IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 56, NO.

12, DECEMBER 2021 3583

A Fill-In Technique for Robust IMD Suppression

in Chopper Amplifiers
Thije Rooijers , Member, IEEE, Shoubhik Karmakar , Student Member, IEEE,
Yoshinori Kusuda , Member, IEEE, Johan H. Huijsing, Life Fellow, IEEE,
and Kofi A. A. Makinwa , Fellow, IEEE

Abstract— In chopper amplifiers, the interaction between the

input signal and the chopper clock can give rise to intermod-
ulation distortion (IMD). This chopper-induced IMD is mainly
due to amplifier delay, which causes large pulses at the output
of the amplifier’s output chopper. This article proposes the use
of a so-called fill-in technique to eliminate these pulses, and thus
the resulting IMD, by multiplexing the outputs of two identical
amplifiers that are chopped in quadrature. A prototype chopper-
stabilized amplifier was implemented in a 180-nm CMOS process.
Measurements show that the fill-in technique suppresses chopper-
induced IMD by 28 dB, resulting in an IMD of −126 dB for input
frequencies near 4FCH (=80 kHz). It also improves the amplifier’s
two-tone IMD (with 79 and 80 kHz inputs) from −97 to −107 dB,
which is the same as that obtained without chopping.
Index Terms— Auto-zeroing, chopping, dynamic offset compen-
sation (DOC), fill-in technique, intermodulation distortion (IMD),
offset.

I. I NTRODUCTION

C MOS amplifiers suffer from offset (several millivolts),

offset drift (several μV/◦ C), and 1/ f noise corner (tens
of kilohertz). Dynamic offset compensation (DOC) techniques,
such as auto-zeroing and chopping, are often used to achieve
low offset (microvolt level), offset drift (<20 nV/◦ C), and 1/ f
noise corners (several hertz). However, these techniques also
have drawbacks. Some of these, such as the introduction of
glitches, increased input current, and residual offset due to
the charge-injection (mismatch) of switches, have been well Fig. 1. Amplitude spectrum for: (a) traditional choice of FCH (FCH  Fin )
studied [1], [2] and can be effectively mitigated [3]–[8]. This and (b) the choice in this work (FCH < FIN ), showing the resulting ripple,
has led to precision CMOS amplifiers with no visible spectral spikes, and IMD tones.
content at the auto-zero/chop frequencies [6]–[8], very low
input current (1 pA untrimmed, 0.2 pA trimmed [8]) and very
low residual offset (100 nV [3]). This article focuses on the will occur at the sum and difference of Fin and multiples of the
mitigation of a lesser-known drawback of DOC techniques: DOC frequency FDOC , i.e., at Fin ± n FDOC and n FDOC ± Fin ,
the generation of intermodulation distortion (IMD). In the where n is integer. In the case of auto-zeroing, this is due
presence of an ac input at a fixed input frequency (Fin ), IMD to its inherent sample and hold operation, which causes
noise folding and large odd-order (n = 1, 3, 5, . . .) IMD
Manuscript received May 10, 2021; revised July 21, 2021; accepted tones [8], [17], [18]. In the case of chopping, the main cause
August 20, 2021. Date of publication September 6, 2021; date of current
version November 24, 2021. This article was approved by Associate Editor of even-order (n = 2, 4, 6, . . .) IMD is amplifier delay, as will
Man-Kay Law. (Corresponding author: Thije Rooijers.) be discussed later in more detail.
Thije Rooijers, Shoubhik Karmakar, Johan H. Huijsing, and In many chopper amplifiers, the chopping frequency (FCH )
Kofi A. A. Makinwa are with the Department of Microelectronics, Delft
University of Technology, 2628 CD Delft, The Netherlands (e-mail: is much higher than Fin . The resulting input-referred amplitude
c.t.rooijers@tudelft.nl). spectrum is shown in Fig. 1(a). It may be seen that the ripple
Yoshinori Kusuda is with Analog Devices, Inc., San Jose, CA 95134 USA. caused by the up-modulated offset and 1/ f noise can then be
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/JSSC.2021.3107350. suppressed by a simple low-pass filter (LPF). The same filter
Digital Object Identifier 10.1109/JSSC.2021.3107350 will also suppress the tones near 2FCH due to charge-injection
0018-9200 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Hanyang University. Downloaded on August 23,2023 at 07:00:53 UTC from IEEE Xplore. Restrictions apply.
3584 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 56, NO. 12, DECEMBER 2021

