sensors

Article
Development of Magnetocardiograph without Magnetically
Shielded Room Using High-Detectivity TMR Sensors
Koshi Kurashima 1, *, Makoto Kataoka 1 , Takafumi Nakano 2 , Kosuke Fujiwara 3 , Seiichi Kato 1 ,
Takenobu Nakamura 1 , Masaki Yuzawa 1 , Masanori Masuda 1 , Kakeru Ichimura 1 , Shigeki Okatake 1 ,
Yoshitaka Moriyasu 1 , Kazuhiro Sugiyama 2 , Mikihiko Oogane 2 , Yasuo Ando 2 , Seiji Kumagai 2,3 ,
Hitoshi Matsuzaki 2,3 and Hidenori Mochizuki 1

1 Device & Process Application Development Unit, Research & Development Center, Asahi Kasei Microdevices
Corporation, Atsugi AXT Maintower 20F, 3050 Okata, Atsugi 243-0021, Kanagawa, Japan
2 Department of Applied Physics, Graduate School of Engineering, Tohoku University, 6-6-05 Aoba-yama,
Aoba-ku, Sendai 980-8579, Miyagi, Japan
3 Spin Sensing Factory Corporation, Research Center for Rare Metal and Green Innovation,
403 468-1 Aramaki Aza-Aoba, Aoba-ku, Sendai 980-0845, Miyagi, Japan
* Correspondence: kurashima.kc@om.asahi-kasei.co.jp

Abstract: A magnetocardiograph that enables the clear observation of heart magnetic field mappings
without magnetically shielded rooms at room temperatures has been successfully manufactured.
Compared to widespread electrocardiographs, magnetocardiographs commonly have a higher spatial
resolution, which is expected to lead to early diagnoses of ischemic heart disease and high diagnostic
accuracy of ventricular arrhythmia, which involves the risk of sudden death. However, as the con-
ventional superconducting quantum interference device (SQUID) magnetocardiographs require large
magnetically shielded rooms and huge running costs to cool the SQUID sensors, magnetocardiogra-
phy is still unfamiliar technology. Here, in order to achieve the heart field detectivity of 1.0 pT without
magnetically shielded rooms and enough magnetocardiography accuracy, we aimed to improve the
detectivity of tunneling magnetoresistance (TMR) sensors and to decrease the environmental and
Citation: Kurashima, K.; Kataoka, M.;
sensor noises with a mathematical algorithm. The magnetic detectivity of the TMR sensors was con-
Nakano, T.; Fujiwara, K.; Kato, S.;
firmed to be 14.1 pTrms on average in the frequency band between 0.2 and 100 Hz in uncooled states,
Nakamura, T.; Yuzawa, M.; Masuda,
thanks to the original multilayer structure and the innovative pattern of free layers. By constructing a
M.; Ichimura, K.; Okatake, S.; et al.
sensor array using 288 TMR sensors and applying the mathematical magnetic shield technology of
Development of
Magnetocardiograph without
signal space separation (SSS), we confirmed that SSS reduces the environmental magnetic noise by
Magnetically Shielded Room Using −73 dB, which overtakes the general triple magnetically shielded rooms. Moreover, applying digital
High-Detectivity TMR Sensors. processing that combined the signal average of heart magnetic fields for one minute and the projection
Sensors 2023, 23, 646. https:// operation, we succeeded in reducing the sensor noise by about −23 dB. The heart magnetic field
doi.org/10.3390/s23020646 resolution measured on a subject in a laboratory in an office building was 0.99 pTrms and obtained
magnetocardiograms and current arrow maps as clear as the SQUID magnetocardiograph does in the
Academic Editor: Fow-Sen Choa
QRS and ST segments. Upon utilizing its superior spatial resolution, this magnetocardiograph has
Received: 7 December 2022 the potential to be an important tool for the early diagnosis of ischemic heart disease and the risk
Revised: 23 December 2022 management of sudden death triggered by ventricular arrhythmia.
Accepted: 29 December 2022
Published: 6 January 2023
Keywords: magnetocardiography; TMR sensors; signal space separation; projection operation;
environmental noise

Copyright: © 2023 by the authors.

Licensee MDPI, Basel, Switzerland.
1. Introduction
This article is an open access article
distributed under the terms and Magnetocardiography, which detects heart magnetic fields outside the body, has a
conditions of the Creative Commons higher spatial resolution than electrocardiography, which measures the electric potential
Attribution (CC BY) license (https:// on the body surface produced by heart activities [1]. While electrocardiography needs
creativecommons.org/licenses/by/ an exercise stress on the subject for detecting ischemic heart disease, especially stable
4.0/). angina, it has been reported that magnetocardiography requires no stresses on the subject

Sensors 2023, 23, 646. https://doi.org/10.3390/s23020646 https://www.mdpi.com/journal/sensors

