TEF 2021: The Road Ahead Next Generation Optical Interfaces
Transcript of TEF 2021: The Road Ahead Next Generation Optical Interfaces
Ethernet Alliance
TEF 2021: The Road Ahead
Next Generation Optical Interfaces
2
The Ethernet AllianceGlobal Community of End Users, System Vendors, Component Suppliers & Academia
Our Mission
• To promote industry awareness, acceptance and
advancement of technology and products based on, or
dependent upon, both existing and emerging IEEE 802
Ethernet standards and their management.
• To accelerate industry adoption and remove barriers to
market entry by providing a cohesive, market-responsive,
industry voice.
• Provide resources to establish and demonstrate multi-
vendor interoperability.
Ethernet Alliance Strategy
3
● Facilitate interoperability testing
○ Industry Plug Fests supporting
member and technology initiatives
● Interoperability Assurance
○ PoE Certification Program
● Collaborative Interaction with
other Industry Organizations
○ Multiple SIGs, Applications and MSAs
○ Industry Consensus Building
● Global Outreach
○ Worldwide Membership
●Thought Leadership
○ EA Hosts Technology Exploration
Forums (TEFs)
○ Technology and Standards
incubation
●Promotion of Ethernet
○ Industry Analysts
○ Education
○ Marketing
■Trade shows & Panel Presentations
■White Papers, Blogs & Social Media
Expanding the Ethernet Ecosystem, Supporting Ethernet Development
NEXT GENERATION OPTICAL INTERFACESPanelists
Scott Schube, Intel “Scaling Bandwidth with Optical Integration”
Matt Traverso, Cisco Systems “Scaling Architecture for Next Gen Optical Links”
David Lewis, Lumentum “Performance Photonics Enabling next Generation Interfaces”
Xinyuan Wang, Huawei “Observations on the Rate of Beyond 400GbE, 800GbE an/or
1.6TbE”
Scaling Bandwidth with Optical Integration
5
Scott Schube, Intel
Scaling Bandwidth With Optical IntegrationScott Schube
Intel Silicon Photonics Products Division
Ethernet Alliance TEF, January 2021
Ethernet Alliance TEF, January 2021Silicon Photonics Product Division
• Cost, cost, cost• Cloud expectation of cost/bit parity or better from Day 1 for a new generation
• Power• Supply scale
Optics Bandwidth Scaling Requirements
7
Ethernet Alliance TEF, January 2021Silicon Photonics Product Division
• Cost, cost, cost• Cloud expectation of cost/bit parity or better from Day 1 for a new generation
• Power• Supply scale
But also:• Availability timing and risk• Backwards compatibility
• across 2 or even 3 speed/technology generations
• Support for various network architectures• E.g. sufficient switch radix
• Configuration/application flexibility• E.g. ability to support multiple reaches, different mixes of interfaces in common platform
• Support for multiple suppliers/sources• For availability + security of supply across multiple optics types
Optics Bandwidth Scaling Requirements
8
Ethernet Alliance TEF, January 2021Silicon Photonics Product Division
• Cost, cost, cost• Cloud expectation of cost/bit parity or better from Day 1 for a new generation
• Power• Supply scale
But also:• Availability timing and risk• Backwards compatibility
• across 2 or even 3 speed/technology generations
• Support for various network architectures• E.g. sufficient switch radix
• Configuration/application flexibility• E.g. ability to support multiple reaches, different mixes of interfaces in common platform
• Support for multiple suppliers/sources• For availability + security of supply across multiple optics types
Optics Bandwidth Scaling Requirements
9
Any next-generation technology or approach need to be evaluated on all of these
Ethernet Alliance TEF, January 2021Silicon Photonics Product Division
• More channels• Fibers• Wavelengths
• Higher baud rate / channel
• Advanced modulation formats (e.g. PAM4, QAM)
Optics Bandwidth Scaling Options
10
Ethernet Alliance TEF, January 2021Silicon Photonics Product Division
• More channels• Fibers
+ Lower baud rate = easier on device technology- Packaging cost/complexity, fiber cost
• Wavelengths+ Lower baud rate, lower fiber cost- Packaging cost/complexity, link budget