glitches and chopper-induced IMD. However, to achieve a

large tone-free signal bandwidth (BW), FCH must be quite
high, and will certainly be higher than the amplifier’s original
1/ f corner frequency. However, increasing FCH results in
lower effective gain, higher input current, and residual offset.
For example, in [9], FCH = 333 kHz, which results in 50
μV offset, 0.25 μV/◦ C offset drift, and 550 pA input current.
In [10], inter-leaved 800 kHz clocks are used to achieve an
effective FCH = 4.8 MHz. This results in 5 μV offset and
0.02 μV/◦ C offset drift, but trimming is required to lower its
input current to acceptable levels (from 1.5 nA to 150 pA).
This work focuses on chopper amplifiers in which FCH is
set only slightly higher than the 1/ f corner frequency, and
so will typically be lower than Fin . This choice results in
the lowest residual offset, lowest input current, and largest
effective gain, while still effectively suppressing 1/ f noise.
For example, in [11], FCH is 30 kHz, which results in 1 μV
offset, 22.5 nV/◦ C offset drift and an input current of 110 pA.
However, as shown in Fig. 1(b), a simple LPF can no longer
be used to filter out chopper ripple, and so other techniques
are required, such as the use of a ripple-reduction loop [11],
trimming [12] or auto-zeroing [13]. In addition, chopper-
induced IMD tones will now be in-band and so cannot be
easily filtered out.
In prior art [14]–[19], spread-spectrum chopper clocks were
used to lower the magnitude of IMD tones by dithering FDOC .
This turns IMD tones, as well as chopper ripple, into noise-
like signals, thus suppressing IMD tones (by ∼10–20 dB) but
increasing the noise floor around FDOC . However, since this
approach does not solve the underlying problem, the IMD is
still visible in the time domain [15].
In this article, a novel fill-in technique is proposed to address
the underlying cause of chopper-induced IMD [20]. When Fig. 2. Chopped OTA with a pure amplifier delay (Tdelay ) and the
corresponding time-domain signals. The large pulses in the output current
used in a prototype chopper-stabilized amplifier, it decreases (Iout ) result in chopper-induced IMD.
IMD tones by 28 dB and results in a two-tone IMD of
−107 dB for input frequencies near 4FCH (=80 kHz), which
is the same as that obtained without chopping. will no longer be aligned with the corresponding transitions
The rest of the article is organized as follows. The origin of in I1 . The demodulated current (Iout ) produced by the output
chopper-induced IMD is discussed in Section II. The operation chopper (CHout ) will then contain large pulses around the
of the proposed fill-in technique and the resulting amplifier chopping transitions. The amplitude of these pulses (2G m V̂in )
architecture are discussed in Section III. Measurement results is equal to the instantaneous amplitude of the input signal and
are shown and discussed in Section IV, and the article ends so can be a significant source of distortion. These pulses are
with the conclusions. the main cause of chopper-induced IMD.
In the frequency domain, the effect of these pulses can be
better understood by considering the simplified model of the
II. O RIGIN OF C HOPPER -I NDUCED IMD
chopped OTA shown in Fig. 3. This consists of a delay-free
The cause of chopper-induced IMD can be understood by OTA with an extra input path, shown in red, that models the
examining the signals in a chopped operational transconduc- IMD pulses. In this path, Vin is multiplied by a sequence
tance amplifier (OTA) in the presence of a sine-wave input of rectangular pulses ( p) with an amplitude of 2, a width
signal Vin (Fig. 2). Due to the action of the input chopper of Tdelay , and a period of 0.5/FCH . The result (pVin ) is then
(CHin ), the OTA’s input V1 will be an up-modulated version subtracted from Vin to model the pulses generated at the
of Vin . Without loss of generality, the worst case scenario is chopping transitions.
shown, in which the chopping transition coincides with the In the frequency domain, it can be seen that the
peak of the input signal (V̂in ). At this point, CHin changes the multiplication in the extra path will fold input signals near
polarity of the input signal, causing a 2V̂in step in V1 . This is multiples of 2FCH back to dc. The spectra of the various
converted into a current (I1 ) by the OTA’s transconductance signals in Fig. 3 are shown in Fig. 4, where the input is
G m , and delayed by its finite BW (here modeled as a pure assumed to have a narrow-band spectrum (Vin ( f )) centered
delay Tdelay ). As a result, the edges of the chopping clock (CH) on 4FCH . At low frequencies, the spectrum of the rectangular

Authorized licensed use limited to: Hanyang University. Downloaded on August 23,2023 at 07:00:53 UTC from IEEE Xplore. Restrictions apply.
ROOIJERS et al.: FILL-IN TECHNIQUE FOR ROBUST IMD SUPPRESSION IN CHOPPER AMPLIFIERS 3585

Fig. 5. Chopper-induced IMD tones versus the OTA delay for a fixed FCH
Fig. 3. Simplified model for a chopped OTA with a pure delay and the of 20 kHz.
resulting waveform.

Fig. 6. Simplified model for a chopped OTA with limited BW and the
resulting waveform.

