Laskin MASc Thesis PDF

O N -C HIP S ELF -T EST C IRCUIT B LOCKS FOR
H IGH -S PEED A PPLICATIONS
BY
E KATERINA L ASKIN
A THESIS SUBMITTED IN CONFORMITY WITH THE REQUIREMENTS

FOR THE DEGREE OF M ASTER OFA PPLIED S CIENCE
G RADUATE D EPARTMENT OF E LECTRICAL AND C OMPUTER E NGINEERING
U NIVERSITY OF TORONTO

c E KATERINA L ASKIN , 2006
O N -C HIP S ELF -T EST C IRCUIT B LOCKS FOR
H IGH -S PEED A PPLICATIONS
Ekaterina Laskin
Master of Applied Science, 2006

Graduate Department of Electrical and Computer Engineering
University of Toronto
A BSTRACT
In this thesis, the advantages of parallel pseudo-random bit sequence (PRBS) generators for
high-speed self-test applications are examined. An ultra-low-power, 4-channel 27 − 1 PRBS
generator with 60 mW per channel was designed, fabricated and measured to work correctly
up to 23 Gb/s. The circuit was based on a 12-Gb/s, 2.5-mW BiCMOS current-mode logic
(CML) latch topology, which, to the best of my knowledge, represents the lowest power for a
latch operating above 10-Gb/s. The fabricated chip also included an integrated PRBS checker
and error counter. Techniques for further power reduction, by eliminating the current source
transistor, and speed improvements, by adding inductive peaking, are presented.
i
ii
Acknowledgements
This work would not be possible without the support and contributions of several people.
I would like to express my thanks to Prof. Sorin P. Voinigescu for providing me with this
opportunity and for giving ample guidance along the way. I also thank all the residents of
BA4182 for their help and support during design, tapeouts, and courses.
I would like to express my deepest appreciation to my family for their love and care. Thanks
to all my friends who tried their best to keep me from working too hard.
The following organizations are acknowledged for their support. STMicroelectronics for
providing design kits and fabrication; CMC for CAD support; NSERC, Micronet, and Gennum
for financial support; and CFI and OIT for equipment.
iii
iv
Table of Contents
Abstract i
Acknowledgements iii
List of Tables vii
List of Figures x
List of Abbreviations xi
1 Introduction 1
1.1 Built-In Self-Test for High-Speed Transceivers . . . . . . . . . . . . . . . . . 1
1.2 Speed and Power Optimization of High-Speed Logic . . . . . . . . . . . . . . 2
1.3 Objective of Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Contributions from this Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 PRBS Generator and Checker Architectures 5

2.1 Definition and Properties of Pseudo-Random Bit Sequences . . . . . . . . . . . 5
2.2 Series PRBS Generators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Parallel PRBS Generators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4 PRBS Checker Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 PRBS Generator Architecture Comparison for High-Speed Operation . . . . . 21
2.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3 27 -1 PRBS Generator and Checker Chip Design 25

3.1 System Description and High-Level Simulations . . . . . . . . . . . . . . . . . 25
3.1.1 MATLAB Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.2 Simulink Functional Simulations . . . . . . . . . . . . . . . . . . . . . 27
3.1.3 Verilog Timing Simulations . . . . . . . . . . . . . . . . . . . . . . . 28
v
vi Table of Contents
3.2 Design of Building Blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.2.1 1-mA and 2-mA Latches . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.2 12-Gb/s DFF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2.3 Selector, XOR, and AND Gates . . . . . . . . . . . . . . . . . . . . . 31
3.2.4 24-Gb/s 2-to-1 MUX . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.5 Clock, Data, and Output Buffers . . . . . . . . . . . . . . . . . . . . . 32
3.3 System-Level Design and Simulations . . . . . . . . . . . . . . . . . . . . . . 33
3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4 Experimental Results 39
4.1 Fabrication and Test Equipment . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 27 -1 PRBS Generator Measurements . . . . . . . . . . . . . . . . . . . . . . . 40
4.2.1 Generator Test Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2.2 Generator Measurement Results . . . . . . . . . . . . . . . . . . . . . 41
4.3 PRBS Checker Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3.1 Checker Test Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.3.2 Checker Measurement Results . . . . . . . . . . . . . . . . . . . . . . 46
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5 Comparison and Power Optimization of High-Speed Logic Topologies 49

5.1 High-Speed Digital Logic Gates . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.2 SiGe BiCMOS CML Logic / CML Latch Design . . . . . . . . . . . . . . . . 52
5.3 Power and Speed Optimizations of CML Latches . . . . . . . . . . . . . . . . 56
5.4 Performance Comparison and Simulation Results . . . . . . . . . . . . . . . . 58
5.5 Future Scaling and CMOS Logic . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.5.1 CMOS and MOS-CML Ring Oscillators . . . . . . . . . . . . . . . . . 62
5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6 Conclusion 67
6.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Bibliography 72
A Appendix 73
A.1 MATLAB Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
A.2 Simulink Schematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
A.3 Verilog Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
List of Tables
2.1 All 18 possible characteristic polynomials for a 27 − 1 PRBS . . . . . . . . . . 9

2.2 Standard characteristic polynomials for various PRBS lengths . . . . . . . . . 10
2.3 PRBS checker behaviour without errors (left), with errors spaced far apart
(right top), and with errors close together (right bottom) . . . . . . . . . . . . . 19
2.4 Comparison of the circuitry required for series and parallel PRBS generators
of different sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1 Maximum allowed block delays for system operation with 15-GHz clock . . . 28
3.2 Power consumption of the chip sub-blocks . . . . . . . . . . . . . . . . . . . . 33
4.1 Performance summary of the PRBS generator at different bit rates . . . . . . . 43
5.1 Model parameters for 0.13 µm and 90 nm SiGe BiCMOS technologies . . . . . 59

5.2 Device sizes for 1-mA latches in each configuration . . . . . . . . . . . . . . . 59
5.3 Simulated eye diagrams for 1-mA latches with different topologies . . . . . . . 60
5.4 Performance comparison of the different latch topologies . . . . . . . . . . . . 61
5.5 Summary of the designed 65-nm oscillators . . . . . . . . . . . . . . . . . . . 63
vii
viii List of Tables
List of Figures
1.1 Use of PRBS generator and checker for testing . . . . . . . . . . . . . . . . . 2
2.1 Diagram of a feedback shift register . . . . . . . . . . . . . . . . . . . . . . . 6

2.2 A 3-stage LFSR and one of its maximal-length sequences . . . . . . . . . . . . 8
2.3 A series generator for a 27 − 1 PRBS . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Algorithm for finding phases of added sequences to produce a sum sequence
with the required phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 A 25 − 1 series PRBS generator with its transition matrix . . . . . . . . . . . . 14
2.6 Series-parallel 25 −1 PRBS generators with their transition matrices and output
sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.7 Examples of practical parallel PRBS generators . . . . . . . . . . . . . . . . . 16
2.8 A simple PRBS checker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.9 A 27 − 1 PRBS checker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.10 A modified 27 − 1 PRBS checker . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.1 Top-level schematic of the designed chip . . . . . . . . . . . . . . . . . . . . . 26

3.2 Parallel and series implementations of 27 − 1 PRBS generators with 8 outputs . 27
3.3 Detailed schematics of the designed latches . . . . . . . . . . . . . . . . . . . 29
3.4 D-flip-flop schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5 Detailed schematics of other blocks . . . . . . . . . . . . . . . . . . . . . . . 31
3.6 2-to-1 MUX schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.7 Detailed schematics of buffers . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.8 Detailed system schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.9 Latch (top) and DFF (bottom) simulation with 15-GHz clock . . . . . . . . . . 35
3.10 System simulation with 15-GHz clock . . . . . . . . . . . . . . . . . . . . . . 35
3.11 Layouts (from left to right) of the clock buffer, XOR gate, and flip-flop (2
latches and emitter followers) . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
ix
x List of Figures
3.12 System simulation after extraction with 11-GHz clock . . . . . . . . . . . . . . 37
4.1 Die photo of the fabricated chip . . . . . . . . . . . . . . . . . . . . . . . . . 40

4.2 Measurement setup for PRBS generator . . . . . . . . . . . . . . . . . . . . . 41
4.3 Measurement results of the PRBS generator at 12-Gb/s . . . . . . . . . . . . . 42
4.5 Measured 23-Gb/s (top), measured 12-Gb/s (middle), and ideal (bottom) time
domain 27 − 1 PRB-sequences . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.7 Measurement setup for PRBS checker . . . . . . . . . . . . . . . . . . . . . . 46
4.8 Bit0 (bottom) and Bit1 (top) outputs of the error counter . . . . . . . . . . . . 47
5.1 CML digital gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

5.2 ECL gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.3 NMOS fT as a function of current density for several technology nodes and
transistor sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.4 HBT fT as a function of bias current for several transistor sizes . . . . . . . . . 52
5.5 CML latch schematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.6 Time-constant derivation for a BiCMOS latch . . . . . . . . . . . . . . . . . . 55
5.7 CML latch schematics without current source and VDD reduced to 1.8V . . . . 56
5.8 CML latch schematics with shunt inductive peaking . . . . . . . . . . . . . . . 57
5.9 CML latch schematics with shunt inductive peaking and without current source 57
5.10 MOS-CML and CMOS inverter schematics . . . . . . . . . . . . . . . . . . . 61
5.11 Top level schematics of the designed ring oscillators and quadrature oscillator . 63
5.12 Schematics of blocks used in the ring oscillators . . . . . . . . . . . . . . . . . 64
5.13 Layouts of the Ring Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . 64
A.1 Simulink test bench for the PRBS generator and checker . . . . . . . . . . . . 75
A.2 A series 27 − 1 PRBS generator in Simulink . . . . . . . . . . . . . . . . . . . 76
A.3 A parallel 27 − 1 PRBS generator in Simulink . . . . . . . . . . . . . . . . . . 77
A.4 A regular 27 − 1 PRBS checker in Simulink . . . . . . . . . . . . . . . . . . . 78
A.5 The implemented 27 − 1 PRBS checker in Simulink . . . . . . . . . . . . . . . 78
List of Abbreviations
BER Bit-Error Rate

BiCMOS Bipolar and CMOS
BIST Built-in Self Test
CDR Clock and Data Recovery
CML Current-Mode Logic
CMOS Complementary Metal Oxide Semiconductor
DEMUX De-multiplexer
DFF Data Flip-flop
DUT Device Under Test
ECL Emitter-Coupled Logic
FET Field-Effect Transistor
FSM Finite State Machine
fT Unity current gain frequency
HBT Hetero-junction Bipolar Transistor
LFSR Linear Feedback Shift Register
MIM Metal-Insulator-Metal
MOSFET Metal Oxide Semiconductor FET
MUX Multiplexer
NFET N-channel Field-Effect Transistor
NRZ Non-Return to Zero
PRBS Pseudo-Random Bit Sequence
SCFL Source-Coupled FET Logic
SERDES Serializer-Deserializer
SiGe Silicon-Germanium
TFF Toggle Flip-flop
TIA Transimpedance Amplifier
xi
xii List of Figures
1 Introduction
1.1. Built-In Self-Test for High-Speed Transceivers
Communication networks continually evolve to cover larger areas, reach more places, and
operate at increasing data rates. Currently, the 10-Gb/s Ethernet standard is being deployed for
industrial and commercial use. Hence, research is focused on next generation 40-Gb/s Sonet
and 100-Gb/s Ethernet systems. Recently, chipsets for 40-Gb/s transceivers [1] have been
demonstrated. Furthermore, digital building blocks for 80-Gb/s applications and above [2, 3]
are being developed in silicon, with increasing integration levels. As the data rates of state-
of-the-art broadband circuits increase, these circuits outperform commercially available test
equipment and verification of error-free operation becomes more challenging. Hence, custom
circuits are needed to generate input data for testing purposes. This input data must resemble
real input signals as closely as possible. Previously, pseudo-random bit sequence (PRBS)
generators with sequence length of 231 − 1 have been developed [4, 5]. However, the testing
circuits need to be compact and low power, suitable for use as circuit blocks that can be placed
on the same chip as the system under test. For bit-error-rate testing, input signals with the
suitable properties can be generated using linear-feedback shift registers (LFSR). The produced
signals are called pseudo-random bit sequences due to their spectral content. Even though the
LFSR structures are relatively simple, the main challenges lie in implementing them at high
data rates with the restrictions of minimizing power and area.
A general setup for testing of broadband circuits is shown in Figure 1.1(a). Here, a test
signal is fed into the device under test (DUT) and then compared with the output of the DUT.
In this context, the term “DUT” refers to a broadband circuit. The general testing configuration
shown in Figure 1.1(a) is not possible for most practical circuits because their input and output
are far apart, such that the test signal cannot be connected directly as shown. On the other hand,
if the input signal is known and if the system clock is recovered, it can be regenerated at the
1
2 Introduction
output side and then compared against the DUT output to check whether errors occur inside the
DUT. This can be done with a PRBS generator, because the PRBS signal is deterministic and
can be recreated exactly, once the clock signals are synchronized. The bit error rate test setup
for a practical system is shown in Figure 1.1(b), where a PRBS generator produces the pseudo-
random signal which is sent into the device being tested. On the output side of this device, a
PRBS checker recreates the same pseudo-random signal as the generator and compares it with
the output of the DUT to find errors. For this to happen the clock signals and data streams in the
PRBS generator and error checker must be synchronized. The device shown in Figure 1.1(b)
is a complete SERDES, however smaller sub-circuits, such as a TIA, driver, re-timer, CDR,
MUX-DEMUX, transmitter, or receiver, can also be tested using the same method.
1.2. Speed and Power Optimization of High-Speed Logic
Although the main application of PRBS generators is testing, they contain digital blocks that
can be reused in other applications. Thus, to achieve high system integration in the future, it
is essential to find how the design of each constituent block can be optimized for speed and
power.
Error
Test
DUT =?
Signal
(a) Conceptual setup for BER test of a device
MUX DRV TIA LA CDR DEMUX

DATA DATA
CLOCK
PRBS
PRBS ÷N
÷N PLL CHECKER
CLOCK
GENE-
Device Under Test
RATOR
ERROR
(b) Use of a PRBS for system BER testing
Figure 1.1: Use of PRBS generator and checker for testing

1.3 Objective of Thesis 3
At data rates above 10 Gb/s, current-mode logic or emitter-coupled logic are used to imple-
ment digital gates. In these types of circuits, power consumption is given by the supply voltage
multiplied by the constant current that biases the gate. Previously, bipolar-only or MOS-only
implementations of the gates have been made. However, bipolar-only implementations require
a high supply voltage. MOS-only implementations operate from a lower supply but require
more current to reach the same speed of operation. Hence, a BiCMOS gate implementation,
which consists of both MOS and HBT devices [6], is explored in this thesis as an alternative
configuration that demands less power to operate at a given speed.
Circuit design is a very challenging task, especially at very high data-rates that approach
the maximum switching frequency of the transistors. To ease this problem, a procedure will
be presented that explains how to choose transistor sizes and bias points to achieve high-speed
gate design with low power consumption.
1.3. Objective of Thesis
The objective of this thesis is two-fold. The first target was to implement a 27 − 1, 23-Gb/s
PRBS generator and error checker blocks, such that they could later be used for testing of
other circuits. The second goal was to find how the design of high-speed digital blocks can be
optimized for minimal power while still operating at a given data rate (above 10 Gb/s).
Both of these objectives are combined in this thesis because the PRBS generator was orig-
inally designed to be the first stage of an 80-Gb/s transceiver with 10-Gb/s inputs. Thus, the
lowest data rate in the generator is above 10 Gb/s, and the output of the generator works up to
23 Gb/s to have margin over 80 Gb/s after 4-to-1 multiplexing.
1.4. Contributions from this Work
The work to be presented in this thesis contains several contributions to the field of high-speed
digital circuit design and testing.
The first contribution is the design of a low-power 27 − 1 PRBS generator with 4, appro-
priately delayed, parallel output streams at 23-Gb/s each. This PRBS generator was fabricated
and verified to operate correctly. Thus it can be integrated as a self-test reusable circuit block
for other systems. The four outputs of the generator can be further multiplexed to an aggre-
gate PRBS output at 92 Gb/s with minimal circuitry. The 4-channel PRBS generator consumes
4 Introduction
235 mW from 2.5 V, which results in only 60 mW per output lane. The entire fabricated chip,
which also includes an error checker and a 5-bit error counter, consumes 940 mW.
The second contribution is the design of a low-power SiGe BiCMOS CML latch, which
worked up to 12 Gb/s, while consuming only 2.5 mW. This latch has enabled the low power
performance of the PRBS generator circuit. It opens the possibility to develop highly integrated
broadband systems that operate above 10 Gb/s, with low power consumption. An investigation
into further power reduction and speed improvement up to 40 Gb/s of high-speed digital gates
in SiGe BiCMOS and 65-nm CMOS technologies is also presented in this thesis.
Additionally, a thorough analysis of series and parallel PRBS generator architectures is
presented, concerning their power requirements, their applicability to high-speed implementa-
tion, and the respective design challenges. The comparison will be done in terms of the circuit
complexity (i. e. number of flip-flops, gates, and their fanout) required by each architecture to
implement the needed function.
The PRBS error checker implementation that is described in this thesis is novel. As opposed
to existing PRBS error checkers that produce three or fewer pulses for each error present in the
signal, the proposed checker circuit generates only one pulse for each error in the input signal.
Thus, it is more useful for precise bit error rate measurements.
PRBS Generator and
2 Checker Architectures
This chapter presents the theory of generating pseudo-random bit sequences. It starts by pre-
senting the mathematical properties of pseudo-random bit sequences in section 2.1. Then, the
two existing series and parallel PRBS generator configurations are described in section 2.2
and section 2.3 respectively. Various PRBS checkers are shown in section 2.4. Finally, sec-
tion 2.5 presents a novel comparison of the applicability of series and parallel pseudo-random
bit sequence generator and checker topologies to high-speed applications.
2.1. Definition and Properties of Pseudo-Random Bit Sequences
Pseudo-Random Bit Sequence generators are finite state machines (FSM) consisting of a linear
feedback shift register (LFSR) circuit. In general, a shift register has one input, and n memory
elements, called the stages of the register. Each memory element of the register stores one bit
of data, resulting in n outputs from the register. The contents of the shift register at any given
time are called the state of the register. Feedback is added to a general shift register by means
of a circuit that realizes a function of n inputs and 1 output. The n inputs of the function are
taken from the n elements of the shift register and its one output is connected to the input of
the shift register, as shown in Figure 2.1. A Feedback Shift Register is called linear when the
feedback function can be expressed in the form [7]
g(x1 , x2 , ...xn ) = c1 x1 ⊕ c2 x2 ⊕ c3 x3 ⊕ ... ⊕ cn xn (2.1)
where each of the constants ci is either 0 or 1, and ⊕ denotes addition modulo 2. When some of
the constants ci are 0, then the number of inputs of g is less than n. Addition of linear feedback
to a shift register makes the output become periodic (theorem 1 in [7]). Each of the n stages of
the LFSR can be either 1 or 0, resulting in a maximum of 2n different states in which the LFSR
5
6 PRBS Generator and Checker Architectures
can be found. Depending on the feedback function g and on the initial state of the LFSR, the
number of different states can be reduced down to 1.
Since any LFSR repeats its states periodically, the output of each stage is also periodic.
The period of the output is the length of the sequence that the LFSR generates. Although the
maximum number of states the LFSR can be in is 2n , the maximum length of the sequence that
can be generated by the same LFSR is 2n − 1. This is because a state with all 0’s stored in the
LFSR can never occur when the output is periodic (with a period greater than 1). Conversely
the all 0 state can be regarded as being periodic with the period equal to 1. Depending on the
feedback function g, it is possible to construct the same n-stage shift register to generate either
sequences with maximum 2n − 1 length (also called ”maximal-length”, or m-sequences) or
sequences with shortened lengths.
It so happens that m-sequences also have randomness properties that make them appear
as noise. However, these sequences are periodic and therefore not truly random. They are
also called pseudo-random bit sequences (PRBS). The randomness properties of PRBS are the
following [7]:
1. A balance of 1 and 0 terms.
2. Two runs of length n for each run of length n + 1.
3. A two-level auto-correlation function.
A balance of 1 and 0 terms implies that the number of 1’s and 0’s in a pseudo-random
sequence is different by at most 1. As a result, for longer sequences, the probability of 1’s
and 0’s gets closer to 1/2.
A run of length n is part of a sequence where n identical bits occur consecutively. Because
in a PRBS there are two runs of length n for each run of length n + 1, 1/2 of the number of
x1 x2 x3 x4 xn-1 xn
D Q g(x1, x2, … , xn)
Figure 2.1: Diagram of a feedback shift register

