Professional Documents
Culture Documents
A Thesis
by
of Graduate Studies of
MASTER OF SCIENCE
December 2002
Submitted to Texas A&M University in partial fulfillment of the requirement for the degree of
MASTER OF SCIENCE
Approved as t
ty
Jose
(Chair
lva-Martinez
of Committee)
anchez-Sinencio ember)
Eva Sevick-Muraca
(Member)
anan Sin
(Head
of Department)
December 2002
ABSTRACT
On-Chip Spectrum/Vector
of Analog
Integrated Circuits.
(December 2002)
Marcia Gisela Mendez Rivera,
Guanajuato, Mexico;
The number
of functions
though the
techniques
have been
circuits due
to the different nature of digital and analog circuits. In the case of analog circuits, the
number
of possible
if the
circuit is working
The complexity
and sensitivity
of the time
and voltage
nature in analog circuits makes the testing task even more difficult. Even more, the
analog circuits have to be tested under different conditions; e.g. , sweeping frequency and
amplitude.
Due to the fact that there are not specific and efficient techniques for testing
analog circuits, every analog circuit requires a particular design in order to be tested. As
a result of this, the test cost is the dominant issue in many products and the investment
made in this stage is not recovered.
of the
fundamental
to characterize and test analog circuits using an external digital tester rather
used is a built-in self-test
that tells us
if the
that
circuit works or not based on certain error margin. In our case, the parameters
concern us are transfer function (magnitude and phase response) and harmonic distortion
components.
The techniques
capability
of the DUT
rely on the
signal to be
of the circuit to
used as stimuli.
synchronization
The switched-capacitor
used ensure
the
controlling these blocks assures that the tracking error could be within
0.5'/o if switched-
To my beloved family:
My husband Ari, my daughter Aimee,
my parents Luts Alfonso and Lydia and my brother Luis Alfonso
ACKNOWLEDGMENTS
I would like to thank my advisor Dr. Jose Silva-Martinez for his support, guidance, and
patience beyond the advisor-student
invaluable
friendship
relationship
during all these years. The advice and time he always shared with
of my education. I also
Sinencio for his support during the time I spent in the AMSC group, as well as express
my gratitude to all the members
of my
I would like to thank in a very special way to the family Silva-Rivas whose support
played a very important role in the culmination
of this
and priceless.
in one way or another during this work. The interaction I had with all
important during my studies here at Texas A&M.
of them
was really
Finally I would like to dedicate this work to the most important persons in my life: my
husband
Ari, my daughter Aimee, my parents Luis Alfonso and Lydia and my brother
to make my dreams came
of you.
vn
TABLE OF CONTENTS
Page
ABSTRACT.
DEDICATION.
ACKNOWLEDGMENTS.
nl
V1
TABLE OF CONTENTS.
V11
LIST OF FIGURES.
LIST OF TABLES.
CHAPTER
1X
X11
INTRODUCTION.
A.
B.
Introduction. Motivation.
4 6 7
II
A.
Introduction.
9
10
15
B. Basic Principles.
C. Proposed Solution. D. Specifications.
19
24 24 24 29 37 39 41
III
B.
C. D.
E. F.
Introduction. Sinewave Generator Bandpass Filter Voltage Gain Amplifier (VGA) . . Operational Amplifier (Opamp) . . Analog-to-Digital Converter.
vnt
CHAPTER
Page
IV
..
45
A.
Introduction.
B.
45 46 50 55
CONCLUSIONS
REFERENCES.
APPENDIX.
61
64
75
VITA.
LIST OF FIGURES
FIGURE
Test at each stage
Page
of the
manufacturing
process. . .
steps. . .
1.2 1.3
production.
10
of a filter
bank . .
spechnm analyzer. . .
13 13
spectrum analyzer.
2.5
2.6
14
16
18
2.7 2.8
Center-to-sampling
f'requency relationship.
25 26
logic.
of the of the
sinewave generator. . . . .
27
harmonic distortion
30
3.6 3.7
Effect
of the
accuracy
of the
31
32
Switched-capacitor
bandpass filter.
FIGURE
Page
bandpass filter.
3.8
.
= '/4 .
33
3.9
fi
and
35
Switcap simulation
36
37 38
40
41
42 44 44
45
(a) Schematic
of the
46
47
48
of the
~+100 mV
..
51 51
52
52
4. 10 4. 11 4. 12 4. 13
4. 14
A. 1
Intermodulation
measurement
53
Noise
of the
bandpass filter.
54
56
The input signal amplitude is +100
mV.
Noise in the VGA.
56 57
bandpass filter.
Switched-capacitor
67
FIGURE
Page
bandpass filter. .
A.2
67
B.I
69
LIST OF TABLES
TABLE
Page
Target specifications for the on-chip network/spectrum
analyzer. .
2. 1
23
Capacitor values for the sinewave generator. . . Capacitor values for the SC bandpass filter.
29
35 35 38
41
.. ..
of
46
50
55
summary
..
summary
57
72
B.1
CHAPTER
INTRODUCTION
A. Introduction
The process of making integrated
specifications must be translated
circuits consists
works within the specs. Once the design is finished, the layout is made: in this step, postlayout simulations
help to determine
if the
to the schematic
specifications
properly,
are within
certain boundaries.
Finally, provided
of
the production
process. Fig
I.l
process of placing a product on the market [1,2]. If in any of these steps a failure exist,
optimization
or modifications
of these
failures can occur at the layout level, on the wafer or in packaged parts. There are many
of the
of
the manufacturing
effective because only non-defective components are permitted to move to the next level
of assembly.
Wafer
Process
Test
Die
Pastose
Bum-ia
Test
Discard
Board
Substrate
Process
Test
Ininal Assembly
System
Test
Fmal Assembly
Fmal
Test
To Fteld
Repair
process.
of Integrated
Circuits (IC)
a large amount
of
money
is invested
not
recovered.