For a fixed FCH of 20 kHz, the IMD predicted by (1) is

plotted in Fig. 5 for various values of Tdelay . It can be seen
that a delay of only 1 ns results in a chopper-induced IMD
of 82 dB. Decreasing the delay causes the amplitude of the
IMD tones to roll-off at the rate of 20 dB/decade.
Rather than the rectangular pulses considered so far,
finite OTA BW will cause exponentially settling pulses
at the output of CHout . This can be incorporated in the
Fig. 4. Amplitude spectra of the different signals in the simplified model of chopped OTA model (Fig. 3) by modifying the shape of
a chopped OTA for a narrow-band triangular input spectrum. the pulse sequence p(t), as shown in Fig. 6. Noting that
Tdelay = 1/(2πBW), (1) can be rewritten in terms of amplifier
BW
pulses ( p( f )) consists of impulses at multiples of 2FCH with  
4FCH
an amplitude of 4Tdelay FCH (see Appendix). As a result, the IMD = 20log . (2)
convolution Vin ∗ p( f ) consists of scaled and frequency-shifted 2πBW
versions of the input spectrum, which are subtracted from Vin This equation predicts that the IMD can be reduced by
and applied to the ideal OTA. The amplitude of the IMD tones increasing the OTA’s BW. However, this comes at the expense
in the resulting output current Iout ( f ) are then given by of significantly higher OTA power dissipation. In this work,
  an OTA with a noise density of 11 nV/Hz and a 15 MHz
IMD = 20log 4Tdelay FCH . (1) BW is employed, which corresponds to a delay of ∼10 ns.

Authorized licensed use limited to: Hanyang University. Downloaded on August 23,2023 at 07:00:53 UTC from IEEE Xplore. Restrictions apply.
3586 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 56, NO. 12, DECEMBER 2021

Fig. 7. Simplified block diagram of the chopper-stabilized amplifier.

To achieve the 28 dB reduction in chopper- induced IMD

reported in this work, a 25× increase in BW, and thus power Fig. 8. Schematic of G m1 and G m2 .
would be required.

III. P ROPOSED A MPLIFIER T OPOLOGY W ITH

F ILL -I N T ECHNIQUE
A simplified block diagram of the proposed chopper-
stabilized amplifier is shown in Fig. 7. It consists of a
two-stage main amplifier with a folded-cascode input stage
(G mMAIN ) and a Class AB output stage (G mout ) with Miller
compensation capacitors (Cm1 and Cm2 , 10 pF each), and
Miller-zero compensation resistor (not shown, 5.4 k). The
offset and 1/ f noise of the main amplifier (VosMAIN ) are
Fig. 9. Zoom-in of the pulse in Iout versus time for: ideal signal (no pulse),
suppressed by a three-stage auxiliary amplifier. To minimize dead-band chopping, delayed output chopping, and normal chopping.
its own offset (Vos1 ) and 1/ f noise, the auxiliary amplifier
employs a chopped OTA (G m1 , folded-cascode Fig. 8), fol-
lowed by an integrator (G mINT , folded-cascode, and Cint1−int2 ,
each 36 pF), and a correction OTA (G mCOR , telescopic).
When used in a negative feedback configuration, the offset
of the main amplifier (VosMAIN ) appears at the input of the
chopped G m1 , whose output current is integrated (G mINT )
to generate, via G mCOR , an offset-correcting signal for the
main amplifier (G mMAIN ). Due to the amplifier’s finite gain,
the application of an ac input signal will give rise to a finite
swing at the input of the chopped auxiliary OTA, whose finite
BW then causes chopper-induced IMD in its output current
(Iout ). The resulting tones will be somewhat suppressed by the
active integrator. However, input frequencies close to 2FCH (or
4FCH , 6FCH , etc.,) will cause IMD tones near dc, and so will Fig. 10. Fill-in technique for the chopped auxiliary OTA.
not be suppressed.
To reduce the effect of near-dc IMD tones, the magnitude of output chopper to compensate for the OTA delay. However,
the pulses in Iout should be reduced. One possibility would be simulations show that the IMD reduction is then limited to
to introduce a dead-band, as is sometimes done to mitigate the about 4 dB for the optimal delay, which again ends at the
effect of chopping glitches [21]. In other words, make Iout = 0 zero-crossing of Iout (Fig. 9). The limited effectiveness of these
for the duration of the pulses (Fig. 9). However, this will only techniques can be attributed to their inability to compensate
reduce the amplitude of the pulses by about half: to G m Vin for the long settling tail of Iout after the zero-crossing.
from 2G m Vin . In the case of exponential settling, simulations The goal of the proposed fill-in technique is to suppress
show that the IMD reduction is limited to about 2 dB for chopper-induced IMD by eliminating the pulses in Iout .
the optimal dead-band, which ends at the zero-crossing of Iout As shown in Fig. 10, this can be done by using two iden-
(Fig. 9). Another solution would be to delay the clock of the tical auxiliary OTAs (G m1 and G m2 ), which are chopped

Authorized licensed use limited to: Hanyang University. Downloaded on August 23,2023 at 07:00:53 UTC from IEEE Xplore. Restrictions apply.
ROOIJERS et al.: FILL-IN TECHNIQUE FOR ROBUST IMD SUPPRESSION IN CHOPPER AMPLIFIERS 3587

Fig. 11. Fill-in technique combined with auto-zeroing (left), timing diagram (middle), and the corresponding signals (right).

