voltage swing
at
when
is not conducting. This eliminates the need for
a tuning inductor at
. A similar voltage coupling does occur
for
as well. However, in that case, the coupling is through
two levels of devices and the resulting series connection of intrinsic capacitances reduces the strength of the voltage coupled
to
. Since
and
can be viewed as a two-stacked PA
with the tuning inductor serving as the choke inductance in large
signal, a tuning inductor is not required at
(as discussed earlier).
Another technique for inducing voltage swing at the intermediary nodes in a stacked configuration is through the use of capacitive charging acceleration. The work in [24] describes two
methods for accomplishing this. The first is by placing an explicit capacitor between every pair of intermediary nodes and
the second is using the inherent drainbulk capacitance of a device by connecting the bulk and source nodes of stacked devices along with appropriate device sizing. The first method is
1696
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 62, NO. 8, AUGUST 2014
operation of the PA. Ignoring these losses in the theoretical analysis results in PAE higher than what is obtained from actual device-based simulations.
For linear stacked PAs, the impact of Elmore delay can
be accounted (and compensated) for theoretically [28] since a
small-signal model for the devices is used for the preliminary
analysis and design procedure. A similar endeavor for stacked
switching PAs is challenging owing to nonlinear operation of the
circuit. Consequently, we adopt a simulation-based approach to
investigate this effect for a four-stacked configuration (without
the tuning inductor) using a switch capacitor-based model for
the devices. The resulting circuit resembles that in Fig. 9(b)
(sans the tuning inductor, the input matching network, and with
a square-wave drive instead of a sinusoidal input) with each device modeled as a switch augmented with intrinsic device capacitances. Layout parasitics (capacitors, resistors, and inductances) based on Fig. 7 are also incorporated in the circuit to
facilitate better correlation with device-based results. The delay
in the switch-voltage waveforms exhibit close correspondence
with those obtained from device-based simulations as well, as
shown in Fig. 16. Since the voltage profiles confirm no significant delay, the phenomenon of Elmore delay is not a concern
in stacked switching PAs. The resulting output power and PAE
are reported in Fig. 3(a). The reduction in efficiency ( 20%) for
the switch capacitor-based model indicates that capacitive discharge loss at intermediate nodes is a more important practical
consideration. These results, in conjunction with the comparison presented in Fig. 15 indicate that ignoring Elmore delay
and the additional loss mechanisms in the theoretical analysis is
not a crippling limitation since it does not significantly alter the
waveform characteristics, and hence, the impact on switching
behavior of the stacked configuration at mmWave frequencies.
One would therefore obtain output power similar to that predicted by theory, but at a lower PAE, which follows the theoretical trend [see Fig. 3(a)]. This also demonstrates the efficacy of
CHAKRABARTI AND KRISHNASWAMY: HIGH-POWER HIGH-EFFICIENCY CLASS-E-LIKE STACKED mmWAVE PAs IN SOI AND BULK CMOS
Fig. 16. (a) Post-layout simulated device voltages for the four-stacked PA prototype without tuning inductor in 45-nm SOI CMOS. (b) Simulated switch voltages for the same four-stacked configuration without tuning inductor, using a
switch capacitor-based model for the devices and layout parasitics from Fig. 7.
1697
A. Small-Signal Measurements
The small-signal measurement setup is calibrated at the probe
tip planes. The small-signal measurements are conducted up to
65 GHz using an Anritsu 37397E Lightning vector network analyzer (VNA). Figs. 18 and 19 illustrate the simulated and measured small-signal -parameters of the two-stacked PA and the
four-stacked PA implemented in 45-nm SOI CMOS. The measured peak gain of the two-stacked PA is 13.5 dB at 46 GHz
1698
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 62, NO. 8, AUGUST 2014
CHAKRABARTI AND KRISHNASWAMY: HIGH-POWER HIGH-EFFICIENCY CLASS-E-LIKE STACKED mmWAVE PAs IN SOI AND BULK CMOS
1699
Fig. 24. Measured gain, drain efficiency and PAE as a function of output power
for: (a) the 45-nm SOI four-stacked Class-E-like PA with the tuning inductor
eliminated through laser trimming at 42.5 GHz and (b) the 45-nm SOI fourV,
V,
stacked Class-E-like PA at 47.5 GHz (
V,
V, and
V for both designs).
