Cmos, Analog, Design-41

ISSCC 2007 / SESSION 24 / MULTI-GB/s TRANSCEIVERS / 24.
1
24.1
A 12.5Gb/s SerDes in 65nm CMOS Using a BaudRate ADC with Digital Receiver Equalization and
Clock Recovery.
Mike Harwood1, Nirmal Warke2, Richard Simpson1, Tom Leslie1, Ajith

Amerasekera2, Sean Batty1, Derek Colman1, Eugenia Carr1, Venu Gopinathan3,
Steve Hubbins2, Peter Hunt1, Andy Joy1, Pulkit Khandelwal1, Bob Killips1,
Thomas Krause2, Shaun Lytollis1, Andy Pickering1, Mark Saxton2, David
Sebastio1, Graeme Swanson1, Andre Szczepanek1, Terry Ward1, Jeff Williams1,
Richard Williams1, Tom Willwerth2
Texas Instruments, Northampton, United Kingdom
Texas Instruments, Dallas, TX
3
Texas Instruments, Bangalore, India
A key challenge in designing high-bandwidth systems such as datarouters and super-computers is to transferring large amounts of
data between ICs either on the same circuit board or between
boards. Analysis of typical backplane channel attenuation (-24dB)
and package losses (-1 to -2dB) in the presence of crosstalk predict
that an un-equalized transceiver provides inadequate performance
and that DFE is needed to achieve error rates of <10-17.
Traditional DFE methods for SerDes receivers rely on either modifying the input signal based on the data history [1-3] or on having
an adaptive analog slicing level [4]. This paper describes an alternative approach. The input data is sampled at the baud-rate, digitized, and the equalization and CDR are performed using numerical digital processing techniques. This approach enables
power/area scaling with process, simplifies production testing,
allows integration of a FFE, and provides a flexible design with a
configurable number of filter taps. Implemented in 65nm CMOS,
the 12.5Gb/s transceiver uses the baud-rate sampling ADC, a RX
with digital 2-tap FFE and digital 5-tap DFE to correct channel
impairments, and a TX with a 4-tap FIR filter to pre-compensate
for channel impairments.
The RX is shown in Fig. 24.1.1. The received data is digitized at the
baud-rate, 1.0 to 12.5Gb/s, using a pair of interleaved T/H stages
and 23-level (4.5b) full-flash ADCs. The 2 T/H circuits enable interleaving of the half-rate ADCs and reduce signal-related aperture
timing errors. Two ADCs, each running at 6.25Gb/s, provide baudrate quantization of the received data. The dynamic range of the
ADC is normalized to the full input amplitude using a 7b AGC circuit. A loss-of-signal indication is obtained by detecting an out-ofrange AGC. An optional attenuator is included in the termination
block to enable reception of large signals whilst minimizing signal
overload.
The detailed full-flash ADC is shown in Fig. 24.1.2(A). The first linear transconductor stage (24.1.2(B)) is optimized for handling large
input signals combined with low supply voltages. Resistive interpolation is used to reduce the input DNL. The second CML latchstage (C) is conventional, but optimized for speed and metastability performance. The final sense-amp latch (D) uses a cross-coupled
NMOS pair with PMOS loads. Cross-coupled inverters further
enhance the gain and reduce the meta-stability occurrence probability by adding hysteresis.
The output from the ADCs is fed into a custom DSP data-path that
performs the numerical FFE and DFE, as shown in Fig. 24.1.3(A).
The digital FFE/DFE is implemented using standard library gates.
An advantage of digital equalization is that it is straightforward to
include FFE as a delay-and-add function without any noise-sensitive analog delay elements. The FFE tap weight is selected to compensate for pre-cursor ISI and can be bypassed to reduce latency.
While many standards require pre-cursor de-emphasis at the TX,
inclusion at the RX allows improved BER performance with existing legacy transmitters.
The DFE uses an unrolled non-linear cancellation method [4]. The
FFE output is compared with a stored slicer-level to generate the
data output. The slicer-level is selected from one of 2n possible
options depending on the previous n bits of data. Unrolled tap
adaption is performed using an LMS method where the optimum
slicing position is defined to be the average of the 2 possible symbol amplitudes (1) when preceded by identical history bits.
Although 5-taps of DFE are chosen for this implementation, this
436
parameter is scaleable and performance can be traded-off against