Fig. 12. Simulated amplitude spectrum without G m -mismatch, 1% of mismatch, and 5%.

Fig. 14. Power breakdown of the most important blocks.

Fig. 13. Die micrograph.

OTA are filled in by the output current of the other. The

in quadrature (CH1 and CH2 ). Although the output current key insight is that the required switches can select the OTA’s
of each channel (Iout1 and Iout2 ) will still contain pulses output currents much faster than the OTAs themselves can
around the chopping transitions, these will only occur in settle.
one channel at a time. The OTAs’ output currents (Iout1,2 ) However, the chopped offset of G m1 and G m2 (Vos1 and
are nominally identical, and so, by selecting the appropriate Vos2 ) will still cause ripple. To suppress this, the proposed fill-
current slightly before each chopping transition, a combined in technique is combined with auto-zeroing (Fig. 11). An AZ
output current can be generated that is free of pulses and, loop is added, which is used to AZ the OTA when it is not
therefore, free of IMD. In other words, the pulses of one connected to G mINT . To prevent input common-mode (CM)

Authorized licensed use limited to: Hanyang University. Downloaded on August 23,2023 at 07:00:53 UTC from IEEE Xplore. Restrictions apply.
3588 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 56, NO. 12, DECEMBER 2021

Fig. 15. Measured amplitude spectrum (ten averages) for a one-tone test of 79 kHz 1 Vrms for fill-in disabled and enabled.

transients, which would cause additional IMD, the OTA inputs 1% mismatch, which is readily achievable. Even with 5%
are shorted to one of the input pins during the AZ phase mismatch, −120 dB IMD can still be achieved, showing that
(via S1 ). Dummy always-closed and always-open switches in practice, G m mismatch is not an important limiting factor.
(in gray) ensure that the input network formed by the switch
resistances and the parasitic capacitance is symmetric. During
IV. M EASUREMENT R ESULTS
the AZ phase, CAZ (25 pF) acts as a passive integrator whose
output drives G mAZ1 (Telescopic) to cancel the OTA’s offset. The prototype chopper-stabilized amplifier with fill-in tech-
The resulting voltage is held by C1,2 (1.8 pF each) during the nique was realized in a 0.18 μm CMOS BCD process. The
amplification phase. To minimize noise folding, the noise BW die micrograph is shown in Fig. 13, with the most important
during the AZ phase should be limited by minimizing G mAZ1 , blocks highlighted. It occupies an active area of 0.54 mm2 and
but this increases the OTA’s worst case output swing. As a draws 550 μA from a 5 V supply. The opamp has a 0–4.5 V
compromise, G mAZ1 is chosen to be ∼50× smaller than G m1 . input CM range, a 15.4 noise efficiency factor (NEF), and a
At the start of the AZ phase, the output current of the OTA 4.2 MHz gain bandwidth product (GBW). A power breakdown
needs to transition from a level that depends on the input is shown in Fig. 14. It is equally split between the main
signal to a level that is nearly zero. To prevent these transients amplifier, Channel 1, Channel 2, and the rest of the stabi-
from reaching the AZ loop and thus causing additional IMD, lization loop. The first three all have low noise input stages
shorting switch S4 allows the OTA current to settle before and so dissipate significant power. Furthermore, at high input
it is connected via switches S6&7 to the integration capaci- frequencies, the signal entering the stabilization loop increases
tor CAZ [22]. S6&7 are controlled by the AZ1S signal, which due to the roll-off of amplifier gain. To handle the resulting
includes a dead-band that disconnects CAZ during this settling high current levels output by the input OTAs, the stabilization
time. The width of the dead-band is set to 100 ns, which loop also dissipates significant power.
guarantees OTA settling in the worst case corner. Similarly, With the prototype amplifier configured as a buffer,
the AZ phase is ended roughly 100 ns before the next chopping a one-tone test was performed using an Audio Precision
phase, allowing G m1,2 to settle before it is connected to G mINT . APx555 analyzer, which provides a 1 Vrms 79 kHz input
To further minimize the voltage transient that occurs when the tone at an input CM level of 2.2 V. To ensure that the
OTA is connected to G mINT , the shorting switch (S4 ) resistance IMD tones are well above the ∼134 dB noise floor, an input
is set to ∼1/G mINT (6.8 k). The remaining voltage transients tone near 4FCH (79 kHz) was used rather than one near
are then mainly due to the charge-injection of the multiplexing 2FCH (39 kHz). The resulting output amplitude spectrum is
switches (S3 and S5 ). shown in Fig. 15. The 79 kHz input tone, together with the
The effectiveness of the fill-in technique is limited by the chopping frequency of 20 kHz, leads to an IMD tone at 1 kHz
G m matching of the two input OTAs, as any mismatch will (4FCH –Fin ). Without the fill-in technique, a large (−97.7 dB)
cause additional transients in their composite output current. IMD tone is present. With the fill-in technique enabled, this
With a 1 Vrms 81 kHz input signal, the results of simulations drops by 28 to −125.9 dB. Some power-line interference is
with various degrees of intentional G m mismatch are shown also visible below 1 kHz. This is present even in the absence
in Fig. 12. Without mismatch, the simulated chopper-induced of the prototype chip and is thus caused by the measurement
IMD is −133 dB, increasing only slightly to −131 dB for setup. Higher-order IMD tones (>1 kHz) are also partly

