MS Thesis ZhinengZhu

download MS Thesis ZhinengZhu

of 81

Transcript of MS Thesis ZhinengZhu

  • 7/29/2019 MS Thesis ZhinengZhu

    1/81

    LOW NOISE LOW OFFSET OPERATIONAL AMPLIFIER FOR

    NANOPORE-BASED GENE SEQUENCER

    By

    Zhineng Zhu

    B.S. Hefei University of Technology, 1998

    A THESIS

    Submitted in Partial Fulfillment of the

    Requirements for the Degree of

    Master of Science

    (in Electrical Engineering)

    The Graduate School

    The University of Maine

    May, 2007

    Advisory Committee:

    David E. Kotecki, Associate Professor of Electrical and Computer Engineering,

    Advisor

    Donald M. Hummels, Professor of Electrical and Computer Engineering

    Rosemary Smith, Professor of Electrical and Computer Engineering

  • 7/29/2019 MS Thesis ZhinengZhu

    2/81

    LIBRARY RIGHTS STATEMENT

    In presenting this thesis in partial fulfillment of the requirements for an advanced

    degree at The University of Maine, I agree that the Library shall make it freely availablefor inspection. I further agree that permission for fair use copying of this thesis for

    scholarly purposes may be granted by the Librarian. It is understood that any copying

    or publication of this thesis for financial gain shall not be allowed without my written

    permission.

    Signature:

    Date:

  • 7/29/2019 MS Thesis ZhinengZhu

    3/81

    LOW NOISE LOW OFFSET OPERATIONAL AMPLIFIER FOR

    NANOPORE-BASED GENE SEQUENCER

    By Zhineng Zhu

    Thesis Advisor: Dr. David E. Kotecki

    An Abstract of the Thesis Presented

    in Partial Fulfillment of the Requirements for the

    Degree of Master of Science

    (in Electrical Engineering)

    May, 2007

    A nanopore-based gene sequencer generates a modulated current signal with the

    expected magnitude to be in the range of several pico-amps to a nano-amp, and the

    bandwidth up to 10MHz. To detect such a weak signal, an ultra-low noise, low offset,

    high precision CMOS operational amplifier is designed in a 0.35m CMOS process.

    Most of the literature on low noise amplifier design emphasizes flicker noise

    reduction. This research analyzed and optimized a proposed differential pair and opera-

    tional amplifier, with both the flicker noise and the thermal noise minimized simultane-

    ously. The size of each MOSFET is optimized, making the input pair the only dominant

    noise source. The input pairs bias current is maximized to reduce the thermal noise.

    Also this bias current is rationed between the active current source load and the current

    mirror load. This ration of bias current makes the optimization of the amplifiers noise

    performance and DC gain separated, which is often a trade-off in normal operational

    amplifier design. An operational amplifier with the input-referred voltage noise Power

    Spectral Density (PSD) of less than 2nV/

    Hzfor frequencies above 1MHz is realized.

    The DC gain of this amplifier is up to 120dB. The extra bias current and the cascode

    structure also result in a high speed design: the unity gain bandwidth of this operational

  • 7/29/2019 MS Thesis ZhinengZhu

    4/81

    amplifier is more than 200MHz with 20pF loading, providing a sufficient close loop gain

    in the MHz range.

    CMOS operational amplifiers generally have a higher offset voltage than bipolar

    ones. In this design, a digital offset trimming method is studied and used to cover 10mV

    input-referred offset. Its effect on noise performance is minimized. This method is

    suitable for low voltage and continuous mode applications.

    Power Supply Rejection Ratio (PSRR) is an important parameter for the low

    noise operational amplifier. The PSRR limitation of the taped-output operational am-

    plifier with common source output stage is analyzed. An operational amplifier with

    cascode compensation scheme is studied, which shows the potential improvement in

    PSRR.

  • 7/29/2019 MS Thesis ZhinengZhu

    5/81

    ACKNOWLEDGMENTS

    A special thanks to my wife, Ming Luan, for her constant support and encour-

    agement.I would like to express my sincere gratitude to Professor Kotecki for his guidance

    and assistance throughout the course of this research work. Thanks to Professor Collins

    and Professor Smith for their advices and support. Thanks to Professor Hummels for

    the time and reviewing my thesis. Thanks to Steve for his kindly help in the CAD and

    chip testing. Thanks to Sanjeev, Alma and Bingxin for their discussions and happy time

    in the past two years.

    This work was supported by the National Institutes of Health, Bethesda, MD,

    under contract 5R01HG003565-03.

    ii

  • 7/29/2019 MS Thesis ZhinengZhu

    6/81

    TABLE OF CONTENTS

    ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. ii

    LIST OF TABLES. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . v

    LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . vi

    Chapter

    1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 1

    1.1. Background . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . 1

    1.2. Purpose of this Research. . . . .. . . . .. . . . .. . . . . .. . . . .. . . . . .. . . . .. . . . .. . . . .. . . . . .. 2

    1.3. Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

    2. Noise in MOSFETs and Basic CMOS Operational Amplifiers. . .. . .. . .. . .. . .. . .. . 6

    2.1. Noise Sources in MOSFETs. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . . 6

    2.1.1. Thermal noise in the channel . . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . . 6

    2.1.2. Flicker noise in the channel. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . . .. . . . .. 7

    2.1.3. Noise of polysilicon gate resistance . . . . .. . . . .. . . . . .. . . . .. . . . . .. . . . .. 8

    2.1.4. Noise from the distributed substrate resistance . . . . . . . . . . . . . . . . . . . . . 10

    2.2. Noise Performance of the CMOS Differential Pair . . . . . . . . . . . . . . . . . . . . . . . . . 11

    3. Novel Structure for a Low Noise Operational Amplifier using MOSFETs . . . . .. . 18

    3.1. Noise Analysis of New Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 203.1.1. The noise contribution of the input stage M1,2 . . . . . . . . . . . . . . . . . . . . . 213.1.2. The noise contribution of the active current source load M3,4 . . . . . 213.1.3. The noise contribution of current mirror load M7,8 . . . . . . . . . . . . . . . . 223.1.4. The noise contribution of cascode stage M5,6 . . . . . . . . . . . . . . . . . . . . . . 223.1.5. The noise contribution of current bias MOSFET M0 . . . . . . . . . . . . . . 233.1.6. Total output noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . 24

    3.2. Noise Reduction Techniques for this New Structure . . . . . . . . . . . . . . . . . . . . . . . . 25

    3.2.1. Determination of the input pair type for M1,2 . . . . . . . . . . . . . . . . . . . . . . 253.2.2. Optimization of the bias current ID2 of the input pair . . . . . . . . . . . . . . 263.2.3. Optimization of the sizes and aspect ratios of MOSFETs. . . . . . . . . . 26

    3.2.4. Noise performance of the folded-cascode differential pair. . . . . . . . . 293.3. DC Gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . 29

    3.4. AC Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 31

    3.5. Bias Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . 35

    3.6. Offset Reduction and Auxiliary Port . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 35

    iii

  • 7/29/2019 MS Thesis ZhinengZhu

    7/81

    4. Testing Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . 47

    4.1. Tape-out . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . 47

    4.2. Noise Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

    4.3. PCB Board Design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

    4.4. Measured Results . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 50

    4.4.1. Input Common Mode Range . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 504.4.2. Output Voltage Swing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . 54

    4.4.3. Transient Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . 54

    4.4.4. AC Performance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . 55

    5. PSRR of Low Noise Operational Amplifier . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 59

    5.1. Modified Output Stage. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

    5.2. PSRR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 59

    6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 64

    6.1. Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . 64

    6.2. Future Work. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . 65

    REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 68

    BIOGRAPHY OF THE AUTHOR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . 70

    iv

  • 7/29/2019 MS Thesis ZhinengZhu

    8/81

    LIST OF TABLES

    Table 3.1. Optimized devices size of the two-stage operational ampli-

    fier . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 31

    Table 3.2. MOSFETsaspect ratio in the bias circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 36

    Table 4.1. Operational amplifiers or their first stage included in the

    chip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . 49

    v

  • 7/29/2019 MS Thesis ZhinengZhu

    9/81

    LIST OF FIGURES

    Figure 1.1. Nanopore gene sequencing . . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . .. . . . . .. . . . .. 2

    Figure 1.2. Typical input-referred voltage noise PSD for a MOSFET . .. . . . . .. . . . . . 4

    Figure 2.1. MOSFET noise model: a) Model of thermal noise in the

    channel, b) Input-referred voltage noise model . . . . .. . . . .. . . . .. . . . .. . . . . 7

    Figure 2.2. Reduction of gate noise through layout, a) single-finger MOS-

    FET, b) single-finger MOSFET with contact at both ends, c)

    multiple-finger MOSFET and d) multiple-finger MOSFET

    with contacts at both ends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . 10

    Figure 2.3. Contribution mechanism of substrate noise to the drain current. . . . . . . . 11

    Figure 2.4. a) Differential pair with current mirror load, b) noise model

    of a) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . 12

    Figure 2.5. Two stage operational amplifier with noise sources . . . . . . . . . . . . . . . . . . . . 14

    Figure 2.6. Four basic CMOS differential pairs, a) resistive load, b)

    diode-connected load, c) active current source load and d)

    folded-cascode differential pair . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 16

    Figure 3.1. Proposed structure of the low noise differential pair . . . . . . . . . . . . . . . . . . . 19

    Figure 3.2. Noise model of differential pair in Figure 3.1 . . . . . . . . . . . . . . . . . . . . . . . . . . 20

    Figure 3.3. Small-signal noise model for input stage M1,2 . . . . . . . . . . . . . . . . . . . . . . . . . 21

    Figure 3.4. Small-signal noise model for current mirror load M7,8 . . . . . . . . . . . . . . . . 22

    Figure 3.5. Small-signal noise model for the cascode stage M5,6 . . . . . . . . . . . . . . . . . . 22