2.1 Definition and Properties of Pseudo-Random Bit Sequences 7
runs is of length 1, 1/4 of the runs is of length 2, 1/8 of length 3, etc. . . Analogously, in a
random coin toss experiment, 2 consecutive heads occur with probability 1/4, 3 consecutive
heads occur with probability 1/8, etc. . .
The auto-correlation function of any binary periodic sequence {an } with period p, which
is composed of +1’s and −1’s is defined as [7]
p
1X
C(τ ) = an an+τ (2.2)
p n=1
where τ is the phase shift in bits. C(τ ) is always highest for τ = 0 and drops off for larger τ .
For a PRBS, C(τ ) = 1 for τ = 0 and C(τ ) < 1 for 0 < τ < p. A two-level auto-correlation
function confirms that the sequence is as close as possible to being random. This is because the
sequence is uncorrelated with phase shifts of itself, unless the phase shift is equal to an integer
number of sequence periods.
Thanks to these randomness properties, PRB-sequences are useful as data sources when
testing other circuits or systems. On one hand they are close to being random and therefore
exercise the device under test (DUT) for many different input combinations. On the other hand,
they are completely defined and repeatable and therefore more useful than just digitized noise.
PRB-sequences can be generated from a LFSR with any number of stages by selecting an
appropriate feedback function [8]. However certain sequence lengths are more widely used
than others because they are standardized [9]. The common PRB-sequence lengths are 27 −
1, 215 − 1, 223 − 1, and 231 − 1 requiring at least 7, 15, 23, and 31 memory elements in
the LFSR, respectively. Longer sequence lengths are desirable because they repeat less often
and have longer run lengths, which subjects the DUT to more exhaustive tests. However, the
main challenge of building PRBS generators that produce longer sequence lengths is that they
consume more power and occupy more die area.
For a LFSR with n stages and a feedback function g(x1 , x2 , ...xn ) = c1 x1 ⊕ c2 x2 ⊕ c3 x3 ⊕
... ⊕ cn xn , the output of each stage is a PRBS {ak } with period p = 2n − 1. As described in [7],
{ak } satisfies the recurrence relation
n
X
ak = ci ak−i (modulo 2). (2.3)
i=1
This can be seen by looking at the example of a PRBS with period p = 7, shown in Figure 2.2.
The produced sequence is shown in columns.
The polynomial
Xn
f (x) = 1 + ci xi (modulo 2) (2.4)
i=1
x1 x2 x3
x0 x1 x2 x3 1 1 1
0 1 1
1 0 1
0 1 0
0 0 1
1 0 0
g = x1 + x3 1 1 0
1 1 1
Figure 2.2: A 3-stage LFSR and one of its maximal-length sequences
is called the characteristic polynomial of the sequence {ak }. The necessary condition for {ak }
to be a PRBS is that f (x) is irreducible, that is it cannot be factored. The necessary and
sufficient condition is that 1 + xm can be divided (modulo 2) exactly by f (x) for m = p, but
for no positive integer m smaller than p [7]. The number of distinct pseudo-random sequences
of length p = 2n − 1 is φ(p)/n, where φ(p) is Euler’s function1 . For example there are 18
distinct characteristic polynomials that can generate a PRBS of length 27 − 1 (Table 2.1). Out
of these 18, the polynomials in rows 1-4 of Table 2.1 require the feedback function to add the
outputs of only 2 stages, thus decreasing the amount of circuitry need to generate the PRBS.
However, the PRBS shown in row 2 is used most often because it is one of the standard testing
sequences [9].
In addition to the above definitions and randomness properties, m-sequences posses two
other useful properties. These properties are very practical in applications for building PRBS
generators, as will be shown later. First, PRB-sequences possess the “cycle-and-add” prop-
erty [7]. This means that when two identical sequences, which are phase shifted with respect
to each other, are added bitwise (modulo 2), then the result is an identical sequence but with
some phase shift. This is true because PRB-sequences always satisfy a linear recurrence rela-
tion. Therefore, given two PRB-sequences Ai and Aj that satisfy a linear recurrence relation
R, then Ai + Aj also satisfies R and therefore is the same PRB-sequence. Here the addition is
bit-wise and modulo 2.
The second property relates to decimation of PRB-sequences [7]. Decimation is defined
as forming a sequence {aqk } from the sequence {ak } by taking every q th bit of {ak } (q is a
1
The totient function φ(p), also called Euler’s totient function, is defined as the number of positive integers
≤ p that are relatively prime to (i.e., do not contain any factor in common with) p, where 1 is counted as being
relatively prime to all numbers. Since a number less than or equal to and relatively prime to a given number is
called a totative, the totient function φ(p) can be simply defined as the number of totatives of p. For example,
there are eight totatives of 24 (1, 5, 7, 11, 13, 17, 19, and 23), so φ(24) = 8 [10].
2.1 Definition and Properties of Pseudo-Random Bit Sequences 9
Row Characteristic Polynomial Pseudo-Random Bit Sequence

7
1 x +x+1
2 x7 + x6 + 1
3 x7 + x3 + 1
4 x7 + x4 + 1
5 x7 + x3 + x2 + x + 1
6 x7 + x6 + x5 + x4 + 1
7 x7 + x4 + x3 + x2 + 1
8 x7 + x5 + x4 + x3 + 1
9 x7 + x5 + x2 + x + 1
10 x7 + x6 + x5 + x2 + 1
11 x7 + x5 + x3 + x + 1
12 x7 + x6 + x4 + x2 + 1
13 x7 + x5 + x4 + x3 + x2 + x + 1
14 x7 + x6 + x5 + x4 + x3 + x2 + 1
15 x7 + x6 + x3 + x + 1
16 x7 + x6 + x4 + x + 1
17 x7 + x6 + x5 + x3 + x2 + x + 1
18 x7 + x6 + x5 + x4 + x2 + x + 1
Table 2.1: All 18 possible characteristic polynomials for a 27 − 1 PRBS
positive integer). Decimation of any PRB-sequence forms another PRB-sequence. However,

the important property is that when q is a power of 2, decimation by q does not change the
order of bits in the PRBS. That is, if {ak } is a PRB-sequence then {aqk } differs from {ak } by
at most a phase shift, for q = 1, 2, 4, 8, . . . , 2n−1. This can be proved as follows. Let D be a
unit delay operator, so that Dak = ak−1 and D 2 ak = ak−2 . Then, from equations 2.3 and 2.4,
X X X
f (D){ak } = (1+ ci D i ){ak } = {ak }+ ci D i {ak } = {ak }+ ci {ak−i } = {ak }+{ak } = 0
(2.5)
So f (D){ak } = 0 is equivalent to the recurrence relation of equation 2.3. Also, from the
simplified binomial theorem2 , it follows that
i i
f (x2 ) = [f (x)]2 . (2.6)
i i i
2
(a ⊕ b)2 = a2 ⊕ b2
Now, consider the sequence {a2k }, formed by taking every second bit of {ak }, or equivalently,
by the operation
f (D 2){ak } = f (D){a2k }. (2.7)
By using equations 2.6 and 2.7 together with f (D){ak } = 0, we obtain
f (D){a2k } = f (D 2){ak } = [f (D){ak }]2 = f (D)[f (D){ak }] = f (D){0} = 0. (2.8)
Thus, {a2k } satisfies f (D){a2k } = 0, which is the same recurrence relation as equation 2.3
for {ak }. Therefore, the sequence {a2k } is identical to {ak }, except for a possible phase shift.
The same proof can be extended to show that {aqk } for q = 1, 2, 4, 8, . . . , 2n−1 is also the same
sequence as {ak }.
The decimation property of PRBS proves to be most useful when applied in reverse. That
is, two appropriately shifted sequences {a2k } and {a2k−j } can be multiplexed into {bk } by
alternating the bits from {a2k } and {a2k−j }. In order for {bk } to also be pseudo-random, the
phase shift must be j = (p − 1)/2 [11], where the length of the sequence is p = 2n − 1 bits.
This operation effectively doubles the bit rate of the generated PRBS.
To the best of my knowledge there exist two different circuit architectures to generate PRB-
sequences. The first one is series architecture, which directly applies the LFSR theory. The
second one still generates the same PRBS, but is optimized to generate several shifted se-
quences in parallel [12].
2.2. Series PRBS Generators
Series PRBS generators are, in effect, just shift registers with a linear feedback function that
causes the shift register to output a maximal length sequence. As mentioned before, for each
shift register, several different characteristic polynomials can generate a PRBS, however, typ-
Shift Register Length n Characteristic Polynomial PRBS Length

7 x7 + x6 + 1 27 − 1 = 127
9 x9 + x5 + 1 29 − 1 = 511
11 x11 + x9 + 1 211 − 1 = 2, 047
15 x15 + x14 + 1 215 − 1 = 32, 767
23 x23 + x18 + 1 23
2 − 1 = 8, 388, 607
29 x29 + x27 + 1 229 − 1 = 536, 870, 911
31 x31 + x28 + 1 231 − 1 = 2, 147, 483, 647
Table 2.2: Standard characteristic polynomials for various PRBS lengths

2.2 Series PRBS Generators 11
ically only certain characteristic polynomials are used. To minimize the number of gates re-
quired for addition, these polynomials are of the form 1 + xi + xn , where n is the number of
stages in the shift register. The PRBS characteristic polynomials that are used in measuring
and testing equipment are summarized in Table 2.2 [9]. Figure 2.3 shows a schematic of a
series PRBS generator, using D-flip flops as memory elements and an XOR gate as the adder,
that generates a sequence 127 bits long.
From Figure 2.3 it can be observed that the PRBS bits are output sequentially (hence the
name ”series generator”) from the generator. That is, during the first clock cycle, the stage
outputs produce bits 1, 2, 3, 4, 5, 6, 7, during the second clock cycle the outputs have bits 2, 3,
4, 5, 6, 7, 8, during the third clock cycle bits 3, 4, 5, 6, 7, 8, 9 of the sequence are produced,
and so on. Note that because the sequence is periodic, the starting bit can be chosen arbitrarily.
Because in series PRBS generators the sequence bits are generated sequentially, they are not
well suited for multiplexing to higher bit rates. In this case, before multiplexing can be applied,
additional circuitry is needed. To multiplex two PRB-sequences into a PRBS at double the
original bit rate, the phase shift between the two incoming sequences must be (p − 1)/2 [11],
where p = 2n − 1 is the length of the sequence in bits. For example for a 27 − 1 PRBS,
bits 1 and 64 of the sequence must be available during one clock cycle, bits 2 and 65 during
the next clock cycle, etc. . . but these bits are not available at the same time in a series PRBS
generator. Fortunately, this problem can be solved by using more circuitry and applying the
“cycle-and-add” property of PRBS.
The “cycle-and-add” property states that by adding two sequences, a third sequence with
a different phase is produced. But here the problem is reversed. We know the required phase
of the sequence and need to find the phases of the other two (or more) sequences that, when
added, give the required phase. Furthermore, the phases of the added sequences must be such
B C D E F G H
A D Q D Q D Q D Q D Q D Q D Q
CLOCK IN
CLOCK 1 7 6 5 4 3 2 1
CLOCK 2 8 7 6 5 4 3 2
CLOCK 3 9 8 7 6 5 4 3
CLOCK 4 10 9 8 7 6 5 4
CLOCK 5 11 10 9 8 7 6 5
... ... ... ... ... ... ... ...
Figure 2.3: A series generator for a 27 − 1 PRBS

Example:
Feedback polynomial: f(x) = x7+x6+1
start Variable Definitions: Required phase shift: d = 25 bits
d– required phase shift in bits d = 1·20 + 0·21 + 0·22 + 1·23 + 1·24

initialization
a0 = a3 = a4 = 1, a1 = a2 = 0, s = 4
i=s
ai – binary digits of d
r(x) = 1
Initialization: i = 4, r(x) = 1
s– highest power of 2 in the Execution:
binary representation of d 1) r(x) = xa4·12 = x
execution 2) r(x) = x
r(x) – a polynomial test: i = 4 ≠ 0 i=3
1) r(x) = xai r2(x)
2) r(x) = r(x) mod f(x)
representing the 1) r(x) = xa3·x2 = x3
connections needed for 2) r(x) = x3
the required phase test: i = 3 ≠ 0 i=2
1) r(x) = xa2·x6 = x6
i=0? i=i-1
f(x) – the feedback polynomial 2) r(x) = x6
test: i = 2 ≠ 0 i=1
xai – terms used to form r(x) 1) r(x) = xa1·x12 = x12
2) r(x) = x6 + x5 + x4 + x3 + x2 + x + 1
correction bi – coefficients of the final r(x) test: i = 1 ≠ 0 i=0
r’(x) = r(x) + b0f(x) terms 1) r(x) = x13 + x11 + x9 + x7 + x5 + x3 + x
2) r(x) = x6 + x3 + x2
test: i = 0 stop
b2 = b3 = b6 = 1, b0 = b1 = b4 = b5 = 0
stop
25-bit delay is achieved by adding
the outputs of stages 2, 3, and 6
Figure 2.4: Algorithm for finding phases of added sequences to produce a sum sequence
with the required phase
that they are directly available from the shift register stages. To do that, an efficient, O(log(n)),
algorithm exists [13]. The algorithm is shown in Figure 2.4. Sometimes, many adders are
needed to produce the required phase for multiplexing, but since the beginning of the sequence
is arbitrary, there is some freedom in the choice of the reference sequence for multiplexing.
For example, to multiplex a 27 − 1 PRBS, any of the pairs with phases 1 and 64, or 2 and 65,
or 3 and 66, . . . , or 7 and 70 can be used. Therefore, the algorithm of Figure 2.4 can be run
iteratively to find the optimal (in terms of the required number of additions) pair of sequences
to multiplex.
Multiplexing can be repeated more than once to increase the bit rate of the final sequence
by 2, 4, 8, . . . , 2n−1 times. To multiplex q times, q PRB-sequences are required at the original
bit rate. Here q is a power of 2 because the basic multiplexing circuit combines 2 inputs into 1
at twice the bit rate. To ensure that the final output is a PRBS, the q original sequences must be
equally spaced apart by (p − 1)/q bits in phase, where p = 2n − 1 is the length of the sequence
in bits. However, generating all the appropriately phased sequences may require a large amount
of additions (XOR gates), and offset the benefit of generating the PRBS at a lower bit rate.
2.3 Parallel PRBS Generators 13
2.3. Parallel PRBS Generators
It is possible to produce a PRBS using a structure different than a LFSR, as long as it im-
plements the correct characteristic polynomial. Parallel PRBS generators are an extension of
series generators and can produce several shifted sequences in parallel. The phase shift be-
tween the parallel output sequences is such that they can be directly multiplexed to higher bit
rates. Any series PRBS generator with n memory elements can be restructured into a series-
parallel PRBS generator with n memory elements and k parallel outputs at the expense of k −1
additional XOR gates. When k = n, the generator is completely parallel [14].
For analyzing and constructing parallel PRBS generators, it is convenient to represent the
circuit using a transition matrix T that represents how data is transferred between the memory
elements of the generator. If U(j) is an n × 1 vector of 1’s and 0’s that represents the state
(what is stored in each memory element) of an n-stage PRBS generator at the j th clock cycle,
then T is an n × n matrix that can be used to find the state of the generator at the next clock
cycle.
U(j + 1) = T · U(j). (2.9)
Consequently, the state of the generator after k clock cycles will be
U(j + k) = Tk · U(j). (2.10)
The transition matrix T is closely related to the characteristic polynomial of the m-sequence.
The columns of T correspond to the data stored in the stages of the PRBS generator, assuming
the stages are numbered in the same direction as the data is shifting. The rows of T correspond
to the connections that exist between stages. That is, each 1 in a particular row corresponds to a
connection made from a stage represented by the column of the 1 to a stage represented by the
row of the 1. In rows where more than one 1 appear, the inputs are summed (modulo 2) before
the connection is made. All other entries of the matrix are 0 [14]. For a series PRBS generator
(LFSR), T contains 1’s in the diagonal below the main diagonal and 1’s in the first row and
columns numbered the same as the exponents of the characteristic polynomial. An example
of a transition matrix T that describes a 25 − 1 series PRBS generator with the characteristic
polynomial f (x) = x5 ⊕ x3 ⊕ 1 is shown in Figure 2.5, which also includes several periods of
the produced sequence.
When the number of outputs of an n-stage PRBS generator is increased to 1 < k < n,
then Tk describes the connections between the PRBS generator blocks; it is constructed as
follows. A diagonal array of (n − k) 1’s appears k rows below the main diagonal. The k th
row of Tk is the same as the first row of T. Row i, for i < k, is the same as row 1 of T
0 0 1 0 1
1 0 0 0 0
T= 0 1 0 0 0
0 0 1 0 0
0 0 0 1 0
x1 x2 x3 x4 x5
out 1
Figure 2.5: A 25 − 1 series PRBS generator with its transition matrix
shifted left by k − i column positions, with zeros shifted in from the right-hand side; if 1’s
are shifted out to the right, then a copy of row k is added (modulo 2) to row i. To illustrate
this process, 25 −1 series-parallel PRBS generators implementing the characteristic polynomial
f (x) = x5 ⊕x3 ⊕1 are shown in Figure 2.6 for all 5 possible values of k with the corresponding
Tk [14, 15]. Beside each generator, the k produced sequences are shown for several periods.
One of the periods is highlighted to emphasize the phase shift between them. Note that only
T2 and T4 of Figure 2.6 produce the exact same sequences as that of T in Figure 2.5, while
T3 and T5 produce different sequences that are also PRBS. This effect is a consequence of the
decimation property of pseudo-random bit sequences.
Parallel PRBS generators operate by decimation. That is, when k parallel outputs are gen-
erated from a PRBS, a transition matrix Tk is used, meaning that in one clock cycle, according
to equation 2.10, the generator goes through k states. As mentioned before, decimating a PRB-
sequence changes its phase. This effect is also shown in Figure 2.6. Conveniently, since the
the k parallel sequences are decimated by k and phase shifted accordingly, they can be easily
serialized into one PRBS stream at k times the original bit rate. This can be done by direct
multiplexing and without requiring additional phase shifting circuitry as in the case of a series
PRBS generator. It should be noted that the k parallel PRBS streams require k − 1 additional
adders (XOR gates) compared to a series PRBS generator.
The small parallel generators shown in Figure 2.6 are not practical. Usually, in practice,
sequence lengths of 27 − 1, 215 − 1, 223 − 1, 231 − 1 are required and k = 2, 4, 8 . . . is used
because it is easy to multiplex. Two examples of practical parallel PRBS generators are shown
in Figure 2.7. The first one (Figure 2.7(a)) is a 27 − 1 PRBS generator with 8 parallel outputs.
The table shows the phase-shift in bits of each output. The second schematic (Figure 2.7(b))
shows a 231 − 1 PRBS generator with 8 parallel outputs. Due to space limitation, the memory
elements (flip-flops) of the chain are shown as small numbered blocks. Each number represents
the delay from the beginning of the chain, analogous to Figure 2.6.
2.3 Parallel PRBS Generators 15
0 1 0 1 0
0 0 1 0 1
T2 = 1 0 0 0 0
0 1 0 0 0
x2 x4 0 0 1 0 0
x1 x3 x5
out 2
out 1
(a) 25 − 1 PRBS generator with 2 parallel outputs
1 0 1 0 0
0 1 0 1 0
T3 = 0 0 1 0 1
x3 1 0 0 0 0
0 1 0 0 0
x2 x5
out 3
x1 x4
out 1
out 2
(b) 25 − 1 PRBS generator with 3 parallel outputs
x2
out 4 0 1 1 0 1
1 0 1 0 0
T4 = 0 1 0 1 0
0 0 1 0 1
x3 1 0 0 0 0
x4 out 3
x1 x5 out 2
out 1
(c) 25 − 1 PRBS generator with 4 parallel outputs
1 1 0 1 0
0 1 1 0 1
5
T = 1 0 1 0 0
x5 0 1 0 1 0
0 0 1 0 1
x3 out 1
x1 out 3
out 5
x4
out 2
x2
out 4
(d) 25 − 1 PRBS generator with 5 parallel outputs
Figure 2.6: Series-parallel 25 −1 PRBS generators with their transition matrices and output
sequences
out 8 out 7 out 6 out 5 out 4 out 3 out 2 out 1

CLOCK 1 112 96 80 64 48 32 16 0
CLOCK 2 113 97 81 65 49 33 17 1
CLOCK 3 114 98 82 67 50 34 18 2
CLOCK 4 115 99 83 68 51 35 19 3
CLOCK 5 116 100 84 69 52 36 20 4
... ... ... ... ... ... ... ... ...
(a) 27 − 1 PRBS generator with 8 parallel outputs
1 2
8
6 4 out 1
1 2 3
7
5 3 1 out 2
1 2 3
6
4 2 0 out 3
1 2 2
5
3 1 9 out 4
1 2 2
4
2 0 8 out 5
1 1 2
3
1 9 7 out 6
1 1 2
2
0 8 6 out 7
1 2
1 9
7 5 out 8
(b) 231 − 1 PRBS generator with 8 parallel outputs
Figure 2.7: Examples of practical parallel PRBS generators

2.4 PRBS Checker Design 17
2.4. PRBS Checker Design
PRBS generators are used as data sources on the transmit side, or on the input of a device
under test (DUT). PRBS checkers are needed on the receive side, or at the output of a DUT,
to verify correct transmission of data or to find the bit-error-rate (BER) of the DUT. To check
a pseudo-random sequence for correctness, it has to be compared to a reference sequence on
the receiver. For that, the incoming sequence and the local reference have to be synchronized.
Furthermore, correct synchronization must be maintained over long periods of time.
Synchronization can be achieved quickly by using the idea presented in [16]. This idea
makes use of the series PRBS generator structure, but with the feedback loop broken. Fig-
ure 2.8 shows how this can be done for a 25 − 1 PRBS, discussed earlier. Here, the feedback
loop is broken between nodes A and G, which used to be one node in the generator case. Node
A becomes an input, and node G becomes an output. If a pseudo-random bit sequence is con-
nected to the created input at node A, the created output G will give an identical sequence
because the input and output used to be one node. Now it becomes easy to check the input
signal for errors. The input signal is simply compared to the output using an XOR gate. When
the input signal is a PRBS, the comparator will produce a zero output. Whenever a wrong bit
appears in the input, it will propagate through the shift register and give rise to a wrong bit,
making the two inputs of the comparator different, which will indicate an error. As shown in
Figure 2.8, a buffer or a retiming flip-flop may be required at the checker input to reduce the
load at that node. Although this checker was shown for a 25 − 1 PRBS, the same idea can be
used to make a checker for a PRBS of any length.
This simple checker has two drawbacks. First, for errors that occur rarely (spaced apart by
more than the shift register length), the checker indicates 3 errors for every error that is actually
INPUT A B C D E
G F
H ERROR
Figure 2.8: A simple PRBS checker

present in the input. Second, for errors that occur more often, the checker indicates 3 or less
errors, depending on the spacing of the errors. For precise BER counts, it would be desirable
to reduce the variability in the number of detected errors. This problem is also illustrated
in Table 2.3 [16]. Letters A-H of the table column names correspond to the node names of
Figure 2.8. The left part of Table 2.3 shows how the checker works with correct PRBS input.
The absence of errors can be seen because the bits of node A are equal to the bits of column G.
The right part of Table 2.3 shows how errors propagate through the checker. Errors e1 and e2
(top part of the table) occur far apart in the input and therefore each of them generates 3 error
bits in the output node H. However, errors e3 and e4 occur close together, and both of them
generate only 4 error bits in the output.
The PRBS checker behaviour can also be analyzed mathematically. Assume that the signal
at each node, Snode , consists of a PRBS bit D i and an error bit E i , where i indicates a delay
of i clock cycles with respect to a set reference. With reference to Figure 2.8, and noting that
D 3 +D 5 = D 0 (the characteristic polynomial for this PRBS), the signals at the important nodes
are:
SA = D 0 ⊕ E 0
SD = D 3 ⊕ E 3
SF = D 5 ⊕ E 5
SG = SD ⊕ SF = D 3 ⊕ D 5 ⊕ E 3 ⊕ E 5 = D 0 ⊕ E 3 ⊕ E 5
SH = SA ⊕ SG = D 0 ⊕ E 0 ⊕ D 0 ⊕ E 3 ⊕ E 5
= E0 ⊕ E3 ⊕ E5 (2.11)
Indicating that when one error appears at the input node A then three errors appear at the output
node H.
An extension of the small 25 − 1 PRBS checker of Figure 2.8 into a more practical 27 − 1
PRBS checker version is shown in Figure 2.9. However, this 27 − 1 PRBS checker, as before,
INPUT A B C D E F G H
J ERROR
Figure 2.9: A 27 − 1 PRBS checker

