You are on page 1of 7

IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 35, NO.

12, DECEMBER 2000 1877

A CMOS Nested-Chopper Instrumentation Amplifier


with 100-nV Offset
Anton Bakker, Kevin Thiele, and Johan H. Huijsing, Fellow, IEEE

Abstract—A CMOS nested-chopper instrumentation amplifier


is presented with a typical offset of 100 nV. This performance is ob-
tained by nesting an additional low-frequency chopper pair around
a conventional chopper amplifier. The inner chopper pair removes
the 1 noise, while the outer chopper pair reduces the residual
offset. The test chip is free from 1 noise and has a thermal noise
of 27 nV/ Hz consuming a total supply current of 200 A.
Index Terms—Chopper amplifiers, instrumentation amplifiers,
low-offset amplifiers, offset cancellation techniques.

I. INTRODUCTION
Fig. 1. Noise power spectrum of standard CMOS operational amplifier.

I N many applications, such as sensor interfaces, the overall


performance of the system is limited by the offset and noise
of the input amplifiers. This problem has been growing in the II. DYNAMIC OFFSET-CANCELLATION TECHNIQUES
past years, because of the shift from bipolar to CMOS pro- A. Offset and Noise in CMOS Amplifiers
cesses which have significantly higher noise and offset.
Also a conventional offset-cancellation technique such as trim- A conventional CMOS amplifier has a typical input-referred
ming, which is widely used in bipolar technology, is much less noise spectrum, as shown in Fig. 1. For rather high frequen-
beneficial in CMOS technology because it can not reduce the cies, the noise can be considered as frequency independent or
noise. Solutions that can remove both offset and noise white. This is usually called the thermal noise floor. At low fre-
are found in the dynamic offset-cancellation techniques. Exam- quencies, the noise power is increasing almost linearly with de-
ples of these are the autozero and chopper techniques, which creasing frequency and is therefore commonly called noise.
will be explained in this paper. Derivatives of these techniques The frequency at which the noise becomes dominant over
are found in all commercial ultra-low-offset CMOS operational the white noise is called the noise corner frequency .
amplifiers. Typical offset figures of these kinds of operational At very low frequencies, offset becomes the dominant error.
amplifiers are in the range of 10 V. Although these offset fig- Although offset is usually modeled as a time-invariant voltage
ures are already very low, still some applications require an even source, it may change due to aging and temperature variations.
lower offset. This implies that it has a certain bandwidth and can therefore be
In this paper, an instrumentation amplifier for read-out of a considered as a very low-frequency noise source.
spinning-current Hall plate is described [1]. This Hall plate is
B. Classification
integrated in a standard CMOS process and has an offset of less
than 500 nV. The offset of the instrumentation amplifier needs Due to historical reasons, some confusion has arisen in the
therefore to be well below this 500 nV. This paper discusses naming conventions of the different dynamic offset-cancellation
the design and realization of this ultra-low-offset instrumenta- techniques. Nowadays, it is generally accepted to distinguish
tion amplifier. A new dynamic offset-cancellation technique is two main groups: autozeroing and chopping [2]–[4]. The fun-
shown that can reduce the offset of a CMOS amplifier to typi- damental difference between them is the offset handling. While
cally 100 nV. the autozero principle first measures the offset and subtracts it in
a next phase, chopping modulates the offset to higher frequen-
cies, which will be explained further in this paper.
In data books and literature, many derivatives of these
two basic offset-cancellation techniques can be found, like
correlated double-sampling [2], [4], ping-pong opamps [5], [6],
self-calibrating opamps [7], synchronous detection, the two-
Manuscript received April 17, 2000; revised July 15, 2000.
A. Bakker and K. Thiele are with Philips Semiconductors, Sunnyvale, CA or three-signal approach [8], and dynamic element matching
94088 USA. (DEM). The name chopper stabilization is even used for both
J. H. Huijsing is with the Electronic Instrumentation Laboratory, Delft Insti- autozeroing and chopping techniques [2], [9].
tute of MicroElectronics and Submicron Technology (DIMES), Delft University
of Technology, The Netherlands. Table I shows the above-mentioned techniques classified into
Publisher Item Identifier S 0018-9200(00)10050-2. the two main groups: autozeroing and chopping. The next para-
0018–9200/00$10.00 © 2000 IEEE
1878 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 35, NO. 12, DECEMBER 2000