• Higher baud rate / channel+ Fewer channels = simpler packaging- Performance feasibility and yield, cost/complexity, lower
switch radix, backwards compatibility challenges
• Advanced modulation formats (e.g. PAM4, QAM)+ Lower baud rate- Performance feasibility and yield, cost/complexity, lower
switch radix, backwards compatibility challenges, higher performance FEC required w/ higher latency
Optics Bandwidth Scaling Options
11
Ethernet Alliance TEF, January 2021Silicon Photonics Product Division
• More channels• Fibers
+ Lower baud rate = easier on device technology- Packaging cost/complexity, fiber cost
• Wavelengths+ Lower baud rate, lower fiber cost- Packaging cost/complexity, link budget
• Higher baud rate / channel+ Fewer channels = simpler packaging- Performance feasibility and yield, cost/complexity, lower
switch radix, backwards compatibility challenges
• Advanced modulation formats (e.g. PAM4, QAM)+ Lower baud rate- Performance feasibility and yield, cost/complexity, lower
switch radix, backwards compatibility challenges, higher performance FEC required w/ higher latency
Optics Bandwidth Scaling Options
12
Traditionally, the optics
imperative has been to
serialize wherever
possible, because of
optical packaging
challenges/cost
Ethernet Alliance TEF, January 2021Silicon Photonics Product Division
Integration Changes the Game
For high-yield process technology, cost per channel drops with the
integration of more channels
13
Source: Internal Intel analysis and estimates; your mileage may vary
Integration changes the tradeoff between optics options, enabling “scale out” as well as “scale up”
Ethernet Alliance TEF, January 2021Silicon Photonics Product Division
Example: 16x100G Silicon Photonics Integrated Circuit
• 16-channel (1.6Tbps, 16x100G / 4x400G DR4+) PSM transmitter PIC
• On-die integrated lasers
• 112G ring-resonator modulators
• Mode-converters and V-grooves for cost-effective high-volume packaging
• Fully integrated Tx optics enables wafer-level test
• Supports redundant lasers if needed
14
• Suitable for 2x400G products starting to be deployed in the market this year
• Monolithically-integrated 8-channel WDM demonstrated to enable 2x400G FR4/LR4 on two fiber pairs for maximum interoperability, or 800G on single duplex fiber for maximum fiber efficiency (see right)
Example: 8x100G Silicon Photonics Integrated Circuit
Silicon Photonics Product Division Ethernet Alliance TEF, January 2021 15
System Level Optical Integration: Co-Packaged Optics• Co-packaged optics targeted to provide both lower power and
cost/bit• Demonstrated Intel 1.6T CPO optics and 12.8T switch system with CPO
shown here as an example
• Support for interoperability and backwards compatibility depends on implementation• To be most useful, need to support standard optical interfaces (e.g. FR4,
DR4, DR1/FR1) and electrical interfaces (e.g. XSR)
• Support for multi-sourcing and configuration flexibility also depends on implementation. A socketed/replaceable engine approach enables
• Multiple supplier sources – can mix and match silicon and optics• Multiple technologies can coexist within this envelope (e.g. remote vs.
integrated laser) for maximum innovation flexibility and options
• “Configure at manufacture” flexibility to support different types of interfaces in the same platform
• Reworkability in system manufacturing
Silicon Photonics Product Division Ethernet Alliance TEF, January 2021
Very Quick Note on Coherent
• As noted by many, coherent optics is getting more attractive for certain applications as its maturity goes up and cost and power comes down
• In addition to the two “marquee” challenges/goals for coherent to penetrate into shorter, less fiber-constrained data center applications (cost and power competitiveness: ultimate intercept for high-BW links likely, timing TBD), other issues that may constrain or delay coherent deployment in non-greenfield data center applications include
• Interoperability with current generations
• Backwards compatibility with prior generations
• Compatibility with network architectures and radix
• Coherent optics are coming into data center networks – but when and how far down?