2.4 PRBS Checker Design 19
No Errors Remote Errors or Close Errors

A B C D E F G A B C D E F G H
1 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0
2 0 0 1 0 0 0 0 1 (e1 ) 0 1 0 0 0 0 1 (e1 )
3 1 0 0 1 0 0 1 1 1 (e1 ) 0 1 0 0 1 0
4 0 1 0 0 1 0 0 0 1 1 (e1 ) 0 1 0 0 0
5 1 0 1 0 0 1 1 1 0 1 1 (e1 ) 0 1 0 1 (e1 )
6 1 1 0 1 0 0 1 1 1 0 1 1 (e1 ) 0 1 0
7 0 1 1 0 1 0 0 0 1 1 0 1 1 (e1 ) 1 1 (e1 )
8 0 0 1 1 0 1 0 0 0 1 1 0 1 0 0
9 1 0 0 1 1 0 1 0 (e2 ) 0 0 1 1 0 1 1 (e2 )
10 1 1 0 0 1 1 1 1 0 (e2 ) 0 0 1 1 1 0
11 1 1 1 0 0 1 1 1 1 0 (e2 ) 0 0 1 1 0
12 1 1 1 1 0 0 1 1 1 1 0 (e2 ) 0 0 0 1 (e2 )
13 1 1 1 1 1 0 1 1 1 1 1 0 (e2 ) 0 1 0
14 0 1 1 1 1 1 0 0 1 1 1 1 0 (e2 ) 1 1 (e2 )
15 0 0 1 1 1 1 0 0 0 1 1 1 1 0 0
16 0 0 0 1 1 1 0 0 0 0 1 1 1 0 0
17 1 0 0 0 1 1 1 1 0 0 0 1 1 1 0
18 1 1 0 0 0 1 1 1 1 0 0 0 1 1 0
19 0 1 1 0 0 0 0 1 (e3 ) 1 1 0 0 0 0 1 (e)
20 1 0 1 1 0 0 1 1 1 (e3 ) 1 1 0 0 1 0
21 1 1 0 1 1 0 1 0 (e4 ) 1 1 (e3 ) 1 1 0 1 1 (e)
22 1 1 1 0 1 1 1 1 0 (e4 ) 1 1 (e3 ) 1 1 0 1 (e)
23 0 1 1 1 0 1 0 0 1 0 (e4 ) 1 1 (e3 ) 1 0 0
24 1 0 1 1 1 0 1 1 0 1 0 (e4 ) 1 1 (e3 ) 1 0
25 0 1 0 1 1 1 0 0 1 0 1 0 (e4 ) 1 0 0
26 1 0 1 0 1 1 1 1 0 1 0 1 0 (e4 ) 0 1 (e)
27 0 1 0 1 0 1 0 0 1 0 1 0 1 0 0
28 0 0 1 0 1 0 0 0 0 1 0 1 0 0 0
29 0 0 0 1 0 1 0 0 0 0 1 0 1 0 0
30 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0
31 1 0 0 0 0 1 1 1 0 0 0 0 1 1 0
32 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0
Table 2.3: PRBS checker behaviour without errors (left), with errors spaced far apart (right
top), and with errors close together (right bottom)
has the problems of indicating 3 error pulses for each error in the input, and producing fewer
error pulses when the errors in the input are very close to each other. To solve this problem,
a new PRBS checker configuration was developed. For a 27 − 1 PRBS, the new configuration
is shown in Figure 2.10. The analysis of error propagations inside the checker nodes can be
repeated in equation 2.12 for the checker of Figure 2.10, noting that in this case the relationship
F H
INPUT A B C D E
G
I
O
P
J K L M
N
S T
U
ERROR
Figure 2.10: A modified 27 − 1 PRBS checker
D 6 + D 7 = D 0 holds.
SA = D 1 ⊕ E 1
SF = D 6 ⊕ E 6
SG = D 7 ⊕ E 7
SH = SF ⊕ SG = D 6 ⊕ D 7 ⊕ E 6 ⊕ E 7 = D 0 ⊕ E 6 ⊕ E 7
SI = D 1 ⊕ E 7 ⊕ E 8
SN = D 6 ⊕ E 12 ⊕ E 13
SO = D 7 ⊕ E 13 ⊕ E 14
SP = SN ⊕ SO = D 6 ⊕ D 7 ⊕ E 12 ⊕ E 13 ⊕ E 13 ⊕ E 14 = D 0 ⊕ E 12 ⊕ E 14
SQ = D 1 ⊕ E 13 ⊕ E 15
SS = SA ⊕ SI = D 1 ⊕ D 1 ⊕ E 1 ⊕ E 7 ⊕ E 8 = E 1 ⊕ E 7 ⊕ E 8
ST = SA ⊕ SQ = D 1 ⊕ D 1 ⊕ E 1 ⊕ E 13 ⊕ E 15 = E 1 ⊕ E 13 ⊕ E 15
SU = SS ∧ ST = (E 1 ⊕ E 7 ⊕ E 8 ) ∧ (E 1 ⊕ E 13 ⊕ E 15 )
= E1 (2.12)
Thus, in this checker configuration, only one error pulse appears for each error in the input
signal. This enables more precise counting of errors.
The PRBS checker of Figure 2.10 was developed as a possible replacement for the checker
of Figure 2.9 to be able to provide more precise bit error rate (BER) counts. The new checker
(Figure 2.10) achieves this goal by ensuring that only one error pulse is produced for each
2.5 PRBS Generator Architecture Comparison for High-Speed Operation 21
error in the input. Another possibility for precise BER counts (which was not explored in this
thesis) is to modify the basic PRBS checker circuit to ensure that always three error pulses are
produced for each error in the input. In this case a precise BER count will be obtained after
division by 3. In addition, the rate of occurrence of errors that are spaced close enough to each
other to affect BER counts has to be significant to justify the extra circuitry in the new PRBS
checker.
2.5. PRBS Generator Architecture Comparison for High-Speed Op-

eration
For very high-speed generation of PRBS sequences, it is useful to know which architecture is
optimal for a particular application. The different options that can be compared are parallel
versus series PRBS generator configuration and the level of multiplexing. The level of mul-
tiplexing determines how much slower, relative to the final output, the core generator can be
operated. However, if the multiplexing is too deep, too much power might be spent in the
multiplexer itself.
Table 2.4 presents a comparison of series and parallel generators, in terms of the number
of gates required and the maximum fanout of gates needed to build the generator. Fanout in
the PRBS generator chain determines the maximum speed at which the core generator can be
operated for a given gate topology. The number of gates determines the area of the PRBS
generator. It is also related to the operation speed because greater area implies the some gates
have to drive longer lines, which limits the overall achievable bit rate. The overall power
that the generator will consume is also directly proportional to the number of blocks required
to build it. Note that the count of blocks indicated in Table 2.4 does not include the blocks
required to build the multiplexers needed to increase the core generator bit rate to the final bit
rate.
From Table 2.4, it is apparent that parallel PRBS generators outperform series generators
in all cases where the PRBS is generated below the full data rate and multiplexing is applied.
In practice, for the sequence to be generated at full rate, a more complex and power hungry
design is required for each flip-flop in the chain. On the other hand, when multiplexing is used,
only the last stage of multiplexing needs to operate at the full data rate. This greatly simplifies
the overall design and results in a smaller and more power efficient circuit.
Parallel PRBS generators have several other advantages over series generators in high-
speed applications. First the fanout of the XOR gates and flip-flops is uniform throughout the
structure, making it easier to design each block and to lay them out. Second, re-timing of each
Series Architecture Parallel Architecture

No. of Blocks Max. No. of Blocks Max.
Generator Type Rate DFFs XORs Fanout DFFs XORs Fanout
full 7 1 2 - - -
27 − 1 1/2 9 2 2 7 2 2
x + x6 + 1
7
1/4 11 5 3 7 4 2
1/8 15 11 3 8 8 2
full 15 1 2 - - -
215 − 1 1/2 17 2 2 15 2 2
x + x14 + 1
15
1/4 19 5 3 15 4 2
1/8 23 15 4 15 8 2
1/16 31 38 8 16 16 2
full 23 1 2 - - -
223 − 1 1/2 25 2 2 23 2 2
x + x18 + 1
23
1/4 27 6 2 23 4 2
1/8 31 18 4 23 8 2
1/16 39 52 6 23 16 2
full 31 1 2 - - -
1/2 33 2 2 31 2 2
231 − 1 1/4 41 6 2 31 4 2
x31 + x28 + 1 1/8 39 17 4 31 8 2
1/16 47 48 7 31 16 2
1/32 63 186 7 32 32 2
Table 2.4: Comparison of the circuitry required for series and parallel PRBS generators of
different sizes
combinational logic gate is essential for correct operation above 10 Gb/s because gate delays
are a large fraction of the clock cycle. Conveniently, parallel generators are structured such that
all parallel outputs are automatically re-timed and there is only one XOR gate between each
two flip-flops. On the other hand, series generators require a very large number of XOR gates
to produce appropriately shifted sequences as the multiplexing ratio is increased (Table 2.4)
making them highly impractical. Third, since all outputs of the parallel PRBS generators are
re-timed, the first stage of multiplexing can be simplified. Instead of the usual high-speed
multiplexer that consists of five latches and a selector [17], only one latch and a selector are
needed in this case. This further saves power and area.
2.6 Summary 23
2.6. Summary
This chapter described how pseudo-random bit sequences can be created in several different
ways and what are the benefits and drawbacks of each approach. Series PRBS generators
have a very simple structure, but are cumbersome to use when multiplexing to higher speeds.
Parallel PRBS generators have more gates in their structure, but generate sequences suitable
for direct multiplexing. These properties make parallel generators more suitable for high-speed
applications. This chapter also described how pseudo-random bit sequences can be checked
for errors.
27-1 PRBS Generator and
3 Checker Chip Design
In this chapter, the procedure used to design a 27 − 1 PRBS generator and checker system will
be outlined. The chip described in this chapter was designed and fabricated in STMicroelec-
tronics’ 0.13 µm SiGe BiCMOS process technology. First, system-level considerations will
be presented in section 3.1. Then, section 3.2 will present the design details of each building
block that was used in the system. Simulation results of the system performance before and
after layout will be shown in section 3.3.
3.1. System Description and High-Level Simulations
A 27 − 1 PRBS generator and checker system was designed to be used as an integrated self-test
block. The primary design goals of built-in self-test (BIST) blocks are low power consumption
and small area. Thus, these were also the requirements of this system. The third requirement
was to make the PRBS generator compatible with testing 80-Gb/s systems. Therefore, the
generator must be able to generate 4 pseudo-random sequences at 20-Gb/s or higher data rates.
A block diagram of the entire system that was implemented on chip is shown in Figure 3.1.
The only high-speed input to the system is a 12-GHz clock signal. The clock signal is
distributed by a tree of clock buffers to each of the chip components. The pseudo-random
bit sequence is generated at half-rate (12 Gb/s) by the 27 − 1 PRBS generator block. The
generator block has 8 outputs and each of them is delayed appropriately for multiplexing. The
eight 12-Gb/s PRBS streams are fed directly to an 8-to-4 multiplexer (MUX), which produces
four PRBS streams at 24 Gb/s each. One of the four streams is provided off-chip for testing.
The other three outputs are terminated on-chip. Even though the four 24-Gb/s signals are
generated such that they are ready to be multiplexed further to 80 Gb/s by a 4-to-1 MUX, the
4-to-1 MUX is not integrated on this chip.
The other part of the system is used for verification. In a real testing system, the checker
25
26 27 -1 PRBS Generator and Checker Chip Design
This Work
SWITCH PRBS
Checker ERROR
SEL
+ COUNT
Counter
PRBS 8 to 4 4 to 1 Output
Gen. MUX MUX Buffer
50 Ω
CLK IN Clock Buffer
Buffers
PRBS
OUT
Figure 3.1: Top-level schematic of the designed chip
takes its input signal from the output of the device that is being tested. Then, this signal is
de-multiplexed down to the data rate at which the checker operates, and the checker tests the
signal for errors. However, in this case, both the PRBS generator and the PRBS checker are on
the same chip. To be able to test the checker, its input is taken directly from the generator.
The input signal to the PRBS checker is at 12 Gb/s. Before entering the checker, this
signal passes through a selector switch (SEL). The selector can be controlled manually to
switch between two PRBS sequences with different phase. At the switching point, errors are
introduced into the signal. This signal with manually controlled errors is fed into the checker
to enable its verification. The PRBS checker operates at 12 Gb/s and outputs one at-speed
pulse for each detected error. These error pulses are then counted by a 5-bit counter. The five
error-count bits are provided as outputs for off-chip error monitoring.
3.1.1. MATLAB Simulations
The first decision in the design of the system was to choose between a series or a parallel PRBS
generator topology. The parallel topology inherently generates 8 PRBS outputs that are ready
for multiplexing. However, for a series topology, the optimal way of generating 8 appropriately
delayed signals has to be found. Therefore, the algorithm to determine the connections needed
for delayed PRBS signals (Figure 2.4) was implemented in MATLAB. The code is attached
in Appendix A.1. This algorithm was run iteratively to find the combination that requires
the minimum amount of XOR gates. Note that, because the zero-delay reference is arbitrary,
several options exist for selecting the delays even though the delay spacing is fixed.
3.1 System Description and High-Level Simulations 27
3.1.2. Simulink Functional Simulations
Simulink was used for functional simulation of the PRBS generator, 8-to-4 multiplexer, and
checker blocks. It was also employed to verify functional correctness of the parallel PRBS
generator by comparing its output to that of the simplest series generator. This was possible to
carry out by visual inspection because the sequence length is only 127 bits. These simulations
were also used to estimate the power and area (in terms of number of gates) requirements of
the parallel and series generator. The final versions of both generators are shown in Figure 3.2.
In the case of the series PRBS generator, re-timing flip-flops are required after the combi-
national logic to align all signals with the clock, before multiplexing. The second problem with
combinational logic is that it requires the fanouts of the shift register flip-flops to be different,
and therefore have different delays. Even very small timing variations can significantly affect
operation at high speeds. The total number of gates needed in this case is 11 XOR gates and
15 D-flip-flops, resulting in an estimated power of 242 mW for 12-Gb/s operation.
The parallel generator avoids the problems mentioned above thanks to its regular structure.
The outputs are automatically re-timed and delayed appropriately. The fanout for all XOR
gates and flip-flops is uniform, thus delays are equalized. The total number of gates needed
in this case is 8 XOR gates and 8 D-flip-flops, consuming approximately 140 mW at 12 Gb/s.
The parallel PRBS generator was chosen to be implemented for this system because it saves
area and 42 % power compared to the series generator.
0° 0°
D Q D Q
D Q
180° 45°
D Q D Q
D Q
315° 90°
D Q D Q
D Q
90° 135°
D Q D Q
D Q
135° 180°
D Q D Q
D Q
225° 225°
D Q D Q
D Q
45° 270°
D Q D Q
D Q
270° 315°
D Q D Q
CLOCK IN CLOCK IN
(a) Series 27 − 1 PRBS generator with 8 (b) Parallel 27 − 1 PRBS

outputs generator with 8 outputs
Figure 3.2: Parallel and series implementations of 27 − 1 PRBS generators with 8 outputs
The schematic of the PRBS checker is shown in Figure 2.10. Its operation was also verified
with Simulink.
The useful attributes of Simulink are that the simulations can be performed quickly and
Simulink’s schematics are useful for debugging Cadence schematics. The Simulink schematics
are given in Appendix A.2.
3.1.3. Verilog Timing Simulations
After functional verification, Verilog was used for preliminary timing simulations of the gen-
erator, the MUX, the PRBS checker, and the clock tree. In these simulations the maximum
allowed delay was found for each building block, depending on its fanout. A clock frequency
of 15 GHz was used to have margin over the target 12-GHz clock frequency. The maximum
allowed block delays, summarized in Table 3.1, were obtained from Verilog simulations and
used as design targets in transistor-level design of the building blocks. The Verilog code is
located in Appendix A.3.
3.2. Design of Building Blocks
Once high-level system simulations were completed, each individual block was designed at
the transistor level and simulated using Spectre. The chip was designed in the STM’s 0.13 µm
SiGe BiCMOS technology with heterojunction bipolar transistor (HBT) fT of 160 GHz [18].
This process technology has 6 metal layers. The design procedure and schematics of each
block will be given in this section.
3.2.1. 1-mA and 2-mA Latches
Three types of latches were designed for different parts of the system. All three employ the
same basic BiCMOS CML topology but have different component values. This is done to
Fanout = 1 Fanout = 2 Fanout = 3

Maximum XOR delay 16 psec 19 psec 22 psec
Maximum Latch delay 16 psec 19 psec 22 psec
Table 3.1: Maximum allowed block delays for system operation with 15-GHz clock
3.2 Design of Building Blocks 29
customize each latch to its load conditions and thus save power where the load (fanout) is
small. Schematics for the three latches are shown in Figure 3.3.
To bias the latches for fastest switching time, the tail current Itail and the transistor sizes
are related by
Itail Itaill
W = = (3.1)
Jpeak−fT ,M OS 0.3mA/µm
for MOSFETs in any technology, and by
Itail Itail
le × we = = (3.2)
Jpeak−fT ,HBT 6mA/µm2
for HBTs in the 0.13 µm SiGe BiCMOS technology [5].

In this work, it was important to achieve the lowest power consumption possible. Therefore,
latches with low fanout, like master latches of D-flip-flops, were designed with the lowest tail
current possible that still obeys equations 3.1 and 3.2. The lowest tail current is set by the
minimum allowed HBT size in the technology, which is le × we = 0.64µm × 0.2µm. Thus, the
tail current was chosen to be 1 mA. Next, the MOSFETs were sized according to equation 3.1.
Simulations indicated that the 1-mA latches worked up to 15 Gb/s, which met the design goal,
so it was not necessary to increase the current for achieving the desired bit rate.
The output swing ∆V = RL × Itail of the latches was changed depending on the next stage
coming after the latch. If the latch was used to drive the HBT pair of a BiCMOS block, ∆V
was set to 300mV , which is adequate to fully switch an HBT differential pair (Figure 3.3(a)).
If the latch was used to drive the MOS pair through a stage of emitter followers, then ∆V was
set to 500mV , which is required to switch a MOSFET differential pair in 0.13 µm technology
(Figure 3.3(b)).
In places where the fanout of a latch was larger than 2, the latch tail current was increased
2.5V 2.5V 2.5V
RL=300Ω RL=500Ω RL=250Ω OUT

OUT OUT
IN IN IN
e=0.64µm e=0.64µm e=1µm
W 2×2µm W 2×2µm W 2×4µm

= = =
CLK L 0.13µm CLK L 0.13µm CLK L 0.13µm
W 2×2µm W 2×2µm W 2×4µm

= = =
VTAIL L 0.15µm VTAIL L 0.15µm VTAIL L 0.15µm
ITAIL=1mA ITAIL=1mA ITAIL=2mA
(a) 1-mA, 300-mV swing latch (b) 1-mA, 500-mV swing latch (c) 2-mA, 500-mV swing latch
Figure 3.3: Detailed schematics of the designed latches

to 2 mA (Figure 3.3(c)). Transistor sizes were increased accordingly. This configuration was
used in the slave latches of D-flip-flops that had to drive two XOR gates, a 2-to-1 MUX, and
the associated interconnect.
All latches and gates in this chip use the BiCMOS CML logic topology, but differ from
previous designs [6]. In the current design the feedback source followers are removed to save
power, and peaking inductors are removed to save area. These changes are permitted because
the parallel PRBS architecture allows the shift register to operate at lower bit rates than in [5].
3.2.2. 12-Gb/s DFF
D-flop-flops are used in the core part of the PRBS generator, the PRBS checker and the error
counter. A schematic of the DFF is illustrated in Figure 3.4, showing the master and slave
latches and the clock emitter followers.
The DFF topology is also an improved version of the one presented in [6]. The clock source
followers are replaced by emitter followers which are able to drive a larger capacitance per unit
current. Notice that the DFF contains only one set of emitter followers for both latches. This is
done to both reduce the load on the clock distribution buffers and to save power compared to a
DFF configuration where each latch has its own emitter followers.
The D-flip-flops used in the PRBS generator have the 1-mA latch of Figure 3.3(a) as the
master and the 2-mA latch of Figure 3.3(c) as the slave. The slave latch of each generator DFF
needs a larger tail current because it has to drive two XOR gates and a 2-to-1 MUX. Together
with the clock emitter-followers, this results in a current of 5 mA from 2.5 V, thus a power
dissipation of only 12.5 mW for a DFF that operates at 12 Gb/s.
2.5V 2.5V 2.5V
RL RL RL RL
OUT
CLK
IN
VTAIL
1mA 1mA 1mA 2mA
Figure 3.4: D-flip-flop schematic

3.2 Design of Building Blocks 31
Unlike the flip-flops of the generator chain, in the PRBS checker and counter most flip-
flops have a fanout of 1. In this case, the latches of Figure 3.3(a) and 3.3(b) are used as the
master and slave, respectively, for a total 12-Gb/s DFF power of 10 mW.
3.2.3. Selector, XOR, and AND Gates
In addition to latches, the other digital blocks that are used in this system include selectors,
XOR gates, and AND gates. Their schematics are shown in Figure 3.5. Selectors (Fig-
ure 3.5(a)) are employed in the final stage of each 2-to-1 MUX. To achieve a 24 Gb/s signal
at the output, the tail current was chosen to be 2 mA, with a single-ended swing of 250 mV.
Transistors were sized by following the same procedure as for the latch.
XOR gates (Figure 3.5(b)) and AND gates (Figure 3.5(c)) are designed with 1-mA tail
currents because their fanout is 1 in most cases. However, they differ from the latch and the
selector topology by having emitter-followers at one of the inputs. These emitter-followers
are necessary to step-down the DC voltage level from the top HBT pair to the bottom MOS
transistor pair; they cannot be shared between gates.
3.2.4. 24-Gb/s 2-to-1 MUX
The 2-to-1 MUX block is repeated four times to build the 8-to-4 MUX that outputs four 24-Gb/s
PRBS streams. The 2-to-1 MUX schematic is shown Figure 3.6. Note that only one latch
and one selector are used to build the MUX, unlike the more common 5 latches and selector
configuration [17]. This is acceptable because the signals going from the PRBS generator into
the MUX are already re-timed.
2.5V 2.5V 2.5V 2.5V 2.5V
RL=125Ω RL=500Ω RL=500Ω

OUT OUT OUT
IN1 IN1
e=0.64µm e=0.64µm
e=0.64µm e=0.64µm
e=1µm
VB
IN1 IN0 IN0 IN0
W 4×2µm W 2×2µm W 2×2µm

= = =
L 0.13µm L 0.13µm L 0.13µm
SEL
W 2×2µm W 2×2µm
W 2×4µm = =
= VTAIL L 0.15µm VTAIL L 0.15µm
VTAIL L 0.15µm
ITAIL=2mA 0.5mA 0.5mA ITAIL=1mA 0.5mA 0.5mA ITAIL=1mA
(a) Selector (b) XOR gate (c) AND gate
Figure 3.5: Detailed schematics of other blocks