TABLE I
CLASSIFICATION OF DYNAMIC OFFSET-CANCELLATION TECHNIQUES

Fig. 3. Noise power spectrum of autozeroed amplifier.

modulated only once and appears at the chopping frequency and


its odd harmonics. These frequency components need to be re-
moved by a low-pass filter. Next to the frequency domain, the
chopping principle can also be explained in the time-domain.
In that case, the input signal is periodically inverted by the
first multiplier or chopper. After amplification, the inverted and
amplified signal is inverted for the second time, resulting again
in a dc signal. The offset is periodically inverted only once and
Fig. 2. Principle of autozero technique. therefore appears as a square wave at the output.
In contrast to the increased white noise component of au-
graphs will discuss the basic characteristics of the autozero and tozero amplifiers, the baseband noise of chopper amplifiers is
the chopper technique. almost equal to the wideband thermal noise, assuming again
that the chopping frequency is higher than the noise corner
C. Autozero Technique frequency. The typical noise power spectrum of a chopper am-
The principle of the autozero technique is shown in Fig. 2. plifier is shown in Fig. 5. The folded noise components
The main characteristic of the autozero technique and its deriva- are omitted for simplicity. The reason why the residual noise of
tives is that the offset cancellation is done in two phases. A sam- a chopper amplifier is fundamentally lower than that of an au-
pling phase when the offset is measured and sampled on tozero amplifier is that the input signal of a chopper amplifier is
and an amplification phase when the sampled offset is sub- not sampled, which makes it impossible for wideband thermal
tracted from the input signal and amplified. This technique is noise to fold back into the baseband.
very well-known and an improved derivative is applied in all The lower noise of the chopper technique is the main reason
commercially available chopper-stabilized opamps. The reason to use this technique for read-out of our spinning-current Hall
to mention it here is the noise performance of this technique. plate. However, the residual offset of the chopper technique is
Besides the offset, the autozero technique also removes the still too high for this application. The rest of this paper will
noise of the amplifier, which makes sense because offset therefore focus on techniques to further reduce this residual
can be considered as low-frequency noise. To remove all the offset of a chopper amplifier.
noise the autozeroing frequency should be higher than
the noise corner frequency. The typical noise power spec- E. Origins of Offset in Chopper Amplifiers
trum of an autozero amplifier is shown in Fig. 3. The residual The residual offset of a chopper amplifier is in the order of a
noise at frequencies lower than is almost white. However, it is few tens of V. To be able to reduce this offset, first the origins
not equal to the thermal noise floor, as it is increased by the ratio of the offset need to be explored.
of the unity-gain bandwidth of the amplifier and the autoze- The residual offset of a chopper amplifier originates mainly
roing frequency . The reason for this is that due to the sam- from the spikes of the input chopper [2]. In turn, these spikes
pling action, high-frequency components are folded back to the originate from the charge injection mismatch of the switches.
baseband. The higher the bandwidth of the amplifier, the more After demodulation of the spikes a residual offset occurs. This
noise is sampled on the capacitor . A more precise explana- is schematically shown in Fig. 6. It can be seen that the residual
tion is given by Enz et al. in [2]. In practical situations, the ratio offset is determined by the number of the spikes and the
is somewhere between three and five. This is the funda- energy content of the spikes. This energy content is dictated
mental reason why, for example, chopper-stabilized opamps al- by the input impedance and the mismatch in parasitic ca-
ways have rather high noise figures in the order of 70 nV/ Hz. pacitive coupling between the chopping signal and the
input lines, which is presented by . There are three main
D. Chopper Technique options to reduce the residual offset: 1) lowering the chopping
The principle of the chopper technique is shown in Fig. 4. frequency; 2) lowering the input impedance; or 3) lowering the
The input signal is modulated to the chopping frequency, charge injection. However, lowering the chopping frequency is
amplified and modulated back to the baseband. The offset is not a real solution, because the chopping frequency should be
BAKKER et al.: CMOS NESTED-CHOPPER INSTRUMENTATION AMPLIFIER 1879