16
Ethernet Alliance TEF, January 2021Silicon Photonics Product Division
Implications for Next-Generation Optics
• With integrated optics, “scaling out” by adding more lanes is an increasingly attractive solution to scale bandwidth
• E.g. for 25T and 51T, adding more lanes of 100G (4x100G > 8x100G > 16x100G or 32x100G)
• Most cost-effective (especially if leveraging high levels of integration)
• Less technologically risky (100G/lane starting to be deployed in the market now), can get to market faster
• Can preserve compatibility and interoperability with 400G infrastructure (Nx400G breakout)
• Preserves switch radix (Nx100G or Nx400G breakout) for full network connectivity
• Channel “scale out” can be combined with lane speed “scale up” to enable even higher bandwidths for 100T+ once 200G/lane is ready (e.g. 8x200G 1.6T, 32x200G 6.4T CPO, etc.)
• Integration at the system level with co-packaged optics integration promises further benefits in power and cost
17
Silicon Photonics Product Division Ethernet Alliance TEF, January 2021 18
Thank You
Scaling Architecture for Next Gen Optical Links
20
Matt Traverso, Cisco Systems
matt traversoSpecial thanks to Cisco development team – many many contributorsJanuary 28, 2021
Scaling Architecture for Next Gen Optical Link
© 2020 Cisco and/or its affiliates. All rights reserved.22
• Scaling to the next gen optical link requires careful optimization of the communication blocks
Digital Communication Fundamentals
channel
Carrier
SourceSourceEncode
ChannelEncode
DigitalModulator
SinkSourceDecode
ChannelDecode
DigitalDemodulator
Ref. Carrier
© 2020 Cisco and/or its affiliates. All rights reserved.23
• High speed optical Ethernet partitions the communications stack
• Digital modulator beings at Serializer in SerDes and extends thru the optical modulator
• Laser provides the carrier frequency
• Digital demodulator beings at detector and extends to the Deserializer which includes a CDR
• Clock & Data Recovery performs the decision and provides clock for parallel digital output
High Speed Optical Ethernet
channel
Carrier
SourceSourceEncode
ChannelEncode
DigitalModulator
SinkSourceDecode
ChannelDecode
DigitalDemodulator
TXPMD
MACPCSPMA
MACPCSPMA
RXPMD
ASIC
SerD
es
SerD
es
Digital
SerD
es
DRV
TIA Detector
Laser+
Modulator
© 2020 Cisco and/or its affiliates. All rights reserved.24
100 Gbps/λ Electro-Optical System
• Industry-first integrated linear CMOS TIA transceiver SoC for 100 Gbps/λ
• K. R. Lakshmikumar et al., "A Process and Temperature Insensitive CMOS Linear TIA for 100 Gb/s / lambda PAM-4 Optical Links," in IEEE Journal of Solid-State Circuits, vol. 54, no. 11, pp. 3180-3190, Nov. 2019, doi: 10.1109/JSSC.2019.2939652.