    Figure 3.6. Small-signal noise model for bias current source M0 . . . . . . . . . . . . . . . . . . 24

    Figure 3.7. Optimum gate length of the input pair . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . 27

    Figure 3.8. The noise performace of the circuit in Figure 3.1 . . . . . . . . . . . . . . . . . . . . . . 27

    Figure 3.9. The noise performace of the circuit with PMOS as the input

    pair . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 28

    Figure 3.10. Modified structure of the operational amplifier with an aux-iliary port . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 30

    Figure 3.11. Two stage low noise low offset operational amplifier . . . . . . . . . . . . . . . . . . 31

    Figure 3.12. Small signal model of the circuit in Figure 3.11 . . . . . . . . . . . . . . . . . . . . . . . . 32

    Figure 3.13. AC performance with parasitic resistances and capacitances. . . . . . . . . . . 34

    Figure 3.14. Noise performance with parasitic resistances and capacitances . . . . . . . . 34

    vi

  • 7/29/2019 MS Thesis ZhinengZhu

    10/81

    Figure 3.15. Self-biasing Vthreshold reference with start-up circuit . . . . . . . . . . . . . . . . . . . 35

    Figure 3.16. Schematic of entire bias circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 36

    Figure 3.17. Offset tuning scheme with operational amplifier in open

    loop mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 39

    Figure 3.18. Offset tuning scheme with operational amplifier inverting-

    configured . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 40

    Figure 3.19. Setup for auxiliary input design . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 40

    Figure 3.20. DC sweep of the auxiliary input . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 41

    Figure 3.21. Noise level with and without auxiliary input . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

    Figure 3.22. Diagram of offset tuning block . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 42

    Figure 3.23. Connection between counter and DAC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 43

    Figure 3.24. Timing of offset tuning . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . 45

    Figure 3.25. Simulation of offset tuning: a) Switching point, b) DAC

    output, c) Calibration output, d) Operational amplifier output . . . . . . . . . 46

    Figure 4.1. Microphotograph of the op amp . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 48

    Figure 4.2. Microphotograph of the chip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 48

    Figure 4.3. Diagram of the noise measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . 49

    Figure 4.4. The diagram of the PCB board . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 51

    Figure 4.5. Bias current generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . 51

    Figure 4.6. Bias voltage generator . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . 51

    Figure 4.7. PCB Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 52

    Figure 4.8. PCB Board . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 53

    Figure 4.9. Common mode input range . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . 54

    Figure 4.10. Sinewave input test setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . 55

    Figure 4.11. Measured sine wave input: a)Input=1kHZ, Gain=50, b) In-

    put=1kHz, Gain=3, and c)Input=1MHz, Gain=3 . . . . . . . . . . . . . . . . . . . . . . . 56

    Figure 4.12. AC gain test setup . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

    Figure 4.13. AC gain . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . 58

    Figure 5.1. AC performance with 20pF load capacitance . . . . . . . . . . . . . . . . . . . . . . . . . . 60

    Figure 5.2. New structure with cascode compensation . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 61

    Figure 5.3. PSRR simulation results: a) Positive PSRR, b) Negative

    PSRR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . 62

    vii

  • 7/29/2019 MS Thesis ZhinengZhu

    11/81

    Figure 5.4. Cascode-compensated operational amplifiers AC performance

    . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . 62

    Figure 5.5. Noise level of the cascode-compensated operational ampli-

    fier . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 63

    viii

  • 7/29/2019 MS Thesis ZhinengZhu

    12/81

    CHAPTER 1

    Introduction

    1.1 Background

    A proposed gene sequencer generates a modulated signal based on the tunneling

    current through a single strand of DNA passing through a nanopore as illustrated in

    Figure1.1. This signal is expected to be in the range of several pico-amps to a nano-

    amp, and the bandwidth can be up to 10MHz. This corresponds to a change of 100-

    1000 electrons being detected as different base pairs pass through the nanopore. To

    detect such a weak signal, an ultra-low noise, low offset, high precision transimpedance

    amplifier is needed [1].

    There are many low noise, high precision amplifiers available. Operational am-

    plifiers designed using bipolar technology have achieved a wide bandwidth and superior

    voltage noise performance. For example, the AD8099 from Analog Devices[2] has

    about 1nV /

    Hz of input voltage noise at 10KHz, and a Gain-Bandwidth Product

    (GBP) as high as 3.8GHz. There are also designs which use the lateral pnp transis-

    tors provided in digital CMOS processes as the input stage of the amplifier to achieve

    very-low voltage noise performance[3]. However, a bipolar operational amplifier needs

    several micro amps of input current to bias the input stage. Such high bias current can in-

    troduce several pA/

    Hzof input-referred current noise. For example, the AD8099 has

    more than 2pA/

    Hznoise Power Spectral Density (PSD) at frequency up to 100MHz.

    The high input-referred current noise prevents bipolar technology from being used in

    the nanopore gene sequencer application.

    Commercially available CMOS operational amplifiers have higher voltage noise

    than bipolar amplifiers. For example, the LM6211 from National Semiconductor[4] has

    5.5nV/

    Hz input voltage noise at 10KHz. Another example is the opa725 from Texas

    1

  • 7/29/2019 MS Thesis ZhinengZhu

    13/81

    Current

    Voltage

    Tunnelling CurrentSilicon Nitride

    Silicon

    Metal

    CurrentCurrent

    VoltageVoltage

    Tunnelling CurrentSilicon Nitride

    Silicon

    Metal

    Figure 1.1: Nanopore gene sequencing

    Instruments[5], which has10nV /

    Hz

    input voltage noise at 10KHz. These CMOS

    operational amplifiers have a limited bandwidth. The GBP of the op725 is only 20MHz.

    This will not provide sufficient close-loop gain at megahertz frequencies need for the

    nanopore application. However, CMOS transistors are voltage controlled devices and

    there is almost no bias current needed for the input stage. For both the LM6211 and

    opa725, the device input bias current is far less than a nano amp and the input-referred

    current noise is less than 10f A/

    Hz at 1KHz. Therefore, to amplify a current signal

    with a magnitude of several pico amps, a CMOS device is preferred over a bipolar

    device.

    1.2 Purpose of this Research

    There are two noise sources in MOSFET devices: flicker noise (below 1MHz)

    and thermal noise. In order to distinguish the flicker noise from the thermal noise, the

    PSD of both types of noise are plotted on the same axis as shown in Figure 1.2. The

    corner frequency is where both PSDs are equal to each other. From Figure 1.2, it is

    observed that flicker noise is inversely proportional to frequency and most significant

    at low frequencies. The thermal noise is independent of frequency and dominant at

    high frequencies. In the modern deep-submicron CMOS processes, the flicker noise

    2

  • 7/29/2019 MS Thesis ZhinengZhu

    14/81

    source may dominate the device noise at frequencies well into megahertz range. For the

    nanopore gene sequencer application, the bandwidth of interest is between 1MHz and

    10MHz. In this frequency range, the thermal noise is dominant, and the flicker noise

    tail will overlap the thermal noise and increase the noise floor. So for deep sub-micron

    CMOS process implementation of low noise operational amplifier, both types of noise

    are significant in this bandwidth and both need to be considered simultaneously.

    Most of the literature addressing the design of low noise operational amplifiers

    deal only in the low frequency range and therefore emphasize flicker noise reduction

    techniques[3][6][7][8][9]. In [3], the BiCMOS process is used and bipolar transistors

    are chosen for the input pair to address the high CMOS input-referred flicker noise prob-

    lem, which is not good for amplifying low current signal. Some dynamic techniques,

    like auto-zeroing and chopper stabilization [9], can achieve flicker noise levels in CMOS

    comparable to bipolar devices. But those methods use switch-capacitor circuits which

    introduce switching noise and alias high frequency thermal noise, and therefore achieve

    low flicker noise at the cost of bandwidth and thermal noise performance. The ther-

    mal noise PSD of these dynamic circuits could be far higher than 10nV/

    Hz [9] at

    frequencies in the MHz range.

    In this research, special circuit techniques are exploited to reduce both thermal

    and flicker noise in a CMOS operational amplifier designed as a current detector capable

    of operating in the frequency range of 1-10MHz. The unity gain bandwidth and DC gain

    are maximized. Since the untrimmed DC offset of a CMOS operational amplifier can be

    up to 20mV, larger than that of bipolar counterpart, a digital trimming method has been

    investigated and implemented to reduce the DC offset voltage. The amplifier design is

    based on a 0.35m CMOS process operated with a 3.3V supply voltage. This amplifier

    has applications to other systems requiring low levels of current detection.

    3

  • 7/29/2019 MS Thesis ZhinengZhu

    15/81

    thermal Noise

    corner frequency

    flicker Noise

    (log scale)

    )log(20 2nv

    cf f

    thermal Noise

    corner frequency

    flicker Noise

    (log scale)

    )log(20 2nv

    cf f

    Figure 1.2: Typical input-referred voltage noise PSD for a MOSFET

    1.3 Thesis Organization

    Chapter 1 introduces the research background and problem being addressed.

    Chapter 2 describes the various noise sources associated with MOSFETs. For

    each noise source a small-signal model is provided. Then the noise performance is ana-

    lyzed for the basic operational amplifiers with four different loads: current mirror load,

    resistor load, diode-connected load and active current source load. Their limitations for

    further noise reduction are analyzed.

    Chapter 3 introduces a new circuit topology for minimizing noise. Its noise per-

    formance is analyzed and the techniques of noise reduction are described and compared

    with those basic operational amplifiers presented in Chapter 2 and the folded cascode

    differential pair. Because of the relatively large offset voltage associated with CMOS

    operational amplifiers, a digital offset trimming method is also described in chapter

    three.

    Chapter 4 describes the noise measurement technique. The test circuit is pro-

    vided. Some test results are provided.