Fig. 1.2 shows the cost associated to each stage of the process of making an
integrated circuit. It can be seen that it takes ten times the effort through each successive
manufactunng
step to detect, locate and repair a faulty component. Therefore, the testing
of the next
stage.
1000
100
10
0. 1
'O
Test
Acttvtty
p
M
steps.
B.
Motivation
Modem integrated circuits are becoming more complex, which does not assure optimal
performance
and the access to internal nodes is difficult due to both the limited number
of pins
available
task. Several
techniques
that involve
self-testing
developed. However,
the growth
in complexity
of functions
integrated on a single-chip have increased rapidly in the last years, making the testing a
difficult task
[3].Even
Albeit the digital testing has reached certain maturity and well-defined
techniques have
been developed, we can not use the same techniques for analog and mixed-signal circuits
due to the different
information
is binary
nature
of
digital
and analog
circuits.
In digital
circuits, the
of analog
of
1.4.
possible values is unlimited and there is not a specific value that tells us
if the
circuit is
Testing in analog circuits faces many problems that digital testing has overcome. The
complexity and sensitivity
of the time
testing task even more difficult. Even more, the analog circuits have to be tested under
different conditions; e.g. , sweeping frequency and amplitude.
There are not specific and
efficient techniques for testing analog circuits: every analog circuit requires a particular
design in order to be tested.
Bad
Good
Digital Function
Good
Bad
Bad
Analog Function
analog part. Mixed signal testing is not only made difficult by the long testing times analog components
which must be tested on expensive automated
of
to
test equipment
with
limited
relatively few inputs and outputs since it is not possible to bnng all the analog inputs and
outputs out to the pins.
There are several factors that have contributed to make the mixed-signal
complex, such as the access to the analog components
testing more
both digital and analog parts. Other issues to consider are the reduction
of the
test time
and the area used by the testing circuit. They should not represent a large area overhead
for the whole system. Also, the presence of the testing circuit must affect as less as
possible the measurements
obtained.
Mixed-signal
[4]. In the
of the Device
(DUT): i.e. operation point. In the AC BIST we try to measure the frequency response of
the system. Traditionally,
the AC characterization
of
frequency frequency
domain:
a sinusoidal
signal is applied
output
to the input
of
is converted
time consuming. Therefore, the need for a solution to this problem is evident.
C. Basics on Testing
Fig. 1.5 shows the block diagram of how the testing is performed
in analog circuits.
There are two blocks: the device under test (DUT) and the circuit performing
Using the switches, normal operation or testing mode can be chosen. The area
the test.
of the test
circuit should be smaller than the DUT. The goal is to test the DUT in an automatic way
without changing its original configuration.
Input
L
some
DUT
Output
Vector
Analyzer
Fig. 1.5. Testing in circuits. Block diagram. Fig. 1.6 shows, some of the many parameters
frequency response, amplitude, application
in analog circuits:
of the DUT,
important
all
than
others and the circuit should allow us to test the most critical parameters,
using an input signal and measuring
of these of
of the
parameters
concern are:
~ ~
Harmonic distortion.
Noise.
Input
DUT
Output
Vector
Analyzer
Amphtude
BIV
Analysis
to integrate
the testing
circuit on the same chip and characterize and test analog circuits using an external digital
tester rather that an analog tester in order to reduce test time and cost [5]. The
architecture to be used is a built-in test circuit. The test should be made in an automatic
that tells us
if the
margin. The tester is designed in such a way that allows us to test the most important
parameters,
such as transfer
function
(magnitude
and harmonic
distortion components.
of the
required specifications.
D. Organization
The thesis is organized
as follows:
Chapter
proposed technique as well as some previous proposals. The third chapter deals with the
design issues
of the
main building
and experimental
results
of
the designed
of the
thesis.
CHAPTER
II
Spectrum
and
of analog
systems.
The network and spectrum analyzers in use today are the result of
of
continuous
development.
available
spectrum analyzers today have a dynamic range greater than 100 dB (corresponding
to
10 orders of magnitude in power) and over seven orders of magnitude in frequency [6].
With this equipment,
it is possible to analyze the transfer function and the distortion
components of the circuit under test. The use of the spectrum analyzers for production
testing
is prohibited
due to large
testing
time
and
high
costs. In general,
the
charactenzation
of the
gain and
Usually
such measurements
are performed
a considerable
parasitic to
to embed the testing circuitry on the same chip in order to reduce the interaction
of the
. Several approaches to
between the stimuli
[7-11], but
the synchronization
of switched-capacitor techniques
a self-synchrouized
10
B. Basic Principles
The traditional way
parameters
of observing
displays,
instead
of
spectrum analyzer takes the analog signal (in the time domain) and converts it into the frequency
domain.
The resulting
components
spectrum
measurement
shows
the energy
of
the
this
different
relationship
frequency
along
the
frequency
[12]; here,
that are
not distinguishable
Ampbtude
eeet
Ii
Amplitude
Amphtude
Time
prequeuey
Many
signals
harmonic
components)
domain.
11
analysis,
in a single frequency
using an
signal, amplitude
and frequency
can be measured
oscilloscope, but for more complex signals, this task is not possible because such signals
generally
components
with different
amplitudes
and phases.
and comparing
the
signal.
of measuring the
the input
spectrum
of a
of
filter banks
[13], where
signal
of
bandpass
of these
filters are
of the
of interest. However,
dependent
of
each filter.
of filters
given by:
= Spectrum bandwidth
(2. 1)
Frequency
of a filter bank.
