Professional Documents
Culture Documents
INTRODUCTION
Course Objective
This course teaches analog integrated circuit design using CMOS technology.
VDD
VPB1
I4 I5
SPECIFICATIONS
M4 M5
I1 I2 VPB2
I6 I7
vOUT
M6 M7
+
M1 M2 VNB2
vIN
- M8 M9 CL
VNB1 M3 I
3 M10 M11
070209-01
Course Prerequisites
• Basic understanding of electronics
- Active and passive components
- Large and small signal models
- Frequency response
• Circuit analysis techniques
- Mesh and loop equations
- Superposition, Thevenin and Norton’s equivalent circuits
• Integrated circuit technology
- Basics process steps
- PN junctions
Appendix E Chapter 9
Switched Capaci- D/A and A/D
tor Circuits Converters
Systems
Chapter 4 Chapter 5
CMOS CMOS
Subcircuits Amplifiers
Simple
Circuits
Chapter
Chapter10
2 Chapter
Chapter11
3
CMOS/BiCMOS
D/A and A/D CMOS/BiCMOS
Analog
Technology
Converters Modeling
Systems
Devices
Introduction
070209-02
References
1.) P.E. Allen and D.R. Holberg, CMOS Analog Circuit Design – 3rd Ed., Oxford
University Press, 2012.
2.) P.R. Gray, P.J. Hurst, S.H. Lewis and R.G. Meyer, Analysis and Design of Analog
Integrated Circuits – 4th Ed., John Wiley and Sons, Inc., 2001.
3.) B. Razavi, Design of Analog CMOS Integrated Circuits, McGraw-Hill, Inc., 2001.
4.) R.J. Baker, H.W. Li and D.E. Boyce, CMOS Circuit Design, Layout, and
Simulation, IEEE Press, 1998.
5.) D. Johns and K. Martin, Analog Integrated Circuit Design, John Wiley and Sons,
Inc., 1997.
6.) K.R. Laker and W.M.C. Sansen, Design of Analog Integrated Circuits and Systems,
McGraw-Hill, Inc., 1994.
7.) R.L. Geiger, P.E. Allen and N.R. Strader, VLSI Techniques for Analog and Digital
Circuits, McGraw-Hill, Inc., 1990.
8.) A. Hastings, The Art of Analog Layout – 2nd Ed., Prentice-Hall, Inc., 2005.
9.) J. Williams, Ed., Analog Circuit Design - Art, Science, and Personalities,
Butterworth-Heinemann, 1991.
10.) R.A. Pease, Troubleshooting Analog Circuits, Butterworth-Heinemann, 1991.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 01 – Introduction (7/6/15) Page 01-6
Course Philosophy
This course emphasizes understanding of analog integrated circuit design.
Although simulators are very powerful, the designer must understand the circuit before
using the computer to simulate a circuit.
M6
Circuit or Analog M3 M4 Cc
vout
systems Integrated -
vin
M1 M2 CL
-
VSS
Topology
The electrical design requires active and passive device electrical models for
- Creating the design
- Verifying the design
- Determining the robustness of the design
Ground
+5V vout
M2
vout
vin
M2
(2.5V) (2.5V)
M1
M1 p+
p+
n+
te
n+
tra
bs
p-well
su
n-
n-substr
ate
031113-01
5V vin
Packaging†
Packaging of the integrated circuit is an important part of the physical design process.
The function of packaging is:
1.) Protect the integrated circuit
2.) Power the integrated circuit
3.) Cool the integrated circuit
4.) Provide the electrical and mechanical connection between the integrated circuit and
the outside world.
Packaging steps:
Attachment Connecting Encapsulating the
Dicing
of the chip to the chip to chip and lead
the wafer
a lead frame a lead frame frame in a package
031115-01
Types of tests:
• Functional – verification of the nominal specifications
• Parametric – verification of the characteristics to within a specified tolerance
• Static – verification of the static (AC and DC) characteristics of a circuit or system
• Dynamic – verification of the dynamic (transient) characteristics of a circuit or system
Additional Considerations:
Should the testing be done at the wafer level or package level?
How do you remove the influence (de-embed) of the measurement system from the
measurement?
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 01 – Introduction (7/6/15) Page 01-14
Understanding Technology
Understanding technology helps the analog IC designer to know the limits of the
technology and the influence of the technology on the design.
Device Parasitics:
Connection Parasitics:
+5V
M2
M2 vout
vin vout vin
+5V
M1
M1
050304-01
Understanding Modeling
Modeling:
Modeling is the process by which the electrical properties of an electronic circuit or
system are represented by means of mathematical equations, circuit representations,
graphs or tables.
Models permit the predicting or verification of the performance of an electronic
circuit or system.
Examples:
Ohm’s law, the large signal model of a MOSFET, the I-V curves of a diode, etc.
Goal:
Models that are simple and allow the designer to understand the circuit performance.
Assumptions
Assumptions:
An assumption is taking something to be true without formal proof. Assumptions in
analog circuit design are used for simplifying the analysis or design. The goal of an
assumption is to separate the essential information from the nonessential
information of a problem.
The elements of an assumption are:
1.) Formulating the assumption to simplify the problem without eliminating the
essential information.
2.) Application of the assumption to get a solution or result.
3.) Verification that the assumption was in fact appropriate.
Examples:
Neglecting a large resistance in parallel with a small resistance
Miller effect to find a dominant pole
Finding the roots of a second-order polynomial assuming the roots are real and
separated
• As analog is combined with more digital, substrate interference will become worse
150623-02
Idm
id
ID iD
t
Fig. 1.4-1
Enhancement Enhancement
G NMOS with G PMOS with
VBS = 0V. VBS = 0V.
S D
D S
Enhancement Enhancement
G B NMOS with G B PMOS with
VBS ¹ 0V. VBS ¹ 0V.
S D
D S
Simple Simple
G NMOS G PMOS
symbol symbol
S D
- - -
Voltage-controlled, Voltage-controlled,
voltage source current source
I1 I1 I2
RmI1 +- V2 AiI1
-
Current-controlled, Current-controlled,
voltage source current source
Three-Terminal Notation
QABC
A = Terminal with the larger magnitude of potential
B = Terminal with the smaller magnitude of potential
C = Condition of the remaining terminal with respect to terminal B
C = 0 There is an infinite resistance between terminal B and the 3rd terminal
C = S There is a zero resistance between terminal B and the 3rd terminal
C = R There is a finite resistance between terminal B and the 3rd terminal
C = X There is a voltage source in series with a resistor between terminal B
and the 3rd terminal in such a manner as to reverse bias a PN
junction. IDSS
S D S D
Examples
D +
- G
VGS CDGS IDS BVDGO
+
G S G -
SUMMARY
• Successful analog IC design proceeds with understanding the circuit before simulation.
• Analog IC design consists of three major steps:
1.) Electrical design Topology, W/L values, component values and dc currents
2.) Physical design (Layout)
3.) Test design (Testing)
• Analog designers must be flexible and have a skill set that allows one to simplify and
understand a complex problem
• Analog IC design has reached maturity and is here to stay.
• The appropriate philosophy is “If it can be done economically by digital, don’t use
analog”.
• As a result of the above, analog finds applications where speed, area, or power result in
advantages over a digital approach.
• Deep-submicron technologies will offer exciting challenges to the creativity of the
analog designer.
• Paradigm for IC design might be changing which would influence analog IC design.
CMOS TECHNOLOGY
Categorization of CMOS Technology
• Minimum feature size as a function of time:
1
Submicron Technology
Minimum Feature Size (µm)
0.1
0.01
1985 1990 1995 2000 2005 2010 2020
Year 120327-01
Therefore,
• Almost every comparison favors the BJT, however a similar comparison made from
digital viewpoint would come up on the side of CMOS.
• Therefore, since large-volume mixed-mode technology will be driven by digital
demands, CMOS is an obvious result as the technology of availability.
Oxidation
Description:
Oxidation is the process by which a layer of silicon dioxide is grown on the
surface of a silicon wafer.
Original silicon surface tox
Silicon dioxide
Uses:
• Protect the underlying material from contamination
• Provide isolation between two layers.
Very thin oxides (100Å to 1000Å) are grown using dry oxidation techniques.
Thicker oxides (>1000Å) are grown using wet oxidation techniques.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 02 – Submicron CMOS Technology (12/9/13) Page 02-6
Diffusion
Diffusion is the movement of impurity atoms at the surface of the silicon into the
bulk of the silicon.
Always in the direction from higher
High Low
concentration to lower concentration. Concentration
Concentration
Fig. 150-04
NB NB
t3 t2 t3
t1 t2 t1
Depth (x) Depth (x)
Infinite source of impurities at the surface. Finite source of impurities at the surface.
Fig. 150-05
Ion Implantation
Ion implantation is the process by Path of
impurity
which impurity ions are accelerated to atom
a high velocity and physically lodged
into the target material. Fixed Atom
Fixed Atom
• Annealing is required to activate the Fixed Atom
Impurity Atom
impurity atoms and repair the physical
final resting place
damage to the crystal lattice. This step Fig. 150-06
is done at 500 to 800°C.
• Ion implantation is a lower temperature
process compared to diffusion. Concentration peak
N(x)
• Can implant through surface layers, thus it
is useful for field-threshold adjustment.
• Can achieve unique doping profile such as
buried concentration peak. NB
Generally implant first then use diffusion
to achieve the well or active area. 0 Depth (x) Fig. 150-07
Deposition
Deposition is the means by which various materials are deposited on the silicon
wafer.
Examples:
• Silicon nitride (Si3N4)
• Silicon dioxide (SiO2)
• Aluminum
• Polysilicon 120521-01
Etching
Etching is the process of selectively Mask
Film
removing a layer of material.
When etching is performed, the etchant Underlying layer
may remove portions or all of: (a) Portion of the top layer ready for etching.
a Selectivity
• The desired material
Mask
• The underlying layer Film c
• The masking layer b Anisotropy
Selectivity Underlying layer
Important considerations:
(b) Horizontal etching and etching of underlying layer.
• Anisotropy of the etch is defined as, Fig. 150-08
Epitaxy
Epitaxial growth consists of the formation of a layer of single-crystal silicon on
the surface of the silicon material so that the crystal structure of the silicon is
continuous across the interfaces.
• It is done externally to the material as opposed to diffusion which is internal
• The epitaxial layer (epi) can be doped differently, even opposite to the material on
which it is grown
• It is accomplished at high temperatures using a chemical reaction at the surface
• The epi layer can be any thickness, typically 1-20 microns
Si Si Si Si Si
- Si Si Si Si Si Si Si
-
Si Si Si Si Si Si Si Si Si Si Si Si
- -
Si Si Si Si Si Si Si Si Si Si Si Si
Fig. 150-09
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 02 – Submicron CMOS Technology (12/9/13) Page 02-11
Photolithography
Components:
• Photoresist material
• Mask
• Material to be patterned (e.g., oxide)
Positive photoresist:
Areas exposed to UV light are soluble in the developer
Negative photoresist:
Areas not exposed to UV light are soluble in the developer
Steps:
1. Apply photoresist
2. Soft bake (drives off solvents in the photoresist)
3. Expose the photoresist to UV light through a mask
4. Develop (remove unwanted photoresist using solvents)
5. Hard bake ( 100°C)
6. Remove photoresist (solvents)
Photoresist
Polysilicon
Fig. 150-10
Develop
Polysilicon
Photoresist Etch
Photoresist
Polysilicon
Remove
photoresist
Polysilicon
Fig. 150-11
SiO2
Photoresist Photoresist
p- substrate
n-well
p- substrate
070523-01
Photoresist
Si3N4 Photoresist Pad oxide (SiO 2)
n-well
p- substrate
Si3N4
Photoresist
n-well
p- substrate
070523-02
FOX FOX
n-well
p- substrate
Step 6.) Growth of the gate thin oxide and deposition of polysilicon. The thresholds
can be shifted by an implantation before the deposition of polysilicon.
Polysilicon
FOX FOX
n-well
p- substrate
070523-03
Step 8.) Implantation of NMOS source and drain and contact to n-well (not shown)
n+ S/D implant
Polysilicon
Photoresist
FOX FOX FOX
FOX
n-well
p- substrate
070523-04
Polysilicon Photoresist
FOX FOX FOX
FOX
n-well
p- substrate
Step 10.) Implant the PMOS source/drains and contacts to the p - substrate (not shown),
remove the sidewall spacers and implant the PMOS lightly doped source/drains
LDD Diffusion
Polysilicon
070209-03
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 02 – Submicron CMOS Technology (12/9/13) Page 02-20
070523-05
BPSG
FOX FOX FOX
n-well
p- substrate
Step 14.) Deposit another interlayer dielectric (CVD SiO 2), open contacts,
deposit second level metal
Metal 2
Metal 1
BPSG
FOX FOX FOX
n-well
p- substrate
070523-06
BPSG
FOX FOX FOX
n-well
p- substrate
070523-07
p-well process is similar but starts with a p-well implant rather than an n-well implant.
Metal 4
Metal 3
Metal 2
2 microns
Metal 1
Polysilicon
Diffusion 070523-08
Planarization
Planarization attempts to minimize the variation in surface height of the wafer.
Planarization techniques
• Repeated applications of SOG Tungsten
Plug
• Resist etch-back – highest areas of oxide
are exposed longest to the etchant and
therefore erode away the most.
Comments:
• Both chemical effect (slurry) and mechanical (pad pressure) take place.
• Although CMP is superior to SOG and resist etchback, large areas devoid of underlying
metal or poly produce low regions in the final surface.
• Challenge: Achieve a highly planarized surface over a wide range of pattern density.
Silicide/Salicide Technology
Used to reduce interconnect resistivity by placing a low-resistance silicide such
as TiSi2, WSi2, TaSi2, etc. on top of polysilicon
Salicide technology (self-aligned silicide) provides low resistance source/drain
connections as well as low-resistance polysilicon.
Polysilicide Polysilicide
Metal Metal
Salicide
SUMMARY
• Fabrication is the means by which the circuit components, both active and
passive, are built as an integrated circuit.
• Basic process steps include:
1.) Oxide growth 2.) Thermal diffusion 3.) Ion implantation
4.) Deposition 5.) Etching 6.) Epitaxy
• The complexity of a process can be measured in the terms of the number of
masking steps or masks required to implement the process.
• Major CMOS Processing Steps:
1.) Well definition
2.) Definition of active areas and substrate/well contacts (SiNi3)
3.) Thick field oxide (FOX)
4.) Thin field oxide and polysilicon
5.) Diffusion of the source and drains (includes the LDD)
6.) Dielectric layer/Contacts (planarization)
7.) Metallization
8.) Dielectric layer/Vias
ion h
Trench Isolation
lat nc
ow
Iso Tre
all
ow
Sh
a ll
Sh
n+ p+ p+ nn++ nnn+++
Substrate 070330-03
Comments:
• If the n+ to p+ spacing is large, the Bird’s beak can be compensated using techniques
such as poly buffered LOCOS
• At some point as the n+ to p+ spacing gets smaller, the restricted bird’s beak leads to
undesirable stress effects in the transistor.
• An important advantage of STI is that it minimizes the heat cycle needed for n+ or p+
isolation compared to LOCOS. This is a significant advantage for any process where
there are implants before STI.
(3)
3.) Grow a thin thermal oxide layer on the trench walls.
(4)
4.) A CVD dielectric film is used to fill the trench.
(5)
5.) A chemical mechanical polishing (CMP) step is used to
polish back the dielectric layer until the nitride is reached.
The nitride acts like a CMP stop layer. (6)
6.) Densify the dielectric material at 900°C and strip the 060203-01
Transistors
fT as a function of gate-source overdrive, VGS-VT (0.13µm):
Typical, 25°C
70
60 NMOS
Slow, 70°C
50
fT (GHz)
40 Typical, 25°C
20
10
0
0 100 200 300 400 500
|VGS-VT| (mV) 030901-07
The upper frequency limit of the transistors varies with overdrive and process corners.
The NMOS transistor has an fT of 40GHz at low overdrives and increases to above
60GHz at the slow-high temperature corner with 0.5V overdrive.
Resistors
1.) Diffused and/or implanted resistors.
2.) Well resistors.
3.) Polysilicon resistors.
4.) Metal resistors.
Capacitors
Polysilicon-polysilicon
capacitors:
Metal-metal capacitors:
Protective Insulator Layer
Metal Via
Top
Vias connecting top Metal
Capacitor plate to top metal
dielectric Capacitor Top Metal Second level
Inter-
Vias connecting bottom from top metal
mediate
Capacitor bottom plate plate to lower metal Third level
Oxide
Layers Vias connecting bottom from top metal
plate to lower metal Fourth level
from top metal
060530-01
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 03 – DSM CMOS Technology (11/16/15) Page 03-10
Inductors
Top view and cross-section of a planar inductor:
Top Metal
Top Metal
W
S Next Level
Metal
Vias Oxide
Next Level Oxide
Metal
D
Silicon Substrate
N turns
030828-01
D
†
M. Raieszadeh, Integrated Inductors on Trenched Silicon Islands, MS Thesis, School of Electrical and Computer Engineering, Georgia Institute of Technology, April 2005
Starting Material
The substrate should be highly doped to act like a good conductor.
Substrate
p+ n+
n-well p-well
Substrate
p+ n+
Shallow Shallow Shallow
Trench Trench Trench
Isolation Isolation Isolation
n-well p-well
Substrate
n+
Shallow Shallow Shallow
Trench Trench Trench
Isolation Isolation Isolation
n-well p-well
Substrate
p+ n+
Shallow Shallow Shallow
Trench Trench Trench
Isolation Isolation Isolation
n-well p-well
Substrate
p+ n+
Shallow Shallow Shallow
Trench Trench Trench
Isolation Isolation Isolation
n-well p-well
Substrate
Sidewall Sidewall
Spacers Spacers
p+ n+
Shallow Shallow Shallow
Trench Trench Trench
Isolation Isolation Isolation
n-well p-well
Substrate
n+ p+ p+ n+ n+ p+
implant implant implant implant implant implant
n+ p+ p+ n+ nn++ p+
Shallow Shallow Shallow
Trench Trench Trench
Isolation Isolation Isolation
n-well p-well
Substrate
Polycide Polycide
Substrate
Completed Fabrication
After multiple levels of metal are applied, the fabrication is completed with a thicker top-
level metal and a protective layer to hermetically seal the circuit from the environment.
Note that metal is used for the upper level metal vias. The chip is electrically connected
by removing the protective layer over large bonding pads.
Tungsten Plug
TEOS
SOG
Polycide
Sidewall
TEOS/BPSG Spacer
Poly
Gate
Fig. 2.8-20
Metal 3
Aluminum
Vias
Metal 2
Tungsten
Plugs
Metal 1
Transistors Fig.180-11
SUMMARY
• DSM technology typically has a minimum channel length between 0.35µm and 0.1µm
• DSM technology addresses the problem of excessive depletion region widths in
junction isolation techniques by using shallow trench isolation
• DSM technology may have from 4 to 8 levels of metal
• Lightly doped drains and sources are a key aspect of DSM technology
220 nm pitch
NMOS
†P. Bai, et. Al., “A 65nm Lobic Technology Featuring 35nm Gate Lengths, Enhanced Channel Strain, 8 Cu Interconnect Layers, Low-k ILD and
0.57 µm2 SRAM Cell, IEEE Inter. Electron Device Meeting, Dec. 12-15, 2005.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 04 – UDSM and BiCMOS Technologies (3/10/14) Page 04-4
†Anne-Johan Annema, et. Al., “Analog Circuits in Ultra-Deep-Submicron CMOS,” IEEE J. of Solid-State Circuits, Vol. 40, No. 1, Jan. 2005, pp.
132-143.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 04 – UDSM and BiCMOS Technologies (3/10/14) Page 04-7
- +
f(vGD) vGD f(vSG) vSG ggd gsg
+ -
+ -
f(vGS) vGS f(vDG) vDG ggs gdg
- +
BiCMOS TECHNOLOGY
Typical 0.5µm BiCMOS Technology
Masking Sequence:
1. Buried n+ layer 9. Base oxide/implant 17. Contacts
2. Buried p+ layer 10. Emitter implant 18. Metal 1
3. Collector tub 11. Poly 1 19. Via 1
4. Active area 12. NMOS lightly doped drain 20. Metal 2
5. Collector sinker 13. PMOS lightly doped drain 21. Via 2
6. n-well 14. n+ source/drain 22. Metal 3
7. p-well 15. p+ source/drain 23. Nitride passivation
8. Emitter window 16. Silicide protection
Notation used in the following slides:
BSPG = Boron and Phosphorus doped Silicate Glass (oxide)
Kooi Nitride = A thin layer of silicon nitride on the silicon surface as a result of the
reaction of silicon with the HN3 generated, during the field oxidation.
TEOS = Tetro-Ethyl-Ortho-Silicate. A chemical compound used to deposit conformal
oxide films.
Starting Substrate:
p-substrate 1mm
BiCMOS-01 5mm
p-substrate 1mm
BiCMOS-02 5mm
Epitaxial Growth
p-type
n-well p-well n-well p-well
Epitaxial
Silicon
n+ buried layer p+ buried n+ buried layer
p+ buried layer
layer
p-substrate 1mm
BiCMOS-03 5mm
Comment:
• As the epi layer grows vertically, it assumes the doping level of the substrate beneath
it.
• In addition, the high temperature of the epitaxial process causes the buried layers to
diffuse upward and downward.
Collector Tub
p-substrate 1mm
BiCMOS-04 5mm
Comment:
• The collector area is developed by an initial implant followed by a drive-in diffusion to
form the collector tub.
Comment:
• The silicon nitride is use to impede the growth of the thick oxide which allows contact
to the substrate
• -silicon is used for stress relief and to minimize the bird’s beak encroachment
Field Oxide
NPN Transistor
FOX PMOS Transistor NMOS Transistor
p-substrate 1mm
BiCMOS-06 5mm
Comments:
• The field oxide is used to isolate surface structures (i.e. metal) from the substrate
p-substrate 1mm
BiCMOS-07 5mm
Base Definition
NPN Transistor
FOX PMOS Transistor NMOS Transistor
p-substrate 1mm
BiCMOS-08 5mm
p-substrate 1mm
BiCMOS-09 5mm
Emitter Implant
NPN Transistor
Emitter Implant PMOS Transistor NMOS Transistor
FOX
FOX Field Oxide Field Oxide Field Oxide
Collector Tub n-well
p-well p-well p-type
Sub-Collector Epitaxial
Silicon
p-substrate 1mm
BiCMOS-10 5mm
Comments:
• The polysilicon above the base is implanted with n-type carriers
Emitter Diffusion
NPN Transistor
FOX PMOS Transistor NMOS Transistor
p-substrate 1mm
BiCMOS-11 5mm
Comments:
• The polysilicon not over the emitter window is removed and the n-type carriers diffuse
toward the base forming the emitter
NPN Transistor
FOX PMOS Transistor NMOS Transistor
p-substrate 1mm
BiCMOS-12 5mm
Comments:
• The surface of the region where the MOSFETs are to be built is cleared and a thin gate
oxide is deposited with a polysilicon layer on top of the thin oxide
• The polysilicon is removed over the source and drain areas
• A light source/drain diffusion is done for the NMOS and PMOS (separately)
NPN Transistor
FOX PMOS Transistor NMOS Transistor
p-substrate 1mm
BiCMOS-13 5mm
Comments:
• The sidewall spacers prevent the heavy source/drain doping from being near the
channel of the MOSFET
Siliciding
p-substrate 1mm
BiCMOS-14 5mm
Comments:
• Siliciding is used to reduce the resistance of the polysilicon and to provide ohmic
contacts to the base, emitter, collector, sources and drains
Contacts
Tungsten Plugs Tungsten Plugs Tungsten Plugs
TEOS/BPSG/SOG TEOS/BPSG/SOG TEOS/BPSG/SOG
FOX
FOX Field Oxide Field Oxide Field
FieldOxide
Oxide
n-well
p-well p-well p-type
Epitaxial
Silicon
p-substrate 1mm
BiCMOS-15 5mm
Comments:
• A dielectric is deposited over the entire wafer
• One of the purposes of the dielectric is to smooth out the surface
• Tungsten plugs are used to make electrical contact between the transistors and metal1
Metal1
Metal1 Metal1 Metal1
p-substrate 1mm
BiCMOS-16 5mm
Metal1-Metal2 Vias
p-substrate 1mm
BiCMOS-17 5mm
Metal2
Metal 2
Oxide/
SOG/
Oxide
TEOS/BPSG/SOG TEOS/BPSG/SOG TEOS/BPSG/SOG
FOX
p-substrate 1mm
BiCMOS-18 5mm
Metal2-Metal3 Vias
Comments:
• The metal2-metal3 vias will be filled with metal3 as opposed to tungsten plugs
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 04 – UDSM and BiCMOS Technologies (3/10/14) Page 04-28
Completed Wafer
Silicon-Germanium
Physical Perspective (130nm):
Electrical:
Max. voltage = 2V
≈ 300
fT ≈ 100 GHz
SUMMARY
• UDSM technology typically has a minimum channel length less than 0.1µm
• UDSM transistors utilize enhanced channel strains to increase drive capability and
reduce off currents
• Advantages of UDSM technology include:
- Smaller devices
- Higher speeds and transconductances
- Improved Ion/Ioff
• Disadvantages of UDSM technology include:
- Gate leakage currents
- Reduced small signal gains
- Increased nonlinearity
• BiCMOS technology
- Offers both CMOS transistors and a high performance vertical BJT
- CMOS is typically a generation behind
- Silicon germanium can be used to enhance the BJT performance
PN JUNCTIONS
How are PN Junctions used in CMOS?
• PN junctions are used to electrically isolate one semiconductor region from another
• PN diodes
• ESD protection
• Creation of the thermal voltage for bandgap purposes
• Depletion capacitors – voltage variable capacitors (varactors)
Components of a pn junction:
1.) p-doped semiconductor – a semiconductor having atoms containing a lack of
electrons (acceptors). The concentration of acceptors is NA in atoms per cubic
centimeter.
2.) n-doped semiconductor – a semiconductor having atoms containing an excess of
electrons (donors). The concentration of these atoms is ND in atoms per cubic
centimeter.
Abrupt PN Junction
Metal-semiconductor junction pn junction Metal-semiconductor junction
p+ semiconductor n semiconductor
Depletion Region
W 060121-02
p+ semiconductor n semiconductor
W1 0 x
W2
W1 = Depletion width on p side W2 = Depletion width on n side
1. Doped atoms near the metallurgical junction lose their free carriers by diffusion.
2. As these fixed atoms lose their free carriers, they build up an electric field, which
opposes the diffusion mechanism.
3. Equilibrium conditions are reached when:
Current due to diffusion = Current due to electric field
ND
W2 = 0.91nm
ND
qND1 + p-side n-side
N A 140310-01
Assume that vD = 0, o = 0.637V and ND = 1017 atoms/cm3. Find the p-side depletion
region width if NA = 1015 atoms/cm3 and if NA = 1019 atoms/cm3:
For NA = 1015 atoms/cm3 the p-side depletion width is 0.90 µm.
For NA = 1019 atoms/cm3 the p-side depletion width is 0.9 nm.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 05 – PN Junction and CMOS Transistors (8/5/14) Page 05-5
xd
ni is the intrinsic concentration of silicon. 060121-04
Reverse-Biased PN Junctions
Depletion region: xd
xd = xp + xn = W1 + W2 Influence
vD of vR on
xp = W1 vR depletion
iD region width
and - vR = 0V +
vR xd
xn = W2 vR
060121-05
- vR > 0V +
Breakdown voltage (BV):
In the reverse direction the current iD
can be written as,
-IR BV
iD = vD
vR n Reverse Forward
1 - BV
Bias Bias
060121-06
†
P. Allen and D. Holberg, CMOS Analog Circuit Design, 3rd ed., Oxford University Press, 2012
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 05 – PN Junction and CMOS Transistors (8/5/14) Page 05-8
Depletion Capacitance
Physical viewpoint of the depletion capacitance:
d
xd
- +
- +
- +
- +
- +
- +
W1 W2 060204-01 + vD -
siA siA
Cj = d = W +W
1 2
siA Cj
=
2si(o-vD) ND NA Ideal
q(ND+NA) NA + ND Cj0 Gummel-
siqNAND 1 Poon Effect
=A 2(NA+ND) o-vD Reverse Bias
Cj0
=
v
yo D
vD 0
1-
o 060204-02
Forward-Biased PN Junctions
When the pn junction is forward-biased, the potential barrier is reduced and significant
current begins to flow across the junction. This current is given by:
vD Dppno Dnnpo qAD ni2 -VGO
iD = Isexp V - 1 where Is = qA L + L ≈ L N = KT exp V 3
t p n t
Graphically, the iD versus vD characteristics are given as:
ln(iD/Is)
Decade current
change/60mV
or
Octave current
change/18mV
vD
0V 060204-03
Graded PN Junctions
In practice, the pn junction is graded rather than abrupt.
Impurity
Concentration Impurity profile
approximates a
p+
constant slope
n+ p+
Intrinsic
x Concentration
x
0
Surface Junction
060204-04
W2 = qN (N +N )
Cj0
D A D
=
v D m
1 -
o
where 0.33 m 0.5.
Metal-Semiconductor Junctions
Ohmic Junctions: A metal-semiconductor junction formed by a highly doped
semiconductor and metal.
Energy band diagram IV Characteristics
I
Vacuum Level 1
qfm Tunneling
Contact
qfs
qfB EC Resistance
EF V
EV
n-type metal Highly doped n-type 140809-02
semiconductor
MOS TRANSISTORS
Physical Structure of MOS Transistors in an n-well Technology
Substrate Salicide Substrate Salicide
Well Salicide
W W
n+ p+ p+ nn++ nnn+++
L L
Shallow Shallow
Trench Trench
Isolation n-well Isolation
p-well
Substrate 070322-02
Enhancement MOSFETs
The channel of an enhancement MOSFET is formed when the proper potential is applied
to the gate of the MOSFET. This potential inverts the material immediately below the
gate to the same type of impurity as the source and drain forming the channel.
VGS=0V 0V<VGS<VT V <V (sat) VGS>VT
VDS<VDS(sat) DS DS VDS<VDS(sat)
S G D S G D S G D
VDS VDS VDS
VT = Gate-bulk work function (MS) + voltage to change the surface potential (-2F)
+ voltage to offset the channel-bulk depletion charge (-Qb/Cox)
+ voltage to compensate the undesired interface charge (-Qss/Cox)
Qb0 Qss Qb - Qb0
VT = MS -2F - C - C - C = VT0 + |-2F + vSB| - |-2F|
ox ox ox
where Qb0 Qss 2qsiNA
VT0 = MS - 2F - C - C , = and Qb ≈ 2qNAsi(|-2F+vSB|)
ox ox Cox
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 05 – PN Junction and CMOS Transistors (8/5/14) Page 05-14
The threshold voltage for a depletion mode NMOS transistor will be negative (a negative
gate potential is necessary to attract enough holes underneath the gate to cause this
region to invert to p-type material).
10-12 VGS
0 VT
L STI L
Well/Bulk Well/Bulk
Drain Drain
W W
n-well p-well
Comments:
• Make sure to contact the source and drain with multiple contacts to evenly distribute
the current flow under the gate.
• Minimize the area of the source and drain to reduce bulk-source/drain capacitance.
041027-03
A B
Dummy Gate
Dummy Gate
Dummy Gate
Dummy Gate
A B B A GA GB
GB GA
B A
GA GB GB GA
Interdigitated, common centroid layout
DB SB/SA DA
041027-04
Cross-Coupled Transistors
Metal 2 Metal 1
Via 1
120328-03
Metal 2
Via 1
Metal 1
120328-04
enhance beta:
ßF 50 to 100 depending on the n+ p+ p+ pp++
Keeps carriers from
process STI flowing at the surface STI
and reduces 1/f noise
n-well
Substrate
060221-02
Base
Area = Vp Area = Vd
0
Distance, x
Pinch-off region xp xd
Channel
Source n+ Drain n+
Source Drain
depletion depletion
region region
Substrate depletion region
p - substrate
040920-01
DSM Architecture:
SUMMARY
• pn junction usage in CMOS include:
- Electrical isolation, pn diodes, ESD protection, depletion capacitors
• Depletion region widths are inversely proportional to the doping
• Depletion region widths are proportional to the reverse bias voltage
• Ohmic metal-semiconductor junctions require a highly doped semiconductor
• MOSFETs can be:
- Enhancement – the applied gate voltage forms the channel
- Depletion – the channel is physically constructed in fabrication
• The threshold voltage of MOSFETs consists of the following components:
- Gate bulk work function (MS)
- Voltage to change the surface potential (-2F)
- Voltage to offset the channel-bulk depletion charge (-Qb/Cox)
- Voltage to compensate the undesired interface charge (-Qss/Cox)
• Weak inversion is MOSFET operation with the gate-source voltage less than the
threshold voltage
• Layout of the MOSFET is important to its performance and matching capabilities
• Extended drain regions lead to higher voltage capability MOSFETs
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 06 – Capacitors (8/18/14) Page 06-1
LECTURE 06 - CAPACITORS
LECTURE ORGANIZATION
Outline
• Introduction
• pn junction capacitors
• MOSFET gate capacitors
• Conductor-insulator-conductor capacitors
• Deviation from ideal behavior in capacitors
• Summary
CMOS Analog Circuit Design, 3rd Edition Reference
Pages 46-52 and 654-657
INTRODUCTION
Types of Capacitors for CMOS Technology
1.) PN junction (depletion) d
xd
capacitors - +
- +
- +
- +
- +
- +
W1 W2 060204-01 + vD -
G D,S,B
n+ n+ p+
Cjunction p-well
060207-01
Characterization of Capacitors
What characterizes a capacitor?
1.) Losses in a capacitor characterized by the quality factor of a capacitor is a measure of
the imaginary to real part of the impedance or admittance
1
Q= = CRp
CRs
where Rp is the equivalent resistance in parallel with the capacitor, C, and Rs is the
electrical series resistance (ESR) of the capacitor, C.
2.) Parasitic capacitors to ground from each node of the capacitor.
3.) The density of the capacitor in Farads/area.
4.) The absolute and relative accuracies of the capacitor.
5.) The Cmax/Cmin ratio which is the largest value of capacitance to the smallest when
the capacitor is used as a variable capacitor (varactor).
6.) The variation of a variable capacitance with the control voltage.
7.) Linearity, q = Cv.
PN JUNCTION CAPACITORS
PN Junction Capacitors in a Well
Generally made by diffusion into the well.
Anode Cathode
Substrate
rD
Cj Cj Cw C VB
VA
n+ p+ n+ p+ Anode Rwj Cathode
Rwj Rwj Rw Rs
Depletion
n-well Region
p- substrate
Fig. 2.5-011
Layout:
Minimize the distance between the p+ and n+ diffusions. n+ diffusion
Fig. 2.5-1A
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 06 – Capacitors (8/18/14) Page 06-5
3 C
Large Islands Anode Cathode
2.5 80
QAnode
Small Islands R-X Cathode
2 60 Bridge Voltage
C
1.5 Anode Cathode
40
Large Islands
1 R-X Cathode
Bridge Voltage 20
0.5
0 0
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5
Cathode Voltage (V) Cathode Voltage (V) 060206-03
Terminal Small Islands (598 1.2µm x1.2µm) Large Islands (42 9µm x 9µm)
Under Test Cmax/Cmin Qmin Qmax Cmax/Cmin Qmin Qmax
Anode 1.23 94.5 109 1.32 19 22.6
Cathode 1.21 8.4 9.2 1.29 8.6 9.5
Electrons as majority carriers lead to higher Q because of their higher mobility.
The resistance, Rwj, is reduced in small islands compared with large islands higher Q
†E. Pedersen, “RF CMOS Varactors for 2GHz Applications,” Analog Integrated Circuits and Signal Processing, vol. 26, pp. 27-36, Jan. 2001.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 06 – Capacitors (8/18/14) Page 06-6
Cox S D
n+ n+ p+
n+ n+ Weak
p+
Accumulation Inv. Strong
Cjunction p-well Inversion
In this configuration, the MOSFET gate capacitor has 5 regions of operation as VGS is
varied. They are:
1.) Accumulation
2.) Depletion
3.) Weak inversion
4.) Moderate inversion
5.) Strong inversion
For the first four regions, the gate capacitance is the series 1
combination of Cox and Cj given as, C gate =
1 1
+
Cox Cj
MOSFET Gate Capacitor as a function of VGS with Bulk Fixed (Inversion Mode)
G D,S B Capacitance
n+ n+ p+ Inversion VT shift
Mode MOS if VBS ≠ 0
Cjunction p-well
0 VG-VD,S
060207-04
Conditions:
• D = S, B = VSS
• Accumulation region removed by connecting bulk to ground
• Nonlinear
• Channel resistance:
L
Ron = 12K '(V -|V |)
P BG T
• LDD transistors will give lower Q because of the increased series resistance
32
VG = 1.8V
QGate
3 30
VG = 1.8V
28
2.5 VG = 1.5V
VG = 1.5V 26
2
24
1.5 22
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5
Drain/Source Voltage (V) Drain/Source Voltage (V) 070617-06
VG =1.8V: Cmax/Cmin ratio = 2.15 (1.91), Qmax = 34.3 (5.4), and Qmin = 25.8(4.9)
†E. Pedersen, “RF CMOS Varactors for 2GHz Applications,” Analog Integrated Circuits and Signal Processing, vol. 26, pp. 27-36, Jan. 2001.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 06 – Capacitors (8/18/14) Page 06-11
Cox Cox
n+ Depletion
n+
Inversion Accumulation
VG-VD,S,B
060207-05
Conditions:
• Build the NMOS in a n-well or the PMOS in a p-well – channel is present with no bias
• Implements a variable capacitor with a larger transition region between the maximum
and minimum values.
• Reasonably linear capacitor for values of VG-VD,S,B > 0
3.2
QGate
VG = 0.6V
35
VG = 0.9V
2.8
2.4 VG = 0.3V 30
2 25
0 0.5 1 1.5 2 2.5 3 3.5 0 0.5 1 1.5 2 2.5 3 3.5
Drain/Source Voltage (V) Drain/Source Voltage (V) 070617-07
VG = 0.6V: Cmax/Cmin ratio = 1.69 (1.61), Qmax = 38.3 (15.0), and Qmin = 33.2(13.6)
†
E. Pedersen, “RF CMOS Varactors for 2GHz Applications,” Analog Integrated Circuits and Signal Processing, vol. 26, pp. 27-36, Jan. 2001.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 06 – Capacitors (8/18/14) Page 06-13
CONDUCTOR-INSULATOR-CONDUCTOR CAPACITORS
Polysilicon-Oxide-Polysilicon (Poly-Poly) Capacitors
LOCOS Technology:
A very linear capacitor
with minimum bottom
plate parasitic.
DSM Technology:
A very linear capacitor with
small bottom plate parasitic.
Metal Via
Top
Vias connecting top Metal
Capacitor plate to top metal
dielectric Capacitor Top Metal Second level
Inter-
Vias connecting bottom from top metal
mediate
Capacitor bottom plate plate to lower metal Third level
Oxide
Layers Vias connecting bottom from top metal
plate to lower metal Fourth level
from top metal
060530-01
Metal Metal
Metal 2 - + - +
Metal 1 + - + - Fig2.5-9
These capacitors are sometimes called fractal capacitors because the fractal patterns are
structures that enclose a finite area with a near-infinite perimeter.
The capacitor/area can be increased by a factor of 10 over vertical flux capacitors.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 06 – Capacitors (8/18/14) Page 06-16
030909-01
†R. Aparicio and A. Hajimiri, “Capacity Limits and Matching Properties of Integrated Capacitors, IEEE J. of Solid-State Circuits, vol. 37, no. 3,
March 2002, pp. 384-393.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 06 – Capacitors (8/18/14) Page 06-17
Vias
030909-03
Vias
030909-04
060207-07
C
A B
C
A B
Replication Principle
Based on the previous result, a way to minimize the matching error between two or more
geometries is to insure that the matched components have the same area to periphery
ratio. Therefore, the replication principle requires that all geometries have the same
area-periphery ratio.
Correct way to match
the previous capacitors
(the two C2 capacitors
are connected
together):
0.02
0.01
Unit Capacitance = 4pF
0.00
1 2 4 8 16 32 64
Ratio of Capacitors
1:2 2 2 1 1 2 2
4 4 4 4 1 1 4 4 4 4
1:4
4 4 4 4 1 1 4 4 4 4
4 4 4 4 4 4 4 4 1 1 4 4 4 4 4 4 4 4
4 4 4 4 4 4 4 4 1 1 4 4 4 4 4 4 4 4
1:8
4 4 4 4 4 4 4 4 1 1 4 4 4 4 4 4 4 4
4 4 4 4 4 4 4 4 1 1 4 4 4 4 4 4 4 4
120625-01
Top
plate Desired
parasitic Capacitor
Bottom
Bottom Plate plate
060702-08
parasitic
? ?
Sensitive to alignment errors in the Insensitive to alignment errors and the
upper and lower plates and loss of flux reaching the bottom plate is lar ger
capacitance flux (smaller capacitance). resulting in large capacitance. 060207-09
Top
Plate
†M.J. McNutt, S. LeMarquis and J.L.Dunkley, “Systematic Capacitance Matching Errors and Corrective Layout Procedures,” IEEE J. of Solid-State
Circuit, vo. 29, No. 5, May 1994, pp. 611-616.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 06 – Capacitors (8/18/14) Page 06-29
Shielding
The key to shielding is to determine and control the electric fields.
Consider the following noisy conductor and its influence on the substrate:
Increased Parasitic Capacitance
Noisy Conductor Noisy Conductor Separate
Ground
Shield
Substrate Substrate
060118-10
2Cpar +1
Bottom Plate
Cpar
2Cpar Shield
Substrate Substrate
060316-02
SUMMARY
• Capacitors are made from:
- pn junctions (depletion capacitors)
- MOSFET gate capacitors
- Conductor-insulator-conductor capacitors
• Capacitors are characterized by:
- Q, a measure of the loss
- Density
- Parasitics
- Absolute and relative accuracies
• Deviations from ideal capacitor behavior include;
- Dielectric gradients
- Edge effects (etching)
- Process biases
- Parasitics
- Voltage and temperature dependence
RESISTORS
Types of Resistors Compatible with CMOS Technology
1.) Diffused and/or implanted resistors.
2.) Well resistors.
3.) Polysilicon resistors.
4.) Metal resistors.
Characterization of Resistors
1.) Value Area = A
Area = A
L nt L
R= A Curre L
Current 050217-02
AC and DC resistance
i
2.) Linearity Linear Resistor
Does V = IR?
Velocity saturation
Velocity saturation of carriers
v
Breakdown
3.) Power Voltage 060211-01
P = VI = I2R
4.) Current
Electromigration Metal 050304-04
5.) Parasitics
R R
R
Cp Cp
060210-01
Cp 2 2
Polysilicon Resistor
N-well Resistor
Metal n+ First Level Metal
Tungsten Intermediate Tungsten
Plug Oxide Layer Plug
FOX FOX FOX n+ n+
n- well L
L STI STI
n-well
p- substrate
060214-04 LOCOS Technology p-substrate
1000-5000 ohms/square
Absolute accuracy = ±40%
Relative accuracy 5%
Temperature coefficient = 4000 ppm/°C
Voltage coefficient is large 8000 ppm/V
Comments:
• Good when large values of resistance are needed.
• Parasitics are large and resistance is voltage dependent
• Could put a p+ diffusion into the well to form a pinched resistor
Metal as a Resistor
Illustration: L1 L5 Second
A B
Inter- Level
mediate Metal
L2 Tungsten Plug Tungsten Plug L4 First
Oxide
Layers Level
L3 Metal
Salicide
Substrate
060214-05
Resistance from A to B = Resistance of segments L1, L2, L3, L4, and L5 with some
correction subtracted because of corners.
Sheet resistance:
50-70 m/ ± 30% for lower or middle levels of metal
30-40 m/ ± 15% for top level metal
Tj(°C) Tr(°C) Dt
Watch out for the current limit for metal resistors.
<85 85 1
Contact resistance varies from 5 to 10.
100 85 0.63
Tempco ≈ +4000 ppm/°C 110 85 0.48
Need to derate the current at higher temperatures: 125 85 0.32
IDC(Tj) = Dt·IDC(Tr) 150 85 0.18
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 07 – Resistors and Inductors (3/10/14) Page 07-8
060612-01
Performance:
Sheet resistivity is approximately 5-10 ohms/square
Temperature coefficients of less than 100 ppm/°C
Absolute tolerance of better than ±0.1% using laser trimming
Selectivity of the metal etch must be sufficient to ensure the integrity of the thin-
film resistor beneath the areas where metal is etched away.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 07 – Resistors and Inductors (3/10/14) Page 07-9
Substrate Substrate
Active area (diffusion) Substrate Well diffusion Active area (diffusion)
L
Metal 1
Metal 1
L
Diffusion or polysilicon resistor Well resistor 060219-01
L3
L1
L2 L4 L4 L4 L4 L4 L2
060220-01
These resistors typically have model problems because of non-uniform current flow at
the corners.
Corner corrections:
0.5
1.45 1.25
Fig. 2.6-16B
L1 W
060220-02
For good matching between link resistors, keep the link length, L1, identical.
Resistor Ending Influence:
050416-02
3.8mm 4mm
1.8mm 2mm
10mm 041020-01
The actual matching ratio due to the etching bias is,
R2 W1 4-0.2 3.8
R1 = W2 = 2-0.2 = 1.8 = 2.11 → 5.6% error in matching
Use the replication principle to eliminate this error.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 07 – Resistors and Inductors (3/10/14) Page 07-13
Dummy
Dummy
A B C A B C
041025-04
The objective is to make A = B = C. In the left-hand case, B is larger due to the slower
etch rates on both sides of B. In the right-hand case, the dummy strips have caused the
etch rates on both sides of A, B and C to be identical leading to better matching.
It may be advisable to connect the dummy strips to ground or some other low impedance
node to avoid static electrical charge buildup.
n-epi
140917-03
If A, B, and C are resistors that are to be matched, we see that the effective concentration
of B is larger than A or C because of diffusion interaction. This would cause the B
resistor to be smaller even though the geometry is identical.
Solution: Place identical dummy resistors to the left of A and right of C. Connect the
dummy resistors to a low impedance to prevent the formation of floating diffusions that
might increase the sensitivity to latchup.
Thermoelectric Effects
The thermoelectric effect, also called the Seebeck effect, is a potential difference that is
developed between two dissimilar materials that are at different temperatures. The
potential developed is given as,
V = S·T
where,
S = Seebeck coefficient ( 0.4mV/°C)
T = temperature difference between the two metals
Thus, a temperature difference between the contacts to a resistor and the resistor of 1°C
can generate a voltage of 0.4mV causing problems in certain circuits (bandgap).
+ + + + + + + +
Cold
Two possible resistor
layouts with regard to
Resistor Segment
Resistor Segment
Resistor Segment
Resistor Segment
Resistor Segment
Resistor Segment
Resistor Segment
Resistor Segment
the thermoelectric
effect:
Hot
- - - - - - - -
Thermoelectric potentials add Thermoelectric potentials cancel 041026-07
n diffused
resistor
INDUCTORS
Characterization of Inductors
1.) Value of the inductor 2r
Spiral inductor†:
L ≈ µ0n2r = 4x10-7n2r ≈ 1.2x10-6n2r
n=3
L
2.) Quality factor, Q = R
060216-02
1
3.) Self-resonant frequency: fself =
LC
4.) Parasitic and inter-winding capacitances
†H.M Greenhouse, “Design of Planar Rectangular Microelectronic Inductors,” IEEE Trans. Parts, Hybrids, and Packaging, vol. 10, no. 2, June
1974, pp. 101-109.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 07 – Resistors and Inductors (3/10/14) Page 07-19
IC Inductors
What is the range of values for on-chip inductors?
I ID I
I
Nturns = 2.5
Typically: 3 < Nturns < 5 and S = Smin for the given current
Select the OD, Nturns, and W so that ID allows sufficient magnetic flux to flow
through the center.