    Chapter 5 presents a low noise operational amplifier with a cascode compensa-

    tion scheme, which result in an improved PSRR.

    4

  • 7/29/2019 MS Thesis ZhinengZhu

    16/81

    Chapter 6 summarizes the conclusions for the design and suggestions for future

    work are provided.

    5

  • 7/29/2019 MS Thesis ZhinengZhu

    17/81

    CHAPTER 2

    Noise in MOSFETs and Basic CMOS Operational Amplifiers

    This chapter provides an overview of noise sources in MOSFETs. The noise

    performance of four basic CMOS operational amplifiers is analyzed. The limiations of

    noise reduction techniques for these basic amplifiers are discussed.

    2.1 Noise Sources in MOSFETs

    The noise sources in MOSFETs include: (a) thermal noise introduced by the

    channel; (b) flicker noise from the channel[10][11]; (c) thermal noise introduced by the

    polysilicon gate resistance[12]; (d) thermal noise introduced by the source/drain resis-

    tance and (e) thermal noise instroduced by the distributed substrate resistance[13]. In

    the design of a low-noise operational amplifier, wide transistors are typically chosen for

    the input pair and there are lots of contacts connected to the source and drain, therefore

    the thermal noise associated with the source/drain resistances can be neglected[14].

    2.1.1 Thermal noise in the channel

    Thermal noise is generated by the random motion of carriers in the channel[ 10]

    which introduces fluctuations, and therefore noise in the drain current. When a MOS-

    FET is biased in the active region, this noise source can be represented by a current

    noise generator connected from the drain to source as shown in Figure 2.1(a). The PSD

    of this current noise generator is i2d,thermal, which is described by

    i2d,thermal = 4kT2

    3gm (2.1)

    where k is Boltzmanns constant, T is the temperature in Kelvin, and gm is the small-

    signal transconductance from the gate to the channel.

    6

  • 7/29/2019 MS Thesis ZhinengZhu

    18/81

    2

    ,thermaldi2

    ,thermaldi

    (a)

    2

    ,thermaldv2

    ,thermaldv

    (b)

    Figure 2.1: MOSFET noise model: a) Model of thermal noise in the channel, b) Input-

    referred voltage noise model

    To compare the noise performance of the different circuits, the concept of input-

    referred noise is introduced. Input-referred noise is a source at the circuit input which

    represents the effect of all noise sources in the circuit. It is fictitious and cannot be

    measured in the real circuit. For a MOSFET, the current noise source introduced by the

    channel is referred back to the gate by dividing the PSD of the current noise generator

    by the square of the MOSFET transconductance gm. This results in an input-referred

    voltage noise source which is connected in series with the gate as shown in Figure 2.1(b).

    Its approximate PSD is given by:

    v

    2

    d,thermal = 4kT

    2

    3gm . (2.2)

    Minimizing the channel thermal noise of a MOSFET is straightforward from the above

    formula: the small-signal transconductance must be maximized. This can be achieved

    by using a large DC bias current and having a large width to length (W/L) ratio for the

    device.

    2.1.2 Flicker noise in the channel

    Flicker noise has been observed in all kinds of devices, from metal film resistors

    to semiconductor devices and even chemical batteries [15]. The MOSFET has one of

    the highest PSD of flicker noise among all active devices. Its precise origin has not yet

    been identified unequivocally. One popular explanation is that in a MOSFET, current

    7

  • 7/29/2019 MS Thesis ZhinengZhu

    19/81

    flows near the surface between SiO2 gate oxide and the crystalline silicon substrate.

    The silicon crystal terminates at this Si/SiO2 interface, producing many unoccupied

    dangling energy bonds [14]. As charge carriers move along this surface, they are trapped

    and released by those energy bonds randomly and introduce flicker noise in the drain

    current. The PSD of the noise depends inversely on the frequency and the input-referred

    flicker noise model[16] is given by:

    v2i,flicker(f) =KF

    2C2oxW L

    1

    f(2.3)

    where

    =carrier mobility in the channel;

    Cox=capacitance per unit area of the gate oxide;

    W=channel width;

    L=channel length;

    f=frequency;

    KF=flicker noise coefficient which is a process-dependant factor on the order of

    1028F

    A and can be different for NMOS and PMOS devices. In the 0.35m process

    used in this study, KF for the NMOS device is less than that of PMOS device.

    From the Equation 2.3, the PSD of the flicker noise is inversely proportional to

    the active gate area. Therefore devices with large gate areas are chosen to reduce the

    flicker noise of the MOSFET. Input-referred flicker noise is often modeled by connecting

    a noise voltage source in series to the gate of a MOSFET like that shown in Figure 2.1(b).

    2.1.3 Noise of polysilicon gate resistance

    The thermal noise and flicker noise in the channel are the two main sources of

    noise in MOSFETs. But for a low-noise amplifier design, other noise sources need to

    be considered. One of these is the noise introduced by the resistance of the polysilicon

    gate. The gate of modern CMOS device is made of polysilicon material often with a

    8

  • 7/29/2019 MS Thesis ZhinengZhu

    20/81

    silicided surface layer instead of metal. However, the polysilicon gate has its drawback,

    in that a polysilicon silicided gate is more resistive than metal [12], and therefore more

    noisy. For example, a silicided polysilicon gate sheet resistance is about 7/2. For

    a transistor with a W/L ratio of 10/1, the noise density from the gate is approximately

    1.077nV/

    Hz. The sheet resistance of aluminum is about 0.05/2 and the noise

    density of aluminum with the same aspect ratio is about 0.0288nV/

    Hz, which is

    negligible when compared to that of a polysilicon gate. In low-noise amplifier design,

    wide transistors are adopted for the input pair and the noise contribution from the gate

    can be reduced by proper layout.

    When the resistance of the gate is calculated, only effective resistance is consid-

    ered. The effective resistance of polysilicon gate is actually less than the multiplication

    of the technologys sheet resistance by the aspect ratio of the gate. This is because the

    gate resistance of a MOSFET is a distributed resistance[12] and its effective value for

    the gate when contacted at only one end as shown in Figure 2.2(a) is given by:

    RG =1

    3RSH,G

    W

    L(2.4)

    where RSH,G is the gate polysilicon sheet resistance with the typical value of 7/2 for

    a silicided polysilicon gate. If both ends of the gate are connected together, like that in

    Figure 2.2(b), the gate resistance becomes:

    RG =1

    12RSH,G

    W

    L(2.5)

    The above equation works for a single-finger device. If the device consists of

    multiple fingers such as that shown in Figure 2.2(c), then the overall gate resistance is

    given by

    RG =1

    3RSH,G

    W

    L

    1

    N(2.6)

    9

  • 7/29/2019 MS Thesis ZhinengZhu

    21/81

    (a) (b)

    (c) (d)

    Figure 2.2: Reduction of gate noise through layout, a) single-finger MOSFET, b) single-

    finger MOSFET with contact at both ends, c) multiple-finger MOSFET and d) multiple-

    finger MOSFET with contacts at both ends

    where N is the number of fingers. If all of the fingers are connected at both ends as

    shown in Figure 2.2(d), then the factor 1/3 should be replaced by 1/12 such that:

    RG =1

    12RSH,G

    W

    L

    1

    N (2.7)

    The noise spectral density associated with the gate resistance is then given by:

    v2g,thermal = 4kT RG (2.8)

    To reduce the thermal noise generated by the gate resistance, the transistor is laid-out in

    a multi-finger version with both ends of the gate connected together.

    2.1.4 Noise from the distributed substrate resistance

    Figure 2.3 shows the noise contribution from the substrate to the drain current

    [13]. The substrate has a distributed resistance and therefore generates thermal noise

    10

  • 7/29/2019 MS Thesis ZhinengZhu

    22/81

    2

    ,thermalsubv

    subR

    2

    ,thermalsubv

    subR

    Figure 2.3: Contribution mechanism of substrate noise to the drain current

    v2sub,thermal that can be approximated by:

    v2sub,thermal = 4kT Rsub (2.9)

    where Rsub is substrate distributed resistance. This noise is amplified by the substrate

    transconductance gmb and coupled to the drain current through a depletion capacitance,

    such that the current noise can be represented by:

    i2sub,thermal = 4kT Rsubg2mb (2.10)

    The direct way to reduce the noise from substrate is to decrease gmb by increasing

    the source bulk bias voltage or using a process with heavily doped substrate.

    2.2 Noise Performance of the CMOS Differential Pair

    To understand the main noise sources involved in an operational amplifier, lets

    consider the noise performance of a differential pair with a current mirror load shown in

    Figure 2.4. The voltage gain of the amplifier, A, measured from the gate of M1,2 to the

    11

  • 7/29/2019 MS Thesis ZhinengZhu

    23/81

    1M

    2M

    3M

    4M

    DDV

    SSI

    outV

    1M

    2M

    3M

    4M

    DDV

    SSI

    outV

    (a)

    1M

    2M

    3M

    4M

    SSI

    2

    1nv

    2

    2nv

    2

    3nv

    2

    4nv

    DDV

    outV

    1M

    2M

    3M

    4M

    SSI

    2

    1nv

    2

    2nv

    2

    3nv

    2

    4nv

    DDV

    1M

    2M

    3M

    4M

    SSI

    2

    1nv

    2

    2nv

    2

    3nv

    2

    4nv

    DDV

    outV

    (b)

    Figure 2.4: a) Differential pair with current mirror load, b) noise model of a)

    output Vout is described by the following equitation:

    A = gm2ro =2

    (nCOX(W/L)2)

    (2 + 4)

    ID2(2.11)

    where is is channel length modulation, is carrier mobility in the channel, Cox is the

    capacitance per unit are of the gate oxide, W and L are MOSFETs width and length

    respectively, ID2 is the bias current of the input pair M1 and M2.