12
We can see that even for small frequency ranges, the number of required filters can
become very large, which makes this option not a good candidate for on-chip testing,
where silicon area must be optimized.
generated from the overlapping
non-overlapping
Another
problem
frequencies.
Another approach
of spectrum
of tunable
of the
incoming signal
For this purpose, a mixer and a local oscillator (LO) are used.
the
The &equency of the LO is chosen in such way that the difference between
frequency of the input signal and the one frequency
at a specific intermediate
bandpass
filter. By
of the LO,
of
pnneiple.
spectrum
analyzer.
The low-pass
of the
mixed with the signal coming from the LO. The envelope of the IF signal is detected and
is used to control the y-axis [6]. The LO frequency is swept in order to downconvert the
frequency of the input signal; the VCO is con&oiled by a ramp generator, which also
In commercial
analyzers,
the IF is chosen
resolution
of
the spectrum
narrow IF bandwidths
are
difficult to achieve in the Gigahertz range, most of the spectrum analyzers use several IF
stages to mix down the input signal to an IF where the narrowest filtering takes place.
(IF stages) in order to reject image frequencies. All the filters are
13
fixed-tuned
so better resolutions
of the
final
IF
filter can be changed to control selectivity. The local oscillator sweeps across the desired
range of frequencies (which is translated into a visual swept between the start and stop
frequencies
of the spectrum being studied). The detector is used after the last IF stage to
signal, where the detected signal is converted
into a dc value,
which represents the amplitude, and is sent to the circuitry that drives the display.
mixer
Input
X
Lo
'"
Envelope Detector
Vertical deflection
Honzontal deflection
vco
Voltage ramp
of a superheterodyne
spectrum analyzer.
LO
d)
IF
if xed
LO~ ~
I
I"Mater
Input
Oet
le
Leman Ftiter
flntaae Reieetteel
I' lP Piller
I" L 0
2 IF alter
IVanaale
3 opC ael
V dee Falter
14
specifications
which
analyzer
signals
should satisfy.
First, resolution
bandwidth,
is the
ability
to resolve
of
different
&equencies, is determined
by the IF bandpass
filter bandwidth.
follows
[6]:
ST=k.
Rzu
(2.2)
where ST is the sweeping time, k is a constant that depends on the filter shape (for a
Gaussian
bandwidth.
filter is about 2.5), sPan is the &equency sPan and Rsrr is the resolution
Usually,
Rsrr and
reasonable
resolution
can be obtained;
if
a higher resolution
If low
R=10kHz,
R s=l
kHz
ST=0 03 sec
ST=0 3 sec
Frequency
of Rsrr
and ST.
description
of
the signal is
of the contents of
of the
15
at
phase
of a
measurement
of the
of
magnitude
because these quantities provide information about the transfer function of the system.
The magnitude and phase extraction can be implemented
using peak detectors and phase
detectors but they usually require calibration and are prone to errors over frequency,
variations.
Calibrated
that use
since the result is given by the ratio V,,/V, . Another proposed solution is the use of
filter banks that are multiplexed
accordingly to the frequency being analyzed [8-11],but
the silicon area required is quite large, which makes them not very good candidates for
built-in testing.
frequency,
magnitude
and
results in a considerable
because there will be no need to use two different tests in order to measure the spectrum
and the transfer function
of the DUT.
C. Proposed Solution
The architecture to be used is a built-in on-chip spectrum analyzer. The parameters
of
is
distortion
analyzer
spectrum/vector
proposed.
2.6 shows
of the
the stimuli that is applied to the DUT and its output is sensed by the detector. Having
both the input and output signals, we can compare and analyze them to decide if the
Device
Under
Test
On-Chip Spectrum/Vector
Analyzer
Source
Signal
Processor
Detector
V
/
External
xz'
Digital Control
Output
Control
The techniques proposed to measure the frequency response of the DUT rely on the
capability of the circuit to generate a low distortion and accurate sinusoidal signal to be used as stimuh. Once this tone has passed through the DUT, it has to be measured and
compared with the stimuli in order to obtain the DUT transfer function.
The process descnbed above is repeated for different frequencies to obtain the full DUT
charactenzation.
In order to perform the harmomc
fixed frequency is applied to the DUT. Then the center frequency of the high-Q filter is
changed to frequencies that are integer multiples of the frequency applied to the DUT
and their respective amplitudes
are measured. A vanable gain amplifier (VGA) is needed
of the
system dynamic
range. Usually
filter is used in the detector in order to minimize the in-band noise and spurious tones.
Therefore the frequency of the stimuli and the filter's center frequency have to be fully
synchronized
in order
to avoid unwanted
attenuations
in the measured
tones. The
switched-capacitor
mentioned
above. It
is shown
if both
signal generator
filter are controlled by the same digital signal, the tracking error of these parameters lies
17
within
techniques
of
the
analyzer
techniques.
A conceptual
schematic
diagram
of
the proposed
technique
to characterize
synthesizer,
analog
of a
digital I'requency
a switchedfilter and an
bandpass
synthesizer
as the non-overlapping
delivers a sinusoidal
amplitude
of the
from the sinusoidal generator is adjustable to provide the proper level to the stimuli. The
output
of the DUT is
of both
switched-capacitor
of the
spectrum-vector
analyzer,
f, to
distortion
of
this system
is its inherent
synchronization.
The sinewave
f, //6.
of the switched-capacitor
18
where
f, is the
the bandpass
defined. Even more, given that the stimuli frequency and the filter center &equency are
tracked to each other (since both are generated from the master clock), the system is selfsynchroiuzed;
the
Filter clocks
reference
clk
Frequency Synthesizer
f,
fsr
Smevrave Generator
Signal
Z'fo
3 sf
VGA
VGA
SC Bandpass Fdter
A/D
Converter
DSP, PC,
Drgnal Tester
From the previous description, it can be noted that the blocks that conform the system
and their main properties are:
19
Digital Frequency Synthesizer: Should be flexible and simple. Since the number
generated
of frequencies to be
used.