Loss Mechanisms:
• Skin effect
• Capacitive substrate losses
• Eddy currents in the silicon
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 07 – Resistors and Inductors (3/10/14) Page 07-22
S Next Level
Metal
Vias Oxide
Next Level Oxide
Metal
D
Silicon Substrate
N turns
030828-01
D
Inductor Modeling
Cp 37.5µ0N2a2 ox
Model: L 11D-14a Cox = W·L· t
ox
L Rs L
Rs (low freq. resistive loss)
Cox Cox W(1-e-t/)
2 2 WLCsub
R1 (eddy current substrate loss)
2
C1 R1 C1 R1 ox
Cp = 2
NW L· t (overlap and coupling)
ox
030828-02
2
where C1 WLC (substrate capacitance)
sub
µ0 = 4x10-7 H/m (vacuum permeability)
= conductivity of the metal
a = distance from the center of the inductor to the middle of the windings
L = total length of the spiral
t = thickness of the metal
= skin depth given by = 2/Wµ0
Gsub(Csub) is a process-dependent parameter
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 07 – Resistors and Inductors (3/10/14) Page 07-24
Example →
Fig. 2.5-12
Inductor Improvements
Symmetrical Layout:
• Good for differential circuits
• Higher Q
• Can achieve a center tap 131001-02
17 µm metal
Q Improvement:
• Substrate replaced with trenched silicon islands†
†
M. Raieszadeh, Integrated Inductors on Trenched Silicon Islands, MS Thesis, School of Electrical and Computer Engineering, Georgia Institute of Technology, April 2005
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 07 – Resistors and Inductors (3/10/14) Page 07-26
Q = 5-6, fSR = 30-40GHz. Q = 10-11, fSR = 15-30GHz1. Good for high L in small area.
1The skin effect and substrate loss appear to be the limiting factor at higher frequencies of self-resonance.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 07 – Resistors and Inductors (3/10/14) Page 07-27
Inductors - Continued
Self-resonance as a function of inductance. Outer dimension of inductors.
Transformers
Transformer structures are easily obtained using stacked inductors as shown below for a
1:2 transformer.
Method of reducing the
inter-winding capacitances.
Transformers – Continued
A 1:4 transformer:
Structure- Measured voltage gain-
Secondary
Summary of Inductors
Scaling? To reduce the size of the inductor would require increasing the flux density
which is determined by the material the flux flows through. Since this material will not
change much with scaling, the inductor size will remain constant.
Increase in the number of metal layers will offer more flexibility for inductor and
transformer implementation.
Performance:
• Inductors
Limited to nanohenrys
Very low Q (3-5)
• Transformers
Reasonably easy to build and work well using stacked inductors
ASITIC† – A CAD tool that aids the RF circuit designer to optimize and model spiral
inductors, transformers, capacitors, and substrate coupling. ASITIC calculations include
the electrically induced losses and coupling as well as the magnetically induced eddy
current losses. Skin effect and proximity effects, or eddy currents in the metallization,
are also included.
†http://rfic.eecs.berkeley.edu/~niknejad/asitic.html
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 07 – Resistors and Inductors (3/10/14) Page 07-31
SUMMARY
• Types of resistors include diffused, well, polysilicon and metal
• Resistors are characterized by:
- Value
- Linearity
- Power
- Parasitics
• Technology effects on resistors includes:
- Process bias
- Diffusion interaction
- Thermoelectric effects
- Piezoresistive effects
• Inductors are made by horizontal metal spirals, typically in top metal
• Inductors are characterized by:
- Value
- Losses
- Self-resonant frequency
- Parasitics
• RF transformers are reasonably easy to build and work well using stacked inductors
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 08 – Latchup and ESD (4/25/16) Page 08-1
Furthermore, the resistance from the bases of the NPN transistors and collectors of the
PNP transistor greatly influences both latchup and ESD. Thus, both latchup and ESD are
influenced by layout.
LATCHUP
What is Latchup?
• Latchup is the creation of a low impedance path
between the power supply rails.
• Latchup is caused by the triggering of parasitic
bipolar structures within an integrated circuit
when applying a current or voltage stimulus on
an input, output, or I/O pin or by an over-voltage
on the power supply pin.
• Temporary versus true latchup:
A temporary or transient latchup occurs only while the pulse stimulus is connected to
the integrated circuit and returns to normal levels once the stimulus is removed.
A true latchup remains after the stimulus has been removed and requires a power
supply shut down to remove the low impedance path between the power supply rails.
Latchup Testing
The test for latchup defines how the designer must think about latchup.
• For latchup prevention, you must consider where a current limited (≥100mA), 10ms
pulse is going to go when applied to a pad when the voltage compliance of the pad is
constrained to 50% above maximum power supply and to 2V below ground. (Higher
temperatures, 85C°and 125°C, are more demanding, since VBE is lower.)
• VDD is increased by 150% (must be careful for low voltage ICs)
100m
A
1ms
100mA ms VDD
1
1.5VDD
• Latchup is sensitive to layout and is most often solved at the physical layout level.
Hold vPNPN
Cathode Cathode Avalanche Sustaining
Voltage, VH voltage, VS
Breakdown Body diode 050414-01
(CMOS)
Important concepts:
• To avoid latchup, vPNPN ≤ VS
• Once the pnpn structure has latched up, the large current required by the above i-v
characteristics must be provided externally to sustain latchup
• To remove latchup, the current must be reduced below the holding current
Latchup Triggering
Latchup of the SCR can be triggered by two different mechanisms.
1.) Allowing vPNPN to exceed the sustaining voltage, VS.
2.) Injection of current by a triggering device (gate triggered)
Anode Pad VDD
pnp Gate
Gate Current
Injector Injector
npn Gate
Gate Current
SCR SCR
Note: The gates mentioned above are SCR junction gates, not MOSFET gates.
From the above considerations, latchup requires the following components:
1.) A four-layer structure (SCR) connected between VDD and ground.
2.) An injector.
3.) A stimulus.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 08 – Latchup and ESD (4/25/16) Page 08-7
+fb ii
VDD bn bp io
loop
050414-04
Loop gain:
io
ii ≈ pn
2.) A bias condition must exist such that both bipolars are turned on long enough for
current through the “SCR” to exceed its switching current.
3.) The bias supply and associated circuits must be capable of supplying the current at
least equal to the switching current and at least equal to the holding current to maintain
the latched state.
VDD
vIN
vOUT
vOUT
vIN
VDD
Rw3
LT2
Rs1 LT1 Rw2
Rs2 Rw4 VT2 VT1 Rw1
Rs3 Rs4
p+ p p- n- n n+ Oxide Poly 1 Poly 2 Nitride Salicide Metal 050416-03
Parasitic components:
Lateral BJTs LT1 and LT2
Vertical BJTs VT1 and VT2
Bulk substrate resistances Rs1, Rs2, Rs3, and Rs4
Bulk well resistances Rw1, Rw2, Rw3, and Rw4
Voltage Compliant
vIN vOUT
Current Source VDD
LT2
Rs LT1 VT2 VT1 Rw
Loop gain:
iout Rw Rs
= P1 N1R +r
iin Rw+rP1 s N1
Rw Rs
= P1N1
P1Vt N1Vt
Rw+ Rs+
IP1 IP2
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 08 – Latchup and ESD (4/25/16) Page 08-12
Voltage Compliant
vIN vOUT Current Sink VDD
Rw3
LT2
Rs LT1 VT2 VT1 Rw
Loop gain:
iout Rw Rs
= P1 N1R +r
iin Rw +r P1 s N1
Rw Rs
= P1N1
P1Vt N1Vt
Rw+ Rs+
IP1 IP2
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 08 – Latchup and ESD (4/25/16) Page 08-13
Clk
Injectors Receiver
Driver
Transmission Gate Clock Driver
050416-09 p+ p p- n- n n+ Oxide Poly 1 Poly 2 Nitride Salicide Metal
The two bold solid bipolar transistors in the transmission gate act as injectors to the npn-
pnp parasitic bipolars of the clock driver and cause these transistors to latchup. The
injector sites are the diffusions connected to the pad.
n-well p-well
Substrate
Preventing Latch-Up
1.) Keep the source/drain of the MOS device not in the well as far away from the well as
possible. This will lower the value of the BJT betas.
2.) Reduce the values of RN- and RP-. This requires more current before latch-up can
occur.
3.) Surround the transistors with guard rings. Guard rings reduce transistor betas and
divert collector current from the base of SCR transistors.
VDD VSS
FOX
FOX FOX FOX FOX FOX FOX FOX
p-well
n- substrate
140805-01
Decreased bulk
Decreased bulk
resistance
resistance
p+ p p- n- n n+
n-
051201-01
p+ p p- n n+ 051201-02
Also, the increased doping level of the n+ (p+)guard ring in n (p) material decreases the
resistance in the area of the guard ring.
Rw
Rs
• The guard rings also help to reduce the effective well and substrate resistance.
• The guard rings reduce the lateral beta
Key: The guard rings should act like collectors
vIN
Rs Rw
050727-01
Risetime
0 t
0 070210-01
ESD Models
• Human body model (HBM): Representative of an ESD
event between a human and an electronic component.
050423-02
11 CDM
2 HBM
-1
MM
-4
0 20 40 60 80 100 120
Time, t (ns)
VSS 041008-01
• ESD protection will require area on the chip (busses
and timing components)
Local Local
Clamp Clamp ESD
Input Internal Output Power
Pad Circuits Pad Rail
Local Local Clamp
Clamp Clamp
040929-06
Local clamp based protection VSS
Local clamps – Conducts ESD current without loading the internal (core) circuits
ESD power rail clamps – Conducts a large amount of current with a small voltage drop
ESD Events:
Pad-to-rail (uses local clamps only)
Pad-to-pad (uses either local or local and ESD power rail clamps)
R
Trigger NMOS
Circuit Clamp
C Inverter
Driver
VSS 041001-03
Operation:
• Normally, the input to the driver is high, the output low and the NMOS clamp off
• For a positive ESD event, the voltage increases across R causing the inverter to turn on
the NMOS clamp providing a low impedance path between the rails
• Cannot be used for pads that go above power supply or are active when powered up
• For power supply turn-on, the circuit should not trigger (C holds the clamp off during
turn-on)
Also, forward biased diodes serve as non-breakdown clamps.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 08 – Latchup and ESD (4/25/16) Page 08-29
Current
Current
ITarget ITarget
ESD Clamp ESD Clamp
Protected Protected
Device Device
Voltage Voltage
Protected ESD
Case 1 - Okay Case 2 - Protected Device Fails
Device Clamp
Current
Current
ITarget ITarget
ESD Clamp ESD Clamp
Protected
Protected Device
Device
Voltage Voltage
Case 3 - Okay Case 4 - Protected Device Fails
070221-02
NMOS W
Increasing
Snapback W
ESD Practice
General Guidelines:
• Understand the current flow requirements for an ESD event
• Make sure the current flows where desired and is uniformly distributed
• Series resistance is used to limit the current in the protected devices
• Minimize the resistance in protecting devices
• Use distributed (smaller) active clamps to minimize the effect of bus resistance
• Understand the influence of packaging on ESD
• Use guard rings to prevent latchup
Check list:
• Check the ESD path between every pair of pads
• Check for ESD protection between the pad and internal circuitry
• Check for low bus resistance
- Current: Minimum metal for ESD 40 x Electromigration limit
- Voltage: 1.5A in a metal bus of 0.03/square of 1000µm long and 30µm wide
gives a voltage drop of 1.5V
• Check for sufficient contacts and vias in the ESD path (uniform current distribution)
SUMMARY
• Latchup is a low impedance path between VDD and ground causing excessive current.
• The conditions for latchup are:
- A four-layer, pnpn structure connected between power supply rails
- An injector (any diffusion connected to a pad)
- A stimulus
• Latchup is prevented by:
- Keeping the NMOS and PMOS transistors separated
- Reducing the well resistance with appropriate well ties
- Surrounding the transistors with guard rings
• ESD is caused by triboelectric charging which discharges through the IC
• The current produced by an ESD event must be controlled – uniform current flow,
minimum voltage drop, and must not flow through sensitive circuitry
• An ESD event turns on very quickly (<1ns), has a high peak current (1A), and lasts for
approximately 100 ns.
• ESD clamps consist of breakdown clamps (snapback) and non-breakdown clamps.
• A 16-lecture, on-line course on the Analog Design viewpoint of ESD can be found at
http://www.udemy.com/esd-an-analog-design-viewpoint
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 09 – Large Signal MOSFET Model (5/14/18) Page 09-1
INTRODUCTION TO MODELING
Models Suitable for Understanding Analog Design
The model required for analog design with CMOS technology is one that leads to
understanding and insight as distinguished from accuracy.
Technology
Understanding
and Usage
Refined and
optimized
design Fig.3.0-02
This lecture is devoted to the simple model suitable for design not using simulation.
Time Dependence
Time Independent Time Dependent
Polysilicon
p+ n+ n+
Threshold (VG=VT)
VB = 0 VS = 0 VG =VT VD = 0
Polysilicon
p+ n+ n+
Polysilicon
p+ n+ n+
Fig.3.1-02
p+ n+ n+
p+ n+ n+
p+ n+ n+
A depletion region
p- substrate
forms between the drain and channel 0 vDS
0 0.5VT VT Fig.3.1-04
p+ n+ n+
p- substrate
0 vDS
0 VT 2VT 3VT
VGS=3VT:
VB = 0 VS = 0 VG = 3VT VD = 2VT iD
VGS =3VT
iD
Polysilicon
p+ n+ n+
Further increase in
p- substrate
VG will cause the FET to become active 0 vDS
0 VT 2VT 3VT
Fig.3.1-05
1500
VGS = 2.5
iD(mA) 1000
VGS = 2.0
500
VGS = 1.5
VGS = 1.0
0
0 1 2 3 4 5
vDS (Volts)
SPICE Input File: Fig. 3.1-6
Output Characteristics for NMOS M5 6 5 0 0 MOS1 w=5u l=1.0u
M1 6 1 0 0 MOS1 w=5u l=1.0u VGS5 5 0 3.0
VGS1 1 0 1.0 VDS 6 0 5
M2 6 2 0 0 MOS1 w=5u l=1.0u .model mos1 nmos (vto=0.7 kp=110u
VGS2 2 0 1.5 +gamma=0.4 lambda=.04 phi=.7)
M3 6 3 0 0 MOS1 w=5u l=1.0u .dc vds 0 5 .2
VGS3 3 0 2.0 .print dc ID(M1), ID(M2), ID(M3), ID(M4),
M4 6 4 0 0 MOS1 w=5u l=1.0u ID(M5)
VGS4 4 0 2.5 .end
0
0 1 2 3 4 5
vGS (Volts) Fig. 3.1-7
SPICE Input File:
Transconductance Characteristics for NMOS M5 5 6 0 0 MOS1 w=5u l=1.0u
M1 1 6 0 0 MOS1 w=5u l=1.0u VDS5 5 0 5.0
VDS1 1 0 1.0 VGS 6 0 5
M2 2 6 0 0 MOS1 w=5u l=1.0u .model mos1 nmos (vto=0.7 kp=110u
VDS2 2 0 2.0 +gamma=0.4 lambda=.04 phi=.7)
M3 3 6 0 0 MOS1 w=5u l=1.0u .dc vgs 0 5 .2
VDS3 3 0 3.0 .print dc ID(M1), ID(M2), ID(M3), ID(M4),
M4 4 6 0 0 MOS1 w=5u l=1.0u ID(M5)
VDS4 4 0 4.0 .probe
.end
Increasing
values of vGS
vDS
Fig. 110-04
The saturation voltage for MOSFETs is the value of drain-source voltage at the peak of
the inverted parabolas. vDS
diD µoCoxW
dvDS = [(vGS-VT) - vDS] = 0
L Cutoff Saturation Active
T
-V
vDS(sat) = vGS - VT S
vG
=
S
Useful definitions: vD
0 vGS
µoCoxW K’W 0 VT Fig. 3.2-4
= L =
L
Note that newest editions of Analysis and Design of Analog ICs, P.R. Gray et.al,
switches the definition for the active and saturation regions.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 09 – Large Signal MOSFET Model (5/14/18) Page 09-12
K’ = 29.6µA/V 2,
5µA
k = 0,
VDS(sat) = 1.0V
0µA
0 0.2 0.4 0.6 0.8 1.0
vDS (volts) 140825-01
This discrepancy is due to the fact that we assumed that the threshold, VT, was constant
over the channel.
If we let VT (y) = VT + kv(y) then the Sah model is exactly the same as the SPICE model.
To find vDS(sat), set the diD/dvDS equal to zero and solve for vDS = vDS(sat),
vGS - VT
vDS(sat) =
1+k
Therefore, in the saturation region, the drain current is
WµoCox
iD = 2(1+k)L (vGS - VT)2
For k = 0.5 and K’ = 44.8µA/V2, excellent correlation is achieved with SPICE 2 as seen
on the previous slide.
p+ n+ n+
Illustration:
Leff
0.4
PMOS
0.3
0.2
NMOS
0.1
0
0 0.5 1 1.5 2 2.5
Channel Length (microns) Fig.130-6
Most analog designers stay away from minimum channel length to get better gains and
matching at the sacrifice of speed.
Influence of the Bulk Voltage on the Large Signal MOSFET Model Polysilic
p+ n+
+ voltage to offset the channel-bulk VSB1 > 0:
VD > 0
depletion charge (-Qb/Cox) p- substrate
Influence of the Bulk Voltage on the Large Signal MOSFET Model - Continued
Bulk-Source (vBS) influence on the transconductance characteristics-
iD
Decreasing values
of bulk-source voltage
VBS = 0
ID
vDS > vGS-VT
VGS
vGS
VT0 VT1 VT2 VT3
060612-02
In general, the simple model incorporates the bulk effect into VT by the previously
developed relationship:
Silicon Constants
MOSFET Parameters
Model Parameters for a Typical CMOS Bulk Process (0.25µm CMOS n-well):
SUBTHRESHOLD MODEL
Large-Signal Model for Weak Inversion
The electrons in the substrate at the source side can be expressed as,
s
np(0) = npoexpV
t
The electrons in the substrate at the drain side can be expressed as,
s-vDS
np(L) = npoexp V
t
Therefore, the drain current due to diffusion is,
np(L)- np(0) W s vDS
iD = qADn = qXDnnpoexp 1 - exp-
L L Vt Vt
where X is the thickness of the region in which iD flows.
In weak inversion, the changes in the surface potential, s are controlled by changes in
the gate-source voltage, vGS, through a voltage divider consisting of Cox and Cjs, the
depletion region capacitance. Poly
ds Cox 1 vGS vGS-VT Oxide Cox vGS
dv = C + C = n → s = n + k1 = n + k2 Channel
GS ox js Dep. Cjs fs
where VT
k2 = k1 + n Substrate
060405-04
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 09 – Large Signal MOSFET Model (5/14/18) Page 09-24
5x103
An expression for the electron drift 105 106 107
velocity as a function of the electric Electric Field (V/m) Fig130-1
field is,
µnE
vd 1 + E/E
c
where
vd = electron drift velocity (m/s)
µn = low-field mobility ( 0.07m2/V·s)
Ec = critical electrical field at which velocity saturation occurs
Saturation Voltage
Differentiating iD with respect to vDS and setting equal to zero gives,
1
(VGS-VT)
V’DS(sat) = 1 + 2(VGS-VT -1 (VGS-VT)1 -
+ ···
2
if
(VGS-VT)
<1
2
Therefore,
(VGS-VT)
V’DS(sat) VDS(sat) 1 - + ···
2
Note that the transistor will enter the saturation region for vDS < vGS - VT in the
presence of velocity saturation.
K’ W
= [ v - V ]2
2[1 + (vGS-VT)] L GS T
However, we continue to use the following to define when the MOSFET is in the
saturation region,
(VGS-VT)
vDS ≥ (VGS-VT) 1 - + ···
2
Note as the velocity saturation effect becomes stronger, that the drain current-gate
voltage relationship becomes linear.
SUMMARY
• The modeling of this lecture is devoted to understanding how the circuit works
• The two primary current-voltage characteristics of the MOSFET are the
transconductance characteristic and the output characteristic
• The simple Sah large signal model is good enough for most applications and
technology
• The Sah model can be improved in the region of the knee and for the weak dependence
of drain current on drain-source voltage in the saturation region
• Most designers do not work at minimum channel length because of the channel length
modulation effect and because worse matching occurs for small areas
• The threshold voltage is increased as the bulk-source is reverse biased
• The subthreshold model accounts for very small currents that flow in the channel when
the gate-source voltage is smaller than the threshold voltage
• The subthreshold current is exponentially related to the gate-source voltage
• Velocity saturation occurs at minimum channel length and can be modeled by
including a source degeneration resistor with the simple large signal model
SiO2
Gate
Source Drain
C1 C2 C3
FOX FOX
C4
CBS CBD
Bulk
Fig120-06
CJ·AS CJSW·PS D C
CBS = MJ + MJSW,
vBS vBS Source Drain
1 - 1 -
PB PB F
E
and A B
2.) vBS> FC·PB SiO2
Bulk
CJ·AS VBS Fig. 120-07
CBS = Drain bottom = ABCD
1+MJ 1 - (1+MJ)FC + MJ PB
1- FC Drain sidewall = ABFE + BCGF + DCGH + ADHE
VBS
CJSW·PS
+ 1+MJSW 1 - (1+MJSW)FC + MJSW PB
1 - FC
where
AS = area of the source
PS = perimeter of the source
CJSW = zero bias, bulk source sidewall capacitance
MJSW = bulk-source sidewall grading coefficient
For the bulk-drain depletion capacitance replace "S" by "D" in the above.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 10 – MOS Capacitor Model and Large Signal Model Dependence (3/10/14) Page 10-5
Gate
Drain-gate overlap
Channel capacitances:
Source-gate overlap
capacitance CGS (C1) capacitance CGD (C3) C2 = gate-to-channel = CoxWeff·(L-2LD)
Gate
FOX FOX = CoxWeff·Leff
Source Drain
Gate-Channel Channel-Bulk
C4 = voltage dependent channel-
Bulk bulk/substrate capacitance
Capacitance (C2) Capacitance (C4)
Fig. 120-09
Gate
FOX C5 Source/Drain C5 FOX
Bulk
Fig120-10
C5 = CGBO
Capacitance values based on an oxide thickness of 140 Å or Cox=24.7 10-4 F/m2:
Type P-Channel N-Channel Units
CGSO 220 10-12 220 10-12 F/m
CGDO 220 10-12 220 10-12 F/m
CGBO 700 10-12 700 10-12 F/m
CJ 560 10-6 770 10-6 F/m2
CJSW 350 10-12 380 10-12 F/m
MJ 0.5 0.5
MJSW 0.35 0.38
C1 and C3 are overlap capacitors due to lateral diffusion of the source and drain
C2 is the gate to channel capacitance
C4 is the depletion capacitance between the channel and the bulk
C5 is the fringing capacitance between the gate and the bulk around the edges of the
channel
As the gate-source voltage varies from 0 to VT, the channel-bulk capacitor varies from a
very large capacitor (because of a very small depletion region) to a capacitor much
smaller than C2.
Capacitors in Cutoff:
CGS C1 = Cox·LD·W = CGSO·W
CGD C3 = Cox·LD·W = CGDO·W
CGB C2 varies from Cox·L·W to 2C5
CBD CBD = (CJ·AD)/[1 – (vBD/PB)]MJ + (CJSW·PD)/[1 – (vBD/PB)]MJSW
CBS CBS = (CJ·AS)/[1 – (vBS/PB)]MJ + (CJSW·PS)/[1 – (vBS/PB)]MJSW
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 10 – MOS Capacitor Model and Large Signal Model Dependence (3/10/14) Page 10-9
070330-06
In the saturation region, C4, becomes small and is not shown above.
Capacitors in Saturation:
CGS C1 = Cox·LD·W + (2/3)Cox·L·W = [CGSO + (2/3)Cox·L]W
CGD C3 = Cox·LD·W = CGDO·W
CGB 2C5 = 2·CGBO·W
CBD CBD = (CJ·AD)/[1 – (vBD/PB)]MJ + (CJSW·PD)/[1 – (vBD/PB)]MJSW
CBS CBS = (CJ·AS)/[1 – (vBS/PB)]MJ + (CJSW·PS)/[1 – (vBS/PB)]MJSW
070330-07
In the saturation region, C4, becomes small and is not shown above.
Capacitors in Active:
CGS C1 = Cox·LD·W + (1/2)Cox·L·W = [CGSO + (1/2)Cox·L]W
CGD C3 = Cox·LD·W + (1/2)Cox·L·W = [CGDO + (1/2)Cox·L]W
CGB 2C5 = 2·CGBO·W
CBD CBD = (CJ·AD)/[1 – (vBD/PB)]MJ + (CJSW·PD)/[1 – (vBD/PB)]MJSW
CBS CBS = (CJ·AS)/[1 – (vBS/PB)]MJ + (CJSW·PS)/[1 – (vBS/PB)]MJSW
tox(min) tox(max)
060225-01
3.) Process biases – differences between the drawn and actual dimensions due to process
(etching, lateral diffusion, etc.)
Drawn Dimension
Actual Dimension
060225-03
Note: Because the large-signal model for the MOSFET includes all the influences of
voltage on the transistor, we will focus on passive components except for breakdown.
Depletion region
STI STI
n- well
n-well
p- substrate
As the voltage at the terminals of the resistor become smaller than the n-well potential,
the depletion region will widen causing the thickness of the resistor to decrease.
L
R= VR
tW
where VR is the reverse bias voltage from the resistor to the well.
This effect is worse for well resistors because the doping concentration of the resistor is
smaller.
Voltage coefficient for diffused resistors ≈ 200-800 ppm/V
Voltage coefficient for well resistors ≈ 8000 ppm/V
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 10 – MOS Capacitor Model and Large Signal Model Dependence (3/10/14) Page 10-18
0.1A
Conductivity
modulation
v
060311-01
As the reverse bias voltage across a pn junction becomes large, at some point, called the
breakdown voltage, the current will rapidly increase. Both transistors, diodes and
depletion capacitors experience this breakdown.
iR
Model for current multiplication factor:
1
iR = M·IR where M =
vR n
1 - BV
Breakdown
voltage
vR
060311-02 BV
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 10 – MOS Capacitor Model and Large Signal Model Dependence (3/10/14) Page 10-20
-4T
VGS – VT0 - (T-To) = 3 VGS(ZTC) = VT0 - To - 3
0.9 25°C
0.7 NMOS
25°C 0.8 50°C
L=50nm
0.6 50°C 0.7 NMOS 100°C
100°C L=500nm 140°C
0.5 140°C 0.6
0.4 0.5
p+ n+ n+
p-well
n- substrate
Fig.3.6-5
VGS<VT: VG <VT VD > VDS(sat)
B S
Depletion
Polysilicon Region
p+ n+ n+
p-well
n- substrate
Fig.3.6-6
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 10 – MOS Capacitor Model and Large Signal Model Dependence (3/10/14) Page 10-25
−iD Is = qA + = KT exp
3
Lp Ln L N Vt
Differentiating with respect to temperature gives,
dIs 3KT 3 −VGo qKT 3VGo −VGo 3Is Is VGo
= exp + exp = +
dT T Vt KT 2 Vt T T Vt
dIs 3 1 VGo
TCF = = +
IsdT T T Vt
Example
Assume that the temperature is 300° (room temperature) and calculate the reverse
diode current change and the TCF for a 5° increase.
Solution
The TCF can be calculated from the above expression as TCF = 0.01 + 0.155 = 0.165.
Since the TCF is change per degree, the reverse current will increase by a factor of 1.165
for every degree (or C) change in temperature. Multiplying by 1.165 five times gives
an increase of approximately 2. Thus, the reverse saturation current approximately
doubles for every 5C temperature increase. (Experimental is closer to 8°C.)
10-5
200°C Data
-6
10 Symbol Min. L
6 mm
250°C 5 mm
10-7
Leakage Current (A)
4 mm
10-8 IR 2 mm
50mm
Lmin 100°C
-9
10 1V
Theory
10-10 matched
at 150°C Generation-
Diffusion Recombination
10-11
Leakage Leakage
Dominant Dominant
10-12
1.8 2 2.2 2.4 2.6 2.8
1000/T (°K -1) 100324-03
Theory:
VG(T)
Is(T) T exp kT 3
MOSFET RELIABILITY
Hot Carrier Injection
Hot carriers depend on channel length. Longer channel lengths minimize hot carrier
effects. The worst-case hot carrier degradation occurs in NMOS devices when the gate
voltage is between VT and 0.5VDS and VDS is large.
Substrate current- Gate current-
Target specifications:
1.) No more than 100mV change in VT within a year of stress.
2.) No more than a 10% change in IDsat or Rdson within a year of stress.
IDSat Degradation in %
VT Degradation in mV
10.0 10.0
Extra-
1.0 Extra- 1.0 polated
polated Data
Data
10 years
10 years
0.1 0.1
0.01 0.01
101 102 103 104 105 106 107 108 109 101 102 103 104 105 106 107 108 109
Time (Seconds) Time (Seconds) 140826-01
Also:
• NBTI (negative bias temperature instability) – PMOS with large gate-source voltage
• GOI (gate oxide integrity) – reliability of dielectrics
Antenna Effect
• The antenna effect is the situation during processing the charged plasmas that are used
put charge on the metal and if this metal is connected to the gate of a MOSFET and the
metal area is large enough, the oxide can breakdown.
• Oxide thicknesses less than 100Å are more susceptible to the antenna effect (plasma
induced discharge).
• A reverse biased diode connected from the metal to the semiconductor can be used to
leak this charge during processing (at high temperatures) and protect the gate oxides.
• Design rules exists to avoid having metal with large areas connected to gates (metal
jumpers).
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 10 – MOS Capacitor Model and Large Signal Model Dependence (3/10/14) Page 10-32
SUMMARY
• The large signal capacitance model includes depletion and parallel plate capacitors
• The depletion capacitors CBD and CBS vary with their reverse bias voltage
• The capacitors CGD, CGS, and CGB have different values for the regions of cutoff,
active and saturated
• The large signal model varies with process primarily through µo and tox
• Voltage dependence of resistors and capacitors is primarily due to the influence of
depletion regions
• The temperature dependent large signal model of the MOSFET yields a gate-source
voltage where the derivative of drain current with respect to temperature is zero
• Other MOSFET temperature dependence comes from the leakage currents across
reverse biased pn junctions
• MOSFET reliability concerns degradation in performance over a specified lifetime
id = gmvgs
ID Q
vgs
vGS
VT VGS 060311-03
DiD DiDʼ Q
vgs
DVGS 060311-04
Be alert for situations where the small signal model will be in error (i.e. slide 25-27).
diD | ID
gds dv = ID
DS Q 1 + V DS
and
dD | diD dvGS | diD dVT | gm
gmbs dv Q = dv dv = - = = gm
BS GS BS Q dV T BSQ 2 2|F| - VBS
dv
NOISE MODELS
Derivation of the MOSFET Thermal Noise Model
In the active region, the channel resistance of the MOSFET is given from the simple large
signal model as,
1 1 1 1
Rchannel = = ≈ =
∂iD | K’W K’W gm(sat)
(V GS - V T - V DS ) (V GS - V T)
∂vDSQ L L
The current thermal noise spectral density of a MOSFET in the active region would be
4kT
in2(active) = R = 4kTgm(sat) (A2/Hz)
channel
In the saturation region, approximate the channel resistance as 2/3 the value in the active
region resulting in 2/3 the noise. Therefore in saturation we have the current thermal
noise spectral density as,
2 2 8kTgm(sat)
2
in (sat) = 3 in (active) = (A 2/Hz)
3
Translating this drain current noise to the gate voltage noise by dividing by gm2 gives
8kT
en2 = 3g (V2/Hz)
m
where
KF = Flicker noise coefficient
S = Slope factor of the 1/f noise
Although we do not have a good explanation for the reason why, the value of the 1/f
noise for a PMOS is typically less than that for an NMOS for the same current and W/L.
f = 10Hz
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 11 – Linear Circuit Models (6/10/14) Page 11-13
Noise Noise
S B Free
Free S
MOSFET MOSFET S
where
8kTgm(1+) KF IDAF
in2 = + S (amperes2/Hz)
3 f CoxL
2
gmbs
= g
m
k = Boltzmann’s constant
KF = Flicker noise coefficient
S = Slope factor of the 1/f noise
AF = Current coefficient
1/f noise
Thermal noise
fCorner log10 f
060311-06
Cp Cp1 Cp2
i
1.) Large signal i= v
R
Conductivity
modulation
v
2.) Small signal 060311-01
v = Ri
3.) Noise
en2 = 4kTR or in2 = 4kTG
Capacitor Models
Rp
One of the parasitic capacitors is i C(v)
the top plate and the other is -
+
associated with the bottom plate. v
Cp Cp
C 060315-03
Linear
1.) Large signal
Nonlinear
v
060315-04
2.) Small signal
q = Cv i = C(dv/dt)
060315-05
INTERCONNECTS
Types of “Wires”
1.) Metal
Many layers are available in today’s technologies:
- Lower level metals have more resistance (70 m/sq.)
- Upper level metal has the less resistance because it is thicker (50 m/sq.)
2.) Polysilicon
Better resistor than conductor (unpolysicided) (135/sq.)
Silicided polysilicon has a lower resistance (5/sq.)
3.) Diffusion
Reasonable for connections if silicided (5/sq.)
Unsilicided (55/sq.)
4.) Vias
Vias are vertical metal (tungsten plugs or aluminum)
- Connect metal layer to metal layer (3.5/via)
- Connect metal to silicon or polysilicon contact resistance (5/contact)
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 11 – Linear Circuit Models (6/10/14) Page 11-19
050319-02
Capacitance of Wires
Self, fringing and coupling capacitances:
Wide Spacing Minimum Spacing
CCoupling CCoupling
CFringe CFringe
Ground plane CSelf
050319-03
Electromigration
Electromigration occurs if the current density is too large and the pressure of carrier
collisions on the metal atoms causes a slow displacement of the metal.
Black’s law:
1 (Ea/kTj)
MTF = e
AJ 2 Metal
Where 050304-04
n-well p-well
DC Ground
Substrate
IA IA+IB IA+IB+IC
Better:
Circuit Circuit Circuit
A B C IC
R
2R IB
3R IA
Best:
Circuit Circuit Circuit
A B C
R IA R IB R IC
050305-04
Kelvin Connections
Avoid unnecessary ohmic drops.
A B A B
X Y
Ohmic Connection Kelvin Connection 041223-12
• Leakage
Also: The substrate BJT
and the inductor create
currents in the substrate.
• Minority Carrier
BJTs:
Injection primarily across the depletion
capacitance.
Passives:
Isolation Techniques
Isolation techniques include both layout and circuit approaches to isolating quiet from
noisy circuits.
†
Electrical Performance of Packages, National Semiconductor Application Note 1205, August 2001.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 11 – Linear Circuit Models (6/10/14) Page 11-32
SUMMARY
• Small signal models are a linear representation of the transistor electrical behavior
• Including the transistor capacitors in the small signal model gives frequency
dependence
• Noise models include thermal and 1/f noise voltage or current spectral density models
• Passive component models include the nonlinearity, small signal and noise models
• Interconnects include metal, polysilicon, diffusion and vias
• Electromigration occurs if the current density is too large causing a displacement of
metal
• Substrate interference is due to interaction between various parts of an integrated circuit
via the substrate
• Method to reduce substrate interference include:
- Physical separation
- Guard rings
- Reduced inductance in the power supply and ground leads
- Appropriate contacts to the regions of constant potential
- Reduce the source of interfering noise
- Use differential signal processing techniques
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 12 – Component Matching (6/10/14) Page 12-1
INTRODUCTION
What is Accuracy and Matching?
The accuracy of a quantity specifies the difference between the actual value of the
quantity and the ideal or true value of the quantity.
The mismatch between two quantities is the difference between the actual ratio of the
quantities and the desired ratio of the two quantities.
Example:
x1 = actual value of one quantity
x2 = actual value of a second quantity
X1 = desired value of the first quantity
X2 = desired value of the second quantity
The accuracy of a quantity can be expressed as,
x - X X
Accuracy = X = X
x 2 X2
The mismatch, , can be expressed as, x1 - X1 X1x2
= X = X x -1
2 2 1
X1
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 12 – Component Matching (6/10/14) Page 12-3
9
8
7
6
5
4
3
2
1
0 X
0 1 2 3 4 5 6 7 8 9 1011121314
041005-01
253
m = 40 = 6.325 s = 2.115
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 12 – Component Matching (6/10/14) Page 12-5
Types of Mismatches
1.) Those controlled or influenced by electrical design
- Transistor operation
- Circuit techniques
- Correction/calibration techniques
2.) Those controlled or influenced by physical design
- Random statistical fluctuations (microscopic fluctuations and irregularities)
- Process bias (geometric variations)
- Pattern shift (misalignment)
- Diffusion interactions
- Stress gradients and package shifts
- Temperature gradients and thermoelectrics
- Electrostatic interactions
ELECTRICAL MATCHING
Matching Principle
Assume that two transistors are matched (large signal model parameters are equal).
Then if all terminal voltages of one transistor are equal to the terminal voltages of the
other transistor, then the terminal currents will be matched.
iC1 iC2 iD1 iD2
iB1 iB2
Q1 Q2 M1 M2
iE1 iE2
041005-02
Note that the terminals may be physically connected together or at the same potential but
not physically connected together.
M3 M4 iD1 iD2
M3 M4
+ M1 VB M2 + M1 M2
Vio Vio
VB M1 M2 - -
M5 M5
041005-03
Cascode current mirror:
The key transistors are M1 and M2. The gates and sources are physically connected
and the drains are equal due to M3 and M4 gate-source drops. As a result, iD1 will be
very close to iD2.
Differential amplifier:
When iD1 and iD2 are equal, the fact that the drains of M1 and M2 are equal should
give the smallest value of the input offset voltage, Vio.
Note: Since the drain voltages of M3 and M4 in both circuits are not necessarily equal,
the gate-source voltages of M3 and M4 are not exactly equal which cause the drain
voltages of M1 and M2 to not be exactly equal.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 12 – Component Matching (6/10/14) Page 12-10
Gate-Source Matching
Not as precise as the previous principle but useful for biasing applications.
A. If the gate-source voltages of two or more FETs are equal and the FETs are matched
and operating in the saturation region, then the currents are
related by the W/L ratios of the individual FETs. The gate- iD1 M1 M2
iD2
source voltages may be directly or indirectly connected. W1 W2
L1 + + L2
K’W1 2K’iD1
iD1 = 2L (vGS1-VT1)2 → (vGS1-VT1)2 = (W /L ) vGS1 vGS2
1 1 1 - -
Fig. 290-02
K’W2 2K’iD2
iD2 = 2L (vGS2-VT2)2 → (vGS2-VT2)2 = (W /L )
2 2 2 iD1
W2 W1 W1/L1 W1
If vGS1 = vGS2, then iD1 = iD2 or iD1 = W /L iD2 + L1
L2 L1 2 2 vGS1
-
B. If the drain currents of two or more transistors are equal and the trans- iD2
M2
W2
istors are matched and operating in the saturation region, then the gate- + L2
source voltages are related by the W/L ratios (ignoring bulk effects). vGS2
-
If iD1=iD2, then Fig. 290-03
W2/L2 W2 W1
vGS1 = VT1+ W1/L1 (vGS2-VT2) or vGS1 = vGS2 if L2 = L1
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 12 – Component Matching (6/10/14) Page 12-11
f
0 fc 2fc 3fc
VB(f)
f
0 fc 2fc 3fc
VC(f)
f
0 fc 2fc 3fc Fig. 7.5-8
Cc vout
Inn clk clkb clk Inp
clk
Inp clk clk Inn
M1 M2
VDD VDD
M5
R1
clkb clkb 140828-01-01
Self-Calibration Techniques
The objective of self-calibration is to increase the matching between two or more
components (generally passive).
The requirements for self-calibration:
1.) A time interval in which to perform the calibration
2.) A means of adjusting the value of one or more of the components.
041007-05
Self-calibration can typically improve the matching by a factor of 2-3 bits (4-8).
Assume the amplifier has a DC input offset voltage of Vio. The following shows how to
calibrate one (or both) of the capacitors.
Variable Components
The correction circuitry should be controlled by logic circuits so that the correction can
be placed into memory to maintain the calibration of the circuit during application.
Implementation for C1 and C2 of the previous example:
C1 C1 C1 C1 C2 C2 C2 C2
C1 1- 1 2K 2K+1 2K+2 2N C2 1- 1 2K 2K+1 2K+2 2N
2K 2K
S1 S2 S3 SN S1 S2 S3 SN
Capacitor C 1 Capacitor C 2
041007-08
Additional circuitry:
Every self-calibration system will need additional logic circuits to sense when the value
of vx changes from positive to negative (or vice versa) and to store the switch settings in
memory to maintain the calibration.
i me t1
S0 S1 S2 S3 S4 S5 S6 S7 S8 S9 T
R0 R1 R2 R3 R4 R5 R6 R7 R8 R9
VRef
R0 R1 R2 R3 R4 R5 R6 R7 R8 R9
Tim S0 S1 S2 S3 S4 S5 S6 S7 S8 S9
e t2
All resistor are approximately equal
valued to within some tolerance
R0 R1 R2 R3 R4 R5 R6 R7 R8 R9
041010-01
†L. R. Carley, “A Noise-Shaping Coder Topology for 15+ Bit Converters, IEEE J. of Solid-State Circuits, vol. 24, no. 2, April 1989, pp. 267-273.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 12 – Component Matching (6/10/14) Page 12-18
+2
Dynamic Element
PHYSICAL MATCHING
Review of Physical Matching
We have examined these topics in previous lectures. To summarize, the sources of
physical mismatch are:
- Random statistical fluctuations (microscopic fluctuations and irregularities)
- Process bias (geometric variations)
- Pattern shift (misalignment)
- Diffusion interactions
- Stress gradients and package shifts
- Temperature gradients and thermoelectrics
- Electrostatic interactions
†Alan Hastings, Art of Analog Layout, 2nd ed, 2006, Pearson Prentice Hall, New Jersey
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 12 – Component Matching (6/10/14) Page 12-22
†Alan Hastings, Art of Analog Layout, 2nd ed, 2006, Pearson Prentice Hall, New Jersey
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 12 – Component Matching (6/10/14) Page 12-24
Mismatched Transistors
Assume two transistors have vDS1 = vDS2, K1’ K2’ and VT1 VT2. Therefore we have
iO K2’(vGS - VT2)2
iI = K1’(vGS - VT1)2
How do you analyze the mismatch? Use plus and minus worst case approach. Define
K’ = K’2-K’1 and K’ = 0.5(K2’+K1’) K1’= K’-0.5K’ and K2’= K’+0.5K’
VT = VT2-VT1 and VT = 0.5(VT1+VT2) VT1 =VT -0.5VT and VT2=VT+0.5VT
Substituting these terms into the above equation gives,
K’ V T 2
1 +
iO (K’+0.5K’)(vGS - VT - 0.5VT )2 2K’ 1 - 2(vGS-VT)
iI = (K’-0.5K’)(vGS - VT + 0.5VT)2 = K’ V T 2
1 -
2K’ 1 + 2(vGS-VT)
Assuming that the terms added to or subtracted from “1” are smaller than unity gives
iO K’ K’ VT 2 VT 2 K’ 2VT
iI ≈ 1 + 2K’ 1 + 2K’ 1 - 2(vGS-VT) 1 - 2(vGS-VT) 1 + K’ - (vGS-VT)
If K’/K’ = ±5% and VT/(vGS-VT) = ±10%, then iO/iI 1 ± 0.05 ±(-0.20) = 1±(0.25)
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 12 – Component Matching (6/10/14) Page 12-25
Pelgrom’s Law
Spatial Averaging: Local and random Threshold mismatch for 0.18µm NMOS
where,
P = mismatch in a parameter, P
WL = width times the length of the device (effective Pelgrom area)
Ap = proportionality constant between the standard deviation of P and the area of
the device
Dx = distance between the matched devices
Sp = proportionality constant between the standard deviation of P and Dx
As Dx becomes large, the standard deviation tends to infinity which is not realistic.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 12 – Component Matching (6/10/14) Page 12-26
Geometric Effects
How does the size and shape of the transistor effect its matching?
Gate Area:
CVth CKp CW/W
Vth = Kp = K’ W/W =
WeffLeff WeffLeff WeffLeff
where CVth, CKp and CW/W are constants determined by measurement.
Values from a 0.35µm CMOS technology:
10.6mV·µm 8.25mV·µm
Vth,NMOS = Vth,PMOS =
WeffLeff WeffLeff
and
W 0.0056·µm W 0.0011·µm
W NMOS = W PMOS =
WeffLeff WeffLeff
The above results suggest that PMOS devices would be better matched than NMOS
devices in this technology.
†Alan Hastings, Art of Analog Layout, 2nd ed, 2006, Pearson Prentice Hall, New Jersey
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 12 – Component Matching (6/10/14) Page 12-28
SUMMARY
• IC technology offers poor absolute values but good relative values or matching
• In analog circuits, gains are determined by ratios (good matching) and time constants
are determined by products (poor matching)
• Electrical matching is determined in the electrical design phase
- Matching due to equal terminal voltages
- Matching due to process independent biasing
- Doubly correlated sampling
- Self-calibration techniques
- Dynamic element matching
• Physical matching is determined in the physical design phase
- Random statistical fluctuations (microscopic fluctuations and irregularities)
- Process bias (geometric variations)
- Pattern shift (misalignment)
- Diffusion interactions
- Stress gradients and package shifts
- Temperature gradients and thermoelectrics
- Electrostatic interactions
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture13 – Computer Models and Extraction of Simple Large Signal Model (6/14/14) Page 13-1
COMPUTER MODELS
FET Model Generations
• First Generation – Physically based analytical model including all geometry
dependence.
• Second Generation – Model equations became subject to mathematical conditioning for
circuit simulation. Use of empirical relationships and parameter extraction.
• Third Generation – A return to simpler model structure with reduced number of
parameters which are physically based rather than empirical. Uses better methods of
mathematical conditioning for simulation including more specialized smoothing
functions.
Performance Comparison of Models (from Cheng and Hu, MOSFET Modeling & BSIM3
Users Guide)
Model Minimum Minimum Model iD Accuracy in iD Accuracy in Small signal Scalability
L (µm) Tox (nm) Continuity Strong Inversion Subthreshold parameter
MOS1 5 50 Poor Poor Not Modeled Poor Poor
MOS2 2 25 Poor Poor Poor Poor Fair
MOS3 1 20 Poor Fair Poor Poor Poor
BSIM1 0.8 15 Fair Good Fair Poor Fair
BSIM2 0.35 7.5 Fair Good Good Fair Fair
BSIM3v2 0.25 5 Fair Good Good Good Good
BSIM3v3 0.15 4 Good Good Good Good Good
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture13 – Computer Models and Extraction of Simple Large Signal Model (6/14/14) Page 13-3
BSIM2 Model
Generic composite expression for the model parameters:
LX WX
X = Xo + +
Leff Weff
where
Xo = parameter for a given W and L
LX (WX) = first-order dependence of X on L (W)
Modeling features of BSIM2:
Mobility
• Mobility reduction by the vertical and the lateral field
Drain Current
• Velocity saturation
• Linear region drain current
• Saturation region drain current
• Subthreshold current
µoCoxWeff kT evGS-Vt-Voff
iDS = · ·1 - eqVDS/kT
Leff q n
where
NB
Voff = VOF + VOFB ·vBS + VOFD ·vDS and n = NO + + ND ·vDS
PHI - vBS
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture13 – Computer Models and Extraction of Simple Large Signal Model (6/14/14) Page 13-7
BSIM3 Model
The background for the BSIM3 model and the equations are given in detail in the text
MOSFET Modeling & BSIM3 User’s Guide, by Y. Cheng and C. Hu, Kluwer Academic
Publishers, 1999.
The short channel effects included in the BSIM3 model are:
• Normal and reverse short-channel and narrow-width effects on the threshold.
• Channel length modulation (CLM).
• Drain induced barrier lowering (DIBL).
• Velocity saturation.
• Mobility degradation due to the vertical electric field.
• Impact ionization.
• Band-to-band tunneling.