Fig. 4. Chopping principle including signals in frequency and time domain.

Fig. 5. Noise power spectrum of chopper amplifier.

higher than the noise corner frequency to remove the


noise. The input resistance is dictated by the input signal
source and can usually not be lowered by the designer. Charge
injection is mainly dictated by the process choice and can be
minimized by the designer by using small transistors that are
well matched. The final conclusion is that except for careful
layout design, the designer can not improve the residual offset Fig. 6. Residual offset caused by spikes. (a) Spike signal. (b) Demodulation
in a straightforward manner. signal. (c) Demodulated spike.

F. Techniques to Reduce the Residual Offset spikes. However, a high introduces gain accuracies if there
An interesting nonstraightforward method to reduce the is a mismatch between and . For a mismatch between
residual offset is shown by Menolfi et al. [10]. They remark and of 1%, a good compromise value for is between
that the energy content of the spikes is mainly located at higher three and five. This value gives a residual offset of 500 nV
harmonics of the chopping frequency, while the energy of the which is the best reported value so far. As already mentioned
modulated signal is mainly located at the fundamental of the before, the major drawback of this circuit is the gain accuracy.
chopping frequency. If the modulated signal that includes the This gain accuracy is dependent on the quality factor and the
spikes is low-pass or bandpass filtered, almost all spikes are matching between and . This implies that already in a
removed, while only a small part of the signal is lost. This relatively small temperature range, large deviations in accuracy
idea is schematically shown in Fig. 7. Compared to a conven- will occur. The accuracy can also not be improved by applying
tional chopper amplifier, a bandpass filter is added within the feedback because of stability problems caused by the phase
amplifier. The center frequency of this bandpass filter is response of the bandpass filter. In conclusion, this technique
equal to the chopping frequency . The quality factor significantly reduces the residual offset at the cost of reduced
should be high, to increase the attenuation of the unwanted gain accuracy.
1880 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 35, NO. 12, DECEMBER 2000

Fig. 7. Chopper amplifier with bandpass filter to improve residual offset.

Fig. 8. Nested chopper amplifier principle.