Sinewave
Generator.
A low distortion
switched-capacitor
50dB) is desired for this block. The digital circuitry used to control the switches
should be as simple as possible.
Switched-Capacitor
implementation
is used; it
g (50). The
capacitor spread
needs to be minimized
Voltage
Gain Amplifier:
This
is a programmable
gain
amplifier
with
an
attenuation/gain
AnaloglDigital
The particular specifications for these blocks will be discussed in the next section and in
Chapter III.
D. Specifications
In this section we will discuss the main design issues and the specifications
builihng block.
for each
The first limitation to be considered is the distortion allowed in the sinewave generator
and bandpass
in
20
robustness
and implementation
resolution of
of the
increased
capacitor mismatch
complexity.
conditions
and
a limited
implementation
A resolution
of 8 bits
corresponds to an error
specification
generator. The third harmonic distortion is set to be I LSB below the resolution
system, such that it is transparent
of the
of 0.2
'/o
(-54dB).
The distortion
of the
of steps used
to
waveform.
If a
small number
of steps
harmonics
of the signal, it is necessary to choose a number of steps that provides a of harmonic components along with a small silicon area. If 8 points are
used, the signal obtained can resemble a sinewave, but the number
high.
of harmonics
can be
If we
use 32 points, the signal will be purer, but the digital control logic increases
in the
(as well as the required silicon area) and there is not a significant improvement
harmonic
components
best option is to use 16 points, which provides a reduced digital logic and therefore a
small area and a proper number ofharmonic
components.
is highly dependent
on the quahty
g of the BP filter.
g,
of the
harmonic distortion is. However a practical limit must be set for the quality factor
of the
the
~-54 dB,
Irom the center frequency fp to the third harmonic 3 fp, yields to:
21
~H(s)
H(s)~,
54dB
or
iH(s)
' &0.005
(2.4)
of a biquadratic
No
H(s) =
S +
Ns
(2.5)
S+No
The transfer function of the filter is evaluated at frequencies, fo and 3fo and a Q that
64 Q', 9
&0.005
(2.6)
The required quality factor to provide an attenuation of -46 dB according to the previous
equation
is 75, which is extremely
translated into a large capacitive spread or a higher order filter. Since one limitation for
this application is the use
of 50 is
assumed,
this attenuation
a reasonable
tradeoff
Another issue to consider is the center to sampling Irequency ratio fg lf, . This ratio can be
of harmonic
components
to be measured.
switching
without
Since the
techniques,
of using
any effect
of
22
/
I
fp=
1
I
/
sf
/
p
2fp
3fp
4fp
3fo
6f
7fo
fo
fo
10fo
/
/
'I
1
I
2fp
3fp
f, = 16fo
4fp
ifo
6fp
fp
1 fo
12fo
(b)
frequency relationship.
(a) f, = 8
Also, the frequency range where the proposed on-chip spectrum analyzer can operate has
to be evaluated. In order to avoid the hard constrain that a high frequency design implies
and since we want to probe the proposed principles only, this particular implementation
product
generated
in order to obtain a
frequencies
of
the transfer
1
of the DUT,
kHz steps which can be achievable using a simple digital &equency synthesizer.
23
analyzer.
On-Chip
Specs
Network Spectrum
Analyzer
Harmonic
Distortion
&-54 dB
&-42 83
(Sinewave Generator)
Harmonic
(Randpass Filter)
Distortion
()uality
Factor
to
&50
16
1
(Randpass Filter)
Center
Sampling
Frequency Redo
(fJf
Frequency Range
Frequency Resolution
kHz-10 kHz
1
1-10 kHz in
kHz steps
24
CHAPTER III
of the
B.
Sinewave Generator
The purpose of this block is to generate the stimuli to be applied to the system under test.
Several approaches for generating sinewave signals have been reported
applications,
sinusoidal
[14-16].In many
Analog Converter (DAC). The signal to be generated is sampled in several points and
the obtained values are stored in a Read-Only Memory (ROM) and translated
into an
analog signal to be applied to the Device Under Test (DUT). Even though the signals
obtained using this technique have very low distortion and are very accurate when the
DAC and the ROM (or PEA) are added up, the resulting area is quite large. Since we
desire a small area, a simpler but efficient approach has to be used.
Fig. 3.1 shows the schematic diagram of the sinewave generator proposed in this work. It
consists of a programmable
gain amplifier
values of an ideal sampled and held sinewave. The amplifier has 4 different gain stages,
which generate a sinusoidal output with 16 steps per period. The switch PZ sets the zero
of the
sinusoidal
sequentially
25
Once the maximum values is obtained, the switches close in the opposite direction (from
PDI to PA1)
2
PI,switches
advantage
from V, r to
Vpr
of
The
of this implementation
of the logic,
PDI
CI
PD2
J
PBI
PC2
V,
~
P
j
PBI
PB2
-P
PA2
Voot
An important
of the
sinusoidal
of the
sinewave. Fig.
3.2
of the
sinusoidal
generator. It is composed
of a
5-bit
shift register with clear and preset. The control signals generated by the
by the non-overlapping
phases
f&
switches
of
the bandpass
required. Fig.
3.4 shows
the simulated results for both the control signals and the sinusoidal output.
0
Ik
0,
0
Ik
D D
Ik
0
Q,
P
0
Ik
Ik
Q.