    The gain from the gate ofM3,4 to the output Vout is gm4ro, where gm2 and gm4 are

    transconductance of M2 and M4, respectively, and ro is small-signal output resistance.

    The output noise is given by:

    v2

    no(f) = 2(gm2ro)2

    v2

    n2(f) + 2(gm4ro)2

    v2

    n4(f) (2.12)

    where vn2 and vn4 represent the noise from M2 and M4 respectively, which include all

    those noise sources discussed above and are referred back to the gate ofM2 and M4.

    12

  • 7/29/2019 MS Thesis ZhinengZhu

    24/81

    This noise v2no can be referred back to the amplifiers input by dividing the

    squared gain from the amplifiers input to its output which results in

    v2

    ni(f) = 2v2

    n2(f) + 2v2

    n4(f)gm4

    gm22

    = 2v2

    n2(f) + 2v2

    n4(f) (W/L)4p

    (W/L)2n

    (2.13)

    Assuming COX,n = COX,p = COX, the flicker noise and thermal noise of this structure

    are given by the following two equations:

    v2flicker(f) =KFn

    nC2OXW2L2f

    1 +

    KFpKFn

    L2L4

    2(2.14)

    v2thermal =16kT

    3

    2nCOX(W/L)2ID2

    1 + p(W/L)4n(W/L)2

    (2.15)where KFn is the flicker noise coefficient of NMOS, KFp is the flicker noise coefficient

    of PMOS, n and p are carrier mobilities for both NMOS and PMOS respectively.

    From the above equations, we find that each MOSFET introduces both flicker noise and

    thermal noise. All CMOS voltage operational amplifiers have at least two MOSFETs

    as the input pair. The more MOSFETs in the circuit, the more noise is introduced.

    From Equation 2.15, it is observed that large bias currents can be used to minimize

    thermal noise at the cost of power dissipation and voltage gain A. The large bias current

    also results in a large overdrive voltage for transistors M3 and M4. The drain to source

    voltage VDS ofM3 and M4 has to be large enough to keep the transistors in the saturation

    region. Hence

    VDS > Vthreshold + Voverdrive (2.16)

    Therefore increasing the bias current reduces the voltage headroom at the output. For

    the differential pair with current mirror load, it is clear from this analysis that optimizing

    the noise performance involves a trade off with other parameters, including the DC gain

    and power.

    13

  • 7/29/2019 MS Thesis ZhinengZhu

    25/81

    A1 A2

    2

    1v

    2

    2v

    A1 A2A1 A2

    2

    1v

    2

    2v

    Figure 2.5: Two stage operational amplifier with noise sources

    Figure 2.5 shows a two stage operational amplifier. The first stage has a gain of

    A1 and an output noise v21; the second stage has a gain of A2 and an output noise v22 .

    The total gain of the amplifier is A = A1 A2. The total output noise of the amplifier is

    v2

    total = v2

    2 + A2

    2 v2

    1 (2.17)

    When this noise is referred back to the input port, the input-referred noise is

    v2i =v22

    A22A21

    +1

    A21v21 (2.18)

    Since the noise from the first stage is amplified by the second stage, the noise of first

    stage is more important than that of the second stage. The larger the gain of first stage

    A1, the less important the noise contribution of second stage. However, the differential

    pair with current mirror load has a disadvantage in that the DC gain is inversely propor-

    tional to the bias current. The higher the bias current, the lower the DC gain of the first

    stage, which will make the noise from later stage more pronounced.

    In this research, the differential pair with more than 4 cascode stages is not pre-

    ferred because we used a 0.35m CMOS process with a working voltage of 3.3V. A

    differential pair with more than 4 cascode stages will have a headroom problem. So, the

    first effort is to analyze some other basic differential pairs to see whether any of them

    can provide a superior noise performance.

    14

  • 7/29/2019 MS Thesis ZhinengZhu

    26/81

    Differential pair with a resistive load:

    The differential pair with resistive load is shown in Figure 2.6(a). Its input-

    referred noise is given by:

    v2ni = 8kT

    2

    3gm+

    1

    g2mRD

    +

    KFnnC

    2OXW2L2f

    (2.19)

    For low-noise design, a large bias current (even to the order of a few milliamps) is

    needed and the resistor value has to be chosen to minimize thermal noise. The opera-

    tional amplifier with a large resistive load can reduce the voltage headroom at the output.

    Moreover, the gain of this structure is relatively low, which make the noise of the later

    stages a larger issue.

    Differential pair with a Diode-connected load:

    An Differential pair with a diode-connected load is shown in Figure 2.6(b), and

    the noise performance is given by Equation 2.14 and 2.15. This structure has the same

    problems as the operational amplifier with a resistive load. For low noise operational

    amplifier design, the bias current should be large and the drain-source voltage also needs

    to meet Equation 2.16, which results in a small output voltage swing range. Also such a

    structure has a small gain.

    Differential pair with an active current source load:

    The fully differential pair with active current source load is shown in Figure

    2.6(c). The structure has a higher output common mode range than those amplifiers dis-

    cussed above since its load drain-source voltage only need to meet the requirement of

    VDS > Voverdrive, a threshold voltage less than that of the differential pair with a current

    15

  • 7/29/2019 MS Thesis ZhinengZhu

    27/81

    1M

    2M

    1R

    2R

    DDV

    SSI

    1M

    2M

    1R

    2R

    DDV

    SSI

    (a)

    1M

    2M

    3M

    4M

    DDV

    SSI

    1M

    2M

    3M

    4M

    DDV

    SSI

    (b)

    CMFB

    1M

    2M

    3M

    4M

    DDV

    SSI

    CMFB

    1M

    2M

    3M

    4M

    DDV

    SSI

    (c)

    1M 2M

    3M

    4M

    SSI

    5M 6M

    7M

    8M

    AB

    In +In

    ddV

    2bV

    4bV

    Out

    1M 2M

    3M

    4M

    SSI

    5M 6M

    7M

    8M

    AB

    In +In

    ddV

    2bV

    4bV

    Out

    (d) Folded-cascode differential pair

    Figure 2.6: Four basic CMOS differential pairs, a) resistive load, b) diode-connected

    load, c) active current source load and d) folded-cascode differential pair

    16

  • 7/29/2019 MS Thesis ZhinengZhu

    28/81

    mirror load and diode-connected structure. The larger headroom provided by this struc-

    ture allows the input pair to sink more DC bias current than the other structures. The

    input-referred noise density is also given by Equations 2.14 and 2.15. Therefore it has

    the potential of the best noise performance among those introduced before. But such

    a structure requires a Common Mode Feedback (CMFB) network to define the output

    common-mode level. Any mismatch in the CMFB circuit can introduce extra noise that

    could outbalance the advantage of the fully differential structure.

    Folded-cascode differential pair:

    Figure 2.6(d) gives the structure of a folded cascode differential pair. This struc-

    ture has the limitation of maximizing the input pairs bias current for a low noise design.

    M3 and M4 source bias current for the input pair M1 and M2, and the current mirror

    load M7 and M8. When the bias current ofM1 and M2 is maximized, the bias current

    ofM3 and M4 is maximized as well. This will reduce the DC voltage level at the drain

    of M3 and M4. This reduced DC voltage level will reduce the voltage headroom at the

    drain of M6, which is especially true when the length of M3 and M4 is chosen larger

    than normal in a low noise operational amplifier design.

    From the above discussion, we can appreciate the trade-offs between noise,

    power dissipation, voltage headroom when designing a low noise operational ampli-

    fier. Chapter three describes a new amplifier structure which combines the features of

    the fully differential pair with active current source load with an amplifier with a current

    mirror load. That structure has the potentiality of improved noise performance.

    17

  • 7/29/2019 MS Thesis ZhinengZhu

    29/81

    CHAPTER 3

    Novel Structure for a Low Noise Operational Amplifier using

    MOSFETs

    This chapter proposes a new operationality amplifier structure offering the po-

    tential of better noise performance. The techniques used to reduce noise are described.

    In Chapter 2, the noise limitations of the basic CMOS differential pair structure

    were discussed. In this chapter a new structure shown in Figure 3.1 is proposed and

    analyzed. This structure combines the merits of a fully differential pair with an active

    current source load and an differential pair with a current mirror load. The active current

    source M3 and M4 can source the maximum current for the input pair to reduce thermal

    noise. The current mirror M7 and M8 combine with the transconductance of the input

    pair to define the DC gain. A cascode stage, M5 and M6, is introduced for the following

    three reason:

    1. This cascode stage can reduce the Miller effect, thereby increasing the bandwidth

    of the amplifier. The input pair (M1, M2) has a gate-drain parasitic capacitance

    Cgd. If there is no cascode stage, the gain Amiller from input (M1, M2) to the drain

    of M7 and M8 will be larger than 1, that is, Amiller > 1. The Miller effect will

    magnify the parasitic capacitance as AmillerCgd, which will limit the amplifiersbandwidth. With the introduction of cascode stage M5 and M6, the voltage at the

    drains ofM1 and M2 is fixed, and the gain from the input to the source of M5,6 is

    reduced and therefore the parasitic capacitance at the gate ofM1,2 is minimized.

    2. This cascode stage increase the output resistance looking into the drain ofM6 and

    therefore increase the gain of this differential pair.

    3. The cascode structure reduces the short channel variation of the input stage. In

    describing the channel length modulation using the channel length modulation

    18

  • 7/29/2019 MS Thesis ZhinengZhu

    30/81

    DDV

    1M

    2M

    3M

    4M

    5M 6M

    7M

    8M

    SSI

    2bV

    4bV

    +In

    In

    Out

    DDV

    1M

    2M

    3M

    4M

    5M 6M

    7M

    8M

    SSI

    1M

    2M

    3M

    4M

    5M 6M

    7M

    8M

    SSI

    2bV

    4bV

    +In

    In

    Out

    Figure 3.1: Proposed structure of the low noise differential pair

    parameter , we assume that the small-signal output impedance ro is constant.