Fig. 23. Measured gain, drain efficiency, and PAE as a function of output power
for: (a) the 45-nm SOI two-stacked Class-E-like PA at 47 GHz (
V,
V,
V) and (b) the 65-nm differential two-stacked
Class-E-like PA at 47.5 GHz (
V,
V,
V).
20.3 dBm at 47.5 GHz at a peak PAE of 19.4%. For the trimmed
version of the four-stacked PA without the tuning inductor, a
peak PAE of 18.3% was achieved at 42.5 GHz along with a
saturated output power of 20.3 dBm. Unlike the two-stacked
PA, the measured performance metrics of the four-stacked
PAs (particularly efficiency) are somewhat lower than those
predicted by simulations. This is an indication of unmodeled
active losses, as there is good correspondence between the
measured and simulated characteristics of the various passive
components [20]. The loss in the active components depends on
a proper choice of device layout, as well as accurate modeling,
as discussed in Section III.
Large-signal measurements were also conducted for the twoand four-stacked PAs across frequency (at the optimal bias
point) and for different supply voltages (at a fixed frequency,
keeping gate biases constant). The results are depicted in
Figs. 25 and 26. Large-signal measurement beyond 48 GHz
was limited by the characteristics of the measurement equipment (specifically, the Quinstar PA used to drive the PAs
under test, as well as the isolator, dual-directional coupler,
and the power sensors used in the measurement setup). Unlike
the two-stacked PA, the output power does not increase with
increasing supply voltage for the four-stacked prototype. Once
again, this can probably be attributed to the device layout
discussed previously that results in lower
.
This hypothesis is tested in measurement. As discussed previously, an important characteristic of switching PAs is linearity
with respect to supply voltage, which causes the average supply
current and the output power to scale linearly and quadratically
with supply voltage, respectively. This unique feature distinguishes switching PAs from the class of linear PAs. At mmWave
1700
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 62, NO. 8, AUGUST 2014
Fig. 27. Measured and expected: (a) average supply current versus
and (b) saturated output power versus
for two-stacked Class-E-like
PA in 45-nm SOI CMOS. The profiles display the linearity with respect to
supply voltage associated with switching Class-E PAs, thereby establishing the
Class-E-like characteristics of the PA even at mmWave frequencies.
Fig. 25. Measured gain, saturated output power, drain efficiency, and PAE:
(a) across frequency (
V,
V,
V) and
(b) across supply voltage at 47 GHz of the 45-nm SOI two-stacked Class-E-like
V,
V).
PA (
Fig. 26. Measured gain, saturated output power, drain efficiency, and PAE:
(a) across frequency (
V,
V,
V,
V) and (b) across supply voltage at 47 GHz of the 45-nm SOI four-stacked
Class-E-like PA.
frequencies, the various sources of nonidealities result in deviation from ideal Class-E characteristics. Thus, the scaling of
supply current and output power with supply voltage can be utilized as a useful metric to determine the extent of switching
characteristics of a PA in the mmWave regime. Fig. 27 illustrates the measured average supply current and saturated output
power of the two-stacked PA in 45-nm SOI CMOS as a function of
and
, respectively. The respective linear trends
Fig. 28. Measured and expected: (a) average supply current versus
and
(b) saturated output power versus
for four-stacked Class-E-like PA in
45-nm SOI CMOS. The profiles do not display the linearity with respect to
supply voltage characteristic of switching Class-E PAs, owing to layout-induced
increased gate resistance, which prevents hard switching at mmWave frequencies.
Fig. 29. (a) Measured small-signal -parameters and (b) measured gain, drain
efficiency, and PAE as a function of output power for the two-stage 45-nm SOI
PA comprising a four-stacked main PA and a two-stacked driver stage at 47 GHz
V,
V,
V,
V,
V,
(
V,
V, and
V). Power consumption
mW under small-signal operation.