Ik
f, A
f, C
f. Z
A,
f,
PAI,
PBI,
PCI
PDI
PZI
PA'
PD
PIZ
PI2,
11
PBI
PCI
PDI
PZI
PA
PDI
ICI
Pl
'
time
le
12
'13
'14
15
PZ
PD
PC
PBI
PAI
27
Sinewave
Generator
Responses
2.6
Pz
88
-2.8
2, 8
pet
88
2.8
2, 8
Pat
88
2.8
28
8.8 2.8
, : Pet
PA1
88
2.8
: trout
588m
8 8
-588
588u
1.2m
em
time
28m
2 em
2 em
(s
Fig. 3.4 Simulated control signals and output of the sinewave generator.
This implementation
and provides
a compact implementation.
The sinusoidal
very precise
using a
and since
digital
and controllable
switched-capacitor
filter is
amplifier,
results.
One
of the
most important
characteristics
of the circuit is
harmonic distortion to be able to measure properly the DUT. The main limitations
of this
block are due to the finite parameters of the operational amplifier and clock feed-through
effects. The maximum operation frequency is given by the maximum frequency at which
the opamp
maximum
of
thumb
for switched
product according
to:
f, =
Since the sinewave
frequency
is generated
using
2n
10
GBW(Hz)
(3.1)
trequency
vs. clock
complexity
and harmonic
distortion.
The maximum
switched-capacitor
enough.
techniques
of
the sinusoidal
generator
output
is
of the
3.1,
of the
sinusoidal
of the
VGA shown
of the
sinusoidal
the magnitude
of the
in the sinewave
= Sin(90') C,
C,'
' = Sm(45') C,
The minimum
capacitors
C,
29
Table
Cr C9 Cs
3. 1. Capacitor
5.22 F 3.7 F 5 22 F
Cr
C4
C(ops
4.83 pF 2 F 20 97
C. Bandpass Filter
Another critical building block
of this
the harmonic
distortion
fmm the DUT. Due to the important role played by the filter in the system,
of
on Fig. 3.5. Consider a bandpass filter with a high quality factor (Q). When measuring
the fundamental
should be attenuated
at 2 fp is made,
is
of the
fundamental
should be attenuated
of 3 fp, where
the component
filter with a
that when a
wider bandwidth
measurement
(low
g)
of the second
of the
fundamental
is
considerably
large, sometimes
not precisely centered. As we can see, the effect of Q is more critical for the second and
third harmonic distortion characterization.
30
2f
3fo
Frequency
2f
3fFrequency
Frequency
2 "o
3&
Frequency
2t;,
3j,
Frequency
Frequency
o o
Frequency
2o
Frequency
Fig. 3.5. Quality factor effects in the selection of the harmonic distortion component. (a)
High
g (b) Low
Q.
A second issue to consider is the precision of the filter's center &equency. A small
deviation of the center frequency can lead to incorrect measurements
as shown in Fig.
31
the measurement
of the
fundamental;
if
of this
even more severe when high Q filters are used. Both magnitude
measurements
+af f; t,
2(,
3fo
Frequency
f.= if+sf
fo
2$
3 f,
Frequency
o o o
Frequency
f,= 3f +sf
2fo
Frequency
a bandpass
of a large capacitor
g; the
in this
need
of a
spread. The second problem can be solved using a very precise approximation:
32
Since the system is intended for on-chip testing, the silicon area required by the filter
should be minimized,
preventing
the use
of a high
approximations,
of higher
~ ~
f, lf,
Q
=1/32
=50
and Fig.
3.8
C~l
C
C,
OUi
33
C,
C,
1
z'
z'
Cz
Ci
1-z '
C,
Ci
1
z
V
C, C,
1 1
z'
C, C,
bandpass filter.
H( )=A
z'+a ='+b
shown in Fig.
(3 3)
3.8 is:
H(z)
2
(3.4)
z
11
By comparing both the bilinear and filter transfer functions (equations 3.3
can find the required values for the capacitors. These values are summarized
for a unity peak gain reahzation. The minimum
design of this filter can be found in appendix A.
and
3.4), we
detailed
on table 3.2
capacitor is set to
0.2 pF. A
34
Cs Cs
Cg
02 F
10.04 F 10.04 F 0.2 F
Cr Cs Cs
Croiag
5116 F
51.16 F 1.02 F
123 82 F
From table 3.2 it can be seen that the capacitor spread is very large (1:250).Notice that
the smallest capacitors are Cg and Cs, those that connol the quality factor and the peak
gain
of the
g, the capacitance
in the design
of the
power consumption.
Second, the
A method that allows to reduce the capacitive spread while maintaining the high
filter is needed
g of the
[17].The
associated resistance
(3 5)
CQ
where
we decrease
If
n as follows:
f/
(3.6)
Co
Since the obtained equivalent resistance is n times larger than the original R,~, and the
time constant
(R,q
should
C&)
of
the integrator
must
be maintained,
capacitor
C&
35
capacitance
an effective reduction
in the
values are much smaller than the original ones, leading to a capacitive spread
of I:64.
Furthermore,
if the
minimum
of all
capacitors is also reduced. The final capacitor values are given on table
3.4.
C
Table
Cz
Cs Cs
Croppy(
3.4. Final
Co
Cs
Cv
C
Figure
Ci
Cs Cs
Cr,l~
3.8 shows
of the described
1' CQ
1'
C1
CO
C3
72
Vin
Voul
cs
Fig. 3.9. Bandpass filter with additional sampling Irequencyin Cg. (fi' = 'l4. fi andfz'='l~ fz).