power consumption and die area. In addition, the digital equalizer
can be tested using standard ATPG and circular BIST approaches.
A Mueller-Muller approach [5] is used for clock recovery where the
timing function adapts the T/H sample position to the point where
the calculated pre-cursor ISI or h(-1) is zero, an example is shown
in Fig. 24.1.3(B). The 2 curves show the post-equalized response for
010 and 011 data sequences, respectively. The intersection at
3440ps occurs when the sample of the second bit is independent of
the third bit that is, h(-1) = 0. This position can be detected by
comparing the post-equalized symbol amplitude with the theoretical amplitude h(0) and using the difference to update the phaseinterpolator of the CDR.
An outline diagram of the TX is shown in Fig. 24.1.4. The data to
be transmitted is sequentially delayed to produce a nibble-wide
word containing the pre-cursor, cursor, and 2 post-cursor components. These allow a 4-tap FIR output waveform to be obtained
from simple current summing of the time-delayed contributions.
The relative amplitude of each contribution is weighted to allow the
FIR coefficients to be optimized for a given backplane and minimize
the overall residual ISI. The line-driver is implemented using CML
techniques and operates using 4 interleaved nibbles at 3.125Gb/s to
ease timing closure. The 4 line drivers are implemented as a single
structure minimizing return-loss.
A PLL is used to generate low-jitter reference clocks for the TX and
RX to meet standards [6, 7]. The PLL uses a ring oscillator to produce 4 clock-phases at a quarter of the line data-rate. The lowerspeed clocks allow power-efficient clock distribution using CMOS
logic levels, but need duty-cycle and quadrature correction at the
point of use. The 3.125GHz clocks are frequency doubled (XOR
function) to provide the 6.25GHz clock for the T/H and the ADC.
The TX uses the 4 separate 3.125GHz phases, but they require
accurate alignment to meet jitter specifications of 0.15UIpp RJ and
0.15UIpp DJ.
The fabricated chip supports error-free operation at 12.5Gb/s over
short channels (two 11mm package traces, 30cm low-loss PCB, and
2 connectors). A legacy channel with -24dB of attenuation at
3.75GHz supports error-free operation at 7.5Gb/s. Figure 24.1.5
shows a 12.5Gb/s 27-1 PRBS transmitted eye-pattern with 20% deemphasis on the first post-cursor. The differential amplitude is
700mVpp (200mV/div). Figure 24.1.5 also shows the ADC output
when a 6.25GHz sine-wave is sampled and the phase between the
sine-wave and RX is incremented using a programmable delayline. The measured codes are within 1 LSB of the expected values.
This level of performance ensures robust operation over a wide
range of cables, green-field, and legacy channels. The worst-case
power of a single TX/RX pair is 330mW/lane and the total macro
area is 0.45mm2 per lane. A chip-plot of a quad (4-lane) SerDes
macro with the key circuit elements identified is shown in Fig.
24.1.6.
Acknowledgements:
This work is the combined effort of TIs UK SerDes Development group, especially the Design, Test-Chip, Physical-Layout, DFT, Verification, Packaging
and Signal Integrity teams. The authors would like to thank Sun
Microsystems for their collaboration with this development. The authors also
thank the TI Dallas characterization & compliance lab.
References:
[1] R. Payne, B. Bhakta, S. Ramaswamy, A 6.25Gb/s Binary Adaptive DFE
with First Post-Cursor tap Cancellation for Serial backplane
Communications, ISSCC Dig. Tech. Papers, pp. 68-69, Feb., 2005.
[2] M. Sorna, T. Beukerna, K. Selander, A 6.4Gb/s CMOS SerDes Core with
Feed-Forward and Decision Feedback Equalization, ISSCC Dig. Tech.
Papers, pp. 2-63, Feb., 2005.
[3] V. Balan, J. Caroselli, J.-G. Chern, et al., A 4.8-6.4Gb/s Serial Link for
Backplane Applications Using Decision Feedback Equalization, IEEE J.
Solid-State Circuits, pp. 1957-1967, Nov., 2005.
[4] S. Kasturia, H. J. Winters, Techniques for High-Speed Implementation of
Non-Linear Cancellation, IEEE JSAC, pp. 711-717, Jun., 1991.
[5] K. Mueller, M. Muller Timing Recovery in Digital Synchronous Data
Receivers IEEE Trans. on Communications, pp. 516-531, May, 1976.
[6] OIF-CEI-02.0 - Common Electrical I/O (CEI) - Electrical and Jitter
Interoperability Agreements for 6G+ bps and 11G+ bps I/O, Optical
Internetworking Forum, Feb., 2005.
[7] IEEE Draft 802.3ap/Draft 3.0 Amendment: Electrical Ethernet
Operation over Electrical Backplanes, IEEE , Jul., 2006.
2007 IEEE International Solid-State Circuits Conference
1-4244-0852-0/07/$25.00 2007 IEEE.
ISSCC 2007 / February 14, 2007 / 8:30 AM
(A) ADC core showing 3 comparators
Figure 24.1.1: A 12.5Gb/s receiver with 2 ADCs, a 2-tap FFE and a 5-tap DFE.
Figure 24.1.2: A) Sub-sections of the ADC including 3 comparator sections, and B) the
linear transconductor.
(A) Equalization implementation
(C) Schematic of CML latch
(B) Schematiic of linear transconductor
(B) Post-equalized signal amplitude
(D) Schematic of sense-amp latch
Figure 24.1.2: C) The CML latch and D) the sense-amp latch.
Figure 24.1.3: A) Digital receiver equalizer implementation, B) Post-equalized signal