• Velocity overshoot.
• Self-heating.
1.) Channel quantization.
2.) Polysilicon depletion.
INTRODUCTION: This report contains the lot average results obtained by MOSIS from measurements of
MOSIS test structures on each wafer of this fabrication lot. SPICE parameters obtained from similar
measurements on a selected wafer are also attached.
COMMENTS: TSMC 0251P5M.
TRANSISTOR PARAMETERS W/L N-CHANNEL P-CHANNEL UNITS
MINIMUM 0.36/0.24
Vth 0.54 -0.50 volts
SHORT 20.0/0.24
Idss 557 -256 uA/um
Vth 0.56 -0.56 volts
Vpt 7.6 -7.2 volts
WIDE 20.0/0.24
Ids0 6.6 -1.5 pA/um
LARGE 50.0/50.0
Vth 0.47 -0.60 volts
Vjbkd 5.8 -7.0 volts
Ijlk -25.0 -1.1 pA
Gamma 0.44 0.61 V0.5
K’ (Uo*Cox/2) 112.0 -23.0 uA/V2
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture13 – Computer Models and Extraction of Simple Large Signal Model (6/14/14) Page 13-11
vDS >VDSAT
Weak inversion
region
K’Weff 1/2
m = 2L
eff
0 vGS
0 VT0 (iD =0) 140909-08
K’Weff 1/2
b=- V T0 = -mVT0
2Leff
Comments:
• Stay away from the extreme regions of mobility degradation and weak inversion
• Use channel lengths greater than Lmin
5.) Iterative procedures can be used to achieve the desired accuracy of and 2F.
Generally, an approximate value for 2F gives adequate results.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture13 – Computer Models and Extraction of Simple Large Signal Model (6/14/14) Page 13-21
vGS
VT0 VT1 VT2 VT3
FigAppB-02
By plotting VT versus x of Eq. (13) one can measure the slope of the best fit line from
which the parameter can be extracted. In order to do this, VT must be determined at
various values of vSB using the technique previously described.
† Anurag Kaplish, “Parameter Optimization of Deep Submicron MOSFETS Using a Genetic Algorithm,” May 4, 2000, Special Project Report,
School of ECE, Georgia Tech.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture13 – Computer Models and Extraction of Simple Large Signal Model (6/14/14) Page 13-28
SUMMARY
• Models have greatly improved over time resulting in efficient computer simulation
• Output conductance model is greatly improved
• Narrow channel transistors have difficulty with modeling
• Can have discontinuities at bin boundaries
• The BSIM model is a complex model, widely used and difficult to understand in detail
• The simple large signal model can be extracted from any computer model
• Extract the model at the desired channel length for the design
• Short channel technology can be modeled by finding the by any optimization routine
Switch Model
• An ideal switch is a short-circuit when ON IAB
and an open-circuit when OFF. VC = A B RAB = 0W
(VC= high)
controlling terminal for the switch (VC high +
VAB
switch ON, VC low switch OFF) VC
RAB = ¥W
-
(VC= low)
060526-03
• Actual switch:
IOFF
ron = resistance of the switch when ON
rOFF
roff = resistance of the switch when OFF VOS
rON
CAB
Ioff = offset current when the switch is OFF
CAC CBC
IA and IB are leakage currents to ground IA C IB
060526-05 C (G)
On Characteristics of a MOS Switch
Assume operation in active region (vDS < vGS - VT) and vDS small.
µCoxW vDS µCoxW
iD = L (vGS - VT) - 2 vDS L (vGS - VT)vDS
vDS 1
Thus, RON ≈ i = µC W
D ox
L (vGS - VT)
OFF Characteristics of a MOS Switch
If vGS < VT, then iD = IOFF = 0 when vDS 0V.
If vDS > 0, then
1 1
ROFF ≈ = ≈∞
iD IOFF
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 14 – The MOS Switch and Diode (6/14/14) Page 14-4
• To insure that the bulk-source and bulk-drain pn junctions are reverse biased, the bulk
voltage must be less than the minimum analog signal for a NMOS switch.
• To insure that the switch is on, the gate voltage must be greater than the maximum
analog signal plus the threshold for a NMOS switch.
Therefore:
VBulk 0V
VGate(on) > 1V + VT
VGate(off) 0V
Unfortunately, the large value of reverse bias bulk voltage causes the threshold voltage
to increase.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 14 – The MOS Switch and Diode (6/14/14) Page 14-5
iD 0mA VGS=1.5V
VGS=1.0V
-50mA
-100mA
-1V -0.5V 0V 0.5V 1V
vDS Fig. 4.1-4
Example
Initially assume the capacitor is uncharged. If VGate(ON) is 5V and is high for 0.1µs,
find the W/L of the MOSFET switch that will charge a capacitance of 10pF in five
time constants.
Solution
The time constant must be 100ns/5 = 20ns. Therefore RON must be less than
20ns/10pF = 2k. The ON resistance of the MOSFET (for small vDS) is
1 W 1 1
RON = K ’(W/L)(V -V ) L = R ·K ’(V -V ) =
N GS T ON N GS T 2k·110µA/V2·4.3
=1.06
Comments:
• It is relatively easy to charge on-chip capacitors with minimum size switches.
• Switch resistance is really not constant during switching and the problem is more
complex than above.
1 gON(0) + gON(∞)
gON(aver.) = r ≈
ON(aver.) 2
K’W K’WVDS(0) K’W
= 2L (VGS-VT) - + 2L (VGS-VT)
4L
K’W K’WVDS(0)
= (VGS-VT) -
L 4L
Gate-source Varying
Typically, no problems occur unless capacitance voltages are held for a long time. For
example,
vout(t) = vCH e-t/(RBulkCH)
If RBulk 109 and CH = 10pF, the time constant is 109·10-11 = 0.01seconds
OFF OFF
Qch = -WLCox(VH-vin-VT)
where VH is the value of the clock waveform when the switch is on (VH ≈ VDD)
When the switch turns OFF, this charge is injected ON
into the source and drain terminals as shown. Clk
vin
Assuming the charge splits evenly, then the change of OFF DV
voltage across the capacitor, CL, is
vin e- e-
CL
Qch -WLCox(VH-vin-VT)
V = 2C = 2CL 060613-04
L
The charge injection does not influence vin because it is a voltage source.
Clock Feedthrough
In addition to the charge injection, the overlap capacitors of the MOSFET couple the
turning off part of the clock to the load capacitor. This is called clock feedthrough.
The model for this case is given as:
The gate decrease from B to C is modeled as a negative step of magnitude VS +VT - VL.
The output voltage on the capacitor after opening the switch is,
CL COL COL COL
vCL = C +C VS-C +C VT -(VS+VT -VL)C +C ≈ VS-(VS+2VT -VL) C
OL L OL L OL L L
if COL < CL.
Therefore the error voltage is,
COL COL
Verror ≈ -(VS + 2VT – VL) C = -(vin + 2VT – VL) C
L L
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 14 – The MOS Switch and Diode (6/14/14) Page 14-15
To begin the model development, there are two cases of charge injection depending
upon the transition rate when the switch turns off.
1.) Slow transition time – the charge in the channel can react instantaneously to changes
in the turning-off, gate-source voltage.
2.) Fast transition time – the charge in the channel cannot react fast enough to respond
to the changes in the turning-off, gate-source voltage.
Charge
injection
vin CL vin CL
Fig. 4.1-13
1.) During the on-to-off transition time from A to B, the charge injection is absorbed by
the low impedance source, vin.
2.) The switch turns off when the gate voltage is vin+VT (point B).
3.) From B to C the switch is off but the gate voltage is changing. As a result charge
injection occurs to CL.
Charge Charge
injection injection
vin CL vin CL
Fig. 4.1-14
vGATE vGATE
Charge
vin+VT vin+VT injection
vin vin due to fast
vCL vCL transition
t t
Slow Transition Fast Transition Fig 4.1-15
The time constant of the channel, Rchannel·Cchannel, determines whether or not the
capacitance, CL, fully charges during each voltage step.
†B.J. Sheu and C. Hu, “Switched-Induced Error Voltage on A Switched Capacitor,” IEEE J. Solid-State Circuits, Vol. SC-19, No. 4, pp. 519-
525, August 1984.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 14 – The MOS Switch and Diode (6/14/14) Page 14-19
W·CGD0 + Cchannel 3
2 VHT W·CGD0
Verror = - VHT - 6U·C - C (VS+2VT -VL)
CL L L
Case 1:
The value of U is equal to 5V/0.2nS or 25x109. Next we must test to see if the
slow or fast transition time is appropriate. First calculate the value of VT as
VT = VT0 + 2|F| -VBS - 2|F| = 0.7 + 0.4 0.7+1 - 0.4 0.7 = 0.887V
Therefore,
2
VHT 110x10-6·3.1132
VHT =VH-VS-VT = 5-1-0.887=3.113V 2C = = 5.32x10 8< 25x109
L 2·1pF
which corresponds to the fast transition case. Using the previous expression gives,
Verror =
176x10-18+0.5(1.58x10-15) 3.32x10-3 176x10-18
- 3.113- - (1+1.774-0) = -3.39mV
1x10 -12 30x10 1x10
-3 -12
Clock
A B A B
VDD
Clock
Advantages:
• Feedthrough somewhat diminished
• Larger dynamic range
• Lower ON resistance
Disadvantages:
• Requires a complementary clock
• Requires more area
2
NVHTN 110x10-6·(5-2.5-0.7)2
= = 1.78x10 8
2CL 2·10-12
For the PMOS transistor, noting that
VHTP = VS - |VTP| - VL = 2.5-0.7-0 = 1.8
2
PVHTP 50x10-6·(1.8)2
we have 2C = -12 = 8.10x107 . Thus, the NMOS transistor is in the
L 2·10
slow transition and the PMOS transistor is in the fast transition regimes.
176x10-18+0.5(1.58x10-15)
·108·10-12 176x10-18
Verror(NMOS) = - - (2.5+1.4-0)
10-12
2·110x10-6 10-12
= -1.840mV
Error due to PMOS (fast transition):
A B
VDD
VA,B 1mA
M2
Fig. 4.1-22
Spice File:
Simulation CMOS transmission switch resistance VDD 3 0
M1 1 3 2 0 MNMOS L=1U W=10U VAB 1 0
M2 1 0 2 3 MPMOS L=1U W=10U IA 2 0 DC 1U
.MODEL MNMOS NMOS VTO=0.7, KP=110U, .DC VAB 0 3 0.02 VDD 1 3 0.5
+LAMBDA=0.04, GAMMA=0.4, PHI=0.7 .PRINT DC V(1,2)
.MODEL MPMOS PMOS VTO=-0.7, KP=50U, .END
+ LAMBDA=0.05, GAMMA=0.5, PHI=0.8
Result:
Low ON resistance over a wide voltage range is difficult as the power supply decreases.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 14 – The MOS Switch and Diode (6/14/14) Page 14-29
C2
vOUT = 2VDD C + C + C
2 L NMOSswitch
Simulation:
3.0
Output
2.0
Input
Volts
1.0
0.0
-1.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0
Time (ms) Fig. 4.1-24
†T.B. Cho and R.R. Gray, “A 10b, 20 Msample/s, 35mW Pipeline A/D Converter,” IEEE J. of Solid-State Circuits, Vol. 30, No. 3m March 1995,
pp. 166-172.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 14 – The MOS Switch and Diode (6/14/14) Page 14-31
low: M7 and M10 make vg=0 and C3 charges to VDD, high: C3 connected to vGS11.
M7 reduces the vDS and vGS of M10 when = 0. M13 ensures that vGS8 ≤ VDD.
The parasitics at the source of M11 require this node to be driven from a low impedance.
†A.M. Abo and P.R. Gray, “A 1.5V, 10-bit, 14.3-MS/s CMOS Pipeline Analog-to-Digital Converter, IEEE J. of Solid-State Circuits, Vol. 34,
No. 5, May 1999, pp. 599-605.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 14 – The MOS Switch and Diode (6/14/14) Page 14-32
MOSFET DIODE
MOS Diode
When the MOSFET has the gate connected to the drain, it acts like a diode with
characteristics similar to a pn-junction diode.
i
+ +
i
vSG = v
vGS = v
i
- -
v
VT Fig. 4-2-1
Note that when the gate is connected to the drain of an enhancement MOSFET, the
MOSFET is always in the saturation region.
vDS vGS - VT vD - vS vG - vS - VT vD - vG -VT vDG -VT
Since VT is always greater than zero for an enhancement device, then vDG = 0 satisfies
the conditions for saturation.
• Works for NMOS or PMOS
• Note that the drain could be VT less than the gate and still be in saturation
If the threshold voltage is less than 0.4V, the MOS diode can provide more current for
the same voltage than a pn junction diode even for modest W/L ratios. However, at the
same value of current, the pn junction always has a larger transconductance.
SUMMARY
• Symmetrical switching characteristics
• High OFF resistance
• Moderate ON resistance (OK for most applications)
• Clock feedthrough is proportional to size of switch (W) and inversely proportional
to switching capacitors.
• Output offset due to clock feedthrough has 2 components:
Input dependent
Input independent
• Complementary switches help increase dynamic range.
• Fully differential operation should minimize the clock feedthrough.
• As power supply reduces, floating switches become more difficult to fully turn on.
• Switches contribute a kT/C noise which can get folded back into the baseband.
• The gate-drain connected MOSFET can make a good diode realization
10µA vDS
1V 10V 060526-10
vds 1
AC resistance = =
id gds
where
gds 2 (VGS-VT)2 = ID
VDS VT 2
DC resistance = I = I +
D D ID
vds 1 1
AC resistance = i = g + g g
d m ds m
where
gm = (VGS-VT) = 2ID
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 15 – Resistor Implementations and Current Sinks and Sources 5/14/18) Page 15-4
20mA
VGS=2V
-20mA
VGS=3V
VGS=4V
-60mA
VGS=5V
VGS=6V
-100mA
-1V -0.6V -0.2V 0.2V 0.6V 1V Fig. 4.2-95
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 15 – Resistor Implementations and Current Sinks and Sources 5/14/18) Page 15-5
1
iAB = 2ß(VC - VT)vAB RAB = 2ß(V - V )
C T
Io
i +
Io v
- v
060527-01
1 1+VDS 1
rout = di /dv = ≈ and VMIN = VDS(sat) = VGS - VT0 = VGG - VT0
D DS D ID
Note: The NMOS current sink can only have positive values of v.
0 vSD = v
0 VGG-|VT0| VDD
0601527-03
ID
Enhance Provide
Channel Current
0 vGS
0 VT VGS Fig. 280-03
VGS = VT0 + VON = Part to enhance the channel + Part to cause current flow
where
VON = VDS(sat) = VGS - VT0
2ID
VMIN = VON = VDS(sat) = for the simple current sink.
K’(W/L)
Note that VMIN can be reduced by using large values of W/L.
100
Slope = 1/Rout
80
iOUT (mA) iOUT
10mm
60 1mm +
VGS1 = vOUT
40 -
1.126V
20
Vmin
0
0 1 2 3 4 5
vOUT (Volts)
Comments:
VMIN is too large - desire VMIN to approach zero, at least approach VCE(sat)
Slope too high - desire the characteristic to be flat implying very large output
resistance
(KN’ = 110µA/V2, VT = 0.7Vand = 0.04V-1) rds = 250k
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 15 – Resistor Implementations and Current Sinks and Sources 5/14/18) Page 15-12
†
R.B. Blackman, “Effect of Feedback on Impedance,” Bell Sys. Tech.J., Vol. 23, pp. 269-277, October 1943.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 15 – Resistor Implementations and Current Sinks and Sources 5/14/18) Page 15-15
Noting that vgs1 = vg2 = vb2 = 0 and writing a loop equation we get,
vout = (iout - gm2vgs2 - gmbs2vbs2)rds2 + rds1iout
However,
vgs2 = 0 - vs2 = -ioutrds1 and vbs2 = 0 - vs2 = -ioutrds1
Therefore,
vout = iout[rds1 + rds2 + gm2rds1rds2 + gmbs2rds1rds2]
or
vout
rout = i = rds1 + rds2 + gm2rds1rds2 + gmbs2rds1rds2 gm2rds1rds2
out
A general principle is beginning to emerge:
The output resistance of a cascode circuit R x (Common source voltage gain of the
cascoding transistor)
M2 +
+ VDS2 ≥VDS2(sat)
VGS2 - vOUT(min) = VDS1(sat)+VDS2(sat)
- +
VGG2
VDS1= VDS1(sat)
VGG1 -
060527-06
Thus, for the previous NMOS current sink, VGG2 would be equal to,
VGG2 = 2(0.426) + 0.7 = 1.552V
iOUT (mA)
10mm/1mm +
signal output resistance for the simple 60 VGG2 =
current sink if IOUT = 100µA and (b) 1.552V vOUT
the small-signal output resistance for 40
the cascode current sink with IOUT = VGG1 =
-
20 1.126V
100µA. Assume that all W/L values Vmin
are 1. 0
0 1 2 3 4 5
vOUT (Volts) Fig. 280-12
Solution
(a) Using = 0.04 V-1 and IOUT = 100µA gives rds1 = 250k = rds2. (b) Ignoring the
bulk effect, we find that gm1 = gm2 = 469µS which gives rout =
(250k)(469µS)(250k) = 29.32M.
2ID
VON = iOUT iOUT
K’(W/L) M2 + VMIN
M4 +
1/1 VON
then if L/W of M4 is 1/4 + +
VT+VON -
quadrupled, VON is VT+2VON M3 - vOUT
M1 +
doubled to get 1/1 + VON
-
VT+VON 1/1 -
VMIN = 2VON. - -
0 2VON vOUT
Example 060527-07
Use the cascode current sink configuration above to design a current sink of 100µA and
a VMIN = 1V. Assume the device parameters of Table 3.1-2.
Solution
With VMIN = 1V, choose VON = 0.5V. Assuming M1 and M2 are identical gives
W 2·IOUT 2·100x10-6 W1 W2 W3 W4
= = = 7.27 L = L = L = 7.27 and L = 1.82
L K’·VON2 110x10-6x0.25 1 2 3 4
Unfortunately, the drain voltages of M1 and M3 are not matched.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 15 – Resistor Implementations and Current Sinks and Sources 5/14/18) Page 15-22
†T.L. Brooks and A.L. Westwick, “A Low-Power Differential CMOS Bandgap Reference,” Proc. of IEEE Inter. Solid-State Circuits Conf., Feb.
1994, pp. 248-249.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 15 – Resistor Implementations and Current Sinks and Sources 5/14/18) Page 15-26
3 +
Rout(3) = g + 4rds VDS(min)=VDSat
m
-
1 +
Rout(2) = g + 2rds +
m VDS(min)=VDSat VDS=0V
- -
+
+ +
VDS(min)=VDSat VDS=0V VDS=0V
- - -
+ + +
VDS=0V VDS=0V VDS=0V
- - -
150527-03
2n-1-1
It can be shown that Rout(n) is g + 2n-1rds if the gates are grounded. Therefore, the
m
output resistance is increasing by a factor of 2n-1 for each cascade device and the
minimum voltage across the sink remains constant at VDSat.
The upper transistor is in saturation while all the other transistors have VDS = 0 which
implies that gm = 0 and rds = 1/gm(sat).
This really only works well if the transistors are isolated and the bulk can be connected
to the source.
The area required for the sink will increase significantly because of the isolation.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 15 – Resistor Implementations and Current Sinks and Sources 5/14/18) Page 15-27
Comments:
• Achieves very high output resistance by increasing the loop gain (return-ratio) due to
the M4-M5 inverting amplifier.
gm4 gm3rds2gm4rds4
LG = gm3rds2g +g
ds4 ds5 2
rds3gm3rds2gm4rds4
If rds4rds5, then rout
2
• M3 maintains “constant” current even though it is no longer in the saturation region.
†E. Sackinger and W. Guggenbuhl, “A Versatile Building Block: The CMOS Differential Difference Amplifier,” IEEE J. of Solid-State Circuits,
vol. SC-22, no. 2, pp. 287-294, April 1987.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 15 – Resistor Implementations and Current Sinks and Sources 5/14/18) Page 15-29
A number of solutions exist. For example, let IB = IREF. This gives ID4A = 5.824IREF
assuming all W/L ratios are identical.
105
BJT Cascode
100 MOS Cascode
Regulated
iOUT (mA)
95 MOS
Cascode
90
85
80
0 0.1 0.2 0.3 0.4 0.5
vOUT (V) Fig. 290-12
Comments:
• The regulated cascode current is smaller than the cascode current because the drain-
source voltages of M1 and M2 are not equal.
• The regulated cascode current sink has a smaller VMIN due to the fact that M3 can
have a drain-source voltage smaller than VDS(sat)
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 15 – Resistor Implementations and Current Sinks and Sources 5/14/18) Page 15-34
SUMMARY
Summary of Both BJT and MOS Current Sinks/Sources
Current Sink/Source rOUT VMIN
Simple MOS Current Sink 1
rds = VDS(sat) =
D
VON
Simple BJT Current Sink VA VCE(sat)
ro =
C 0.2V
Cascode MOS gm2rds2rds1 2VON
Cascode BJT Fro 2VCE(sat)
Regulated Cascode Current Sink rds3gm3rds2gm4(rds4||rds5) VT +VON
Minimum VMIN Regulated rds3gm3rds2gm4(rds4||rds5) VON
Cascode Current Sink
Resistor Implementations
• MOSFET resistors may use less area than actual resistors
• Linearity is the primary issue for MOSFET resistor realizations
iIN iOUT
Current Current
Mirror Mirror
060528-01
The above current mirrors are referenced with respect to ground. Current mirrors can
also be referenced with respect to VDD and can source input and output currents.
Therefore, Rout, Rin, VMIN(out), VMIN(in), and Ai will characterize the current mirror.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 16 – Current Mirrors and Simple References (6/14/14) Page 16-4
+ M1 M2 +
vDS1 + vDS2
Assume that vDS2 > vGS - VT2, then - vGS
- -
-
iO L1W2VGS-VT221 + vDS2 K2’ Fig. 300-02
iI = W1L2VGS-VT1 1 + vDS1 K1’
If the transistors are matched, then K1’ = K2’ and VT1 = VT2 to give,
iO L1W21 + vDS2
iI = W1L21 + vDS1
If vDS1 = vDS2, then
iO L1W2
iI = W1L2
Therefore the sources of error are:
1.) vDS1 vDS2
2.) M1 and M2 are not matched.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 16 – Current Mirrors and Simple References (6/14/14) Page 16-5
Key: Make the part of VGS causing the current to flow, VON, more significant than VT.
We note that M1 M2
+ +
the tolerance VDS1 VDS2
+
is not multi- GND
-
VGS
-
plied by the -
factor of 4.
The ratio of W2 to W1 and consequently the gain of the current amplifier is
iO W2 20 ± 0.1 1 ± (0.1/20)
0.1 ±0.1
0.1 ±0.4
iI = W1 = 5 ± 0.1 = 4 1 ± (0.1/5) ≈ 41 ± 20 1 - 5 ≈ 41 ± 20 - 20 = 4 - (±0.03)
where we have assumed that the variations would both have the same sign (correlated). It
is seen that this ratio error is 0.75% of the desired current ratio or gain.
M1 M2
GND
GND
Fig. 300-6
Ib Ib
Ib
M3 M4
iI iO
iI VT iO VT
+ M1 M2
VON +
+ M1
VON + M2 - VT+VON Ib
- VT+VON -
-
Fig. 300-7
IIN IOUT
R
• Rout gm2rds2rds1 iin iout
D5=G3
M4 M5 M2 +
1/1 rds5
vin rds5 + rds3 + rds3gm5rds5 1 1/4 1/1 gm5vgs5
• Rin = i = g r (1+g r ) ≈ g iin vin
in m3 ds3 m5 ds5 m3 M3 M1
D3=S5 +
1/1 1/1 gm3vgs3
rds3 vs5
An easier way to find Rin: = gm3vin
- S3=G5 -
060528-02
1.) Apply a small voltage change, vin, at the input.
VDD
2.) Note that this voltage is equal to vgs3.
IIN
3.) This small voltage change causes a current change iin
• VMIN(in) = VT + VON
• Current gain is excellent because vDS1 = vDS3.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 16 – Current Mirrors and Simple References (6/14/14) Page 16-11
+ iin R g v
• Rin = ? R
+ +
m3 gs3
M3 M4
vin = iinR + rds3(iin-gm3vgs3) vin rds3
+
vin v2
+ rds1(iin-gm1vgs1) M1 M2 gm1vgs1 v1 rds1
-
But, - - -
vgs1 = vin-iinR
Self-biased, cascode current mirror Small-signal model to calculate Rin.
and Fig. 310-03
M3
M1
M4 M2
FIG. 310-11
• Rout gm2rds3
• Rin 1
gm4
• VMIN(out) = VT+2VON (Can be reduced to 2VON)
• VMIN(in) = VT+VON (Can be reduced to VON)
• Current gain matching - good as long as vDS4 = vDS2
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 16 – Current Mirrors and Simple References (6/14/14) Page 16-13
M3 M4 M5
i2 i2=Vtln(i1/Is)/R
M6 I
1 Desired i2=i1
I2 I5 operating
point
M1 M2
+
VEB1
Undesired
+ operating
M7 Q1 - point
R VR
Startup
i1
- -
070621-01
VEB1
Iout = I2 =
R
M3 M4 VT+VON
VON VT
I1
VT I2
VON
M1 M2
VT+VON
R VR
However, not only is VREF a function of T, but R and Is are also functions of T.
dVREF k VDD-VREF kT RIs -1 dVREF VDD-VREF dR dIs
dT = q ln
+ q V -V - + I dT
RI s DD REF s RI dT RI s RdT s
VREF Vt dVREF dR dIs VREF-VGO Vt dVREF 3Vt Vt dR
= T - V -V - Vt RdT + I dT = - V -V - T - R dT
DD REF dT s T DD REF dT
VREF-VGO dR 3Vt
dVREF - Vt RdT - T V -V
T dR 3Vt
REF GO
dT = - Vt RdT - T
Vt T
1 + V -V
DD REF
5 − 1.281 1.5
dVREF −2.3x10-3 + − 1500x10 -6
222 300
Now, = = -1.189x10-3V/°C
dT 1
1+
222 (5 - 1.281)
The fractional temperature coefficient is given by
1
TCF = −1.189x10 1.281 = −928 ppm/°C
-3
PTAT Voltage
The principle illustrated on the last slide requires perfectly linear positive and negative
temperature coefficients to work properly. We will now show a technique of generating
PTAT voltages that are linear with respect to temperature.
Implementation of a PTAT voltage:
I1 I2
VPTAT = VD = VD1 – VD2 = Vt ln I - Vt ln I
s1 s2
I1 Is2 Is2 A2 kT A2
= Vt ln I = Vt ln
2 Is1 Is1 = Vt ln A1 = q ln A1
if I1 = I2.
Psuedo-PTAT Currents
In developing temperature independent voltages, it is useful to show
how to generate PTAT currents. A straight-forward method is to
superimpose VPTAT across a resistor as shown:
Because R is always dependent on temperature, this current is called a pseudo-PTAT
current and is designated by IPTAT’.
When a pseudo-PTAT current flows through a second
resistor with the same temperature characteristics as the
first, it creates a new VPTAT voltage.
The new VPTAT voltage, VPTAT2 is equal to,
R2
VPTAT2 = R VPTAT1
1
Differentiating with respect to temperature gives
dVPTAT2 R2 dR2 dR1 dVPTAT1
= - +
dT R1 R2dT R1dT dT
Therefore, if the temperature coefficient of R1 and R2 are equal, then the temperature
dependence of VPTAT2 is the same as VPTAT1.
M1 M2
M3 M5 M3
I1 M4 I2 I1 I2
IPTAT’ IPTAT’
- +
M1 + + M2
VGS1 VGS2
- -
R IPTAT’ R IPTAT’
D1 D1
A1 D2 A1 D2
A2 A2
In these circuits, I1 = I2 and the voltage across D1 is made equal to the voltage across the
series combination of R and D2 to create the pseudo-PTAT current,
VD1 - VD2 kT A2
IPTAT’ = = Rq ln A
R 1
where VGS1 = VGS2 for the MOSFET only version.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 17 – Temperature Stable References (5/15/18) Page 17-6
CTAT Voltage
This becomes more challenging because a true CTAT voltage does not exist. The best
approach is to examine the pn junction (can be a diode or BJT).
The diode voltage can be written as
æ iD ö
vD = Vt ln ç ÷ = Vt ln(iD ) -Vt ln(I s )
è Is ø
where g
æ -VBG ö
I s = AT exp ç ÷ and iD = BT a
è Vt ø
and where A and B are temperature independent constants, is the temperature
coefficient for Is ( ≈ 3), is the temperature coefficient for iD ( =1 for PTAT), and VBG
is the bandgap voltage of silicon (1.205V at 27°C).
The diode voltage as a function of temperature is,
é g æ VBG öù
÷ú = VBG -Vt (g - a )ln(T ) -Vt ln(A / B)
a
vD (T ) = VCTAT = Vt ln(BT ) -Vt ln êAT exp ç -
ë è Vt øû
Note that the term Vt(-)ln(T) is not linear with temperature and cannot completely
cancel the perfectly linear PTAT voltage.
IPTAT’
M3 M4 M1
M4 M5 M2
M5 M6
IPTAT’ IPTAT’ I2
ICTAT’ ICTAT’ ICTAT’ ICTAT’
ICTAT’ - +
M2 M3
M1 M2
Q1 + + R ICTAT’
VBE R VD R D2
A2
- -
Generation of a pseudo CTAT current Generation of a pseudo CTAT current Generation of a pseudo CTAT current using
using a bipolar transistor. using a diode. MOSFETs, an op amp and pn junctions.
120326-01
The negative feedback loop shown causes the current designated as ICTAT’ to be,
VBE VD
ICTAT’ = R = R
†
I.M. Gunawan, G.C.M. Jeijer, J. Fonderie, and J.H. Huijsing, “A Curvature-Corrected Low-Voltage Bandgap Reference, IEEE J. Solid-state Circuits, vol. SC-28, No. 6, June
1993, pp. 677-670.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 17 – Temperature Stable References (5/15/18) Page 17-8
Series form:
R2
VREF = IPTAT’R2 + VD = R VPTAT + VCTAT
1
Parallel form:
R3 R3 R3R2
VREF = (IPTAT’ + ICTAT’)R3 = R VPTAT + R VCTAT = R R VPTAT + VCTAT
1 2 2 1
To achieve temperature independence, VREF must be differentiated with respect to
temperature and set equal to zero. The resistor ratios and other parameters can be used
to achieve temperature independence.
dVREF V V V V
(T = T0 ) = K ln(A2 / A1 ) t 0 - t 0 ln(A/B) - t 0 (g - a )ln(T0 ) - (g - a ) t 0
dT T0 T0 T0 T0
Equating the derivative to zero gives,
Kln(A2 / A1 )- ln(A/B) = (g - a )[1+ ln(T0 )]
Substituting back gives,
é æ T0 öù
VREF = (g - a )Vt [1+ ln(T0 )]+VBG -Vt (g - a )ln(T ) = VBG + (g - a )Vt ê1+ ln ç ÷ú
ë è T øû
At T = T0, assuming = 3.2 and = 1, the reference voltage is
VREF = VBG + Vt(-) = 1.205V + 0.057V = 1.262V (T0 = 27°C)
Because VREF ≈ VBG, this voltage reference is called the “bandgap reference”.
Example 17-1 – Temp. Independent Constant for Series and Parallel References
(a.) Design the ratio of R2/R1 for the series configuration if VCTAT = 0.6V and A2/A1 = 10
for room temperature (Vt = 0.026V). Assume = 3.2 and = 1. Find the value of VREF.
R2 VGO - VCTAT + (-)Vt0 1.205 - 0.6 + 2.2(0.026)
R1 = = = 11.05
VPTAT 0.026(2.3026)
VREF = 1.205 + 2.2(0.026) = 1.262V
If R1 = 1k, then R2 = 11.05k
(b.) For the parallel configuration find the values of R2/R1 and R3/R2 if VREF = 0.5V.
From (a.) we know that R2/R1 = 11.05. We also know that,
R3 R3 R3R2
VREF = R VPTAT + R VCTAT = R R VPTAT + VCTAT
1 2 2 1
= (R3/R2)[11.05ln(10)(0.026) + 0.6] = (R3/R2)1.262 = 0.5
(R3/R2) = 0.3963
If R1 = 1k, then R2 = 11.05k and R3 = 4.378k
†
A.P. Brokaw, “A Simple Three-Terminal IC Bandgap Reference,” IEEE J. Solid-State Circuits, Vol. SC-6, No. 1, 1971, pp. 2-7.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 17 – Temperature Stable References (5/15/18) Page 17-14
VBE2 - VBE1 Vt I2 I1
VD1 = I2R1 + VD2
I1 = IPTAT’ = = ln - ln
R2 R2 Is2 Is1
Vt Is1 Vt AE1 Vt
= R lnI = R lnA I3 = I2 = IPTAT’ = R ln(n)
2 s2 2 E2 1
R1 AE1 Let R1 = R and R2 = kR,
Since I1= I2, VREF = VBE2 + I1R1 = VBE2 + R lnA Vt
2 E2 VREF = VD3 + I3(kR) = VD3 + kVt ln(n)
R1
= VCTAT + R VPTAT = VCTAT + kVPTAT
2
R3 R3
VREF = R VPTAT + R VCTAT
1 2
Comments:
• The BJT of the ICTAT’ generator can be replaced with an MOSFET-diode equivalent
• Any value of VREF can be achieved
• Part (b.) of Example 17-1 showed how to design the resistors of this implementation
VDD VDD
IPTAT IREF
IVBE
+
R3 VREF =VGS(ZTC)
-
060529-09
Comments:
• Ability of the ZTC point not to drift with temperature restricts the temperature range
• The reference voltage must be equal to the ZTC voltage
• The voltage VREF will suffer the bandgap curvature problem which can be translated
into IREF.
Comments:
• True temperature independence is only achieved over a small range of temperatures
• References that do not correct this problem have a temperature dependence of 10
ppm°/C to 50 ppm/°C over 0°C to 70°C.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 17 – Temperature Stable References (5/15/18) Page 17-20
• VBE loop
Voltage
M. Gunaway, et. al., “A Curvature- VPTAT2
Corrected Low-Voltage Bandgap
Reference,” IEEE Journal of Solid- VRef = VBE + VPTAT + VPTAT2
State Circuits, vol. 28, no. 6, pp. 667- Temperature
670,
Fig. 400-01
June 1993.
• Nonlinear cancellation
G.M. Meijer et. al., “A New Curvature-Corrected Bandgap Reference,” IEEE
Journal of Solid-State Circuits, vol. 17, no. 6, pp. 1139-1143, December 1982.
47.16k(3.2)
2.) R3 = R2 = = 68.60 k
-1 2.2
R4 VGO VREF 0.72
3.) VREF = 0.72V = R ⇒ R4 = V R2 = 1.205 (47.16 k)) = 90.17 k
2 GO
10GH
• Influence of the op amp VOS vout
vin
IX PTAT Block
111208-09
RX R4 VREF
DQX
x1
C1 E2
Since iC2R3 = iC1R2 - VOS
iC2 R2 VOS R2 VOS
then = - = 1+
iC1 R3 iC1R3 R3 iC1R2
R A VOS
2 E1
Therefore, VR1 = -VOS + Vt ln 1 +
R3AE2 iC1R2
V R
VREF = VBE2 - VOS + iC1R2 = VBE2 - VOS + R R2 = VBE2 - VOS + R
R1 2
1 1
R2 R2 R A VOS
2 E1
VREF = VBE2 - VOS1+ R + R Vt ln R A 1 - i R
CMOS Analog Circuit Design 1 1 3 E2 C1 2 © P.E. Allen - 2016
Lecture 17 – Temperature Stable References (5/15/18) Page 17-29
ID1 ID2
VBias 100µA
M1 M2
100µA
050716-01 1mm
If the bus metal is 50m/sq. and is 5µm wide, the resistance of the bus in one direction
is (50m/sq.)x(1000µm/5µm) = 10 The difference in drain currents for an overdrive
of 0.1V is,
VGS1 = 1mV + VGS2 + 1mV = VGS2 + 2mV
ID1 (VGS1-VTN)2 (VGS2-VTN+2mV)2 0.1+0.0022
ID2 = (VGS2-VTN)2 = = 0.1 = 1.04
(VGS2-VTN)2
The currents are used to distribute the bias voltages to remote sections of the chip.
Ib Ib VPBias2
VNBias2
VNBias1
Fig. 400-08
INTRODUCTION
Types of Amplifiers
Most CMOS amplifiers fit naturally into the transconductance amplifier category as they
have large input resistance and fairly large output resistance.
If the load resistance is high, the CMOS transconductance amplifier is essentially a
voltage amplifier.
Characterization of an Amplifier
1.) Large signal static characterization:
• Plot of output versus input (transfer curve)
• Large signal gain
• Output and input swing limits
2.) Small signal static characterization:
• AC gain
• AC input resistance
• AC output resistance
3.) Small signal dynamic characterization:
• Bandwidth
• Noise
• Power supply rejection
4.) Large signal dynamic characterization:
• Slew rate
• Nonlinearity
vOUT
D at ve
1 s acti
2 M 1
E M
F
1 G H I J K
Fig. 320-02 0
0 1 2v
IN
3 4 5
The boundary between active and saturation operation for M1 is
vDS1 vGS1 - VTN → vOUT vIN - 0.7V
M2 gm2vgs2
ID vOUT G1 rds2 Rout
D1=D2=G2
+ + + +
vIN vin gm1vgs1 vout vin vout
M1 rds1 gm1vin rds1 gm2vout rds2
- - - -
S1=B1 Fig. 320-03
Sum the currents at the output node to get,
gm1vin + gds1vout + gm2vout + gds2vout = 0
Solving for the voltage gain, vout/vin, gives
vout −gm1 gm1 K'NW1L2
= − =−
vin gds1 + gds2 + gm2 gm2 P 1 2
K' L W
The small-signal output resistance can also be found from the above by letting vin = 0 to
get,
1 1
Rout = g + g + g g
ds1 ds2 m2 m2
+ GoutVout + sCoutVout = 0
Vout(Gout + sCM + sCout) = - (gm – sCM)Vin
sCM
1- g s
−gmRout 1 -
Vout
=
-(gm – sCM)
= -gmRout
m =
z1
Vin Gout+ sCM + sCout 1+ sRout(CM + Cout) 1-
s
p1
−1 gm1
where gm = gm1, p1 = R and z1 = C
(C
out out +CM) M
1
and Rout = [gds1+gds2+gm2] gm2 , CM = Cgd1 , and
-1 Cout = Cbd1+Cbd2+Cgs2+CL
It is useful to plot the roots of the transfer function on the complex frequency plane.
For the previous T(s), the roots are:
The numerator root (zero) is s = z1 = +(gm/C1)
The denominator root (pole) is s = p1= -[1/R2(C1+ C2)]
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 18 – Inverting Amplifiers (8/14/17) Page 18-11
To construct a Bode asymptotic magnitude plot for a low pass transfer function in the
form of products of roots:
1.) Start at a low frequency and plot 20 log10(|T(0)| until you reach the smallest root.
2.) At the frequency equal to magnitude of the smallest root, change to a line with a
slope of +20dB/decade if the root is a zero or -20dB/decade if the root is a pole.
3.) Continue increasing in frequency until you have plotted the influence of all roots.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 18 – Inverting Amplifiers (8/14/17) Page 18-14
Note: The roots maximally influence the magnitude when is such that the angle
between the vector and the horizontal axis is 45°. This occurs at j1 for p1 and j10 for z1.
So, back to the frequency response of the active load inverter, we find that if |p1| < z1,
then the -3dB frequency is approximately equal to the magnitude of the pole which is
[Rout(Cout+CM)]-1.
dB
20log10(gmRout)
z1
0dB log10w
|p1| » w-3dB
0512-06-02.EPS
Observation:
In general, the poles in a MOSFET circuit can be found by summing the capacitance
connected to a node and multiplying this capacitance times the equivalent resistance
from this node to ground and inverting the product.
ID (mA)
M1 +
KJIH F E vIN=2.0V vOUT
0.2
M2: vSD2 vSG2 - |VTp| 0.1
G M2
D
+
vIN
W1 = 2mm
L1 1mm
vIN=1.5V - -
C
or 0.0
0 1 2 3 4
A,B
5
vIN=1.0V
vOUT 5 A B C
VDD-vOUT VDD -VGG2 - |VTp| D
4
or M2 active
3 M2 saturated
t u rat
ed
a ve
vOUT
1 s ti
vOUT 3.2V 2 M 1 ac
M
E
Swing limits: 1 F
G H I J K
0
vOUT (max) VDD 0 1 2v
IN
3 4 5 Fig. 5.1-5
2
VDD - VGG - |VT2|2
vOUT(min) = (VDD - VT1)1 - 1-
1 V DD - V T1
M2 rds2
VGG2 ID vOUT G1 Rout
D1=D2
+ + + +
vIN vin gm1vgs1 vout vin vout
M1 rds1 gm1vin rds1 rds2
- - - -
S1=B1=G2 Fig. 5.1-5B
Midband Performance:
vout −gm1 2K'NW1 −1 1 1 1
= = !!! and Rout =
vin gds1 + gds2 L1ID 1 + 2 D gds1 + gds2 ID(1 + 2)
vout
vin Strong Inversion
Weak
Invers-
ion
log(IBias)
» 1µA
060614-01
s Fig. 5.1-4
−gmRout 1 -
Vout(s) z1
Vin(s) = s
1-p
1
−1 gm
where gm = gm1, p1 = and z1 =
Rout(Cout+CM) CM
1
and Rout = g and Cout = Cgd2 + Cbd1 + Cbd2 + CL CM = Cgd1
ds1 + gds2
Therefore, if |p1|<|z1|, then the −3 dB frequency response can be expressed as
gds1 + gds2
-3dB 1 =
Cgd1 + Cgd2 + Cbd1 + Cbd2 + CL
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 18 – Inverting Amplifiers (8/14/17) Page 18-21
In terms of the current sink CMOS inverter, these limits are subtracted from 5V to get
vOUT(max) = 4.26V and v OUT (min) = 0V.
To find the small signal performance, first calculate the dc current. The dc current, ID, is
KN’W1 110·1
ID = 2L (VGG1-VTN)2 = 2·1 (3-0.7)2 = 291µA
1
vout/vin = −9.2V/V, Rout = 38.1 k and f-3dB = 2.78 MHz.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 18 – Inverting Amplifiers (8/14/17) Page 18-22
vIN=2.5V + M1 +
vIN=2.5V vOUT
0.4
F vIN W1 = 1mm
G vIN=3.0V E vIN=2.0V L1 1mm
0.2 - -
H vIN=3.5V vIN=4.5V D vIN=1.5V
I
0.0 vIN=1.0V
0 J,K 1 2 3 4 CA,B 5
vOUT A B C D
E
4 d
u rat e Note
at v e the rail-
3 1 s acti
M 1 to-rail
vOUT
e M
c tiv ted output
2 2 a ra
M satu F voltage
2 swing
1 M
G
H I J K
0
0 1 2v 3 4 5 Fig. 5.1-8
IN
eout2 eout2
en12 eeq2
vin M1 vin M1
* *
Fig. 5.1-10
Approach:
1.) Assume a mean-square input-voltage-noise spectral density en2 in series with the gate
of each MOSFET.
(This step assumes that the MOSFET is the common source configuration.)
2.) Calculate the output-voltage-noise spectral density, eout2 (Assume all sources are
additive).
3.) Refer the output-voltage-noise spectral density back to the input to get equivalent input
noise eeq2.
4.) Substitute the type of noise source, 1/f or thermal.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 18 – Inverting Amplifiers (8/14/17) Page 18-26
1/f Noise
KF B
Substituting en2= 2fC WLK’ = fWL , into the above gives,
ox
B1 K'2B2 L1 B1 1/2 K'2B2 L11/2
eeq(1/f) = fW L 1 + K' B L → eeq(1/f) = fW L 1 + K' B L
2
1 1 1 1 2 1 1 1 1 2
To minimize 1/f noise, 1.) Make L2>>L1, 2.) Increase W1 and 3.) choose M1 as a PMOS.
Thermal Noise
2 8kT
Substituting en = into the above gives,
3gm
8kT W2L1K'21/21/2
eeq(th) = 3[2K' (W/L) I ]1/2 1+ L W K'
1 1 1 2 1 1
* M1
Fig. 5.1-13.
SUMMARY
Table of Performance
AC Voltage AC Output Equivalent,
Inverter Bandwidth (CGB=0) input-referred,mean-
Gain Resistance
square noise voltage
p-channel -gm1 1 gm2
gm22
active load gm2 gm2 CBD1+CGS1+CGS2+CBD2 en12 + en22
inverter gm1
Current -gm1 1 gds1+gds2
gm2
source load gds1+gds2 gds1+gds2 CBD1+CGD1+CDG2+CBD2 en12 + en22g 2
inverter m1
Push-Pull -(gm1+gm2) 1 gds1+gds2 gm1en1 2 gm1en1 2
inverter gds1+gds2 gds1+gds2 CBD1+CGD1+CGS2+CBD2 +
m1 m2 gm1+ gm2
g + g
1.) Bulks connected to the sources: No modulation of VT but large common mode
parasitic capacitance.
2.) Bulks connected to ground: Smaller common mode parasitic capacitors, but
modulation of VT.
What are the implications of a large common mode capacitance?
+
R - R
vIN
vIN 0V
Little -
+ Large charging
charging of of capacitance
capacitance 070416-02
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 19 – Differential Amplifier (6/24/14) Page 19-5
Voltage Transfer Function of the Differential Amplifer with a Current Mirror Load
C3
M3 M4 D1=G3=D3=G4
iD3 iD4 iout G1 G2 rds1 S1=S2 rds2 D2=D4
+ vid -
+ + i3 +
iD1 iD2 +
vg1 vg2 rds5 i3 vout
rds4 C2
C1 1
M1 M2 gm3 rds3 gm1vgs1 gm2vgs2
- - -
vid vout S3 S4
M5 iout'
ISS G1 G2 D1=G3=D3=G4 D2=D4
- + vid -
VBias + + i3 +
vgs1 vgs2 C3
vout
gm1vgs1 1
C1 gm2vgs2 i3 rds2 rds4 C2
- - rds1 rds3 gm3 -
S1=S2=S3=S4
Differential Transconductance: Fig. 330-03
†It can be shown that the current mirror causes this requirement to be invalid because the drain loads are not matched. However, we will continue to
use the assumption regardless.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 19 – Differential Amplifier (6/24/14) Page 19-10
Common Mode Analysis for the Current Mirror Load Differential Amplifier
The current mirror load differential amplifier is not a good example for common mode
analysis because the current mirror rejects the common mode signal.