2.5V 2.5V 2.5V
RL RL RL RL
OUT
CLK
IN1 IN0
VTAIL
1mA 1mA 1mA 2mA
Figure 3.6: 2-to-1 MUX schematic
Since the non-latched input to the selector comes from the generator DFFs, which have
500 mV swing, the latched input must also have 500 mV swing. Therefore, a 1-mA latch with
500 mV swing (Figure 3.3(b)) is used in the 2-to-1 MUX in front of the selector. The clock
emitter followers are shared between the latch and the selector of the 2-to-1 MUX, as in the
DFF.
3.2.5. Clock, Data, and Output Buffers
One of the most important parts of the PRBS generator and checker system is the clock tree.
It is a tree of CML buffers designed to deliver the 12-GHz clock signal synchronously to all
latches in the system. The schematic of one clock buffer is shown in Figure 3.7(a). It consists
of an HBT differential pair preceded by emitter followers. The swing is set to 500 mV, to be
able to switch the MOS transistors at the clock inputs of the latches. The tail current in the
differential pair is set to 2.5 mA for adequate bandwidth.
To reduce the number of clock buffers in the system, and thus to save power, the fanout
of each buffer is set to 4. This high fanout is possible because in each flip flop, the emitter
followers on the clock path are shared among the two latches. They also serve as the final stage
of clock buffering.
It is very important to ensure that the paths traveled by the clock signal have identical
delays. This ensures that the clock arrives to all flip-flops with the same phase. Thus, a lot
of attention was payed in the layout to ensure equal-length connections between clock buffers
and from the clock buffers to the flip-flops.
In addition to clock buffers, two other types of buffers are used in the system. These are
data buffers and 50-Ω output buffers, shown in Figure 3.7(b) and Figure 3.7(c) respectively.
3.3 System-Level Design and Simulations 33
2.5V 2.5V 2.5V 2.5V 2.5V 2.5V 2.5V 2.5V 2.5V 2.5V 2.5V 2.5V
RL=220Ω RL=550Ω RL=50Ω
OUT OUT OUT
e=4µm e=0.64µm e=6µm
IN IN IN IN IN IN
e=2µm e=0.64µm e=11µm
VTAIL VTAIL VTAIL
I=1.25mA ITAIL=2.5mA I=1.25mA I=0.5mA ITAIL=1mA I=0.5mA I=3mA ITAIL=12mA I=3mA
(a) Clock buffer schematic (b) Data buffer schematic (c) 50Ω output buffer schematic
Figure 3.7: Detailed schematics of buffers
Data buffers are employed as intermediate buffers to enhance the signal, or before driving a
large load. 50-Ω output buffers are used only on the outputs, to drive external 50-Ω loads. The
50-Ω load and the 300-mV swing requirement restrict the tail current in these buffers to be
12 mA.
3.3. System-Level Design and Simulations
A detailed schematic of the entire system that was implemented on-chip is illustrated in Fig-
ure 3.8. The power consumption of each sub-block is summarized in Table 3.2.
Throughout the design, every building block, such as a latch or an XOR gate, was ex-
tensively simulated at the transistor level using Spectre to ensure that it behaves as desired.
Simulated eye diagrams of a latch and a D-flip-flop with a 15-GHz clock input are shown in
Figure 3.9.
Power Consumption
27 − 1 PRBS Generator 145 mW
8-to-4 MUX 50 mW
PRBS Generator Clock Tree 75 mW
27 − 1 PRBS Error Checker 200 mW
5-Bit Error Counter 80 mW
PRBS Checker Clock Tree 75 mW
Output Buffers 300 mW
Total 925 mW
Table 3.2: Power consumption of the chip sub-blocks

SWITCH
27-1 PRBS Error Checker RESET
Error Counter
D Q
D
DATA
EN
SEL BUFFER D Q D Q D Q D Q D Q D Q D Q
BIT0
Q
DATA 50 Ω
BUFFER Buffer
D
EN
BIT1
Q
D Q D Q D Q D Q D Q D Q D Q DATA 50 Ω
BUFFER Buffer
D
CLOCK
EN
BUFFER
BIT2
Q
DATA 50 Ω
BUFFER Buffer
D
CLOCK CLOCK CLOCK CLOCK
EN
BUFFER BUFFER BUFFER BUFFER
BIT3
Q
CLOCK DATA 50 Ω
BUFFER BUFFER Buffer
D
EN
OV
Q
DATA 50 Ω
BUFFER Buffer
Parallel 27-1 PRBS Generator
D Q
2:1
D Q MUX
D Q
2:1
SEL MUX
D Q
D Q
2:1
MUX 480 µm
D Q 50 Ω line PRBS
50 Ω
Buffer OUT
D Q 2:1
MUX
D Q
8-to-4 MUX
CLOCK
BUFFER
225 µm line
CLOCK CLOCK
BUFFER BUFFER
CLOCK CLOCK
BUFFER BUFFER
200 µm line 280 µm line
CLOCK
BUFFER
CLOCK
IN
Figure 3.8: Detailed system schematic

3.3 System-Level Design and Simulations 35
Figure 3.9: Latch (top) and DFF (bottom) simulation with 15-GHz clock
Figure 3.10: System simulation with 15-GHz clock