CHAKRABARTI AND KRISHNASWAMY: HIGH-POWER HIGH-EFFICIENCY CLASS-E-LIKE STACKED mmWAVE PAs IN SOI AND BULK CMOS
1701
TABLE III
COMPARISON OF FABRICATED CLASS-E-LIKE PAs WITH STATE-OF-THE-ART CMOS AND SiGe mmWAVE PAs
(REFERENCES ARE ORGANIZED IN ORDER OF DECREASING PAE, FOR CMOS AND SiGe PAs SEPARATELY)
Defined as
dBm
Gain dB
Ideal external lossless output balun assumed.
Uses off-chip bias-T for providing power supply.
GHz
Gain(dB)
(10)
where is the operating frequency in gigahertz, takes into account four important performance metrics of a PA. The implemented single-stage prototypes achieve ITRS FOM comparable
to current state-of-the-art mmWave CMOS PAs and the highest
1702
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 62, NO. 8, AUGUST 2014
amongst fully integrated PAs, which do not employ power combining. In particular, the two-stage cascade PA in 45-nm SOI
CMOS achieves the highest ITRS FOM amongst PAs, which
do not employ power combining, and second highest overall.
(14)
VI. CONCLUSION
(15)
This work indicates that stacked switching CMOS PAs potentially take us one step closer to implementing efficient PAs
in CMOS with watt-level output power at mmWave frequencies
for the first time. Topics for future research include large-scale
low-loss power-combining techniques so that multiple such PAs
may be combined to approach watt-class output power, and linearizing architectures that enable such mmWave switching-type
PAs to be used with complex modulation formats with high average efficiency.
(16)
The expressions for
(17)
APPENDIX
and
(18)
The loss in the switch and the dc-feed inductance are given by
(19)
(11)
and
(20)
respectively, while the input power
required to switch a
device between the triode and cutoff regions is approximated as
(21)
(12)
where
where
is the input capacitance of the
bottom device in the triode region,
is the input drive level
in the ON half-cycle and is a fitting parameter determined
from schematic simulations. The foregoing lead to a complete
expression for PAE,
(22)
where
(23)
and
(24)
(25)
(13)
,
and
are constants dewhile
termined by imposing continuity of dc-feed inductor current
CHAKRABARTI AND KRISHNASWAMY: HIGH-POWER HIGH-EFFICIENCY CLASS-E-LIKE STACKED mmWAVE PAs IN SOI AND BULK CMOS
REFERENCES
[1] P. Haldi, G. Liu, and A. Niknejad, CMOS compatible transformer
power combiner, Electron. Lett., vol. 42, no. 19, pp. 10911092, Sep.
2006.
[2] Y. Zhao, J. Long, and M. Spirito, Compact transformer power combiners for millimeter-wave wireless applications, in IEEE Radio Freq.
Integr. Circuits Symp., May 2010, pp. 223226.
[3] D. Chowdhury, C. Hull, O. Degani, Y. Wang, and A. Niknejad, A fully
integrated dual-mode highly linear 2.4 GHz CMOS power amplifier for
4G WiMax applications, IEEE J. Solid-State Circuits, vol. 44, no. 12,
pp. 33933402, Dec. 2009.
[4] J. W. Lai and A. Valdes-Garcia, A 1 V 17.9 dBm 60 GHz power amplifier in standard 65 nm CMOS, in IEEE Int. Solid-State Circuits Conf.
Tech. Dig., Feb. 2010, pp. 424425.
[5] D. Zhao, S. Kulkarni, and P. Reynaert, A 60 GHz dual-mode power
amplifier with 17.4 dBm output power and 29.3% PAE in 40-nm
CMOS, in Proc. ESSCIRC, Sep. 2012, pp. 337340.
[6] M. Bohsali and A. Niknejad, Current combining 60 GHz CMOS
power amplifiers, in IEEE Radio Freq. Integr. Circuits Symp., Jun.
2009, pp. 3134.