    This is not really true. This output impedance varies with the drain-source voltage.

    When the drain-source voltage increases, the pinch-off point will move towards

    source region resulting in a wider depletion region around the drain and a higher

    ro. When the drain-source voltage decreases, a lower value ofro results. With thecascode stage M5,6, the drain-source voltage of M1,2 doesnt experience a large

    variation and the small-signal resistance ofM1,2 remains almost constant.

    Lets examine how the disadvantages of achieving low noise performance for

    those differential pairs discussed in Chapter 2 are avoided in this new structure. In this

    new structure, the bias current of the amplifiers input pair is rationed between M3,4

    and M7,8. M3,4 is sized to source most of the bias current. For example, if the bias

    current ofM1,2 is chosen to be 3mA, I3,4 ofM3,4 can be 2.85mA and I7,8 can be set as

    0.15mA. The common-mode level at the drain of M3,4 is defined by the cascode stage

    M5,6, instead of common mode feedback circuit used in the differential pair with active

    current source load. To reduce the thermal noise, the large bias current of 3mA is chosen

    19

  • 7/29/2019 MS Thesis ZhinengZhu

    31/81

    1M

    2M

    3M

    4M

    5M

    6M

    7M

    8M

    DDV

    0M

    2

    1nv

    2

    2nv

    2

    3nv

    2

    4nv

    2

    5nv

    2

    6nv

    2

    7nv

    2

    8nv

    2

    0nv

    1M

    2M

    3M

    4M

    5M

    6M

    7M

    8M

    DDV

    0M

    2

    1nv

    2

    2nv

    2

    3nv

    2

    4nv

    2

    5nv

    2

    6nv

    2

    7nv

    2

    8nv

    2

    0nv

    Figure 3.2: Noise model of differential pair in Figure 3.1

    for the input pair M1,2. This current is mainly provided by the active load M3,4. The

    output of the common-mode level is defined by M7,8, which takes a very small current

    and provides a wider output swing than that provided by a differential pair with a current

    mirror load. In the analysis that follows, we show that this structure has the advantage

    of increasing the small-signal gain without deteriorating the noise performance. The

    large 3mA bias current results in the input pair M1,2 with a high aspect ratio and a large

    transconductance gm, these improves the bandwidth and other performances, like low

    flicker noise and high DC gain.

    3.1 Noise Analysis of New Structure

    Figure 3.2 shows the noise model for the structure in Figure 3.1. The noise con-

    tribution of each gate is referred back to their own gate input, like the noise contribution

    ofM3, which is represented by v2n3 connected in series to its gate. In the following noise

    calculation, symmetry is assumed, such that the noise contributions ofM1, M3, M5 and

    M7 are the same as those ofM2, M4, M6 and M8.

    20

  • 7/29/2019 MS Thesis ZhinengZhu

    32/81

    42|| rr

    16vg

    m

    22 nmvg

    8r

    1v

    2nov

    2nv

    42|| rr

    16vg

    m

    22 nmvg

    8r

    1v

    2nov

    2nv

    Figure 3.3: Small-signal noise model for input stage M1,2

    3.1.1 The noise contribution of the input stage M1,2

    The small-signal model shown in Figure 3.3 is for the calculation ofM2s noise

    contribution, where the parasitic capacitances are omitted for simplicity. From this

    model, we find the output noise from M2 is:

    vno2 = gm2r8vn2 gm2rovn2 (3.1)

    where ro is the small-signal output resistance of the circuit in Figure 3.1.

    3.1.2 The noise contribution of the active current source load M3,4

    The small-signal model to calculate the noise contribution of the active current

    source load stage is the same as that in Figure 3.3 withvn

    2 andgm

    2 replaced withvn

    4

    and gm4. So the output noise from the active current load M4 is:

    vno4 = gm4r8vn4 gm4rovn4 (3.2)

    21

  • 7/29/2019 MS Thesis ZhinengZhu

    33/81

    8r

    16vg

    m

    88 nmvg 42 || rr

    1v

    8nov

    8nv

    8r

    16vg

    m

    88 nmvg 42 || rr

    1v

    8nov

    8nv

    Figure 3.4: Small-signal noise model for current mirror load M7,8

    3.1.3 The noise contribution of current mirror load M7,8

    The small signal model for noise contribution of the current mirror load is shown

    in Figure 3.4. The noise contribution from M8 is approximately:

    vno8 = gm8r8vn8 gm8rovn8 (3.3)

    3.1.4 The noise contribution of cascode stage M5,6

    The cascode stages small-signal noise model is shown in Figure 3.5. Here

    vgs6 = vn6 v1. The output noise from this stage is given by:

    42|| rr

    1v

    6r

    66 gsmvg

    8r

    6nov

    6nv

    42|| rr

    1v

    6r

    66 gsmvg

    8r

    6nov

    6nv

    Figure 3.5: Small-signal noise model for the cascode stage M5,6

    22

  • 7/29/2019 MS Thesis ZhinengZhu

    34/81

    vno6 =gm6r6r8

    r6 + r8 + gm6r6r2vn6 r8

    r2vn6 (3.4)

    where vn6 is noise contribution of M6 referred back to its gate. The effective transcon-

    ductance of M6 according to Equation 3.4 is approximately 1/r2 and is much smaller

    than transconductance gm6 of M6 and gm2 of M2. The operational amplifiers small-

    signal output resistance ro is close to r8. The noise generated by the cascade current

    buffers and referred back to the amplifiers input is vni5,6 =1

    gm2ro

    r8r2

    vn6 1gm2r2 vn6, andis much smaller than those generated by input stage, active current and current mirror

    loads, therefore it is omitted in the later noise analysis.

    3.1.5 The noise contribution of current bias MOSFET M0

    M0 provides the bias current for the input pair, and noise associated with this

    transistor will modulate the bias current. Ideally, the active current source load M3,4

    will not see noise from M0 since their gate voltage experience no change. So this noisewill run through the input pair M1,2 and the cascade stage M5,6 to the current mirror

    load M7,8. The other observation is that the drains voltage of M8 will track that ofM7.

    So, the noise from M0 at the output (drain of M8) is equal to the noise at the drain of

    M7. As a result, the small-signal model shown in Figure 3.6 can be used to analysis the

    noise contribution ofM0. The noise from M0 is approximately vno0 = gm0gm7vn0, and theamount of noise referred back to the operational amplifiers input is vni0 =

    1

    gm2ro

    gm0gm7

    vn6.

    This is a very small value when compared with other stages input referred noise due to

    the large gain gm2ro from the input pair to the output.

    23

  • 7/29/2019 MS Thesis ZhinengZhu

    35/81

    1v

    00nm

    vg0r

    11vg

    m

    1r

    3r

    2v

    7

    1

    mg

    0nov

    5r

    25vg

    m

    1v

    00nm

    vg0r

    11vg

    m

    1r

    3r

    2v

    7

    1

    mg

    0nov

    5r

    25vg

    m

    Figure 3.6: Small-signal noise model for bias current source M0

    3.1.6 Total output noise

    Combining all those noise models discussed above, the total output noise is given

    by

    v2no(f) = v2

    no0(f) + 2v2

    no2(f) + 2v2

    no4(f) + 2v2

    no6(f) + 2v2

    no8(f)

    2(gm2ro)2v2n2(f) + 2(gm4ro)2v2n4(f) + 2(gm8ro)2v2n8(f) (3.5)

    The input referred noise is

    v2ni(f) = 2v2

    n2(f) + 2v2

    n4(f)

    gm4gm2

    2

    + 2v2n8

    gm8gm2

    2

    = 2v2n2(f) + 2v2

    n4(f)

    (W/L)4pID4(W/L)2nID2

    + 2v2n8(f)

    (W/L)8pID8(W/L)2nID2

    (3.6)

    The following two equations show the input referred flicker noise and thermal noise for

    this design respectively:

    v2flicker(f) =KFn

    nC2OXW2L2f

    1 + KF

    p

    KFn

    L2L4

    2ID4ID2

    + KFp

    KFn

    L2L8

    2ID8ID2

    (3.7)

    v2thermal =16kT

    3

    2nCOX(W/L)2ID2

    1 +

    p(W/L)4ID4n(W/L)2ID2

    +

    p(W/L)8ID8n(W/L)2ID2

    (3.8)

    24

  • 7/29/2019 MS Thesis ZhinengZhu

    36/81

    3.2 Noise Reduction Techniques for this New Structure

    From the above two equations, we find there are three techniques that can be

    used to reduce the noise:

    1. Determination of the input pair type for M1,2

    2. Optimization of the bias current ID2 of the input pair

    3. Optimization of the sizes and aspect ratios of the MOSFETs

    3.2.1 Determination of the input pair type for M1,2

    In this structure, NMOS should be chosen for the input pair M1,2 for the follow-

    ing three reasons:

    1. The bandwidth we are dealing with is between 1MHz to 10MHz. Over this band-

    width range, thermal noise is often dominant. Selecting a NMOS transistor for

    the input pair reduces thermal noise according to Equation 3.8, since NMOS tran-

    sistors have a lager mobility n(often 2 to 3 times) than PMOS transistors.

    2. Flicker noise is still important in this design. In a submicron process, the corner

    frequency could be greater than 1MHz. In Figure 1.2 we can see that as the

    thermal noise floor is reduced, the corner frequency is increased. The flicker

    noise tail will overlap with the thermal noise floor. In the 0.35m CMOS process

    used, the NMOS flicker noise coefficient KFn is smaller than that of PMOS flicker

    noise coefficient KFp, which is helpful for achieving lower flicker noise according

    to Equation 3.7.

    3. NMOS transistors have a higher transition frequency fT than PMOS transistors,

    which help achieve a higher bandwidth.