After the operation of building blocks was verified, the system sub-blocks, like the gener-
ator and the checker, were simulated by themselves. Finally, the sub-blocks were assembled,
and the entire system was simulated, including a model of the test setup. A simulation of the
system with input clock frequency of 15 GHz is shown in Figure 3.10. Figure 3.10 shows
(from bottom to top) the 15-Gb/s error-free 27 − 1 PRBS signal coming from the generator; the
30-Gb/s output signal from the 8-to-4 MUX; the distorted 15-Gb/s input to the PRBS checker;
the errors detected by the checker; and the 5-bit count of the error counter.
During the layout design process, a lot of attention was paid to the symmetry of each
building block, and to the interconnect between blocks to minimize the impact of transistor
variations and mismatches. All blocks used unit-size transistors with identical orientation to
form all other transistors sizes. Since all signals are differential, the path lengths of the two
wires carrying the signal were equalized, especially in the clock distribution tree. 1-pF MIM
capacitors were added in the current mirror for decoupling. Additional decoupling capacitors
were connected between ground and VDD for filtering of power supply noise. Metal 1 and
metal 2 were used for ground and VDD, respectively. All high-speed signals were routed in
metals 5 and 6 to reduce their capacitance to ground. Control signals were routed using metal 3.
The layouts of the clock buffer, the XOR gate, and the flip-flop and given in Figure 3.11.
After the layout was completed, simulations with extracted parasitics were carried out.
The extracted parasitics included series resistances, shunt, and coupling capacitances of all
nodes. A simulation of the system after extraction is shown in Figure 3.12. In this figure, the
111111111111
11111111111111111111111111111111111
11111111 1111111111111
1111111 111111111111
11111111 111111111111
111111 111111111111
111111111111 111111 1111111111111
11111111111111111111111111111111111
11111111 11111111 1111111 111111111111
111111 1111111111111
111111 111111111111
11111111111111111111111111111111111
11111111 11111111 11111111
111111
111111111111
11111111111111111111111111111111111
11111111 1111111 1111111
11111
111111
1111111111111
11111111111111111111111111111111111
11111111 11111111
111111111111 11111111
111111111111 1111111
1111111111111
11111111111111111111111111111111111
11111111 11111111
111111
111111111111
111111111111 111111
11111
111111 1111111111111
1111111
111111
11111
1111111111111
11111111111111111111111111111111111
11111111 1111111
111111
111111111111
11111111
111111
111111111111
11111111 1111111
111111
111111111111
1111111
111111
111111111111
11111111
111111
111111111111 1111111
11111
111111
1111111111111
11111111111111111111111111111111111
11111111
11111111111111111111111111111111111
11111111
11111111
111111
111111111111 111111
1111111
1111111111111
11111111111111111111111111111111111
11111111 11111111
111111
111111111111
111111 1111111111111
111111111111 111111
11111
11111111111111111111111111111111111
11111111 1111111
111111 1111111111111
111111111111
11111111111111111111111111111111111
11111111 111111
11111
1111111
111111
111111111111
11111111
111111
111111111111
11111111 1111111
111111111111
111111
111111 111111111111
11111111 11111
111111
111111 1111111111111
1111111
11111
111111
11111111111111111111111111111111111
11111111 111111111111
11111111 1111111111111
1111111
111111
111111111111
111111111111
111111
1111111
111111 1111111111111
11111111111111111111111111111111111
11111111 11111111
1111111
11111111111111111111111111111111111
11111111 1111111111111
11111111
111111
111111 111111111111
11111111
111111
111111111111
11111111
111111 1111111111111
111111111111 111111
11111
1111111111111
11111111111111111111111111111111111
11111111 1111111 111111
111111111111
11111111 1111111
111111
111111111111
111111111111
11111111
111111
111111111111 1111111
11111
111111
1111111111111
11111111111111111111111111111111111
11111111 11111111
111111
111111111111 111111
1111111
1111111111111
11111111111111111111111111111111111
11111111 11111111
111111
111111111111
111111111111
11111111111111111111111111111111111
11111111 1111111 111111111111
11111111 1111111
111111 111111111111
111111 1111111111111
11111
111111
11111111111111111111111111111111111
11111111 11111111 1111111 111111 1111111111111
111111111111 111111 111111111111
1111111
11111111111111111111111111111111111
11111111 11111111 111111
11111111
111111111111
11111111 1111111111111
1111111 111111111111
11111111111111111111111111111111111 11111111 111111111111
11111111 111111111111
11111111 1111111111111
1111111 111111111111
11111111111111111111111111111111111 11111111 1111111111111
11111111 111111111111
11111111111111111111111111111111111
11111111
11111111111111111111111111111111111
11111111 1111111 11111111 11111111111111111111111111111111111
11111111 11111111 1111111
11111111111111111111111111111111111
11111111 11111111 11111111111111111111111111111111111
11111111 11111111
1111111 11111111111111111111111111111111111
11111111 11111111 11111111
11111111
11111111111111111111111111111111111
11111111 1111111
11111111111111111111111111111111111
11111111 1111111 11111111
11111111 11111111111111111111111111111111111
11111111 111111111111
11111111 1111111111111
1111111 111111111111
11111111111111111111111111111111111
11111111 1111111111111
11111111 111111111111
11111111
11111111111111111111111111111111111
111111111111
11111111 1111111111111
1111111 111111111111
11111111 11111111111111111111111111111111111
111111111111
11111111 11111111 1111111
11111111111111111111111111111111111
111111111111
11111111
1111111 11111111111111111111111111111111111
11111111 11111111 11111111
11111111111111111111111111111111111
111111111111
11111111
111111 1111111111111
11111111 1111111
111111
1111111
11111 111111111111
11111111111111111111111111111111111
111111111111 1111111111111
11111111
11111111
111111
111111111111 111111 111111111111
11111111
111111 1111111111111
11111111
1111111
11111
111111
11111111111111111111111111111111111
111111111111
11111111
1111111 111111111111 1111111111111
1111111
11111111111111111111111111111111111
111111111111
11111111
111111 1111111111111
11111111
1111111
111111 111111111111
11111111
11111111111111111111111111111111111
111111111111
11111111 1111111111111
11111111
1111111 111111
111111111111
11111111
11111111
111111
11111111111111111111111111111111111
111111111111 111111 111111
11111111111111111111111111111111111
111111111111
11111111
1111111
111111 111111111111
11111111 11111
111111
111111 1111111111111
1111111 111111
11111111111111111111111111111111111
111111111111
11111111 111111
111111 1111111111111
11111111
1111111 111111
111111 111111111111
11111111
111111 1111111111111
111111
1111111
11111 11111111
111111
111111111111 11111111111111111111111111111111111
111111111111
11111111
1111111
111111 111111111111
11111111
111111 1111111111111
1111111
11111
111111 11111111111111111111111111111111111
111111111111
11111111
111111 1111111111111
11111111
1111111
111111 111111111111
11111111
11111111
11111111111111111111111111111111111
111111111111 111111
111111 1111111111111
111111
1111111
11111 11111111
111111
111111111111 11111
111111
11111111111111111111111111111111111
111111111111
11111111
1111111 111111111111
11111111 1111111111111
1111111 11111111111111111111111111111111111
111111111111
11111111 1111111111111
11111111
1111111 111111
111111111111
11111111
111111
11111
11111111111111111111111111111111111
111111111111
11111111 1111111111111
1111111 111111
111111111111
11111111 111111 111111
11111111111111111111111111111111111
111111111111
11111111
1111111 111111111111
11111111 11111
111111
1111111111111
1111111 111111
11111111111111111111111111111111111
111111111111
11111111 111111
1111111111111
11111111
1111111 111111
111111111111
11111111
11111111
111111 111111
1111111
11111
11111111111111111111111111111111111
111111111111 1111111111111 11111111
111111 11
111111111111 11
11111111
11111111111111111111111111111111111
111111111111
11111111
1111111 1111111111111111
111111111111
11111111 111111111
1111111111111
1111111 11111111111111
11
11111111
11111111111111111111111111111111111
11111111 111111111111111
111111111111111 11111111
1111111111111
11111111
1111111 111111111111
11111111
11
11111111111111111111111111111111111
111111111111
11111111
11111111 111111111111111
1111111111111
1111111 11
111111111
1111111111111111111 1111111
11111111111111111111
111111111
111111111111
11111111
111 11111111
11111111 11 11
111111111111111111111111111111111
11
11111111 1111111111111111
1111111111111
1111
11111111 111111111111111
11111111 1111111111
1111111111 111111111111111
1111111
111111111
11111111 1111111
111111111
11111111 11111111 11 11
11111111 1111111111111111111111111 1111
11111111 111111111111111 11111111 11
111111111111111 11111111
1111111111111111111
11111111 111111111
11111111 11111111 11 11
1111111111111111111 111111111
11111111 11111111
11111111111111111 11 11
111111111111111111111111111111111
1111111111111
11111111 111111111111111 1111111111
11111111111111111
11111111 11111111 11111111 11111111 1111111111111111 11111111 11111111
11111111 111111111
11111111
1111111 111111111
11111111
111 1111111 11111111 11111111 1111111111111111
111111
1111111 111111
1111111111111
111111111111111111111111
111111111
1111111111111
111111
111
1111111 111111111 11111111 111111111111111
111111
11111111111
111111111
111111 111111111 11111111111111111 111111
1111
11 111
11111111
11111111111111111
111111
1111 11111111
11111 11111111
111111111111111
1111111 1111111
11111111
111111111
11111111111111111
1111111 11111111
111111111
111 11111111
1111111 11111111 11111111 1111111
1111111111111
111111 1111111 111111111 11111111 11111111
111111111111111
1111111 1111111 11111111
11111111
11111111111 11111111
1111111
1111111111111111111
11111111
11111111
111111111
111 1111111
11111111
111111111 11111111111 11
11111111 1111111 1111111 1111111111111
11111111
11111111111 11111111
111111111 11111111 11111111
111111111111 11
11111111111
1111111111111111111
11111111111111111
11111111
111111111 11111111
11111111111
11 11 11111111111111
11111111 1111111 1111111 111111111 11
1111111111111
11111111111 1111111111111111
111111111
11111111111111111 11111111 11111111
111111111111 11
11111111 11 11
11111111 1111111111111
11111111 11111111
1 111111111111
11111111111
11 1111111111
1111111111 11111111
111111111 111111111
11111111111
11111111 11 11
11111111 1111111
111111
1 1111111
11 11111111
111111111 11
1111111111111
11111111
11111111111 11111111
111111111
1111111
11 11111111
1111111 11111111
111111111111 11
11111111111
1111111111 11111111
111111111
1111111 1111111111
11111111
1111111
1 11111111111
11111111 11 11 11111111
11111111 11111111
1 111111111
11111111
11 11111111111
11111111111
111111111
11111111
11111111
11111111111
11 111111111
11111111
11 1
11111111
1111111111111
111 11111111
11
1 1
1111
11111111
1111
11111111
11111111
1 11
11111111111
111111111
11111111
11111111
111111111
1111111111111
1111111111111
11111111
1
11
11111
11 11 11 1
11111111
11
1 1111111 11111111
11 11
11111111111111111111111
11111111111
11 11
1111 111111111
1111 11111111
11
1 11
111 1111
11111
1 11111111111
1111111111
11111111
111111111
11111111
1 1111111
11111111111
111111111
11111111
11
111111111
11111111
11
1
11 1
111
111111111
11111111
1
111111111
1111111111
1 1111111
1 1 11111111
11111111111111111
11111 11111111
111111111111
11111111
1 11
111111111111
111
11
11
11111
1
11111111111
11 111111111111111111111111
11111
111
1111111111111
111 11
1 11111
1111 1 11 11
11111
1 11111111111
111
1111111111111
111 1 11111 11111
11 11 11 111111111111111111
11 1 111111
1111111
1 11111 1111
11111111111111111111111
111 11111
1111 11111
1111
1111111
11 11 1111 11111
1111111 1111111111111
11111
1 11
11 11111
111 11111111
11111
11111
111111 111111
1111111
11 111111
1111111111111
1111111
111111111111111111
111 111111111
1111111
11111 111 11
11111
1 11
11 11111
11111111111
11 1111111111
111111111111111111
11
111
11111 11 111111
1111
11111
111111 11 11
111
11111 11111111111
111 111 11111
11111
111111 11
11 11 111111111111111111
11
1
11 11 111
1111 11 11
111111111111111111 111
111 11
1 11
111 1 11111111
1 111
111
1111
11111111111 11
1
11 111
11
1111
1 11 11
11111 1 11
1111 1
111111111111
11111111111
11 111111
11
11111
11111 11
1 11
11111
1 1 1 1 11
11111
1111 1111111111111
111
1 11111
1 11111
11 1 11111
11
1111
1 111111
1 11
111
11 11 11111
11 1111111111111111111111 111
1
11111 11
1 11111
11
111
11
1111 11111
1 1111111111111
11111
11111
111111
111111
111
111
1
1111
1 11
1111111
11
11111
11
1111 111111111
11
11111 1111111
11
1111 11111
111
111111111111 11
11111111111
11 11 11
111111
11111111
11111 11
1 11
11111 111
11111
11 111111
11
111
1111
11111
111111
11111 1111111111111
1 11111
11
11111
111
11111
111111 11
11 11 11 111 1
1 1111
11 11111
1111 11111
11
111111
1111
111 111
111 11111
111
1111111111111111111111
11
1 11111
11 1
11111111 11 1111
1111111
111
1 11
111
11111
111 11111
11
11111
1 11
1111
11111
1111
11111111111
11 1
1111111111111
1111111
1 11111
1111111 11111111
11111
111
11
1111111111
111111
1 111
1111 111
11111111
111
11 11
1111111
11
111111
1111 111111111
1111111 111111
111111111111111
1111 111
1111111
1111 11
111111
11111
111 11
11111111
11111111111
11 111
11
11111111
11 11111 1111
11
1111111
11 11111
1 1 1 11111111
111 1111111111111
1 1
11 11
11111 11 1111111111111111111111
11
1 11111
11
1111 11111
111111111111
1
111
1
11111111 111
111 11111 111
11111
1111111111111
111
1 11
1
11111111 1 111 1111
111
11
1
11111111111
1111111111111
11111111111111111
111111
1
111
111
1
11
11111111111
11
111111111111
11
1111111
11
111
11111111 111 1111
11
111111 1
1111111 1111
1111111111111
111
11 1 11
111111
1111111 1111 1
111111111111
11111
1
111111111111 11
11111111
11111111111
11 1111111111
11111 11
11111111
11111
1
111 1 11111
111
1 11111111
1111111111111
1 11111
111
1 11 11
11111 11 11111111
11 11111111
1 11
1111 11111111
111111111111
111
111
1 111
11111111 111111111
11111111
111111111
11 11111111
1111111111111
111
1 1111
1
11111111 1111111111111
1 11
1111111111111111
11111111
1111111111111111111
11111111
1 11111111
1
111
111111111111
11
1111 111111111
11
111
11111111 111111111
1111111111
1 11111111
1111111111111
111
11 11111
1111111
111111111111 11
111111111
1 11
11111111111
11 11111111
111111111
11 11111111
11111111
11
1 1 1111111111111111
11111111111
111111111
111111111111111111
111 11111111
11 11111111
111 1111111111
1 111
1 11111 11
11 11 11 11111111111111111
11
111
11111
11 111111111111
1111
11
11111111
11111111 11 11111
111111111 11
111111
11111111111 11111111
1111111
11111111111 1111111111111
11111111
11 1111
11111
11111111
1111111
11111111
11111111111
11 111111111
11111111111
1
111111 11111111
11111111111 1
11111
11 11111111
11 1111
1 11111111111
111111111
1 11111111
111
1 1 11 11
11111
11111
111
1111111111111111
11
11111111 111
11 1
11
1111
11
11 1
11111111 1111 11
11111
11
1 1
11
1
11
1
111111
1111111
1
111111111
11 1 11
1111111111111111111
11111111111 11111
111
1 11
111
1
1111111 1
11
111111111111 11
1111111111
111111111111111
1111111111
11111111
1111111111111 1111111111
111111111111111111111 11111
11111111111
11111111111
11111111111
111111111
11111111111
11111111111111
11111111111
111111
111 11
111
1
11111111111111111
1
11
1
11
11111
11 11111111
11
1111
1 111
11
1 11111111111
1
111 1111 11 11 111
11
11111
11111 11111111111
1111
11111
1
1111111111111111
1111111111111 111
11
1111
1111
1111
11 1111111111111
11 11
11111
11
1 111
11
11
11
11
111111
1111111
1 11111
11
11
1 111111
11
1111
1111111
11
11 11
11
1
11 1111
1 11
1111
1
1111111
1
1111111 1
111111111111 11
11 111111
1111111111
11111111
1111111111111
11111111111111
111111111111111
1111111111111
1111111111
111111111111111111111
1111111111
111111111111111111111
11111
11111
11111111111
111111111
11111111111
11111111111111
111111111111111
11111111111
11111111111111 111
1
11 11
111111111111111
111
11
11111 1111
11
11111 11111
1111
111 1111
11
111 1111
1111 11
1111
1 11111111111
111
111
1
111 111111 11
111111
11111111111111
111111111111111
1111111111111
111111111111111
11111111111111 1111111111
111111111111111111111
1111111111
11111
11111
111111111111111
11111111111
11111111111111
11111111111111
111111111111111
11111111111 11111111111
111111
111 1
111
11
1
11
11111
11
111 1111
1111
1 11111
1111
11
1 11
1 1111
111
1 1 11
11
1111
1 11111111111
11
11
11
1111 1111 11 111
11
11111
11111
11 11 11
1111
1111
1
11 11 1
1111
11
11
111 1 11111111 11
1 1
11
1
111111
1111111
111
11111
11111111111 1111
11
11
1 11
11
111111
1111
1 1111
11 11
11
1 111
1 11
11
1
1111111
1
1111111 11
111111111111
11111111111111
111111111111111
11111111111111
111111111111111 111111111111111
111111111111111 11111
11111111111111
111111111111111
11111111111
11111111111111
111111111111111
11111111111 11111111111
11 1111
111
11111111111
111111
11111
11
111111111
11111
11
11111
11 1
111111111
1111
11
1111
1 11111
11
1
1111
11 11
111
1 11
1
1111111
1 11111
11
1111
111
1
1111
11
111
1 11
1111
11
1111
1
11111111111
11
111111111
1 11
1111
1 11
1111
11 11111111111
111111111
1111
1 1 111111
1111
111
11111111111
11 11 11111
11111 11111111111
1111
11 11
111
1
111
11
1111
1 1 1111
11111
111
11 1 11111111
11111111111 11
11111111111
111111
1111111
111111 1111
11
11
1 11
11
111111
1111
1 11111
11
1 11
111
11111111111
11111111
1
11 1
11111 11 1111
111111111111 11
11111
11111111111111
111111111111111
11111111111111
111111111111111 1111111111
111111111111111
11111111111111
111111111111111
11111111111
11111111111111
111111111111111
11111111111 11 11 11111111111
1111 11
1111
1111
1
1111
11 1111 1111111111111
111111 11 11
11111111111 1111
11 11111
11
11
11111
1 1111
11
11
1 11
1111111111
11111111111
1 1 11
111111
111111111111
11111111111111
111111111111111
11111111111111
111111111111111 111111111111111
1111111111
11111 11111111111111
111111111111111
11111111111
11111111111111
111111111111111
11111111111 11111111111
111111 11
11111111 1111
11
11
1 11
1111
11
1 11
1 1111
111
1 111111111
1 1
11111111
11
1111111 1111 11 11 11111111111
1111 11
111
11
1111
1111
1 1 1111 11111111 11
11111111
1
11111111111 1111
1 11111
11
11
11111
1 11111
11 11
11
1111111111
11 1
11111111 11
11111
111111111111
11111111111
1111111111111111
11
11111 111111111
1
1111
11
1111
11111
111 1111
11
11111 11111111111
11111111
11
11111 11111111 11 11111111111
1111 1111
111 1 1111 1111111111111
1111 1
11111111111 11
111111
11111111 1111
11
1 11
11
111111
1111
1 11111
11
1 11
11
111 1
11111
11111111 11
1
111111111
1111
111111111111
11111
11111111111111
111111111111111
1111111111111
111111111111
11111111111111
111111111111111
1111111111111 111111111111111 111111111111111
11111111111111
1111111111111
11111111111
111111111111
11
111111111111 1111111111 11111 111111111111111
11111111111111
1111111111111
11111111111
111111111111 11
11
111111111111
11111111111111111
111111111111111111
11 11 111111111111111
111
11 11111
11111111111
111111111111 11
1111111111111111
111111111111111
111 11 111
11 11111111111
111111
1111 1111111
11111
11111111111
11 11111
111 111111111
1 111 11 111 11111111111
111111
11
111 111111
11111 11
111111 11 11111111111
1111 11 1111
111 1111111111111
11111 1111111
1111
111111111111111111111111111111111
111111111
1111 11111
1111 1111
11 11
11 11
1111
11111
1111111
11
111 1 11
111111
11111
111
1111 11111
111111111
111111111111 11
11111111111111111
111
111
111111111111111
11 11 111
11
11 11111111111111111111111
11 11111111111 111
11
11111
1111111
1111111
1111111
1111111111111111
111111111111111
111
111
11
11 111
11
111
11 111 11
11111
11111111111
111111 111
11
1111111
111111111
111 11111111111
11111
11
11111111
11
11111
11111 11 11111111111
1111
111 1111
111111111111111111111111111111
11 111
1111 111
1111111
1111
11111111111 11 111111111
1111
11
1 11111
11
11 1111
11
1 11 1
11
1111
111 11
1 111
111111111111 11
111
11
111
11
11
11111111111
11 11111111111
111
11
11111111
11
111111
11
11111111111111111111111
111
11
111111
111
11 11
111111
11 11111 11
11111
1111111
1111111
1111111
1111
1111111111
1
111111
1111
11
111
11
1
1 111111111111
11
111
1111111111
1
111111
11 11
11
11
111111111111
1 111111111111
111
111
11
111
11 111
11111111111 111
11
111
1111111111111 11
11111 111111111
1 111 11 11111111111
111
11
11111 111
1111 11 11 11111111111
11 1111
1111 1111111
1111
11111111111 11111111
11 111111111
1 111111111
11
1 11
11111
1111
111 11
1 111
111111111111 11
111
11
111
11
11 1 11111111111111111111111
11111111
11111
11
11111111
11111
11
11 1111
11111111111
111 111
1 111
11
11111
1111
11 111
111111
11
11111
11111111111111111111111
11 111
11
1111
111
1111
111111
11
111
111
11111
11111
1111111
11111
1111111
1111111
11
111
1111
11
1
111111
11
1
1111111111
1 11
1
111
1111111111
11
111
1111
11
11111111
1
111 111 11
11111111
11
11111 1111
11111111
11
11111
111111111111
111 111
111
11
111
11
11111111111
11 11 11111111111111111
11 111111111
1
111111 111111111 11111111111
11111111
11111
11 11111111 11 11111111111
11 1111111111111
11
1111 1 11
1111111
111111
1111 111111111
111111
11 1 1111111
111111111
1111111111111111 11
111
111111
11111
111111111111
1111 11 11
1 11111
111
111
111
11
11111111
11111
11
11
11111111111
111 111
11 11111111111
11111111
11111
11
11
111 1111
11 11111111111
11111111
11111
11
11
111 1
111
11111111111111111111111
11
1111
11
111
11
11
111
11
11
111
11
11
111
1111
11
111
1111
111
111111
11
111111
11
11
111
1111
111111
11
111
1
11
1111
111
1
11
11
1111111
11111
1111111
11111
1111111
1
111111
11111111111111
111
1111
11
111111
1111
1
1
11
111
111
1111
11
1
1111
11
111111
1111
111 11
1
111
11111111
11
11111
111
111
11
11111
111
11
111
11
1 111111111111
111 111111111111
111
11
1 111111111111
111 11111111
11
11111111
11
11111
111 1111
11
111
11
111
11
111
11
11 111
111111111111111
11111
11111111
11 11111
111111111
1
111111 111
111111111
1
111111 111
11
111
111111 111
11111111111
11111
11111111
1
11111111111
11
11 11
11 11 11111111111
11 1111
11111111111
1111 111111111
111111
1 11
1111111111111111
1 11
1
111111111111
1111
111
111
11111111
11111
11
11 1111
11111111111
111
11 11111111111
11111111
11111
11
111
11 11111111111
111
11111
111 1111
111
11
11
111
11
11
11
111
1111
111111
11
11
111
1111
111111
11
111
1111
11
11111
1111
1111
1111111
1111111
1111111
1111111
11
1
11
1
111
111
1111
11
111111
111
111
1111
11
111111
11111111
11
11111
111
111
11
111
111
1111
11
1 111111111111
111 111111111111
11
1 111111111111
111 11111111
11
11111 111
11
111
11
1111111111
11 111111
11111111
111111111111 111111111
11111111
1111111111111 1111111111
111111111
1111111111 111111111111
11
11111111 11 11 11111111111
11
1111111111 111111111111
111111111111 11111111111
111111111111 1111 11111111
1 111111111
111111
111111 1 11 111111111
11111111
1111111
11
1 11111 11
1 11
111
11
111
11111111
11111
11
111
11111111
11111
11
11111111111
11 11111111111
11
11111 1 111
11
11
111
11
11
11
111
1111
111111
11
11
111
1111
111111
11 1111
11 1111
1111
11 11111
1111111
1111111
1111111
11
1
11
1
1
111
1111
11
111111
11 1
111
1111
11
111111
11
1
111
11
11111111
11
11111
111
11 111
11111111
11
11111 1
111111111111
1 111111111111
111
111
11
111
11
111111 111111 111111
11111111111111111111111111111111111
11111111
111111111111 1111111
111
1111111111111
111111 111111
11111111
111111111111 111111111111 11
11111111 1111111 1111111111111111111111111111111111111
1111111111111
11111111111111111111111111111111111
11111111
1111111111
1
11111111
111111111111 111111111
1
1111111111111
111111 111111111
1 111111111111
11111111
1111111
111111111
11111111 1
111111111111
11111111 11
111111111
111
11
111111111
11 11111111111
111
1111
1111 11 111
11
11
111
11
111
1111
111111
11
11111111111111
111
1111
111111
11 11
1 11 11 11111
1111111
11111
111 11
1111111
1111
111
1111
11
111111
1111
1111
111
1111
111111
1
111 111111111
11111
111111111
1 111111111111
111 11111 111
11
111
11
111111 111111
11111
11111111111111111111111111111111111
11111111
111111111111 1111111
1111111111111
111111 111 111111 111111
11111111
111111111111 111111
111111111111
11
11111111
111111 1111111111111
1111111
111111
11111
11111111111111111111111111111111111
11111111
111111111111
11111111
111111 1111111111111
111111 11
11111111
111111
11111 11111111
111111
111111111
111
11
11111111
111111111
111
11111111111111111
11
111111111111111
11111111
111111111
111111111111111111111111111111
11 11111111111111111111111111
111
11111111111111111
11
111111111111111
1111111111111111
11 1111111111111111111
11
111
11
11
111
11
11
111
11
1111
11111111111111
111
111111
11
111111
111111
11
111
111111
1
11
11111111111111111111111111
111111111111111111111111111111
11 11111111
11111111111111111111111111
11
11
11111
111
1111111
11111
11111
111
111111
11
111 11
1111111
1111
11 11
1
111
111111
111111 1
111
111111111
11111111
111111111
1111111111111111
111111111111111
11111111111111111111111111
111111111111111111111111111111
1111111
11111
111
1111111111111
11 11
11111111111
1
111
111111 11111111
111111111
1111111111111111
111111111111111
111
11
111111
11
111
11
111111111111111
11111111
111111111
111
11111111111111111111111111
111111111111111111111111111111 111 111 111111 111111 11111
11111111111111111111111111111111111
11111111
111111111111 111111
1111111
1111111111111 111111
11111111
111111111111111111 111111
111111111111
11111111
111111111111
11
11111111
111111
111111111111
11111111
111111 111111111111
1111111111111
1111111
111111
11111
1111111111111
1111111
111111 1111111111111
111111111111
111111111111
11111111
111111 111111111111
11111 1111111111111
111111
111111 111111 1111111
11111111
111111
11111
111111111111
11111111111111111111111111111111111
11111111
111111 1111111111111
111111 111111111
11111111
111111
11111 11111111
111111
111111111111
11111111
111111
1111111111111111
111
11 11111111
111111111
11 111111111
1111111111111111
111
11 11
11 11111
11
11111111111111111111111111
111111111111111111111 11111
111
1111111
11111
111
1111111
111111
111111 11111111
1
111111111111111
111111111
111
11111111111111111
11111 111111111 111
111111111111111
111
11 111
111
111 111111 111111
11111111
111111111111 111111
11111
1111111
1111111111111
111111 111111
11111111
111111111111 111111111111
11111111
111111 11111111
111111 1111111
111111
11111 11111111
111111 1111111111111
1111111
11111111
111111
11111 111111111111
11111111
111111
111
11 11
1111
111111111
111111111
1111111111111111
111111111111111
11 11 11 111111111
11 11 1111111111111111
11111111
11111
111
1111111
111
11111
1111111
111111111 111111111
111
11
111111111111111
1111111
111111111 11
111111111
111 11 111
11 111111111111
111
111
111 111111 111111
11111111
111111111111 111111
11111
1111111
1111111111111 111111
11111111
111111111111111111 111111111111
11111111
111111 111111111111
11111111
111111 1111111111111
1111111
111111
11111 111111111111
11111111
111111 1111111111111
111111 1111111
11111111
111111
11111 111111111111
11111111
111111
111
11
111111111
11111111
11111
1111 11
111111111111111
11 1111 11 111111111
111111111
11111111
1111111
11111111
111111111111 11111111 11
111111111
11111111 111111
111111
11 1111
1111 11111
111 1111111
11111
11111
11111 1111
111111
111111111
111
1 1
1111
111111
11 11
111111111
111
11 111111111111
111111111111111111
11111111111 111
11 11111111
1
11111
11111
111
11111111
1
11111
111
1111
111111111111
111
111 111111
11111111
111111111111
11111111 111111
111111111111
111111
11111
1111111
1111111111111
111111 111111 111111
11111111
111111111111 111111111111
11111111 111111111111
11111111 1111111111111
1111111 111111111111
11111111 111111
1111111111111
1111111
11111111 111111111111
11111111
11111111 1111111111111
111
11
111111111
11111111
11111
11
1111
1111111111 11
11
111111111111111
11 11111111
1111111
11111111
11
111111111111
11 111111111
11111111 11
111111111
11111111 111111
111111
11
1111
1111 11111
1111
11
111
11111
111 11
11
11111
11111
111
11111
111
1111
1
1111
1
111111
111
1111
1
1111
111111
11 11 111111111111111111
111111111111
11111111111111
111
111
11 11111111
1
11111
11111
111
11 111
11 111111111111
11111111
1
11111
11
111
11
111111111111
111
111 1111111 11111111
111111111111111111 111111111111
11111111 111111111111
11111111 1111111111111
1111111 111111111111
11111111 1111111111111
111111 1111111
11111111 111111111111
11111111
111
11 11111111
1111111
11
11111111
11
1111111111111 11 11111111111111111 111111
11 1111 11 111 11111
11111 111
1111
1
1111
11
111111
11 11 11 11111111
1
11111
111
1111111
111111111111 111
1111111 11111111 11111111
11111111 11111111 1111111 11111111
11111111 11111111 11111111
11111111
11111
11
1111
1111111111 11
11
111111111111111
11 11 111111111
11111111 11
111111111 111111 1111 11111
1111 111 11111
11111
11 111
1111
1
1111
1111111111111
111111 111111111111111 11111111
1
11111
11111
111
11 111
11 111111111111 111
111
11 11111111
1111111
11111111
11
111111111111
11111
11
1111111 11
11
111111111111111
11 11111111
1111
111
111111111111
11111111
11 111111111
11111111 11
111111111 111111
11
111111
1111
1111
11
111
11111
11111 11111111
11111
11111
1111111111
1111
1
1111
1
111
1
111111
11
1111
1
1111
1111111111111111
111
111111
11 11111111
1
11111
111
1111
111111111111
11111111
1
11111
11111
111
11 111
11 111111111111
111111111111
111
111 11111111 1111111 11111111 11111111 1111111 11111111 11111111
111
11
111
11
11111111
1111111
11111111
11
11111111
11111
11
1111
1111111111 11
11
111111111111111
11 111111111111
11111111
1111111
11111111
11
11111111
11 111111111
11111111
111111111
11111111
11111111111
1111111
111111
11
111111
11
111111
11
1111
1111
1111
11
1111
11111
1111
11
1111 111
111
11111
1111111
11111
11111
11111
11
111
1111
1
1111
1
111111
11 11
111
1111
1
111
111
11 111111111
111111111111
111111
1111
1
1111
1
111111
11
111111111
1111111 111
111111111111
111111111111
111
11
111
11
11111111
1
11111
111
1111
11111111
1
11111
11111
111
11 111
11 111111111111
111111111111
11111111
1
11111
111
1111
111
111
111 11111111
11111111 1111111
1111111 11111111
11111111 11111111 11111111 1111111 11111111 11111111 11111111
111
11
111
11
111
11
11111111
1111111
11
11111111
11
111111111111
11111
11
1111
111111
11111111
11111
11
111111111111
11111
111111
11
1
111 11
11
111111111111111
11 11111111
111111111111
1
11 11111111
111111111
11111111111
1111111
1111111111111
11
1111111111111
111111111111111111111
11
111111
11
111111
11
111111
11
111111
11
1111
1111
1111
11
111
1 11
1 11111
11111
11111
11111111111
11
11111 11111
111
11111111111
11 11111
11
11111
11111
111
1111
1
11
111111
11 1
111111111111
111111 1111111
1 1111111111111
1111
1111
111
1 1111111111111
111111
11
111111
11 11
111
11
111
111
11
1111111111111111111111
111111
11 11 111
11
11111111
1
11111
111
1111111
111111111111
11111111
1
11111
11111
111
11
11 111111111111
111111111111
11111111
1
11111
11
111111111111
11111
11111
1111
1 11
111
111
111
111 111111111111
11111111 1111111111111
1111111 111111111111
11111111 111111111111
11111111
111111111111 111111111111
11111111
11111111 111111111111 1111111111111
1111111
11111111 1111111111111 111111111111
11111111
1111111 111111111111 1111111111111
11111111
11111111 1111111111111 111111111111
11111111
11111111 111111111111
11111111
111
11
1111111111111111111111111
111111111111111
11 111111 11
11111111111111111111
11
1111111111111111111111111
11111
11111 111
111111111111111111111111111 111111111111
111
11
11111
11111
1111 111
111 111111111111
11111111
111111 1111111111111
1111111
111111
11111 111111111111
11111111
111111 111111111111
11111111
111111 111111111111
11111111
111111 1111111111111
1111111
111111
11111 111111111111
11111111
111111 1111111111111
11111111
111111
11111 111111111111
11111111
111111
111
11
111
11
11111111111111111111
1111111111111111 11
11111111111111111111
1111111111111111 11
11111
111
11111
1111111111111111111111
111
11
1111111111111111111111
1111111111111111
111111111111111
111
11
1111111111111111
111111111111111
111
111 111111111111
11111111
111111111111
11111111
111111 1111111111111
1111111
111111 1111111111111
111111
11111
1111111
111111
11111 111111111111
11111111
111111
111111111111
11111111
111111 111111111111
11111111 111111111111
11111111
111111 111111111111 1111111111111
1111111
111111 1111111111111 111111111111
11111111
111111 111111111111
11111 1111111111111
11111111
111111 1111111111111
111111 111111111111
11111111
11111 111111111111
111111
11111111111111111111
11111111111111111111 11111
11111
1111111111111111111111
1111111111111111111111
111111111111
11111111
111111 1111111111111
1111111
111111
11111 111111111111
11111111
111111 111111111111
11111111
111111 11111111
111111 1111111
111111
11111 11111111
111111 11111111
111111
11111 11111111
111111
1111111111111111111111111
11111111111111111111 1111111111111111111111
1111111111111111111111
111111111111
11111111 1111111111111
1111111
111111 1111111111111
111111
11111 111111111111
11111111
111111 111111111111
11111111
111111
111111111111
11111111 111111111111
11111111
111111 1111111111111
1111111
111111
11111 111111111111
11111111
111111 1111111111111
11111111
111111
11111 111111111111
11111111
111111
11111111111111111111
11111111111111111111
11111111111111111111
1111111111111111111111
1111111111111111111111
1111111111111111111111
111111111111
11111111 1111111 111111111111
11111111 111111111111
11111111 111111111111
11111111
111111 111111111111
11111111 1111111111111
1111111
111111 1111111111111
1111111 11111111
111111 111111111111
11111 111111111111
11111111 1111111111111
11111111
111111 1111111111111
111111
11111111 111111111111
11111111
11111 111111111111
111111
11111111
(a) Clock buffer layout (b) XOR gate layout (c) D-flip-flop layout (60µm × 63µm)
(30µm × 25µm) (30µm × 63µm)
Figure 3.11: Layouts (from left to right) of the clock buffer, XOR gate, and flip-flop (2
latches and emitter followers)
3.4 Summary 37
Figure 3.12: System simulation after extraction with 11-GHz clock
signals (from bottom to top) are: two 11-Gb/s error-free 27 − 1 PRBS signals coming from the
generator; the 22-Gb/s output signal from the 8-to-4 MUX; the distorted 11-Gb/s input to the
PRBS checker and the errors detected by the checker.
3.4. Summary
This chapter described the design considerations for the 27 − 1 PRBS generator and checker
chip, starting from high-level MATLAB and Verilog simulations. Detailed schematics of all
system blocks were shown, together with the ideal and extracted simulation results of the entire
chip. The chip was designed and fabricated in STMicroelectronics’ 0.13 µm SiGe BiCMOS
process technology.
4 Experimental Results
This chapter will describe the experiments performed to measure the performance of the fabri-
cated 27 − 1 PRBS generator and checker. The chip is shown in section 4.1. Section 4.2 will
present the measurement procedure of the PRBS generator and the obtained results. Similarly,
the measurement procedure and results for the PRBS checker will be described in section 4.3.
4.1. Fabrication and Test Equipment
The chip was designed in the STM’s 0.13 µm SiGe BiCMOS technology with HBT fT of
160 GHz [18]. The die photo of the fabricated chip is shown in Figure 4.1, with the PRBS
generator and checker identified. The total, pad-limited chip area is 1mm × 0.8mm. The
PRBS generator and 8-to-4 MUX together occupy and area of 393µm × 178µm and consume
235 mW. The PRBS checker and error counter have an area of 308µm × 349µm and power
consumption of 350 mW. The rest of the power is consumed in the output buffers, adding up
to a total measured power consumption of 940 mW.
The fabricated chip was tested using an Agilent E4448A PSA series spectrum analyzer
for verifying the bit rate and periodicity of the generated PRB-sequence on one of the two
differential outputs. Furthermore, an Agilent 86100C DCAJ oscilloscope was employed to
monitor the other differential output. The oscilloscope is capable of identifying, locking, and
characterizing the jitter of digital sequences as long as 215 − 1 at data rates beyond 40 Gb/s.
In the absence of a 40-Gb/s BERT, use of the oscilloscope was essential for confirming the
correctness of the generated sequence.
39
Figure 4.1: Die photo of the fabricated chip
4.2. 27 -1 PRBS Generator Measurements
On-chip testing of the 27 − 1 PRBS generator was performed first. The test setup for generator
measurements and the results are described next.
4.2.1. Generator Test Setup
A detailed measurement setup for the PRBS generator circuit is shown in Figure 4.2. A 20-GHz
signal source is used to generate the clock, which is passed through a 40-GHz-bandwidth power
splitter. One output of the splitter is used to synchronize the oscilloscope, and the other output
is used as the clock input to the chip. The clock signal is applied to only one side of the
differential clock input. To be able to bias the clock input at the right DC voltage, it is passed
through a bias-tee before connecting it to the chip. The other clock input is also connected to a
bias-tee, and then terminated in 50 Ω.
The input clock and the output PRBS signal are provided onto and off the chip using differ-
ential 67-GHz GSGSG probes. The output signal was taken from both sides of the differential
output. One side was connected through a DC-blocking capacitor to the remote head of the
4.2 27 -1 PRBS Generator Measurements 41
GEN_RESET
CNT_RESET
SWITCH
VDD
VDD
VDD
CLOCK IN
Spectrum
Analyzer
Oscilloscope
VDD (Vtail)
Figure 4.2: Measurement setup for PRBS generator
digital oscilloscope. The other output was connected through another blocking capacitor to the
spectrum analyzer.
To provide all the necessary DC signals and supply current to the generator, a GPPGPPGPPGPPGPPG
DC probe was used on the top side of the chip and a PGPPG probe was used on the bottom.
The top DC probe used is bigger than required because there was no DC probe of the right size
available. The unconnected pads on the bottom are the outputs of the error counter. A second
DC probe (like the one on top), could not be connected on the bottom, because it did not fit on
the probe station. Therefore, during the first round of testing, only the PRBS generator could
be tested.
4.2.2. Generator Measurement Results
The 27 − 1 PRBS generator (together with the 8-to-4 MUX) was tested by first applying a
relatively slow clock signal and verifying the correctness of the generated sequence. The mea-
surement results of the 12-Gb/s PRBS are shown in Figure 4.3 with a 6-GHz clock signal.
Figure 4.3(c) shows the spectrum of the 12-Gb/s PRBS output. It has a sin(x)/x-type shape
with nulls at multiples of the clock frequency, indicating NRZ logic. A zoomed-in version of
the same spectrum is shown in Figure 4.3(d), with spectral tones spaced apart by 94.5 MHz.
12 Gb/s
This tone spacing is equal to the bit rate divided by the sequence length 94.5MHz = 127 bits
,
(a) Eye diagram of the generated PRBS at 12 Gb/s (b) Locked time-domain sequence at 12 Gb/s
(c) Spectrum of the generated PRBS at 12 Gb/s (d) Spectrum of the generated PRBS at 12 Gb/s
(zoomed)
Figure 4.3: Measurement results of the PRBS generator at 12-Gb/s
indicating that the correct pattern length of 127 bits is achieved. Figure 4.3(a) shows a fully-
open eye diagram at 12 Gb/s. However, this does not guarantee that every bit of the generated
sequence is correct. To confirm the correctness of the sequence, the oscilloscope was locked
to a 127-bit long pattern, and the pattern was checked bit-by-bit by scrolling through it (Fig-
ure 4.3(b)).
The same procedure was repeated for output data rates ranging from 8 Gb/s up to 24 Gb/s
in 2-Gb/s steps. Output characteristics at these bit rates are summarized in Table 4.1. The
highest bit rate at which the PRBS generator was found to work correctly is 23 Gb/s. Plots
confirming correct operation at 23 Gb/s are shown in Figure 4.4. The time-domain sequence
Gb/s
is correct (Figure 4.4(b)) and the spectral tones are spaced apart by 180.9MHz = 23 127 bits
(Figure 4.4(d)).
4.2 27 -1 PRBS Generator Measurements 43
Output Bit Rate Jitter (rms) Eye Amplitude Rise Time Fall Time Eye SNR
8 Gb/s 1.39 psec 299 mV 23.3 psec 21.1 psec 14.58
20 Gb/s 1.337 psec 251 mV - - 11.07
24 Gb/s 1.276 psec 268 mV - - 8.02
Table 4.1: Performance summary of the PRBS generator at different bit rates
(a) Eye diagram of the generated PRBS at 23 Gb/s (b) Locked time-domain sequence at 23 Gb/s
(c) Spectrum of the generated PRBS at 23 Gb/s (d) Spectrum of the generated PRBS at 23 Gb/s
(zoomed)

