Power Optimized Transmitter for Future Switched Network

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 09 | Sep -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 569
Power Optimized Transmitter for Future Switched Network
Omkar C. Mane1, Prof. Usha Jadhav2
1Dept. of Electronics and Telecommunication Engg. D. Y. Patil College of Engineering, Akurdi Pune, India
2Faculty of Dept. of Electronics and Telecommunication Engg. D. Y. Patil College of Engineering, Akurdi
Pune, India
---------------------------------------------------------------------***-------------------------------------------------------------------
Abstract - Network equipment power consumption is under
increased scrutiny. To understand transmitter power
consumption, Combination of CMOS and MOS current mode
logic (MCML) is used and characterize power consumption
using Tanner EDA Tool 13.0. For optical transmitters, weshow
that photonic components and front end drivers only consume
a small fraction (<22%) of total serial transmitter power. This
implies that the power of optical transmitter is reduced can
only be obtained by paying attention to the physical layer. We
propose a physical layer protocol suitable for optically
switched links that retains the beneficial transmission
characteristics of 8b/10b, but, even without power gating and
voltage controlled oscillator power optimization, reduces the
power consumption during idleperiods by29%compared with
a conventional 8b/10b transmitter. We have made the toolkit
available to the community at large in the hope of stimulating
work in this field.
1. INTRODUCTION
The persistent growth in network traffic advanced by recent
developments, such as video sharing, IPTV, and cloud-based
storage, is causing increased demands on the network
switching capacity and energy consumption at the Internet
core and within data centers. Increasing the capacity of
current high-bandwidth electronic switches is not only
technically demanding it also leads to higher thermal
dissipation. This leads not only to interconnect technologies
with high connectivity and capacity, but also lower latency,
power consumption, and cost. Among these, the energy
performance of networked systems has become a first class
property of interest to industry and researchers. It has been
shown that to make large energy savings through energy-
proportionality current computer systems must be made to
do nothing well: minimizing consumption when not in use.
Optical networks continue to deliver on the promise of
bandwidth, latency, and low power utilization but if optical
switch fabrics are to continue meeting their promise asa key
component in future energy-proportional systems.
Then, we need a generation of high-speed transmitter
designed with energy proportionality as the first class
property.
Transceiver designhasbeen focuseduponproviding
high reliability with ever-higher levels of link capacity
(bandwidth) to meet ever-growing needs to interconnect
computer devices. This has led to optical transceivers that
are always on, exchanging information to remain
synchronized even when carrying no data. Such designs suit
point-to-point link communications, providing implicit
information about the point-to-point link status, even when
no data is carried. There is a wide range of transceivers, with
electronics to drive twisted paircopper,multichannel coaxial
copper, and a range of optical systems. Current commercial
optical 10 Gb/s transceivers have lower power consumption
than twisted-pair serial transceivers due to a lower
complexity physical (PHY) layer,1. Yet,inweshowedthat the
popular 8b/10b coding scheme can consume more power
when transmitting idle frames than when transmitting data.
Finally, a further power consumption incentive comes from
the increasing move of communication endpoints to on-chip
in silicon-on-chip (SoC) processors. With predictions that a
growing proportion of the chipwill needto bepowergatedat
any one time, the so called dark silicon effect. The serial
electronic transceivers, which provide several Tb/s of off-
chip bandwidth required in high-performance SoC
processors, are already consuming >20% of the total power.
Silicon photonics has been widely proposed as one of the
solutions to the processor communications bottleneck and
energy issues. However, we show that optical transceiver
power is dominated by other physical layer (PHY) functions
such as serialization/deserialization (SERDES), clock
recovery, and line coding. Hence, a simplistic change from
electronic to optical transmitter will not reduce power
consumption without an accompanying change to the PHY
layer. Furthermore, at the packet timescale, an optical
switched system sets a new optical pathway to each
destination. Thus, the physical layer need not remain
operating when idle and, without system wide time
synchronization, an optical packet-switch proposal uses
burst-mode receivers capable of fast locking to incoming
packets each having different frequency, phase, and
amplitude. This requires per packetclock recoverydesignsin
a new PHY implementation.
Synchronization, an optical packet-switch proposal
uses burst-mode receiverscapableoffastlockingtoincoming
packets each having different frequency, phase, and
amplitude. This requires per packetclock recoverydesignsin
a new PHY implementation.