• Segmented MZI driver to achieve linear PAM-4 transmission
• Lowest power 100 Gbps/λ electro-optical system
10km
Golden
Eye
4x25 Gbps
4x25 Gbps
IntegratedLinear TIA
Segmented MZI Driver
Optical ICElectrical IC
ADCCDR
PAMDecoder
PAMEncoder
Serializer
HOSTSERDES
World’s
First
© 2020 Cisco and/or its affiliates. All rights reserved.25
Cisco Silicon Photonics Optics PDK
Segmented PAM-4 MZI
400um
ER=1
0 d
B
100Gbps PAM-4 Eye
PDK Element In Production
Impact / Benefit
100G Modulator ✔ Golden Eye
Integrated Ge PD ✔
Fiber Coupler ✔ Wide band & low loss
Silicon & Nitride Waveguides
✔ Low loss & multiple layers
VOA ✔
1:2 Switch ✔ Low power & low loss
Fully qualified 300mm Silicon Photonic process in volume production at North American foundry
© 2020 Cisco and/or its affiliates. All rights reserved.26
Optical Coupling
Silicon nitridenanotapers
TM
TE
Edge coupler – “prong coupler”:
• Mode matched to SMF for <1dB coupling loss
• Supports entire CWDM band & beyond
• Qualified & in production on 100G Single l
• Tolerant to dicing – see ref:•R. S. Tummidi and M. Webster, "Multilayer Silicon Nitride-Based
Coupler Integrated into a Silicon Photonics Platform with <1 dB
Coupling Loss to a Standard SMF over O, S, C and L Optical
Bands," 2020 Optical Fiber Communications Conference and
Exhibition (OFC)
Measured Data
© 2020 Cisco and/or its affiliates. All rights reserved.27
Optical Receiver
100G-LR1-20
100GBASE-DR1100GBASE-LR1
100GBASE-FR1
Measured Module Data
S. Rauch, D. Lee, A. Vert, L. Jiang and B. Min, "Reliability Failure
Modes of an Integrated Ge Photodiode for Si Photonics," 2020
Optical Fiber Communications Conference and Exhibition (OFC),
San Diego, CA, USA, 2020, pp. 1-3
Qualified Integrated Detector
© 2020 Cisco and/or its affiliates. All rights reserved.28
QSFP28 100G-DR/FR/LR
28
1 l
53Gbaud PAM4
Transmission Results(Live demo at OFC)
2km 10km
Silicon Photonics IC
Elec IC CeramicLaser
Fiber Array
© 2020 Cisco and/or its affiliates. All rights reserved.29
• Scaling number of lasers
• Scaling number of lanes
Scaling 100G/Lambda →More lanes
TXPMD
MACPCSPMA
MACPCSPMA
RXPMD
ASIC
SerD
es
SerD
es
Digital
SerD
es
DRV
TIA Detector
Laser+
Modulator
4x 100Gbps per l
Fiber Array aligned & attached to Chip on Chip Assembly
© 2020 Cisco and/or its affiliates. All rights reserved.30
Challenges in Scaling w/ Parallel
• Careful design of parallel lanes required to minimize Crosstalk
LinearIntegrated
TIA
ADCCDR
LinearIntegrated
TIA
ADCCDR
Detector
Crosstalk
𝑆𝑁𝑅1 =𝑆
𝑁1𝑆𝑁𝑅2 =𝐺 ∙ 𝑆
𝐺 ∙ 𝑁1 + 𝑁2
++ G
N2 N1
S
0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5
Nec
essa
ry S
NR
imp
rove
men
t [d
B]
SNR Degradation [dB]
Necessary SNR Improvement of one source in case of SNR Degradation through another source
© 2020 Cisco and/or its affiliates. All rights reserved.31
Challenges in scaling w/ Rate
Electrical Channel suffers from Frequency Dependent loss• Using PAM4 a ~17.5dB channel loss is approximated for
112G VSR length
• Using PAM2 the loss would be >30dB
Optical Channel• The fiber insertion loss does not change with data rate
• Components (Driver/Modulator/TIA) introduce bandwidth dependent loss
– Electrical channel within optical packaging also has BW dependent loss
freq
S21
DRV Mod PD TIAdigital… …digital
© 2020 Cisco and/or its affiliates. All rights reserved.32