Authorized licensed use limited to: Hanyang University. Downloaded on August 23,2023 at 07:00:53 UTC from IEEE Xplore. Restrictions apply.
ROOIJERS et al.: FILL-IN TECHNIQUE FOR ROBUST IMD SUPPRESSION IN CHOPPER AMPLIFIERS 3589

Fig. 16. Measured amplitude spectrum (ten averages) for a two-tone test of 79 and 80 kHz tone 0.5 Vrms each, for the case of the un-chopped amplifier,
chopped without fill-in, and with fill-in.

Fig. 18. Input current for different chopping frequencies.

Fig. 17. Histogram of the offset voltage and input current for 15 samples.
The opamp’s voltage noise density is shown in Fig.
√ 19. With
chopping disabled, its white-noise level is 16 nV Hz and its
suppressed by the fill-in technique. Some residual ripple is 1/ f corner frequency is ∼6 kHz. With chopping, auto-zeroing
also present at multiples of FCH . and fill-in enabled, the 1/ f corner frequency is less than a
A two-tone test was also performed using tones at 79 and few Hertz. The noise folding associated with auto-zeroing
80 kHz, both with an amplitude of 0.5 Vrms . The resulting each OTA at 20 kHz causes a slight noise bump around this
amplitude spectrum is shown in Fig. 16. As a baseline, the two- frequency. Some tones can also be seen at the chopping/auto-
tone IMD was first measured with chopping disabled. Due to zeroing frequencies, which is due to PCB-mediated crosstalk
the amplifier’s own non-linearity, an IMD tone of −107 dB between the 5 V reference clock (80 kHz) and the prototype
can be seen at 1 kHz. When chopping is enabled, but without chip.
the fill-in technique, this tone increases to −97 dB, which is To verify the amplifier’s ability to handle rapidly changing
mainly due to chopper-induced IMD. With the fill-in technique signals, its slew rate was measured by applying a 4 V input
enabled, the IMD drops back to −107 dB, demonstrating the step (Fig. 20). For both rising and falling steps, the speed at
effective suppression of chopper-induced IMD. which the amplifier’s output transitions from 10% to 90% of
With a 2.5 V input CM voltage and FCH = 20 kHz, its final value corresponds to a slew rate of 1.7 V/μs.
measurements on 15 samples show that the offset is less Fig. 21 shows the measured power supply rejection ratio
than 0.8 μV, while the input current is less than 600 pA (PSRR) of the amplifier using a 1 Vrms disturbance added to
(Fig. 17). The input current was measured by a Keithley the 5 V supply at different frequencies. At low frequencies,
6514 electrometer, and guarding was used to minimize printed the PSRR is 124 dB, rolling off at higher frequencies.
circuit board (PCB) leakage. The input current is a linear Table I summarizes the performance of the amplifier and
function of FCH (Fig. 18), indicating that it is mainly due to the compares it to other DOC amplifiers using chopping, auto-
charge-injection (mismatch) of the chopper and AZ switches. zeroing, or a combination of the two. Table I reports the

Authorized licensed use limited to: Hanyang University. Downloaded on August 23,2023 at 07:00:53 UTC from IEEE Xplore. Restrictions apply.
3590 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 56, NO. 12, DECEMBER 2021

TABLE I
P ERFORMANCE S UMMARY AND C OMPARISON W ITH THE S TATE - OF - THE -A RT

Fig. 20. Slew-rate measurement for a step-up and step-down.

Fig. 19. Voltage noise density versus frequency for the case without any
dynamic techniques and with CH, AZ, and fill-in.