2. Transmitter design
This section describes the main functional circuit blocks of
the transceiver. Top-level representations of the transmitter
and receiver are shown in Figs. 1 . Initially, the transmitter
design was aimed at a payload bit rate (before adding coding
overhead) of 10 Gb/s. In a later work, we characterized the
circuits at different bit rates to investigate optimum bit rate
versus power operating points.
Fig 1. Proposed block diagram of optical transmitter
I. Line Coding
The functions of the coding block include dc balance, byte
alignment within the serial stream, and error detection. We
consider two popular encoding schemes: 8b/10b block code
and the scrambler-based64B/66B.8b/10b representsa class
of parity–disparity dc-balanced codes that map arriving 8-b
symbols into a 10-b code words using predefined code
groups at run time on the word-by-word basis. The code
limits the run length of identical symbols to remove baseline
wander in AC-coupled receivers and guarantee the required
transition density for clock synchronization. 8b/10b coding
has excellent transmission properties but has a bit rate
overhead of 25%. For this reason, the hybrid-scrambled
64B/66B encoding scheme was selected for 10 Gb/s
Ethernet, which reduces the overhead to 3%. The encoding
module performs a framing function by transformingthe 64-
b data and 8-b control inputs into a 66-b block. Each 64-b
data word is scrambled with a 58th degree polynomial to
ensure statistical dc-balance and transition densityanda 2-b
synchronization headerisappendedtoallowframedetection
and alignment to be performed. Fig. 3 shows block diagrams
of the two alternative coding schemes in the transmitter. In
the 64B/66B case, the transmitter accepts 64-b data at
156.25 MHz and carries out encoding and scrambling. The
resulting 66-b are converted by the gearboxto64-binterface
at 161.13 MHz for more efficient serialization. In the 8b/10b
case, we implemented versions with both 8-b wide client-
side data interface running at 1.25 GHz and a dual encoder
with a 16-b interface operating at 625 MHz. Figs.3 show the
16-b interface. In all cases, phase differences between the
coding block and client-side interface are compensated by a
first-in first-out (FIFO) buffer.
2. Scrambler
Fig2. Scrambler operation
The first bit in sequence s1is summed modulo-2 with the
modulo-2 sum of location 2 and 5 in the shift register. This
sum becomes the first bit in bit sequence s2. As this bit is
presented to the channel, the contents of the shift register
are shifted up one stage as follow: 5 out, 4 goes to 5, 5goes
to 3, 3 goes to 4, 2 goes to 3, 1 goes to 2. The first bit in s2 is
also placed in shift register stage 1.The next bit of sequence
s1 arrives, and the procedure is repeated.
II. Serialization and Deserialization
The SERDES circuits convert between the low-speedparallel
data and a high-speed serial bit stream. The multiplexing
ratios depend on the coding scheme used. In the case of
64B/66B, 64-b sequences at 161.13 MHz are converted to
10.3125 Gb/s using a 64:1 ratio. In contrast, a transceiver
with 8b/10b coding performs either 10:1 or 20:1
multiplexing producing a line rate of 12.5 Gb/s. As shown in
Figs 3 the SERDES circuits are implementedina combination
of static CMOS and MCML. To find a power-efficient SERDES
design, we engineered a variety of configurations. For
example, for 64B/66B, we investigated 64:1 SERDES based
on 64:N CMOS and N:1 MCML circuits where N = 2, 4, or 8
(referred to throughout the rest of this paper as 64:N:1). In a
similar way, we investigated 8b/10b SERDES using 20:N:1
(dual encoder) and 10:N:1 (single encoder) cases, with N = 2
or 4.