III. NESSTED-CHOPPER TECHNIQUE


A new technique to reduce the residual offset of a chopper
amplifier is shown in Fig. 8, which will be referred to as the
nested-chopper technique. The basic idea is to consider a con-
ventional chopper amplifier as a regular amplifier without
noise and a reduced offset. The offset of this amplifier can be re-
duced by applying another pair of choppers, but now operating
at a much lower frequency. This frequency can be much lower
than the noise corner frequency, because the noise is al-
ready removed by the inner pair of choppers. Because the outer
pair is working at a much lower frequency the residual offset due
to spikes of these choppers is much lower. The corresponding
signals of the nested-chopper amplifier are shown in Fig. 9. The
phase relationship of both signals is not important. This implies
Fig. 9. Reduction of residual offset by nested chopper amplifier. (a) Spikes
in practical situations that both signals will have synchronous after first demodulator. (b) Low-frequency modulation signal. (c) Spikes after
edges. The spikes that are generated by the high chopping fre- second demodulator.
quency are modulated by the output chopper with a
frequency . The average energy of the spikes has be-
come zero now, resulting in a residual offset that is theoreti- 100 nV, without increasing the noise and with the possibility to
cally zero. If we take into account the spikes generated by the apply feedback.
low-frequency choppers, the theoretically achievable improve-
ment in residual offset is the ratio of and . For IV. REALIZATION
practical values of and of 2 kHz and 20 Hz,
respectively, the residual offset will be 100 times less. This im- To test and prove this theory, an instrumentation amplifier
plies that it should be possible to reduce the residual offset to employing the proposed nested-chopper technique is de-
less than 100 nV. Compared to the above-mentioned bandpass signed. The schematic is shown in Fig. 10. A major problem
chopper amplifier, the nested-chopper amplifier can be fed back, in testing low-offset amplifiers is the effect of parasitic
which implies that the gain-accuracy can be very high over a thermocouples at the input when connecting an external test
wide temperature range. Also the implementation is very simple signal. To avoid disturbances by these thermocouple effects,
as only one extra pair of choppers and a clock signal are needed. we tested the system with an on-chip spinning-current
A disadvantage is that the maximum input signal frequency is magnetic Hall sensor [1]. Switching off the Hall plate’s bias
reduced to half . However, a bandwidth of a few tens of current makes the signal at the input of the amplifier zero
Hertz is sufficient for many data-acquisition applications. without changing the source impedance. The amplification
The conclusion on the proposed nested-chopper amplifier factor of the amplifier is determined by resistors , ,
technique is that the offset can be reduced to values as low as and and is set to 100.
BAKKER et al.: CMOS NESTED-CHOPPER INSTRUMENTATION AMPLIFIER 1881

Fig. 10. Nested chopper instrumentation amplifier with spinning-current Hall sensor input.

Fig. 11. Detailed schematic of chopper opamp including feedback resistors.

The low-pass filter is a first-order one with a 3-dB roll-off mines the overall noise performance and has an equivalent input
frequency of 3 Hz. This low-pass filter is added externally, be- thermal noise of approximately 15 nV/ Hz while drawing only
cause of the required large RC constant. In a commercial ver- 35- A tail current. The second stage consists of ,
sion, this low-pass filter can be integrated on the chip by making and and is optimized for high gain and low transconduc-
it a part of an integrating A/D converter, such as a sigma–delta tance. The high gain is necessary to reduce the influence of off-
A/D converter. sets of the subsequent stages outside the choppers. The reason
The instrumentation amplifier can be divided in two equal for the low transconductance is to have a low unity-gain fre-
amplifiers. One of these amplifiers is shown in detail in Fig. 11. quency which implies an intrinsic filtering of the modulated
The amplifier is a four-stage operational amplifier. The first offset. This will be discussed more elaborately later. The low
stage which is formed by , and is a low-gain transconductance is achieved by using a small tail current of
low-noise stage. This stage has a gain of approximately twenty. only 500 nA and a very low W/L ratio. The first two stages of the
The noise performance is optimized firstly by using resistors amplifier are kept fully differential to reduce effects of charge
and instead of an active load and secondly by choosing injection of the chopper switches. To define the common-mode
a high over ratio of input transistors and in such a voltage on the drains of and , a common-mode control
way that they are biased in weak inversion. The first stage deter- circuit is necessary. This is done by measuring the common-
1882 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 35, NO. 12, DECEMBER 2000

Fig. 13. Input referred offset versus chophigh frequency; resolution of


measurement is 50 nV.
Fig. 12. Microphotograph of the test chip.
TABLE II
PERFORMANCE SUMMARY