    25

  • 7/29/2019 MS Thesis ZhinengZhu

    37/81

    3.2.2 Optimization of the bias current ID2 of the input pair

    To further reduce the thermal noise, the bias current ID2 of the input pair can be

    increased up to several milliamps in this new structure. With such large bias current,

    the basic CMOS differential pair like the one with current mirror load can have good

    noise performance but at the cost of gain and voltage headroom at the output. However,

    for this new structure, the increased bias current does not deteriorate either the gain or

    voltage headroom . This is one of the highlights of this new structure which is explained

    in the DC gain section 3.3 below.

    3.2.3 Optimization of the sizes and aspect ratios of MOSFETs

    According to Equation 3.7, choosing a large size (W L)2 for the input pair, and

    making the channel length L of the active load M3,4 and current mirror M7,8 larger than

    that of input pair M1,2 can help to reduce the flicker noise. The aspect ratio (W/L)2

    is chosen to be larger than (W/L)4 and (W/L)8, so the second and third terms in the

    bracket of Equation 3.8 is less than 1. This helps to reduce the thermal noise. After

    those choices, we can find that the input pair contributes most of the noise of both types.

    The gate length L1,2 of the input pair could increase or decrease flicker noise

    according to Equation 3.7. The optimum value is determined by:

    v2flickerL2

    = 0 L2 =

    KFnKFp

    1

    L24

    ID4ID2

    +

    1

    L28

    ID8ID2

    (3.9)

    When the gate lengths of M3,4 and M7,8 are first chosen to be 2.1m to avoid a large

    overdrive voltage, ID2 and ID4 chosen to be 3mA and 2.85mA respectively, the optimum

    gate length ofM1,2 is found to be 0.7m as shown in Figure 3.7.

    The simulated input-referred voltage noise performance of the circuit of Figure

    3.1 using a 0.35m CMOS process is shown in Figure 3.8. The total noise level is

    simulated to be 1.796nV /

    Hzat 1MHz and 1.184nV/

    Hzat 10MHz.

    26

  • 7/29/2019 MS Thesis ZhinengZhu

    38/81

    0 0.5 1 1.5 2 2.5 3

    x 10-6

    2.5

    3

    3.5

    4

    4.5

    5

    5.5

    6

    6.5

    7x 10

    -19

    Input Gate Length (m)

    Noise

    Level(v

    2)

    X: 7e-007

    Y: 2.873e-019

    Figure 3.7: Optimum gate length of the input pair

    100

    101

    102

    103

    104

    105

    106

    107

    0

    1

    2

    3

    4

    5

    6

    7

    8

    9x 10

    -7

    X: 1e+006

    Y: 1.796e-009

    Frequency(Hz)

    noise

    level(v/(Hz))

    X: 1e+007

    Y: 1.184e-009

    Figure 3.8: The noise performace of the circuit in Figure 3.1

    27

  • 7/29/2019 MS Thesis ZhinengZhu

    39/81

    100

    101

    102

    103

    104

    105

    106

    107

    0

    0.5

    1

    1.5

    2

    2.5

    3x 10

    -6

    X: 1e+007

    Y: 2.427e-009

    Frequency(Hz)

    noise

    level(v/(Hz))

    X: 1e+006

    Y: 2.306e-009

    Figure 3.9: The noise performace of the circuit with PMOS as the input pair

    Figure 3.9 shows the simulated noise level of the same structure when using

    PMOS transistors as the input pair. The size, aspect ratio and bias current are all

    the same as those in Figure 3.8. From this figure we conclude that the PMOS input

    pair shows poorer noise performance: the noise level is 2.306nV /

    Hz at 1MHz and

    2.427nV/

    Hzat 10MHz.

    From the previous analysis, we find that the current mirror load M7,8 sources

    a very small amount of bias current compared with the input pair. Thus, ID8ID2

    120

    ,

    ID4 ID2 and Equation 3.7 and 3.8 are reduced to:

    v2flicker(f) KFn

    nC2OXW2L2f

    1 +

    KFpKFn

    L2L4

    2(3.10)

    v2thermal 16kT

    22nCOX(W/L)2ID2

    1 +

    p(W/L)4n(W/L)2

    (3.11)

    From these two equations, we can find that the four MOSFETs ( M1,2,3,4) contribute most

    of the noise in this new structure. Also from these two equations, we can find that the

    noise performance of this new structure is comparable with the differential pair with

    active load, but it doesnt need CMFB circuit.

    28

  • 7/29/2019 MS Thesis ZhinengZhu

    40/81

    3.2.4 Noise performance of the folded-cascode differential pair

    Figure 2.6(d) shows a folded-cascode differential pair. It has the same small-

    signal model as that of Figure 3.1. And its flicker and thermal noise are also represented

    by Equation 3.7 and 3.8 respectively. But the new structure has better noise performance

    than this folded-cascode differential pair. Assuming ID2, ID4 and ID8 are the bias cur-

    rents of M2, M4 and M8 in Figure 3.1 respectively, and IfD2, I

    fD4 and I

    fD8 are the bias

    currents ofM2, M4 and M8 in Figure 2.6(d) respectively, we find ID2 = ID4 + ID8 and

    IfD2 = IfD4 IfD8. If we choose ID2 = IfD2, then ID4 < IfD4 and the drain of M4 in

    Figure 2.6(d) will have smaller DC level than that of the drain of M4 in Figure 3.1. This

    smaller DC level in folded-cascode differential pair results in smaller voltage headroom

    at the drain ofM6. This is especially true when a long gate length is chosen for M3 and

    M4 and a large bias current is chosen for the input pair in a low noise design. Therefore,

    this folded-cascode structure has smaller output swing. Lets think about it in another

    way, if we apply the same power to both the circuits in Figure 3.1 and Figure 2.6(d),

    which means ID2 = IfD4 > I

    fD2, then new structure has better noise performance since

    its input pair has larger bias current which results in lower thermal noise. Hence, the

    new structure offers an improvement over the standard folded cascode structure.

    3.3 DC Gain

    The small-signal DC gain of the differential pair of Figure 3.1 is given by:

    A1 = gm2((gm6ro6(ro2 ro4 r1a)) ro8) (3.12)

    where roa is the small signal resistance of the auxiliary port consisting ofM1a and M2a

    as shown in Figure 3.10. This auxiliary port is used to reduce offset and its detailed

    analysis is given in the offset reduction section 3.6 below. Even though the large bias

    current reduces ro2 and ro4, the introduction of the cascode stage increases the output

    29

  • 7/29/2019 MS Thesis ZhinengZhu

    41/81

    1M

    2M

    3M

    4M

    5M

    6M

    7M 8M

    DDV

    1SSI

    2SSI

    aM

    1a

    M2

    2bV

    4bV

    +InIn +Ina

    Ina

    Out

    1M

    2M

    3M

    4M

    5M

    6M

    7M 8M

    DDV

    1SSI

    2SSI

    aM

    1a

    M2

    2bV

    4bV

    +InIn +Ina

    Ina

    Out

    Figure 3.10: Modified structure of the operational amplifier with an auxiliary port

    resistance looking into the drain of M6 and makes the ro8 the dominant parameter in

    determining the DC gain in Equation 3.12. ro8 can be approximated by ro8 1ID8 andcan be increased in two ways. First the channel length of the transistors M7,8 can be

    increased to reduce the length modulation parameter which is inversely proportional

    to the MOSFET channel length for the first order approximation [14]. In this design, the

    current is radioed between the active load M3,4 and current mirror M7,8. So the second

    way to increase ro8 is to minimize the current ID8 in M7,8. Increasing the bias current ID2

    increases the transconductance gm2 and therefore DC gain. The active load M3,4 sources

    this extra bias current through the input pair M1,2, and therefore the thermal noise is

    reduced as described by Equation 3.8. By now it can be observed from the previous

    discussion that M1, M2, M3 and M4 are primarily responsible for noise reduction, while

    M1, M2, M5, M6, M7 and M8 are responsible for the DC gain. The DC gain and noise,

    which is often a trade-off between noise level and gain in differential pair with current

    mirror load, can be optimized independently for this proposed structure .

    30

  • 7/29/2019 MS Thesis ZhinengZhu

    42/81

    1M

    2M

    3M

    4M

    5M

    6M

    7M

    8M

    DDV

    aM

    1 aM

    2

    0M

    aM

    0

    9M

    10M11M

    12M

    13M

    14M

    Rc

    C

    Out

    1bI

    2bI

    3bI

    4b

    V

    5bI

    +In In

    +a

    In aIn1M

    2M

    3M

    4M

    5M

    6M

    7M

    8M

    DDV

    aM

    1 aM

    2

    0M

    aM

    0

    9M

    10M11M

    12M

    13M

    14M

    Rc

    C

    Out

    1bI

    2bI

    3bI

    4b

    V

    5bI

    +In In

    +a

    In aIn

    Figure 3.11: Two stage low noise low offset operational amplifier

    3.4 AC Performance

    The common source configuration is chosen for the output stage of the opera-

    tional amplifier. Such a stage can achieve about 20-30 dB of gain. A Miller compen-

    sation scheme is chosen to achieve reasonable phase margin. A two stage low noise

    operational amplifier incorporating the new structure for the differential pair is shown in

    Figure 3.11. Sizes of the transistors, and the Miller compensation resistor and capacitor

    are provided in Table 3.1.

