520 JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 39, NO. 2, JANUARY 15, 2021
Analog Coherent Detection for Energy Efficient
Intra-Data Center Links at 200 Gbps Per Wavelength
Takako Hirokawa , Student Member, IEEE, Student Member, OSA, Sergio Pinna , Member, IEEE,
Navid Hosseinzadeh , Student Member, IEEE, Aaron Maharry , Hector Andrade , Junqian Liu,
Thomas Meissner, Stephen Misak , Student Member, IEEE, Ghazal Movaghar,
Luis A. Valenzuela , Student Member, IEEE, Yujie Xia, Shireesh Bhat, Fabrizio Gambini,
Jonathan Klamkin, Senior Member, IEEE, Senior Member, OSA,
Adel A. M. Saleh , Life Fellow, IEEE, Fellow, OSA, Larry Coldren , Life Fellow, IEEE, Fellow, OSA,
James F. Buckwalter , Senior Member, IEEE, and Clint L. Schow, Fellow, IEEE, Fellow, OSA
Abstract—As datacenters continue to scale in size, energy effi-
ciency for short reach (<2 km) links is a major factor for networks
that may connect hundreds of thousands of servers. We demon-
strate that links based on analog coherent detection (ACD) offer
a promising path to simultaneously achieving significantly larger
link budgets and improved link energy efficiency. A complete anal-
ysis is presented that considers the power consumption of all the
photonic and electronic components necessary to realize an ACD
link architecture based on 50 Gbaud (GBd) quadrature phase-shift
keying (QPSK) signaling combined with polarization multiplexing
to achieve 200 Gb/s/λ. These links utilize receivers that incorporate
an optical phase-locked loop (OPLL) to frequency- and phase-
lock the local oscillator (LO) laser to the incoming signal. QPSK
modulation offers compelling advantages both in achievable link
budget and in energy efficiency. Indeed, low-complexity electronics
based on limiting amplifiers can be used as opposed to the linear
front-ends, A/D converters, and digital signal processing (DSP)
required for higher-order QAM or PAM formats. Our analysis
indicates that links with 13 dB of unallocated budget operating
at error rates of <10
-12
can be achieved and is compatible with
higher error rates that require forward error correction (FEC).
We present a comparison of silicon and InP platforms and evaluate
both traveling-wave and segmented modulator designs, providing
an illustration of the wide design space before converging on the
Manuscript received July 14, 2020; revised September 21, 2020; accepted
October 5, 2020. Date of publication October 12, 2020; date of current version
January 15, 2021. This work was supported in part by the Advanced Research
Projects Agency-Energy under Grant ARPA-E, and in part by the U.S. Depart-
ment of Energy under Award DE-AR0000848. (Corresponding author: Takako
Hirokawa.)
Takako Hirokawa, Sergio Pinna, Navid Hosseinzadeh, Aaron Maharry,
Hector Andrade, Junqian Liu, Thomas Meissner, Stephen Misak, Ghazal
Movaghar, Luis A. Valenzuela, Yujie Xia, Fabrizio Gambini, Jonathan Klamkin,
Adel A. M. Saleh, Larry Coldren, James F. Buckwalter, and Clint L. Schow
are with the Electrical and Computer Engineering Department, University of
California, Santa Barbara, CA 93106-9560 USA (e-mail: takako@ucsb.edu;
pinna@ece.ucsb.edu; hosseinzadeh@ucsb.edu; amaharry@ucsb.edu; han-
drade@ucsb.edu; junqian@ucsb.edu; thomas_meissner@ucsb.edu; smisak@
ucsb.edu; ghazalmovaghar@ucsb.edu; valenzuela@ucsb.edu; yujiexia@
ucsb.edu; fgambini@ucsb.edu; klamkin@ece.ucsb.edu; adelsaleh@ece.
ucsb.edu; coldren@ece.ucsb.edu; buckwalter@ece.ucsb.edu; schow@ece.
ucsb.edu).
Shireesh Bhat was with the University of California, Santa Barbara, CA
93106-9560 USA. He is now with Juniper Networks, Sunnyvale, CA 94089
USA (e-mail: sbhat@ece.ucsb.edu).
Color versions of one or more of the figures in this article are available online
at https://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/JLT.2020.3029788
most promising architectures that maximize energy efficiency and
minimize laser power. We establish the theoretical potential to
achieve picojoule-per-bit energy efficiency targets.
Index Terms—Coherent detection, data center, energy efficiency.
I. INTRODUCTION
W
ITH ever-increasing demand for cloud services, evaluat-
ing interconnect technology benefits and tradeoffs antic-
ipates future deployments of the data center through scaling baud
rates, higher order modulation formats with more bits/symbol,
polarization multiplexing, and adding additional wavelength
division multiplexed (WDM) channels. Current data center links
rely on intensity-modulated direct detection (IMDD) schemes
due to their relative simplicity and correspondingly relatively
low cost and power consumption. However, scaling IMDD links
to 200 Gbps/lane will require a large jump in complexity and
power consumption. A recent study showed the potential of a
100 GBd PAM-4 link to operate over a 400 m link distance [1].
However, heavy equalization was required, with 71 feedforward
equalizer (FFE) taps and 15 decision feedback equalizer (DFE)
taps, just to achieve a pre-FEC (Forward Error Correction) bit
error ratio (BER) slightly below the soft decision (SD-FEC)
limit of 2 × 10
-2
. With such power-hungry equalization, the
required received optical power was > +7 dBm, likely de-
manding an unfeasible output power from the transmitter (TX)
source laser [1]. The limited prospects for scaling IMDD links
to 200 Gbps/lane and beyond have driven substantial interest in
developing a new generation of energy-efficient coherent links
designed specifically for intra-datacenter applications [2]–[5].
A recent paper by authors from the Alibaba Group presents
a detailed comparison of several variants of IMDD (PAM4,
CAP16, DMT) against digital coherent (PDM-16QAM) for
400G links, backed up with experimental results, using metrics
of minimizing laser and ASIC power consumption [4]. The
authors conclude that coherent links have lower laser power
requirements and comparable ASIC power dissipation and digi-
tal signal processing (DSP) complexity compared to the IMDD
approaches. Recent work from Google provides a comparison up
to 1.6 Tb/s, analyzing in detail multiple digital coherent (16, 32,
0733-8724 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Univ of Calif Santa Barbara. Downloaded on January 25,2021 at 22:50:15 UTC from IEEE Xplore. Restrictions apply.