• Next generation optical components necessary to scale link rates
• Focusing on component BW to increase baud rate is critical
• Careful integration with multi-channel implementations is necessary to maintain link budget
Summary
Performance Photonics Enabling next Generation Interfaces
34
David Lewis, Lumentum
Performance photonics enabling
next-generation interfaces
January 28th, 2021
David Lewis
© 2021 Lumentum Operations LLC 36
Outline
▪ Possible changes in component performance in the next few years:– DML (Directly Modulated Laser) 25 GBd (today) to 50 GBd
– EML (Externally Modulated Laser) 50 GBd (today) to 100 GBd
– Coherent Components 64 – 96 GBd (today) to 128 GBd
▪ These components will enable Ethernet PMDs beyond 400 Gb/s, for example:– 8 x 50 GBd DML for 800 Gb/s interface
– 4 x 100 GBd EML for 800 Gb/s interface
– 8 x 100 GBd EML for 1600 Gb/s interface
– 128 GBd with DP-16QAM for 800ZR interface
© 2021 Lumentum Operations LLC 37
Leading wafer fab infrastructure
North America EMEA APAC
San Jose, California USA• Gallium Arsenide
• Indium Phosphide
• Lithium Niobate
• Planar waveguides
• Flagship products
- Pump lasers
- VCSELs
- ROADMs
- Modulators
Caswell, UK• Indium Phosphide
• Flagship products
- Coherent components
- Tunable lasers
- Photonic integrated circuits
Sagamihara, Japan• Indium Phosphide
• Flagship products
- Data center laser chips DMLs
- PAM4 EMLs
© 2021 Lumentum Operations LLC 38
1971: Room Temperature CW Operation of GaAlAs Lasers
1973: First to realize CW operation of DFB lasers at room temperature
1974: First operation of BH lasers
1975: First operation of DFB lasers under current-injection
1978: First operation of 1.3μm BH lasers
1979; First MP of 1.3μm lasers
1982: MP of 1.3μm BH-lasers for TAT-8 submarine systems
1985: First demonstration of high-speed properties by MQW lasers
1987: Proposal & demonstration of MD-MQW lasers with fr up to 30GHz
1989: Ultrahigh-speed 1.55μm lasers with 17GHz–BW & 16Gb/s modulation
1990: First MP of 2.5Gb/s 1.55µm MQW-DFB lasers
1991: Proposal & demonstration of innovative EML
1992: Record 16λ-WDM 10Gb/s DFB lasers
1994: Record-ultralow threshold current of 1.3μm MQW-DFB lasers with 0.5mA
1995: First demonstration of 40Gb/s EML, proposal and operation of GaInNAs lasers
1996: First MP of innovative 2.5Gb/s 640km EA-DFB lasers
1999: First MP of 10Gb/s-40km EML
2002: First MP of Uncooled 10Gb/s DFB lasers, First MP of 10Gb/s APDs w/GB of 120GHz
2003: First demonstration of uncooled 10Gb/s DFB-LD beyond 115°C2004: First MP of 40Gb/s EML
2007: First demonstration of uncooled 10Gb/s-80km EML up to 85°C2008: First demonstration of 1.3um CWDM 4ch 25Gb/s uncooled EML for 100GbE
2010: First MP of LAN-WDM 25Gb/s EML for 100G CFP-LR4
2011: First demonstration of 25Gb/s LISEL and LIPD
2014 : First demonstration of 1.3μm uncooled 50Gb/s DFB-LD beyond 80°C
2015: First MP of 1.3μm uncooled 25Gb/s DML for 100G-CWDM4
2016: First demonstration of 1.3μm 100G-PAM4 (53Gbaud) EML with 10km transmission
2016: First demonstration of 1.3μm uncooled 25Gb/s DFB-LD beyond 120°C
2018: First uncooled operation of 53-Gbaud PAM4 (106-Gb/s) EML from 25°C to 85 °C2018: First 53-Gbaud PAM4 (106-Gb/s) operation of 1.3-mm DML from 25°C to 80°C