36
Fig. 3.9 shows the modified schematic. The filter was simulated in Switcap. Fig. 3.10
shows
a comparison
of
the bandpass
filter when
the capacitive
spread
reduction
technique is used. Since the effective Q is increased by factor 4, the peak gain increases
3.10, since
3.11
input signal at
Khz.
ste it ta
estusqes
st)ter
u nng
tqoo sana)ang
sequent)ca
's, s t"
qtt 0 1
10
-10
-10
-30
5(q)
No)
Fmqqqrtto
100)
Fn)
3000
37
Bandpass Filter
Transient
Response
58m
28m
18m
-48m
78m
78m
48m 1 8m
28m
-58m
32m
time
38m (s )
48m
42m
44m
the
of the
range
of the
ADC and to avoid a reduction in resolution due to a small input signal. For
amplification
'g,. factors in 4 dB steps ar e p ro p osed.. Fi
this application,
3.12
shows the
u Cq through
for the
of errors
scheme
of the
C7 sample
the offset
and the
I, the
polarity
of C7 is reversed
38
measured
is subtracted
C6
C3
J
1
c7
Vin
C1
vout
Ct
Table
3.5. Capacitor
0.34 pF 0.2 F 05 F 1.241 F 0.2 F
C6
Ct~&
39
E.
of the
opamp are based on the features required by the filter because this is the block with more
stringent constrains.
gain
of the
I
P
where
(3 7)
s is
the maximum
P is
0.5 and A is the opamp dc gain. For an error of I'/w we have that the minimum dc gain
should be 200.
The gain-bandwidth
as follows:
by the maximum
sampling
frequency
GBW &10
f,
(3.8)
Then, for a 32 kHz sampling frequency, the minimum GWB should be 320 kHz.
The load capacitor is estimated adding all the capacitors that will be connected at the
output
of the
Form the previous discussion, we can set the design specifications for the opamp:
~ ~
I MHz.
Ct = 10pF.
40
Fig. 3.13 shows the schematic of the opamp. It consists of a differential pair (Mq-Mz) and
three current mirrors
(Mq-Ms) with a cascode output
B. The
3.6.
Voo
M,
M4
M4
M4
Ib
M4c
Msc
M, 4
M4
~ss
Figure
3.13. Opamp
schematic.
41
8'(pm)
5, 52
L(~)
1.2 1.2 1.2
1.2 1.2
Mn Ms
13.2
3.86
MsrMsrsMa Mdc
4.8
4.8
3.86
1.316
1.62
2.63
3.3
F.
Analog-to-Digital
Converter
3.14. This
of
being halved, and the reference voltage is unchanged. time and a moderate circuit complexity.
Vm
Track
aod Hold
V,
Comparator
Senal
Output
V,
Va
~vm
/2
42
The conversion is made as follows: first, the input signal is sampled by a track and hold
circuit and then passed to a comparator.
If the
it is set to 0.
(if the
bit is
I) to
of bits
3.15,
is obtained. As we can see, this converter requires small area because it reuses the same
circuitry to perform
the conversion.
in Fig.
Qs~.
V&=v, to
Vot
) [ Btt=1
/2) ,
V=2(V-V
V=2(V+V
j2)
I=i+i
No
i&N
Vo.
Stop
is the multiplier
in order to reduce errors in the final result input and output capacitors will give us
of the
of the
43
of the
nature
can be generated. Therefore, we need to find a way to reduce the error introduced by this possible mismatch.
A common source
of error
is the offset generated by the opamp due to its finite gain and
of techniques
The architecture used in the multiplier by two is shown in Fig. 3.16(a) [21].It consists of
an opamp, two capacitors and seven switches. The circuit works as follows: This circuit
the
input voltage is sampled in Ci as well as the input offset voltage; in the second one, the
at the
of the
multiplication
made. Fig.
3.17 shows
the circuit
be discussed.
44
CI
v. ~
J
V
(a)
(b)
aX
a
C
b
Vinn
J
Vout
Vfn n
Vout
Phase
a
I&
C
b
Vtn n
VinW
Vout
J
Vout
Phase 3
Phase 4
46
B.
Sinewave Generator
of this block
be applied to the system under test. Figure 4.2 shows the simulated
sinewave generator wtth a 500 mV reference voltage. Since the amplitude
for the
output
of the
waveform
is proportional
Table 4. 1 shows the simulated and measured results for the spectrum at fa 2 f0, and 3 f0
of 1 kHz.
688m
8.M
6Mm
of the
Simulated
of the
oscillating frequency is
%+V/8v.
ir
I t I
t
1
3
I
I-W
4'
t *M-+-
"I""4-,
I I
t
i
t
4
I I
---kf
t
I
f
'I
Betet 88
tttRCE
&)2
Bffstt
Taaet NBSB N
8:
A
Cht Beeetrun
1,088 d8
08tr
/dt's
.10
d8
-100
dBJt
8'tg 'tt
0 Ht
Btedt 10 kdt
ete: IPHS-82
faetex 6$83P.
N
1
000,
Ht
-89 888
88tt/Nt
dde/BB
-180
49
The rms value for the 200 mV~ input signal is given by:
V, (rms)
Vp
= 200m V = 141.42m
V,
(4.1)
From Fig. 4.5, the noise power measured (in dBm) in the bandwidth of interest is given by:
P = 59.26+10
log(BW)
(4.2)
(4 3)
(P
= 7.55n W, m gJJ P,
(4 4)
D resistor:
V,
(4.5)
DB =20
log,
(4.6)
50
summary.