mode current through and , and use this value to


control the currents and . However, in the figure this is
omitted for simplicity.
The third stage is formed by , and and
acts as a Miller stage. This stage splits the dominant poles of the
high-gain second and third stages. The fourth stage is formed by
and and has a gain of approximately one.
One of the reasons for the Miller stage is to assure stability
of the opamp. However, this stage also acts as a low-pass filter
for the modulated offset of the first two stages. To have the
most aggressive filtering, the transconductance of the first two
stages should be as low as possible. However, the first stage Commond mode Rejection
should have a high transconductance to achieve low noise
performance, which implies a trade-off. The second stage has V. MEASUREMENT RESULTS
therefore a transconductance that is made as low as possible Nine chips from two different batches have been tested. The
without being a dominant noise source for the whole amplifier. input referred offset versus has been measured for
With a gain of 20 of the first amplifier, the transconsuctance of values of between 2 and 50 kHz. Experimental results
the second amplifer can be 400 times lower to still not be the show that for is 2 kHz and a corresponding
dominant noise source. For our circuit, the transconductance of 16 Hz, the input referred offset is below 100 nV for all nine
of the first stage, is set to 700 A/V and for the second samples. It also shows a significant increase for higher frequen-
stage, equals 2 A/V. The unity-gain frequency is given by cies. The results for values of between 2 and 8 kHz are
, which equals 400 kHz. For an amplication shown in Fig. 13. The minimal value of 2 kHz is exactly equal
set by R41 and R42 ton100, this gives an amplifier bandwidth to the noise corner frequency.
of 4 kHz. The open-loop dc gain is more than 130 dB. A remarkable result is that the offset is not dependent on
The current consumption of the instrumentation amplifier , but on . The reason for this seems to be the
is 200 A. A chip microphotograph is shown in Fig. 12. The mismatch of the on-resistance of the low-frequency chopper
circuit is implemented in a single-poly double-metal 1.6- m switches. This is explained in Fig. 14. If the input impedances
CMOS process. The die area is 6 mm . A large part of the die are not exactly equal, the area under the spikes do not com-
area is occupied by the two 16-pF metal1–metal2 capacitors. pletely cancel. This results in a residual offset due to the high-
These capacitors could not be made out of MOS capacitors, frequency spikes that is not exactly zero. If this is true, the dom-
because they need to be very linear, because of the switching. inant contribution to the residual offset does not come from the
Because our process lacks the availability of double-poly linear low-frequency choppers but from the high-frequency ones. This
capacitors, metal1–metal2 capacitors were the only possible explains why the residual has approximately the value that could
implementation. be expected from theory, but is still dependent on .
The higher chopping frequency is applied externally. The The noise is measured in a bandwidth of 0.1–3 Hz and was
lower chopping frequency is derived from found to be 27 nV/ Hz, which is in very good concordance with
by a frequency divider of 128. The divider value can also be the simulated thermal noise. The common-mode rejection ratio
changed to 512 for test purposes. is also measured in a 0.1–3-Hz band and was found to be over
BAKKER et al.: CMOS NESTED-CHOPPER INSTRUMENTATION AMPLIFIER 1883

Fig. 14. Explanation of the result that the residual offset is dependent on f .

140 dB. This shows again the excellent common-mode rejec- [10] C. Menolfi and Q. Huang, “A fully integrated CMOS instrumentation
tion ratio (CMRR) performance of chopper amplifiers, which amplifier with submicrovolt offset,” IEEE J. Solid-State Circuits, vol.
34, pp. 415–420, Mar. 1999.
was already known from previous designs [2]. A performance
summary is shown in Table II.