IMD of amplifiers that use both fixed DOC frequencies

(Single) [8], [17] and spread-spectrum DOC frequencies
(Spread) [18], [19]. For the latter, the IMD with a fixed DOC
frequency is also reported (Single), where it should be noted
that [17] and [18] describe the same amplifier. Although the
reported IMD of the proposed amplifier is obtained from a Fig. 21. Measured PSRR versus frequency.
single sample, the variation between five samples has been
measured. Without the fill-in technique, the single-tone IMD
varies between −97.2 and −98.2 dB. With the fill-in technique higher input frequency (79 kHz), which is more difficult to
enabled, the IMD varies between −125 and −131 dB. The achieve due to the roll-off of amplifier gain with frequency.
amplifier achieves the lowest IMD (125.9 dB) at a much Furthermore, at worst case input frequencies ( f in = n fAZ with

Authorized licensed use limited to: Hanyang University. Downloaded on August 23,2023 at 07:00:53 UTC from IEEE Xplore. Restrictions apply.
ROOIJERS et al.: FILL-IN TECHNIQUE FOR ROBUST IMD SUPPRESSION IN CHOPPER AMPLIFIERS 3591

n = 1, 3, 5, . . . or f in = n fCH with n = 2, 4, 6, . . .), which will negligible for a small Tdelay . The multiplication pVin leads to
cause near-dc IMD tones, an 81.9 dB improvement is obtained. the convolution
For low input frequencies (<1 kHz), the IMD tones are below
the −134 dB noise floor. Even though an additional low- p ∗ Vin ( f )
∞

noise input stage is required to implement the fill-in technique,
= 4Tdelay FCH sinc(Tdelay f ) δ( f − n2F CH ) ∗ Vin (f)
the amplifier’s total supply current (0.55 mA) is comparable
n=−∞
with that of other designs. Each fill-in channel uses 24% of ∞

the power and 10% of the total active area. = 4Tdelay FCH sinc(Tdelay f ) Vin ( f − n2F CH ). (7)
n=−∞

V. C ONCLUSION The frequency shifted versions of the input spectrum lead

In chopper amplifiers, finite amplifier delay gives rise to to the IMD tones. pVin is subtracted from Vin and multiplied
short pulses around the chopping transitions, which in turn by G m , leading to an output current spectrum (Iout ( f )) of
give rise to significant amounts of chopper-induced IMD. Iout ( f ) = G m Vin − 4Tdelay FCH sinc(Tdelay f )
To address this problem, a novel fill-in technique is proposed, ∞

in which two identical amplifiers are chopped in quadrature × Gm Vin ( f − n2F CH ). (8)
such that a pulse-free output signal can be obtained by n=−∞
selecting the output of the appropriate amplifier. As a proof of
concept, the fill-in technique was used in a chopper-stabilized R EFERENCES
amplifier, resulting in a 28 dB reduction of its chopper-
induced IMD. This corresponds to less than −125 dB of [1] C. C. Enz and G. C. Temes, “Circuit techniques for reducing the effects
of op-amp imperfections: Autozeroing, correlated double sampling, and
chopper-induced IMD for a single-tone near 4FCH (=80 kHz), chopper stabilization,” Proc. IEEE, vol. 84, no. 11, pp. 1584–1614,
while the two-tone IMD (with 79 and 80 kHz tones) is less Nov. 1996.
than −105 dB, mainly limited by the amplifier’s own linearity. [2] Q. Fan, J. H. Huijsing, and K. A. A. Makinwa, “Input characteristics
of a chopped multi-path current feedback instrumentation amplifier,” in
√ amplifier achieves low offset (<1 μV) and low
In addition, the Proc. 4th IEEE Int. Workshop Adv. Sens. Interfaces (IWASI), Jun. 2011,
noise (16 nV Hz), while its supply current (0.55 mA from a pp. 61–66.