M3 M4 M3 M4 M3 M4
vo1 vo2 vo1 vo2 vo1 vo2
v1 v2
M1 M2 M1 M2 M1 M2
ISS 1
M5x 2 ISS
vid vid ISS 2 2
2 2 vic vic
M5
VBias VBias
Differential-Mode Analysis:
vo1 gm1 vo2 gm2
vid ≈ -2gm3 and v ≈ + 2g
id m4
Note that these voltage gains are half of the active load inverter voltage gain.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 19 – Differential Amplifier (6/24/14) Page 19-13
R1 M2
vo1 gm2vo1 vout
gm1vin
vin M1 R2
Fig. 5.2-10C
2.) i3 = i1 = 0.5gm1vid M3 M4
3.) i4 = i3 = 0.5gm1vid gm1vid gm1vid rout
2 2
4.) The short-circuit output current is +
M1 gm1vid gm2vid M2
i4 - i2 = 0.5gm1vid + 0.5gm2vid = gm1vid + 2 2 -
vid vid
vout
4.) The resistance at the output node, rout, is 2 - + 2
1 + -
rds2||rds4 or g M5 vid
-
ds2 + gds4 VBias
140624-02
5.) vout = (0.5gm1vid+0.5gm2vid )rout
gm1vin gm2vin vout gm1
= g +g =
ds2 ds4 gds2+gds4 vin = gds2+gds4
vin
vgs2 = vg2 - vs2 = vin - (gm2rds1)vgs2 vgs2 = 1+g r
m2 ds1
gm2vin
Thus, iout = 1+g r = gm2(eff) vin
m2 ds1
* M1 M2 * * M1 M2
ito2 vOUT
en32 en42
M3 * * M4 Vout M3 M4
Fig. 5.2-11C
Solve for the total output-noise current to get,
ito 2 = gm12en12 + gm22en22 + gm32en32 + gm42en42
This output-noise current can be expressed in terms of an equivalent input noise voltage,
eeq2, given as ito2 = gm12eeq2
Equating the above two expressions for the total output-noise current gives,
gm3 2
eeq = en1 + en2 + g en32 + en42
2 2 2
m1
1/f Noise (en12=en22 and en32=en42): Thermal Noise (en12=en22 and en32=en42):
2BP K’N BN L12 16kT W3L1K'3
eeq (1/f) = fW L 1+ K’ B L
2 2
eeq (th) = 1+
1 1 P P 3 3[2K'1 (W/L) I
1 1 ] 1/2 L3 1 1
W K'
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 19 – Differential Amplifier (6/24/14) Page 19-20
V1 VOUT
+ +
VGS1 VGS2
-
V2 -
140423-04
ISS
Repeating the previous analysis with the following model for the transistors
W æV - V ö
iD = IT 1 exp çç GS T ÷
÷
L è nVt ø
gives,
æi L ö æi L ö æi LI W ö
VIO = VGS1 - VGS2 = VT 1 + nVt ln çç D1 1 ÷ - V + nV ln ç
÷ T 2 t
D2 2
çI W ÷
÷ = DV + nV ln ç D1 1 T 2 2 ÷
T t çi L I W ÷
I W
è T1 1 ø è T2 2 ø è D2 2 T 1 1 ø
But iD1RD1 = iD2R D2 and W1/L1 = W2/L2 = W/L which gives,
æR I ö
VIO = DVT + ln çç D2 T 2 ÷÷
Define the following, è RD1IT 1 ø
RD1 = R + 0.5R, RD2 = R - 0.5R, IT1 = IT + 0.5IT, and IT2 = IT - 0.5IT
where R = 0.5(RD1 + RD2), R = RD1 - RD2, IT = 0.5(IT1 + IT2), and IT = IT1 – IT2.
Substituting these relationships into the expression for VIO gives,
æ (R - 0.5DR)(I - 0.5DI ) ö æ (1 - 0.5DR R)(1 - 0.5DI I ) ö
VIO = DVT + nVt ln çç T T ÷
÷ = DVT + nVt ln çç T T ÷
÷
è (R + 0.5DR)(IT + 0.5DIT ) ø è (1 + 0.5DR R)(1 + 0.5DIT IT ) ø
» DVT + nVt ln éë(1 - 0.5DR R)2(1 - 0.5DIT IT )2 ùû » DVT + nVt ln éë1 - DR R - DIT IT ùû
æ DR DI ö
» DVT - nVt çç + T ÷÷
è R IT ø
Linearization
vin vin
-ISS -ISS
060608-03
Method (degeneration):
VDD VDD
M3 M4 M3 M4
iout iout
M1 M2 M1 M2
+ RS RS VDD or + VDD
vin 2 2 2 vin RS 2
- -
M5
VNBias1 M5 VNBias1 M6
060118-10
M1 VBias M2 M1 M2
+ VDD + VDD
or
vin M6 2 M6 2
vin
- M7
M5x1/2 -
VNBias1 M5x1/2 M5
VNBias1 M6
Note that these transconductors on this slide and the last can all have a varying
transconductance by changing the value of ISS.
Current Current
I1 I3
I3 I1
0 vDS1 0 vDS1
0 VDS1<VDS(sat) VDD 0 VSD3<VSD(sat) VDD
(a.) I1>I3. (b.) I3>I1. Fig. 5.2-13
Operation:
• Common mode output voltages are sensed at the gates of MC2A and MC2B and
compared to VCM.
• The current in MC3 provides the negative feedback to drive the common mode output
voltage to the desired level.
• With large values of output voltage, this common mode feedback scheme has flaws.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 19 – Differential Amplifier (6/24/14) Page 19-29
MC5 M5
MB
Note that RCM1 and RCM2 must not load the output of the differential amplifier.
(We will examine more CM feedback schemes in Lecture 28.)
Constraints Specifications
Power supply Small-signal gain M3 M4
Technology Frequency response (CL) vout
CL
Temperature ICMR +
Slew rate (CL) vin M1 M2
-
Power dissipation
I5
Relationships
VBias M5
Av = gm1Rout
VSS ALA20
-3dB = 1/RoutCL
VIC(max) = VDD - VSG3 + VTN1
VIC(min) = VSS +VDS5(sat) + VGS1 = VSS +VDS5(sat) + VGS2
SR = ISS/CL
Pdiss = (VDD+|VSS|)x(All dc currents flowing from VDD or to VSS)
Procedure:
1.) Pick ISS to satisfy the slew rate knowing CL or
the power dissipation
2.) Check to see if Rout will satisfy the frequency
response, if not change ISS or modify circuit
3.) Design W3/L3 (W4/L4) to satisfy the upper ICMR
4.) Design W1/L1 (W2/L2) to satisfy the gain
5.) Design W5/L5 to satisfy the lower ICMR
6.) Iterate where necessary
Example 19-1 - Design of a MOS Differential Amp. with a Current Mirror Load
Design the currents and W/L values of the current mirror load MOS differential amplifier
to satisfy the following specifications: VDD = -VSS = 2.5V, SR 10V/µs (CL=5pF), f-
3dB 100kHz (CL=5pF), a small signal gain of 100V/V, -1.5VICMR2V and Pdiss
mW. Use the parameters of KN’=110µA/V2, KP’=50µA/V2, VTN=0.7V, VTP=-0.7V,
N=0.04V-1 and P=0.05V-1.
Solution
1.) To meet the slew rate, ISS 50µA. For maximum Pdiss, ISS 200µA.
2
2.) f-3dB of 100kHz implies that Rout 318k Therefore Rout = 318k
(N+P)ISS
ISS 70µA Thus, pick ISS = 100µA
3.) VIC(max) = VDD - VSG3 + VTN1 → 2V = 2.5 - VSG3 + 0.7
2·50µA
VSG3 = 1.2V = + 0.7
50µA/V2(W3/L3)
W3 W4 2
L =L = 2 =8
3 4 (0.5)
SUMMARY
• Differential amplifiers are compatible with the matching properties of IC technology
• The differential amplifier has two modes of signal operation:
- Differential mode
- Common mode
• Differential amplifiers are excellent input stages for voltage amplifiers
• Differential amplifiers can have different loads including:
- Current mirrors
- MOS diodes
- Current sources/sinks
- Resistors
• The small signal performance of the differential amplifier is similar to the inverting
amplifier in gain, output resistance and bandwidth
• The large signal performance includes slew rate and the linearization of the
transconductance
• The design of CMOS analog circuits uses the relationships of the circuit to design the
dc currents and the W/L ratios of each transistor
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 20 Low Input Resistance Amplifiers (6/24/14) Page 20-1
VNBias1
vIN IBias vIN M1
060609-01
Rout ≈ rds2||rds3
From the previous page, the input resistance to the common gate configuration is,
rds2 + RLoad
Rin = 1 + g r
m2 ds2
For the various loads shown, Rin becomes:
rds2 1 rds2+rds3 2 rds2+rds4gm3rds3
Rin1 = 1+g r ≈ g Rin2 = 1+ g r ≈ g Rin3 = 1+ g r ≈ rds!!!
m2 ds2 m2 m2 ds2 m2 m2 ds2
The input resistance of the common gate configuration depends on the load at the drain.
M1
vIN
060609-05
†“Cascode” = “Cascaded triode” see H. Wallman, A.B. Macnee, and C.P. Gadsden, “A Low-Noise Amplifier, Proc. IRE, vol. 36, pp. 700-708, June
1948.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 20 Low Input Resistance Amplifiers (6/24/14) Page 20-6
vOUT
M2 active
2
F
G H
1
M1 sat- M1 I J K
urated active
0
Fig. 5.3-2 0 1 2v
IN
3 4 5
M1 sat. when VGG2-VGS2 VGS1-VT → vIN 0.5(VGG2+VTN) where VGS1=VGS2
M2 sat. when VDS2VGS2-VTN → vOUT-VDS1VGG2-VDS1-VTN → vOUT VGG2-VTN
M3 is saturated when VDD-vOUT VDD - VGG3 - |VTP| → vOUT VGG3 + |VTP|
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 20 Low Input Resistance Amplifiers (6/24/14) Page 20-7
Assuming that the poles are split allows the use of the previous technique to get,
−1 −1 −gm1C2
p1 = and p2
R1(C1+C2)+R3(C2+C3)+gm1R1R3C2 gm1R1R3C2 C1C2+C1C3+C2C3
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 20 Low Input Resistance Amplifiers (6/24/14) Page 20-14
The Miller effect causes Cgs1 to be increased by the value of 1 + (v1/vin) and appear in
parallel with the gate-source of M1 causing a dominant pole to occur.
The cascode amplifier eliminates this problem by keeping the value of v1/vin small by
making the value of Rs2 approximately 2/gm2.
CURRENT AMPLIFIERS
What is a Current Amplifier?
• An amplifier that has a defined output-input current relationship
• Low input resistance
• High output resistance
Application of current amplifiers:
The above illustration assumes that the GB of the voltage amplifier realizing the voltage
buffer is greater than the GB achieved from the above method.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 20 Low Input Resistance Amplifiers (6/24/14) Page 20-21
1 1 W2/L2
Rin = g Rout = and Ai = W /L .
m1 Io 1 1
Frequency response:
-(gm1+gds1) -(gm1+gds1) -gm1
p1 = = ≈
C1+C2 Cbd1+Cgs1+Cgs2+Cgd2 Cbd1+Cgs1+Cgs2+Cgd2
Note that the bandwidth can be almost doubled by including the resistor, R.
(R removes Cgs1 from p1)
IIN IOUT
iin iout
+ M3 +
VNBias2 M4
vIN vOUT
M1 M2
- -
060610-01
1 W2/L2
Rin g , Rout rds2gm4rds4, and Ai = W /L
m1 1 1
Therefore, we want a configuration where the return ratio (RR) goes to zero when the port
is shorted. We know that the shunt configuration shown below accomplishes this.
It is easy to see that the return ratio for the input shorted is zero and the return ratio for
the input open is,
RR(port opened) = Agm1rds1 ≠ 0
Therefore based on these ideas, a low-input resistance realization is proposed on the next
slide.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 20 Low Input Resistance Amplifiers (6/24/14) Page 20-26
vc
RR(vx = 0): - = 0 RR(ix = 0): vc = - vgs3(1+ gm3rds3) = - gm1rds1 (1+ gm3rds3)vc’
v c'
vc
RR(ix = 0) = -v ' = gm1rds1 (1+ gm3rds3)
c
1+0 1
Finally, Rx = Rin = rds1 1 + g r (1+ g r ) ≈ g g r
m1 ds1 m3 ds3 m1 m3 ds3
Small signal analysis gives the same result and is much easier to calculate.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 20 Low Input Resistance Amplifiers (6/24/14) Page 20-27
i1+i2
iO = AIDiID ± AICiIC = AID(i1 - i2) ± AIC 2
Implementations:
VDD
VDD VDD VDD M3 M4
iO
I 2I I
i1 iO
M1 M2
i2 i1 i2
i2 VGG1
i1-i2
M1 M2 M3 M4
M5 M6
VGG2
Fig. 5.4-7
SUMMARY
• Low input resistance amplifiers use the source as the input terminal with the gate
generally on ground
• The input resistance to the common gate amplifier depends on what is connected to the
drain
• The voltage driven common gate/common source amplifier has one dominant pole
• The current driven common gate/common source amplifier has two dominant poles
• The cascode amplifier eliminates the input dominant pole for the current driven
common gate/common source amplifier
• Current amplifiers have a low input resistance, high output resistance, and a defined
output-input current relationship
• Input resistances less than 1/gm require feedback
However, all feedback loops have internal poles that cause the benefits of negative
feedback to vanish at high frequencies.
In addition, feedback loops can have a slow time constant from a pole-zero pair.
• Voltage amplifiers using a current amplifier have high values of gain-bandwidth
• Current amplifiers are useful at low power supplies and for switched current
applications
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 21 – Output Amplifiers (6/24/14) Page 21-1
INTRODUCTION
General Considerations of Output Amplifiers
VDD
Requirements:
1.) Provide sufficient output power in the form of f1(vIN) i1
iOUT
vIN
voltage or current.
+
2.) Avoid signal distortion. f2(vIN) i2 RL vOUT
Buffer -
3.) Be efficient Class A VSS
4.) Provide protection from abnormal conditions i1
Current
(short circuit, over temperature, etc.)
Types of Output Amplifiers: t
i2=IQ iOUT
1.) Class A amplifiers Class AB
i1
2.) Source followers iOUT
Current
t
3.) Push-pull amplifiers
i2
4.) Substrate BJT amplifiers Class B
i1
5.) Amplifiers using negative iOUT
Current
t
shunt feedback i2
Fig. 5.5-005
iOUT
Output Imax due to CL
t
Amplifier +
CL RL vOUT
- Imax due to RL 070422-01
Result:
|iOUT| > CL·SR
vOUT(peak)
|iOUT| >
RL
Fortunately, the maximum current for the resistor and capacitor do not occur at the same
time.
Volts
Output
vIN Amplifier R
out
vOA t
+
RL vOUT
-
070422-02
CLASS A AMPLIFIERS
Current source load inverter
VDD i
A Class A circuit has VDD+|VSS| D
RL
current flow in the MOSFETs M2
during the entire period of a VGG2 IQ iOUT RL dominates
vOUT
sinusoidal signal. IQ as the load line
iD1
Characteristics of Class A
amplifiers: vIN M1CL RL vOUT
IQRL IQRL
VSS VDD
• Unsymmetrical sinking and
VSS Fig. 5.5-1
sourcing
• Linear
• Poor efficiency
vOUT(peak)2 vOUT(peak)2
PRL 2RL 2RL vOUT(peak)2
Efficiency = P = (V -V )I = = V
Supply DD SS Q (V -V
DD SS ) -V
DD SS
(VDD -VSS) 2R
L
Maximum efficiency occurs when vOUT(peak) = VDD = |VSS| which gives 25%.
vgs1
+ -
+ C1 +
vin rds1 rds2 RL C2 vout
- gm1vin gm1vout gmbs1vout -
Fig. 040-04
Vout gm1 gm1 gm1RL
Vin = gds1 + gds2 + gm1 + gmbs1+GL gm1 + gmbs1+GL 1 +gm1RL
If VDD = -VSS = 2.5V, Vout = 0V, W1/L1 = 10m/1 m, W2/L2 = 1m/1 m,
and ID = 500 A, then:
For the current sink load follower (RL = ):
Vout Vout
= 0.869V/V, if the bulk effect were ignored, then = 0.963V/V
Vin Vin
For a finite load, RL = 1000
Vout
Vin = 0.512V/V
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 21 – Output Amplifiers (6/24/14) Page 21-14
PUSH-PULL AMPLIFIERS
Push-Pull Source Follower VDD VDD
VDD
Can both sink and source M1
M6
VGG
current and provide a slightly M5 M1 VSS
VBias VSS
lower output resistance. VSS iOUT
vIN iOUT vOUT
VBias vOUT
RL VDD M4 M2 VDD RL
M2 VDD
Efficiency: vIN M3
Depends on how the transistors VSS
VSS VSS Fig. 060-01
are biased.
• Class B - one transistor has current flow for only 180° of the sinusoid (half period)
vOUT(peak)2
PRL 2RL vOUT(peak)
Efficiency = P = 1 2v
=2 V
VDD OUT (peak) DD -VSS
(VDD -VSS)2
R L
Maximum efficiency occurs when vOUT(peak) =VDD and is 78.5%
• Class AB - each transistor has current flow for more than 180° of the sinusoid.
Maximum efficiency is between 25% and 78.5%
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 21 – Output Amplifiers (6/24/14) Page 21-16
0V 0mA 0V 0mA
vout vG2 vout vG2
-1V -1V
iD2 iD2
-2V -1mA -2V -1mA
-2 -1 0 1 2 -2 -1 0 1 2
Vin(V) Vin(V)
Class B, push-pull, source follower Class AB, push-pull, source follower Fig. 060-02
Comments:
• Note that vOUT cannot reach the extreme values of VDD and VSS
• IOUT+(max) and IOUT-(max) is always less than VDD/RL or VSS/RL
• For vOUT = 0V, there is quiescent current flowing in M1 and M2 for Class AB
• Note that there is significant distortion at vIN =0V for the Class B push-pull follower
vgs1
+ -
+ C1 +
vin 1 RL C2 vout
g
- gm1vin gm1vout gmbs1vout rds1 gm2vin gm2vout m2gmbs2vout rds2 -
Fig. 060-03
vout gm1 + gm2
vin = gds1+gds2+gm1+gmbs1+gm2+gmbs2+GL
1
Rout = g +g +g +g (does not include RL)
ds1 ds2 m1 mbs1+gm2+gmbs2
If VDD = -VSS = 2.5V, Vout = 0V, ID1 = ID2 = 500µA, and W/L = 20µm/2µm, Av = 0.787
(RL=) and Rout = 448.
A zero and pole are located at
-(gm1+gm2) -(gds1+gds2+gm1+gmbs1+gm2+gmbs2+GL)
z= and p=
C1 C1+C2
These roots will be at high frequencies because the associated resistances are small.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 21 – Output Amplifiers (6/24/14) Page 21-18
M2
VTR2 iOUT
vIN vOUT
VTR1
M1CL RL
Comments:
• Note that there is significant distortion at vIN =0V for the Class B inverter
• Note that vOUT cannot reach the extreme values of VDD and VSS
• IOUT+(max) and IOUT-(max) is always less than VDD/RL or VSS/RL
• For vOUT = 0V, there is quiescent current flowing in M1 and M2 for Class AB
VDD
M5 M6
M1 M3 VGG3
iOUT
vIN vOUT
M2 M4 VGG4
CL RL
M7 M8
VGG3 and VGG4 can be used to bias this amplifier in class AB or class B operation.
Note, that the bias current in M6 and M8 is not dependent upon VDD or VSS (assuming
VGG3 and VGG4 are not dependent on VDD and VSS).
VDD
VDD -VT+VSat
VDD
IBias
VDD -VT+2VSat
VT+2VSat vIN
vIN
050423-10
050423-08
M3
Q1 M2
iB vout vout
iB
Comments: M2 Q1
CL M3 CL
• Can use either substrate
VSS VSS VSS
or lateral BJTs. p-well CMOS n-well CMOS Fig. 5.5-8A
• Small-signal output resistance is 1/gm which can easily be less than 100.
• Unfortunately, only PNP or NPN BJTs are available but not both on a standard CMOS
technology.
• BJTs will cause substrate current unless they surrounded by a deep well
• In order for the BJT to sink (or source) large currents, the base current, iB, must be
large. Providing large currents as the voltage gets to extreme values is difficult for
MOSFET circuits to accomplish.
• If one considers the MOSFET driver, the emitter can only pull to within vBE+VON of
the power supply rails. This value can be 1V or more.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 21 – Output Amplifiers (6/24/14) Page 21-24
Consequently, the driver for the BJT should be a MOS follower as shown:
VDD VDD
r1 + 1/gm3 1 1 1
Rout = =g + ≈g M3
1+F m1 gm3(1+F) m1 vIN
Q1
Rout
VBN1
M2
M4
070423-03
M1 M5 M6 M1
070423-01
rds1||rds2 1
Rout = 1+Loop Gain ≈ ≈ 10 if gm = 500µS and gmrds ≈ 100.
2gm2rds
The actual value of Rout will be influenced by the value of RL, particularly if it is small.
Push-Pull Implementation
rds1||rds2
Rout = 1+Loop Gain
Comments:
• Can achieve output resistances as low as 10.
• If the error amplifiers are not balanced, it is difficult to control the quiescent current in
M1 and M2
• Great linearity because of the strong feedback
• Can be efficient if operated in class B or class AB
• We will consider this circuit in more detail in a later lecture.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 21 – Output Amplifiers (6/24/14) Page 21-27
vIN
M1
vOUT
VBN1 Rout
M2
070423-04
1
Rout =
gm1K
SUMMARY
• The objectives are to provide output power in form of voltage and/or current.
• In addition, the output amplifier should be linear and be efficient.
• Low output resistance is required to provide power efficiently to a small load resistance.
• High source/sink currents are required to provide sufficient output voltage rate due to
large load capacitances.
• Types of output amplifiers considered:
Class A amplifier
Source follower
Class B and AB amplifier
Use of BJTs
Negative shunt feedback
OP AMPS
What is an Op Amp?
The op amp (operational amplifier) is a high gain, dc coupled amplifier designed to
be used with negative feedback to precisely define a closed loop transfer function.
The basic requirements for an op amp:
• Sufficiently large gain (the accuracy of the signal processing determines this)
• Differential inputs
• Frequency characteristics that permit stable operation when negative feedback is
applied
Other requirements:
• High input impedance
• Low output impedance
• High speed/frequency
Why Op Amps?
The op amp is designed to be used with single-loop, negative feedback to accomplish
precision signal processing as illustrated below.
Single-Loop Negative Feedback Network Op Amp Implementation of a Single-Loop
Negative Feedback Network
Feedback Network
Vf(s) Vf(s)
F(s) F(s)
- Vout(s) Vout(s)
Vin(s) + -
S A(s) Vin(s)
+
Av(s)
Op Amp 060625-01
Vout(s)
The voltage gain, V (s) , can be shown to be equal to,
in
Vout(s) Av(s)
Vin(s) = 1+Av(s)F(s)
If the product of Av(s)F(s) is much greater than 1, then the voltage gain becomes,
Vout(s) 1
The precision of the voltage gain is defined by F(s).
Vin(s) ≈ F(s)
OP AMP CHARACTERIZATION
Linear and Static Characterization of the CMOS Op Amp
A model for a nonideal op amp that includes some of the linear, static nonidealities:
where
Rid = differential input resistance
Cid = differential input capacitance
Ricm = common mode input resistance
Ricm = common mode input capacitance
VOS = input-offset voltage
CMRR = common-mode rejection ratio (when v1=v2 an output results)
en2 = voltage-noise spectral density (mean-square volts/Hertz)
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 22 – Compensation of Op Amps (6/24/14) Page 22-5
OP AMP CATEGORIZATION
Classification of CMOS Op Amps
Conversion Hierarchy
Current
Voltage Transconductance Transconductance Stage
to Current Grounded Gate Grounded Source
Second
Voltage
Current Class A (Source Class B Stage
to Voltage or Sink Load) (Push-Pull)
Table 110-01
COMPENSATION OF OP AMPS
Compensation
Objective
Objective of compensation is to achieve stable operation when negative feedback is
applied around the op amp.
Types of Compensation
1. Miller - Use of a capacitor feeding back around a high-gain, inverting stage.
• Miller capacitor only
• Miller capacitor with an unity-gain buffer to block the forward path through the
compensation capacitor. Can eliminate the RHP zero.
• Miller with a nulling resistor. Similar to Miller but with an added series resistance
to gain control over the RHP zero.
2. Self compensating - Load capacitor compensates the op amp (later).
3. Feedforward - Bypassing a positive gain amplifier resulting in phase lead. Gain can
be less than unity.
Because compensation plays such a strong role in design, it is considered before design.
-20dB/decade
A measure of stability is given by the phase when |A(j)F(j)| = 1. This phase is called
phase margin.
Phase margin = M = Arg[-A(j0dB)F(j0dB)] = Arg[L(j0dB)]
A “good” step response is one that quickly reaches its final value.
Therefore, we see that phase margin should be at least 45° and preferably 60° or larger.
(A rule of thumb for satisfactory stability is that there should be less than three rings.)
Note that good stability is not necessarily the quickest rise time.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 22 – Compensation of Op Amps (6/24/14) Page 22-14
M3 M4 Q3 Q4
M6 Q6
vout vout
- M1 M2 - Q1 Q2
vin vin
+ +
+ M7 + Q7
VBias M5 VBias Q5
- -
VSS VEE Fig. 120-04
Small-Signal Model:
Note that this model neglects the base-collector and gate-drain capacitances for purposes
of simplification.
|A(jw)|
GB
0dB log10(w)
Phase Shift -40dB/decade
-45/decade
180°
Arg[-A(jw)]
135°
-45/decade
90°
45°
0° log10(w)
|p1'| |p2'| w0dB 150128-02
If we assume that F(s) = 1 (this is the worst case for stability considerations), then the
above plot is the same as the loop gain.
Note that the phase margin is much less than 45° (≈ 6°).
Therefore, the op amp must be compensated before using it in a closed-loop
configuration.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 22 – Compensation of Op Amps (6/24/14) Page 22-17
MILLER COMPENSATION
Miller Compensation of the Two-Stage Op Amp
VDD VCC
M3 M4
Q3 Q4
CM M6 CM Q6
Cc vout Cc vout
M1 M2 Q1 Q2
- -
vin CI CII vin CI CII
+ +
+ M7 + Q7
VBias M5 VBias Q5
- -
VSS VEE Fig. 120-08
Cc
v2
+ +
vin gm1vin CI CII vout
rds2||rds4 gm6v2 rds6||rds7
- -
Fig. 120-09
Same circuit holds for the BJT op amp with different component relationships.
-1 -1 gmII
p1 = R (C +C )+R (C +C )+g R R C ≈ g R R C , z= C
I I II II II c mII 1 II c mII 1 II c c
Avd(0) dB Uncompensated
|A(jw)F(jw)| -20dB/decade
F(jw)=1
Compensated
GB
0dB log10(w)
Phase Shift -40dB/decade
Uncompensated
180°
Arg[-A(jw)F(jw)|
-45°/decade
135°
F(jw)=1
90° -45°/decade
Compensated Phase
45°
No phase margin Margin
0° log10(w)
|p1| |p1'| |p2'| |p2|
150128-04
1
|p1| ≈ R (g R C )
I m6 II c
gm6
|p2| ≈ C
II
3.) Right-half plane zero (One source of zeros is from multiple paths from the input to
output): VDD
gm6 RII
-RIIsC - 1 Cc
-gm6RII(1/sCc) RII c vout
vout = R + 1/sC v’ + R + 1/sC v’’ = R + 1/sC v
II c II c II c v''
M6
v'
where v = v’ = v’’. Fig. 120-15
Further Comments on p2
The previous observations on p2 can be proved as follows:
Find the resistance RCc seen by the compensation capacitor, Cc.
Cc VDD
vx
RCc RII
RCc
ix ix
M6 +
RI RI vgs6 RII
- gm6vgs6
060626-02
Thus, at the frequency where CII begins to short the output, Cc is acting as a short.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 22 – Compensation of Op Amps (6/24/14) Page 22-24
Cc
A
+ +
Vi gmIIVi CII RII Vout
s + gmII/ACc
- -
Vout(s) ACc Fig.430-09
Self-Compensated Op Amps
Self compensation occurs when the load capacitor is the compensation capacitor (can
never be unstable for resistive feedback)
Voltage gain:
vout
vin = Av(0) = GmRout
Dominant pole: Stability:
-1 Large load capacitors simply reduce
p1 = R GB but the phase is still 90° at GB.
outCL
Unity-gainbandwidth:
Gm
GB = Av(0)·|p1| = C
L
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 22 – Compensation of Op Amps (6/24/14) Page 22-29
C2 C2
-A -A
R1 C1 R1 C1 C3(1+A)
RootID01
C2 C2
+A +A
R1 C1 R1 C1 C3(1-A)
RootID02
p1 +1 Cc
RII
vout
2.) Zeros are also created by two paths from the input to the M6
output and one or more of the paths is frequency dependent. v''
v'
3.) Zeros also come from simple RC networks. 070425-01
C1
+ + Vout s + 1/(R1C1)
Vin R1 R2 Vout =
Vin s + 1/(R1||R2)C1
- -
070425-02
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 22 – Compensation of Op Amps (6/24/14) Page 22-31
I5 I6-I5-I7 I5 I5 I7-I5 I5
SR = minC , C = C because I6>>I5
+ SR = minC , C = C if I7>>I5.
-
c L c c L c
Therefore, if CL is not too large and if I7 is significantly greater than I5, then the slew
rate of the two-stage op amp should be, I5/Cc.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 22 – Compensation of Op Amps (6/24/14) Page 22-32
SUMMARY
• Op amps achieve accuracy by using negative feedback
• Compensation is required to insure that the feedback loop is stable
• The degree of stability is measured by phase margin and is necessary to achieve small
settling times
• A compensated op amp will have one dominant pole and all other poles will be greater
than GB
• A two-stage op amp requires some form of Miller compensation
• A high output resistance op amp is compensated by the load capacitor
• Poles of a CMOS circuit are generally equal to the negative reciprocal of the product of
the resistance to ground from a node times the sum of the capacitances connected to
that node.
• The slew rate of the two-stage op amp is equal to the input differential stage current
sink/source divided by the Miller capacitor
or systems M3 M4 Cc
M6
Topology
specifications
vout
-
vin
M1 M2 CL L
+
+ M7
VBias
-
M5 W
VSS
DC Currents
Design of 50µA
CMOS
Op Amps
W/L ratios
Component C R
values
060625-06
M6
M3 M4 Cc
vout
- M1 M2 CL
vin
+
+ M7
VBias M5
-
VSS Fig. 6.3-1
Notation:
Wi
Si = L = W/L of the ith transistor
i
Op Amp Specifications
The following design procedure assumes that specifications for the following parameters
are given.
1. Gain at dc, Av(0)
2. Gain-bandwidth, GB
3. Phase margin (or settling time)
4. Input common-mode range, ICMR
5. Load Capacitance, CL
6. Slew-rate, SR
7. Output voltage swing
8. Power dissipation, Pdiss
Inverting vOUT
High-Gain
Stage
120523-01
† W.J. Parrish, “An Ion Implanted CMOS Amplifier for High Performance Active Filters”, Ph.D. Dissertation, 1976, Univ. of CA, Santa Barbara.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 23 – Design of Two-Stage Op Amps (3/11/16) Page 23-19
A Design Procedure that Allows the RHP Zero to Cancel the Output Pole, p2
We desire that z1 = p2 in terms of the previous notation.
Therefore,
1 -gmII
Cc(1/gmII - Rz) = CII
The value of Rz can be found as
Cc + CII
Rz = C (1/gmII)
c
With p2 canceled, the remaining roots are p1 and p4(the pole due to Rz) . For unity-gain
stability, all that is required is that
Av(0) gmI
p4 Av(0)p1 = g R R C = C and (1/RzCI) (gmI/Cc) = GB
mII II I c c
Substituting Rz into the above inequality and assuming CII >> Cc results in
gmI
Cc gmII CICII
This procedure gives excellent stability for a fixed value of CII ( CL).
Unfortunately, as CL changes, p2 changes and the zero must be readjusted to cancel p2.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 23 – Design of Two-Stage Op Amps (3/11/16) Page 23-20
M9 M5
M12 M7
For the zero to be on top of the second pole (p2), the following relationship must hold
1 CL + Cc Cc+CL 1
Rz = g C = C
m6 c c 2K’PS6I6
The resistor, Rz, is realized by the transistor M8 which is operating in the active region
because the dc current through it is zero. Therefore, Rz, can be written as
vDS8 1
Rz = =
iD8 V =0 K’PS8(VSG8-|VTP|)
DS8
The bias circuit is designed so that voltage VA is equal to VB.
W11 I10 W6
VGS10 − VT = VGS8 − VT VSG11 = VSG6 =
L11 I6 L6
In the saturation region
2(I10)
VGS10 − VT =
K'P(W10/L10) = VGS8 − VT
1 K’PS10 1 S10
Rz =
K’PS8 2I10 = S8 2K’PI10
W8 Cc S10S6I6
Equating the two expressions for Rz gives =
L8 CL + Cc I10
3pF 1·190·95µA 100327-03
VSS
(W/L)8 = 3pF+10pF
=8
15µA
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 23 – Design of Two-Stage Op Amps (3/11/16) Page 23-23
M10 M9 M10 M9
The roots become, VSS VSS
120523-03
1.) The dominant pole increased slightly because RI (output of first stage) is decreased.
VDD VDD
2.) The output pole is increased by a rds7 rds7
Cc
-Agm6 vout A vout
factor of A to get new p2 ≈ C 1
II M8 GB·Cc » 0 M6
CII
M6 CII
3.) The pole at the source of M8 (-gm8/Cc)
becomes a zero on the negative real axis. 120523-04
Roots: jw
s
-Agm6 -gm8 -1 gm6
C2 Cc gm6rds2Cc Cgd6 120523-05
† B.K. Ahuja, “An Improved Frequency Compensation Technique for CMOS Operational Amplifiers,” IEEE J. of Solid-State Circuits, Vol. SC-18,
No. 6 (Dec. 1983) pp. 629-633.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 23 – Design of Two-Stage Op Amps (3/11/16) Page 23-26
The solution proposed in the reference below is to decrease the impedance at the source
of M8 by using a negative feedback loop. Below is a possible solution that will have
better phase margin. VDD
M11 M12 M16 M7
M13
vOUT
M8 Cc
M6
160311-02
VSS
† Uday Dasgupta, “Issues with ‘Ahuja’ Frequency Compensation Technique,” Proc. of IEEE Inter. Symposium on Radio Frequency Integration
Technology, Jan. 9, 2009, pp. 326-329.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 23 – Design of Two-Stage Op Amps (3/11/16) Page 23-27
M5 rds7
Vss Path through Cgd7
VBias M7
is negligible
VBias connected to VSS VSS
Fig. 180-11
What is Zout?
Vt gmIVt
Zout = I It = gmIIV1 = gmIIG +sC +sC Cc CII+Cgd7 It
t I I c
+ rds6||rds7 +
GI+s(CI+Cc) CI RI V1 gmIIV1 Vout Vt
Thus, Zout = g g gmIVin
- -
mI MII
150131-01
rds7
1+ Z
Vout out s(Cc+CI) + GI+gmIgmIIrds7 -GI
= = Pole at
Vss 1 s(Cc+CI) + GI Cc+CI
The negative PSRR is much better than the positive PSRR.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 23 – Design of Two-Stage Op Amps (3/11/16) Page 23-30
SUMMARY
• The output of the design of an op amp is
- Schematic
- DC currents
- W/L ratios
- Component values
• Design procedures provide an organized approach to creating the dc currents, W/L
ratios, and the component values
• The right-half plane zero causes the Miller compensation to deteriorate
• Methods for eliminating the influence of the RHP zero are:
- Nulling resistor
- Increasing the magnitude of the output pole
• The PSRR of the two-stage op amp is poor because of the Miller capacitance, however,
methods exist to eliminate this problem
• The two-stage op amp is a very general and flexible op amp
Cascode Op Amps
Why cascode op amps?
• Control of the frequency behavior
• Can get more gain by increasing the output resistance of a stage
• In the past section, PSRR of the two-stage op amp was insufficient for many
applications
• A two-stage op amp can become unstable for large load capacitors (if nulling resistor
is not used)
• The cascode op amp leads to wider ICMR and/or smaller power supply requirements
Where Should the Cascode Technique be Used?
• First stage -
Good noise performance
Requires level translation to second stage
Degrades the Miller compensation
• Second stage -
Self compensating
Increases the efficiency of the Miller compensation
Increases PSRR
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 24 Cascode Op Amps (9/7/17) Page 24-3
MB5
MC1 MC2 MC1 + MC2
M1 M2 M1 VBias M2
VBias MB1 MB2
+v - +v
in vin in - vin -
2- + 2 2- + 2
+ M5 + M5
VNBias1 VNBias1
- -
VSS VSS 060627-01
M7 M8 M7 M8
M15
-A M5 M6
VNB1
M5 M6 M16
vOUT VDD VDD vOUT
VPB1
M3 M4
M13 M14
-A -A M3 M4
M11 M12
M1 M2 M1 M2
+ +
vIN vIN
- -
M9
VNB1 M9 VNB1 M10
060627-02
From inspection, we can write the voltage gain as,
vOUT
Av = v = gm1Rout where Rout = (Ards6gm6rds8)|| (Ards2gm4rds4)
IN
If rdsn ≈ rdsp, then A ≈ gmrds/2 and the voltage gain would be equal to 100K to 500,K.
Output is not optimized for maximum signal swing.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 24 Cascode Op Amps (9/7/17) Page 24-6
120523-06
-
VSS Fig. 6.5-3
Comments:
• The second-stage gain has greatly increased improving the Miller compensation
• The overall gain is approximately (gmrds)3 if rdsn >> rdsp or if rdsp >> rdsn
• Output pole, p2, is approximately the same if Cc is constant
• The zero RHP is the same if Cc is constant
• PSRR is poor unless the Miller compensation is removed (then the op amp becomes
self compensated)
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 24 Cascode Op Amps (9/7/17) Page 24-9
A
A B C D
B Thin
oxide Poly I Poly II
n+ n-channel n+
C p substrate/well
D
Fig. 6.5-5
If a double poly CMOS process is available, inter-node parasitics can be minimized.
As an alternative, one should keep the drain/source between the transistors to a minimum
area.
Minimum Poly
A separation
A B C D
B Thin
oxide Poly I Poly I
n+ n-channel n+ n-channel n+
C p substrate/well
D
Fig. 6.5-5A
Input Common Mode Range for Two Types of Differential Amplifier Loads
VDD-VSD3+VTN
VDD VDD
VDD-VSG3+VTN
+ + + +
VSG3 VSD4 V VSD4
Input SD3
Input
- M3 M4 - Common - M3 M4 -
VBP
Common Mode
Mode Range
Range M1 M2 M1 M2
VSS+VDS5+VGS1 VSS+VDS5+VGS1
+ M5 vicm + M5 vicm
VBias VBias
- -
VSS VSS
Differential amplifier with Differential amplifier with
a current mirror load. current source loads. Fig. 6.5-6
In order to improve the ICMR, it is desirable to use current source (sink) loads without
losing half the gain.
The resulting solution is the folded cascode op amp.
RA VPB2 RB
I1 I2
I6 I7
M6 vOUT
M7
+
M1 M2 VNB2
vIN
- M8 M9 CL
M3
VNB1 I3 M11
M10
Comments: 060628-04
• I4 and I5, should be designed so that I6 and I7 never become zero (i.e. I4=I5=1.5I3)
• This amplifier is nearly balanced (would be exactly if RA was equal to RB)
• Self compensating
• Poor noise performance, the gain occurs at the output so all intermediate transistors
contribute to the noise along with the input transistors. (Some first stage gain can be
achieved if RA and RB are greater than gm1 or gm2.
With the output short-circuited, RA ≈ 1/gm6 and RB ≈ 1/gm7. Therefore the currents i7
and i9 can be written as,
gm2(rds2||rds5)vin gm2vin gm1(rds1||rds4)vin gm1vin
i7 = ≈ and i9 ≈ -i10 = ≈
2[RB + (rds2||rds5)] 2 2[RA + (rds1||rds4)] 2
The output resistance with the short-circuit removed is,
Rout ≈ (gm9rds9rds11)||[ gm7rds7(rds2|| rds5)]
Finally,
gm1vin gm2vin
vout = (i7 + i9)Rout = + 2 Rout = gm1Rout = gm2Rout
2
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 24 Cascode Op Amps (9/7/17) Page 24-14
where the approximate expressions are found by the reciprocal product of the resistance
and parasitic capacitance seen to ground from a given node. One might feel that because
RB is approximately rds that this pole also might be small. However, at frequencies
where this pole has influence, Cout, causes Rout to be much smaller making pB also non-
dominant.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 24 Cascode Op Amps (9/7/17) Page 24-16
Vss
Fig. 6.5-9A
This model assumes that gate, source and drain of M11 and the gate and source of M9 all
vary with VSS.
We shall examine Vout/Vss rather than PSRR-. (Small Vout/Vss will lead to large PSRR-.)
The transfer function of Vout/Vss can be found as
Vout sCgd9Rout
Vss ≈ sCoutRout+1 for Cgd9 < Cout
We see that the PSRR of the cascode op amp is much better than the two-stage op amp
without any modifications to improve the PSRR.
M10 M11
VPB1 M3
-A
+
vIN M8 M9
- vOUT
M1 M2
M6 M7
-A -A
If rdsn >> rdsp or if rdsp >> rdsn, then A ≈ gmrds and the voltage gain would be in the
range of 100,000 to 500,000.
Note that to achieve maximum output swing, it will be necessary to make sure that M5
and M11 are biased with VDS = VDS(sat).
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 24 Cascode Op Amps (9/7/17) Page 24-25
VDD VDD
VPB1 VPB1
vin M4 vin M6 M3
vout
vin
M6
-A VPB2 VNB2 M5
M1 M2
vout -A vin M1 M2
M5 VDD
M9
vout VNB1 -VSD(Sat) vout VNB1
M4 VDS(Sat)
M6 M3
140713-02
SUMMARY
• Cascode op amps give additional flexibility to the two-stage op amp
- Increase the gain
- Control the dominant and nondominant poles
• Enhanced gain, cascode amplifiers provide additional gain and are used when high
gains are needed
• Folded cascode amplifier is an attractive alternate to the two-stage op amp
- Wider ICMR
- Self compensating
- Good PSRR
INTRODUCTION
Simulation and Measurement Considerations
Objectives:
• The objective of simulation is to verify and optimize the design.
• The objective of measurement is to experimentally confirm the specifications.
Similarity between Simulation and Measurement:
• Same goals
• Same approach or technique
Differences between Simulation and Measurement:
• Simulation can idealize a circuit
- All transistor electrical parameters are ideally matched
- Ideal stimuli
• Measurement must consider all nonidealities
- Physical and electrical parameter mismatches
- Nonideal stimuli
- Parasistics
VDD
vIN vOUT
CL RL VSS
C R
Fig. 240-02
Resulting Closed-Loop Frequency Response:
dB Op Amp
Av(0) Open Loop
Frequency
Response
0dB
1 Av(0) log10(w)
RC RC Fig. 240-03
VOS VDD
+ -
vout
+
vcm VSS
- CL RL
Fig. 6.6-5
Make sure that the output voltage of the op amp is in the linear region.
Divide (subtract dB) the result into the open-loop gain to get CMRR.
V1+V2
Vout = Av(V1-V2) ±Acm 2 = -AvVout ± AcmVcm
±Acm ±Acm
Vout = V ≈ V
1+Av cm Av cm
Av Vcm
|CMRR| = =
Acm Vout
Vout Vout
Avd = V = V
id i
Vos ≈ 1000Vi
1000Vout
Therefore, Avd = V
os
Sweep Vout as a function of frequency, invert the result and multiply by 1000 to get
Avd (j).
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 25 – Measurement and Simulation of Op amps (6/25/14) Page 25-10
Measurement of CMRR
Measurement Configuration:
Note that the whole amplifier is stimulated by
Vicm while the input responds to this change.
The definition of the common-mode rejection
ratio is
Avd (vout/vid)
CMRR = =
Acm (vout/vicm)
However, in the above circuit the value of vout
is the same so that we get
vicm
CMRR = v
id
vos
But vid = vi and vos 1000vi = 1000vid vid = 1000
vicm 1000 vicm
Substituting in the previous expression gives, CMRR = v =
os vos
1000
Make a frequency sweep of Vicm, invert the result and multiply by 1000 to get CMRR.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 25 – Measurement and Simulation of Op amps (6/25/14) Page 25-11
Measurement of PSRR
100kW
Measurement Configuration: Vos
IDD 1
VDD
vOUT 1
+ vIN
vIN ICMR
- VSS
C RL
ISS L
Also, monitor
IDD or ISS. Fig.240-11
If the slew rate influences the small signal response, then make the input step size small
enough to avoid slew rate (i.e. less than 0.5V for MOS).
20
Overshoot (%)
Peak Overshoot 50
5
40 Phase Margin Overshoot
30
t 1.0
150303-01 20
For example, a 5% overshoot
10
corresponds to a phase margin
of approximately 64°. 0 0.1
0 0.2 0.4 0.6 0.8 1
z= 1 070429-03
2Q
.SUBCKT OPAMP 1 2 6 8 9
M1 4 2 3 3 NMOS1 W=3U L=1U AD=18P AS=18P PD=18U PS=18U
M2 5 1 3 3 NMOS1 W=3U L=1U AD=18P AS=18P PD=18U PS=18U
M3 4 4 8 8 PMOS1 W=15U L=1U AD=90P AS=90P PD=42U PS=42U
M4 5 4 8 8 PMOS1 W=15U L=1U AD=90P AS=90P PD=42U PS=42U
M5 3 7 9 9 NMOS1 W=4.5U L=1U AD=27P AS=27P PD=21U PS=21U
M6 6 5 8 8 PMOS1 W=94U L=1U AD=564P AS=564P PD=200U PS=200U
M7 6 7 9 9 NMOS1 W=14U L=1U AD=84P AS=84P PD=40U PS=40U
M8 7 7 9 9 NMOS1 W=4.5U L=1U AD=27P AS=27P PD=21U PS=21U
CC 5 6 3.0P
.MODEL NMOS1 NMOS VTO=0.70 KP=110U GAMMA=0.4 LAMBDA=0.04 PHI=0.7
+MJ=0.5 MJSW=0.38 CGBO=700P CGSO=220P CGDO=220P CJ=770U CJSW=380P
+LD=0.016U TOX=14N
.MODEL PMOS1 PMOS VTO=-0.7 KP=50U GAMMA=0..57 LAMBDA=0.05 PHI=0.8
+MJ=0.5 MJSW=.35 CGBO=700P CGSO=220P CGDO=220P CJ=560U CJSW=350P +LD=0.014U TOX=14N
IBIAS 8 7 30U
.ENDS
2.5
2
VOS
1
vOUT(V)
-1
-2
-2.5
-2 -1.5 -1.0 -0.5 0 0.5 1 1.5 2
vIN(mV) Fig. 240-18
80 200
150
60
40
50
20 0
-50
0
-100
-20
-150 Phase Margin
GB GB
-40 -200
4 5 6 7 8
10 100 1000 10 10 10 10 10 10 100 1000 104 105 106 107 108
Frequency (Hz) Frequency (Hz) Fig. 6.6-16
ID(M5) mA
. 3 30
(Subcircuit of Table 6.6-1)
.. 2 20
. Input CMR
vOUT (V)
.DC VIN+ -2.5 2.5 0.1 1 10
.PRINT DC V(3)
.TRAN 0.05U 10U 0 10N 0 0
.PRINT TRAN V(3) V(1)
.AC DEC 10 1 10MEG -1
.PRINT AC VDB(3) VP(3)
.PROBE (This entry is unique to PSPICE) -2
.END -3
-3 -2 -1 0 1 2 3
vIN(V) Fig. 240-21
Note the usefulness of monitoring the
current in the input stage to determine the lower limit of the ICMR.
This PSRR+ is poor because of the Miller capacitor. The degree of PSRR+ deterioration
will be better shown when compared with the PSRR-.
1 0.1 vin(t)
0.5 0.05
vout(t)
vout(t)
Volts
Volts
0 0
-0.5 -0.05
vin(t)
-1 -0.1
-1.5 -0.15
2.5 3.0 3.5 4.0 4.5
0 1 2 3 4 5
Time (Microseconds) Time (Microseconds) Fig. 240-24
SUMMARY
• Simulation and measurement of op amps has both similarities and differences
• Measurement of open loop gain is very challenging – the key is to keep the quiescent
point output of the op amp well defined
• The method of stimulating the output of the op amp or power supplies and letting the
input respond results in a robust method of measuring open loop gain, CMRR, and
PSRR
• Carefully investigate any deviations or aberrations from expected behavior in the
simulation and experimental results
• Be alert for when the small-signal model calculations are influenced by the large signal
performance
INTRODUCTION
Buffered Op Amps
What is a buffered op amp?