VI. CONCLUSION Anton Bakker was born in Amsterdam, The Nether-


lands, on July 11, 1968. In 1991, he received the
In this paper, a nested-chopper technique is presented and M.Sc. degree in electrical engineering from the Delft
compared with other dynamic offset-cancellation techniques. It University of Technology, Delft, The Netherlands.
In 1996, he started his Ph.D. project on CMOS smart
is shown that this new technique can significantly reduce the temperature sensors. During this research period, he
offset of conventional chopper amplifiers at the cost of only one designed a number of temperature sensors for Philips
additional chopper pair. A nested-chopper instrumentation am- Semiconductors, Sunnyvale, CA. He received the
Ph.D. degree from Delft University in 2000.
plifier for a spinning-current Hall plate is realized and measure- In 1991, he joined the Werkgroep Elektrotechnisch
ment results show a residual offset of only 100 nV. This offset Practicum, where he developed a laboratory course
value is the lowest ever reported. A disadvantage is the limita- for second-year students on the design of complex integrated circuits. In 1993,
he joined the Electronic Instrumentation Laboratory where he was involved in
tion of the maximum input signal frequency to half . a European Project (ESPRIT) on the design of an ultra-low-power tempersature
However, a bandwidth of a few tens of hertz is sufficient for sensor. During this project, he spent three months at CSEM, Neuchâtel, Switzer-
many data-acquisition applications. land, to implement his design. He is currently with Philips Semiconductors.

REFERENCES
[1] A. Bakker, A. A. Bellekom, S. Middelhoek, and J. H. Huijsing,
“Low-offset low-noise 3.5-mW CMOS spinning-current Hall effect Kevin Thiele, biography and photograph not available at time of publication.
sensor with integrated chopper amplifier,” in Proc. Eurosensors XIII,
Sept. 1999, pp. 1045–1048.
[2] C. C. Enz and G. C. Temes, “Circuit techniques for reducing the effects
of opamp imperfections: Autozeroing, correlated double sampling, and
chopper stabilization,” Proc. IEEE, vol. 84, pp. 1584–1614, Nov. 1996. Johan H. Huijsing (SM’81–F’97) was born in Ban-
[3] C. C. Enz, E. A. Vittoz, and F. Krummenacher, “A CMOS chopper am- dung, Indonesia, on May 21, 1938. He received the
plifier,” IEEE J. Solid-State Circuits, vol. SC-22, pp. 335–342, June M.Sc. degree in electrical engineering from the Delft
1987. University of Technology, Delft, The Netherlands, in
[4] K. C. Hsieh, P. R. Gray, D. Senderowicz, and D. G. Messerschmitt, 1969, and the Ph.D. degree from the same university
“A low-noise chopper stabilized differential switched-capacitor filtering in 1981.
technique,” IEEE J. Solid-State Circuits, vol. SC-16, pp. 708–715, Dec. He has been an Assistant and Associate Professor
1981. in electronic instrumentation with the Faculty
[5] C. G. Yu and R. L. Geiger, “An automatic offset compensation scheme of Electrical Engineering, Delft University of
with ping-pong control for CMOS operational amplifiers,” IEEE J. Technology, since 1969, where he became a full
Solid-State Circuits, vol. 29, pp. 601–610, May 1994. Professor in the Chair of electronic instrumentation
[6] I. E. Opris and G. T. A. Kovacs, “A rail-to-rail ping-pong opamp,” IEEE in 1990. From 1982 through 1983, he was a Senior Scientist at Philips Research
J. Solid-State Circuits, vol. 31, pp. 1320–1324, Sept. 1996. Labs, Sunnyvale, CA. Since 1983, he has been a Consultant for Philips. His
[7] (1999, May) TLC4501, Self-calibrating operational amplifier. Texas In- research work is particulary focused on the systematic analysis and design of
struments Inc., Dallas, TX. [Online]. Available: http://www.ti.com operational amplifiers and integrated smart sensors. He is author or co-author
[8] G. C. M. Meijer, “Concepts and focus point for intelligent sensor sys- of some 150 scientific papers, 20 patents and four books, and co-editor of
tems,” Sensors and Actuators, vol. 41, pp. 183–191, 1994. five books. He is initiator and Co-Chairman of the International Workshop on
[9] (1999, June) LTC1050, Precision chopper stabilized operational ampli- Advances in Analog Circuit Design, which has been held annually since 1992.
fier with internal capacitors. Linear Technology, Milpitas, CA. [Online]. He is Chairman of the biannual National Workshop on Sensor Technology,
Available: http://www.linear.com since 1991, and Chairman of the Dutch STW Platform on Sensor Technology.

You might also like