[7] T. Dickson, K. H. K. Yau, T. Chalvatzis, A. Mangan, E. Laskin, R.
Beerkens, and P. Westergaard et al., The invariance of characteristic
current densities in nanoscale MOSFETs and its impact on algorithmic
design methodologies and design porting of Si(Ge) (Bi)CMOS highspeed building blocks, IEEE J. Solid-State Circuits, vol. 41, no. 8, pp.
18301845, Aug. 2006.
[8] A. Niknejad, S. Emami, B. Heydari, M. Bohsali, and E. Adabi,
Nanoscale CMOS for mm-wave applications, in IEEE Compound
Semicond. Integr. Circuits Symp., Oct. 2007, pp. 14.
[9] B. Heydari, M. Bohsali, E. Adabi, and A. Niknejad, Millimeter-wave
devices and circuit blocks up to 104 GHz in 90 nm CMOS, IEEE J.
Solid-State Circuits, vol. 42, no. 12, pp. 28932903, Dec. 2007.
[10] S. Nicolson, A. Tomkins, K. Tang, A. Cathelin, D. Belot, and S.
Voinigescu, A 1.2 V, 140 GHz receiver with on-die antenna in 65 nm
CMOS, in IEEE Radio Freq. Integr. Circuits Symp., Jun. 2008, pp.
229232.
[11] E. Johnson, Physical limitations on frequency and power parameters
of transistors, in IRE Int. Convention Rec., Mar. 1965, vol. 13, pp.
2734.
[12] I. Sarkas, A. Balteanu, E. Dacquay, A. Tomkins, and S. Voinigescu, A
45 nm SOI CMOS class-D mm-wave PA with 10 Vpp differential
swing, in IEEE Int. Solid-State Circuits Conf. Tech. Dig., Feb. 2012,
pp. 8890.
[13] A. Mazzanti, L. Larcher, R. Brama, and F. Svelto, Analysis of reliability and power efficiency in cascode class-E PAs, IEEE J. Solid-State
Circuits, vol. 41, no. 5, pp. 12221229, May 2006.
[14] A. Ezzeddine and H. Huang, The high voltage/high power FET
(HiVP), in IEEE Radio Freq. Integr. Circuits Symp., Jun. 2003, pp.
215218.
[15] S. Pornpromlikit, J. Jeong, C. Presti, A. Scuderi, and P. Asbeck, A
33-dBm 1.9-GHz silicon-on-insulator CMOS stacked-FET power
amplifier, in IEEE MTT-S Int. Microw. Symp. Dig., Jun. 2009, pp.
533536.
[16] J. McRory, G. Rabjohn, and R. Johnston, Transformer coupled
stacked FET power amplifiers, IEEE J. Solid-State Circuits, vol. 34,
no. 2, pp. 157161, Feb. 1999.
[17] A. Balteanu, I. Sarkas, E. Dacquay, A. Tomkins, and S. Voinigescu,
A 45-GHz, 2-bit power DAC with 24.3 dBm output power,
14 Vpp differential swing, and 22% peak PAE in 45-nm
SOI CMOS, in IEEE Radio Freq. Integr. Circuits Symp., Jun.
2012, pp. 319322.
[18] S. Pornpromlikit, H.-T. Dabag, B. Hanafi, J. Kim, L. Larson, J. Buckwalter, and P. Asbeck, A -band amplifier implemented with stacked
45-nm CMOS FETs, in IEEE Compound Semicond. Integr. Circuit
Symp., Oct. 2011, pp. 14.
[19] A. Chakrabarti, J. Sharma, and H. Krishnaswamy, Dual-output
stacked class-EE power amplifiers in 45 nm SOI CMOS for -band
applications, in IEEE Compound Semicond. Integr. Circuit Symp.,
Oct. 2012, pp. 14.
[20] A. Chakrabarti and H. Krishnaswamy, High power, high efficiency
stacked mmWave class-E-like power amplifiers in 45 nm SOI CMOS,
in IEEE Custom Integr. Circuits Conf., Sep. 2012, pp. 14.
1703
1704
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 62, NO. 8, AUGUST 2014