Buffered op amps are op amps with the ability to drive a low output resistance and/or
a large output capacitance. This requires:
- An output resistance typically in the range of 10 Ro 1000
- Ability to sink and source sufficient current (CL·SR)
Rout Rout
+ Large Small
vIN vOUT
- vOUT’
Op Amp Buffer 070430-01
2I1 2I2
= Kn'(W1/L1) + Kp'(W2/L2)
Use the W/L ratios to define I1 and I2 from I5 and I6
• Maximum positive and negative output voltages are limited
M6
M3 M4 Cc M8
vout
- M1 M2
vin
+ CL
M5 I5 M7 I7 M9 I9
VNBias
060706-03
1
Rout ≈ 1000, Av(0)=65dB (IBias=50µA), and GB = 60MHz for CL = 1pF
gm21+gm22
Note the bias currents through M18 and M19 vary with the signal.
• This buffer trades gain for the ability to drive a low load resistance
• The load resistance should be fixed in order to avoid changes in the buffer gain
• The push-pull common source output will give good output voltage swing capability
VDD vout
M6
M7 M3 M4 M6 Active
IBias M1-M3 M2-M4
Inverter M6 Satur- Inverter
C1 C2 vout
M5 Saturated ated
M1 M2 M5 RL M5 Active
vin'
0 vin'
VSS VA VB VDD
VSS 060706-05
M1
vin vout vin vout
M8 M3 M4 M9
M2
M6
Ib I=2Ib M10
If W4/L4 = W9/L9 and W3/L3 = VSS
W8/L8, then the quiescent VSS 070430-07
- Ro ROUT
vIN + vOUT
Av
RL
070430-02
• Output resistance
Ro
ROUT = 1 + A
v
• Watch out for when small RL causes Av to decrease.
• The bandwidth will be limited by the feedback (i.e. at high frequencies, the gain of Av
decreases causing the output resistance to increase.
070430-03
1
Rout ≈ g
m8K
Power dissipation now becomes (I5 + I7 + I11)VDD
Gain becomes,
gm1 gm6 gm8K
Av = g +g g +g g K+ g
ds2 ds4 ds6 ds7 m8 mbs8 K +G L
1 1000
Rout ≈ ≈ ≈ 100
K(gm21+gm22) K
Av(0)=65dB (IBias=50µA)
Note the bias currents through M18 and M19 vary with the signal.
Comments:
• The output resistance will be equal to rds1||rds2 divided by the loop gain
• If the error amplifiers are not perfectly matched, the bias current in M1 and M2 is not
defined
The feedback circuitry of the two error amplifiers tries to insure that the voltages in
the loop sum to zero. Without the M9-M12 feedback circuit, there is no way to adjust the
output for any error in the loop. The circuit works as follows:
When VOS is positive, M6 tries to turn off and so does M6A. IM9 reduces thus reducing
IM12. A reduction in IM12 reduces IM8A thus decreasing VGS8A. VGS8A ideally decreases
by an amount equal to VOS. A similar result holds for negative offsets and offsets in EA2.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 26 – Buffered Op Amps (6/25/14) Page 26-16
VDD
+
VBias
M6 M5A -
M3 M4 Cc1
vin
vin MR1 M2A M1A
M1 M2 vout MR2
Cc2
+ M6A M4A M3A
VBias M5
-
A1 amplifier VSS A2 amplifier
070430-05
Basically a two-stage op amp with the output push-pull transistors as the second-stage of
the op amp.
RC CC
Short circuit protection(max. output 60mA):
MP3-MN3-MN4-MP4-MP5 gm1 gm6
MN3A-MP3A-MP4A-MN4A-MN5A
C1 CL
rds6||rds6A 50k R1 RL
Rout Loop Gain ≈ 5000 = 10
M8 M3 M4 M6
200mA 10/1 1/1 1/1 10/1
vout
Output Resistance:
M1 M2 CL Ro
+ Rout =
vin 10/1 10/1 1+LG
-
1/1 where
M5 1
M10 1/1 Ro =
10/1 10/1 gds6+gds7
M9 M7 and
gm2
VSS Fig. 7.1-9 |LG| = 2g (gm6+gm7)Ro
m4
Therefore, the output resistance is:
1
Rout =
gm2
(gds6+gds7) 1 + (gm6+gm7)Ro
2g
m4
Example 26-1 - Low Output Resistance Using Shunt Negative Feedback Buffer
Find the output resistance of above op amp using the model parameters of KN’ =
120µA/V2, KP’ = 25µA/V2, N = 0.06V-1 and P = 0.08V-1.
Solution
The current flowing in the output transistors, M6 and M7, is 1mA which gives Ro of
1 1000
Ro = = 0.14 = 7.143kΩ
(N+P)1mA
To calculate the loop gain, we find that
gm2 = 2KN’·10·100µA = 490µS
gm4 = 2KP’·1·100µA = 70.7µS
and
gm6 = 2KP’·10·1000µA = 707µS
Therefore, the loop gain is
490
|LG| = 2·70.7 (0.707+0.071)7.143 = 19.26
Solving for the output resistance, Rout, gives
7.143k
Rout = 1 + 19.26 = 353 (Assumes that RL is large)
n+ (Emitter) p+ n+
Base
p- well (Base)
n- substrate (Collector)
Fig. 7.1-10
Emitter
Comments:
• gm of the BJT is larger than the FET so that the output resistance w/o feedback is lower
• Collector current will be flowing in the substrate
• Current is required to drive the BJT
• Only an NPN or a PNP bipolar transistor is available
Base
Base
r10 + (1/gm9) 1 1
Small-signal output resistance : Rout ≈ = +
1+ßF gm10 gm9(1+ßF)
= 51.6+6.7 = 58.3 where I10=500µA, I8=100µA, W9/L9=100 and ßF is 100
2KP’ Ic10
vOUT(max) = VDD - VSD8(sat) - vBE10 = VDD -
I8(W8/L8) - Vt lnIs10
Voltage gain:
vout gm1 gm6 gm9 gm10RL
vin ≈ gds2+gds4gds6+gds7gm9+gmbs9+gds8+g101+gm10RL
Compensation will be more complex because of the additional stages.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 26 – Buffered Op Amps (6/25/14) Page 26-24
requirements. 070430-06
SUMMARY
• A buffered op amp requires an output resistance between 10 Ro 1000
• Output resistance using MOSFETs only can be reduced by,
- Source follower output (1/gm)
- Negative shunt feedback (frequency is a problem in this approach)
• Use of substrate (or lateral) BJT’s can reduce the output resistance because gm is larger
than the gm of a MOSFET
• Adding a buffer stage to lower the output resistance will most likely complicate the
compensation of the op amp
-gm6
2.) Output pole p2 = C
L
-gm3
3.) Mirror pole p3 = C +C
gs3 gs4
and z3 = 2p3
-1
4.) Nulling pole p4 =
RzCI
-1
5.) Nulling zero z1 =
RzCc-(Cc/gm6)
Higher-Order Poles
For reasonable phase margin, the smallest higher-order pole should be 2-3 times larger
than GB if all other higher-order poles are larger than 10GB.
Av(0) dB
Larger non- Smallest non- Dominant
dominant poles dominant pole pole
-10GB -GB
10GB
0dB
GB GB log10w
Av(0)
060709-01
If the higher-order poles are not greater than 10GB, then the distance from GB to the
smallest non-dominant pole should be increased for reasonable phase margin.
When p2 is cancelled, the next smaller pole is p4 which will define the new GB. 2.)
Using the nulling zero, z1, to cancel p2, gives p4 as the next smallest pole.
For 60° phase margin GB = |p4|/2.2 if the next smallest pole is more than 10GB.
GB = 0.680x109/2.2 = 0.309x109 rads/sec. or 49.2MHz.
This value of GB is designed from the relationship that GB = gm1/Cc. Assuming gm1 is
constant, then Cc = gm1/GB = (94.25x10-6)/(0.309x109) = 307fF. It might be useful to
increase gm1 in order to keep Cc above the surrounding parasitic capacitors (Cgd6 =
18.7fF). The success of this method assumes that there are no other roots with a
magnitude smaller than 10GB.
The result of this example is to increase the GB from 5MHz to 49MHz.
-1
pB ≈ R C (the pole at the source of M7)
B B
-gm10 M6
p6 ≈ C (the pole at the drain of M6)
6 IT = gm8VT rds8 gm10
IT M8
-gm8rds8gm10 VT 1 M10
p8 ≈ (the pole at the source of M8) R8 = = +
C8 IT gm8rds8 gm10 VT
- 150708-01
-gm9
p9 ≈ C (the pole at the source of M9)
9
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 27 – High Speed Op Amps (6/25/14) Page 27-10
4.) Next, we consider the pole, p8. The capacitance connected to this node is M10
C8= Cbd10 + Cgd10 + Cgs8 + Cbs8 150504-02
The smallest of these poles is pA or pB. Since p6 is not much larger than pA or pB, we
will find the new GB by dividing pA or pB by 4 (which is a guess rather than 2.2) to get
364x106 rads/sec. Thus the new GB will be 364/2 or 58MHz.
Checking our guess gives a phase margin of,
PM = 90° - 2tan-1(0.364/1.456) - tan-1(0.364/2.62) = 54° which is okay
The magnitude of the dominant pole is given as
pdominant = GB/Avd(0) = 364x106/3,678 = 99,000 rads/sec.
The value of load capacitor that will give this pole is
CL = (pdominant·Rout)-1 = (99x103·7.44M)-1 = 1.36pF
Therefore, the new GB = 58MHz compared with the old GB = 10MHz.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 27 – High Speed Op Amps (6/25/14) Page 27-14
vin + VPB1
M4 Non-
dominant
VPB2 M3 Pole
vout
Dominant Pole
VNB2 M2 CL
Non-
M1 dominant
vin + VNB1 Pole
060710-01
If the source-drain area between M1 and M2 and M3 and M4 can be minimized, the non-
dominant poles will be quite large.
Operation:
VDD VDD
+
VB2 VDD-VB2-(vin+-vin-)
M8 - M4
M7 + + M3
C1 VDD-VB2-vin+ VDD-VB2-vin+ C1
- vin- - vout
IB vin+
+ +
C2 vin+-VSS-VB1 vin+-VSS-VB1 C2
M6 - - M2
M5 + M1
VB1 VSS+VB1-(vin+-vin-)
-
VSS VSS
Equivalent circuit during the f1 clock period Equivalent circuit during the f2 clock period.
120523-07 120523-08
† S. Masuda, et. al., “CMOS Sampled Differential Push-Pull Cascode Op Amp,” Proc. of 1984 International Symposium on Circuits and Systems,
Montreal, Canada, May 1984, pp. 1211-12-14.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 27 – High Speed Op Amps (6/25/14) Page 27-17
Ao n
s/w1+1 060710-02
Voltage Gain:
Vout gm1 Kn'(W1/L1)(I3+I5)
Vin = gm3 = Kp' (W3/L3)I3
gm3
-3dB ≈ C
gs1
Requirements:
io = Ai(i1-i2)
Ri1 = Ri2 = 0
Ro =
Ideal source and load requirements:
Rsource =
RLoad = 0
+
Vin GM
- Ai Vout
060711-04
Vout -GMRFAi
Vin = 1 +Ai
VPB2
Rin
Vin+ Vin-
M1 R Iin M2
F RF
+ Vout -
I nI nI
I
VNB2
nIin nIin
1:n 1:n
150504-03
R1
vin+ vin-
M1 M2
R2 R2
+1 + vout - +1
VBias
x4 x2 x1 VSS x1 x2 x4
=1/8 = 1/4 =1/2 =1/2 = 1/4 =1/8
Fig. 7.2-135A
R1 and the current mirrors are used for gain variation while R2 is fixed.
SUMMARY
• Increasing the GB of an op amp requires that the magnitude of all non-dominant poles
are much greater than GB from the origin of the complex frequency plane
• The practical limit of GB for an op amp is approximately 5-10 times less than the
magnitude of the smallest non-dominant pole (≈ 100MHz)
• To achieve high values of GB it is necessary to eliminate the non-dominant poles
(which come from parasitics) or increase the magnitude of the non-dominant poles
• The best way to achieve high-bandwidth amplifiers is to cascade high-bandwidth
voltage amplifiers
• If the gain of the high-bandwidth voltage amplifiers is well defined, then it is not
necessary to use negative feedback around the amplifier
• Amplifiers with well-defined gains are achievable with a -3dB bandwidth of 100MHz
INTRODUCTION
Why Differential Output Op Amps?
• Cancellation of common mode signals including clock feedthrough
• Increased signal swing
v1
A v1-v2
2A
t
v2 -A t
A
t -2A Fig. 7.3-1
-A
• Cancellation of even-order harmonics
Symbol:
0 t 0 t 0 t
VSS VSS VSS
CM output voltage properly defined, CM output voltage too large, CM output voltage too small,
Vcm = 0 Vcm= 0.5VDD Vcm= 0.5VSS 070506-01
Remember that:
vOUT = Avd(vID) ± Acm(vCM)
vi1 M1 M2 vi2
M5
M9 + M7
VBN
-
VSS Fig. 7.3-3
(OCMR) = VDD+ |VSS| - VSDP(sat) - VDSN(sat)
The maximum peak-to-peak output voltage 2·OCMR
Conversion between differential outputs and single-ended outputs:
M5
M9 M10 + M12 M8
VBN
-
VSS Fig. 7.3-6
Comments:
• Able to actively source and sink output current
• Output quiescent current poorly defined
CL - vOUT +
M1 M2 CL
+ VNB2
vIN
- M8 M9
M3 VNB1
VNB1 I3 M11
M10
060717-01
M7 M3 M4 M6 M8
M5
M21 M20
M9 M10
vi1 M1 M2 vi2 M19
vo1 R2 M22 R1 vo2
M15 M16
M17
VBias M18
M23
M13 M12
M11 M14
M10 M11 M6 M7
VPB2
M12 M13 +A -
+ - +
vIN M1 M2 M3 M4
- M8 M9
M5
VNB1 CL - vOUT + CL
M15 M18
VNB2 - + VNB2
M14 +A - M19
+ M1 M2 +
VGS1 vGS1 vGS2 VGS2
- -
vi1 vi2
+ +
VSG3 vSG3 vSG4 VSG4
- -
M3 M4
i2 i1
Fig. 7.3-9
Operation:
Voltage loop vi1 - vi2 = -VGS1+ vGS1 + vSG4 - VSG4 = VSG3 - vSG3 - vGS2 + VGS2
Using the notation for ac, dc, and total variables gives,
vi2 - vi1 = vid = (vsg1 + vgs4) = -(vsg3 + vgs2)
If gm1 = gm2 = gm3 = gm4, then half of the differential input is applied across each
transistor with the correct polarity.
gm1vid gm4vid gm2vid gm3vid
i1 = 2 = and i2 = - 2 = - 2
2
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 28 – Differential-In, Differential-Out Op Amps (6/25/14) Page 28-12
M9 M7 M8 M10
M25
M26
M13 M24
vi1 vi2
M1 M2
M21 M22 R1
M14
vo1
vo2
M19 M3 M4 M20
M15 R2
M16
M27
M17 M18
M23
M28 +
VBias
M11 M12
- M5 M6
VSS Fig. 7.3-10
Quiescent output currents are defined by the current in the input cross-coupled
differential amplifier.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 28 – Differential-In, Differential-Out Op Amps (6/25/14) Page 28-13
M7 - M6
Cc Rz
M3 M4
Rz Cc vo2
vo1
vi1 M1 M2 vi2
M5
M9 + M8
VBN
-
VSS Fig. 7.3-12
Comments:
• Simple
• Unreferenced – value of common mode output voltage determined by the circuit
characteristics
060718-10
This scheme can be applied to any differential output amplifier.
CM Loop Gain = -gmC1Ro1 which can be large if the output of the differential output
amplifier is cascaded or a gain-enhanced cascode.
The common-mode loop gain may need to be compensated for proper dynamic
performance.
MC5 M5
MB
060718-11
M5 M6
CM Correction Circuitry
vo1 M1 M2 M3 M4 vo2
RCM RCM
VCMREF
060718-12
This circuit is capable of sustaining a large differential voltage without loading the
output of the differential output op amp.
070506-02
The CM feedback path has two poles – one at the gates of M10 and M11 and the
dominant output pole of the differential output op amp.
Can compensate with Miller capacitors as shown.
• The need for compensation of the common mode loop no longer exists since there is
only one dominant pole
• The dominant pole of the differential amplifier becomes the dominant pole of the
common mode feedback
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 28 – Differential-In, Differential-Out Op Amps (6/25/14) Page 28-21
Operation:
1.) During the 1 phase, both Ccm are charged to the desired value of Vocm and CMbias
= Vocm.
2.) During the 2 phase, the Ccm capacitors are connected between the differential
outputs and the CMbias node. The average value applied to the CMbias node will be
Vocm.
not at VCM, the currents I12 and I13 will change to force the value of the common mode
output voltage back to VCM.
SUMMARY
• Advantages of differential output op amps:
- 6 dB increase in signal amplitude
- Cancellation of even harmonics
- Cancellation of common mode signals including clock feedthrough
• Disadvantages of differential output op amps:
- Need for common mode output voltage stabilization
- Compensation of common mode feedback loop
- Difficult to interface with single-ended circuits
• Most differential output op amps are truly balanced
• For push-pull outputs, the quiescent current should be well defined
• Common mode feedback schemes include,
- Continuous time
- Discrete time
The model that has been developed for the large signal sub-threshold operation is:
W vGS-VT vDS
iD = It L exp nV 1 + V where vDS > 0 and VDS(sat) = VON = VGS -VT = 2nVt
t A
Small-signal model:
diD | W It vGS-VT vDS ID qID ID Cox
gm = =I exp 1 + = = =
dvGS Q t L nVt nV t V A nV t nkT Vt Cox+Cjs
diD | ID
gds = dv
DS Q VA
+ VDS
VBS VGS
-
111130-03
M6
M3 M4 Cc
vout
- M1 M2 CL
vin
+
+ M7
VBias M5
-
VSS Fig.7.4-1
Low frequency response:
ro2ro4 ro6ro7 1 1
Avo = gm2gm6 = (No longer )
o2
r + ro4 o6
r + ro7 n n
2 6 (kT/q) 2 ( 2 + 4 )( 6 + 7 ) I D
GB and SR:
ID1 ID5 ID1 kT
GB = and SR = =2 = 2GB n1 = 2GBn1Vt
(n1kT/q)C C C q
M8 M6
Total gain is, vi2
M1 M2
gm1(S6/S4) (S6/S4) vout
Avo = =
(gds6 + gds7) (6 + 7)n1Vt Cc
At room temperature (Vt = 0.0259V) and + M5
VBias
for typical device lengths, gains of 60dB M9 M7
can be obtained. -
The GB is, VSS Fig. 7.4-2
0
0 1 2
vIN nVt Fig. 7.4-5
i1 i2
Current
i2 for W2/L2 = 5.3(W1/L1)
M1 M2 +
i2 for W2/L2 = W1/L1
+ Vds2
VGS 100µA VGS
- -
Volts
0.1Vds2(sat) Vds1(sat)=Vds2(sat)
070507-02
M25 M26
+ i2 i1
M15 M23 VBias M5 M24 M16
M11 M20
M19 M12
- M6
VSS Fig.7.4-7
iin+IB iin IB
kiin
M3
50/1
M5 M4
1/1 1/1
M1 M2
1/1 210/1
Fig. 7.4-7A
M1 is M1 is
noisy S noiseless S Fig. 7.5-0A
Noise Analysis
1.) Insert a noise generator for each transistor that contributes to the noise. (Generally
ignore the current source transistor of source-coupled pairs.)
2.) Find the output noise voltage across an open-circuit or output noise current into a
short circuit.
3.) Reflect the total output noise back to the input resulting in the equivalent input noise
voltage.
+ * M1 M2 *
vin eto2
2 2
-
M1 M2 Cc en8 M8 M9 en9
vout
M11 * *
VBias 2 VBias
+ en6
VBias M8 M9 2 2
- M6 en3 en4 M6
*
M3 M4 M3 * * M4
2
The total output-noise voltage spectral density, eto, is as follows where gm8(eff) 1/rds1,
2 2 2 2 2 2 2 2 2
eto = gm62RII2en6+en7 +RI2gm12en1+gm22en2+gm32en3+gm42en4 + (en8/rds12) + (en9/rds22)
2
Divide by (gm1RIgm6RII)2 to get the eq. input-noise voltage spectral density, eeq, as
2 gm32en3
2 eto
2
2en6
2 2 2
en8 2 gm32en32
eeq = (g g R R )2 = g 2R 2 + 2en11+g 2 + 2 2en11+gm1 2
m1 e
m1 m6 I II m1 I g n1 m12rds1 en1
2 en1
2 = e 2 , e 2 = e 2 , e 2 = e 2 and e 2 = e 2 and g R is large.
where en6 n7 n3 n4 n1 n2 n8 n9 m1 I
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 29 – Low Power and Low Noise Op Amps (6/25/14) Page 29-20
2 2 gm32 n3
e
2
2 KNW3L1
eeq = 2en11+g 2 = 2en1 1 + (V2/Hz)
KPW1L3
m1 en1
Comments:
• The choices that reduce the 1/f noise also reduce the thermal noise.
Noise Corner:
Equating the equivalent input-noise voltage spectral density for the 1/f noise and the
thermal noise gives the noise corner, fc, as
3gmB
fc = 8kTWL
Example 29-3 Design of A Two-Stage, Miller Op Amp for Low 1/f Noise
Use the model parameters of KN’ = 120µA/V2, KP’ = 25µA/V2, and Cox = 6fF/µm2
along with the value of KF = 4x10-28 F·A for NMOS and 0.5x10-28 F·A for PMOS and
design the previous op amp with ID5 = 100µA to minimize the 1/f noise. Calculate the
corresponding thermal noise and solve for the noise corner frequency. From this
information, estimate the rms noise in a frequency range of 1Hz to 100kHz. What is the
dynamic range of this op amp if the maximum signal is a 1V peak-to-peak sinusoid?
Solution
1.) The 1/f noise constants, BN and BP are calculated as follows.
KF 4x10-28F·A
BN = = = 1.33x10-22 (V·m)2
2CoxKN’ 2·60x10 F/m ·120x10 A/V
-4 2 -6 2
and
KF 0.5x10-28F·A
BP = 2C K ’ = -4 2 -6 2 = 1.67x10-22 (V·m)2
ox P 2·60x10 F/m ·25x10 A/V
2.) Now select the geometry of the various transistors that influence the noise
performance.
2 small, let W = 100µm and L = 1µm. Select W = 10µm and L =
To keep en1 1 1 3 3
20µm and letW8 and L8 be the same as W1 and L1 since they little influence on the
noise.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 29 – Low Power and Low Noise Op Amps (6/25/14) Page 29-23
10
Experimental noise
performance: 8
Eq. input noise voltage of low-noise op amp
Noise (nV/ Hz)
4
Voltage noise of lateral BJT at 170 mA
2
0
10 100 1000 104 105
Frequency (Hz) Fig. 7.5-7
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 29 – Low Power and Low Noise Op Amps (6/25/14) Page 29-26
f
0 fc 2fc 3fc
VB(f)
f
0 fc 2fc 3fc
VC(f)
f
0 fc 2fc 3fc Fig. 7.5-8
Chopper-Stabilized Amplifier
With chopper
fc = 16kHz
nV/ Hz
100
With chopper fc = 128kHz
10
0 10 20 30 40 50
Frequency (kHz) Fig. 7.5-11
Comments:
• The switches in the chopper-stabilized op amp introduce a thermal noise equal to kT/C
where k is Boltzmann’s constant, T is absolute temperature and C are capacitors
charged by the switches (parasitics in the case of the chopper-stabilized amplifier).
• Requires two-phase, non-overlapping clocks.
• Trade-off between the lowering of 1/f noise and the introduction of the kT/C noise.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 29 – Low Power and Low Noise Op Amps (6/25/14) Page 29-31
† C. Menolfi and Q. Huang, “A Fully Integrated, Untrimmed CMOS Instrumentation Amplifier with Submicrovolt Offset,” IEEE J. of Solid-State
Circuits, vol. 34, no.8, March 1999, pp. 415-420.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 29 – Low Power and Low Noise Op Amps (6/25/14) Page 29-32
SUMMARY
• Operation of transistors for low power op amps is generally in weak inversion
• Boosting techniques are needed to get output sourcing and sinking currents that are
larger than that available during quiescent operation
• Be careful about using circuits at weak inversion, i.e. the self-biased cascode will
cause the resistor to be too large
• Primary sources of noise for CMOS circuits is thermal and 1/f
• Noise analysis:
1.) Insert a noise generator for each transistor that contributes to the noise.
(Generally ignore the current source transistor of source-coupled pairs.)
2.) Find the output noise voltage across an open-circuit or output noise current into a
short circuit.
3.) Reflect the total output noise back to the input resulting in the equivalent input
noise voltage.
• Noise is reduced in op amps by making the input stage gain as large as possible and
reducing the noise of this stage as much as possible.
• The input stage noise can be reduced by using lateral BJTs (particularily the 1/f noise)
• Doubly correlated sampling can transfer the noise at low frequencies to the clock
frequency (this technique is used to achieve low input offset voltage op amps).
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-1
INTRODUCTION
Implications of Low-Voltage, Strong-Inversion Operation
• Reduced power supply means decreased dynamic range
• Nonlinearity will increase because the transistor is working close to VDS(sat)
• Large values of because the transistor is working close to VDS(sat)
• Increased drain-bulk and source-bulk capacitances because they are less reverse biased.
• Large values of currents and W/L ratios to get high transconductance
• Small values of currents and large values of W/L will give small VDS(sat)
• Severely reduced input common mode range
• Switches will require charge pumps
supply is:
VDD
VDD(min.) = 3VON M3 M4 +
However, to have any input common mode range, the VPB1 VON
-
effective minimum power supply is, M1 + M2
VDD(min.) = VT + 2VON VON
+ +
VT+VON - VT+VON
- -
M5 +
VNB1 VON
-
060802-02
Therefore,
VDD(min.) = VT + 3VON
This could be reduced to 3VON with the floating battery but its implementation probably
requires more than 3VON of power supply.
Note the output signal swing is VT + VON while the input common range is VON.
-VT1
Input common-mode range: vicm M1 M2
Vicm(upper) = VDD - VSD3(sat) + VT1 VGS1
Vicm(lower) = VDS5(sat) + VGS1
VDS5(sat) +
VBias M5
-
Fig. 7.6-3
If the threshold magnitudes are 0.7V, VDD =
1.5V and the saturation voltages are 0.3V, then
Vicm(upper) = 1.5 - 0.3 + 0.7 = 1.9V
and
Vicm(lower) = 0.3 + 1.0 = 1.3V
giving an ICMR of 0.6V.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-6
VDD > Vicm > Vonp: (n-channel on and p-channel off) gm(eq) = gmN
Vonp Vicm Vonn: (n-channel on and p-channel on) gm(eq) = gmN + gmP
Vonn > Vicm > 0 : (n-channel off and p-channel on) gm(eq) = gmP
where gm(eq) is the equivalent input transconductance of the above input stage, gmN is
the input transconductance for the n-channel input and gmP is the input trans-
conductance for the p-channel input.
gm(eff)
gmN+gmP
gmP gmN
n-channel off Vonn n-channel on Vonp n-channel on
p-channel on p-channel on p-channel off
Vicm
0 VSDP5(sat)+VGSN1 VDD-VSDP5(sat)+VGSN1 VDD Fig. 7.6-5
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-7
gmN=gmP
0 Vicm
0 Vonn Vonp VDD Fig. 7.6-7
The above techniques and many similar ones are good for power supply values down to
about 1.5V. Below that, different techniques must be used or the technology must be
modified (natural devices).
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-9
Natural Transistors
Natural or native NMOS transistors normally have a threshold voltage around 0.1V
before the threshold is increased by increasing the p concentration in the channel.
If these transistors are characterized, then they provide a means of achieving low voltage
operation.
Minimum power supply (ICMR = 0):
VDD(min) = 3VON
Input common mode range:
Vicm(upper) = VDD – VON + VT(natural)
Vicm(lower) = 2VON + VT(natural)
If VT(natural) ≈ VON = 0.1V, then
Vicm(upper) = VDD
Vicm(lower) = 3VON = 0.3V
Therefore,
ICMR = VDD - 3VON = VDD – 0.3V VDD(min) ≈ 1V
Matching tends to be better (less doping and magnitude is smaller).
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-10
Bulk-Driven MOSFET
A depletion device would permit large ICMR even with very small power supply
voltages because VGS is zero or negative.
When a MOSFET is driven from the bulk with the gate held constant, it acts like a
depletion transistor.
Cross-section of an n-channel vBS VDS VGS VDD
bulk-driven MOSFET: Bulk Drain Gate Source Substrate
Channel
p+ n+ n+ n+
QP
Depletion p-well
Region QV
n substrate
Large signal equation: Fig. 7.6-8
KN’W
iD = 2L VGS - VT0 - 2|F| - vBS + 2|F|2
Small-signal transconductance:
(2KN’W/L)ID
gmbs =
2 2|F| - VBS
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-11
1000
Saturation: VDS > VBS – VP gives,
VBS = VP + VON 500
IDSS
Gate-source
VBS2
iD = IDSS 1 - V driven
P 0
-3 -2 -1 0 1 2 3
Comments: Gate-Source or Bulk-Source Voltage (Volts)
Fig. 7.6-9
200nA
Bulk-Source Current
150nA
100nA
50nA
-50nA
-0.50V -0.25V 0.00V 0.25V 0.50V
Input Common-Mode Voltage Fig. 7.6-10A
Comments:
• Effective ICMR is from VSS to VDD -0.3V
• The transconductance of the input stage can vary as much as 100% over the ICMR
which makes it very difficult to compensate
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-14
†T. Lehmann and M. Cassia, “1V Power Supply CMOS Cascode Amplifier,” IEEE J. of Solid-State Circuits, Vol. 36, No. 7, 2001.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-15
VDD
Current-Driven Bulk Technique
Bias circuit for keeping Imax defined VBias1
M7
independent of BJT betas. M3
IS,E =1.3ID
Note:
ID,C = ICD + ID M6 ID IE IR
IS,E = ID + IE + IR R
IBB M1 M2
The circuit feedback causes a bulk bias ID,C =1.1ID ICD M5
M4
current IBB and hence a bias voltage VBIAS M8 +
such that VBias
IR =
IS,E = ID + IBB(1+CS + CD) + IR VBias2
0.1ID
-
Use VBias1 and VBias2 to set ID,C 1.1ID, VSS
130418-01
IS,E 1.3ID and IR 0.1ID which sets IBB at 0.1ID assuming we can neglect ICS with
respect to ICD. This is illustrated as follows,
IS,E ≈ ID + IBB(1+CD) + IR = ID + IBB + ICD + IR = ID + IBB + 0.1ID + 0.1ID = 1.3ID
For this circuit to work, the following conditions must be satisfied:
VBE < VTN + IRR and |VTP| + VDS(sat) < VTN + IRR
If |VTP| > VTN, then the level shifter IRR can be eliminated.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-16
The problem with this approach is the number of poles that occur (one per stage) if the
amplifier is to be used in a closed loop application. Instability or poor transient response
will result.
This approach is complicated by the feedforward paths which create RHP zeros.
Unfortunately, the analysis becomes quite complex - for the details refer to the reference
below.
† R. Hogervorst and J. H. Huijsing, Design of Low-Voltage, Low-Power Operational Amplifier Cells, Kluwer Academic Publishers, 1996, pp. 127-
131.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-21
VDD VDD
M7
M3 M4 M7 M3 M4
or
M6 M6 M5
M1 M2 M1
M5 IB2 M2
Fig. 7.6-13A
M4
iout M3
+ + +
VGS3 VBS3 VGS4
- -M2
M1 M2 - M1
+ + + + + +
VGS V BS VGS VGS1 VBS1 VGS2
- - - - - -
Simple bulk-driven Cascodebulk-driven
current mirror current mirror. Fig.7.6-11
The cascode current mirror gives a minimum input voltage of less than 0.5V for currents
less than 100µA
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-23
R1
Voltage-mode bandgap topology. Current-mode bandgap topology. Voltage-current mode bandgap topology.
Fig. 7.6-14
Current
K2IVBE K1IPTAT
I2 INL K3INL
INL
IVBE K1IPTAT
Temperature
Circuit to generate nonlinear correction term, INL. Illustration of the various currents.
Fig. 7.6-16
0 , K2IVBE > K1IPTAT
INL =
K1IPTAT - K2IVBE , K2IVBE < K1IPTAT
The combination of the above concept with the previous slide yielded a curvature-
corrected bandgap reference of 0.596V with a TC of 20ppm/C° from -15C° to 90C°
using a 1.1V power supply.† In addition, the line regulation was 408 ppm/V for
1.2VDD10V and 2000 ppm/V for 1.1VDD10V. The quiescent current was 14µA.
† G.A. Rincon-Mora and P.E. Allen, “A 1.1-V Current-Mode and Piecewise-Linear Curvature-Corrected Bandgap Reference,” J. of Solid-State
Circuits, vol. 33, no. 10, October 1998, pp. 1551-1554.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-25
VPB2 vOUT
Cc
+ M1 M2
vIN M6 M7
-
VNB1
M5 M8 M9 M10
060804-01
Performance:
Gain ≈ gm2rds2
Miller compensated
Output swing is VDD -2VON
Max. CM input = VDD
Min. CM input = 2VON + VT
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 30 – Low Voltage Op Amps (6/25/14) Page 30-26
Performance:
Gain ≈ gm2rds2, self compensated, and output swing is VDD -4VON
VPB2 VPB2
+ - vOUT
VNB2 VNB2
VNB1
1:3
060804-02
M6 M13
Cx M9 M10
M17
+
vin M1 M2 vout
-
CL
M7 M8
VBiasN M3 M5 M4 M14
M15
M16
SUMMARY
• Integrated circuit power supplies are rapidly decreasing (today 2-3Volts)
• Classical analog circuit design techniques begin to deteriorate at 1.5-2 Volts
• Approaches for lower voltage circuits:
- Use natural NMOS transistors (VT 0.1V)
- Drive the bulk terminal
- Forward bias the bulk
- Use depeletion devices
• The dynamic range will be compressed if the noise is not also reduced
• Fortunately, the threshold reduction continues to allow the techniques of this section to
be used in today’s technology
CHARACTERIZATION OF COMPARATORS
What is a Comparator?
The comparator is a circuit that compares one analog signal with another analog signal or
a reference voltage and outputs a binary signal based on the comparison.
The comparator is basically a 1-bit analog-to-digital converter:
Comparator symbol:
vP-vN vP-vN
VOL VOL
Noninverting Comparator Inverting Comparator
Fig. 8.1-2A
vP-vN
Model:
vP
+ +
vP-vN f0(vP-vN) vO
- -
vN
Comparator
VOH for (vP-vN) > 0
f0(vP-vN) =
VOL for (vP-vN) < 0 Fig. 8.1-3
VOH-VOL
Gain = Av = lim where V is the input voltage change
V
V0
VOH+VOL
VOS = the input voltage necessary to make the output equal when vP = vN.
2
Model:
vP
+vP' +
±VOSv '-v ' f1(vP'-vN') vO
P N
-v ' -
vN N
Comparator Fig. 8.1-7
Other aspects of the model:
ICMR = input common mode voltage range (all transistors remain in saturation)
Rin = input differential resistance
Ricm = common mode input resistance
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 31 – Open-Loop Comparators (6/26/14) Page 31-7
Comparator Noise
vo
Noise of a comparator is modeled as if the VOH
comparator were biased in the transition region. Rms Noise
vP-vN
Noise leads to an uncertainty in the transition
region causing jitter or phase noise. VOL
Transition Uncertainty Fig. 8.1-8
Step Response:
vo(t) = Av(0) [1 - e-t/c]Vin
where
Vin = the magnitude of the step input.
Maximum slope of the step response:
dvo(t) Av(0)
= e-t/cVin
dt c
The maximum slope occurs at t = 0 giving,
dvo(t) | Av(0)
dt t=0 = c Vin
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 31 – Open-Loop Comparators (6/26/14) Page 31-11
Thus, if k = 1, tp = 0.693c.
vout
Illustration:
Vin > Vin(min)
VOH
Obviously, the more overdrive vin + vout VOH+VOL
2
applied to the input, the smaller - VOL Vin = Vin(min)
the propagation delay time. 0 t t (max) t
0 p p Fig. 8.1-10
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 31 – Open-Loop Comparators (6/26/14) Page 31-12
VPBias2
MC3 MC4 vo
CL
MC1 MC2
vp M1 M2 v
VBias n
-
+ M5
VNBias1
-
060808-02
• Gain ≈ gm2rds2
• Slew rate = I5/CL
• Dominant pole = -1/(RoutCL) = -1/(gmrds2CL)
Folded-Cascode Comparator
VDD
VPB1
M4 M5
VPB2
M6 vOUT
vP M7
M1 M2 VNB2
vN
M8 M9 CL
M3
VNB1 I3 M11
M10
060808-03
• Gain ≈ gm2rds2
• Slew rate = I3/CL
• Dominant pole = -1/(RoutCL) ≈ -1/(gmrds2CL)
• Slightly improved ICMR
M10 M11
VPB1 M3
vP -A
M8 M9
vOUT
vN M1 M2
M6 M7 CL
-A -A
VNB1 M4
M5
060808-04
• Gain ≈ gm1Rout
• Rout ≈ [Ards7gm7(rds1||rds5)]|| (Ards9gm9rds11)
• Slew rate = I3/CL
• Dominant pole = -1/(RoutCL)
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 31 – Open-Loop Comparators (6/26/14) Page 31-18
M3 M4
M6
vn vout
M1 M2
vp CL
+ M7
VNB1 M5
-
060808-05
• Much faster linear response – the two poles of the comparator are typically much larger
than the dominant pole of the self-compensated type of comparator.
• Be careful not to close the loop because the amplifier is uncompensated.
I7 I6-I7
• Slew rate: SR = C and SR = C
- +
II II
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 31 – Open-Loop Comparators (6/26/14) Page 31-19
Normalizing gives,
vout(t) m -t 1 -mt p2
vout’(tn ) = A (0)V = 1 - m-1e n + m-1e n where m = p 1 and tn = -tp1
v in 1
If p1 = p2 (m=1), then vout’(tn) = 1 - etp1 + tp1etp1 = 1 - e-tn - tne-tn
1
m=4
Normalized Output Voltage
0.8
m=2 m = 1 m = 0.5
m = 0.25
0.6
0.4
p2
m= p
1
0.2
0
0 2 4 6 8 10
CMOS Analog Circuit Design Normalized Time (tn = -tp1 ) Fig. 8.2-2 © P.E. Allen - 2016
Lecture 31 – Open-Loop Comparators (6/26/14) Page 31-22
• For the two-stage comparator using NMOS input transistors, the slew rate is
I7
-
SR = C
II
I6-I7 0.56(VDD-VG6(min)-|VTP|)2 - I7
SR+ = C =
II CII
can’t be easily solved so approximate the step response as a power series to get
m tn2 1 m2tn2
vout(tn) Av(0)Vin 1 - m-1 1-tn+ 2 + ··· + m-1 1-mtn+ 2 +···
mtn2Av(0)Vin
2
Therefore, set vout(tn) = 0.5(VOH-VOL)
VOH-VOL mtpn2Av(0)Vin
2 2
or
VOH-VOL Vin(min) 1
tpn mAv(0)Vin = =
mVin mk
This approximation is particularly good for large values of k.
VOH-VOL 1 1 = 0.031
when vn is equal to 2A (0)V = 2k 2k 0
v in 0 2 4 6 8 10
0.52 Normalized Time (tn = tp1 = t/t1)
This corresponds well with the tp = 0.52 = 77ns
120524-01 6.75x106
figure shown where the normalized
propagation time delay is the time at which the amplitude is 1/2k or 0.031which
corresponds to tpn of approximately 0.52 compared with 0.491 of above.
Similarly, for Vin = 100mV and 1V we get a propagation time delay of 23ns and
7.3ns, respectively.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 31 – Open-Loop Comparators (6/26/14) Page 31-26
i3 i4
VOH = VDD – (VDD-VG6(min)-|VTP|)
M3 M4 vo1
M6
8I7 i1 i2 CI
×1 - 1- vout
6(VDD-VG6(min)-|VTP|)2 vG1 M1 M2 vG2
CII
ISS
+ M7
VBias M5
CMOS Analog Circuit Design - © P.E. Allen - 2016
VSS 120524-02
Lecture 31 – Open-Loop Comparators (6/26/14) Page 31-27
2.5V
1.) Total delay = sum of the first and second stage delays, t1 and t2
2.) First, consider the change of vG2 from -2.5V to 2.5V at 0.2µs.
The last row of table on Slide 31-26 gives vo1 = +2.5V and vout = -2.5V
3.) tf1, requires CI, Vo1, and I5. CI = 0.2pF, I5 = 30µA and V1 can be calculated by
finding the trip point of the output stage.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 31 – Open-Loop Comparators (6/26/14) Page 31-30
5.) The rising propagation time delay of the second stage requires CII, Vout, and I6. CII
is given as 5pF, Vout = 2.5V (assuming the trip point of the circuit connected to the
output of the comparator is 0V), and I6 can be found as follows:
VG6(guess) 0.5[VG6(I6=234µA) + VG6(min)]
2·15
VG6(min) = VG1 - VGS1(ISS/2) + VDS2 -VGS1(ISS/2) = -0.7 - 110·3 = -1.00V
VG6(guess) 0.5(1.304V-1.00V) = 0.152V
6 38·50
Therefore VSG6 = 2.348V and I6 = 2 (VSG6-|VTP|)2 = 2
2 (2.348 - 0.7) = 2,580µA
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 31 – Open-Loop Comparators (6/26/14) Page 31-31
Thus the total propagation time delay of the rising output of the comparator is
approximately 13.3 ns and most of this delay is attributable to the first stage.
7.) Next consider the change of vG2 from 2.5V to -2.5V at 0.4µs. We shall assume that
vG2 has been at 2.5V long enough for the conditions of the table on Slide 31-26 to be
valid. Therefore, vo1 VSS = -2.5V and vout VDD. The propagation time delays for the
first and second stages are calculated as 3V
vout
1.304V-(-1.00V)
tro1 = 0.2pF
= 15.4 ns
2V
VTRP6 = 1.304V
30µA
2.5V 1V
SUMMARY
• The two-stage, open-loop comparator has two poles which should as large as possible
• The transient response of a two-stage, open-loop comparator will be limited by either
the bandwidth or the slew rate
• It is important to know the initial states of a two-stage, open-loop comparator when
finding the propagation delay time
• If the comparator is gainbandwidth limited then the poles should be as large as
possible for minimum propagation delay time
• If the comparator is slew rate limited, then the current sinking and sourcing ability
should be as large as possible
AUTOZEROING
Principle of Autozeroing
Use the comparator as an op amp to sample the dc input offset voltage and cancel the
offset during operation.
Comments:
• The comparator must be stable in the unity-gain mode (self-compensating comparators
are ideal, the two-stage comparator would require compensation to be switched in
during the autozero cycle.)
• Complete offset cancellation is limited by charge injection
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 32 – Improved Open-Loop Comparators and Latches (6/26/14) Page 32-3
Inverting:
Comment on autozeroing:
Need to be careful about noise that gets sampled onto the autozeroing capacitor and
is present on the comparison phase of the process.
HYSTERESIS
Influence of Input Noise on the Comparator
Comparator without hysteresis: Comparator with hysteresis:
Comparator vin vin
threshold VTRP+
t t
VTRP-
vout vout
VOH VOH
t t
VOL Fig. 8.4-6A VOL
Fig. 8.4-6B
Voltage Regulator with input voltage having too large of source resistance, RS:
+ -
RS
+ Enable Voltage
VIN - Regulator
VON CL RL
150604-01
VTRP+ R1(V -V )
R2 OH OL VTRP+
vIN 0 vIN
0
VTRP- VTRP-
VOL VOL
IBias M3 M6 M7 M4
vo1 vo2
vi1 M1 M2 vi2
M8 M5
Fig. 8.4-11
VSS
2.4
Remember the simple
2.2
SAH model does not do
a good job of modeling 2
the knee or saturation vo2
1.8
(volts)
voltage.
1.6
1.4
1.2
1
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5
vin (volts) Fig. 8.4-13
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 32 – Improved Open-Loop Comparators and Latches (6/26/14) Page 32-17
IBias M3 M6 M7 M4
M9 M8
vi1 vi2
M1 M2 vout
M10 M11
M8 M5 2.6
2.4
VSS Fig. 8.4-14 2.2
2
vout
1.8
(volts)
1.6
1.4
1.2
1
-0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5
vin (volts) 120524-03
Schmitt Trigger
The Schmitt trigger is a circuit that has better defined switching points.
Consider the following circuit:
VDD How does this circuit work?
Assume the input voltage, vin, is low and the output voltage,
M5
vout , is high.
M4 M3, M4 and M5 are on and M1, M2 and M6 are off.
vin M3 vout When vin is increased from zero, M2 starts to turn on causing M3
M2 to start turning off. Positive feedback causes M2 to turn on
M6
further and eventually both M1 and M2 are on and the output is at
zero.
M1 The upper switching point, VTRP+ is found as follows:
120524-04
When vin is low, the voltage at the source of M2 (M3) is
vS2 = VDD-VTN3
VTRP+ = vin when M2 turns on given as VTRP+ = VTN2 + vS2
VTRP+ occurs when the input voltage causes the currents in M3 and M1 to be equal.
0 0 vin
VTRP- VTRP+ VDD
Fig. 8.4-16
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 32 – Improved Open-Loop Comparators and Latches (6/26/14) Page 32-20
SIMPLE LATCHES
Regenerative Comparators
Regenerative comparators use positive feedback to accomplish the comparison of two
signals. Latches can have a faster switching speed than the previous comparators.
NMOS and PMOS latch:
VDD VDD
I1 I2 M1 M2
vo1 vo2 vo1 vo2
M1 M2 I1 I2
I1 I2 M1 M2
Enable Enable Enable Enable
Vo1ʼ Vo2ʼ Vo1ʼ Vo2ʼ
M1 M2 I1 I2
VOH- VOL
The propagation delay time is tp = L ln
2 V i
Note that the larger the Vi, the faster the response.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 32 – Improved Open-Loop Comparators and Latches (6/26/14) Page 32-25
Since the propagation time delay is the time when the output is 0.5(VOH-VOL), then
using the above results or Fig. 8.5-5 we find for Vi = 0.01(VOH-VOL) that tp = 3.91L =
0.512ns and for Vi = 0.1(VOH-VOL) that tp = 1.61L = 0.211ns.
†A. Coban, “1.5V, 1mW, 98-dB Delta-Sigma ADC”, Ph.D. dissertation, School of ECE, Georgia Tech, Atlanta, GA 30332-0250.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 32 – Improved Open-Loop Comparators and Latches (6/26/14) Page 32-28
Tail-Referenced Latch
The previous two latches experience poor input offset VDD
voltage characteristics because the input devices are Latch Latch
working in the linear region during the latch phase.
vout- vout+
The latch below keeps the input devices in the sat-
uration region. The resulting larger gain of the input
devices reduces the input offset voltage as shown.
vin+ Vref+ Vref- vin-
The input offset voltage of the tail referenced M1 M2
latch is compared between two latches with the
referenced latch for 100 samples. The x-axis is the Latch
All transistors are
3.5µm/0.4µm except
deviation from the mean of the first latch and the y- 070511-01 M1 and M2
axis is the deviation of the mean of the second latch.
CMOS Latch
Circuit:
When Latch_bar is high, M5 and M6 are off, M7 is on, and the latch is disabled and the
outputs are shorted together.
When Latch_bar is low, the input voltages stored at the sources of M1 and M2 will cause
one of the latch outputs to be high and the other to be low.
The source of M1 and M2 that is higher will have a larger source-gate voltage
resulting in a larger transconductance and more gain than the other transistor.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 32 – Improved Open-Loop Comparators and Latches (6/26/14) Page 32-31
Metastability
Metastability is the condition where the latch cannot make a decision in the time
allocated. Normally due to the fact that the input is small (within the input resolution
range).
Metastability can be improved (reduced) by increasing the gain of the comparator by
preceding it with an amplifier to keep the signal input to the latch as large as possible
under all conditions. The preamplifier also reduced the input offset voltage.