Sagamihara Fab ~50 Years of Innovation in Optical Communications Technology
BH structure origin
DFB technology origin
10G-DML foundation
25G-EML foundation
© 2021 Lumentum Operations LLC 39
Datacom Optics for 800G and Beyond
Lumentum is the market leader for datacom lasers:▪ Scale and performance leader
▪ Pioneer in uncooled, self-hermetic lasers
▪ Continued investment in our laser technology
Lasers for 800G and beyond:▪ 200G EML: Enabling high performance, low power consumption for 2km PAM4 modules
▪ 100G DML: Lower power, lower cost, smaller footprint than EML
▪ 100G VCSEL: Leverage high-volume 3D sensing manufacture foundry for leading performance with
the industry’s best cost structure and massive production capacity
▪ CW lasers for SiPh: Based on Lumentum’s EML chip leadership
▪ CW lasers for CPO: Families of lasers covering 20mW to over 400mW
DMLVCSEL EML CW laser
DML Status
© 2021 Lumentum Operations LLC 41
DML Leadership
Design features:
▪ InGaAlAs MQW active layer for reliable high temperature operation
▪ Ridge waveguide structure for manufacturability with high yield
▪ Corrugation Pitch Modulated (CPM) grating and shorter cavity for higher bandwidth - 25G x 4λ
▪ Sophisticated cavity design for wide temperature operation, high reliability
▪ Self-hermetic chip for GR-468 damp heat environments
▪ Grating-pitch designs for CWDM(HL13BF) and LAN-WDM(HL13BE)
▪ PAM4 DMLs
HL13BF DFB Laser
p-Electrode
n-
Electrode
Ridge-Waveguide
n-Electrode
TLD=50ºC, If=60mA
© 2021 Lumentum Operations LLC 42
26 Gbaud
DML vs EA-DFB PAM4 Eye Comparison
53 Gbaud
▪ Choice defines power consumption, dispersion tolerance, and cost
▪ Lumentum EMLs have historically provided the most cost and power efficient solution for leading edge interface rates
▪ Lumentum DMLs provide power and cost reduction path as ecosystem matures at a given rate
© 2021 Lumentum Operations LLC 43
50 GBd (100 Gb/s PAM4) DML Progress
Presented by N. Sasada et al., ECOC2018 Th3F.3 (2018)
EML Status
© 2021 Lumentum Operations LLC 45
EML Leadership
30 years of EML technology expertise dating back to Hitachi▪ Leadership position in high-speed EMLs since 2.5G and 10G era ▪ World’s first uncooled EML, now GR-468 self-hermetic devices in production
Best support for the 200G/400G PAM4 module applications▪ 1.3um LAN-WDM/CWDM wavelengths, cooled LAN-WDM and uncooled 28GBaud/53Gbaud PAM4▪ Bare chip for cost effective design and COC for quicker evaluation and production usage
Design features▪ Butt-joint structure for designing LD and EA independently▪ Buried Heterostructure with semi-insulating InP layer for high speed ▪ High temperature operation for low power consumption TEC, or “coolerless”▪ Optimized p-i-n structure in EA modulator for higher extinction ratio
n-InP sub.EA
modulator
DFB laser
WG
p-electrode
SI-InP Layer
n-electrode
MQW layer
DFB laserEA modulator
© 2021 Lumentum Operations LLC 46
HL13B6: 53Gbaud uncooled EML performance snapshot
▪ Basic design is based on HL13B5 with high reliability and high productivity. – Achieved high BW of 42GHz and high Po at 85C compared to HL13B5
– MQW structure is optimized to achieve low TDECQ over the temperature range
HL13B6-b HL13B5
20C 70C 85C 85C
with
equalizer
(5tap)
Vpp (V) 1.2 1.2 1.2 1.2
Vmid (V) -2.5 -1.3 -1.1 -0.4
OuterER
(dB)3.5 4.6 6.2 10.4
PoAve.