Spec
D3
cise
ynamic range
aximum input
Value
51 dB (Fig. 4.4)
47 dBc at
kHz (Fig. 4.5) 7.25 dB
00 mVp
eltage
C. Bandpass Filter
As discussed before, the most critial block of the system is the bandpass filter. A very
and high
quahty
to provide a reliable
and are
of the
of this
Fig. 4.6
frequency
is 32 kHZ (1 kHz
center trequency)
f'requency
4tH 8P~Z'f&Z
RACK
Tice.
fZ". N PN
003.$$
Hg
4t
'
Af)V
52
follows an almost constant value up to 8 kHz; after this value, the error increases due to
If a higher
of the
to 9 kHz is shown. The value moves between 380 and 420, but this is mainly due to
measurement
39 5
38 5
nl
38
'o
r
375
37 36 5
u-
35 5 1000
2000
3000
4000
500D
60DO
TDOD
800D
9000
Center frnquencyfo
fHz)
500
ty
490
rr 460
440
4
420
400
3000
TDOO
8000
900D
ate:
08-38HZ,
Vice'. 84
% PN
C&at er '
YRRCK'
R'
IHr HRHHHz
CHR
Irttt Hzrkzr
IC
/div
118
Czniart
HHz
55
(4.11)
=20 log
of these
',
=84. 17dB
(4. 12)
summary.
Spec
Value
cise
1'
uxituutu iuput
302.6 Vp
1
altage
requeucy rauge
kHz
9 kHz
of 4 dB each
of
values. It can be noted that the obtained results have a gain error below 0.25 dB.
Y w-~-*"
-s.f
+w
T
I
RACE
Rg
888/HX
-RS
$2
PS8
8 AS 8
Hx
-137 817 de
Hz
ORHQR
18
/III
CIR Y
!
.
818PI IC .RHR
CHAPTER V
CONCLUSIONS
Integrated circuits are becoming more complex as the featured size decreases. This trend
allows us to implement more functions on a single chip, making them more prone to fail.
Usually, each analog circuit requires a "custom made" testing circuit. Then, the need of
finding a umversal test technique for analog circuits is a must.
This work deals with some of the many problems that analog testing engineers find very
often. Some of the goals to achieve are to use an external digital tester instead
of the
analog or mixed-signal tester to reduce the test nme as well as facilitate the test program
generation. An important
issue to consider is the silicon area used for the testing circuit,
since it should not represent a large area overhead for the whole circuit. Also, the testing
when it is working
in normal
mode.
of
analog ICs has been proposed. The idea behind this work is based on the most popular
test techniques: spectrum and network analyzers. Using this approach there is not need
to reconfigure the DUT. Since one
the use
of the
of this
generator, a high-g bandpass filter and a voltage gam amplifier (VGA). An analog-todigital converter can be added, but it is not included in this system. All
use switched-capacitor
techniques to ensure synchronization.
of these blocks
59
The required features for the sinewave generator included a dynamic range larger than
45 dB, center frequency and amplitude programmabihty
all the other blocks. The sinusoidal as well as synchronization
with
of an
ideal
sinewave and since its amplitude is set by a reference voltage and can provide the proper
dynamic range for the DUT, the input VGA was removed. The proposed architecture
requires a very simple digital logic to control the step sizes of the sinusoidal signal and
results.
The bandpass filter was the block with more constrains. Since a high quality factor and a
very precise center frequency are a must, special care has to be taken in its design. The
bilinear transformation;
factor yield to a very high capacitive spread, but this can be reduced by using a different
sampling
frequency
(fin) for the capacitor that controls the Q of the filter: in order to
the equivalent
resistance
is increased,
reducing
the
The voltage gain amplifier is used to provide a proper dynamic range to the ADC, as
well as to avoid a reduction in resolution due to small input signals. For this purpose, a
variable gain amplifier with six gain steps of 4 dB each (from zero to 20 dB) is used. The
structure
used is offset and gain compensated
of
errors due to the dc offset and finite gain of the opamp. The gain is set externally
The complete system was fabricated through the MOSIS service, using the AMI 0.5 )im
analyzer
was
in Chapter
III.
In summary,
has been
60
synthesizer.
Magnitude
of the DUT
used,
full
can
be
measured.
Since
switched-capacitor
techniques
are
being
synchronization
of the
61
REFERENCES
"Metrics, Techniques
in Mixed-Signal
at the International
Test Conference
1999, IEEE
I, 1999.
Testing:
An
"Metric
and
Techniques
of Mixed-Signal
presented
[3] A.
Mixed-Signal
Testing", presented
[4] B. Provost, "Design Methodologyfor Mixed-Signal AC BIST and ADC SelfCahbration" Ph. D. Dissertation, Texas A&M University, May 2002.
1998.
Mountreux,
[6] J. M.
Byrd,
F. Caspers,
Japan-Russia
Switzerland,
[7] J.
Cowles and
B. Gilbert, "A
Cahbrated
2001 IEEE MTT-S Microwave Symposium Digest, Vol. 3, Phoenix, Arizona, pp.
2163-2161, 2001.
[8] L. T. Lin,
Monolithic
FL Tseng, D. Cox,
S. S. Viglione, D. P. Conrad
and
R. G. Runge, "A
State Circuits,
[9) Y. Kuraishi, K.
Channel
Filter Bank",
IEEE Journal of Solid State Circuits, Vol. SC-19, No. 6, pp. 964-
62
[10]J. S.
Chang,
Time Multiplexed
SC
[11]R. Courteau, T. K.
Synchronous
Measurement,
Sampling",
IEEE Transactions
Instrumentation
and
[12] HP
March
1990.
Short Time Founer Transform
Department
Spectruni Analyzer:
A
[13]L.
Chan, A Continuous
of Electrical
Engineering,
1995.
Chiappano,
[14]G. Boarin, G.
Implementation
F.
Maloberti,
S.
Napolitano,
M. Porta, "VLSI
of the
Terminations",
[15]J.
Silva-Martinez,
E. Sanchez-Sinencio,
"Excess Phase-Jitter
Cancellation
Method
Systems,
for SC Relaxation
on Circuits
and
[16]H.C.
Patangia,
B.
Zenone,
"A Programmable
the 37th Midwest
Switched-Capacitor
Symposium
Sinewave
Generator",
Systems,
Proceedings
of
on Circuits
and
F. Malobern.
Circuits",
Individually
Switched-Capacitor
Symposium
in Switched-Capacitor
on
63
W. [19]P.
Li, M.