SUMMARY
• The performance of open-loop comparators can be improved by the use of autozeroing
and hysteresis
• Discrete-time comparators must work with clocks
• Regenerative comparators (latches) use positive feedback
• The propagation delay of the regenerative comparator is slow at the beginning and
speeds up rapidly as time increases
• The highest speed comparators will use a combination of open-loop comparators and
latches
One stage of this amplifier had a gain of 10 and a dominant pole at 551MHz. The
response of this amplifier to a step input is
Vout(t) = 10Vin (1-e-p1t)
If the output signal swing is 1V and the step is 0.1V, the propagation time delay is,
Vin(min) = 1/10 = 0.1V k = 1
2k 2
1
tp = ln = ln = 0.20 ns
p 2k-1 2π·551x106 2-1
Assuming a W/L ratio of 42 for M1 and 200 for M2, if the input can swing to VDD
(=2.5V) and ground, the sourcing and sinking currents are:
Kp'W 25·200
ISourcing = (VDD – |VTP|)2 = (2.5V-0.5)2 µA = 10.0 mA
2L 2
Kn'W 120·42
ISinking = 2L (VDD – VTN)2 = 2 (2.5V-0.5)2 µA = 10.1 mA
If larger currents are required, cascaded stages can be used to optimize the delay versus
the current output.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 33 – High Speed Comparators (6/26/14) Page 33-6
If the effective resistance of the driver is 30k, then the delay is 29 ns which is much too
large.
ln(CLoad /Cin)
From the above figure we see that CLoad = f NCin → N = ln f
The delay of a single, push-pull inverter can be expressed as,
Cj
tinv = invC + inv
j-1
where
inv = ReffCin (Reff is the effective output resistance of the inverter)
Cself Cjunction
inv = C = C (Cjunction is the bulk-drain capacitances)
in in
† D.A. Hodges, H.G. Jackson, and R.A. Saleh, Analysis and Design of Digital Integrated Circuits in Deep Submicron Technology, 3rd ed.,
McGraw-Hill Book Co., 2004, Chapter 6.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 33 – High Speed Comparators (6/26/14) Page 33-9
M1 M2
M1 M2
M5
VBias M5
VSS Fig. 8.3-4
VSS
Advantage:
Large sink or source current with out a large quiescent current.
Disadvantage:
Poor common mode range (vin+ slower than vin-)
†M. Bazes, “Two Novel Full Complementary Self-Biased CMOS Differential Amplifiers,” IEEE Journal of Solid-State Circuits, Vol. 26, No. 2,
Feb. 1991, pp. 165-168.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 33 – High Speed Comparators (6/26/14) Page 33-11
M8 M10
M3 M4
M6
vn
M1 M2 vout
vp CL
+ M7 M9 M11
VNB1 M5
-
060808-08
Comments:
• Slew rate = 3V/µs into 50pF
• Linear rise/fall time = 100ns into 50pF
• Propagation delay time 1µs
• Loop gain 32,000 V/V
• The quiescent dc currents in the output stages are not well defined
• Use the principle of optimizing the delay in cascaded inverters
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 33 – High Speed Comparators (6/26/14) Page 33-12
Fast rising
0 t
060810-04
0 t
tL 060810-05
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 33 – High Speed Comparators (6/26/14) Page 33-13
+ + +
Vin Ao Vo1 Latch Vout
- - -
060810-06
In order to keep the bandwidth of the amplifier large, the gain will be small. To achieve
Preamplifier 1 Preamplifier 2 Preamplifier n
+ + + + + +
Vin Ao1/n Vo1 Ao1/n Vo2 Von-1 Ao1/n Von Latch Vout
- - - - - -
Gain = Ao 060810-08
Therefore, the question is how many stages of the amplifier and what is the gain of each
stage for optimum results?
Enable
Solution
The solution is based on the figure shown.
Amplifier
We note that, VOH
voa(t) = 10[1-e--3dBt]0.05(VOH-VOL). Latch
x(VOH-VOL)
If we define the input voltage to the latch as,
t2
vil = x·(VOH-VOL) VOL t
t1 S01E3S1
then we can solve for t1 and t2 as follows:
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 33 – High Speed Comparators (6/26/14) Page 33-15
1
1 1 dtp
+ L ln
tp = t1 + t2 = ln → = gives
-3dB 1-2x 2x dx
2L-3dB
2x = = = 0.3859 (x = 0.1930)
2+2L-3dB 2+
10ns 1
t1 = ln 1-0.3859 = 1.592ns·0.4875 = 0.7762 ns
2
1
and t2 = 1ns ln 0.3859= 0.9522ns
Gain = Ao 070509-06
Comments:
• Autozero and reset phase followed by comparison phase
• In the autozero phase, switches labeled “Reset” and “FB” are closed.
• In the sample phase, switches labeled “Sample” and “ FB ” are closed.
• Can run as high as 200Msps
An Improved Preamplifier
Circuit:
VDD
VBiasP VBiasP
M3 M4
vout- M5 M6 vout+
Reset
M10 M12
FB M11 FB
M7 M8
VBias
vin+ vin-
M1 M2
VBiasN
M9
Fig. 8.6-5
Gain:
gm1 KN’(W1/L1)I1 KN’(W1/L1) I5
Av = - g = - =- 1+I
m3 KP’(W3/L3)I3 KP’(W3/L3) 3
If I5 = 24I3, the gain is increased by a factor of 5
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 33 – High Speed Comparators (6/26/14) Page 33-21
Gain = 20dB
f-3dB = 551MHz
Vref1 Vref2
VNB1
†T. Liechti, “Design of a High-Seed 12-bit Differential Pipelined A/D Converter,” Diploma Project, Feb. 2004, Microelectronic Systems
Laboratory, Swiss Federal Institute of Technology, Lausanne.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 33 – High Speed Comparators (6/26/14) Page 33-23
Clock waveforms:
Mean comparator power
dissipation is 140µW
under typical conditions
SUMMARY
• Comparators are limited in speed either by bandwidth or slew rate
• Increasing the magnitude of the poles improves the bandwidth limitations
• Increasing the current sinking/sourcing ability improves the slew rate limitation
• Most high speed comparators use a combination of preamplifier followed by a latch
- The preamplifier uses bandwidth to quickly build up the input
- The latch uses positive feedback to take the signal to its final state
INTRODUCTION
Importance of Data Converters in Signal Processing
Digital-Analog Converters
Digital Signal Characteristics:
Processing
System • Can be asynchronous or
Microprocessors DIGITAL- synchronous
Compact disks ANALOG Filter Analog
Amplifier
Read only memory
CONVERTER Output • Primary active element is
Random access memory
Digital transmission the op amp
Disk outputs
Digital sensors • Conversion time can vary
from fast (one clock period,
Reference Fig. 10.1-01 T) to slow (2No. of bits*T)
Analog-Digital Converters
Characteristics:
Digital Signal
Processing • Can only be synchronous (the analog
ANALOG- System signal is sampled and held during
Analog Sample Microprocessors
and
DIGITAL Compact disks conversion)
Input CONVERTER Read only memory
Hold Random access memory
Digital transmission
• Primary active element is the
Disk outputs
Digital sensors
comparator
• Conversion time can vary from fast
060922-01 Reference (one clock period, T) to slow (2No. of
bits*T)
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 34 – Characterization of DACs and Current and Voltage Scaling DACs (7/10/15) Page 34-4
Binary Switches
Input-Output Characteristics
Ideal input-output characteristics of a 3-bit DAC
1.000
0.875
Infinite Resolution
Analog Output Value Normalized to VREF Characteristic
0.750
0.625 1 LSB
0.500
Vertical Shifted
0.375 Characteristic
0.250
0.125
0.000
000 001 010 011 100 101 110 111
Digital Input Code Fig. 10.1-4
Definitions
• Resolution of the DAC is equal to the number of bits in the applied digital input word.
• The full scale (FS):
FS = Analog output when all bits are 1 - Analog output all bits are 0
VREF 1
FS = (VREF - N ) - 0 = VREF1 - N
2 2
• Full scale range (FSR) is defined as
FSR = N∞lim (FS) = V
REF
• Quantization Noise is the inherent uncertainty in digitizing an analog value with a finite
resolution converter.
Quantization Noise
1LSB
0.5LSB
Digital
0LSB Input
000 001 010 011 100 101 110 111
Code
-0.5LSB Fig. 10.1-5
More Definitions
• Dynamic Range (DR) of a DAC is the ratio of the FSR to the smallest difference that
can be resolved (i.e. an LSB)
FSR FSR
DR = LSB change = N = 2N
(FSR/2 )
or in terms of decibels
DR(dB) = 6.02N (dB)
• Signal-to-noise ratio (SNR) for the DAC is the ratio of the full scale value to the rms
value of the quantization noise.
T
1 t LSB FSR
2 2
rms(quantization noise) = T LSB T - 0.5 dt = 12 = 2N 12
0
vOUT(rms)
SNR =
(FSR/ 12 2N)
• Maximum SNR (SNRmax) for a sinusoid is defined as
vOUTmax(rms) FSR/(2 2) 6 2N
SNRmax = = = 2
(FSR/ 12 2N) FSR/( 12 2N)
or in terms of decibels
62N
SNRmax(dB) = 20log10 2 = 10 log10(6)+20 log10(2N)-20 log10(2)= 1.76 + 6.02N dB
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 34 – Characterization of DACs and Current and Voltage Scaling DACs (7/10/15) Page 34-8
• The uncertainty of each bit must be less than ±0.5 LSB (assuming all other bits are ideal.
Must use ±0.25 LSB if each bit has a worst case error.)
• The accuracy of the i-th bit is equal to the uncertainty divided by the output giving:
±0.5 LSB 1 100
Accuracy of the i-th bit = 2n-i-1 LSB = 2n-i = 2n-i %
Actual Gain
7/8 7/8
Characteristic Error
6/8 6/8 Actual
5/8 Characteristic
5/8
Offset
4/8 Error 4/8
Infinite Infinite
3/8 Resolution 3/8 Resolution
Characteristic Characteristic
2/8 2/8
Ideal 3-bit Ideal 3-bit
1/8 Resolution 1/8 Resolution
Characteristic Characteristic
0 0
000 001 010 011 100 101 110 111 000 001 010 011 100 101 110 111
Digital Input Code Digital Input Code
Offset Error in a 3-bit DAC Gain Error in a 3-bit DAC
Fig. 10.1-6
6
8
5 Nonmonotonicity
8
-1 LSB INL
4
8 +1.5 LSB INL A
3 -1.5 LSB DNL
8
2 Ideal 3-bit Characteristic
8
1
8 Actual 3-bit Characteristic
0
8 000 001 010 011 100 101 110 111
Digital Input Code Fig. 10.1-7
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 34 – Characterization of DACs and Current and Voltage Scaling DACs (7/10/15) Page 34-12
TESTING OF DACs
Input-Output Test
Test setup:
ADC
Digital N-bit Vout
ADC with Output
Digital Digital
Word DAC more resolution
Subtractor Error
Input under than DAC
(N+2 bits) Output
test (N+2 bits)
(N+2 bits) (N+2 bits)
Fig. 10.1-9
Comments:
Sweep the digital input word from 000...0 to 111...1.
The ADC should have more resolution by at least 2 bits and be more accurate than the
errors of the DAC
INL will show up in the output as the presence of 1’s in any bit.
If there is a 1 in the Nth bit, the INL is greater than ±0.5LSB
DNL will show up as a change between each successive digital error output.
The bits which are greater than N in the digital error output can be used to resolve the
errors to less than ±0.5LSB
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 34 – Characterization of DACs and Current and Voltage Scaling DACs (7/10/15) Page 34-18
Spectral Test
Test setup:
Comments:
Digital input pattern is selected to
have a fundamental frequency which
has a magnitude of at least 6N dB
above its harmonics.
Length of the digital sequence
determines the spectral purity of the
fundamental frequency.
All nonlinearities of the DAC (i.e. INL and DNL) will cause harmonics of the
fundamental frequency
The THD can be used to determine the SNR dB range between the magnitude of the
fundamental and the THD. This SNR should be at least 6N dB to have an INL of less
than ±0.5LSB for an ENOB of N-bits.
Note that the noise contribution of VREF must be less than the noise floor due to
nonlinearities.
If the period of the digital pattern is increased, the frequency dependence of INL can
be measured.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 34 – Characterization of DACs and Current and Voltage Scaling DACs (7/10/15) Page 34-19
Digital-Analog Converters
Serial Parallel
Comments:
1.) RF can be used to scale the gain of the DAC. If RF = KR/2, then
-KRb0 b1 b2 bN-1
vOUT=-RFIO = 2 R + 2R + 4R +···+ 2N-1RVREF
b0 b1 b2 bN-1
vOUT=-K 2 + 4 + 8 +···+ 2N VREF
where bi is 1 if switch Si is connected to VREF or 0 if switch Si is connected to ground.
RMSB R 1
2.) Component spread value = R = 2N-1R = 2N-1
LSB
3.) Attributes:
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 34 – Characterization of DACs and Current and Voltage Scaling DACs (7/10/15) Page 34-22
Operation:
vOUT = R2(bN-1·I + bN-2·2I + bN-3·4I + ··· + b0·2N-1·I)
VREF b0 b1 b2 bN-3 bN-2 bN-1
If I = IREF = N , then vOUT = 2 + 4 + 8 + ··· + N-2 + N-1 + N VREF
2 R2 2 2 2
Attributes:
Fast (no floating nodes) and not monotonic
Accuracy of MSB greater than LSBs
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 34 – Characterization of DACs and Current and Voltage Scaling DACs (7/10/15) Page 34-25
RL RL
+
vOUT
-
b0 b0 b1 b1 b2 b2 bN-1 bN-1
I I I I
2 4 8 2N
060926-01
b0 b1 b2 bN-1
vOUT = IRL 2 + 4 + 8 + ··· + + N
2
where
+1 if the bit is 1
bi =
-1 if the bit is 0
A single-ended DAC can be obtained by replacing the left RL by a short.
RL RL
+
vOUT
-
d0 d0 d1 d1 d2 d2 d3 d3 d 4 d4 d2N d2N
I I I I I I
2N 2N 2N 2N 2N 2N
d0 d1 d2 d3 d4 d2N
N to 2N Encoder
b0 b1 b2 bN 060926-02
I I I I I I
2N 2N 2N 2N 2N 2N
q0 q1 q2 q3 q4 q2N
N to 2N Encoder
b0 b1 b2 bN 060926-03
V1
V2
Voltage
VREF V3 Decoder
Scaling vOUT
Logic
Network
V2N
Fig. 10.2-6
Operation:
Creates all possible values of the analog output then uses a decoding network to
determine which voltage to select based on the digital input word.
vOUT
5 4V REF
large R 8
3VREF
• Sensitive to 4 vOUT
8
R 2VREF
parasitics 3 8
R V REF
• Requires a buffer 2 8
• Large current can R
1
0
000 001 010 011 100 101 110 111
flow through the R/2
Digital Input Code
resistor string.
(a.) (b.)
Figure 10.2-7 - (a.) Implementation of a 3-bit voltage scaling DAC. (b.) Input-output
characteristics of Fig. 10.2-7(a.)
R/2 b2 b1 b0
8
R
7
R 3-to-8 Decoder
6
R
5
R
4
R
3
R vOUT
2
R
1
R/2
Fig. 10.2-8
SUMMARY
• DACs scale a voltage reference as an analog output according to a digital word input
• Quantization noise is an inherent ±0.5 LSB uncertainty in digitizing an analog value with
a finite resolution converter
• The MSB requires the greatest accuracy with the LSB requiring the least accuracy
• Integral Nonlinearity (INL) is the maximum difference between the actual finite
resolution characteristic and the ideal finite resolution characteristic measured vertically
(% or LSB)
• Differential Nonlinearity (DNL) is a measure of the separation between adjacent levels
measured at each vertical jump (% or LSB)
• The limits to DAC speed include the parasitic capacitors, the op amp gain-bandwidth,
and the op amp slew rate
• Current scaling DACs scale the reference voltage into binary-weighted currents that are
summed into to a resistor to obtain the analog output voltage.
• Current scaling DACs are generally fast but have large element spreads and are not
monotonic
• The voltage scaling DAC creates all possible analog voltages and selects which one
corresponds to the digital input. The voltage scaling DAC is a monotonic converter.
Charge
VREF Scaling vOUT
Network
Fig. 10.2-9
Operation:
1.) All switches connected to
ground during 1.
2.) Switch Si closes to VREF if bi = 1 or to ground if bi = 0.
Equating the charge in the capacitors gives,
b1C b2C bN-1C
VREFCeq = VREF b0C + 2 + 22 + ... + N−1 = Ctot vOUT = 2C vOUT
2
which gives
vOUT = [b02-1 + b12-2 + b22-3 + ... + bN-12-N]VREF
Equivalent circuit of the binary-weighted, charge scaling DAC is:
Attributes: Ceq.
• Accurate
+
• Sensitive to parasitics
VREF 2C - Ceq. vOUT
• Not monotonic
• Charge feedthrough occurs at turn on of switches -
Fig. 10.2-11
where Ceq are capacitors whose bits are 1 and (2C - Ceq) are capacitors whose bits are 0.
The worst case DNL can be expressed as
vstep(worst case) vOUT(1000....) - vOUT(0111....)
DNL = - 1 = - 1 LSBs
vstep(ideal) 1 LSB
The worst case choice for the capacitors is to choose C1 larger by C and the remaining
capacitors smaller by C giving,
1 1 1 1
C1=C+C, C2 = 2(C-C),...,Cn-1= n-2(C-C), Cn= n-1(C-C), and Cterm= n-1(C-C)
2 2 2
n
Note that Ci + Cterm = C2+ C3+···+ Cn-1+ Cn+ Cterm = C-C
i=2
Example 35-1 - DNL and INL of a Binary Weighted Capacitor Array DAC
If the tolerance of the capacitors in an 8-bit, binary weighted, charge scaling DAC are
±0.5%, find the worst case INL and DNL.
Solution
For the worst case INL, we get from above that
INL = (27)(±0.005) = ±0.64 LSBs
For the worst case DNL, we can write that
DNL = (28-1)(±0.005) = ±1.275 LSBs
Attributes:
• No floating nodes which implies insensitive to parasitics and fast
• No terminating capacitor required
• With the above configuration, charge feedthrough will be Verror -(COL/2CN)V
• Can totally eliminate parasitics with parasitic-insensitive switched capacitor circuitry
but not the charge feedthrough
Approaches:
• Combination of similarly scaled subDACs
Divider approach (scale the analog output of the subDACs)
Subranging approach (scale the reference voltage of the subDACs)
• Combination of differently scaled subDACs
k-bit
k-LSB LSB ¸ 2m
bits DAC
Fig. 10.3-1
m-bit
m-MSB + vOUT
MSB S
bits +
DAC
VREF/2m
k-bit
k-LSB LSB
bits DAC
Fig. 10.3-2
Charge Amplifier DAC Using Two Binary Weighted Charge Amplifier SubDACs
Implementation:
Attributes:
• MSB subDAC is not dependent upon the accuracy of the scaling factor for the LSB
subDAC.
• Insensitive to parasitics, fast
• Limited to op amp dynamics (GB)
• No ICMR problems with the op amp
Voltage Scaling MSB SubDAC And Charge Scaling LSB SubDAC - Continued
Equivalent circuit of the voltage scaling (MSB) and charge scaling (LSB) DAC:
Ck = Ck-1 = C2 C1 Bus A
C
Bus A 2k-1C 2k-2C =2C =C Ceq.
vOUT 2-mVREF 2kC - Ceq. v'OUT
Sk,A Sk-1,A S2A S1A Bus B
2-mVREF vOUT
Sk,B Sk,B S2B S1B
V'REF
Bus B
V'REF
where, Fig. 10.3-8
C1 = C2 = Cm-1 Cm Cm R1
2m C 2m-1C =21C =C =C
R2
vOUT
S1,A S2,A Sm-2A Sm-1A k- R3
VREF vk VREF
to-
S1,B S2,B Sm-2B Sm-1B
2k
Decoder R2k-2
m-bit, MSB charge scaling subDAC
R2k-1
k-bit,
R2k LSB
voltage
Fig. 10.3-9A
scaling
k-LSB bits subDAC
b0 b1 bm-2 bm-1 vk bm bm+1 bm+k bm+k-1
vOUT = 21+22+···+2m-1+ 2m VREF+2m where vk = 21 + 22 +···+ 2k-1 + 2k VREF
b0 b1 bm-2 bm-1 bm bm+1 bm+k bm+k-1
vOUT =21 + 22 + ··· + 2m-1 + 2m + 2m+1 + 2m+2 + ··· + 2m+k-1 + 2m+k VREF
Attributes:
• MSBs have good accuracy
• LSBs are monotonic, have poor accuracy - require trimming for good accuracy
Example 35-4 – DAC with Voltage Scaling for MBSs and Charge Scaling for LSBs
Consider a 12-bit DAC that uses voltage scaling for the MSBs charge scaling for the
LSBs. To minimize the capacitor element spread and the number of resistors, choose m =
5 and k = 7. Find the tolerances necessary for the resistors and capacitors to give an INL
and DNL equal to or less than 2 LSB and 1 LSB, respectively.
Solution
Substituting n = 12 and k = 7 into the previous equations gives
11
R 6 C 7
R 7
C
2=2 R +2 C and 1 = 2 R + (2 -1) C
Solving these two equations simultaneously gives
C 24-2 C
C = 211 - 26 - 24 = 0.0071 C = 0.71%
R 28 - 26 -2 R
= 18 13 11 = 0.0008 = 0.075%
R 2 -2 -2 R
We see that the capacitor tolerance will be easy to meet but that the resistor tolerance will
require resistor trimming to meet the 0.075% requirement. Because of the 2n-1
multiplying R/R in the relationship, we are stuck with approximately 0.075%.
Therefore, choose m = 2 (which makes the 0.075% easier to achieve) and let k = 10
which gives R/R = 0.083% and C/C = 0.12%.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 35 – Parallel DACs, Improved Resolution DACs and Serial DACs (6/26/14) Page 35-25
Example 35-5 - DAC with Charge Scaling for MBSs and Voltage Scaling for LSBs
Consider a 12-bit DAC that uses charge scaling for the MSBs voltage scaling for the
LSBs. To minimize the capacitor element spread and the number of resistors, choose m =
7 and k = 5. Find the tolerances necessary for the resistors and capacitors to give an INL
and DNL equal to or less than 2 LSB and 1 LSB, respectively.
Solution
Substituting the values of this example into the relationships developed on a previous
slide, we get
R C R C
2 = 24 R + 211 C and 1 = R + (212-1) C
Solving these two equations simultaneously gives
C 24-2 C R 3 R
= = 0.000221 = 0.0221% and = 0.0968
C 216-211-24 C R 25-1 R = 9.68%
For this example, the resistor tolerance is easy to meet but the capacitor tolerance will
be difficult. To achieve accurate capacitor tolerances, we should decrease the value of m
and increase the value of k to achieve a smaller capacitor value spread and thereby
enhance the tolerance of the capacitors. If we choose m = 4 and k = 8, the capacitor
tolerance is 0.049% and the resistor tolerance becomes 0.79% which is still reasonable.
The largest to smallest capacitor ratio is 8 rather than 64 which helps to meet the
capacitor tolerance requirements.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 35 – Parallel DACs, Improved Resolution DACs and Serial DACs (6/26/14) Page 35-26
Switch S1 is the redistribution switch that parallels C1 and C2 sharing their charge
Switch S2 precharges C1 to VREF if the ith bit, bi, is a 1
Switch S3 discharges C1 to zero if the ith bit, bi, is a 0
Switch S4 is used at the beginning of the conversion process to initially discharge C2
Conversion always begins with the LSB bit and goes to the MSB bit.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 35 – Parallel DACs, Improved Resolution DACs and Serial DACs (6/26/14) Page 35-28
vC2/VREF
vC1/VREF
is given as b0 = 1, b1 = 1, b2 = 0, 1/2 1/2
and b3 = 1. Follow through the 1/4 1/4
sequence of events that result in 0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
the conversion of this digital t/T t/T Fig. 10.4-2
input word.
Solution
1.) S4 closes setting vC2 = 0.
2.) b3 = 1, closes switch S2 causing vC1 = VREF.
3.) Switch S1 is closed causing vC1 = vC2 = 0.5VREF.
4.) b2 = 0, closes switch S3, causing vC1 = 0V.
5.) S1 closes, the voltage across both C1 and C2 is 0.25VREF.
6.) b1 = 1, closes switch S2 causing vC1 = VREF.
7.) S1 closes, the voltage across both C1 and C2 is (1+0.25)/2VREF = 0.625VREF.
8.) b0 = 1, closes switch S2 causing vC1 = VREF.
9.) S1 closes, the voltage across both C1 and C2 is (0.625+1)/2VREF = 0.8125VREF =
(13/16)VREF.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 35 – Parallel DACs, Improved Resolution DACs and Serial DACs (6/26/14) Page 35-29
Pipeline DAC
The pipeline DAC is simply an extension of the sub-DACs concept to the limit where the
bits converted by each sub-DAC is 1.
Implementation:
6.) The MSB = 1, switch A is closed and vOUT = VREF + 0.5(0.375VREF) = 1.1875VREF =
(19/16)VREF. (Note that because the actual VREF of this example if ±VREF or 2VREF, the
analog value of the digital word 11001 is 19/32 times 2VREF or (19/16)VREF.)
SUMMARY
• Voltage scaling DACs are monotonic, use equal resistors but are sensitive to capacitve
parasitics
• Charge scaling DACs are fast with good accuracy but have large element spread and are
nonmonotonic
• DAC resolution can be increased by combining several subDACs with smaller
resolution
• Methods of combining include scaling the output or the reference of the non-MSB
subDACs
• SubDACs can use similar or different scaling methods
• Tradeoffs in the number of bits per subDAC and the type of subDAC allow
minimization of the INL and DNL
• Serial, charge redistribution DAC is simple and requires minimum area but is slow and
requires complex external circuitry
• Pipeline DAC has a latency of N+1 clock cycles but gives an analog output for each
clock
• Serial, algorithmic DAC is simple and requires minimum area but is slow and requires
complex external circuitry
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 36 – Characterization of ADCs and Sample and Hold Circuits (6/26/14) Page 36-1
INTRODUCTION
General Block Diagram of an Analog-Digital Converter
Digital
x(t) y(kTN)
Processor
• Prefilter - Avoids the aliasing of high frequency signals back into the baseband of the
ADC
• Sample-and-hold - Maintains the input analog signal constant during conversion
• Quantizer - Finds the subrange that corresponds to the sampled analog input
• Encoder - Encoding of the digital bits corresponding to the subrange
f
-fB 0 fS fS 2fS
2
Use of an antialiasing filter to avoid aliasing.
Antialiasing
Filter
f
-fB 0 fB fS fS Fig. 10.5-2
2
010
1 LSB
001
000
Quantization
1.0
Noise LSBs
0.5
vin
0.0 VREF
-0.5
0 1 2 3 4 5 6 7 8
8 8 8 8 8 8 8 8 8
Analog Input Value Normalized to VREF
CMOS Analog Circuit Design Figure 10.5-3 Ideal input-output characteristics of a 3-bit ADC. © P.E. Allen - 2016
Lecture 36 – Characterization of ADCs and Sample and Hold Circuits (6/26/14) Page 36-6
Definitions
• The dynamic range, signal-to-noise ratio (SNR), and the effective number of bits
(ENOB) of the ADC are the same as for the DAC
• Resolution of the ADC is the smallest analog change that distinguishable by an ADC.
• Quantization Noise is the ±0.5LSB uncertainty between the infinite resolution
characteristic and the actual characteristic.
• Offset Error is the difference between the ideal finite resolution characteristic and
actual finite resolution characteristic
• Gain Error is the difference 111 111
Gain Error = 1.5LSBs
Note that INL and DNL of an analog-digital converter will be in terms of integers in
contrast to the INL and DNL of the digital-analog converter. As the resolution of the
ADC increases, this restriction becomes insignificant.
111 Ideal
Characteristic
110
INL =
Digital Output Code
101
+1LSB Actual
100 Characteristic
INL = DNL =
011
-1LSB +1LSB
010
001 DNL =
0 LSB vin
000
0 1 2 3 4 5 6 7 8 VREF
8 8 8 8 8 8 8 8 8
Example of INL and DNL for a 3-bit ADC.) Fig.10.5-5
Note that the DNL and INL errors can be specified over some range of the analog input.
Monotonicity
A monotonic ADC has all vertical jumps positive. Note that monotonicity can only be
detected by DNL.
Example of a nonmonotonic ADC:
111
Actual
110 Characteristic
Digital Output Code
101
100
DNL =
011 -2 LSB
Ideal
010 Characteristic
001
vin
000
0 1 2 3 4 5 6 7 8 VREF
8 8 8 8 8 8 8 8 8
Fig. 10.5-6L
If a vertical jump is 2LSB or greater, missing output codes may result.
If a vertical jump is -1LSB or less, the ADC is not monotonic.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 36 – Characterization of ADCs and Sample and Hold Circuits (6/26/14) Page 36-10
Comparator
The comparator is the quantizing unit of ADCs.
Open-loop model:
Nonideal aspects:
• Input offset voltage, VOS (a static characteristic)
• Propagation time delay
- Bandwidth (linear)
Av(0) Av(0)
Av(s) = s =
sc + 1
+1
c
- Slew rate (nonlinear)
C·V V
T = I (I constant) = Slew Rate
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 36 – Characterization of ADCs and Sample and Hold Circuits (6/26/14) Page 36-13
Sample-and-Hold Circuit
Waveforms of a sample-and-hold S/H Command
Hold Sample Hold
circuit:
Output of S/H
Definitions:
Amplitude
ta ts valid for ADC
vin*(t)
• Acquisition time (ta) = time required conversion
Attributes:
• Fast, open-loop
• Requires current from the input to charge CH
• DC voltage offset of the op amp and the charge feedthrough of the switch will create dc
errors
Settling Time
Assume the op amp has a dominant pole at -a and a second pole at -GB.
GB2
The unity-gain response can be approximated as, A(s) ≈ 2
s + GB·s + GB2
4 3
The resulting step response is, vout(t) = 1 - e -0.5GB·t sin GB·t +
3 4
Defining the error as the difference between the final normalized value and vout(t), gives,
4 -0.5GB·t
Error(t) = = 1 - vout(t) = 3e
In most ADCs, the error is equal to ±0.5LSB. Since the voltage is normalized,
1 4 -0.5GB·ts 0.5GB·ts = 4 2N
= e → e
2N+1 3 3
Solving for the time, ts, required to settle with ±0.5LSB from the above equation gives
2 4 1
ts = GB ln 2N = GB [1.3863N + 1.6740]
3
Thus as the resolution of the ADC increases, the settling time for any unity-gain buffer
amplifiers will increase. For example, if we are using the open-loop, buffered S/H circuit
in a 10 bit ADC, the amount of time required for the unity-gain buffer with a GB of
1MHz to settle to within 10 bit accuracy is 2.473µs.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 36 – Characterization of ADCs and Sample and Hold Circuits (6/26/14) Page 36-17
C
+ f1 f2 -
+ -
vin C vout
-+
f1
- C +
f2 f1
120524-06
D3 D4 CH rd rd CH CH
RON = rd ROFF = ¥
Clock IB
Sample phase - diodes Hold phase - diodes
060927-01 forward biased. reversed biased.
MOS diode bridge S/H circuit:
VDD
Clock IB
Blowthru Capacitor
1 1
gm gm
vIN(t) M1 M2 vOUT(t) vIN(t) vOUT(t) vIN(t) vOUT(t)
M3 M4
1 1
CH gm gm CH CH
RON = 1/gm ROFF = ¥
Clock IB
Sample phase - MOS Hold phase - MOS
060927-02 diodes forward biased. diodes reversed biased.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 36 – Characterization of ADCs and Sample and Hold Circuits (6/26/14) Page 36-19
During the hold mode, the diodes D5 and D6 become forward biased and clamp the upper
and lower nodes of the sampling bridge to the sampled voltage.
Attributes:
• Accurate
• First circuit has signal-dependent feedthrough
• Slower because of the op amp feedback loop
Attributes:
• Accurate
• Signal-dependent feedthrough eliminated by a delayed clock
• Differential circuit keeps the output of the op amps constant during the 1 phase
avoiding slew rate limits
Attributes:
• Fast
• Requires current in and out
• Good for low voltage implementations
Op Amp Design
Gain:
1 1
Gain error = 1+Loop Gain ≤ 0.5 LSB = 11
2
Therefore, the op amp gain ≥ 211 = 2048 V/V
Choose the op amp gain as ≥ 5000 V/V
Gainbandwidth:
For a dominant pole op amp with unity-gain feedback, the relationship between the
gain-bandwidth (GB), accuracy (N) and speed (ts) is
N+1 N+1
ts = GB ln(2) = 0.693 GB
061021-01
Bias Currents:
The 100V/µs slew rate requires I3 = 100µA. Setting I4 = I5 = 125µA gives a power
dissipation of 0.875mW with VDD = 2.5V.
2·10 W6
VSG6 = |VT| + 2VON VSG6 - |VT| = 0.5V = L =
25·(W6/L6) 6
3.20
SUMMARY
• An ADC is by nature a sampled data circuit (cannot continuously convert analog into
digital)
• Two basic types of ADCs are:
- Nyquist – analog bandwidth is as close to the Nyquist frequency as possible
- Oversampled – analog bandwidth is much smaller than the Nyquist frequency
• The active components in an ADC are the comparator and the sample and hold circuit
• A sample and hold circuit must have at least the accuracy of 100%/2N
• Sample and hold circuits are divided into two types:
- Open loop which are fast but not as accurate
- Close loop which are slower but more accurate
• An example of designing a sample and hold amplifier was given to illustrate the
electrical design process for CMOS analog circuits
TESTING OF ADCs
Input-Output Test for an ADC
Test Setup:
2.0 LSB
1.5 LSB
Quantization Noise (LSBs)
+2LSB
1.0 LSB INL
Comments:
• Input sinusoid must have less distortion that the required dynamic range
• DAC must have more accuracy than the ADC
Occurances
code.
Number of
Sinusoidal Input
Illustration: Triangular Input
Output
0
0 Mid Full Code
Scale Scale
Comments: Fig.10.5-20
• Emphasizes the time spent at a given level and can show DNL and missing codes
• DNL
Width of the bin as a fraction of full scale H(i)/Nt
DNL(i) = Ratio of the bin width to the ideal bin width -1 = P(i) -1
where
H(i) = number of counts in the ith bin
Nt = total number of samples
P(i) = ratio of the bin width to the ideal bin width
• INL is found from the cumulative bin widths
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 37 – Testing of ADCs and Moderate Speed Nyquist ADCs (6/26/14) Page 37-7
Attributes:
• Simplicity of operation
• Subject to error in the ramp generator
• Long conversion time ≤ 2NT
Dual-Slope ADC
vin
Block diagram: Waveforms: VREF+Vth t1 = NREFT NREFT
vin'''
vin''' > vin'' > vin'.
vin''
vin'
Vth
0 t
0 t2'
Operation: Reset t0(start) t2''
t2'''
Fig.10.6-3 t2= NoutT
*
1.) Initially vint = 0 and vin is sampled and held (vIN > 0).
2.) Reset the positive integrator by integrating a positive voltage until vint (0) = Vth.
3.) Integrate vin* for NREF clock cycles to get,
NREFT
vint(t1) = K vin* dt + vint(0) = KNREFTvin* + Vth
0
4.) After NREF counts, the carry output of the counter closes switch 2 and-VREF is
applied to the positive integrator. The output of the integrator at t = t1+t2 is,
NoutT
vint(t1+t2) = vint(t1)+K (−VREF)dt =Vth → KNREFTvin*+Vth -KNoutTVREF = Vth
t
1
5.) Solving for Nout gives, Nout = NREF (vin*/VREF)
Comments: Conversion time ≤ 2(2N)T and the operation is independent of Vth and K.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 37 – Testing of ADCs and Moderate Speed Nyquist ADCs (6/26/14) Page 37-12
0.25VREF
0 t
0 1 2 3 4 5 6 T
Fig.10.7-2
†R. Hnatek, A User's Handbook of D/A and A/D Converters, John Wiley and Sons, Inc., New York, NY, 1976.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 37 – Testing of ADCs and Moderate Speed Nyquist ADCs (6/26/14) Page 37-14
Delay
G1 G2 G3 G4 G5
Delay
Clock pulses
1 1 1 1 1
SR1 SR2 SR3 SR4 SR5
Start pulse
The delay allows for the circuit transients to Shift Register
settle before the comparator output is sampled. Fig.10.7-3
Operation:
• Each stage multiplies its
input by 2 and adds or Vi/VREF
subtracts VREF depending 1.0
1 1 1 1
Vanalog = 5 2 − 4 + 8 + 16
= 5(0.4375) = 2.1875
An error will occur if the output voltage of one stage exceeds ±VREF (saturates).
[0010]
[0110]
[1010]
[1110]
[0000]
[0001]
[0011]
[0100]
[0101]
[0111]
[1000]
[1001]
[1011]
[1100]
[1101]
[1111]
[00] [01] [10] [11]
060930-01
[0110]
[1010]
[1110]
[0001]
[0101]
[1001]
[1101]
060930-02
The output bits can be used to determine the error. If these bits are 00, then 0.5LSB must
be added to get the correct digital output. If the bits are 11, then 0.5LSB must be
subtracted to get the correct digital output.
[0001]
[0010]
[0101]
[1001]
[0000]
[0100]
[1000]
[1010]
[0001]
[0010]
[0101]
[1001]
[0110]
[0000]
[0100]
[1000]
[1010]
[0110]
[1100]
[1101]
060930-03
To obtain code 11 out of the stage after correction, the correction logic must increment
the output of the stage.
To obtain code 00 from this stage after correction, the correction logic need do nothing.
Therefore, only two comparators are needed to produce outputs of (00, 01, 10) as shown
on the right-hand characteristic.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 37 – Testing of ADCs and Moderate Speed Nyquist ADCs (6/26/14) Page 37-24
When the shift is to the left, the comparator will not be in error until the shift is greater
than 0.25 VREF. This is because the comparator thresholds were shifted to the right by
0.5 VREF.
When the shift is to the right, the input to the next stage will be greater than 0.50VREF.
This will cause the output code 10 which indicates that the digital word should be
incremented by 1 bit.
The range of correction ±VREF /2B+1 where B is the number of bits per stage.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 37 – Testing of ADCs and Moderate Speed Nyquist ADCs (6/26/14) Page 37-25
Operation:
1.) Sample the input by connecting switch S1 to Vin*.
2.) Multiply Vin* by 2.
3.) If Va > VREF, set the corresponding bit = 1 and subtract VREF from Va.
If Va < VREF, set the corresponding bit = 0 and add zero to Va.
4.) Repeat until all N bits have been converted.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 37 – Testing of ADCs and Moderate Speed Nyquist ADCs (6/26/14) Page 37-29
SELF-CALIBRATION TECHNIQUES
Self-Calibrating Analog-Digital Converters
Self-calibration architecture for a m-bit charge scaling, k-bit voltage scaling successive
approximation ADC
Comments:
• Self-calibration can be accomplished during a calibration cycle or at start-up
• In the above scheme, the LSB bits are not calibrated
• Calibration can extend the resolution to 2-4 bits more that without calibration
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 37 – Testing of ADCs and Moderate Speed Nyquist ADCs (6/26/14) Page 37-31
SUMMARY
• Tests for the ADC include:
- Input-output test
- Spectral test
- FFT test
- Histogram test
• Moderate Speed ADCs:
Type of ADC Advantage Disadvantage
Serial ADC High resolution Slow
Voltage-scaling, charge- High resolution Requires
scaling successive considerable digital
approximation ADC control circuitry
Successive approximation Simple Slow
using a serial DAC
Pipeline ADC Fast after initial
Accuracy depends
latency of NT
on input
Iterative algorithmic ADC Simple
Requires other
digital circuitry
• Successive approximation ADCs also can be calibrated extending their resolution 2-4
bits more than without calibration.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 38 – High Speed Nyquist ADCs (6/26/14) Page 38-1
PARALLEL/FLASH ADCs
Parallel/Flash ADC Architecture
Analog Sample
vin(t)
Input and Hold
VREF Circuit
vin*(t)
V1 d1
Voltage
V2 d2 b1
Scaling
V3 d3 b2
Network
V4 2N-1 d4 2N-1 b3 Digital
creating Word
Compar to N
all possible Output
ators Decoder
discrete bN
analog V2N-1 d2N-1
voltages
Phase 1 Phase 2
060928-01 One Clock Period, T
• The notation, vin*(t), means the signal is sampled and held.
• The sample and hold function can be incorporated into the comparators
• The digital words designated as di form a thermometer code
General Comments: R - 1
0.875VREF +
• Fast, in the first phase of the clock the
R - 1
analog input is sampled and applied to the 0.750VREF +
comparators. In the second phase, the R - 0
digital encoding network determines the 0.625VREF + 2N-1 Output
correct output digital word. R to N Digital
- 0 encoder
0.500VREF + Word
• Number of comparators required is 2N-1
R - 0 101
which can become large if N is large 0.375VREF +
• The offset of the comparators must be less R - 0
than ±VREF/2N+1 0.250VREF +
R - 0
• Errors occur as “bubbles” in the 0.125VREF +
thermometer code and can be corrected R
with additional circuitry
Fig.10.8-1
• Typical sampling frequencies can be as
high as 1000MHz for 6-bits in sub-micron CMOS technology.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
1 2 3 4 5 6 7 8 Comparator
Vth
Threshold
V1c
V1b
V1a
V1
0 Vin
0 0.5VREF VREF
Fig.10.8-4
Comments:
• Capacitive loading at the input is reduced from 8 comparators to two amplifiers.
• The comparators no longer need a large ICMR
• V1 and V2, are interpolated through the resistor string and applied to the comparators.
• Because of the amplification of the input amplifiers and a single threshold, the
comparators can be simple and are often replaced by a latch.
• If the dots in Fig. 10.8-4 are not equally spaced, INL and DNL will result.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 38 – High Speed Nyquist ADCs (6/26/14) Page 38-8
Active Interpolation
Example of a 3 level current interpolation:
This type of interpolation works well with current processing, i.e., current comparators.
Vin Interpolating
Preamplifiers Amplifiers
++ ++
Vy Vo2
VR,j+1 - - - -
Aj+1
++
Vo3
- -
++ ++
Vx Vo1
VR,j - - - -
Aj
060928-03
Averaging†
VDD VDD VDD
In many cases, the comparators
Termination Termination Termination
consist of a number of pre- Resistors Resistors Resistors
amplifiers followed by a latch.
A11 A12 A13
Averaging is the result of
interconnecting the outputs of
each stage of amplifiers so that
the errors in one amplifier A21 A22 A23
chain are balanced out by
adjacent amplifier chains.
†
P.C.S. Scholtens and M. Vertregt, “A 6-b 1.6-Gsample/s Flash ADC in 0.18 µm CMOS Using Averaging Termination, IEEE J. of Solid-State
Circuits, vol. 37, no. 12, Dec. 2002, pp. 1599-1609.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 38 – High Speed Nyquist ADCs (6/26/14) Page 38-12
FOLDING
Folding Analog-Digital Converters
Allows the number of comparators to be reduced below the value of 2 N-1.
Architecture for a folded ADC:
n1
Coarse bits
Preprocessor Quantizer n1+n2
bits Digital
Encoding
v*in Output
n2 Logic
Folding Fine bits
Preprocessor Quantizer
Operation: 120525-01
The input is split into two or more parallel paths.
• First path uses a coarse quantizer to quantize the signal into 2 n1 values
• The second path maps all of the 2n1 subranges onto a single subrange and applies this
analog signal to a fine quantizer of 2n2 subranges.
Thus, the total number of comparators is 2n1-1 + 2n2-1 compared with 2n1+n2-1 for a
parallel ADC.
I.e., if n1 = 2 and n2 = 4, the folding ADC requires 3 + 15 = 18 compared with 63
comparators.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 38 – High Speed Nyquist ADCs (6/26/14) Page 38-14
No
n1 = 2 Folding
n2 = 3
32
VREF Folding
4
8
0 VREF
0 Analog Input
MSBs = 00 01 10 11 Fig.10.8-9
Problems:
• The sharp discontinuities of the folder are difficult to implement at high speeds.
• Fine quantizer must work at voltages ranging from 0 to VREF/4 (subranging).
• The actual frequency of the folding signal is F times the input frequency where F is the
number of folds
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 38 – High Speed Nyquist ADCs (6/26/14) Page 38-15
In the second case, the reference voltage for all comparators is identical which removes
any ICMR problems.
Comments:
• Number of comparators is 7 for the fine quantizer and 3 for the coarse quantizer
• The zero crossings of the folders must be equally spaced to avoid linearity errors
• The number of folders can be reduced and the comparators simplified by use of
interpolation
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 38 – High Speed Nyquist ADCs (6/26/14) Page 38-17
Folding Circuits
+VREF
Implementation R/8
R/8
VDD
R/8
of a times 4 R/8 RL RL Folding
R/8 Outputs
+V To com-
folder used R/8 - out parator
R/8 I
in a 3-bit R/8 V8
R/8
quantizer: R/8
R/8 V3
R/8 I V1 I V2 I V7 I V8
R/8
R/8
R/8 Vin
R/8
R/8 Vout
R/8 V2 +IRL
R/8
R/8
R/8
R/8 0 Vin
V1 V2 V3 V4 V5 V6 V7 V8 VREF
R/8
R/8
R/8 V1
Comments: -IRL 060928-06
061002-02
k-bits k-bits
Voltage
101 101 101
VREF 100 100 100
2 011 011 011
010 010 010
001 001 001
0 000 000 000 Time
Clock 1 Clock 2 Clock 3
Digital output = 011 111 001
MSB LSB Fig.10.8-14
Converted word is 011 111 001
Comments:
• Only 21 comparators are required for this 9-bit ADC
• Conversion occurs in three clock cycles
• The residue amplifier will cause a bandwidth limitation,
50MHz
GB = 50MHz → f-3dB = 6MHz
23
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 38 – High Speed Nyquist ADCs (6/26/14) Page 38-20
Note: the reference voltage of the previous stage (i-1) is divided by 2k to get the
reference voltage for the present stage (i), VREF(i), and so forth.
11
0.7500VREF
10
Comments:
Voltage
0.5000VREF
• Resolution of the 0.4375VREF
11
comparators for the 0.3750VREF 01 10
0.3125VREF 01
following stages increases 0.2500VREF
00
but fortunately, the
tolerance of each stage 00
decreases by 2k for every 0 Time
additional stage. Clock 1 Clock 2
Digital output word = 01 10 Fig.10.8-15
• Removes the frequency
limitation of the amplifier
Comparing the actual digital output word with the ideal output word gives the following:
+INL = 0LSB, -INL = 0111-0101 = -2LSB, +DNL = (1000-0101) - 1LSB = +2LSB,
and -DNL = (0101-0100) - 1LSB = 0LSB.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 38 – High Speed Nyquist ADCs (6/26/14) Page 38-25
Increment
DAC by 1
Vr1
LSB
ADC LSBs
Vr2
Fig.10.8-21
DAC
Features:
• Requires only 2n/2-1 comparators
• LSBs decoded using 31 preset charge redistribution capacitor arrays
• Reference voltages used in the LSBs are generated by the MSB ADC
• No op amps are used
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 38 – High Speed Nyquist ADCs (6/26/14) Page 38-26
Operation:
1.) Sample Vin* on
each 32C
capacitance
autozeroing the
comparators
Comments:
• Requires two full clock cycles
• Reuses the comparators
• Accuracy limited by resistor string
and its dynamic loading
• Accuracy also limited by the capacitor
array
• Comparator is a 3-stage, low-gain,
wide-bandwidth, using internal
autozeroing
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 38 – High Speed Nyquist ADCs (6/26/14) Page 38-28
Digital KB
Logic bits
041007-11
Operation:
1.) Stage 1 resolves the analog input signal to within one of B subranges which
determines the first B bits.
2.) Stage 1 then creates the analog residue (analog input – quantized analog output) and
passes on to Stage 2 by either amplifying or subranging.