    M1 816/0.7 M6 66/0.7 M0a 66/0.7 M13 220/0.35M2 816/0.7 M7 132/2.1 M9 385/0.7 M14 198/2.1

    M3 198/2.1 M8 132/2.1 M10 220/0.35 R 940M4 198/2.1 M1a 105/0.7 M11 408/0.7 Cc 5pFM5 66/0.7 M1b 105/0.7 M12 66/0.7

    Table 3.1: Optimized devices size of the two-stage operational amplifier

    31

  • 7/29/2019 MS Thesis ZhinengZhu

    43/81

    1R

    1C

    cC6r

    16vgm

    29vg

    m 3R 3C

    R

    2

    2 im vg2

    R2

    C

    2

    1 im vg

    1v

    2v outv

    1R

    1C

    cC6r

    16vgm

    6r

    16vgm

    29vg

    m 3R 3C

    R

    2

    2 im vg2

    R2

    C

    2

    1 im vg

    1v

    2v outv

    Figure 3.12: Small signal model of the circuit in Figure 3.11

    The small-signal model for the circuit in Figure 3.11 is given in Figure 3.12:

    Here R1 = r2a r2 r4C1 = Cgd2 + Cgd4 + Cgd2a + Cgs6

    R2 = r8

    C2 = Cgd6 + Cgd8 + Cgs9

    R3 = r9 r10C3 = Cgd9 + Cgd10

    Analysis of ac small-signal model, after considerable algebra work and discard-

    ing some minor parameters such as C1 and C2, results in the following equation:

    voutvi

    =gm2gm9R2R3 sgm2R2R3Cc(1 gm9R)

    1 + s(RCc + R3C3 + gm9R2R3Cc) + s2RR3CcC3(3.13)

    With the assumption that the second pole is far larger than the dominant pole, the dom-

    inant pole of above equation can be approximated by:

    p1 = 1gm9R2R3Cc

    . (3.14)

    Since the DC gain of this amplifier is:

    A(0) = A1A2 = (gm2((gm6ro6(ro2 ro4 r2a)) ro8)) gm9R3= gm2R2 gm9R3 (3.15)

    32

  • 7/29/2019 MS Thesis ZhinengZhu

    44/81

    and the unity gain bandwidth is still given by GBP = A(0)|p1| = gm2CcThe second dominant pole is:

    p2 gm9R2RC3

    . (3.16)

    and the zero is:

    z =gm9

    Cc(1 gm9R) . (3.17)

    For unity-gain stability, the magnitude of the second dominant pole should be greater

    than the GBP such that:

    |p2| gm9R2RC3

    > GBP =gm2Cc

    . (3.18)

    Therefore

    Cc >gm2RC3gm9R2

    . (3.19)

    From Equations 3.16 and 3.17, we can find that the zero-nulling resistor R can control

    the zero and second dominant pole positions and therefore change the phase margin.

    One observation from Equation 3.19 is that in this low noise operational amplifier de-

    sign, the input transconductance can be larger than the transconductance of the output

    driver. If R is chosen close to 1/gm9, thenp2 is maximized and stability is achieved with

    a minimum Cc.

    For the values given in Table 3.1, the circuit of Figure 3.11 was simulated. The

    simulated AC results and noise performance including parasitic resistances and capaci-

    tances introduced by the layout are shown in Figure 3.13 and Figure 3.14 respectively.

    Simulation results in Figure 3.13 show the DC gain of the design is around 120dB with

    unity-gain bandwidth up to 380MHz. The phase margin is around 11o. The simulated

    total input-referred noise level is lower than 2nV /

    Hzfrom 1MHz to 10MHz.

    33

  • 7/29/2019 MS Thesis ZhinengZhu

    45/81

    100

    101

    102

    103

    104

    105

    106

    107

    108

    109

    1010

    -100

    -50

    0

    50

    100

    Frequency(Hz)

    Gain(dB) X: 3.802e+008

    Y: -0.1363

    100

    101

    102

    103

    104

    105

    106

    107

    108

    109

    1010-400

    -300

    -200

    -100

    0

    Frequency(Hz)

    Phase

    X: 3.802e+008

    Y: -169.4

    Figure 3.13: AC performance with parasitic resistances and capacitances

    100

    101

    102

    103

    104

    105

    106

    107

    0

    0.1

    0.2

    0.3

    0.4

    0.5

    0.6

    0.7

    0.8

    0.9

    1x 10

    -6

    X: 1e+006

    Y: 1.939e-009

    Frequency(Hz)

    noise

    level(v/(Hz))

    X: 1e+007

    Y: 1.297e-009

    Figure 3.14: Noise performance with parasitic resistances and capacitances

    34

  • 7/29/2019 MS Thesis ZhinengZhu

    46/81

    6bM

    5bM

    7bM

    1bM

    2bM

    3bM 4bM

    R

    6bM

    5bM

    7bM

    1bM

    2bM

    3bM 4bM

    R

    Figure 3.15: Self-biasing Vthreshold reference with start-up circuit

    3.5 Bias Circuit

    The bias network is an important component in the low noise amplifier design.

    The bias core is shown in Figure 3.15. The advantage of this bias circuit is that the

    currents through Mb3,b4 are insensitive to the supply voltage to the first order, which

    means such circuit has superior PSRR performance. A combination of current- and

    voltage-routing techniques [10] is used to reduce the mismatch and supply resistance.

    The MOSFETs Mb5,b6,b7 consist of a start-up circuit with Mb5 being a long channel

    device.

    The entire bias network is shown in Figure 3.16. Cascode stages are imple-

    mented to increase the output resistance and the values of the transistors are given in

    Table 3.2.

    3.6 Offset Reduction and Auxiliary Port

    In general CMOS operational amplifiers exhibit higher input offset voltage than

    their BJT counterparts. The photolithography, ion implantation, etching and other process-

    related factors can cause a mismatch in the threshold voltage Vt and gain factor between

    35

  • 7/29/2019 MS Thesis ZhinengZhu

    47/81

    7bM

    5bM

    6bM

    1bM

    2bM

    3bM

    4bM

    8bM

    9bM

    10bM

    11bM

    12bM

    13bM

    14bM

    15bM

    16bM

    17bM

    18bM

    19bM

    20bM

    21bM

    22bM

    23bM

    1bV

    2bV

    3bV

    5bV

    R

    DDV

    7bM

    5bM

    6bM

    1bM

    2bM

    3bM

    4bM

    8bM

    9bM

    10bM

    11bM

    12bM

    13bM

    14bM

    15bM

    16bM

    17bM

    18bM

    19bM

    20bM

    21bM

    22bM

    23bM

    1bV

    2bV

    3bV

    5bV

    R

    DDV

    Figure 3.16: Schematic of entire bias circuit

    Mb1 220/0.7 Mb7 2.2/2.1 Mb13 220/0.7 Mb19 72/0.7Mb2 220/0.7 Mb8 220/0.7 Mb14 660/0.7 Mb20 22/2.1Mb3 220/0.7 Mb9 55/2.1 Mb15 660/0.7 Mb21 110/0.7Mb4 220/0.7 Mb10 690/0.7 Mb16 22/2.1 Mb22 310/0.7Mb5 22/0.7 Mb11 690/0.7 Mb17 110/0.7 Mb23 310/0.7

    Mb6 2.2/14 Mb12 44/2.1 Mb18 72/0.7 R 940

    Table 3.2: MOSFETsaspect ratio in the bias circuit

    36

  • 7/29/2019 MS Thesis ZhinengZhu

    48/81

    the input pairs and generate a random voltage. Typically, a CMOS operational ampli-

    fiers offset can be up to 20mV. In this design, there have been many efforts made to

    reduce the amplifiers noise level. These efforts are inherently helpful in minimizing the

    operational amplifiers random offset.

    Because of the high gain of the first stage, the offset introduced by the second

    stage is negligible. So, the input-referred random offset for the operational amplifier

    shown in Figure 3.11 is approximated by [10]:

    VOS Vt1,2 +Vt3,4 gm4gm2

    +Vt7,8 gm8gm2

    +Vov1,2

    2

    (WL

    )1,2

    (WL

    )1,2+(W

    L)3,4

    (WL

    )3,4+(W

    L)7,8

    (WL

    )7,8

    (3.20)

    The first term in Equation 3.20 is due to the input pairs threshold mismatch. The second

    term is due to the active current source load threshold mismatch referred back to the in-

    put of amplifier. Having gm4 smaller than gm2 to reduce noise is also helpful in reducing

    this second term. The third term is negligible because ofgm8 is designed far less than

    gm2 fr noise performance. In this design, the large aspect ratio of the input pair achieves

    a small overdrive voltage Vov1,2 and therefore helps to reduce the last term. The standard

    deviation of the difference between the input pairs threshold is used as a measure of the

    mismatch. This is defined as [17]

    Vt =AVtW L

    (3.21)

    and the mismatch in the current gain factor is given by [17]:

    = ATW L

    (3.22)

    where AVt , AT are process-dependent constants and AVt decreases with gate oxide

    thickness. From the above equations, it is observed that the large gate area used to reduce

    the flicker noise can also reduce the threshold and gain factor mismatches, and therefore

    37

  • 7/29/2019 MS Thesis ZhinengZhu

    49/81

    the offset between the input pair. This can be intuitively understood if we consider the

    flicker noise as a low frequency signal and the offset also as a low frequency signal.

    The efforts to reduce flicker noise will also reduce offset. So, the mechanisms to reduce

    noise are consistent with reduction of the input-referred offset.

    Even though the input-referred random offset of this design is theoretically low,

    it still deserves further reduction because of the systematic offset and the remaining ran-

    dom offset. Conventional offset reduction techniques such as on-wafer trimming used

    in bipolar technology is very expensive and prevents real-time trimming. Some state-of-

    art dynamic offset reduction schemes such as auto-zeroing [9] and chopper stabilization

    [18] methods can reduce the offset down to several microvolts, but they often alias high-

    frequency noise down to the base band. Moreover, there is unavoidable charge injection

    and clock feed through. These issues limit dynamic offset reduction schemes to low

    frequency applications. Some other methods, such as the ping-pong amplifier [19], use

    cascode transistors as a load, and adjust the current running through those cascode tran-

    sistors. This method will reduce the voltage headroom at the output of the first stage and

    is problematic in low voltage designs.