J.
Chin,
P. R.
Gray and
Algorithmic
Analog-to-Digital
Conversion
of Sohd
[20]R. H.
McCharles,
Algorithmic
Analog-to-Digital
IEEE International
Solid State
February 1975.
NY, 1997.
64
APPENDIX
A. Bandpass Filter Design
In Chapter III we discussed the filter architecture
filter
using
swtiched-capacitors
and
a bilinear
is proposed.
The
~ ~
f, =1kHz
f, = 32 kHz g =50
HD3
m (T, = 31.25x10-6 s. )
~
~
( -42 dB
of a bandpass
filter is given by:
I')
s +
where ra, is given by
s+co
(A. 1)
ra.
=2
rr
f, =2
rr
(1000)=628318x10' rad(sec
(A. 2)
In order to simplify the calculations, we will use the previous equation in the following
form:
H(s) =
s'+a s+b
(A. 3)
65
where a
and b are
given by:
a=
rad / sec
(A. 4)
b = m,
=(6.28318x10') = 39.478x10'rad/sec
of
the filter can be found with the
(7=
BW
' mBW=
(A. 5)
As we know, a very precise aproximation is needed in order to obtain the desired center
frequency;
hence the bilineal transformation
function
in the z-
2 'I S=
'
I+z'
(A.6)
-2
H(z) =2
T.
b. T' 2
T+4
'+
2 b
T 8 z
'+
T'+2
a T+4
(A7)
Now, we
will apply
the prewarping
as follows
66
For the
central frequency
of 1 kHz
of 32 KHz:
CrP
31.25 x10 ,
tan
( 6.28318x10'
2
31.25x10 '
(A. 9)
of the
a = 126.0689963
b = 39.73347x106
Substituting
the values
of
transfer
algebra, we obtain:
H(c) = 6.419x10
'.
z
' 1.8596
'
+1.01284
(A. 10)
The previous transfer function is compared with the block diagram of the switchedcapacitors
implementation
in order
to obtain
the capacitor
values
required.
The
implementation
of
respectively.
67
c
2
C
2
V,,
C
C,
z'
i-z-'
Cz
=' C, i
C,
V,
z' C, i
C, C,
i-z-'
C, C,
bandpass filter.
of the
68
" (-'):::((-'-I)
And after some algebra:
C4 C5
r
C,
1-z'
C, C,
1-z'
(A. 1 1)
Co
C4 C5
Co
H(z)
C, C3
Ci
Ci C3
C,
(A. 12)
z'+
or
'
"
z'+
1+
(Co 'Ci
Co 'C~
H(z)
4
(A. 13)
By comparing both the block diagram and the filter transfer functions, we can easily find
the required values for the capacitors. The obtained values summarized
on table
3.1 are
C, C,
C4 C,
and
C,
=C,
C, =C4
(A. 14)
B.
69
B.
The operational amplifier is a common element is all the block of the system proposed in
this work.
~ ~ ~
Figure
Bl
and three
current mirrors with cascode stage output. The design procedure is carried out next.
M4
M4
M4
Ib
M4c
Msc
Mio
M,
M4
Figure
B.l. Opamp
schematic.
70
V,
=0.79V
=0,93 V
A/V
Vip
~
~
K= 114.6x10
K~
= 38x10 A/V
1'4' Visas
t,
'
CL
(B.1)
where Vqu is fixed to 50 mV because we are using a cascode structure and we need to
ensure there is enough room to keep all the transistors in the saturation region. t, is the
0.720
of 8 liA
can be used.
The
required
for the differential pair in order to charge the capacitive load in 4.r is
given by:
gmci
= 16*C,
(B.2)
M&
and
Mi. Using:
L,
ID
(B.3)
71
Mq, we
using:
"34
Is
Vdsof3, 4
(B 4)
L,
2 K ID
(B.S)
(B.6)
The cascode transistors Mi c and M6i- have the same sizes as M5 and M6. The gain of the opamp is related to the minimum
by:
length
of transistor M6
which is given
1x10 ' 3
6
ID
(B.7)
g~i Vs
Even though the Early Voltage V,~ does not have an exact value, it gives a good
approach to a good value for this transistor length.
For M5
and Mrs
72
Vd, , and
using:
2. ID
Vaoa4
(B.9)
L,
K.io
(B.10)
Table
opamp transistors.
Transistor
Mn Mt
M3, M4
8'(Ian)
13.2
L(tom)
1.2 1.2
3.86
3.86
1.316
2.63
4.8 4.8
1.2
1.2 1.2
Mz Mg, Msc
My, Mts
1.62
3.3
73
C.
Abbreviations
and Acronyms
Error
ADC:
Ave
Analog-to-Digital
Converter
DC gain
Built-in self testing
Load capacitance
BIST:
Cr. r
clkr
Clock
Digital-to Analog Converter
Desig for testability
DAC:
DFTr
DUT:
for
GBP't
Gain-Bandwidth
product
HD3:
IC:
Integrated circuit
Intermediate
frequency
IF:
LO:
Local oscillator
Least significant bit
Operational amplifier
LSB:
Opampt
Programmable
Quality factor
Logic Array
74
R88'r
RFr
ROM:
STr
THDr
VCOr
VGA:
V.e'er
Reference voltage
75
VITA
B.E. degree
the M. S. degree
in Electronics
Nacional
de Astrofisica,
Optics y
Center in 2002. She is working towards her Ph. D. degree in Electronics in the Instituto
Nacional de Astrofisica, Optics y Electronics. During her studies at INAOE she worked
interests
are analog
testing
and switched-
Her permanent
address
is Otono
Irapuato,