3.) Stage 2 repeats this process which ends with Stage K.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 38 – High Speed Nyquist ADCs (6/26/14) Page 38-29
Comments:
• Adds a correcting bit to the following stage to correct for errors in the previous stage.
• The subranging or amplification of the next stage does not include the correcting bit.
• Correction can be done after all stages of the pipeline ADC have converted or after
each individual stage.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 38 – High Speed Nyquist ADCs (6/26/14) Page 38-30
SUMMARY
Typical Performance:
• 6-8 bits
• 500-2000 Msamples/sec.
• The ENOB at the Nyquist frequency is typically 1-2 bits less that the ENOB at low
frequencies.
• Power is approximately 0.3 to 1W
INTRODUCTION
What is an oversampling converter?
An oversampling converter uses a noise-shaping modulator to reduce the in-band
quantization noise to achieve a high degree of resolution.
• What is the range of oversampling?
The oversampling ratio, called M, is a ratio of the sampling frequency to the Nyquist
frequency of the input signal. The Nyquist frequency is twice the bandwidth of the
input signal. This oversampling ratio can vary from 8 to 256.
- The resolution of the oversampled converter is proportional to the oversampled ratio.
- The bandwidth of the input signal is inversely proportional to the oversampled ratio.
• What are the advantages of oversampling converters?
Very compatible with VLSI technology because most of the converter is digital
High resolution
Single-bit quantizers use a one-bit DAC which has no INL or DNL errors
Provide an excellent means of trading precision for speed (16-18 bits with a signal
bandwidth of 50kHz to 8-10 bits with a signal bandwidth of 5-10MHz).
• What are the disadvantages of oversampling converters?
Difficult to model and simulate
Limited in bandwidth to the clock frequency divided by the oversampling ratio
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 39 – Oversampling ADCs – Part I (6/26/14) Page 39-3
Components:
• Filter - Prevents possible aliasing of the following sampling step.
• Sampling - Necessary for any analog-to-digital conversion.
• Quantization - Decides the nearest analog voltage to the sampled voltage (determines
the resolution).
• Digital Coding - Converts the quantizer information into a digital output signal.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 39 – Oversampling ADCs – Part I (6/26/14) Page 39-4
The noise-shaping oversampling ADCs are also known as delta-sigma ADCs. We will
only consider the delta-sigma type oversampling ADCs.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 39 – Oversampling ADCs – Part I (6/26/14) Page 39-8
DELTA-SIGMA MODULATORS
General block diagram of an oversampled ADC
Components of the Oversampled ADC:
1.) Modulator - Also called the noise shaper because it can shape the quantization
noise and push the majority of the inband noise to higher frequencies. It modulates the
analog input signal to a simple digital code, normally a one-bit serial stream using a
sampling rate much higher than the Nyquist rate.
2.) Decimator - Also called the down-sampler because it down samples the high
frequency modulator output into a low frequency output and does some pre-filtering on
the quantization noise.
3.) Digital Lowpass Filter - Used to remove the high frequency quantization noise and to
preserve the input signal.
Note: Only the modulator is analog, the rest of the circuitry is digital.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 39 – Oversampling ADCs – Part I (6/26/14) Page 39-9
-0.5
-1
-1.5
0 50 100 150 200 250
Tme (Units of T, clock period) Fig.10.9-09
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 39 – Oversampling ADCs – Part I (6/26/14) Page 39-10
Higher-Order Modulators
A second-order, modulator:
q[nTs]
Multibit Quantizers
A single-bit quantizer:
= VREF
Advantage is that the DAC is inherently linear.
Multi-bit quantizer:
Consists of an ADC and DAC of B-bits.
VREF
= B fS
2 -1
Disadvantage is that the
DAC is no longer perfectly v
y
A/D
linear. To get large
resolution delta-sigma
ADCs requires highly
precise DACs. u
D/A
(b.) and (c.) For part (b.) and (c.) we obtain a minimum oversampling rates of M = 32.53
and 96.48, respectively. These values correspond to oversampling rates of 32 and 128,
respectively. The bandwidth of the converters is 312kHz for (b.) and 78kHz for (c.).
Q(z)
Q(z)
X(z) + z-1 + Y(z)
- 1-z-1
Fig.10.9-16
z-1 1 z-1
Y(z) = Q(z) + -1 [X(z) - Y(z)] → Y(z) -1 = Q(z) + -1 X(z)
1-z 1-z 1-z
Y(z) = (1-z-1)Q(z) + z-1X(z) → NTFQ (z) = (1-z-1) for L = 1
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 39 – Oversampling ADCs – Part I (6/26/14) Page 39-19
X(z) + + Y1(z) -1 +
- + Y(z)
z-1 z-1 z z-1 +
X2(z) = -1 (X(z) -Y1(z) - 1-z-1
1-z Fig.10.9-17
z-1 z-1
= -1 X(z) - -1 [(1-z-1)Q1(z) + z-1X(z)]
1-z 1-z
z-2 z-2
Y2(z) = (1-z )Q2(z) + z X2(z) = (1-z )Q2(z) + -1 X(z) - z Q1(z) - -1 X(z)
-1 -1 -1 -2
1-z 1-z
= (1-z-1)Q2(z) - z-2Q1(z)
Y(z) = Y2(z) - z-1Y2(z) + z-2Y1(z) = (1-z-1)Y2(z) + z-2Y1(z)
= (1-z-1)2Q2(z)-(1-z-1)z-2Q1(z)+(1-z-1)z-2Q1(z)+z-3X(z) = (1-z-1)2Q2(z)+z-3X(z)
Y(z) = (1-z-1)2Q2(z) + z-3X(z)
• The above structures that eliminate the noise of all quantizers except the last are called
MASH or multistage architectures.
• Digital error cancellation logic is used to remove the quantization noise of all stages,
except that of the last one.
1-bit
D/A Fig.10.9-20
Amplitude of integrator outputs:
1-bit
A/D Fig.10.9-20
Amplitude of integrator outputs (Integrator constants have been optimized to minimize
the integrator outputs):
Comments:
• The stability is guaranteed for cascaded structures
• The maximum input range is almost equal to the reference voltage level for the
cascaded structures
• All structures are sensitive to the circuit imperfection of the first stages
• The output of cascaded structures is multi-bit requiring a more complex digital
decimator
= = e
o jT
Vin( e ) C2 j2 sin( T/2) T j TC2 sin( T/2)
or
Vout(e jT)
o
C1 I
= (Ideal)x(Magnitude error)x(Phase error) where I = TC Ideal =
Vin( e jT)
o
2 j
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 39 – Oversampling ADCs – Part I (6/26/14) Page 39-26
SUMMARY
• Oversampled ADCs allow signal bandwidth to be efficiently traded for resolution
• Noise shaping oversampled ADCs preserve the signal spectrum and shape the noise
quantization spectrum
• The modulator shapes the noise quantization spectrum with a high pass filter
• The quantizer can be single or multiple bit
- Single bit quantizers do not require linear DACs because a 1 bit DAC cannot be
nonlinear
- Multiple bit quantizers require ultra linear DACs
• Modulators consist of combined integrators with the goal of high-pass shaping of the
noise spectrum and cancellation of all quantizer noise but the last quantizer
IMPLEMENTATION OF MODULATORS
Modulators – The Analog Part of the Oversampling ADC
Most of today’s delta-sigma modulators use fully differential switched capacitor circuits.
Advantages are:
• Doubles the signal swing and increases the dynamic range by 6dB
• Common-mode signals that may couple to the signal through the supply lines and
substrate are canceled
• Charge injected by the switches are canceled to a first-order
Example:
First integrator
dissipates the most
power and requires the
most accuracy.
X a1 a2 a3 a4 1-bit Y
S z - 1 y1 z - 1 y2 S z - 1 y3 S z - 1 y4 A/D
b1 b2
S
1-bit
Fig. 10.10-06
D/A
where a1 = 1/3, a2 = 3/25, a3 = 1/10, a4 = 1/10, b1= 6/5, b2= 1 and = 1/6
Advantages:
• The modulator combines the advantages of both DFB and DFF type modulators:
Only four op amps are required. The 1st integrator’s output swing is between ±VREF
for large input signal amplitudes (0.6VREF), even if the integrator gain is large (0.5).
• A local resonator is formed by the feedback around the last two integrators to further
suppress the quantization noise.
• The modulator is fully pipelined for fast settling.
†A.L. Coban and P.E. Allen, “A 1.5V, 1mW Audio Modulator with 98dB Dynamic Range, “Proc. of 1999 Int. Solid-State Circuits Conf., Feb.
1999, pp. 50-51.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 40 – Oversampling ADCs – Part II (7/17/15) Page 40-4
DR = 98 dB
BW = 20 kHz
Cs = 5 pF
0.5 µm CMOS
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 40 – Oversampling ADCs – Part II (7/17/15) Page 40-5
OSR = 64
OSR = 32
OSR = 16
OSR = 8
Capacitor Values 1
Capacitor Integrator 1 Integrator 2 Integrator 3 Integrator 4
Cs 5.00pF 0.15pF 0.30pF 0.10pF 1d
Ci 15.00pF 1.25pF 3.00pF 1.00pF 2
Ca - - 0.05pF - 2d
Cb1 - - - 0.12pF Fig.10.9-25
Cb2 - - - 0.10pF
Digital Filter
Implements a low pass filter by sampling the modulator stream of the 1-bit or multi-bit
code (PPM)
First-order averaging filter:
Modulator Delay Delay Delay
Input (PPM)
b1 b2 b3 bi
150717-02
Output
S S S
Comments: (PCM)
z-1 z-1
h(2) h(2)
z-1 z-1
h(3) h(3)
z-1 z-1
h(N-1) h(N-1)
†S.R. Norsworthy, R. Schreier, and G.C. Temes, Delta-Sigma Data Converters-Theory, Design, and Simulation, IEEE Press, NY, Chapter 13, 1997.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 40 – Oversampling ADCs – Part II (7/17/15) Page 40-13
Digital Decimator
The purpose of the decimator is to:
1.) Reduce the sample rate from fS down to the Nyquist frequency, 2fB.
2.) Help to remove the quantization noise.
3.) Perform the anti-aliasing filtering.
Challenges for the decimator:
1.) The input sampling rate is very high which makes it difficult to implement an efficient
digital decimation filter.
2.) Higher-order modulators produce highly shaped noise and require the filter to remove
this noise with not much frequency transition region.
3.) Should not distort the magnitude and phase characteristics of the input signal in the
baseband.
Goal:
Implement the digital decimator in a minimum amount of logic and make it feasible for
integrated circuit implementation.
fs fs/D 2fN fN fN
L+1-th First-half Second-half Droop
order band filter band filter correction
Fig.10.9-26
1.) For modulators with (1-z-1)L noise shaping comb filters are very efficient.
• Comb filters are suitable for reducing the sampling rate to four times the Nyquist
rate.
• Designed to suppress the quantization noise that would otherwise alias into the
signal band upon sampling at an intermediate rate of fs1.
2.) The remaining filtering is performed in stages by FIR or IIR filters.
• Suppresses out-of-band components of the signal
3.) Droop correction - may be required depending upon the ADC specifications
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 40 – Oversampling ADCs – Part II (7/17/15) Page 40-15
Comb Filters
A comb filter that computes a running average of the last D input samples is given as
1 D-1
y[n] = D x[n-i]
0
i=0 -20
K=1
where D is the decimation factor given as
|HD(f)| dB
-40
K= 2
fs
D=f -60
K=3
s1
-80
The corresponding z-domain expression is,
D -100
1 1 - z-D fs 2 fs 3 fs 4 fs
HD(z) = z-i = D 1 - z-1
0 fb
D D D D
Frequency Fig.10.9-27
i=1
The frequency response is obtained by evaluating HD(z) for z = ej2fTs,
1 sinfDTs -j2fT /D
HD(f) = D e s
sinfTs
where Ts is the input sampling period (=1/fs). Note that the phase response is linear.
For an L-th order modulator with a noise shaping function of (1-z-1)L, the required
number of comb filter stages is L+1. The magnitude of such a filter is,
1 sinfDTs K
|HD(f)| = D
sinfTs
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 40 – Oversampling ADCs – Part II (7/17/15) Page 40-16
Comments:
1.) The L+1 integrators operating at the sampling frequency, fs, realize the denominator
of HD(z).
2.) The L+1 differentiators operating at the output rate of fs1 (= fs/D) realize the
numerator of HD(z).
3.) Placing the integrator delays in the feedforward path reduces the critical path from
L+1 adder delays to a single adder delay.
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 40 – Oversampling ADCs – Part II (7/17/15) Page 40-17
-20
Magnitude (dB)
-50
-80
- 1+z-2
Fig.10.9-27B
Comments:
• Designed by applying a lowpass to bandpass transform to a second-order lowpass
modulator
• The stabilty and SNR characteristics are the same as those of a second-order lowpass
modulator
• The z-domain output is given as,
Y(z) = z-4X(z) + (1+z-2)2Q(z)
• The zeros are located at z = ±j which corresponds to notches at fs/4.
Operation:
1.) Interpolate a digital word at the conversion rate of the converter (fN) up to the sample
frequency, fs.
2.) The word length is then reduced to one bit with a digital sigma-delta modulator.
3.) The one bit PDM signal is converted to an analog signal by switching between two
reference voltages.
4.) The high-frequency quantization noise is removed with an analog lowpass filter
yielding the required analog output signal.
Sources of error:
• Device mismatch (causes harmonic distortion rather than DNL or INL)
• Component noise
• Device nonlinearities
• Clock jitter sensitivity
• Inband quantization error from the - modulator
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 40 – Oversampling ADCs – Part II (7/17/15) Page 40-26
Lowpass
filter Quantization noise after
filtering
output
A multi-bit output would consist of more parallel, controlled current sources and sinks.
SUMMARY
Comparison of the Various Types of ADCs
Speed Area Dependence
A/D Converter Type Maximum (Expressed in terms on the number of
Practical Number of T a clock period) bits, N, or other
of Bits (±1) ADC parameters
Dual Slope 12-14 bits 2(2NT) Independent
Successive Approximation 12-15 bits NT N
with self-correction
1-Bit Pipeline 10 bits T (After NT delay ) N
Algorithmic 12 bits NT Independent
Flash 6 bits T 2N
Two-step, flash 10-12 bits 2T 2N/2
Multiple-bit, M-pipe 12-14 bits MT 2N/M
- Oversampled (1-bit, L
loops and M= oversampling
ratio = f clock/2fb) 15-17 bits MT L
1.E+06
1.E+05
1.E+04
P/fsnyq [pJ]
1.E+03
1.E+02
ISSCC 2015
VLSI 2015
1.E+01
ISSCC 1997-2014
VLSI 1997-2014
1.E+00
FOMW=5fJ/conv-step
FOMS=175dB
1.E-01
10 20 30 40 50 60 70 80 90 100 110 120
SNDR @ fin,hf [dB]
1.E+08
1.E+07
1.E+06
1.E+05
1.E+04
1.E+03
10 20 30 40 50 60 70 80 90 100 110 120
CONCLUDING THOUGHTS
• What is analog circuit design?
The complex process of creating circuit solutions using analog circuit techniques.
• What is the analog integrated circuit design process?
The even more complex process of combining analog design with IC technology
which includes electrical, physical and test design.
• What are the key principles, concepts and techniques for analog IC design?
Key principles – Fundamental laws
Key concepts – Important relationships and ideas
Key techniques – Tools that allow simplification
or insight
• How can the analog IC designer enhance creativity
Technology changes but principles, concepts
and solve new problems in today’s industrial and techniques remain the same.
environment?
Learn the key principles, concepts and techniques
of analog circuit design
Learn from mistakes
Learn the technology
Always try to understand the concept and operation
of the circuit, never rely on a computer or someone else for this understanding
CMOS Analog Circuit Design © P.E. Allen - 2016
Lecture 01 – (8/9/18) Page 1-1
Circuits PLL
Perspective Components
Technology CMOS
Perspective Technology
Suggested References
Phase Locked Loops:
1. F.M Gardner, Phaselock Techniques, 2nd ed., John-Wiley & Sons, Inc., NY, 1979.
2. B. Razavi (ed.), Monolithic Phase-Locked Loops and Clock Recovery Circuits, IEEE
Press, 1997.
3. R.E. Best, Phase-Locked Loops: Design, Simulation, and Applications, 4th edition,
McGraw-Hill, 1999
4. A. Hajimiri and T.H. Lee, The Design of Low Noise Oscillators, Kluwer Academic
Publishers, 1999.
5. B. Razavi, Design of ICs for Optical Communications, McGraw-Hill, 2003.
6. T.H. Lee, The Design of CMOS Radio-Frequency Integrated Circuits, 2nd edition,
Cambridge University Press, NY, 2004.
7. C. Quemada, et al, Design Methodology for RF CMOS Phase Locked Loops, Artech
House, Norwood, MA, 2009.
• Phase/frequency detector determines the difference between the phase and/or frequency
of two signals
• The loop filter removes the high-frequencies from the voltage-controlled oscillator
(VCO) controlling voltage
• The VCO produces and output frequency controlled by a voltage
PLL Operation
Locked Operation:
• The loop is locked when the frequency of the VCO is exactly equal to the average
frequency of the input signal.
• The PLL has the inherent ability to suppress noise superimposed on its input signal.
• To maintain the control voltage needed for locked conditions, it is generally necessary
for the output of the phase/frequency detector to be nonzero.
Unlocked Operation:
• The VCO runs at a frequency called the free running frequency, o, which corresponds
to zero control voltage.
• The capture process is the means by which the loop goes from unlocked, free-running
state to that of the locked state.
by an amount .
1.) in increases by at to.
vosc(t) t
2.) The input signal leads the VCO and vd
begins to increase.
3.) After a delay due to the loop filter, the
VCO increases osc. t
wo Dw
to t
140420-06
The digital PLL (DPLL) has been the mainstay of most PLLs and is called the “classical”
digital PLL.
Input Error
Signal Signal Analog
Analog
Loop
Multiplier
Filter
Oscillator
Output Controlling
Voltage Voltage
Signal
Controlled
Oscillator 140418-04
Input
Signal Error Controlling
Signal Analog
Digital Voltage
Loop
Detector
Filter
Oscillator
Output
¸N Signal Voltage
Counter Controlled
(optional) Oscillator 140418-05
Input Digital
Signal Error
Digital Signal Digital Loop
Detector Filter
Oscillator Controlling
Output Digital Signal
Digitally
Signal
Controlled
Oscillator
Fixed
Oscillator
(Clock) 140418-06
Input Output
Signal Analog- Software Digital- Signal
Digital PLL Analog
Converter Converter
Clock
140418-07
Circuits PLL
Perspective Components
Technology CMOS
Perspective Technology
140418-08
Outline
• LPLL Blocks
• Locked State
• Order of the LPLL System
• The Acquisition Process - Unlocked State
• Noise in the LPLL
• LPLL System Design
• Simulation of LPLLs
LPLL BLOCKS
Building Blocks of the LPLL
v1(t) w1
Phase
vd(t) Low Pass
w2 Detector
Filter
(Multiplier)
Loop Filters
In the PLL, there are many high frequencies including noise that must be removed by
the use of a low pass filter in order to achieve optimum performance.
Types of Loop Filters:
1.) Passive lag filter (lag-lead)
1 + s2
F(s) = where 1 = R1C and 2 = R2C
1 + s(1 + 2)
|F(jw)| dB
R1
0dB
+ + -20 dB/decade
R2
Vd Vf
C t2
- - t1+t2 dB
1 1 log10(w)
140419-01 t1+t2 t2
• Has large open loop gain at low frequencies Large hold range
• Limited by the linearity and noise of the op amp
• Gain limits at the op amp open loop gain
Stability:
To keep the loop stable, it is important to pick the loop filter so that it does not
introduce more than a 90° phase shift in the loop.
Phase Signals
It is important to remember that frequency and phase are related as
d
dt = → = ·dt
Transfer functions:
V2(s)
H(s) = V (s)
1
where V2(s) and V1(s) are the Laplace transforms of v2(t) and v1(t).
Next, we consider some simple phase signals that are used to excite a PLL.
DF
t
140419-04
Dw
t
140419-05
SUMMARY
• LPLL blocks are:
1.) Multiplying phase detector
2.) Low pass filter
3.) Voltage controlled oscillator
• Locked state: Input frequency = VCO frequency
The phase response is low pass
The phase error response is high pass
Circuits PLL
Perspective Components
Technology CMOS
Perspective Technology
140418-02
1 w
Ko
140419-07
z = 5.0
0
z = 3.0
|H(jw)| dB
-10
z = 0.3
z = 0.5
-20
z = 0.707
z = 1.0
-30
-40
0.1 1 w 10 100
wn 140419-10
e(t)
nt
e(t)
/n
nt
Example 1 - Continued
where Kv = KoKd. Solving for the phase transfer function gives,
out(s) out(s) Kv/
= =
in(s) in(s) s2 + (s/) + (Kv/)
Kv 1
Therefore, n = = 89.148 krads/sec. = = 0.701 and H =1
2 Kv
(b.) The frequency response can be written as
1
nt sinn 1-2t + µ(t)
fout(t) = 200kHz 1 - e-
1-2
Setting fout(ts) = 200kHz – 100Hz, gives
1
nts sinn 1-2ts +
200x103-100 = 200x1031 - e-
1-2
This equation simplifies to the following assuming the value of the sin (x) is 1.
100Hz 1 1
= e-nts sinn 1-2ts + e-nts
200kHz 1-2 1-2
2000 1
e n t s = = 2800 → ts = ln(2800) = 2( 7.9375) = 127µsec.
1-2 n
· t2 ·
However, 1(t) = 2 → 1(s) = s3
Phase error:
· · s2 ·
e(s) = He(s) s3 = 3 2 =
s (s + 2ns + n2) s(s2 + 2ns + n2)
· ·
e(t) = L-1[e(s)] = 2 - 2 cos 1- 2nt + sin 1- 2 t e-nt , < 1
n n 1+2 n
· ·
= 2 - 2 (1 + nt)e-nt, =1
n n
· ·
2-1 t e-nt , > 1
= 2 - 2 cosh 2-1nt + sinh
n n 2-1
n
Steady state error:
· · t ·
e() = lim se(s) = 2 (High loop gain) e() = K K F(0)2 + 2 (Low loop gain)
s→0 n d o n
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 02 – (8/9/18) Page 2-17
e(t)
· /n2
nt
First-Order PLL
A first–order PLL occurs when F(s) = 1. From previous results we have,
2(s) KoKd s
H(s) = = Also, He(s) = 1- H(s) = s + K K
1(s) s + KoKd o d
The –3dB bandwidth of H(s) is KoKd.
Comments:
• F(s) causes the –3dB bandwidth to be reduced in higher-order systems which means
that the first-order PLL has a wider bandwidth
• The hold range of the first-order PLL will be larger than for higher-order PLLs
• The first-order PLL will track the signal variations more quickly than higher-order
PLLs
• The first-order PLL does not suppress noise superimposed on the input signal to the
extent of higher-order PLLs.
Higher-Order PLLs
Comments:
• Generally F(s) has a pole and a zero in order to get better noise rejection without
sacrificing speed.
Open Loop Gain
-20dB/decade
-40dB/decade First-order
Bandwidth
0 dB
Pole Zero
-20dB/decade
140420-02
• If the phase shift of the open loop system is more than 90°, the stability of the loop may
be poor ( is small).
PLL shown with the buffer outside of the PLL loop. Give an approximate sketch for
magnitude response of out(j)/Vb,n assuming = 0.707.
Buffer Vb,n
(b.) Find the output phase, out(s), as a qin LPF VCO
+
+ 1 Ko + qout
function of the input phase in(s) and - st+1 s +1
Example 2 – Continued
Solution
Ko KvF(s)
(a.) out’(s) = s F(s)Kd[in(s)-out’(s)] → out’(s) = s+K F(s) in(s)
v
Substituting for F(s) gives
Kv/ n2
out’(s) = 2 (s) = 2 (s)
s +(s/)+(Kv/) in s +s2n+n2 in
n2
out(s) = out’(s) + Vb,n = 2 (s) + Vb,n(s)
s +s2n+n2 in
Ko
(b.) out(s) = s F(s)Kd[in(s)-out(s)] + Vb,n(s)
KvF(s) KvF(s)
out(s)1+ s = s in(s) + Vb,n(s)
KvF(s) s Kv/ s2+(s/)
out(s)=s+K F(s) in(s)+s+K F(s) Vb,n(s) = 2 in(s)+ 2 Vb,n(s)
v v s +(s/)+(K v/ ) s +(s/)+(K v/ )
n2 s2+s2n
out(s) = 2 (s) + 2 V (s)
s +s2n+n2 in s +s2n+n2 b,n
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 02 – (8/9/18) Page 2-23
Example 2 - Continued
The sketch for both parts (a.) and (b.) is shown below.
20
Noise from buffer (part a.)
0
Closed-loop response
-20
dB
Noise from buffer (part b.)
-40
-60
-80
0.01 0.1 1 10 100
w/wn 140420-05
(c.) Obviously, part (b.) leads to an output spectrum with less noise. Part (a.) has the
same noise contribution from the buffer regardless of the frequency. If the input is noisy
then it will have a spectrum shown above similar to the closed-loop response. When the
input noise is larger than the buffer noise, there is not much difference between the two
architectures.
SUMMARY
• Unlocked state:
- Hold range (H) – frequency range over which a PLL can statically maintain phase
- Pull-in range (P) - frequency range within which a PLL will always lock
- Pull-out range (PO) – dynamic limit for stable operation of a PLL
- Lock range (L) - frequency range within which a PLL locks within one single-beat
note between reference frequency and output frequency
• The order of a PLL is equal to the number of poles in the open-loop PLL transfer
function
Circuits PLL
Perspective Components
Technology CMOS
Perspective Technology
140418-02
The mathematics behind the unlocked state are beyond the scope of this presentation. In
the section we will attempt to answer the following questions from an intuitive
viewpoint:
1.) Under what conditions will the LPLL become locked?
2.) How much time does the lock-in process require?
3.) Under what conditions will the LPLL lose lock?
2.) The pull-in range (P) is the range within which an LPLL will always become
locked, but the process can be rather slow.
Pull-in Range
w
wo-DwP wo wo+DwP 140421-02
3.) The pull-out range (PO) is the dynamic limit for stable operation of a PLL. If
tracking is lost within this range, an LPLL normally will lock again, but this process can
be slow.
(Dynamic Limits of Stability)
w
wo-DwPO wo wo+DwPO 140421-03
4.) The lock range (L) is the frequency range within which a PLL locks within one
single-beat note between reference frequency and output frequency. Normally, the
operating frequency range of an LPLL is restricted to the lock range.
Lock Range
w
wo-DwL wo wo+DwL 140421-04
Hold-in Range
Pull-in Range
wo-DwH wo-DwP
win
wo+DwP wo+DwH
140421-05
The following pages will attempt to relate the key parameters of hold range, pull-in
range, pull-out range, and lock range to the time constants, 1 and 2 and the gain factors
Kd, Ko, and Ka.
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 03 – (8/9/18) Page 3-5
w2(t)
t
140421-06
Note: No locking occurs in the above illustration because > KoKd |F(j)|.
t
140421-07
w2
wo
w2(t)
140421-08
t
Since min is less than max, the frequency of the positive going sinusoid is less than
the frequency of the negative going sinusoid. As a consequence, the average value of the
__
VCO output, 2 , “pulls” toward 1.
wo w2(t)
Pull-in Time, TP t
140421-09
Type of Filter P (Low Loop Gains) P (High Loop Gains) Pull-In Time, TP
Passive Lag
4
2nKoKd - n
4 2 2 o2
2 nKoKd = 16
n3
Active Lag 4 n2 4 2 2 o2Ka
2nKoKd - K nKoKd = 16
a n3
Active PI Lag → → 2 o2
= 16
n3
† R.M. Best, Phase-Locked Loops – Design, Simulation, and Applications, 4th ed., McGraw-Hill Book Co., 1999, Appendix A.
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 03 – (8/9/18) Page 3-12
Example 3
A second-order PLL having a passive lag loop filter is assumed to operate at a center
frequency, fo, of 100kHz and has a natural frequency, fn, of 3 Hz which is a very narrow
band system. If = 0.7 and the loop gain, KoKd = 2·1000 sec.-1, find the lock-in time,
TL, and the pull-in time, TP, for an initial frequency offset of 30 Hz.
Solution
1 1
TL = = 0.333 secs.
fn 3
2 o2 44 fo2 302
TP = = = = 4.675 secs.
16 n3 16·83 fn3 32(0.7)33
Comments:
• Note that for the active PI filter, N = 1.
• For N >1, it becomes difficult to maintain stability.
• In most cases, P(s) is a first-order polynomial and Q(s) is a polynomial of order 0 or 1.
To find the steady-state error, the input, (s) must be known. We will consider several
inputs on the following slide.
qn1(t)
v1(t)+qn1(t)
t
Phase error
140421-10
Attenuation (dB)
Input
Signal Phase Output
Filter
Bi Detector
f
fo
VCO
140421-11
Spectral Power
frequency spectrum:
Density
Power spectra of the reference Area = P n Bi
= WiBi
signal, v1(t), and the superimposed Wi
noise signal, vn(t). fo Frequency
qn1(jw)2
Area = vn12
n2
2 = 2 (j2f) df
n2
0
where n2
2 is the area under the output phase noise plot in a previous slide.
n2
2
=
|H(j)|2df = |H(j)| d
2
0
2 0
Rsmall(not locked)
v1(t) Phase vf(t)
Detector
Rlarge(locked)
v2(t)
Switched Loop Filter
VCO
140421-14
In the unlocked state, the filter bandwidth is large so that lock range exceeds the
frequency range within which the input is expected.
In the locked state, the filter bandwidth is reduced in order to reduce the noise.
SUMMARY
• Acquisition process – the PLL in the unlocked state
• Influence of noise on the linear PLL
• Pull-in techniques for noisy signals
Circuits PLL
Perspective Components
Technology CMOS
Perspective Technology
140418-02
wo
w2min
6.) Determine the value of Kd from the
vf
data sheet. Kd will depend upon the 140421-15
vf(min) VB vf(max) VB
2
signal level. It is preferred to have a
large value of Kd.
7.) Determine the natural frequency, n.
a.) Lock range has been specified in step 3.).
L
n =
2
b.) Noise bandwidth has been specified in step 3.)
2BL
n =
+ 0.25
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 04 – (8/9/18) Page 4-4
Frequency Spectrum
S1 E1 Channel Channel Channel Channel
1 2 3 N
Bi
S2 E2
f
f01 f02 f03 f0N
300 Hz 2DwLmin 3 kHz
SN EN 140421-16
2DwLmax
Each transmitter is to transmit a binary signal with a baud rate of 50 bits/sec. The signal
is encoded in a non-return to zero format which means that the bandwidth required is half
the baud rate or 25 Hz. The spectrum of the FM-modulated carrier consists of the carrier
frequency and a number of sidebands displaced by ±25 Hz, ±2·25 Hz, etc. from the
carrier frequency.
Assuming that a narrow-band FM is used, the channel spacing will be selected as 60
Hz. The channel is assumed to be an ordinary telephone cable with a bandwidth of 300
Hz to 3000 Hz giving Bi = 2700 Hz. Therefore, the maximum number of channels is
Max. no. of channels = Bi/Channel spacing = 2700/60 = 45 channels.
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 04 – (8/9/18) Page 4-6
† Phase-Locked Loop Data Book, Exar Integrated Systems, Sunnyvale, CA, 1981.( http://www.exar.com/products/XR215A.html)
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 04 – (8/9/18) Page 4-7
60 Phase
Margin
40 79°
20
Magnitude
0 Cutoff
-20 Frequency
5Hz
-40
0.01 0.1 1 10 100
Frequency (Hz)
Cutoff frequency 5Hz
Phase margin 79°
C= C=
Phase 18.1 mF 18.1 mF Rx
Co = 0.27mF
Detector R2 = 5kW
R2 =
Outputs 3.35kW Timing
3.35kW
Capacitor
VCC
Phase 16 2 3
+15V Detector Range 10 13 14
vd(t)
mV
vf(t)
vf(t)
V
SUMMARY
• LPLL design –Design the parameters Ko, Kd, , and the filter F(s) of the LPLL for a
given performance specification
• PLL system simulation methods include:
1.) Behavioral modeling of PSPICE (illustrated on the previous example)
2.) PC-based simulator developed by R.M. Best
¸N Counter
(Optional)
Fig. 2.2-01
• The only digital block is the phase detector and the remaining blocks are similar to the
LPLL
• The divide by N counter is used in frequency synthesizer applications.
2
2’ = 1 = N → 2 = N 1
VOH VOH
VIH
VIL
t vin t
VIL VIH
VOL VOL
Fig. 2.2-02
v1 v1 v2’ vd
vd
G1 0 0 0
v2' 0 1 1
Fig. 2.2-03
1 0 1
1 1 0
Zero Phase v1
Error: t
v2'
t
vd
vd
t Fig. 2.2-04
Positive Phase v1
Error: t
v2'
t
qe>0
vd
vd
t Fig. 2.2-05
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 05 – (8/9/18) Page 5-5
-p V -V
2
p qe Kd = OHp OL
-p p
2
If v1 and v2’ are asymmetrical (have different duty cycles), then vd becomes,
vd
VOH
v1
t
-p
2
v2' -p p p qe
t 2
vd
vd
VOL Fig. 2.2-07
t
The effect of waveform asymmetry is to reduce the loop gain of the DPLL and also
results in a smaller lock range, pull-in range, etc.
JK Flip-Flop
The JK Flip-Flop is not sensitive to waveform asymmetry because it is edge-triggered.
v1 v2’ Qn+1
vd 0 0 Qn
v1 J Q
FF 0 1 0
v2' K Q 1 0 1
Zero Phase Error (Assume Fig. 2.2-08 rising edge 1 1
triggered): Qn
v1
t
v2'
t
vd
vd
t Fig. 2.2-09
Positive Phase Error:
v1
t
v2'
qe>0 t
vd
vd
t Fig. 2.2-10
V -VOL
qe Kd = OH
-p p 2p
VOL
Fig. 2.2-11
Comments:
• Symmetry of v1 and v2’ is unimportant
• Both the EXOR and the JK flip-flop have a severely limited pull-in range if the loop
filter does not have a pole at zero.
Phase Feedback
Vin Vout
wout VCO
win
Frequency Feedback
FD LPF2 VLPF2
The output signal of the PFD depends on the phase error in the locked state and on the
frequency error in the unlocked state.
Consequently, the PFD will lock under any condition, irrespective of the type of loop
filter used.
A B Fig. 2.2-13A
Unlike the EXOR gates and the R-S latches, the PFD generates two outputs which are not
complementary.
Illustration of a PFD
PFD (A = B):
A QA
PFD
B QB
(Rising edge triggered) Fig. 2.2-14
fA>fB: fA<fB:
A A
B B
QA QA
QB QB
Time Time
Fig. 2.2-15
wA<wB: wA>wB:
A A
B B
QA QA
QB
QB
Time Time
Fig. 2.2-16
PFD – Continued
Plot of the PFD output versus phase error:
vd
VOH
-4p -2p
qe
2p 4p
VOL
Fig. 2.2-17
When e exceeds ±2, the PFD behaves as if the phase error recycled at zero.
VOH-VOL
Kd =
4
Average Duty Cycle
A plot of the averaged duty cycle of vd
versus 1/2’ (A/B) in the unlocked 1
Fraction of time state
of the DPLL: 0.5 QA=1 and QB=0
(+1 state)
0
w1/w2'
1 2
Fraction of time
-0.5 QA=0 and QB=1
(-1 state)
-1
Fig. 2.2-18
CHARGE PUMPS
What is a Charge Pump?
A charge pump consists of two switched current sources controlled by QA and QB
which drive a capacitor or a combination of a resistor and a capacitor to form a filter for
the PLL with a pole at the origin.
A
VDD
I1 B
QA X QA and QB are
A S1 Vout QA simultaneously
PFD QB high for the
B S2 QB duration given
Y Cp
I1=I2=I by the delay
Vout
I2 of the AND gate
and the reset path
of the flip-flops.
t Fig. 2.2-19
A > B or A = B but A > B: S1 is on and Vout increases.
A < B or A = B but A < B: S2 is on and Vout decreases.
A Charge-Pump PLL
Block diagram:
VDD
I1
x(t) QA
S1 y(t)
PFD VCO
QB
S2
Cp
I2
Fig. 2.2-20
The charge pump and capacitor Cp serve as the loop filter for the PLL.
The charge pump can provide infinite gain for a static phase shift.
QB qe
Ip e Ip T1e
Amount of vd(t) increase per period (T1) = C x =
p 2/T1 2Cp
Ip T1e 1 Ip e
Average slope per period = x =
2Cp T1 2Cp
Ip
vd(t) = Average Slope· = · µ(t)
2Cp e
Ip e Ip V
Taking the Laplace transform gives, Vd(s) = → Kd =
2Cp s 2Cp rads
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 05 – (8/9/18) Page 5-16
Optional
¸ N Counter
1
N Fig. 2.2-25
Loop Filters
1.) Passive lag-
1 + s2
PD → F(s) =
1 + s(1 + 2)
1 + s2
PFD → F(s)
s(1 + 2)
Experimental results using the PFD with a passive lag filter show that the gain of the
passive filter is not constant. As a result, the filter dynamics become nonlinear.
2.) Active lag-
1 + s2
PD → F(s) = Ka
1 + s1
1 + s2
PFD → F(s)
s1
3.) Active PI-
1 + s2
PD or PFD → F(s) =
s1
v2'
vd
VDD
vd High
0.5VDD Impedance
State
0 t
vd(eq.)
VDD
0.5VDD
(If the filter time constant >> the duty cycle, this waveform simplifies the analysis.)
0 t
vf
VDD
w1
Dw
Ko
0.5VDD
t
TP Fig. 2.2-30
vd(eq.) is a 50% duty cycle model of the PFD to find Tp.
3.) PFD:
1-2
po = 2n exp tan
-1 , < 1
1- 2
po = 2ne, =1
1-2 po n( + 0.5) for all
po = 2n exp tanh
-1 , > 1
1- 2
detector is a simple CMOS EXOR whose logic levels qin EXOR R1 - qout
VCO
are ground and VDD = 5V, 2.) both the input to the loop PD +
and the VCO output are square waves that swing VDD
2 Part (d.)
between ground andVDD, and 3.) that the VCO has a
¸N
perfectly linear relationship between the control SU03E1P2
voltage and output frequency of 10 MHz/V. The polarities are such that an increase in
control voltage causes an increase in the VCO frequency.
(a.) Derive the expression for the open-loop transmission and the transfer function
out(s)/in(s).
(b.) Initially assume R2 = 0 and R1 = 10k What value of C gives a loop crossover
frequency of 100kHz? What is the phase margin? Assume the op amp is ideal.
(c.) With the value of C from part (b.), what value of R2 will provide a phase margin of
45° while preserving a 100 kHz crossover frequency?
(d.) Now assume that a frequency divider of factor N is inserted into the feedback path.
With the component values of part (c.), what is the largest value of N that can be tolerated
without shrinking the phase margin below 14°?
Example 1 - Continued
Solution
Ko out(s) 5Ko out(s)
(a.) out(s) = F(s)Kd in(s) +
= F(s) in(s) +
s N s N
5V R2+(1/sC) sR2C+1 s2+1
Kd = and F(s) = - sR C = - sR C = - , 1 = R1C and 2 = R2C
1 1 s1
5Ko s2+1 out(s) 5Kos2+1 5Kos2+1
out(s)= - (s) +
N → out(s) 1 + sN s1 = s s1 in(s)
s s1 in
5Ko
- (s2+1)
out(s) 1
=
in(s) 5Ko 2 5Ko
2
s+ s+
N 1 N1
5Ko
- (s2+1)
out(s) 1 5Kos2+1
= and the loop gain = LG = -
in(s) 5Ko 2 5Ko sN s1
s2+ s+
N 1 N1
Assume N = 1 to get the answer to part (a.).
Example 1 - Continued
(b.) With R2 = 0, 2 = 0 so that the loop gain becomes,
5Ko 5·2x107 108 108
LG = - 2 = 2 = 2 = 1 → 1 = = 253.3µsec.
s 1N s 1 c 1 (2·105)2
1 = R1C → 253.3µsec. = 10kC → C = 25.3nF
The phase margin is 0°.
(c.) The phase margin is totally due to 2. It is written as,
1 1
PM = tan-1(c2) = 45° → c2 = 1 → 2 = = = 1.5915µs = R2C
c 2x105
1
R2 = = 62.83
2x10525.3x10-9
(d.) N does not influence the phase shift so we can write,
tan-1(c2) = 14° → c’2 = 0.2493 → c’ = 0.2493c = 156,657 rads/sec.
Now the loop gain at c’ must be unity.
5Ko (c’2)2+1 5Ko
LG = - =1 → N= (c’2)2+1
c’N c’1 (c’) 1
2
108
N= 2 -6 (0.2493)2+1 = 16.58 = 16
(156.657krads/sec.) 253.3x10
SUMMARY
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 05 – (8/9/18) Page 5-30
• The DPLL has a digital phase detector and the remainder of the blocks are analog
• Digital phase detectors
- EXOR Gate
- JK Flip-Flop
- Phase-Frequency Detector
• Charge pump – a filter implementation using currents sources and a capacitor that
works with the PFD
• Charge pumps implement a pole at the origin to result in zero phase error
vd vd is proportional
100% to the phase noise.
\ LPLL noise theory
50%
» DPLL noise theory.
0% t Fig. 2.2-32
t
W
C1A C1B
v1(10kHz) SIGin 74HC4046A Data N
PC1
COMPin (EXOR) P0····P7 PE
v2'
PC2 VCOout 74HC40102 TC v '
(PFD) PCPout VCO (40103) 2
CP
PC3
TE PL MR
(JK)
PC2out VCOin R1 R2
+5V
R1 = R2=
R1=567W
47kW 130kW
R2=1.35kW
C = 0.33mF
Fig. 2.2-36
Optional
¸ N Counter
1
N Fig. 2.2-25
† R.E. Best, “Phase-Locked Loops – Design, Simulation, and Applications,” 4th Ed., McGraw-Hill, NY, p. 103
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 06 – (8/9/18) Page 6-17
40
LG Phase
20
Phase |LG|
0 Margin
Note that the phase is very » 84°
-20 wc
close to 0° and |LG|>>1 at
low frequencies which is -40
10 100 1000 10 4 10 5
typical of type II systems. Frequency (Hz)
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 06 – (8/9/18) Page 6-18
† Roland E. Best, Phase-Locked Loops – Design, Simulation, and Applications, 4th ed., McGraw-Hill Book Co., 1999, New York, NY
CMOS Phase Locked Loops © P.E. Allen - 2018
Lecture 06 – (8/9/18) Page 6-19
vd
vd(mV)
vf
t(µs)
Case 2 - f = 8000Hz
vd
Phase error
vd(V) ≥ 90° vf
vf
vf
vd
vd
vd(V)
vf
4.5
4.0
3.5 vd
3.0
vd(V) 2.5
2.0
1.5
1.0
0.5
vf
vd(V) vd
Tp 1.5ms
3.0
vd(V) 2.5
2.0
vd
1.5
1.0
0.5
Tp 5ms
SUMMARY
• Illustrated the Noise Performance of the DPLL
• Presented a DPLL Design Procedure
• Showed how to do DPLL System Simulation
• The DPLL is much more compatible with IC technology and is the primary form of
PLL used for frequency synthesizers
fp ≈ 10.051MHz
SPICE Simulation:
Homework H01P1 - Crystal Impedance
IIN 0 1 AC 1.0
CP 1 0 6PF
CS 1 2 30FF
LS 2 3 8.4MH
RS 3 0 5.3OHM
RBIG 1 0 1GOHM
.AC LIN 101 9.5MEG 10.5MEG
.PRINT AC V(1)
.PROBE
.END
100kΩ
Impedance
10kΩ
1kΩ
100Ω
9.6 9.8 10 10.2 10.4
Frequency (MHz) SU03H01S1
PLL Problems and Solutions (9/6/03) Page 2
Problem 2
A simple, doubly balanced passive CMOS mixer is shown along with the local oscillator
waveform, vOL(t). Assume that vRF(t) = ARFcos(ωRFt) and vLO(t) is the waveform shown
below. (a.) Find the mixer gain, Gc, in dB if the switches are ideal. (b.) Find the mixer gain in
dB if the switches have an ON resistance of Rs/2.
vLO(t)
Rs vLO(t) vLO(t) Switch
2 ON
vRF(t) +vIF(t) -
0 t
Rs
vLO(t) vLO(t)
Rs Switch
2 OFF 1
fLO F99E2P1
Solution
Assume the switches have an ON resistance of RON and work both parts (a) and (b)
simultaneously. Also, The equation for vIF(t) can be written as,
Rs
vIF(t) = 2R +2R vRF(t) · sgn[vLO(t)]
s ON
Rs 4 4
VIF(jω) = 2R +2R ARFcos(ωRFt) · cos( ω LO t) + cos (3 ω LO t) + ···
s ON π 3π
Rs 4ARF
∴ VIF(jω) ≈ 2R +2R cos(ωRFt) · cos(ωLOt)
s ON π
Rs 2ARF
= 2R +2R cos[ωRF -ωLO)t]
s ON π
The conversion gain in general is written as
|VIF| Rs 2
Gc = |V | = 2R +2R
RF s ON π
1 1
(a.) For RON = 0, Gc = → Gc = = -9.943dB
π π
2 2
(b.) For RON = 0.5Rs, G c = → Gc = = -13.465dB
3π 3π
PLL Problems and Solutions (9/6/03) Page 3
Problem 3
Use SPICE to demonstrate that the
following circuit is a frequency doubler. 2V
If vin(t) is a sinusoid of 10kHz and 1.5V M1 M2
peak, show vin(t) and vout(t) as a 10µm
function of time. The model parameters 1µm
of the MOSFETS are K N ’ = 110µA/V2, vin(t) vin(t)
vout(t)
VTN = 0.7V, and λN = 0.04V-1.
Solution 100kΩ
The results of this problem are below. -2V SU03H01P3
SPICE Input File:
Homework H01P3 - Frequency Doubler
VIN 1 0 DC 0.0 SIN(0 1.5 10KHz)
EVIN 0 2 1 0 1.0
VDD 4 0 DC 2.0
VSS 5 0 DC -2.0
M1 4 1 3 3 NMOS1 W=10U L=1U
M2 4 2 3 3 NMOS1 W=10U L=1U
RTAIL 3 5 100K
.MODEL NMOS1 NMOS VTO=0.7 KP=110U LAMBDA=0.04
.OP
.TRAN (10U 1000U)
.PRINT TRAN V(1) V(2) V(3)
.PROBE
.END
Output Plots:
1.5V
vin(t)
1.0V
vout(t)
0.5V
-0.5V
-1.0V
-1.5V
0 200 400 600 800 1000
SU03H01S3
Time (µsec.)
PLL Problems and Solutions (9/6/03) Page 4
Problem 4
An 10nH inductor has a Q of 5 and is used to create a tank circuit with a 10pF capacitor.
Assume the capacitor is ideal. (a.) What is the resonant frequency of this circuit? (b.) What
value of parallel negative resistance should be used to create an oscillator? (c.) If C is changed
to 20 pF, what is the new value of the parallel negative resistance?
Solution
C = 10pF:
1 26
Lp = 1+ 2 = 25 ·10nH = 10.4nH
Q
1 1
ωo = = = 3.1623x109 radians/sec.
L pC 10.4nH·10pF
ωoLs ω oL s
Q= R → Rs = Q = 6.201Ω
s
∴ Rp = (1+Q2)Rs = 26·6.201Ω = 161.245Ω
C = 20pF:
1 1
ωo = = = 2.1926x109 radians/sec.
L pC 10.4nH·20pF
ωoLs ω oL s
Q= R → Rs = Q = 4.3853Ω
s
∴ Rp = (1+Q2)Rs = 26·4.3853Ω = 114.017Ω
Problem 5
Give a block diagram of simple brute-force coherent direct synthesizer that will generate 1.75f
from f. The input frequency f is to vary from 12 MHz to 15MHz. Since f is variable, you
cannot use frequency multipliers (integer frequency dividers and mixers are allowed) in your
design. A simple design will receive more credit. What other frequencies will be present at the
output?