5 V supply) is comparable with that of other state-of-the-art [3] C. Menolfi and Q. Huang, “A fully integrated, untrimmed CMOS
precision amplifiers. instrumentation amplifier with submicrovolt offset,” IEEE J. Solid-State
Circuits, vol. 34, no. 3, pp. 415–420, Mar. 1999.
[4] A. Bakker, K. Thiele, and J. H. Huijsing, “A CMOS nested-chopper
instrumentation amplifier with 100-nV offset,” IEEE J. Solid-State
A PPENDIX Circuits, vol. 35, no. 12, pp. 1877–1883, Dec. 2000.
[5] R. Burt and J. Zhang, “A micropower chopper-stabilized operational
In this appendix, the amplitude spectra for the different amplifier using a SC notch filter with synchronous integration inside
signals in Fig. 3 are determined. the continuous-time signal path,” IEEE J. Solid-State Circuits, vol. 41,
no. 12, pp. 2729–2736, Dec. 2006.
The sequence of rectangular pulses ( p) can be described by
[6] T. Rooijers, J. H. Huijsing, and K. A. A. Makinwa, “A quiet digitally
the convolution assisted auto-zero-stabilized voltage buffer with 0.6 pA input current
∞   and 0.6 μV offset,” in IEEE Int. Solid-State Circuits Conf. (ISSCC)
n Dig. Tech. Papers, Feb. 2018, pp. 50–52.
p(t)= 2rect(Tdelay t) ∗ δ t− (3) [7] T. Rooijers, J. H. Huijsing, and K. A. A. Makinwa, “An auto-zero
n=−∞
2F CH stabilized voltage buffer with a quiet chopping scheme and constant
input current,” in IEEE Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech.
where the rectangular function represents the short pulse Papers, Feb. 2019, pp. 298–299.
caused by amplifier delay and the impulse train represents its [8] T. Rooijers, J. H. Huijsing, and K. A. A. Makinwa, “An auto-zero
stabilized voltage buffer with a trimmed input current of 0.2 pA,” in
repetition rate (every 0.5/FCH ). The Fourier transform of the Proc. IEEE 45th Eur. Solid State Circuits Conf. (ESSCIRC), Sep. 2019,
first term is pp. 257–260.
[9] Texas Instruments. (Oct. 2009). OPA378 Data Sheet. [Online]. Available:
2Tdelay sinc(Tdelay f ) (4) https://www.ti.com/lit/ds/symlink/opa378.pdf
[10] Y. Kusuda, “A 60 V auto-zero and chopper operational ampli-
fier with 800 kHz interleaved clocks and input bias current trim-
and the Fourier transform of the second term is ming,” IEEE J. Solid-State Circuits, vol. 50, no. 12, pp. 2804–2813,
∞ Dec. 2015. √
 [11] Q. Fan, J. H. Huijsing, and K. A. A. Makinwa, “A 21 nV/ Hz Chopper-
2F CH δ( f − n2F CH ). (5) stabilized multi-path current-feedback instrumentation amplifier with 2
n=−∞ μV offset,” IEEE J. Solid-State Circuits, vol. 47, no. 2, pp. 464–475,
Feb. 2012. √
The convolution in the time-domain leads to a multiplication [12] I. Akita and M. Ishida, “A 0.06 mm2 14nV/ Hz Chopper instrumen-
in the frequency-domain, which results in tation amplifier with automatic differential-pair matching,” in IEEE
Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers, Feb. 2013,
∞
 pp. 178–179. √
p( f ) = 4Tdelay FCH sinc(Tdelay f ) δ( f − n2F CH ). (6) [13] A. T. K. Tang, “A 3 μV-offset operational amplifier with 20 nV/ Hz
input noise PSD at DC employing both chopping and autozeroing,”
n=−∞ in IEEE Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers,
Feb. 2002, pp. 386–387.
The notches of sinc(Tdelay f ) occur at n/Tdelay with n = 1, [14] E. C. J. Brown, “Chopper-stabilized amplifier with spread-spectrum
2, 3, . . .. At low frequencies, the effect of the sinc function is clocking,” U.S. Patent 5 115 202 A, May 19, 1992.