(dBm)10.5 7.1 4.2 2.0
TDECQ
(dB)2.2 2.2 2.6 unmeasurable
IF=100mA
-15
-12
-9
-6
-3
0
3
0 10 20 30 40 50
S21
(dB)
Frequency (GHz)
Fig. S21 @85C
:HL13B5
:HL13B6
42 GHz @ 85C
© 2021 Lumentum Operations LLC 47
100 GBd EML Progress
Ref: K. Adachi et al, Mo.2.B.6, ECOC2020
InP PIC and Coherent Components
© 2021 Lumentum Operations LLC 49
InP Optical Devices Technology for Moving Networks Forward Faster
NLL-Tunable
ILMZ
MZ-SOA
Receiver PIC
NLL-Tunable
ILMZ
MZ-SOA
Receiver PIC
Intra DC
5G-
Wireless
ILMZ
DML
EML
PD/APD
High-Power DFB
DML
EML
PD/APD
High-Power DFB
© 2021 Lumentum Operations LLC 50
Coherent InP PIC Trends
TL
MZ
Rx
100kHz LW, 17dBm fibre, 100ch
▪ Reduced power dissipation
▪ High temperature operation
▪ 50kHz LW, 18dBm fibre
▪ 120ch operation
64-96 Gbd dual-IQ fold-MZ with SOA
(40GHz BW)
64Gbd dual-IQ Rx with VOA
(45GHz BW)
▪ 128Gbd (70GHz BW)
▪ Reduced power dissipation
▪ High temperature operation / Uncooled
▪ Reduced Vpi / Driverless
▪ 128Gbd
▪ Smaller chips for lower cost
Today Next
ILMZ 25Gb/s NRZ-ILMZ
Hermetic die
▪ 100Gbd PAM4 (60GHz BW)
▪ 1300nm PICs as capable as 1500nm
▪ Smaller chips for lower cost
© 2021 Lumentum Operations LLC 51
Coherent InP OSA Trends
HB-CDM
TROSA
ROSA
64Gbd, 96Gbd
(OIF IA class 40 & 60)
▪ 128Gbd and beyond (OIF IA class 80)‒ Very high-speed RF transition & interconnections
▪ Increased thermal dissipation
43Gbd, 64Gbaud
32Gbd
▪ 64Gbaud
▪ 96Gbd and beyond‒ Very high-speed RF transition & interconnections
▪ Reduced power dissipation
▪ 128Gbd and beyond‒ Very high-speed RF transition & interconnections
▪ Improved responsivity
Today Next
© 2021 Lumentum Operations LLC 52
Summary
▪ The components for the next speed are coming:– DML chips for the module industry
– EML chips for the module industry
– InP PICs for coherent transmission
– InP packaged components for coherent transmission
▪ Timeframe is aligned with other components:– Switch / SerDes / DSP silicon
Thank you
Observations on the Rate of Beyond 400GbE,
800GbE an/or 1.6TbE
54
Xinyuan Wang, Huawei
Observations on the Rate of Beyond 400GbE,
800GbE and/or 1.6TbE
Xinyuan Wang, Huawei Technologies
Advanced Technology Drives 800GbE and 1.6TbE Standard
Eth
ern
et
Sta
nd
ard
1995
100M
1000M
1998
10GbE
25GbE
400
/200GbE
100
/40GbE
20102002
X4 Lan
Lower Co
es/FEC
mplexity
PAM4/Co
High Order
herent
Modulation
~2.5-3X $/bit
~2X $/bit
~1X $/bit
<1X $/bit
Forecasting IEEE 802.3 Beyond 400GbE
➢ Optical/Electrical Innovation to increase per
lane rate with low cost and power
➢ Advanced technology forhigher reliability and
density of optical module
Estimated cost data based on 1M ports shipped
2017 2025?
SiP/CPO/
Higher RateFEC
Modulation
1.6TbE
/800GbE200GbE
Ratified Year
50
/100GbE
5
6
DCN Application: IM-DD for 2km, 500m, 100/50m?
100Gb/s PAM4• More comprehensive discussion at 802.3cu
• Ratified in Jan 2021
• Modulation, Channel/Link Parameter
• SNR, BER, Low Latency FEC
50Gb/s PAM4
• First adopted at 802.3bs with ratified in late 2017
200Gb/s PAM#
200Gb/s per optical lane is expected with following challenge.
➢ X4 for 800GbE, 2X rate from 400GbE!
➢ X8 for 1.6TbE is a sweet point, maximum reach?