Fig 3. CMOS and MCML combination for serialization and
deserialization.
The CMOS SERDES circuits are implemented as shift
registers. The MCML circuits were implemented as binary
tree multiplexers constructed by cascading 2:1 multiplexer
cells, frequency dividers, and delay lines, which were
manually optimized for the required bandwidth and timing
operation. Fig. 5 shows an example of a 64:8:1 SERDES.
III. Transmitter Phase-Locked Loop
Commonly used phase-locked loop (PLL) and CDR circuits
often use multiple stages to facilitate a stable and consistent
operation. This redundancy usually delivers high
performance but the synchronization process takes a
relatively long time to achieve a stable lock. The simplicity of
our CDR design guarantees a fast locking time (≤10 clock
cycles) and maximum power and area efficiency. Although
realistic CDR implementation may require some
modifications to the design to account for factors such as
minor impedance mismatch, capacitive and inductive
resistance variations, and so on, we believe that the power
figures will be representative of real circuits.
IV. Channel Bonding
To find the power consequences of using multiple lower bit
rate serial streams rather than a single serial channel, we
designed a channel bonding circuit in Verilog, which
eliminates skew between multiple channels using a separate
FIFO. In an optical link, these channels could be either space
or wavelength division multiplexed. We tested the circuit
operating on the output of two 8b/10b client-side streams,
but the Verilog model isparameterizedforhighernumbersof
channels. The circuit is designed for burst-mode operation.
3. Circuit design
To find the power consequences of using multiple lower bit
rate serial streams rather than a single serial channel, we
designed a channel bonding circuit in Verilog, which
eliminates skew between multiple channels using a separate
FIFO. In an optical link, these channels could be either space
or wavelength division multiplexed. We tested the circuit
operating on the output of two 8b/10b client-side streams,
but the Verilog model isparameterizedforhighernumbersof
channels. The circuit is designed for burst-mode operation.
A. Design of CMOS Circuits
Design of the static CMOS circuits started with Register
Transfer Level (RTL) Verilog hardware descriptionlanguage
descriptions and synthesized using Synopsys Design
Compiler with a commercially available 45-nm standard cell
library. Constraints wereset tominimize powerconsumption
at the required operating frequency. The typical clock
frequency margin used for synthesis is considered to be at
least 15% faster than the nominal frequency value. The
synthesized Verilog netlist was simulated using Mentor
Graphics ModelSim to verify correct operation and store
activity data for dynamic power analysis. The input stimulus
for the simulations was extracted from realistic 10 Gb/s
Ethernet trace files and analyzed under: 1) continuous data
transmission and 2) continuous idle transmission input
setups. Synopsys Prime Time was used to generate power
consumption data for each circuit block.
B. Design of MCML Circuits
Although new generations of CMOS technologies
continuously improve their performance and power
characteristics due to scaling, CMOS circuits are prone to
generate a highlevel supply noise while operating at high
speeds. The noise factor limits the on-chip integration of
digital blocks with their analog counterparts. Logic families
with differential signaling, such as MCML are characterized
by an improved noise immunity and high-speed operation.
The speed advantage is achieved by the fact that the current,
generated by a constant current source, is steered betweena
pair of fully differential transistors and produces a reduced
swing voltage drop at outputs (in combination with specific
voltage gains), reducing the generation of logic level
switching noise. It must be noted though that the presenceof
the current sink implies a constant power dissipation
irrespective to the operating frequency or input sequence
applied. Power dissipation in MCML circuitisdominated bya
static power (P = Vdd× Iss) and is independent of the
operating frequency. In this paper, an MCML cell library was
developed. The design process used the transistor models
supplied with the 45-nm CMOS standard cell library and a