    In this design, a digital trimming method for offset reduction is introduced and

    analyzed. Such a method can provide offset trimming at power-up or upon request. A

    schematic representation [19] of this method is shown in Figure 3.17. During the offset

    calibration phase, the operational amplifiers inputs are connected to a common voltage

    source. The counter keeps counting and a new controlling voltage is applied to the

    auxiliary port of the operational amplifier. Then, the operational amplifier generates a

    new output which is compared with a reference voltage. The counter will stop counting

    when comparator experiences a zero crossing. When the output of comparator changes

    its state, the calibration phase is concluded and the output of the DAC is set to the

    desired controll voltage. In Figure 3.17, the operational amplifier is in the open loop

    mode. When the open loop DC gain of the amplifier is very large, such a configuration

    38

  • 7/29/2019 MS Thesis ZhinengZhu

    50/81

    Comparator

    DAC

    CO=1?

    Vc

    Cross Zero?

    Counter

    12-bit

    Vref

    CO yes

    Stop

    Op AmpVOS

    ComparatorComparator

    DAC

    CO=1?

    Vc

    Cross Zero?

    Counter

    12-bit

    Vref

    CO yes

    Stop

    Op AmpVOSOp AmpVOS

    Figure 3.17: Offset tuning scheme with operational amplifier in open loop mode

    is not very efficient to reduce the offset to make the output fixed within the supply rail

    unless using the big size auxiliary input pair at the cost of more noise. But in practical

    application, the operational amplifiers are seldom used in an open-loop configuration.

    They are often used in an inverting, non-inverting or transimpedance configuration with

    a gain much less than open-loop DC gain. An example of inverting configuration is

    shown in Figure 3.18. The amplifier can keep inverting, non-inverting or transimpedance

    configuration when offset trimming is taking place.

    For this offset calibration, an auxiliary port in introduced to this operational

    amplifier [20] like that shown in Figure 3.10. The size and bias current of the auxiliary

    port are chosen to cover a 10mV input referred offset without adding too much extra

    noise. The auxiliary input size and offset adjustable range are determined by the setup

    shown in Figure 3.19. The operational amplifiers primary port is in the unity-gain

    configuration. A 10mV offset is applied to the operational amplifiers positive input of

    the primary port. The desired DC output level is 0V. The voltage at the positive input of

    auxiliary port is biased at a DC level and the negative input is DC swept from 0V to 3.3V.

    The simulation result in Figure 3.20 shows that the output experiences a change from

    39

  • 7/29/2019 MS Thesis ZhinengZhu

    51/81

    12-bitOp Amp

    Comparator

    DAC

    CO=1?

    Vc

    Cross Zero?

    Counter

    VOS

    Vref

    CO yes

    Stop

    12-bitOp Amp

    ComparatorComparator

    DAC

    CO=1?

    Vc

    Cross Zero?

    Counter

    VOS

    Vref

    CO yes

    Stop

    Figure 3.18: Offset tuning scheme with operational amplifier inverting-configured

    VOS

    Vsweep

    Vref

    VOutVOS

    Vsweep

    Vref

    VOut

    Figure 3.19: Setup for auxiliary input design

    40

  • 7/29/2019 MS Thesis ZhinengZhu

    52/81

    0 0.5 1 1.5 2 2.5 3

    0

    0.002

    0.004

    0.006

    0.008

    0.01

    0.012

    0.014

    0.016

    0.018

    0.02

    utput

    Input(V)

    Output Sweep

    Desired Output Level

    Figure 3.20: DC sweep of the auxiliary input

    20mV down to less than desired DC output level (0V), which means the chosen size of

    auxiliary port can cover the selected offset range. The bias current in this auxiliary port

    is 10 times less than that of primary port; the aspect ratio is 8 times less than that in

    the primary port. Figure 3.21 shows the pre-layout noise performance of the circuits in

    Figure 3.1 and Figure 3.10 with the auxiliary port, just a little difference in noise level.

    The detailed offset tuning and zero-crossing detection circuit is shown in Figure

    3.22. In this figure, there is a DAC, which consists of a segmented wide-swing 12-

    bit R-2R configuration [21] shown in Figure 3.23. The 12-bit counters upper 5 bits

    are thermo-coded. This makes the accuracy requirement for 1/2 LSB DNL in a 12-bit

    converter to be set by 7-bit matching instead of 12-bit. Such segmentation is also helpful

    in reducing the glitch area associated with the changing DAC output. Since the output

    of DAC drives a capacitive load, the operational amplifier as a buffer in this DAC is

    omitted.

    CNT includes a positive edge triggered 2-bit counter, the counters two output

    bits are anded to give C1 output. Therefore receiving three clocks after reset, C1 is set

    to 1 to enable DAC and disable itself. During this three-clock period D2, which is two

    D flip-flops connected in series, stores the comparators initial status and XOR becomes

    41

  • 7/29/2019 MS Thesis ZhinengZhu

    53/81

    100

    101

    102

    103

    104

    105

    106

    107

    0

    1

    2

    3

    4

    5

    6

    7

    8

    9x 10

    -7

    Frequency(Hz)

    noise

    level(v/

    (H

    z))

    W ith Auxi l iary Port

    W i thout Auxi l iary Port

    Figure 3.21: Noise level with and without auxiliary input

    CO

    Op Amp

    Comparator

    1D

    1D

    DAC

    Vref+

    Vref-Clr

    Clk

    Vc

    S

    CAL

    Out

    A1

    XORD2

    Clkin

    D1

    Clk

    Clr

    Clk

    D1

    Clr

    ClkC1

    CNTClr

    Clk

    A2En

    I1

    I2

    I3

    I4

    CO

    Op AmpOp Amp

    ComparatorComparator

    1D

    1D

    1D1D

    1D1D

    DAC

    Vref+

    Vref-Clr

    Clk

    VcDAC

    Vref+

    Vref-Clr

    Clk

    Vc

    S

    CAL

    Out

    A1A1

    XORD2

    Clkin

    D1

    ClkClkin

    D1

    Clk

    Clr

    Clk

    D1

    Clr

    Clk

    D1

    Clr

    Clk

    Clr

    ClkC1

    CNTClr

    Clk

    A2En

    I1

    I2

    I3

    I4

    Figure 3.22: Diagram of offset tuning block

    42

  • 7/29/2019 MS Thesis ZhinengZhu

    54/81

    5-bit

    Counter

    7-bit

    Counter

    Thermo-

    Decoder

    31-bit

    Register

    7-bit

    Register R

    esistor

    S

    tring

    315

    7

    38

    5-bit

    Counter

    7-bit

    Counter

    Thermo-

    Decoder

    31-bit

    Register

    7-bit

    Register R

    esistor

    S

    tring

    315

    7

    38

    Figure 3.23: Connection between counter and DAC

    0. If CLK[n-1] comparators output changes from 1 to 0, D2 is still 1 and XOR

    gates output becomes 1. Therefore, there is zero crossing for comparator within two

    continuous clocks when the XOR equals 1. D1 is a one-bit D flip-flop, which turns off

    the clock signal Clkin when offset tuning phase is finished to reduce switching noise

    in the chip. Also D1 is used to switch the operational amplifiers output into or out of

    the calibration circuit. The Schmidt Trigger is used to reduce the switching noise in the

    comparators output.

    The timing for the offset tuning is shown in Figure 3.24. At the beginning, a

    Clr signal (command to start offset tuning) is activated, all the registers are cleared,

    and DAC is disabled. The period of Clr should be long enough for the operational

    amplifier and comparators output to become stable. In this figure, the initial output of

    the operational amplifier is larger than the reference voltage and the comparators output

    is 1. At the time 1, the reset signal is released and the counter CNT begins to count.

    The comparators initial status is clocked into D2 at time 2 and the output of the XOR

    gate is

    0

    . This

    0

    will be kept until comparator changes its output status by having azero crossing. At time 3, which is the third clock edge after the reset signal, the counters

    output is set to 1 to disable the counter and enable the DAC. The DAC starts to increase

    its output. When the DAC increases its output, the operational amplifier generates a

    new output, which is compared with a reference DC voltage through comparator. This

    43

  • 7/29/2019 MS Thesis ZhinengZhu

    55/81

    process continues until at time 4, the operational amplifiers output becomes less than

    the reference voltage. At this time, the comparator changes its output to low and D2

    is still comparators previous status, which makes the XOR change its status to logical

    1

    to indicate a zero-crossing. Then A1 is set to

    0

    to disable the DAC and conclude

    the offset tuning process. At the time 5, D1 changes to 1 to switch the operational

    amplifiers output back to the circuit and disable Clkin.

    Figure 4.11 shows the simulation of the offset tuning. In this simulation, the

    operational amplifier works with a single supply voltage with the midpoint voltage at

    1.65V. Figure 3.25(a) shows the timing of signal D1 in Figure 3.22. It shows that the

    calibration phase is concluded at the time of0.2805s. In Figure 3.25(b), we can see that

    the DAC keeps increasing its output until at the time 0.2805s when the output of the

    operational amplifier becomes less than the reference voltage. Figure 3.25(c) shows the

    waveform of signal CAL in Figure 3.22. This waveform from 0s to 0.2805s resembles

    the waveform in Figure 3.20. After the calibration phase, signal CAL becomes float-

    ing, and is no longer used. Figure 3.25(d) shows the waveform Out in Figure 3.22.

    The signal Out is floating during the time from 0s to 0.2805s. After the time 0.2805s,

    the operational amplifiers output is switched back to the Out signal and its value is

    close to 1.65V as shown in Figure 3.25(d).

    44

  • 7/29/2019 MS Thesis ZhinengZhu

    56/81

    Clk

    _in

    Clk

    Clr

    C1

    A1

    A1

    D1

    D2

    S XOR

    CO

    Vc

    1

    2

    3

    4

    5

    OpAmp

    Clk

    _in

    Clk

    _in

    ClkClk

    ClrClr

    C1