➢ As AI, HPC, URLLC, AR/VR applications emerge, lower latency is desired
◆ ~100ns latency for 2-Way KP4 FEC of 400GbE
◆ Similar FEC latency to enable 2X/4X
throughput for Beyond 400GbE
5
7
Carrier Network Application: PnP Coherent for 10km+ Fatter Pipe will drive lower Capital expenditures (CAPEX) and Operating expenses (OPEX)
comparing to L2 Link Aggregation without Hashing efficiency issue.
➢ 400GbE→1.6TbE with 4X can benefit more than 800GbE.
Supporting compatible upgrade to currently deployed network to Beyond 400GbE with
following considerations:➢ Operate over Duplex SMF with 10 and 40km reach, compatible with current fiber link infrastructure
➢ 80km is also suggested, possible coverage down to 40km reach with cost and power advantage
➢ Plug and Play optical module, avoiding field engineering, configuration, debug and test
➢ For 800Gbps solution, suggest to investigate 16QAM feasibility
Access/
Aggregate
Core/
M etro
DC10km 40km 80km
Fixed IP Network 60% 30% 10%
Mobile Backhaul Network 45% 45% 10%
5
8
To Achieve Flexible Solution of Beyond 400GbE
5
9
Based on 200Gb/s per lane, 800GbE and 1.6TbE should be a family standard.
1TbE with X5/X10 lanes will impact off-the-shelf module form factor and ASIC
architecture, not recommended.
At lease to support 100Gb/s SerDes based AUI interface.
Logic layer, PCS/FEC/PMA architecture, should support future optical
evolution.
Balance on cost, power and performance
Friendly support Breakout
Copyright©2018 Huawei Technologies Co., Ltd.
All Rights Reserved.
The information in this document may contain predictive
statements including, without limitation, statements regarding
the future financial and operating results, future product
portfolio, new technology, etc. There are a number of factors that
could cause actual results and developments to differ materially
from those expressed or implied in the predictive statements.
Therefore, such information is provided for reference purpose
only and constitutes neither an offer nor an acceptance. Huawei
may change the information at any time without notice.
把数字世界带入每个人、每个家庭、
每个组织,构建万物互联的智能世界。
Bring digital to every person, home, and organization for a fully connected, intelligent world.
Thank you.
Q & A
61
Scott SchubeIntel
Matt TraversoCisco Systems
David LewisLumentum
Xinyuan WangHuawei
Mark NowellCisco Systems
62
Join Us Tomorrow
Friday, January 29th
Panel discussion – Test & Measurement: Planning for PerformanceModerator – David J. Rodgers, Ethernet Alliance Events Chair and Teledyne LeCroy
Panelists –
John Calvin, Keysight Technologies – “Validation Methods for Emerging 106Gbps Electrical and Optical Specifications relating to IEEE P802.3cu/P802.3ck”
Steve Rumsby, Spirent – “Recommended Design Practices for the Next Generation Ethernet Rate”
Francois Robitaille, EXFO – ” Full Compliance Validation of Next-Gen Transceivers”
If you have any questions or comments, please email [email protected]
For our TEF 2021 on-demand content go to
www.ethernetalliance.org
www.ethernetalliance.org 63
Reference Only
64
© 2020 Cisco and/or its affiliates. All rights reserved.65
100G-xR (single l 100G) – Block Diagram
QSFP 100G• Same hardware/software for 500m (DR), 2km (FR),
10km (LR), 20km (LR1-20)
Elec IC• Custom 16nm CMOS IC
• Flip-Chip’d onto SiPho
• Driver & TIA integrated
Silicon Photonics• Integrated Modulator & Photodiode
• Integrated optical coupling features
• Acts as substrate for Elec IC
Laser• Single Laser – DC power supply
MicroController
QSFP28 module
MUX, TX DSP
& Encoding
SerDes4x 25Gbps
optical TX path
DeMUX, Equalization,
RX DSP,Decoding
SerDes4x 25Gbps
optical RX path
1x100GADC
SPI interface
1x 100GTIA
1x 100GDriver
Elec IC
Silicon Photonics
100GModulator
100GPhotodiode
Laser
Power Supplies RefClk
ASI
C /
Ho
st
4x 25GbpsNRZ
65