Beruflich Dokumente
Kultur Dokumente
Image
Agenda
Introductions
Market Overview
Design Engineering
Test & Measurement
Panel Discussion
Market Adoption
400G 100G
The top-to-bottom 400G standards
Distance Standard Modulation/signaling e.g.
X,000 km OIF, OTN, ITU Complex optical DP-QPSK
100M (MMF) Ethernet PAM2 at 25 GBd 400GBASE-SR16
10 km Ethernet PAM4 at 25 GBd 400GBASE-LR8
2 km Ethernet PAM4 at 25 GBd 400GBASE-FR8
500 m Ethernet PAM4 at 56 GBd 400GBASE-DR4
Backplane < 1m OIF CEI PAM4 at 25 GBd CEI LR
Interconnect Ethernet NRZ CDAUI-16,
module to chip, OIF CEI PAM4 CDAUI-8
chip to chip CAUI-4
CEI VSR
Ransom Stephens
How we got here
Photonics ? 100G: 425 Gb/s
1000
PAM4
Serdes
Serdes
FEC
Data rate (Gb/s)
100
Serdes
Serdes
1 Differential signaling + clock recovery
2000 2005 2010 2015 2020
PAM4
4 level Pulse Amplitude Modulation
T
T
12
RX: 56Gb/s, 25dB Insertion Loss, PRBS31, Pre-FEC BER = 7E-12
eBert
Inner Eye
RX: 51.56Gb/s, 32dB Insertion Loss, PRBS31, Pre-FEC BER = 1E-6
eBert
Inner Eye
Mark Marlett
Near End Eyes: short channel
Near End Eyes: Lots of Pre and Post Cursor EQ needed for long channel
Channel response: ~-33dB at Nyquist
Left: Scope Near and Far end channel
Right: Chip Eye Contour (raw BER ~1e-7)
Note: Channel running error free
Mike Peng Li
FEC Kills DJ and RJ ?
False !!
FEC is NOT everywhere
FEC latency prohibitive for high performance computing (HPC) and high performance memory
(HPM)
For those links, BER of 1e-15 is required, so does DJ and RJ
Even with FEC
Pre-FEC BER needs to be at 1e-6, and DJ and RJ are needed to provide statistical RSS link jitter
margin maximizing, even though TJ could be measured directly with an sampling scope
H FER L FER
FEC
1e-6 <1e-15
Ref: http://www.ieee802.org/3/bs/public/15_01/nicholl_3bs_01b_0115.pdf
FEC is Deterministic and Equal?
False !!
Smiths FEC model (A) differs from Johnsons FEC model (B)
FEC model assumptions are deterministic, but bit error is statistical
Who is right ?
Needs FEC measurement data
(B)
(A)
=
First FEC model vs measurement comparison by Altera/Intel
Marty Miller
T&M challenges
How to know if your Tx is doing what you intend?
In this case 1-pre, 1-post emphasis closes eye (left)
2
So it is painful to move to PAM4.
We can measure instead of extrapolate the bathtub
Actually the bad PAM4 eye is better than it looks:
most designs add FEC (Forward Error
Correction) code, and the link works even if the
raw BER (before FEC) is BER= 10-6 (instead of
the BER= 10-12 or BER= 10-15)
And the oscilloscope can now measure the
bathtub @BER, not just extrapolate it.
So measuring the bathtub is an option instead
of mask. Or some form of bathtub, like TDEC .
But its still harder.
The eye closes fast!
The vertical eye closure
vs the T&M BW:
Paper Wed. at 8:30 am
2
Problem of what to measure
We measure so many things (Rise-times fall-times measurement)
We emulate the channel
We de-embed the fixture
We equalize with FFE/DFE/CTLE
Is that the forest, or the trees?
BUT
What is the best measure?
Is it still the eye diagram?
If yes, with a mask?
Or is it TDEC?
Or is it COM?
2
Greg LeCheminant
We still need to test the PAM4 receiver
The basic premise: Will the receiver achieve
the desired BER when paired with the worst
case allowed TX paired with the worst case
allowed channel
What did we learn from NRZ?
Stressed Receiver Sensitivity
Creating the precision degraded eye is
1) expensive to build
2) difficult to verify
3) difficult to maintain over time (especially for
optical systems)