amplitude against time in psindicates ideal eye sampling point at 3440ps where
h(-1)=0.
24
Figure 24.1.4: Transmitter with weighted 4-tap FIR filter and current-summing CML
output stage.
Figure 24.1.5: Lab measurements of the 12.5Gb/s transmitter (vertical scale

200mV/div) left, and Interleaved ADCs -right.
Continued on Page 613
DIGEST OF TECHNICAL PAPERS
437
ISSCC 2007 PAPER CONTINUATIONS
Figure 24.1.6: Quad SerDes macro with the key components of a single lane and the
central PLL indicated.
613
2007 IEEE International Solid-State Circuits Conference
1-4244-0852-0/07/$25.00 2007 IEEE.

Cmos, Analog, Design-41

Uploaded by

Copyright:

Available Formats

Cmos, Analog, Design-41

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cmos, Analog, Design-41

Uploaded by

Copyright:

Available Formats

ISSCC 2007 / SESSION 24 / MULTI-GB/s TRANSCEIVERS / 24.

Mike Harwood1, Nirmal Warke2, Richard Simpson1, Tom Leslie1, Ajith

parameter is scaleable and performance can be traded-off against

2007 IEEE International Solid-State Circuits Conference

1-4244-0852-0/07/$25.00 2007 IEEE.

ISSCC 2007 / February 14, 2007 / 8:30 AM

(A) ADC core showing 3 comparators

(A) Equalization implementation

(C) Schematic of CML latch

(B) Schematiic of linear transconductor

(B) Post-equalized signal amplitude

(D) Schematic of sense-amp latch

Figure 24.1.2: C) The CML latch and D) the sense-amp latch.

Figure 24.1.3: A) Digital receiver equalizer implementation, B) Post-equalized signal

Figure 24.1.5: Lab measurements of the 12.5Gb/s transmitter (vertical scale

Continued on Page 613

DIGEST OF TECHNICAL PAPERS

ISSCC 2007 PAPER CONTINUATIONS

2007 IEEE International Solid-State Circuits Conference

1-4244-0852-0/07/$25.00 2007 IEEE.

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.