To further illustrate that the correct sequence is generated at bit rates up to 23 Gb/s, the
output was saved using the oscilloscope, and plotted against an ideal 27 − 1 PRBS generated
using MATLAB, as shown in Figure 4.5. Correct PRBS generation was also obtained with
clock frequencies as low as 100 MHz, demonstrating the very wide bandwidth of the PRBS
generator.
With a 12-GHz input clock and a 24-Gb/s output, an open eye was obtained (Figure 4.6(a)).
Gb/s
Also, the spectrum tones have the right spacing of 189.2MHz = 24
127 bits
(Figure 4.6(c)). How-
ever, the oscilloscope could not be locked to the sequence to observe it in time domain. There-
fore, even though all logic blocks inside the generator operate up to 24 Gb/s, their delay relative
to the clock cycle time limits error-free PRBS generation to 23 Gb/s. The 27 − 1 PRBS gen-
erator produces 4, appropriately delayed, parallel output streams at 23 Gb/s each, which can
be further multiplexed to an aggregate PRBS output at 92 Gb/s with minimal circuitry. The
4-channel PRBS generator consumes 235 mW from 2.5 V, which results in only 60 mW per
output lane.
The PRBS generator produces an output at 24 Gb/s, which means that, in the generator
core, latches that consume 2.5 mW are switching at 12 Gb/s. To the best of my knowledge,
this is the lowest power latch operating above 10 Gb/s in any technology [19]. This BiCMOS
CML latch implementation works with 1-mA tail current from a 2.5-V supply. Other recently
reported sub-3.3V bipolar logic families [20–22] consume significantly more power because
they require doubling the tail current for a given logic function. While 130-nm or 90-nm MOS
CML latches operate from 1.5-V or lower supplies, they require more than 2 times higher tail
currents and inductive peaking to operate above 10 Gb/s, thus offsetting the advantage provided
by the lower supply voltage [23].
Generated 27−1 PRBS at 23 Gb/s

0.2
Amplitude (V)
−0.2
4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8
time (sec) x 10−8
7
Generated 2 −1 PRBS at 12 Gb/s
Amplitude (V)
0.2
−0.2
4.5 5 5.5 6
time (sec) x 10−8
7
Ideal 2 − 1 PRBS
0.2
−0.2
20 40 60 80 100 120 140 160 180
bit number
Figure 4.5: Measured 23-Gb/s (top), measured 12-Gb/s (middle), and ideal (bottom) time
domain 27 − 1 PRB-sequences
4.3 PRBS Checker Measurements 45
(a) Eye diagram of the output at 24 Gb/s
(b) Spectrum of the generated PRBS at 24 Gb/s (c) Spectrum of the generated PRBS at 24 Gb/s
(zoomed)
4.3. PRBS Checker Measurements
This section presents the test setup and measurement results of the PRBS checker and error
counter. These tests were performed a few months after the generator measurements.
4.3.1. Checker Test Setup
The test setup for the PRBS checker, shown in Figure 4.7, is similar to the generator test setup.
In this case GPPGPPGPPG DC probes were used both on the top and at the bottom to be able to
apply all DC inputs to the chip and to have access to all counter outputs. As before, the output
PRBS was observed using a spectrum analyzer and a digital oscilloscope to confirm correct
operation of the generator during checker measurements. The counter outputs were connected
to another, low speed, oscilloscope.
4.3.2. Checker Measurement Results
As was shown previously in the chip schematic of Figure 3.8, access to the PRBS checker is
possible only through the generator on the input side, and only through the error counter on the
output side. The input to the checker comes from the PRBS generator though a selector. The
selector is manually controlled by a power supply to switch between two possible inputs to the
checker. At the switching time, the checker is not synchronized to the input from the generator,
and errors result. The output of the checker is accessible only as a bit count from the 5-bit error
counter.
When measuring, the PRBS generator was first set up to generate a correct output, which
means that a correct PRBS was applied to the checker. At this point all outputs of the counter
have to be at logic 0. Then, the voltage value of the SWITCH input was changed to introduce
errors into the checker. However, as shown in Figure 4.8, the error counter outputs were ob-
served to switch all the time, without regard to the correctness of the generator output. This
GEN_RESET
CNT_RESET
SWITCH
VDD
VDD
VDD
CLOCK IN
Spectrum
Analyzer
Oscilloscope
VDD (Vtail)
OV
BIT3
BIT2
BIT1
BIT0
Figure 4.7: Measurement setup for PRBS checker

4.4 Summary 47
Figure 4.8: Bit0 (bottom) and Bit1 (top) outputs of the error counter
means that errors were detected by the checker, and error pulses were produced constantly,
even with a correct PRBS input.
The possible cause of this problem is that there is no synchronization of the clock with the
data on-chip. Instead, the PRBS output from the generator is connected directly to the error
checker by a 200 µm-long transmission line. However, interconnect and circuit delays may
offset the checker clock signal with respect to its data input. In this case, the checker would
always observe a wrong PRBS input and produce pulses, which are then counted by the error
counter.
Originally, the plan was to have a separate input into the checker. However, due to the lack
of high-speed probes, the design was changed such that the error checker input comes directly
from the PRBS generator. A variable delay block is needed between the PRBS generator and
the error checker to compensate for the propagation delay of the logic and along the connecting
line.
Thus, correct functionality of the PRBS checker could not be detected during this round of
testing. More measurements will have to be performed to clarify the exact cause of the problem
and solve it.
4.4. Summary
Measurement results of the fabricated PRBS generator and error checker chip were presented
in this chapter. The 27 − 1 PRBS generator was found to operate correctly up to 23 Gb/s.
The individual generator blocks switch at a bit rate of 24 Gb/s. Thus, individual latches that
consume 2.5 mW worked at 12 Gb/s. To the best of my knowledge, this represents the lowest
power latch operating above 10 Gb/s in any technology. Unfortunately, there were problems
with testing the error checker and counter.
Comparison and Power
5 Optimization of High-Speed
Logic Topologies
In the previous chapters various PRBS generator topologies were compared in terms of their
power and speed characteristics, and the fabricated chip was described. In this chapter avenues
for further speed improvements and possible power reduction of high speed digital blocks
will be presented. This chapter will concentrate on the transistor-level gate design. Several
high-speed logic families (such as CML and CMOS) will be compared in terms of their speed
performance, power and area requirements.
In section 5.1 an overview of existing high-speed digital gates will be given. Current-mode
logic latch design will be outlined in section 5.2. Possible improvements to the existing latch
topology will be presented in section 5.3. The performance of different BiCMOS latches will
be compared in section 5.4 and a comparison with 65-nm CMOS logic will be done section 5.5.
5.1. High-Speed Digital Logic Gates
Usually, in high-speed digital applications, CML logic gates are implemented as cascoded
differential pairs. The logic function is achieved by switching a constant current Itail from one
side of the differential pair to the other side. The currents are transformed into low and high
voltage levels when they pass through the load resistors. The difference between the low and
high levels (logic swing) is smaller than the supply voltage, which is one of the reasons for this
topology being fast. The collection of logic gates implemented in this manner is called current-
mode logic (CML) [24]. Some of the logic gates of this family are shown in Figure 5.1. An OR
gate is not shown because it can be formed from an AND gate by DeMorgan’s Theorem, that
is, by negating the inputs and the output. Negation of signals does not require extra circuitry
because all gates are fully differential. Figure 5.1 shows the logic blocks composed of both
NMOS and SiGe hetero-junction bipolar transistors (HBT), hence, this logic topology will be
referred to as SiGe BiCMOS CML logic.
49
50 Comparison and Power Optimization of High-Speed Logic Topologies
VDD VDD VDD VDD
OUT OUT OUT OUT
IN1 IN0 IN0 IN0 IN
SEL IN1 IN1 CLK
Itail Itail Itail Itail
(a) Selector gate (b) XOR gate (c) AND gate (d) Latch
Figure 5.1: CML digital gates
CML gates can also be implemented with only NMOS devices, in which case they will
be referred to as MOS-CML logic [25]. The original implementation of CML gates was with
only bipolar devices. Nowadays bipolar devices usually require higher supply voltages than
MOSFETs due to larger VBE , so their use is becoming infrequent.
Another variation that is often done to CML gates is to add source- or emitter followers on
the inputs and/or the outputs of the gate. When implemented with MOS transistors, this logic
family is called Source-Coupled FET Logic (SCFL) [24]. In the bipolar version, it is called
Emitter-Coupled Logic (ECL) [26]. ECL logic can usually operate faster, because the added
emitter/source followers decouple the node capacitances, at the expense of power consumption.
Some ECL gates are shown in Figure 5.2. Similarly, gates with double emitter followers can be
built, with even higher supply voltage and current consumption. It should be noted that circuits
with followers may become unstable, and that source followers degrade the signal. Due to the
high power consumption of ECL gates, they will not be described here further.
The performance of CML digital gates highly depends on the fT of the transistors. The
VDD VDD VDD VDD VDD VDD
IN0 OUT
IN1 IN
OUT
CLK
Itail Itail
(a) ECL XOR gate (b) ECL latch
Figure 5.2: ECL gates

5.1 High-Speed Digital Logic Gates 51
frequency fT is defined as the frequency at which the current gain (h21 ) of the transistor is equal
to 1. fT depends on the dc bias current through the transistor, and reaches a maximum for some
bias current. The bias current for which fT is maximum, when normalized to the transistor size,
is called the peak-fT current density. For optimal performance, the transistors in a logic gate
have to be biased such that they are at the peak-fT current density. Optimal performance means
that the maximum switching speed is obtained for a given current consumption.
For NMOS transistors, the peak-fT current density, JpfT M OS , stays constant at 0.3 mA/µm
as long as the constant field scaling rules apply [27]. This implies that transistors in different
technology nodes, or with different widths and lengths at the same technology node, must be
biased at peak-fT current density when the bias current IDS and the transistor width W are
related by IDS /W = 0.3 mA/µm. Figure 5.3 shows plots of fT for several technology nodes
and NMOS transistor sizes [6].
For HBT transistors, the peak-fT current density (JpfT HBT in mA/µm2 ) varies across tech-
nology nodes but remains constant to a first order approximation for transistors with different
emitter lengths in a given technology. The fT versus transistor bias current is plotted in Fig-
ure 5.4 for 0.25 µm SiGe HBTs and several emitter lengths [24].
Figure 5.3: NMOS fT as a function of current density for several technology nodes and
transistor sizes

!

!

!

!

Figure 5.4: HBT fT as a function of bias current for several transistor sizes
5.2. SiGe BiCMOS CML Logic / CML Latch Design
This section will present several possible BiCMOS CML latch topologies. The latch is chosen
as a representative block because it contains the largest output capacitance. Furthermore, the
latch operates at the full clock frequency, rather than at the data rate. Design of the latch is
critical because it is needed in data re-timing and synchronization, which is essential to ensure
the correct operation of high-speed logic. The design of latches shown in Figure 5.5 will be
described in this section, and followed by a performance comparison.
The design of a CML logic gate starts by selecting dc voltage levels at each node. The
dc levels have to be such that when the inputs and output nodes are balanced (zero differential
signal) then all transistors are in saturation. Thus the VDS of NMOS and VCE of HBT transistors
have to be approximately 0.7 V and 0.9 V respectively (in 0.13 µm technology). Next, the tail
current Itail and load resistors RL are chosen to produce the desired voltage swing ∆V .
∆V = Itail RL (5.1)
∆V is the single-ended voltage difference between logic-low and logic-high of the gate. When
the inputs and output are balanced, the voltage drop across RL is 0.5 · ∆V . The supply voltage
VDD required for this gate is given by the sum of all VBE or VGS voltage drops in the transistor
stack and the voltage drop across RL . The power consumption of the gate is then Itail VDD .
5.2 SiGe BiCMOS CML Logic / CML Latch Design 53
VDD VDD
OUT OUT
IN IN
CLK CLK
Itail Itail
(a) BiCMOS CML latch (b) MOS CML latch
Figure 5.5: CML latch schematics
The switching speed of the gate depends on ∆V , Itail , and node capacitances:
∆V × C/W
switching time ∝ (5.2)
Itail /W
Where C/W and Itail /W are technology parameters. Hence, to increase the switching speed,
the bias point must be chosen such that the ∆V needed to fully switch the transistors is small.
Also, the transistors themselves must be small to minimize capacitance, but the current must
be as large as possible. However, increasing the current density beyond the peak-fT current
density increases ∆V . For MOSFETs, the region below the peak-fT current density bias cor-
responds to operation according to the square law model. In the square law region, the swing
required to fully switch the transistor is given by [28]
√
∆V > 2VEF F (5.3)
However, when the device is biased at or above the peak-fT current density, the square law no
longer applies. In this case the swing required to switch the transistor is approximately [6]
∆V > 2VEF F (5.4)
The ∆V of eq. 5.4 is larger than that of eq. 5.3 both because the coefficient is larger and VEF F is
larger. From this discussion it follows that, for best performance, MOSFETs have to be biased
close to, but below the peak-fT current density. Therefore, the bias current Itail is chosen to be
Itail = Wgate × JpfT M OS (5.5)

With this bias current, the MOSFETs are biased at half peak-fT current density when the
inputs are balanced, and at zero or full peak-fT current density when the inputs are switched
to one side. VEF F that corresponds to half peak-fT current density is 300 mV in a 0.13 µm
technology. To account for temperature and process variations, ∆V is chosen to be 400 mV to
500 mV .
The swing needed to fully switch a bipolar transistor can be as low as four times the ther-
mal voltage [28], but in practice needs to be 200 mV to 300 mV when temperature, process
variations, and RE Itail voltage drop are taken into account. The tail current for SiGe HBT
transistors is chosen such that it corresponds to 0.75 times peak-fT current density when the
inputs are balanced, or to 1.5 times peak-fT current density when the inputs are switched [29].
Itail = 1.5 × we × le × JpfT HBT (5.6)
Even though in this case the peak-fT current density is not constant across technologies, it is
still constant for different-size HBTs, when the bias current is normalized to the emitter area.
In addition to the latch topologies shown in Figure 5.5, other modifications are possible for
increasing speed and/or reducing power. To increase the switching speed, peaking inductors
can be added to the load. Inductors reduce the capacitance term of equation 5.2 by resonating
it out. For power reduction it is possible to operate the latches without the current source
transistor. This allows to decrease the power supply voltage while using the same total current,
thus the power consumption is reduced [19].
The performance of latches can be compared based on their time constant τ . The time
constant is suitable for comparison because both the propagation delay through the latch and
the rise and fall times are proportional to it. An approximation of τ for latches can be derived
similarly to τ for cascode circuits [6]. τ is a sum of open-circuit-time-constants at the input
and output nodes and accounts for the fanout k. Equation 5.7 gives an approximation of τ for
a BiCMOS cascode latch. When deriving equation 5.7, it is assumed that the latch is loaded by
a similar latch, but with a tail current of k × Itail in which all transistors and their capacitances
are k times larger. Also, it is assumed that the output of the latch is connected to the top
(bipolar) pair and not to the bottom MOSFET pair. Figure 5.6 illustrates the relevant parasitic
capacitances used to derive the latch time constant τ . The first term of equation 5.7 is the time
constant (τin ) at the clock input of the latch. The second term (τmid ) is the time constant in
the middle (cascode) node of the latch. The third term (τout ) represents the charging of output
capacitances by the tail current. It takes into account the Miller capacitance of the latching
5.2 SiGe BiCMOS CML Logic / CML Latch Design 55
VDD VDD
RL RL/k
RL/k
Q6 k·CBC
RL
Q5 CBC Q6
Q1 Q2 Q3 Q4 Q3 Q4
CCS CBC CBE +
Q1 k·CBE
+CCS (1-AV)CBC
CBE
Q2
M1 M2
CBE
CGD
CDB
M1
RG
Itail k·Itail CGS
(a) BiCMOS CML latch with load (b) One side of (a) with relevant capacitances
Figure 5.6: Time-constant derivation for a BiCMOS latch
pair. The fourth term (τf anout ) describes the fanout of the latch.
τBiCM OS−Latch ≈ τin + τmid + τout + τf anout =

RG ∆V gm,M OS
= CGS + 1 + CGD
RL Itail gm,HBT
2CBE + CDB + CGD
+
gm,HBT
∆V
+ {2CBC + 2CCS + CBE + (1 + gm,HBT RL ) CBC + CIN T }
Itail
∆V
+k {CBE + (1 + gm,HBT RL ) CBC } (5.7)
Itail
A similar expression can be derived for MOS-CML latches, where all devices are NMOS
transistors (Figure 5.5(b)). It is given in equation 5.8.
τM OS−Latch ≈ τin + τmid + τout + τf anout =

RG ∆V
= {CGS + 2CGD }
RL Itail
2CGS + 2CSB + CDB + CGD
+
gm,M OS
∆V
+ {2CGD + 2CDB + CGS + (1 + gm,M OS RL ) CGD + CIN T }
Itail
∆V
+k {CGS + (1 + gm,M OS RL ) CGD } (5.8)
Itail
5.3. Power and Speed Optimizations of CML Latches
After presenting the basic CML latches and how they are designed, this section will describe
the possible modifications that can be done to further reduce power consumption and improve
the speed at which these latches operate.
The CML latches presented in the previous section require a supply voltage of 2.5 V . As
seen in Figure 5.5, 0.4 V - 0.7 V are spent on the transistor that sets the tail current in the
latch. This transistor can be eliminated to reduce power consumption without sacrificing per-
formance. The new latch configuration is shown in Figure 5.7. Now, the latch can operate
from 1.8 V , or lower, with the same current as before. The speed performance is maintained
because ∆V , Itail , and all capacitances are kept constant. However, precaution must be taken
in the design process to ensure that the current through the latches of Figure 5.7 is the same
as in the latches of Figure 5.5. The supply voltage can be further reduced with newer process
technologies, in which smaller voltage drops are needed for each stacked MOSFET transistor.
Biasing of the CML latches of Figure 5.7 proceeds similarly to the earlier case. Since
there now are two separate branches that go to ground, the current in each branch is Ibranch =
0.5 · Itail , where Itail is the corresponding tail current of the latches in Figure 5.5. The MOS
transistors are sized such that Wgate = Ibranch /0.5JpfT M OS when there is zero differential input
to the latch. The SiGe HBT transistors are sized such that we ×le = Ibranch /0.75JpfT HBT when
there is zero differential input to the latch. This choice of Ibranch and transistor sizes results
in peak-fT current density biasing and thus maintains the optimal switching characteristics
described in section 5.2. The time constant τ for each latch of Figure 5.7 can also be calculated
using equations 5.7 and 5.8.
VDD VDD
OUT OUT
IN IN
CLK CLK
I I
(a) BiCMOS CML latch without current (b) MOS CML latch without current
source source
Figure 5.7: CML latch schematics without current source and VDD reduced to 1.8V
5.3 Power and Speed Optimizations of CML Latches 57
VDD VDD
OUT OUT
IN IN
CLK CLK
Itail Itail
(a) BiCMOS CML latch with inductive (b) MOS CML latch with inductive
peaking peaking
Figure 5.8: CML latch schematics with shunt inductive peaking
VDD VDD
OUT OUT
IN IN
CLK CLK
I I
(a) BiCMOS CML latch with inductive (b) MOS CML latch with inductive
peaking and without current source peaking and without current source
Figure 5.9: CML latch schematics with shunt inductive peaking and without current source
The second modification that can be applied improves speed without requiring more power.
This is achieved by adding shunt peaking inductors to the latches discussed so far. The induc-
tors are added as part of the load and are used to extend the bandwidth of the circuit by reducing
the effect of the output capacitance (which is dominant). Since the inductors are connected in
series with the load resistors, they can be added to the latches of both Figures 5.5 and 5.7. The
latch topologies with inductors are shown in Figures 5.8, and 5.9.
The peaking inductors do not affect the biasing of the transistors, but they reduce the time
constant of the latch, and thus allow it to operate faster. For flat group delay response (i.e.
minimum deterministic jitter), the inductor value is selected according to
Cout RL2
L= (5.9)
3.1
where RL is the load resistance and Cout is the total capacitance at the output node. This value
of L improves the output time constant 1.6 times. For a BiCMOS CML latch, the improved τ
becomes
τBiCM OS−Latch−Ind ≈ τin + τmid + τout + τf anout =