semianalytical methodology developed in HSPICE
environment for cells opti-mization. To satisfy the required
performance criteria of high- speed operation and minimize
power dissipation of individual gates, we used HSPICE
optimization solver. This allowed us to produce the bestcase
parameter variation model for a specific subset of supply
voltages, voltage swings and biasing currents selected as the
input characteristics. Appendix describes the design and
optimization process for the MCML cell libraryindetail.Once
the MCML cell library optimization process was complete,
design of serialization, deserialization, and CDR circuits was
performed. Correct operation was verified and power
measured using SPICE simulation.
4 MCML DESIGN AND OPTIMIZATION-
Design of MCML circuits requires optimization of a large
number of parameters. Previous work in the field provided
an analytical description of all parameters used in the MCML
logic design process and reviewed the impact of these on
performance/power response. In this paper, we developed
an optimization toolkit, which allows deriving an MCML cell
library parameters in an automated way via using standard
SPICE descriptions of MOSFET transistors and satisfying the
specific criteria in power efficiency and performance
measured as system’s outputs. In the following section, we
review the major operation principles and properties of a
typical MCML cell and provide the optimization procedure
used throughout the cell development process.
A typical MCML gate is composed of three main blocks the
pull-up network, implemented as a set of resistors or active
p-MOS loads, the fully differential pull-downnetwork,which
steers the current between the branches, and the current
source. The performance of a gate is a function of various
metrics and is determined/evaluated by the corresponding
adjustments made in transistor sizing, biasing voltages,
reference currents/voltages, and differential voltageswings.
The operation of a standard MCML inverter cell can be
described as follows. Due to presence of active loads R, a
voltage drop V = I × R is produced, permitting logical 1 and 0
states to be represented as V dd and V dd – V voltages,
respectively. The use of active loads, implemented as p-MOS
transistors conducting in the linear region (assumed to
provide a roughly linear transfer function response), allows
online adaptabilitythathelpscompensatinganyspontaneous
variations inside the circuit. Typical resistance values are in
the order of 10 s of Ks and require sink currents to be in the
order of couple of hundreds microamperes. The increase in
transistor sizing, i.e., WP/L P ratio, lowers the load
resistance, and, as a rule, propagation delay of inverter
circuit; it is also followed by reduction in saturation voltage
of the p-MOS loads causing degradation in linear response.
An example of biasing circuit that is used for parameter’s
adjustment
Fig. 4. MCML inverter cell.
5. PREPROCESSING RESULTS AND HARDWARE
IMPLEMENTATION
A . Front end
Proposed transmitter is consist of both front end and back
end. Front end is consist of different block such as FIFO,
encoder, bitslip, this block’s are implemented by using
Verilog code. RTL schematic of top level module is shown in
fig 5.
Fig 5. RTL schematic of frontend
Front of proposed transmitter is designed by using VERILOG
code to understand the performance of understand. The
design of front end is accomplished by using XILINX 14.7
tool. Simulation result is shown in fig. 6

Fig. 6 Final simulation result of Front end.
Backend result-
Backend of proposed transmitter is consist of 64:1
multiplexer and it is a combination of combination of MCML
and CMOS circuit and it is implemented byusingTannerEDA
tool.
Fig. 7 Schematic of 2: 1 MUX MCML circuit
Backend of proposed transmitter is designed to understand
performance of MCML and CMOS circuit Simulation
accomplished in Tanner EDA 13.0byusing45nmtechnology.
S-edit of Tanner EDA is used in order to create the
schematics of the circuit. Simulation result shown in fig.7
Fig. 15 Final simulation result of 64:N bit MUX.
Table 1 : Simulation results of MCML and CMOS 64
MUX.
45nm
process
vdd Delay
[ns]
Power
[μW]
AND gate 0.9 0.019 0.38
NAND gate 0.9 0.019 0.38
INVERTER 0.9 0.00023 0.00198
2:1 MUX
MCML
0.9 4.75 41.67
6. Conclusion
We note that, as ultralow energy silicon photonic
communication components become commonplace, the
power consumption of the other transceiver components
must become the focus for major reductions in transceiver
power. Such reductions can only be obtained with attention
to the physical layer circuits and protocols of which SERDES
is the largest component. Our results show that the high-
speed subsystem, incorporating SERDES, CDR, and clock
recovery, can, despite relatively simple logic, consume50%–
60% of the total power. This is largely due to the integration
of standard CMOS and differential MCML components
operating at a high clock rate

References
1. R. S. Tucker, “Green optical communications—Part
II: Energy limitations in networks,” IEEE J. Sel.
Topics Quantum Electron., vol. 17, no. 2, pp. 261–
274, Mar./Apr. 2011.
2. D. A. Miller, “Rationale and challenges for optical
interconnects to electronic chips,”Proc. IEEE,vol.88,
no. 6, pp. 728–749, Jun. 2000
3. D. Huang, T. Sze, A. Landin, R. Lytel, and H. L.
Davidson, “Optical interconnects: Out of the box
forever?” IEEE J. Sel. Topics QuantumElectron.,vol.9,
no. 2, pp. 614–623, Mar./Apr. 2003.
4. L. A. Barroso and U. Holzle, “The case for energy-
proportional
5. computing,” IEEE Comput., vol. 40, no. 12,pp.33–37,
Dec. 2007.
6. O’Connor, “Optical solutions for system-level
interconnect,” in Proc. Int. Workshop Syst. Level
Interconnect Predict., 2004, pp. 79–88.

Power Optimized Transmitter for Future Switched Network

More Related Content

What's hot (20)

Similar to Power Optimized Transmitter for Future Switched Network (20)

More from IRJET Journal (20)

Recently uploaded (20)

Power Optimized Transmitter for Future Switched Network