Solution
Approach: fout = fxf – f/4 = 1.75f
f 2f
2f±0.25f
f
f/4
÷ 4
S03H01S5
The frequency 2.25f will also be present at the output.
PLL Problems and Solutions (9/6/03) Page 5
Problem 6
A phase-locked loop has a center frequency of 105 rads/s, a Ko of 103 rad/V-s, and a Kd of 1
V/rad. Assume there is no other gain in the loop. Determine the loop bandwidth in the first-
order loop configuration. Determine the single-pole, loop-filter pole location to give the closed-
loop poles located on 45° radials from the origin of the complex frequency plane.
Solution
103
The loop bandwidth = Kv = KoKp = s
In order to produce poles at 45° to the axis, we add a loop filter pole at ω1 where
ω1 = 2Kv = 2000 rads/sec.
The filter transfer function becomes,
ω1 2000
F(s) = = s+2000
s+ω1
Problem 7
For the same PLL of the previous problem, design a loop filter with a zero that gives a crossover
frequency for the loop gain of 100 rads/sec. The loop phase shift at the loop crossover
frequency should be –135°.
Solution
A plot of the desired loop gain is shown below.
-20dB/dec.
Loop Gain (dB)
40dB
20dB
-40dB/dec.
0dB ω1 = ω(rads/sec)
ω2 = Kv = 1000
10 100 SU03H02S2
If ω2 (the zero frequency) is at the unity gain point, then the loop phase shift will be
–135° at this point. Therefore, we require that ω2 = 100 radians/sec.. If ω1 = 10 radians/sec.,
the requirement will be satisfied as shown in the above plot.
The design of the filter becomes, R1
1 1
ω2 = R C and ω1 = (R +R )C
2 1 2 C
ω2 R1 R2
∴ = 1 + R = 10 → R1 = 9R2
ω1 2
Problem 8
Estimate the capture range of the PLL of the previous problem assuming that it is not artificially
limited by the VCO frequency range.
Solution
For capture we need
π
|(ωi - ωo)| < 2 Kv|F(j(ωi - ωo)|
If we assume that
π
|(ωi - ωo)| = 2 Kv|F(j(ωi - ωo)|
|F(jω)|
1 10 100
1 ω
0.1
SU03H02S3
From this figure we can solve the above equation to find that
(ωi - ωo) = 157 rads/sec.
which is the capture range.
PLL Problems and Solutions (9/6/03) Page 7
Problem 9
A filter for a phase locked loop is specified as C=10pF
10ω1 1,000,000 R R
|F(s)| = = s+100,000 R = 10kΩ
s+ω1
R2 Vout
and must be implemented on a CMOS chip using Vin R R
resistors no larger than 10kΩ and capacitors no larger _
than 10pF. Using the circuit shown, find the values R1
of R1 and R2 that will satisfy the component value +
constraints. SU03H02P4
Solution
Find the currents i1 and i2, C=10pF
v in R R 1 vi n v in R R
i1 = = = R = 10kΩ
RR 1 R+R 1 2RR + R 2 R T1
R + R+R 1 i2
1 R2 Vout
and Vin R R
vout R _
i2 = RR 1 R+R 1 + sCvout R1 i1
R + R+R +
1
R 2 v out vout SU03H02S4
= + sCv out R T2 + sCvout
=
2RR + R 2
2
Solving for the sum of the currents flow toward the minus op amp input terminal gives,
v in vout vout RT2 1 1
+
R T1 R T2 + sCv out = 0 → v in = -R T1 sCRT2+1 = -10 s
+1
105
10-5
∴ CRT2 = 10-5 → RT2 = -11 = 106
10
2RR 2 + R 2 R2 6
R T2 = = 2R + = 20x10 3 + 100x10 = 106
R2 R2 R2
100x106
∴ R2 = = 100Ω
106-20x103
R T2 R2 100x106
R T1 = 10 = 105 → 2R + R = 20x103 + R = 105
1 1
100x106
∴ R1 = = 1000Ω
105-20x103
This problem shows how a clever circuit technique can make a filter suitable for integrated circuit
implementation.
PLL Problems and Solutions (9/6/03) Page 8
Problem 10
This homework is designed to provide practical inductor design experience for students. Use
ASITIC for the design and analysis. However, other tools are acceptable if they give all the
results including layout.
A 5GHz LC tank will be designed as a part of LC oscillator. C value is given as 1pF.
(a) Find L value. (b) Design and simulate a spiral inductor with this L value (± 5% range).
Optimize design parameters, W, S, D and N to get a high Q (Qmin = 5). Show L, Q, fSR value
obtained from simulation. (c) Show the layout. (d) Give a lumped circuit model with component
values.
Solution
(a) LC tank oscillation frequency is given as 5GHz.
1 1
L= 2 = 9 -12 = 1.01x10-9
ωosc ·C (2π·5x10 )(1x10 )
(b) One possible solution is
Parameters: W = 16um, S = 2um, D = 150um, N =2.5
Resulting inductor: L = 0.952nH, Q = 8.54, fSR = 19.35GHz @ 5GHz
(c) Layout
(d) Pi model from ASITIC is shown below. This is the analysis result from ‘pix’
command.
0.952nH 3.27
71.1fF 65.1fF
-1.3
3.03
PLL Problems and Solutions (9/6/03) Page 9
Problem 11
Assume an LPLL has F(s) =1 and the PLL parameters are Kd = 0.8V/radians, Ko = 100 MHz/V,
and the oscillation frequency, fosc = 500MHz. Sketch the average control voltage at the output of
the phase detector if the input frequency jumps from 500MHz to 550MHz.
Solution
Find the transfer function from the input frequency, fin, to the
output of the phase detector, vd. ω1 vd
K d
K dK o
Vd = Kd(θ1-θ2) = Kdθ1- s Vd ω2 Ko
s
ω1 SU03H03S1A
K dK o
Vd1+ s = Kdθ1 = Kd s
Vd Kd Kd Kd ∆ω1 k1 k2
∴ = s+K K → Vd(s) = s+K K ω1(s) = s+K K s = s + s+K K
ω1 d o d o d o d o
Kd∆ω1 Kd∆ω1
By partial fraction expansion we can show that k1 = - k2 = K K = K = 0.4V
d o v
Kd∆ω1 (V/rad)(rad/sec)
Note the units of K are 1/sec =V
v
and Kv = (2π·100MHz/V)(0.8V/rad.) = 502.65x106 (1/sec.)
Kd∆ω1
∴ vd(t) = K (1-e-Kvt) = 0.4(1-e-502.65x106t)
v
A plot of vd(t) is shown below.
0.400V
0.267V
vd(t)
0.133V
0V
0 2 4 6 8 10
Time (ns) Fig. SU03H03S1B
PLL Problems and Solutions (9/6/03) Page 10
Problem 12
A Type I PLL incorporates a VCO with Ko = 100MHz/V, a phase detector with Kd = 1V/rad,
and a first-order, lowpass filter with ωLPF = 2π x106 radians/s shown below. A divider of 100
has been placed in the feedback path to implement a frequency synthesizer. (a.) Find the value
of the natural damping frequency, ωn, and the damping factor, ζ, for the transfer function
φout(s)/φin(s), for this PLL. (b.) If a step input of ∆φin is applied at t = 0, what is the steady-
state phase error at the output of the phase detector, φe? The steady-state error is evaluated by
multiplying the desired phase by s and letting s→0.
Solution
Ko 1 K φ - φ out → φ 1+ K o Kd K o Kd
(a.) φout = s s d in N out sN s = s s φin
ω + 1 1 + + 1
LPF ωLPF ωLPF
φout(s) K oK d KoKdωLPF ωn2
∴ = s K oK d = =
φin(s) K o K d ω L P F s 2 + 2 ζω n s + ω n 2
s 1 + + 2
ωLPF N s + ω LPF s + N
KoKdωLPF 2πx106·2πx108
Thus, ωn2 = N = 100 = 4π2x1012 → ωn = 2πx106
Problem 13
Modify the active filter shown of Problem 9 to C2
design the lag-lead loop filter shown below. The
capacitors can be no larger than 10pF. Give the R R
values of R1, R2, C1 and C2. R = 10kΩ
|F(jω)| dB R2 Vout
Vin R R
10K 100K
0dB _
ω(rads/sec.) R1
+
SU03H03P3A
-20dB
S03H03P3
Solution
The transfer function corresponding to the above Bode plot is,
s
+1
105
F(s) = 1
+1
104 C1 C2
The modification of the filter is vd R R vc
shown where from Prob. 9,
2RR i+ R 2 - -
R Ti = RT1 RT2
Ri + +
The transfer function of this Loop Filter
filter is found as, SU03H03S2A
Vc(s)
F(s) = V (s) =
d
R T2 sR T1 C 1 +1
⇒ RT2 = RT1 = RT , RTC1 = 10-5 and RTC2 = 10-4
R T1 sRT2C2+1
We see if RT2 = RT1, then C2 = 10C1. Choosing C2 = 10pF gives C1 = 1pF. This gives
10-4 10-4
RT = C = -11 = 107
2 10
2RR i+ R 2 R2 100x106 100x106
RT = 3
= 2R + R = 20x10 + R 7
= 10 ⇒ R1 = 7 = 10.02Ω
Ri 1 1 10 -20x103
Therefore, R1 = R2 = 10.02Ω, C1 = 1pF and C2 = 10pF
The realization is completed by replacing each of the RT resistors with the following equivalent:
10.02Ω
SU03H03S3B
PLL Problems and Solutions (9/6/03) Page 12
Problem 14
Using the filter of Problem 13, find the value of ωn and ζ of the PLL if Kd = 1V/radians, K o =
2Mradians/V·sec. What is the steady state phase error in degrees if a frequency ramp of 109
radians/sec.2 is applied to the PLL?
Solution
Using the definition give in the notes for the time constants of the passive lag-lead filter we get,
s
+1
105 sτ2 + 1
F(s) = 1 = ⇒ τ2 = 10-5 sec. and τ1 = 9x10-5 sec.
+1 s( τ +
1 2τ ) + 1
104
K oK d 2x106
∴ ωn = = = 2 x105 = 141.4x103 radians/sec.
τ1+τ2 10-4
ωn 1 2x105 -5 1 1 1
ζ = 2 τ2 + K K = 2 10 + = 1 + 20 = 0.742
o d 2x106 2
Assuming the PLL has a high loop gain, then the steady-state phase error can be found as
∆ω· 109 1
θe(∞) = 2 = = 20 radians = 2.86°
ω n 2x1010
PLL Problems and Solutions (9/6/03) Page 13
Problem 15
Solve for the crossover frequency of the PLL of Problems 13 and 14 and find the phase margin.
Use SPICE to find the open-loop frequency response of the PLL and from your plot determine
the crossover frequency and phase margin and compare with your calculated values.
Solution
The crossover frequency can be found as,
60
|F(jω)| Phase
40 Margin
≈ 69°
20
0
-20
10 100 1000 104 105
SU03H03S5 Frequency (Hz) ωc ≈ 36kHz
The simulation results agree well with the calculated results.
PLL Problems and Solutions (9/6/03) Page 14
Problem 16
For the DPLL shown assume that N = 1000 and the –3dB bandwidth is 1000 Hz. (a.) Assume
that ζ = 0.2 and solve for the natural pole frequency, ωn, the filter time constant, τ = RC, and the
phase margin. (b.) Repeat part (a.) if ζ = 0.7. (c.) Repeat part (a.) if ζ = 1. Verify your
answers with PSPICE.
VDD
I1
v1(t) QA
S1 v2(t)
PFD VCO
v2'(t) QB
S2 C
R
I2
1/N
SU03H04P1
Solution
The filter output can be written as,
Kd Kd θ2
Vf(s) = s (sτ +1)(θ1-θ2’) = s (sτ +1)θ 1 + N where τ = RC
Ko K oK d θ 2 K v (s τ +1) K v (s τ +1)
θ2(s) = s Vf(s) = 2 (sτ +1)θ 1 + N = θ 1 (s) + θ2(s)
s s2 Ns 2
The closed-loop response is given as,
θ2(s) K v (s τ +1) K v (s τ +1)
= = 2
θ1(s) K vτ K v s + 2ω nζs + ω n2
2
s + N s + N
Kv 2ζ
∴ ωn = N and τ=
ωn
We know that the loop bandwidth, ω-3dB, can be expressed as
ω-3dB
ω-3dB = ωn 2 ζ 2 +1+ (2ζ2+1)2 +1 → ωn =
2 ζ 2 +1+ (2ζ 2+1)2 +1
ζ = 0.2:
ωn = 3933 rads/sec., τ = 102µs and PM = 0°+tan-1(2000π·102µs) = 32.6°
ζ = 0.7:
ωn = 3066 rads/sec., τ = 457µs and PM = 0°+tan-1(2000π·457µs) = 70.8°
ζ = 1:
ωn = 2531 rads/sec., τ = 790µs and PM = 0°+tan-1(2000π·790µs) = 78.6°
PLL Problems and Solutions (9/6/03) Page 15
Problem 16 – Continued
PSPICE Input File:
Homework4, Problem 1
VS 1 0 AC 1.0
R1 1 0 10K
ELPLL1 2 0 LAPLACE {V(1)}= {5044*5044*(1+102E-6*S)/(S+0.01)/(S+0.01)}
R2 2 0 10K
ELPLL2 3 0 LAPLACE {V(1)}= {3513*3513*(1+457E-6*S)/(S+0.01)/(S+0.01)}
R3 3 0 10K
ELPLL3 4 0 LAPLACE {V(1)}= {2531*2531*(1+790E-6*S)/(S+0.01)/(S+0.01)}
R4 4 0 10K
*Steady state AC analysis
.AC DEC 20 1 100K
.PRINT AC VDB(2) VP(2) VDB(3) VP(3) VDB(4) VP(4)
.PROBE
.END
Plot of Results:
100
ζ = 0.2 ζ=1
ζ = 0.7 ζ = 0.2
dB or Degrees
50
PM PM
ζ=1 = 72° = 79°
PM
= 32°
0
ωc
-50
1 10 100 1000 10 4 10 5
Frequency (Hz) SU03H04S1
PLL Problems and Solutions (9/6/03) Page 16
Problem 17
A type-I, second-order DPLL synthesizer is to be made with components having the following
values:
Ko = 4x108 rads/sec./V fref = 12.5 kHz Kd = 0.15 V/rad β = 2π
Design a type-I, second-order synthesizer having the following specifications:
1.) Output frequency range = 50MHz
2.) Lock range = 10MHz at the output
3.) Damping factor = 0.75.
Determine the components for the loop filter. Let C = 0.5µF. Make a sketch of your filter with
all components carefully labeled. Once your design is complete, determine the pull-in range in
Hz (at the output) and the lock time of your loop.
Solution
fout 50MHz K oK d 4x108·0.15
N = f = 12.5kHz = 4000 and Kv = N = 4000 = 15,000 sec.-1
r
∆ωH = βKvN = 2π·15x103·4000 = 377x106 rads/sec.
τ2 τ2 τ2 ∆ωL 62.8Mrads/sec 1
∆ωL = ∆ωH = 377 Mrads/sec. → = = =
τ1 τ1 τ1 ∆ωH 377Mrads/sec 6
∴ τ1 = 6τ2
1 1 1
ζ = 0.5 (1+τ2Kv) → 1.5 = (1+τ2Kv) → 2.25 = (1+τ2Kv)2
K vτ1 K vτ1 K vτ1
2.25·Kv(6τ2) = 13.5 Kvτ2 = 1 +2 Kvτ2 + (Kvτ2)2 → 0 = 1 – 11.5x + x2
11.5 1
where x = Kvτ2. Solving for x gives x = Kvτ2 = 2 ±2 11.5 2 - 4 = 0.0876
0.0876 5.84
∴ τ2 = 15,000 = 5.84µs = R2C = R2(0.5µF) → R2 = 0.5µF = 11.7Ω
35
τ1 = 6τ2 = 35µs = (R1+R2)C → R1+R2 = 0.5µF = 70.08Ω → R1 = 58.4Ω
Filter schematic:
R1 = 58.4Ω
R2 =
11.7Ω
C=
0.5µF
SU03H04S2
Kv
∆ωP = Nβ 2 2ζωnKvF(0) -ωn2 and ωn = = 20,702 rads/sec
τ1
∴ ∆ωP = 4000·2π 2 2·0.75·20,702·15,000 - (20,702)2 = 216.85 Mrads/sec.
∆ fP = 34.51 MHz
2π 6.283
TL = = 20,702 = 303.5µs
ωn
PLL Problems and Solutions (9/6/03) Page 17
Problem 18
Given the DPLL described by
1+5x10-6s
1+τ2s
Kd = 2.2 V/rad F(s) = =
1+τ1s 1+2x10-5s
fref = 12 kHz Ko = 25 MHz/V β = 2π Ν = 15,000
Determine the type number and order of the system and then find:
(a.) The output frequency in Hz.
(b.) The crossover frequency in Hz.
(c.) The noise bandwidth (Hz).
(d.) The closed-loop phase –3dB bandwidth in Hz
(e.) The steady-state phase error in response to a phase step of 0.1 radian.
(f.) The hold range (±Hz at the output).
(g.) The lock range (±Hz at the output).
(h.) The lock time.
(i.) The pull-in range (±Hz at the output)
(j.) The steady-state phase error in radians in response to a frequency step equal to the lock
range.
Solution
This is a type-I, second-order system. The closed loop transfer function is,
1+τ2s
K v
KvF(s) θ2 θ2 K vF(s) 1+τ1s Kv(1+τ2s)
θ2 = s θ 1 - N → = K v F(s) = =
θ1 1+τ2s K v (1+ τ 2 s)
s+ N s(1+ τ s) +
Kv 1 N
1+τ1s
s + N
Kv
(1+τ2s)
θ2 τ1 Kv(1+τ2s)
= Kv =
K v s2+2ζω s+ω 2
θ1 s
s 2 + 1 + N τ2 + n n
τ1 τ1N
Kv 2π·25x106·2.2
∴ ωn = = = 33.94 Krads/sec
N τ1 2x10-5·15,000
Problem 18 - Continued
ωn 1 33,940 1
(c.) Bn = 2 ζ + = 2 0.821 + = 19.1kHz
4ζ 4·0.821
Nωn N ω n
(d.) ω-3dB = ωn b + b2+1 where b = 2ζ2 + 1 - K 4 ζ - K
v v
33,940·15,000 33,940·15,000
b = 2(0.821)2 + 1 - 4·0.821 - = -0.320
2π·25x106·2.2 2π·25x106·2.2
Problem 19
Construct an accurate Bode plot of the synthesizer in Problem 18. Use this Bode plot to
determine the phase margin.
Solution
PSPICE was used to solve this problem. The input file and the results are shown below.
100
dB or Degrees
50 Phase
Margin
= 72°
-50
10 100 1000 10 4 10 5
Frequency (Hz) SU03H04S4
PLL Problems and Solutions (9/6/03) Page 20
Problem 20
Write the transfer functions giving: (1) The VCO phase noise in the output, (2) the reference
oscillator phase noise in the output. Use the literal form of the equations. The phase noise of the
VCO used in the synthesizer of Problem 3 is shown below. Make an accurate plot of the VCO
phase noise in the output of the synthesizer.
-50
SSB Phase Noise (dBc/Hz)
-100
-150
-200
10 100 1000 10 4 10 5 10 6 10 7
Frequency Offset from Carrier (Hz) SU03H04P6
Solution
The following block diagram will be Phase Detector
used to find the phase noise in the +
output due to the VCO phase noise. θr θe ωo
K d F(s) Ko
K v F(s) θ o -
θo = θ o,n - sN
θo' θo,n
1 θo + + 1
θo s s
= N
θ o,n K vF(s) SU03H04P5
s+ N
θo s s(1+τ1s)
= =
θ o,n K v 1+τ2s K vτ2 K v
s+ N s 2 τ 1 + s 1+ N + N
1+τ1s
From Problem 3 of this assignment we get,
Kv
(1+τ2s)
θo 1 τ1
= N Kv Kv
θr,n s
s 2 + 1 + N τ2 +
τ1 τ1N
PLL Problems and Solutions (9/6/03) Page 21
Problem 20 – Continued
The following PSPICE input file gives the results plotted below.
Homework 4, Problem 5 -In/Out VCO Phase Noise, Transfer Function
.PARAM N=15000, KVCO=157.1E6, T1=2E-5, T2=5E-6, KD=2.2, E=0.001
*Input Phase Noise
vphasenoise 1 0 ac 1.0
R1 1 0 10k
EPN 2 0 freq {v(1)} = (1,-40,0) (10,-70,0) (100,-100,0)
+(1E5,-160,0) (1E6,-160,0)
RPN 2 0 10k
*VCO Noise Transfer Function
EDPLL1 3 0 LAPLACE {V(1)}=
+{S*(T1*S+1)/(S*S*T1+KD*KVCO*T2/N*S+S+KD*KVCO/N)}
RDPLL1 3 0 10K
*VCO Noise at the Output
EDPLL2 4 0 LAPLACE {V(2)}=
+{S*(T1*S+1)/(S*S*T1+KD*KVCO*T2/N*S+S+KD*KVCO/N)}
RDPLL2 4 0 10K
*Reference Noise Transfer Function
EDPLL3 5 0 LAPLACE {V(1)}=
+{KD*KVCO*(1+T2*S)/(S*S*T1+S+KD*KVCO/N*S+KD*KVCO/N)/N}
RDPLL3 5 0 10K
*Reference Noise at the Output
EDPLL4 6 0 LAPLACE {V(2)}=
+{KD*KVCO*(1+T2*S)/(S*S*T1+S+KD*KVCO/N*S+KD*KVCO/N)/N}
RDPLL4 6 0 10K
*Steady state AC analysis
.AC DEC 20 1 1000K
.PRINT AC VDB(2) VDB(3) VDB(4) VDB(5) VDB(6)
.PROBE
.END
VCO Output Noise (and Reference Output Noise):
50
Reference Transfer Function VCO Transfer Function
0
-50
dB or dBc
Problem 21
Sketch the time variation and frequency spectrum of an RF signal with 75 percent amplitude
modulation. Show several cycles of the modulated wave. Make the modulation frequency 1/10
of the carrier frequency. The unmodulated carrier has a peak amplitude of 1.0V.
Solution
The expression for the general form of amplitude modulation is,
ωct
v(t) = 1.01 + m a cos 10 cos(ωct) = [1 + 0.75cos(0.1ωct)]cosωct
1.5
1
0.5
v(t) 0
-0.5
-1
-1.5
-2
0 π 2π 3π 4π 5π 6π 7π 8π
ωct Fig. SU03H05S1
Problem 22
The level of an SSB AM spur is observed to be –75 dBc. If the carrier has a peak amplitude of
1V, what is the variation of the carrier in ±V needed to produce the observed spur?
Solution
Vc = 0dBc
The observed spectrum is
Vc
ma m = -75dBc
SSB = 20 log10 2 → ma = 2·10SSB/20 2 a
f
fc SU03H05S2
∴ ma = 2·10-75/20 = 335.6x10-6
∆v
If Vpeak = 1V, then ma = V → ∆ v = 3.35.6µV
p
PLL Problems and Solutions (9/6/03) Page 23
Problem 23
A pair of 5 kHz PM/FM spurs appear on a 10 MHz carrier. The level of each spur is –50dBc.
(a.) What phase deviation in ±degrees is need to produce the spurs? (b.) What frequency
deviation in ±Hz is needed to produce the spurs?
Solution
(a.) The single sideband spurs can be expressed as,
θd β
SSB = 20 log10 2 = 20 log102
Problem 24
The carrier and spurs of Problem 3 above are passed through a frequency tripler. Make a sketch
of the output spectrum of the tripler. Label and show all important features of the spectrum.
Solution
After passing through a tripler, the SSB spur is increased by 20log10(3) or +9.54dB.
The resulting spectrum is shown as,
0dBc 0dBc
-50dBc+20log10(3) = -40.5dBc
5kHz -50dBc 5kHz
10MHz-5kHz x3 30MHz-5kHz
Frequency Frequency
-40.5dBc 10MHz 10MHz+5kHz 10MHz 30MHz+5kHz
-40.5dBc SU03H05S4
PLL Problems and Solutions (9/6/03) Page 24
Problem 25
A 100 MHz carrier having a –40 dBc upper sideband at 100.002 MHz and a –47 dBc lower
sideband at 99.998 MHz is passed through an ideal limiter followed by a bandpass filter centered
at 100 MHz with a 10 kHz total bandwidth. Make a sketch of the spectrum at the output of the
filter. Label all frequencies and amplitudes.
Bandpass
Ideal Limiter Filter
10kHz
100MHz
SU03H05P5
Solution
Asymmetrical sidebands imply the presence of both AM and FM as show below.
Carrier Carrier
AM FM
-40dBc
-47dBc SA SA SF
fc-fm
f f f f
fc-fm fc fc+fm fc fc-fm fc fc+fm fc fc+fm
SF SU03H05S5
Carrier
-51.16dBc
fc-fm
f
fc fc+fm
-51.16dBc SU03H05S5A
where fc = 100MHz and fm = 2kHz.
PLL Problems and Solutions (9/6/03) Page 25
Problem 26
An LC oscillator is shown. The value of the inductors, L, are 5nH VDD
and the capacitor, C, is 5pF. If the Q of each inductor is 5, find (a.)
the frequency of oscillation, (b.) the value of negative resistance that L L
should be available from the cross-coupled, source-coupled pair (M1 C
and M2) for oscillation and (c.) design the W/L ratios of M1 and M2
to realize this negative resistance.
Solution M1 M2
(a.) The equivalent circuit seen by the negative resistance circuit is:
2L 2Rs The frequency of oscillation is given as
1/ 2LC or ωo = 2πx109 radians/sec. 2mA
C
Therefore the series resistance, R s , is found F00E2P2
as
ωL 2πx109·5x10-9
F00E2S2A Rs = Q = 5 = 2π Ω
Converting the series impedance of 2L and 2Rs into a parallel impedance gives,
Problem 27
An LC oscillator is shown. Find an expression for the VDD
frequency of oscillation and the value of gmRL necessary for
oscillation. Assume that the output resistance of the FET, rds, RL
can be neglected. L Vout
Solution
An open-loop, small-signal model of this oscillator is shown M1 C1
below. C2
L F99E2P4
+ +
V' gmV' RL C2 Vo
- C2 -
F99P3S2
Writing a nodal equation at the output and input gives,
V o-V' Vo-V’
gm V’+G LV o+sC2V o+ sL = 0 and sL = sC2V’ → Vo = V’(1+s2LC2)
1 V’
∴ gmV’ + G L +sC 1 + sL (1+s2LC2)V’ - sL = 0
or (gm+GL-ω2LGLC2) + jω[C1-ω2LC1+C2] = 0
Therefore, the frequency of oscillation is,
C1+C2 1
ω osc = LC1C2 = LC1C2
C1+C2
C 2 C2
gm+GL = LGLC2ωosc2 = GL 1 + C → gmRL = C
1 1
PLL Problems and Solutions (9/6/03) Page 27
Problem 28
A Clapp oscillator which is a version of the Colpitt’s
oscillator is shown. Find an expression for the VDD
frequency of oscillation and the value of gmRL necessary RLarge C1
for oscillation. Assume that the output resistance of the
FET, rds, and RLarge can be neglected (approach
C2
infinity). VBias
L
Solution
C3 RL
The small-signal model for this problem is IBias
shown below.
The loop gain will be defined as Vgs/Vgs’. C2 F02FEP5
C1
Therefore, - +
V gs = Vgs
gmVgs' RL C3 L
-g m V gs' RL||(1/sC 3) 1
1 1 sC2
R L ||(1/sC 3 ) + sC + sC + s L F02FES5
1 2
RL(1/sC3) 1
-g m V gs ' R +(1/sC ) sC
L 3 2
= RL(1/sC3) 1 1
R L +(1/sC 3 ) + sC1+ sC2 + s L
-g m R L 1 -g m R L
Vgs sRLC3+1 sC2 sC2
T(s) = V = R 1 1 = 1 1
gs L
sR L C 3 + 1 + sC1+ sC2 + s L R L + (sR L C 3 +1) sC + sC + s L
1 2
-g m R L
T(s) = C2
sC 2 R L + (sR L C 3 +1)(s 2 LC 2 + C + 1 )
1
-g m R L
T(s) = C 2C 3 C2
sC 2 R L +s 3 R L C 3 LC 2 + sR L C + sC 3 R L + s 2 L C 2 + C + 1
1 1
-g m R L
T(jω) = C2 C 2C 3 = 1 + j0
2 2
[1+ C - ω LC 2 ] + j ω [R L (C 2 +C 3 ) + R L C -ω R LC 3LC 2]
1 1
C2C3 1 1 1 1
∴ C2+C3 + C = ωosc2C3LC2 → ω osc = L C1 + C2 + C3
1
C2 1 1 1 C2 C2 C2
Also, gm RL = ωosc2LC2 –1 - C = C2 C + C + C - C -1 = C → g m R L = C
1 1 2 3 1 3 3
PLL Problems and Solutions (9/6/03) Page 28
Problem 29
The objective of this problem is to use passive LC tank and negative feedback circuit to
design an LC oscillator that meets the GSM specification. At first, show the condition that the
1
ideal circuit oscillates at ωosc = and find quality factor, Q. The transistors should be
LC
modeled with the standard small-signal model using gm and rds or rout in this part of the problem.
Second, use SPICE to obtain a transient simulation. Third, simulate the oscillator that replaces
the ideal inductor with the lumped inductor model shown, and use the program referenced below
[1] to layout the inductor. Use the model parameters given in [2] for this problem.
Fig.1. Ideal LC VCO Fig.2. Lumped Inductor Model
GSM specifications:
Frequency range = 935 ~ 960MHz vc = 0.75 ~ 1.75V
Switching time = 800µsec VDD = 2.5V
Technology parameter:
Metal sheet resistance = 35 mΩ/sq.
Substrate layer resistivity = 0.015 Ω-cm
Metal to substrate capacitance = 5.91 aF/µm2
Metal to metal capacitance = 98.0 aF/µm
Csub, Rsub, Cp can be ignored
PLL Problems and Solutions (9/6/03) Page 29
Problem 29 - Continued
Solution
Problem 29 – Continued
Inductor Layout:
PLL Problems and Solutions (9/6/03) Page 31
Problem 30
A four-stage ring oscillator used as the VCO in a PLL is shown. Assume that M1 and M2 are
matched and M3 and M4 are matched. Also assume that
W
gm = 2 K' L ID where K’N = 100µA/V2 and K’P = 50µA/V2
and that rds = ∞. The parasitic capacitors to ground at the outputs are 0.1pF each.
(a.) If I =2mA, find the frequency of oscillation in Hertz. (b.) Find the W/L ratio of M1 (M2)
necessary for oscillation when I =2mA. (c.) If the current I is used to vary the frequency,
express the relationship between ωosc and I. In otherwords, find ωosc = f(I).
VDD
10µm 10µm
vi+ vo+ vi+ vo+ vi+ vo+ vi+ vo+ 1µm 1µm
vo- M3 M4 vo+
F02E2P2
I
Solution
(a.) The small-signal transfer function of the stages can be written as,
Vout(s) gm1/gm3 Vout(jω) ωC
= → Arg = -tan-1
Vin(s) C gm3
s g +1 Vin(jω)
m3
From the above, we see that each stage must contribute –45° of phase to oscillate. Therefore,
gm3 2K' 10·0.5I 2·50x10-610·10-3
ωosc = C = C = = 1010 rads/s → f osc = 1.59GHz
10-13
(b.) The gain of the 4-stage ring oscillator at ωosc should be equal to 1 so we can write,
4
gm1/gm3 4 (gm1/gm3)
1= = 4 → gm1 = 40.25 gm3 = 2 gm3 = 2 mS
1+1
2 mS = 2K' N (W/L)·1mA = 2·100x10-6 (W/L)·1mA
2mS
∴ (W/L)1 = 0.2mS = 10 → (W / L ) 1 = 1 0
(c.) From part (a.) we get,
gm3 2K' 10·0.5I 2·50x10-610·0.5I
ωosc = C = C = = 2.36x1011 I
10-13
ω osc =2.36x1011 I
PLL Problems and Solutions (9/6/03) Page 32
Problem 31
How does the oscillation VDD
frequency depend on I SS for a
ring oscillator using the stage - M3 M4
shown? Express your answer in W4
+ W3 L4
terms of VDD, V REF , I SS , the V vo1 vo2
REF M5 L3
simple large signal model
parameters of the MOSFETs (K’, VDD vi1 M1 W M2 v
1 W2 i2
VT, λ) and the W/L values of the 0.5ISS L1 L2
MOSFETs. ISS
M7 W7
Solution M6 ISS
W8 L7
This topology uses a replica L8 W5
biasing circuit to define the on- M8 L5
resistance of M3 and M4 based SU03H07P3
on the on-resistance of M5. The
on-resistance of M5 is
V DD -V REF
R on5 = 0.5I
SS
We can either assume that the W/Ls of M3, M4 and M5 are equal or since we know that Ron is
inversely proportional to the W/L ratio, we can write that,
W 5/L5
Ron3 = Ron4 = W /L Ron5
3 3
where W3/L3 = W4/L4.
Assuming a capacitance at each output of CL, allows us to write the transfer function of the ring
oscillator stage as,
Vo2-Vo1 gm1Ron3
=
V i1-V i2 sR on3 C L + 1
The phase shift due to a stage can be written as,
θi(jω) = -tan-1(ωRon3CL)
To oscillate, this phase shift needs to be equal to some value, say k (in degrees). Therefore we
can write that,
k 0.5I SS k
ωosc = R C = W /L
on3 L 5 5
W3/L3(V DD -V REF )C L
Therefore, the oscillation frequency varies linearly with ISS.
PLL Problems and Solutions (9/6/03) Page 33
Problem 32
In every practical oscillator, the LC tank is not the only source of phase shift. Hence, the actual
oscillation frequency may differ somewhat from the resonant frequency of the tank. Using the
time-varying model, explain why the oscillators’s phase noise can degrade if such off-frequency
oscillations occur.
Solution
If there is any off-frequency oscillations that are close to the actual oscillation frequency
or harmonics of it, we know from the LTV theory that these frequencies and their associated
noise will “fold” into the noise spectrum around the actual frequency and degrade the oscillator’s
phase noise. The following diagram illustrates the process.
in2
(ω) 1/f noise
∆f
∆ω ∆ω ∆ω
ω
∆ω ωo 2ωo 3ωo
Sφ(ω)
c0 c1 c2 c3
-∆ω ∆ω ω
Sv(ω)
Phase
Modulation
ωo 2ωo 3ωo ω
ωo-∆ω ωo+∆ω Fig. 3.4-32
PLL Problems and Solutions (9/6/03) Page 34
Problem 33
Assume that the steady-state output
amplitude of the following oscillator is 1V. k= M
L1L2 Comparator
Calculate the phase noise in dBc/Hz at an -
offset of 100kHz from the carrier from the vout
signal coming out of the ideal comparator. 2 +
Assume that L1 = 25nH, L2 = 100nH, M = in1 L1 L2 C
10nH, and C = 100pF. Further assume that
the noise current is SU03H07P4
2 = 4kTG ∆ f
in1 eff
Problem 34
A crystal reference oscillator and its associated transistor have the following specifications at
290°K.
Output frequency: 6.4MHz
Power output: +10 dBm
Noise figure: 2.0 dB
Flicker corner: 15 kHz
Loaded Q: 12x103
(a.) Determine and plot the SSB phase noise in dBc as a function of the frequency offset from the
carrier. Include the frequency range from 10Hz to 10MHz.
(b.) Suppose that this reference oscillator is used with a frequency synthesizer whose transfer
function from the reference to the output is
θn,o(s) N 2 ζω n s + ω n 2
=
θn,ref(s) Nref s 2 + 2 ζω n s + ω n 2
where N = 19,000, Nref = 256, ζ = 0.7, and ωn = 908 sec.-1. Make a plot of the SSB reference
noise in the output of the synthesizer.
Solution
(a.) NF = 2.0dB, F = 102.0/10 = 1.585, and Po = 10 dBm = 0.01W
FkT 1 fo 2 fc
L{fm} = 10 log P 1 + 1 + fm
s 4Q2 f m
1.585·1.38x10-23·290 1 6.4x106 2 15kHz
= 10 log
0.01 1 +
4(12x103)2 f m
1 + fm
71.11x10 3 1.5x10 4
L{fm} = 10 log 6.348x10-19 1 + 2 1 + f m
f m
-120
-130
-140
-150
dBc
-160
-170
-180
-190
10 100 1000 10 4 10 5 10 6 10 7
Offset from carrier, fm (Hz) SU03H07S5A
PLL Problems and Solutions (9/6/03) Page 36
Problem 34
(b.) The VCO phase noise transfer function is
θn,o(s) N 2 ζω n s + ω n 2 1271.2s 2 + 8.245x10 5
= Nref 2 = 74.219
θ (s)
n,ref s + 2 ζω s + ω 2
n n s 2 + 635.6s + 8.245x10 5
θn,o(jω) 2
θn,ref(dBc) = 10 log L{fm }
θn,ref(jω)
Below is a plot of the above equation as well as the transfer function, θn\,o(s)/θn\,ref(s), and the
input reference noise.
50
Transfer Function
0
-50
Input Reference Noise
-100
dBc Output Noise
-150
Noise Floor
-200
This region is not possible
-250
10 100 1000 10 4 10 5 10 6 10 7
fm SU03H07S5B
PLL Problems and Solutions (9/6/03) Page 37
Problem 35
Use the National Semiconductor website (www.national.com) to design a DPLL
frequency synthesizer for the GSM (935-960MHz) application. The channel spacing is 200kHz.
Choose an appropriate VCO from a manufacturer. Assume a 0.25µm CMOS process with a
3.3V power supply.
Your homework should show a block diagram for the resulting frequency synthesizer
with the blocks identified. Give the following parameters that you selected for your design:
1.) N, the divider ratio.
2.) ζ, the damping ratio
3.) The type of PD/PFD and the value of Kd.
4.) The type of VCO, Ko, and Vmin and Vmax.
5.) τL, the lock-in time or settling time and ωn, the natural frequency of the PLL
6.) Design of the loop filter including the time constants and component values.
Solution
The problem specifications call for the following:
• Standard : GSM
• Frequency band : 935 MHz ~ 960 MHz
• Channel spacing : 200 kHz
• Power supply : 3.3 V
• Technology : 0.25 µm CMOS
• Switching time : < 800 µs (by GSM standard)
Design
The block diagram for this design is as follows:
The central frequency to use is the geometric mean of the extreme frequencies (947 MHz).
Problem 35 - Continued
The comparison frequency at the input of the phase/frequency detector was chosen to be equal
to the channel spacing, i.e., 200 kHz. Therefore, the reference divider —ifusing a 10 MHz
crystal source at the input— and the feedback divider ratio can be found as:
f CRYSTAL 10 MHz
R= = = 50
f COMPARISON 200 kHz
f OUTPUT 947 MHz
N= = = 4735
f COMPARISON 200 kHz
VCO
• K0 = 18 MHz/V
• Vmin = typ. 0.8 V @ 934 MHz (min. 0.4 V)
• Vmax = typ. 2.2 V @ 960 MHz (min. 2.6 V)
PFD
• Kφ (= Kd) = 4 mA ( or 4/2π [mA/rad] )
General
• Lock-in time = τL = 200 µs
The filter components were found to be (standard values given, ideal values in parenthesis):
Simulation Results
Simulation was performed using the computed standard values for the filter components
and are as follows:
Problem 35 — Continued
Phase Noise
Problem 35 - Continued
Frequency analysis (Bode plots):
Spur Offset Description Spur Gain Leakage Component Pulse Component Spur Level
(kHz) (dB) (dBc) (dBc) (dBc)
200 1st Spur 33.7 -90.5 -76.3 -76.1
400 2nd Spur 21.7 -102.4 -81.2 -81.1
600 3rd Spur 14.7 -109.4 -84.2 -84.2
Therefore, this design is suitable for use in the proposed GSM application.
PLL Problems and Solutions (9/6/03) Page 41
Problem 37
On page 160-33 of the class lecture notes, the approximate rms value of the impulse sensitivity
function for single-ended ring oscillators is given as
2π2 1
Γrms ≈
3η3 N 1.5
Derive this approximate impulse sensitivity function.
Solution
This derivation follows that given in A. Hajimiri, et. al., “Jitter and Phase Noise in Ring
Oscillators,” IEEE J. of Solid-State Circuits, vol. 34, no. 6, June 1999, pp. 790-804.
The approximate waveform f(x)
and the ISF for a single-ended ring
oscillator is shown below and is 1 Slope
based on the assumptions that the
Slope = -fmax'
sensitivity during the transition is = fmax'
inversely proportional to the slope
and the rise and fall times are x
symmetrical. Γ(x) 2π
1 2
fmax' fmax'
x
2 1 2π
fmax' fmax'
SU03H08P3A
PLL Problems and Solutions (9/6/03) Page 42
Problem 37 - Continued
The Γrms can be estimated as,
2π 1/fmax '
1 ⌠ 2 1 ⌠ 2 1 3
Γrms2 ≈ ⌡Γ (x)dx = ⌡x2dx =
2π 0 4π 0 3π f max '
The normalized delay per stage is given as
η
^t =
D f max '
which is found from the following waveforms of the single-ended ring oscillator.
f(x) 1 1
fmax' fmax'
1 η
fmax'
η
fmax'
x
SU03H08P3B 2π
The period of the ring oscillator is 2N times larger than the normalized delay per stage and is
2Nη 1 π
2π = 2Nt^D = f ' → f max ' = N η
max
2 π 3 2π2 1
∴ Γrms2 ≈ =
3π N η 3η3 N 3
The result is obtained as,
2π2 1
Γrms ≈
3η3 N 1.5
Problem 38
A frequency synthesizer has a reference frequency of 5kHz and uses a 64/65 dual-modulus
prescaler. Determine the values of the A and M counters to give an output frequency of 555.015
MHz.
Solution
fo = Nfr
fo
N = f = 111003
r
N = MP+A
N 111003
M = Interger P = Interger 64 =1734, A = N –MP = 27
∴ A = 27 and M = 1734
PLL Problems and Solutions (9/6/03) Page 43
DE-FF DE-FF up
A C
I-clk D Q D Q
data
B D down
Q-clk D Q D Q
DE-FF DE-FF
Figure 1: Rotational frequency detector. I and Q clocks come from the VCO. Data is the non-
return to zero (NRZ) data to be resampled by the clock and data recovery circuit (CDR).
The flip flops are double edge sampling FFs. States A and B hold the present sampled
I and Q clocks whereas the C and D hold the previously sampled inputs (A and B are
resampled) UP is 1 when AB CD = 00 10. DOWN is 1 when AB CD = 10 00.
A typical frequency detector waveform for data slower than VCO clock is shown below. Note that,
anytime when AB changes from 00 to 10 a DOWN pulse is generated. In this example, there is no
UP pulse since no 10 to 00 transition occurs.
data
I_clk
Q_clk
A
B
up 10 11 01 00 10 10 11 00 10 11 01 01 10 01 00 10 11 01 01 10 11 11 01 00 10
down
Q Q Q Q Q Q Q Q Q
01 11 01 11 01 11 01 11 01 11 01 11 01 11 01 11 01 11
I I I I I I I I I
00 10 00 10 00 10 00 10 00 10 00 10 00 10 00 10 00 10
down
Figure 3: IQ clocks are 12.5% faster than sampling clock.
PLL Problems and Solutions (9/6/03) Page 45
Problem 42 - Continued
In the example below, VCO clock is 25% faster. Therefore, the beat frequency completes its full
rotation in 4 cycles. In 8 cycles there are two 00 to 10 transitions or equivalently two DOWN pulses
are produced by the frequency detector.
Q Q Q Q Q Q Q Q Q
01 11 01 11 01 11 01 11 01 11 01 11 01 11 01 11 01 11
I I I I I I I I I
00 10 00 10 00 10 00 10 00 10 00 10 00 10 00 10 00 10
down down
Figure 4: IQ clocks are 25.0% faster than sampling clock.
Now, let s look at what happens if the VCO clock is 37.5% faster than the sampling clock. The beat
frequency rotation vector comes to its starting position in 8 cycles, and during which only one 00 to
10 transition is made. Note that from the 4th position to 5th position, the beat frequency vector, skip
the quadrant 00. Therefore, the past state of the sampling state CD and present sampling state
AB which goes to the four input AND signals are 01 and 10. As a result, both outputs remain at 0.
No UP and DOWN generated when one of the decision quadrants are skipped.
Q Q Q Q Q Q Q Q Q
01 11 01 11 01 11 01 11 01 11 01 11 01 11 01 11 01 11
I I I I I I I I I
00 10 00 10 00 10 00 10 00 10 00 10 00 10 00 10 00 10
down
Figure 5: IQ clocks are 37.5% faster than sampling clock.
When the speed difference is 50%, there is no 00 to 10 transition. As a result no UP/DOWN pulses
generated.
Q Q Q Q Q Q Q Q Q
01 11 01 11 01 11 01 11 01 11 01 11 01 11 01 11 01 11
I I I I I I I I I
00 10 00 10 00 10 00 10 00 10 00 10 00 10 00 10 00 10
Figure 6: IQ clocks are 50% faster (or slower) than sampling clock.
The case where VCO is 62.5% percent faster than the sampling clock: During the 8 sampling
period in which the beat frequency vector comes to its initial starting point, there is only one
transition between quadrants 3 and 4. This transition, however is on the reverse direction. That is
CD=10 to AB=00. The AND gates in this case generate an UP pulse. To the frequency detector,
VCO appears to be 37.5% slower instead of 62.5% faster. A wrong pulse is generated.
Q Q Q Q Q Q Q Q Q
01 11 01 11 01 11 01 11 01 11 01 11 01 11 01 11 01 11
I I I I I I I I I
00 10 00 10 00 10 00 10 00 10 00 10 00 10 00 10 00 10
up
Figure 7: IQ clocks are 50% faster (frequency detector interprets this as VCO is 37.5% slower)
than sampling clock.
The above examples is for the case when VCO is faster. The case in which the VCO is slower can
be plotted similarly. When VCO is slow, the beat frequency vector traverses the IQ quadrant
planes in clockwise direction. In the light of above vector diagrams, the following frequency detector
output vs. frequency input waveform can be plotted.
PLL Problems and Solutions (9/6/03) Page 46
Problem 42 - Continued
Normalized frequncy
+1 detector gain
fc 3fc fc fc
4 2 4 frequency
0 fc fc 3fc fc error
4 2 4
-1
Figure 8: Frequency detector characteristics when the sampling input is clock instead of
NRZ data.
Note that, the pulling range of this frequency detector is +/-50% when a full rate clock signal is
rising (or falling) edge samples the I and Q VCO clocks, instead of data sampling the I and Q clocks
at both rising and falling edges. This case is explained below.
From Figure 8, the detector gain is maximum for +/-25% frequency offset. (Two DOWN pulses in
Figure 4 above). Above +/-50% frequency offset, the output changes polarity and VCO frequency
is pulled to the wrong direction. The useful range is, therefore only +/-50%.
The above phase diagram example is for the case if the frequency detector input is a full-rate clock
instead of NRZ data. We further assumed that, I and Q clocks are sampled only at one edge of
the clock (either rising or falling). A pseudo random NRZ data resembles to a clock with 1/4th of the
full speed clock as far as the transition density is concerned. If this fact is combined with the
double edge sampling nature of the actual frequency detector, the data sampling the I and Q
clocks can be assumed as half the full speed clock. That is, in above phasor diagrams, the IQ
clocks are effectively sampled every other time. In this case, the frequency detector characteristics,
changes polarity when VCO range exceeds +/-25% of the data rate. For actual data inputs,
therefore the frequency characteristics resembles to the following figure. The rounded edges of
the gain characteristics is due to the pseudo random nature of the input bit sequence (PRBS).
Normalized frequncy
+1 detector gain
-1
Figure 10: Rotational frequency detector characteristics for PRBS NRZ data.