RG ∆V gm,M OS
= CGS + 1 + CGD
RL Itail gm,HBT
2CBE + CDB + CGD
+
gm,HBT
∆V {2CBC + 2CCS + CBE + (1 + gm,HBT RL ) CBC + CIN T }
+
Itail 1.6
∆V {CBE + (1 + gm,HBT RL ) CBC }
+k (5.10)
Itail 1.6
And for a MOS-CML latch the improved τ is
τM OS−Latch−Ind ≈ τin + τmid + τout + τf anout =

RG ∆V
= {CGS + 2CGD }
RL Itail
2CGS + 2CSB + CDB + CGD
+
gm,M OS
∆V {2CGD + 2CDB + CGS + (1 + gm,M OS RL ) CGD + CIN T }
+
Itail 1.6
∆V {CGS + (1 + gm,M OS RL ) CGD }
+k (5.11)
Itail 1.6
5.4. Performance Comparison and Simulation Results
This section presents a performance comparison of all the latches described in this chapter.
The comparison is carried out both with hand calculations based on technology data and with
simulations of the latches under identical conditions. The calculations and the simulations
are conducted for two technologies, to be able to predict the feasibility of the proposed latch
topologies for future applications. The first is a production 0.13 µm SiGe BiCMOS technology
5.4 Performance Comparison and Simulation Results 59
with transistor fT of 160 GHz [18]. The second is a 90 nm SiGe BiCMOS technology under
development with transistor fT of 230 GHz [30].
To make the comparison fair, all latches were designed to operate with a total current
consumption of 1 mA. A current of 1 mA was chosen because it is the current that keeps a
minimum-size HBT in the 0.13 µm SiGe BiCMOS technology biased as outlined in the previ-
ous section. Then, the maximum bit rate at which the latch operated properly was observed.
Proper operation condition is reached when the output swing is equal to the designed swing.
For hand calculations, equations 5.7, 5.8, 5.10, and 5.11 were used. The model parameters
of the two technologies that were used in calculations are summarized in Table 5.1. The device
sizes used to realize the 1-mA latches are given in Table 5.2 for each latch configuration.
Finally, Table 5.4 compares the performance of the latches based on power consumption, the
calculated time constant τ , and the maximum simulated speed of operation.
Simulated eye diagrams of the various BiCMOS CML latch topologies are demonstrated in
Table 5.3. Eye diagrams at the maximum bit rate where the output swing is correct are shown.
The bit rate and power consumption obtained in these simulations are summarized in Table 5.4.
The latches in each simulation had a fanout of k = 1.
The latch shown in Figure 5.5(a) was fabricated in 0.13 µm SiGe BiCMOS technology,
as part of the PRBS generator that was described in chapters 3 and 4. It was found to work
160-GHz 230-GHz
Device Parameter 0.13 µm NMOS 90 nm NMOS
SiGe HBT SiGe HBT
gm at peak fT 1.1 mS/µm 38 mS/µm 1.4 mS/µm 74 mS/µm
CGD , CBC 0.5 f F/µm 11 f F/µm2 0.4 f F/µm 8 f F/µm2
CGS , CBE 1.25 f F/µm 2.75 f F/µm 1.0 f F/µm 2.0 f F/µm
CDB , CCS 1.7 f F/µm 1.1 f F/µm 1.1 f F/µm 0.8 f F/µm
JpfT M OS , JpfT HBT 0.3 mA/µm 6 mA/µm2 0.3 mA/µm 13 mA/µm2
we - 0.2 µm - 0.15 µm
Table 5.1: Model parameters for 0.13 µm and 90 nm SiGe BiCMOS technologies
0.13µm SiGe BiCMOS 90nm SiGe BiCMOS

Latch NMOS ( W
L
) HBT (le ) RL L NMOS ( W
L
) HBT (le ) RL L
2 µm 2 µm
Fig. 5.5(a) 2× 0.13 µm
0.64 µm 400 Ω - 2× 90 nm
0.50 µm 200 Ω -
1 µm 1 µm
Fig. 5.7(a) 2× 0.13 µm
0.64 µm 400 Ω - 2× 90 nm
0.50 µm 200 Ω -
2 µm 2 µm
Fig. 5.8(a) 2× 0.13 µm
0.64 µm 400 Ω 2 nH 2× 90 nm
0.50 µm 200 Ω 500 pH
1 µm 1 µm
Fig. 5.9(a) 2× 0.13 µm
0.64 µm 400 Ω 2 nH 2× 90 nm
0.50 µm 200 Ω 500 pH
Table 5.2: Device sizes for 1-mA latches in each configuration

0.13µm SiGe BiCMOS 90nm SiGe BiCMOS

Latch Simulated Eye Diagrams Simulated Eye Diagrams
Fig. 5.5(a)
Fig. 5.7(a)
Fig. 5.8(a)
Fig. 5.9(a)
Table 5.3: Simulated eye diagrams for 1-mA latches with different topologies. The corre-
sponding bit rates are summarized in Table 5.4
5.5 Future Scaling and CMOS Logic 61
0.13 µm SiGe BiCMOS 90 nm SiGe BiCMOS

Latch τ Power ∆V Bit Rate τ Power ∆V Bit Rate
Fig. 5.5(a) 14.1 ps 2.5 mW 400 mV 15 Gb/s 6.3 ps 2.2 mW 200 mV 30 Gb/s
Table 5.4: Performance comparison of the different latch topologies
correctly up to 12 Gb/s. The performance of this latch after fabrication agrees closely with
simulation results that include layout parasitics. This confirms the validity of the simulation
results presented for the other latch topologies. The analysis presented in this chapter demon-
strates that it is possible to reduce the supply voltage without increasing the tail current, thus
saving power. Furthermore, it is possible to significantly increase the speed of a latch, with the
sacrifice of some area, by adding peaking inductors.
5.5. Future Scaling and CMOS Logic
To further investigate the bit rates that will be possible with future technology scaling, it is
useful to look at how CML logic compares with CMOS logic. This section will compare
MOS-CML and CMOS ring oscillators built from inverters in a 65 nm CMOS technology.
Their oscillation speed is indicative of the logic speed attainable in this technology.
MOS-CML and conventional CMOS inverters are shown in Figure 5.10. For these circuits
too, the time constants can be derived. They are given in equations 5.12 and 5.13. In the
derivation of these equations, it is assumed that a unit-size inverter has a fanout of k. For the
MOS-CML inverter the relation RL = ∆V /Itail is used. For the CMOS inverter, it is assumed
VDD VDD
OUT
IN IN OUT
Itail
(a) MOS-CML inverter (b) CMOS inverter
Figure 5.10: MOS-CML and CMOS inverter schematics

that Wp = 2Wn , so that the NMOS and PMOS devices have the same gm .

∆V ∆V RG
τM OS−CM L−IN V ≈ (CGD + CDB ) + k 1+ [CGS + (1 + gm RL ) CGD ]
Itail Itail RL
(5.12)
3 3 RG
τCM OS−IN V ≈ ro (CGD + CDB ) + kro 1 + [CGS + (1 + gm ro ) CGD ] (5.13)
2 2 ro
These equations have almost the same form if ro = RL . Since, usually, ro is larger than
RL , pure CMOS inverters are slower (have larger τ ) than CML inverters. This statement also
applies to the speeds of general CMOS gates and MOS-CML / BiCMOS-CML gates, because
the inverters are basic units of the logic gates.
The average dynamic power consumption of a CMOS inverter can be approximated by [31]

2 trise + tf all
Pdyn−avg = CL VDD fosc ≈ VDD Ipeak fosc (5.14)
2
If the maximum operating frequency fosc = 1/T occurs when T = (trise + tf all ) then the
CMOS power consumption is approximated by
VDD Ipeak
Pdyn−avg ≈ (5.15)
2
Where Ipeak is directly proportional to the transistor width. This figure is much smaller than the
power consumption of MOS-CML inverters, latches, and other gates. However, the achievable
speed in CMOS logic is also much smaller than that of MOS-CML circuits. As shown in [32],
with the benefit of scaling and using the design methodology presented earlier, 65-nm latches
can operate at bit rates up to 60 Gb/s.
5.5.1. CMOS and MOS-CML Ring Oscillators
The PRBS generator and checker chip was based entirely on BiCMOS CML logic. This logic
topology was chosen to achieve the required speed, and designed to minimize power. As part
of this thesis I have looked at the speed and power of other logic topologies. To verify the pre-
dictions that are made regarding the power and speed performance of CML and CMOS logic,
test chips were designed with both logic topologies. The test chips were designed in TSMC’s
65-nm CMOS process technology. They include CMOS and MOS-CML ring oscillator circuits
that can be easily used to evaluate the maximum logic speed that can be achieved.
In total, 5 different ring oscillator circuits were designed. Their schematics, simulation
results, and layouts are summarized in Table 5.5. For all 5 circuits V DD = 1V was used.
5.5 Future Scaling and CMOS Logic 63
The inverters in the 7-stage CMOS ring oscillators were designed with Wp = 2×Wn , which
reflects the measured Ion ratio of 90-nm PMOS and NMOS transistors. Two different inverter
sizes were used in the two CMOS ring oscillators to explore the effect of transistor width on
Wn 0.5 µm Wn 1 µm
speed and power. One with L
= 65 nm
(Figure 5.12(a)) and the other with L
= 65 nm
(Figure 5.12(b)).
Also, two different MOS-CML 7-stage ring oscillators were designed. Both oscillators
consist of inverters with a tail current of 1 mA, with transistors sized for peak-fT current den-
sity (Figure 5.12(c)). One of the designs includes 700-pH peaking inductors (Figure 5.12(d)).
Notice that, because the quality factor of the inductors is not important in this case, very high
inductance per unit area can be achieved by using narrow metal width and spacing. Thus the
700-pH, 3-turn, 3-layer stacked inductor occupies an area of only 10 µm × 10 µm.
Finally, a 60-GHz quadrature oscillator, capable of generating 8 different output phases
(because it is differential), was created. It is built from the MOS-CML inverters with induc-
tive peaking (Figure 5.12(d)). The layouts of all the designed 65-nm circuits are shown in
Figure 5.13.
These 65-nm CMOS circuits did not come back from fabrication yet.
OUT
OUT OUT
(a) 7-Stage ring oscillator (b) Quadrature ring oscillator
Figure 5.11: Top level schematics of the designed ring oscillators and quadrature oscilla-
tor
Oscillator Inverter Simulated Simulated Layout

Schematic Schematic Oscillation Frequency Inverter Power Size
1 Fig. 5.11(a) Fig. 5.12(a) fosc = 8.7 GHz Pinv = 37 µW 8.8 µm × 4.4 µm
2 Fig. 5.11(a) Fig. 5.12(b) fosc = 8.8 GHz Pinv = 65 µW 8.8 µm × 5.4 µm
3 Fig. 5.11(a) Fig. 5.12(c) fosc = 28 GHz Pinv = 1 mW 48 µm × 44 µm
4 Fig. 5.11(a) Fig. 5.12(d) fosc = 51 GHz Pinv = 1 mW 108 µm × 93 µm
5 Fig. 5.11(b) Fig. 5.12(d) fosc = 66 GHz Pinv = 1 mW 120 µm × 92 µm
Table 5.5: Summary of the designed 65-nm oscillators

1V 1V 1V 1V
1V 1V 310 Ω
400 Ω
W 1µm W 2µm 700 pH

= =
L 65nm L 65nm 4 × 1 µm
65 nm 4 × 1 µm
65 nm
W 0.5µm W 1µm
= =
L 65nm L 65nm
(a) CMOS inverter (b) CMOS inverter (c) CML inverter (d) CML inverter with in-
ductors
Figure 5.12: Schematics of blocks used in the ring oscillators
(a) 2 CMOS Ring Oscilla- (b) CML Ring Oscillator (c) CML Ring Oscillator (d) CML Quadrature Os-
tors with Inductors cillator
Figure 5.13: Layouts of the Ring Oscillators
Several observations can be made from the simulation results of Table 5.5. First, CMOS
ring oscillators produce an output signal with a smaller frequency, but also consume a lot less
power and occupy a very small area. Conversely, CML ring oscillators operate much faster
with a higher power consumption. By comparing rows 3 and 4 of Table 5.5, it can be seen that
inductive peaking significantly improves the operation speed with the expense of area.
5.6. Summary
This chapter has introduced two transistor-level design techniques for improving the perfor-
mance of high-speed digital gates. First, the power consumption of the basic CML gates can
be reduces significantly by removing the current source transistor and halving the sizes of the
other transistors. Second, the operation speed of these gates can be increased while maintain-
5.6 Summary 65
ing the same power consumption by introducing inductive peaking; this increases the circuit
area. In addition, this chapter has compared the speed, power and area characteristics of CML
and CMOS inverters in 65-nm CMOS technology.
6 Conclusion
6.1. Summary
In this thesis, the development process of a self-test IP block for high-speed applications was
described. Self-test blocks consist of PRBS generators and error checkers. To be able to place
these self-test blocks on the same chip as the circuit to be tested, the blocks have to be small
in size and consume low power. Furthermore, they must operate at a higher, or comparable,
speed to the circuit being tested.
To achieve the above goals, the design was optimized both at the system level, and at the
transistor level. At the system level an extensive comparison was performed between series
and parallel PRBS generator architectures. It was found that, for operation above 80 Gb/s,
the parallel generator architecture is much more suitable. This is due to the constant and
low fanout of the flip-flops in the parallel generator, and due to the readily available re-timed
outputs. Another very important property of parallel PRBS generators is that all outputs are
appropriately delayed one with respect to the other such that direct multiplexing is possible.
When multiplexing is used, the core generator can operate at a fraction of the final bit rate, thus
saving power.
At the transistor level, a procedure for low-power latch design was described. This proce-
dure uses the CML latch topology to avoid spending extra current in emitter or source follow-
ers. The CML latch is based on the BiCMOS cascode configuration [6]. With zero differential
input to the latch, the MOS transistors are sized to be biased at half peak-fT current density.
The HBT devices are sized at 1.5 peak-fT current density when fully switched. Using this
procedure, a latch was designed that operated at 12 Gb/s while consuming only 2.5 mW of
power. To the best of my knowledge, this is the lowest power latch operating above 10 Gb/s.
Several improvements to the basic CML latch topology have been presented. It is possible
to further reduce the power used by the latch without sacrificing speed by removing the tail
67
68 Conclusion
current transistor, thus lowering the supply voltage and still drawing the same current. Also, it
is possible to increase the speed of the latch, while consuming the same power as before, by
adding peaking inductors.
A test chip that employs the above concepts was designed and fabricated in STMicroelec-
tronics’ 0.13 µm SiGe BiCMOS technology. The chip included a 27 −1 PRBS generator, PRBS
error checker, and a 5-bit error counter. The PRBS generator worked correctly up to 23 Gb/s. It
had four parallel, appropriately shifted, outputs, which can be directly multiplexed to 92 Gb/s.
This makes the generator suitable for testing 80 Gb/s circuits. The PRBS generator consumes
235 mW, resulting in only 60 mW per 23-Gb/s output lane.
6.2. Future Work
The work presented in this thesis can be continued in several directions. These include further
circuit development together with an investigation of power reduction in high-speed digital
circuits.
For example, the 27 − 1 PRBS generator presented in this thesis was designed to be used
together with a 4:1 multiplexer, to produce a final pseudo-random sequence at a bit rate higher
than 80 Gb/s. The multiplexer was not part of the chip described previously. Using the parallel
generator approach combined with multiplexing, generators with sequence lengths longer than
27 − 1 can be designed at very high bit rates, without consuming excessive power.
The new latch topologies presented in Chapter 5 appear to have very promising perfor-
mance according to simulations. Further investigation of these topologies and fabrication of
test chips that utilize these latches is required. This will enable the development of low power
broadband systems operating above 80 Gb/s and open the way for the next generation 100 Gb/s
Ethernet [33].
Bibliography
[1] A. Koyama, T. Harada, H. Yamashita, R. Takeyari, N. Shiramizu, K. Ishikawa, M. Ito,

S. Suzuki, T. Yamashita, S. Yabuki, H. Ando, T. Aida, K. Watanabe, K. Ohhata,
S. Takeuchi, H. Chiba, A. Ito, H. Yoshioka, A. Kubota, T. Takahashi, and H. Nii, “43Gb/s
Full-Rate-Clock 16:1 Multiplexer and 1:16 Demultiplexer with SFI-5 Interface in SiGe
BiCMOS Technology,” in 2003 IEEE ISSCC Digest of Technical Papers, San Francisco,
CA, Feb. 2003, pp. 232–233.
[2] M. Meghelli, “A 108Gb/s 4:1 Multiplexer in 0.13µm SiGe-Bipolar Technology,” in 2004

IEEE ISSCC Digest of Technical Papers, San Francisco, CA, Feb. 2004, pp. 236–237.
[3] A. Rylyakov and T. Zwick, “96 GHz static frequency divider in SiGe bipolar technology,”
in IEEE GaAs IC Symposium Technical Digest, San Diego, CA, Feb. 2003, pp. 288–290.
[4] T. O. Dickson, E. Laskin, I. Khalid, R. Beerkens, J. Xie, B. Karajica, and S. P. Voinigescu,

“A 72 Gb/s 231 − 1 PRBS generator in SiGe BiCMOS technology,” in 2005 IEEE ISSCC
Digest of Technical Papers, San Francisco, CA, Feb. 2005, pp. 342–345.
[5] ——, “An 80-Gb/s 231 − 1 Pseudorandom Binary Sequence Generator in SiGe BiCMOS
Technology,” IEEE Journal of Solid State Circuits, vol. 40, pp. 2735 – 2745, Dec. 2005.
[6] T. O. Dickson, R. Beerkens, and S. P. Voinigescu, “A 2.5-V 45-Gb/s Decision Circuit

Using SiGe BiCMOS Logic,” IEEE Journal of Solid State Circuits, vol. 40, pp. 994–
1003, Apr. 2005.
[7] S. W. Golomb, Shift Register Sequences. San Francisco, CA: Holden-Day, Inc., 1967.
[8] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recipes

in C, The Art of Scientific Computing, Second Edition. New York, NY: Cambridge
University Press, 1992.
69
70 Bibliography
[9] Error Performance Measuring Equipment Operating at the Primary Rate and Above, In-
ternational Telecommunications Union CCITT Series O Recommendation, O.151, Rev. 1,
1992.
[10] E. W. Weisstein, “Totient Function,” From MathWorld–A Wolfram Web Resource.

[Online]. Available: http://mathworld.wolfram.com/TotientFunction.html
[11] F. Sinnesbichler, A. Ebberg, A. Felder, and R. Weigel, “Generation of High-Speed

Pseudorandom Sequences Using Multiplex Techniques,” IEEE Trans. Microwave The-
ory Tech., vol. 44, pp. 2738–2742, Dec. 1996.
[12] W.-Z. Chen and G.-S. Huang, “A Parallel Multi-pattern PRBS Generator and BER Tester
for 40+ Gbps Serdes Applications,” in Proceedings of the 2004 IEEE Asia-Pacific Con-
ference on Advanced Systen Integrated Circuits, Aug. 2004, pp. 318–321.
[13] A. N. Van-Luyn, “Shift Register Connections for Delayed Versions of m-Sequences,”

Electronics Letters, vol. 14, pp. 713–715, Oct. 1978.
[14] J. J. O’Reilly, “Series-Parallel Generation of m-Sequences,” The Radio and Electronic

Engineer, vol. 45, pp. 171–176, Apr. 1975.
[15] S. Kim, M. Kapur, M. Meghelli, A. Pylyakov, Y. Kwark, and D. Friedman, “45-Gb/s

SiGe BiCMOS PRBS Generator and PRBS Checker,” in Proc. IEEE Custom Integrated
Circuits Conference (CICC’2003), 2003, pp. 313–316.
[16] R. Westcott, “Testing Digital Data Transmission Systems,” United Kingdom Patent
1 281 390, 12, 1972.
[17] B. Razavi, Design of Integrated Circuits for Optical Communications. New York, NY:
McGraw-Hill, 2003.
[18] M. Laurens, B. Martinet, O. Kermarrec, Y. Campidelli, F. Deleglise, D. Dutarte, G. Troil-

lard, D. Gloria, J. Bonnouvrier, R. Beerkens, V. Rousset, F. Leverd, A. Chantre,
and A. Monroy, “A 150GHz fT /fmax 0.13µm SiGe:C BiCMOS technology,” in Bipo-
lar/BiCMOS Circuits and Technology Meeting, 2003. Proceedings of the 2003 Meeting,
Toulouse, 2003, pp. 199–202.
[19] E. Laskin and S. P. Voinigescu, “A 60 mW per Lane, 4 × 23-Gb/s 27 −1 PRBS Generator,”

in 2005 IEEE Compound Semiconductor Integrated Circuit Symposium, Palm Springs,
CA, Oct. 2005, pp. 192–195.
Bibliography 71
[20] Y. Amamiya, Y. Suzuki, J. Yamaraki, A. Fujihara, S. Tanaka, and H. Hida, “1.5-V Low
Supply Voltage 43-Gb/s Delayed Flip-Flop Circuit,” in Gallium Arsenide Integrated Cir-
cuit (GaAs IC) Symposium, 2003. 25th Annual Technical Digest 2003. IEEE, Nov. 2003,
pp. 169–172.
[21] D. Kucharski and K. Kornegay, “A 40Gb/s 2.5V 27 − 1 PRBS Generator in SiGe Using
a Low-Voltage Logic Family,” in 2005 IEEE ISSCC Digest of Technical Papers, San
Francisco, CA, Feb. 2005, pp. 340–341.
[22] H. Knapp, M. Wurzer, W. Perndl, K. Aufinger, T. F. Meister, and J. Bock, “100-Gb/s 27 −1

and 54-Gb/s 211 − 1 PRBS generators in SiGe bipolar technology,” in IEEE Compound
Semiconductor IC Symposium, Monterey, CA, Oct. 2004, pp. 219–222.
[23] K. Kanda, D. Yamazaki, T. Yamamoto, M. Horinaka, J. Ogawa, H. Tamura, and H. On-

odera, “40Gb/s 4:1 MUX/1:4 DEMUX in 90nm Standard CMOS,” in 2005 IEEE ISSCC
Digest of Technical Papers, San Francisco, CA, Feb. 2005, pp. 152–153.
[24] S. P. Voinigescu, D. S. McPherson, F. Pera, S. Szilagyi, M. Tazlauanu, and H. Tran, “A