Authorized licensed use limited to: Hanyang University. Downloaded on August 23,2023 at 07:00:53 UTC from IEEE Xplore. Restrictions apply.
3592 IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 56, NO. 12, DECEMBER 2021

[15] A. T. K. Tang, “Bandpass spread spectrum clocking for reduced clock Yoshinori Kusuda (Member, IEEE) received the
spurs in autozeroed amplifiers,” in Proc. IEEE Int. Symp. Circuits Syst. M.S. degree in physical electronics from Tokyo
(ISCAS), vol. 1, May 2001, pp. 663–666. Institute of Technology, Tokyo, Japan, in 2004, and
[16] Maxim Integrated. (Feb. 2018). MAX4238/MAX4239 Data Sheet. the Ph.D. degree with a focus on the subject of
[Online]. Available: https://datasheets.maximintegrated.com/en/ds/MAX reducing switching artifacts in chopper amplifiers
4238-MAX4239.pdf from TU Delft, Delft, The Netherlands, in 2018.
[17] Analog Devices. (Jun. 2015). AD8551 Data Sheet. [Online]. Available: In 2004, he joined Analog Devices Japan Design
http://www.analog.com/media/en/technical-documentation/datasheets/ Center, Tokyo. From 2015 to 2018, he was a
AD8551_8552_8554.pdf Guest with the Electronic Instrumentation Labora-
[18] Analog Devices Inc. (1999). AD8571 Data Sheet. [Online]. Available: tory, TU Delft. Since then, he has been with Analog
http://www.analog.com/media/en/technical-documentation/data- Devices and is currently in San Jose, CA, USA,
sheets/AD8571_8572_8574.pdf working with the Linear and Precision Technology Group. He has been
[19] V. Ivanov and √M. Shaik, “A 10 MHz-bandwidth 4 μs-large-signal- working on precision CMOS analog designs, including stand-alone amplifiers
settling 6.5nV/ Hz-noise 2 μV-offset chopper operational amplifier,” and application specific mixed-signal products. This work has resulted in
in IEEE Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers, presentations and papers at IEEE conferences and journals, as well as
Feb. 2016, pp. 88–89. 14 issued U.S. patents.
[20] T. Rooijers, S. Karmakar, Y. Kusuda, J. H. Huijsing, and
K. A. A. Makinwa, “A Chopper-stabilized amplifier with -107 dB IMD Johan H. Huijsing (Life Fellow, IEEE) received the
and 28 dB suppression of Chopper-induced IMD,” in IEEE Int. Solid- M.Sc. degree in electrical engineering and the Ph.D.
State Circuits Conf. (ISSCC) Dig. Tech. Papers, Feb. √
2021, pp. 438–440. degree from Delft University of Technology, Delft,
[21] Q. Huang and C. Menolfi, “A 200 nV offset 6.5 nV/ Hz noise PSD 5.6 The Netherlands, in 1969 and 1981, respectively.
kHz Chopper instrumentation amplifier in 1 μm digital CMOS,” in IEEE He has been an Assistant Professor and an Asso-
Int. Solid-State Circuits Conf. (ISSCC) Dig. Tech. Papers, Feb. 2001 ciate Professor of electronic instrumentation with the
pp. 362–363. Faculty of Electrical Engineering, Delft University
[22] M. A. P. Pertijs and W. J. Kindt, “A 140 dB-CMRR current-feedback of Technology, since 1969. In 1990, he joined the
instrumentation amplifier employing ping-pong auto-zeroing and chop- Chair of Electronic Instrumentation, as a Full Pro-
ping,” IEEE J. Solid-State Circuits, vol. 45, no. 10, pp. 2044–2056, fessor. He has been a Professor Emeritus since 2003.
Oct. 2010. From 1982 to 1983, he was a Senior Scientist with
Philips Research Laboratories, Sunnyvale, CA, USA. From 1983 to 2005, he
was a Consultant with Philips Semiconductors, Sunnyvale. Since 1998, he
has been a Consultant with Maxim, Sunnyvale. His research work is focused
on operational amplifiers, analog-to-digital converters, and integrated smart
sensors. He has supervised 30 Ph.D. students. He has authored or coauthored
over 300 scientific articles, 40 U.S. patents, and 15 books. In 1992, he initiated
the International Workshop on Advances in Analog Circuit Design. He co-
organized it yearly until 2003.
Thije Rooijers (Member, IEEE) was born in Leiden, Dr. Huijsing was a Program Committee Member of the European Solid-State
The Netherlands, in 1991. He received the B.Sc. and Circuits Conference from 1992 to 2002. He was awarded the title of Simon
M.Sc. degrees in electrical engineering from Delft Stevin Meester by the Dutch Technology Foundation. He was the Chairman
University of Technology, Delft, The Netherlands, of the Dutch STW Platform on Sensor Technology and the Biannual National
in 2013 and 2016, respectively, where he is currently Workshop on Sensor Technology from 1991 to 2002.
pursuing the Ph.D. degree with a focus on reducing
the imperfections of dynamic offset compensated Kofi A. A. Makinwa (Fellow, IEEE) received the
amplifiers. B.Sc. and M.Sc. degrees from Obafemi Awolowo
Mr. Rooijers was a recipient of the ADI Outstand-
University, Ife, Nigeria, in 1985 and 1988, respec-
ing Student Designer Award in 2018, the ProRISC tively, the M.E.E. degree from Philips International
Best Oral Presentation Award in 2019, and the Institute, Eindhoven, The Netherlands, in 1989, and
ProRISC Best Poster Award in 2021. the Ph.D. degree from Delft University of Technol-
ogy, Delft, The Netherlands, in 2004.
From 1989 to 1999, he was a Research Scien-
tist with Philips Research Laboratories, Eindhoven,
where he worked on interactive displays and digital
recording systems. In 1999, he joined Delft Uni-
versity of Technology, where he is currently an Antoni van Leeuwenhoek
Professor and the Head of the Microelectronics Department. His research
Shoubhik Karmakar (Student Member, IEEE) interests include the design of mixed-signal circuits, sensor interfaces, and
received the B.E. degree in electrical and electron- smart sensors. This has led to 16 books, over 300 technical articles, and over
ics engineering from Birla Institute of Technology 30 patents.
and Science, Pilani, India, in 2012, and the M.Sc. Dr. Makinwa has been on the program committees of several IEEE con-
degree from Delft University of Technology, Delft, ferences, and was the Analog Subcom Chair of ISSCC. He has also served
The Netherlands, in 2017, where he is currently the Solid-State Circuits Society as a Distinguished Lecturer and as an Elected
pursuing the Ph.D. degree. Member of its Adcom. He is currently one of the organizers of the Advances
His current research interests include energy- in Analog Circuit Design Workshop and the Sensor Interfaces Meeting. He
efficient data converters and high-performance is an ISSCC top-10 contributor, and a co-recipient of 16 best paper awards,
class-D amplifiers for audio applications. from the JSSC, ISSCC, VLSI, ESSCIRC, and Transducers, among others. He
is a member of the Royal Netherlands Academy of Arts and Sciences.

Authorized licensed use limited to: Hanyang University. Downloaded on August 23,2023 at 07:00:53 UTC from IEEE Xplore. Restrictions apply.