Comparison of Silicon and III-V Technology Performance and Building Block Imple-
mentations for 10 and 40 Gb/s Optical Networking ICs,” International Journal of High
Speed Electronics and Systems, vol. 13, Mar. 2003.
[25] M. Green, “Current-controlled CMOS circuits with inductive broadbanding,” U.S. Patent
06 525 571, Feb. 25, 2003.
[26] H.-M. Rein, “Si and SiGe Bipolar ICs for 10 to 40 Gb/s Optical-Fiber TDM Links,”
International Journal of High Speed Electronics and Systems, vol. 9, pp. 347–384, 1998.
[27] S. P. Voinigescu, T. Dickson, R. Beerkens, I. Khalid, and P. Westergaard, “A Compar-

ison of Si CMOS, SiGe BiCMOS, and InP HBTs Technologies for High-Speed and
Millimeter-wave ICs,” in Si Monolithic Integrated Circuits in RF Systems, Atlanta, GA,
2004, pp. 111–114.
[28] A. S. Sedra and K. C. Smith, Microelectronic Circuits, 5th ed. New York, NY: Oxford
Press.
[29] T. O. Dickson, M.-A. LaCroix, S. Boret, D. Gloria, R. Beerkens, and S. P. Voinigescu,

“30-100-GHz Inductors and Transformers for Millimeter-Wave (Bi)CMOS Integrated
Circuits,” IEEE Transactions on Microwave Theory and Techniques, vol. 53, pp. 123–
133, Jan. 2005.
72 Bibliography
[30] P. Chevalier, C. Fellous, L. Rubaldo, D. Dutartre, M. L. T. Jagueneau, F. Leverd, S. Bord,

C. Richard, D. Lenoble, J. Bonnouvrier, M. Marty, A. Perrotin, D. Gloria, F. Saguin,
B. Barbalat, R. Beerkens, N. Zerounian, F. Aniel, and A. Chantre, “230 GHz self-aligned
SiGeC HBT for 90 nm BiCMOS technology,” in Bipolar/BiCMOS Circuits and Technol-
ogy, 2004. Proceedings of the 2004 Meeting, 2004, pp. 225–228.
[31] K. Martin, Digital Integrated Circuit Design. New York, NY: Oxford University Press,
Inc., 2000.
[32] S. P. Voinigescu, T. O. Dickson, T. Chalvatzis, A. Hazneci, E. Laskin, R. Beerkens, and

I. Khalid, “Algorithmic design methodologies and design porting of wireline transceiver
ic building blocks between technology nodes,” in IEEE Custom Integrated Circuits Con-
ference, 2005.
[33] M. Duelk, “Next Generation 100G Ethernet,” in 31 European Conference on Optical

Communication, Glasgow, Scotland, 2005.
A Appendix
A.1. MATLAB Code
function mux_delays(m, phi, fname)

% mux_delays
% uses the function srconns() to find all possible shift register
% tap combinations to generate m equally-spaced phases of a PRBS
% that can be used as inputs to a m-to-(m/2) mux.
% it also writes the results to file.
% m - the number of equally spaced phase shifts to find
% phi - an array that contains the numbers of the taps of the
% shift register that generate the feedback polynomial,
% e.g., for 2ˆ7-1 PRBS: phi = [0, 6, 7].
% fname - name of the file where to write the results
%
% find the total number of possible delays

delays = 2ˆmax(phi);
fid = fopen(fname, ’w’);
% loop over the possible combinations

for d = 0:(delays/m)
disp(sprintf(’combination # %d’, d));
fprintf(fid, ’combination # %d\n’, d);
% find the m delays for that combination
for i = 0:(delays/m):delays-1
[b, Rd] = srconns(phi, d+i);
disp(sprintf(’\t%d delay (%d):\t%s’, i, d+i, char(Rd)));
fprintf(fid, ’\t%d delay (%d):\t%s\n’, i, d+i, char(Rd));
73
74 Appendix
end
end
fclose(fid);
function [b, Rd] = srconns(phi, d)

% implementation of the algorithm described in
% A. N. Van Luyn, "Shift Register Connections for
% Delayed Versions of m-Sequences", Electronics
% Letters, Vol 14, Oct 1978.
%
% phi - an array that contains the numbers of the switches
% to generate the feedback polynomial, e.g.,
% for 2ˆ7-1 PRBS: phi = [0, 6, 7].
% d - the required phase shift to be generated, e.g.,
% for 2ˆ7-1 PRBS, d is a number between 1 and 127.
% b - an array that indicates the shift register connections
% required to produce a sequence with the delay d.
% Rd - returns the same information as b, but in polunomial
% form
% form the feedback polynomial:

x = sym(’x’);
p = sum(x.ˆphi);
% find s - the highest power of 2 in the binary

% representation of d
a = fliplr(dec2bin(d));
s = length(a);
% initialization:
i = s;
Rd = xˆ0;
% execution
while i > 0
Rd = xˆa(i) * Rdˆ2;
Rd = expand(Rd);
Rd = poly2sym(mod(sym2poly(Rd), 2)) % mod2 addition
[Q, R] = deconv(sym2poly(Rd), sym2poly(p));
Rd = poly2sym(mod(R, 2))
i = i - 1
end
A.2 Simulink Schematics 75
% correction 1
[Q, R] = deconv(sym2poly(Rd), sym2poly(p));
Rd1 = poly2sym(mod(R, 2));
b1 = mod(sym2poly(Rd1), 2);
% correction 2
b2 = mod(sym2poly(Rd + p), 2);
Rd2 = expand(poly2sym(b2));
b = b1;
Rd = Rd1;
if sum(b1) > sum(b2)
b = b2;
Rd = Rd2;
end
A.2. Simulink Schematics
boolean signal1
Clock
0/8 delay Clock
1/8 delay Error Error

Error
Input
2/8 delay
3/8 delay Regular Checker

Clock
4/8 delay 2/8 delay
5/8 delay
6/8 delay
7/8 delay XOR signal4

[0 2 4 6 8 10 12 14 16]
Series Generator
boolean signal5 double
Scope
Clock
Error signal6
XOR Input
0/8 delay
1/8 delay My Checker

2/8 delay 2/8 delay
3/8 delay
Clock
4/8 delay
5/8 delay signal8

6/8 delay
7/8 delay
Parallel Generator 7/8 delay
Figure A.1: Simulink test bench for the PRBS generator and checker
76 Appendix
XOR 8
7/8 delay
7/8 delay
to produce 8 phases with a series 2^7-1 prbs: 6/8 delay

need at least 11 2-input XOR gates, 7
possibly retiming 6/8 delay
XOR 6
5/8 delay
5/8 delay
XOR 5
4/8 delay
XOR 4/8 delay
XOR
6/8 delay XOR 4
3/8 delay
3/8 delay
XOR XOR 3
2/8 delay
2/8 delay
XOR 2
1/8 delay
XOR 1/8 delay
6/8 delay
0/8 delay
1
NOR 0/8 delay
XOR In0
In1 Out D
Q D
Q D
Q D
Q D
Select CLK Q D
CLK Q D
CLK Q
CLK 0/8 delay
Mux !Q CLK
!CLR !Q CLK
!CLR !Q CLK
!CLR !Q
!CLR !Q
DFF1 !CLR !Q
DFF2 !CLR !Q
DFF3 !CLR
DFF4 DFF5
DFF6 DFF7
1 boolean
Clock
Clock
(1) boolean
Figure A.2: A series 27 − 1 PRBS generator in Simulink

A.2 Simulink Schematics 77
XOR D
Q 1
0/8 delay
0/8 delay
CLK
!Q
!CLR
DFF1
XOR D
Q 2
1/8 delay
1/8 delay
CLK
!Q
!CLR
DFF2
XOR D
Q 3
2/8 delay
2/8 delay
CLK
!Q
!CLR
DFF3
XOR D
Q 4
3/8 delay
3/8 delay
CLK
!Q
!CLR
DFF4
XOR D
Q 5
4/8 delay
4/8 delay
CLK
!Q
!CLR
DFF5
XOR D
Q 6
5/8 delay
5/8 delay
CLK
!Q
NOR !CLR
DFF6
XOR D
Q 7
6/8 delay
In0 6/8 delay
CLK
In1 Out !Q
!CLR
Select DFF7
Mux
XOR D
Q 8
7/8 delay
7/8 delay
CLK
!Q
!CLR
1 boolean
Clock DFF8
Clock
(1) boolean
Figure A.3: A parallel 27 − 1 PRBS generator in Simulink

78 Appendix
XOR 1
Error
XOR D
Q D
Q D
Q 2 D
Q D
CLK Q D
CLK Input Q D
CLK Q
CLK
!Q CLK
!CLR !Q CLK
!CLR !Q CLK
!CLR !Q
!CLR !Q
DFF1 !CLR !Q
DFF2 !CLR !Q
DFF3 !CLR
DFF4
DFF5 DFF6
DFF7
1 boolean
Clock
Clock
(1) boolean
Figure A.4: A regular 27 − 1 PRBS checker in Simulink
XOR
2 D
Q D XOR D
Input Q D Q
Q D
Q D
CLK Q D
CLK Q D CLK
CLK Q
CLK
!Q CLK
!CLR !Q CLK !Q
!CLR !Q CLK !CLR
!CLR !Q
!CLR !Q
DFF1 !CLR !Q
DFF2 !CLR !Q DFF8
DFF3 !CLR
DFF4
DFF5
DFF6
DFF7
1 boolean
Clock AND 1
Clock
Error
(1) boolean
XOR
D XOR D
Q D Q
Q D
Q D
Q D
CLK Q D CLK
CLK Q
CLK
CLK
!Q CLK !Q
!CLR !Q CLK !CLR
!CLR !Q
!CLR !Q
!CLR !Q
DFF9 DFF10
!CLR
!CLR
!Q DFF15
DFF11
DFF12
DFF13
DFF14
Figure A.5: The implemented 27 − 1 PRBS checker in Simulink

A.3 Verilog Code 79
A.3. Verilog Code
‘timescale 1ps/10fs
/*
parallel architecture with 8 DFFs and 8 XORs (all outputs are retimed)
All xor gates have fanout of 1, 2 of the DFFs have fanout of 3
includes gate delays
*/
///////////////////// random testing stuff ///////////////////////////////
///////////////////////////// generator //////////////////////////////////
module gen;
reg Clock, Clock2, Clock4, vdd, gnd; // inputs to the circuit
wire [2:0] out; // the fast output
wire [15:0] q; // outputs of latches
wire [6:0] mux_out; // outputs of multiplexers
wire [7:0] xor_out; // outputs of xor gates
//wire [1:0] buf_out; // outputs from the delay buffers
wire [14:0] clk_buf; // outputs of the clock buffers (the clock tree)
wire [7:0] clk_buf2;
wire [7:0] not_out; // outputs of not gates to invert the clock (no delay)
wire not_out2;
wire [7:0] temp; // inverters
wire [9:0] retimer; // retiming latches after first stage of multiplexing
assign out[2] = mux_out[6]; // the fastest output

assign out[1] = mux_out[4]; // the medium output
assign out[0] = /*mux_out[0]*/retimer[1]; // the slowest output
// instantiation of components
// clock tree :
BUF_d0 cb0(clk_buf[0], Clock);
80 Appendix
/*BUF_d0 cb0(clk_buf[0], clk_buf[8]);

BUF_d0 cb1(clk_buf[1], clk_buf[8]);
BUF_d0 cb14(clk_buf[14], Clock);*/
not(not_out[0], clk_buf[0]);
not(not_out2, Clock2);
// the shift register :

LCH_d1 latch0(q[0], not_out[0], xor_out[0]);
LCH_d1 latch1(q[1], Clock, q[0]);
A.3 Verilog Code 81

/*LCH_d0 latch0(q[0], not_out[0], xor_out[0]);

LCH_d0 latch1(q[1], clk_buf[0], q[0]);
LCH_d0 latch15(q[15], clk_buf[7], q[14]);*/
// xor wiring :
not(temp[0], q[1]);
not(temp[1], q[3]);
not(temp[2], q[5]);
not(temp[3], q[7]);
not(temp[4], q[9]);
not(temp[5], q[11]);
XNOR_d3 xnor0(xor_out[0], q[3], q[5]);

XNOR_d1 xnor6(xor_out[6], q[15], xor_out[0]);
XNOR_d1 xnor7(xor_out[7], xor_out[0], xor_out[1]);
/*XNOR_d3 xnor0(xor_out[0], q[3], q[5]);

82 Appendix

XNOR_d1 xnor7(xor_out[7], q[3], q[7]);*/
// mux wiring :
SEL_d1 mux0(mux_out[0], temp[0], temp[4], Clock);
SEL_d0 mux4(mux_out[4], retimer[1], retimer[4], Clock2);
SEL_d0 mux5(mux_out[5], retimer[6], retimer[9], Clock2);
SEL_d0 mux6(mux_out[6], mux_out[4], mux_out[5], Clock4);
LCH_d1 latch16(retimer[0], Clock2, mux_out[0]);

LCH_d1 latch17(retimer[1], not_out2, retimer[0]); // MS
LCH_d1 latch19(retimer[3], not_out2, retimer[2]);
LCH_d1 latch20(retimer[4], Clock2, retimer[3]); // MSM
LCH_d1 latch22(retimer[6], not_out2, retimer[5]); // MS
LCH_d1 latch24(retimer[8], not_out2, retimer[7]);
LCH_d1 latch25(retimer[9], Clock2, retimer[8]); // MSM
initial begin
Clock = 1;
Clock2 = 0;
Clock4 = 0;
vdd = 1;
gnd = 0;
end
always #60 Clock = !Clock; // generate the clock
always #30 Clock2 = !Clock2; // double the speed
always #15 Clock4 = !Clock4; // 4 times the speed
initial begin
$shm_open("signals.shm"); // waveforms
$shm_probe(Clock, out, q, mux_out, xor_out, temp, retimer, not_out);
$shm_probe(clk_buf, Clock2, Clock4, clk_buf2);
end
initial begin
$display("Time Clock in0 in1 sel m0 q1 q2 q3 q4 q5 q6 q7 slow1 slow2 fast");
$monitor("%4g",$time,,,,Clock,,,,,xor_out[0],,,,vdd,,,,
A.3 Verilog Code 83
mux_out[0],,,q[1],,,q[2],,,q[3],,,q[4],,,q[5],,,
q[6],,,q[7],,,,q[8],,,,,,q[3],,,,,,out);
$dumpvars;
#10000 $shm_close();
$finish;
end
endmodule
//////////////////////////// LATCH //////////////////////////////////////////
module LCH_d3(Q, clk, D); // latch with delays (fanout=3)

output Q;
input clk, D;
latch(Q, clk, D); // instantiate a latch
specify // timing specifications

specparam tPLH_clk_Q = 0;//22; // rising Q after clk edge
specparam tPHL_clk_Q = 0;//22; // falling Q after clk edge
specparam tPLH_D_Q = 0;//22;
specparam tPHL_D_Q = 0;//22;
(clk => Q) = (tPLH_clk_Q, tPHL_clk_Q);
(D => Q) = (tPLH_D_Q, tPHL_D_Q);
endspecify
endmodule

output Q;
input clk, D;

endspecify
endmodule

84 Appendix
output Q;
input clk, D;

endspecify
endmodule
module LCH_d0(Q, clk, D); // latch without delays

output Q;
input clk, D;
endmodule
primitive latch(Q, clk, D) ;

output Q; reg Q;
input clk, D;
initial Q = 0; // <-------- !!!
table
// clock data : q : q+
0 1 : ? : 1 ;
0 0 : ? : 0 ;
1 ? : ? : - ; // - = no change
endtable
endprimitive
//////////////////////////// XNOR //////////////////////////////////////////
module XNOR_d1(c, a, b); // XNOR gate with delays (fanout=1)

output c;
input a, b;
xnor(c, a, b);
A.3 Verilog Code 85
specify
specparam tPLH_a_c = 0;//16;
specparam tPHL_a_c = 0;//16;
specparam tPLH_b_c = 0;//16;
specparam tPHL_b_c = 0;//16;
(a => c) = (tPLH_a_c, tPHL_a_c);
(b => c) = (tPLH_b_c, tPHL_b_c);
endspecify
endmodule
output c;
input a, b;
xnor(c, a, b);
specify
endspecify
endmodule
output c;
input a, b;
xnor(c, a, b);
specify
endspecify
endmodule
module XNOR_d0(c, a, b); // XNOR gate without delays
output c;
input a, b;
xnor(c, a, b);
86 Appendix
endmodule
///////////////////////////// SEL /////////////////////////////////////////
module SEL_d1(out, in0, in1, sel); // SEL with delays (fanout=1)

output out;
input in0, in1, sel;
MUX(out, in0, in1, sel);
specify
specparam tPLH_in0_out = 0;//16;
specparam tPHL_in0_out = 0;//16;
specparam tPLH_sel_out = 0;//16;
specparam tPHL_sel_out = 0;//16;
(in0 => out) = (tPLH_in0_out, tPHL_in0_out);
(sel => out) = (tPLH_sel_out, tPHL_sel_out);
endspecify
endmodule
module SEL_d2(out, in0, in1, sel); // SEL with delays (fanout=2)

output out;
specify
specparam tPLH_sel_out = 0;//8;
specparam tPHL_sel_out = 0;//8;
(sel => out) = (tPLH_sel_out, tPHL_sel_out);
endspecify
endmodule
module SEL_d0(out, in0, in1, sel); // SEL without delays

A.3 Verilog Code 87
output out;
endmodule
primitive MUX(out, in0, in1, sel);

output out; //reg out; // <-------- !!!
//initial out = 0; // <-------- !!!
table
// in0 in1 sel : out
1 ? 0 : 1 ; // ? = 0,1,x
0 ? 0 : 0 ;
? 1 1 : 1 ;
? 0 1 : 0 ;
0 0 x : 0 ;
1 1 x : 1 ;
endtable
endprimitive
///////////////////////////// BUF /////////////////////////////////////////
module BUF_d1(out, in); // buffer with delay (has same delay as mux) (fanout=1)
output out;
input in;
buf(out, in);
specify
specparam tPLH_in_out = 0;//6;
specparam tPHL_in_out = 0;//6;
(in => out) = (tPLH_in_out, tPHL_in_out);
endspecify
endmodule
module BUF_d2(out, in); // buffer with delay (has same delay as mux) (fanout=2)
output out;
input in;
88 Appendix
buf(out, in);
specify
specparam tPLH_in_out = 0;//2.0;
specparam tPHL_in_out = 0;//2.0;
(in => out) = (tPLH_in_out, tPHL_in_out);
endspecify
endmodule
module BUF_d0(out, in); // buffer without delays

output out;
input in;
//assign out = in;

buf(out, in);
endmodule

Laskin MASc Thesis PDF

Uploaded by

Copyright:

Available Formats

Laskin MASc Thesis PDF

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

Laskin MASc Thesis PDF

Uploaded by

Copyright:

Available Formats

O N -C HIP S ELF -T EST C IRCUIT B LOCKS FOR

H IGH -S PEED A PPLICATIONS

A THESIS SUBMITTED IN CONFORMITY WITH THE REQUIREMENTS

Master of Applied Science, 2006

List of Tables vii

2 PRBS Generator and Checker Architectures 5

3 27 -1 PRBS Generator and Checker Chip Design 25

3.2 Design of Building Blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5 Comparison and Power Optimization of High-Speed Logic Topologies 49

2.1 All 18 possible characteristic polynomials for a 27 − 1 PRBS . . . . . . . . . . 9

4.1 Performance summary of the PRBS generator at different bit rates . . . . . . . 43

5.1 Model parameters for 0.13 µm and 90 nm SiGe BiCMOS technologies . . . . . 59

1.1 Use of PRBS generator and checker for testing . . . . . . . . . . . . . . . . . 2

2.1 Diagram of a feedback shift register . . . . . . . . . . . . . . . . . . . . . . . 6

3.1 Top-level schematic of the designed chip . . . . . . . . . . . . . . . . . . . . . 26

3.12 System simulation after extraction with 11-GHz clock . . . . . . . . . . . . . . 37

4.1 Die photo of the fabricated chip . . . . . . . . . . . . . . . . . . . . . . . . . 40

5.1 CML digital gates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

BER Bit-Error Rate

1.1. Built-In Self-Test for High-Speed Transceivers

1.2. Speed and Power Optimization of High-Speed Logic

(a) Conceptual setup for BER test of a device

MUX DRV TIA LA CDR DEMUX

(b) Use of a PRBS for system BER testing

Figure 1.1: Use of PRBS generator and checker for testing

1.3. Objective of Thesis

1.4. Contributions from this Work

2.1. Definition and Properties of Pseudo-Random Bit Sequences

g(x1 , x2 , ...xn ) = c1 x1 ⊕ c2 x2 ⊕ c3 x3 ⊕ ... ⊕ cn xn (2.1)

1. A balance of 1 and 0 terms.

2. Two runs of length n for each run of length n + 1.

3. A two-level auto-correlation function.

D Q g(x1, x2, … , xn)

Figure 2.1: Diagram of a feedback shift register

Figure 2.2: A 3-stage LFSR and one of its maximal-length sequences

Row Characteristic Polynomial Pseudo-Random Bit Sequence

Table 2.1: All 18 possible characteristic polynomials for a 27 − 1 PRBS

positive integer). Decimation of any PRB-sequence forms another PRB-sequence. However,

By using equations 2.6 and 2.7 together with f (D){ak } = 0, we obtain

f (D){a2k } = f (D 2){ak } = [f (D){ak }]2 = f (D)[f (D){ak }] = f (D){0} = 0. (2.8)

2.2. Series PRBS Generators

Shift Register Length n Characteristic Polynomial PRBS Length

Table 2.2: Standard characteristic polynomials for various PRBS lengths

Figure 2.3: A series generator for a 27 − 1 PRBS

d– required phase shift in bits d = 1·20 + 0·21 + 0·22 + 1·23 + 1·24

2.3. Parallel PRBS Generators

Consequently, the state of the generator after k clock cycles will be

U(j + k) = Tk · U(j). (2.10)

Figure 2.5: A 25 − 1 series PRBS generator with its transition matrix

(b) 25 − 1 PRBS generator with 3 parallel outputs

(c) 25 − 1 PRBS generator with 4 parallel outputs

out 8 out 7 out 6 out 5 out 4 out 3 out 2 out 1

(a) 27 − 1 PRBS generator with 8 parallel outputs

(b) 231 − 1 PRBS generator with 8 parallel outputs

Figure 2.7: Examples of practical parallel PRBS generators

2.4. PRBS Checker Design

Figure 2.8: A simple PRBS checker

Figure 2.9: A 27 − 1 PRBS checker

No Errors Remote Errors or Close Errors