Sensors 2023, 23, 646 2 of 18

to diagnose them [2,3]. Magnetocardiography shows clear changes before and after percuta-
neous coronary intervention and coronary revascularization, whereas electrocardiography
shows no significant differences [4,5]. The diagnostic accuracy of magnetocardiography
on ischemic heart disease is reported to be as high as myocardial scintigraphy [6,7]. More-
over, relationships between the abnormality in the spatial mappings of heart magnetic
fields obtained with magnetocardiography and Brugada syndrome [8], left intraventricular
disorganized conduction in patients with dilated cardiomyopathy [9], right ventricular
cardiomyopathy [10] and malignant arrhythmias [11] have been reported, attracting at-
tention to magnetocardiography for the early detection of arrhythmias and ventricular
fibrillation. These reports suggest that the magnetocardiography may play an important
role in predicting ischemic heart disease, which is the highest cause of death in the world,
as well as the ventricular arrhythmia, which triggers sudden death.
The only practicalized magnetocardiographs with superconducting quantum inter-
ference devices (SQUIDs) have not been widely used yet due to the huge running costs
of liquid helium to cool the sensors down to ultra-low operating temperatures (−269 ◦ C).
As these magnetocardiographs need Dewar vessels with thick vacuum walls containing
liquid helium, sensors are kept away from the body surface, which causes a limitation of
the spatial detectivity [12]. Therefore, magnetocardiographs with small sensors that work
at room temperatures are desired. As magnetic sensors are suitable for detecting minute
magnetic fields such as heart magnetic fields, optically pumped magnetometers [13,14]
and nitrogen-vacancy center magnetometers [15,16] have recently been attracting interest.
However, as the former sensors have to be heated above 100 ◦ C and the latter still remain at
the research level, there are difficulties with magnetocardiography, which is an urgent issue.
On the other hand, tunneling magnetoresistance (TMR) sensors are easily produced with
a common process technique for integrated circuits and have been very widely used for
commercial sensors, such as magnetic heads and current sensors, as they can be driven at
room temperatures. Moreover, the smallness of the sensors and the needlessness of Dewar
vessels essential for SQUIDs realize a high spatial resolution by placing the sensors close to
the subjects’ body surfaces. In fact, the quadrupole has been observed in the magnetocar-
diogram of patients with Brugada syndrome [8], which suggests a higher spatial resolution
brings a better diagnostic accuracy. In addition, it has been reported that brain magnetic
fields, which are far smaller than heart magnetic fields, have been measured with scalp
attached magnetoencephalography with TMR sensors [17]. Thus, TMR sensors, which are
approachable to the signal source, are adequate for magnetocardiographs.
For conventional magnetocardiographs, magnetically shielded rooms are essential
to reduce environmental noise. While heart magnetic fields are up to several tens of pico
teslas at the body surface, the terrestrial magnetic field is as high as 50 µT. There have been
several reports in which environmental noises are reduced using magnetically shielded
rooms when applying magneto-resistance sensors to the magnetocardiography [18,19];
however, the magnetically shielded rooms require special operation and high initial costs.
Therefore, the development of a mathematical algorithm to establish a system without
magnetically shielded rooms is desirable to spread the use of magnetocardiography.
In order to measure the tiny heart magnetic fields, an algorithm to decrease the sensor
noises down to 1 pT, which is lower than the signal, is also favorable. Although it has
recently been reported that heart magnetic fields were observed with TMR sensors under
an unshielded condition by the AC modulation and the impedance compensation [20], the
detectivity was not enough for accurate diagnostics. As TMR sensors have higher sensitivity
and higher 1/f noise in the low frequency domain compared to giant magnetoresistance
(GMR) sensors [21], the suppression of sensor noises may be effective to improve the
accuracy of magneto-cardiometry.
The purpose of this work was to fabricate a magnetocardiograph that allows us
to quickly obtain clear heart magnetic field maps without liquid helium and magneti-
cally shielded rooms. According to previous studies, early diagnoses of ischemic heart
disease [22], follow-up before and after coronary revascularizations [5] and prognosis
Sensors 2023, 23, 646 3 of 18

prediction of dilated cardiomyopathy [9] are expected to be possible when the magnetic
field detectivity reaches 1.0 pTrms . Here, we aimed to obtain the heart magnetic detectivity
of 1.0 pTrms in the frequency domain of 0.2–100 Hz with integration for 1 min. The issues
to be solved in this work were the following:
• Improving the magnetic detectivity of TMR sensors with high sensitivity and low noises;
• The suppression of environmental noises with the algorithm;
• The suppression of sensor noises with the algorithm.
The target values of sensor noises and the suppression of environmental noises were
1.0 pTrms and −60 dB, respectively. The latter was set to suppress the assumed environ-
mental noises of 1 nT to be lower than 1 pT. We report that we successfully fabricated a
magnetocardiograph whose heart magnetic field detectivity was 0.99 pTrms in an office
building under the unshielded condition. Moreover, we obtained magnetocardiograms
and current arrow maps as clear as SQUID magnetocardiographs in the QRS and ST seg-
ments. These results may lead to the common use of high-accuracy magnetocardiographs
in hospitals and clinics in the future.

2. Materials and Methods

The thin films of TMR sensors were deposited on 5-inch Si/SiO2 (1 µm) wafers using
sputtering apparatus (HC7200, Canon Anelva Corp., Kanagawa, Japan). TMR sensors were
microfabricated using the common process technique for general integrated circuits. After
the microfabrication, the wafers were first annealed at 325 ◦ C in the external magnetic field
of 1 T vertical to the sensing axis, followed by the second annealing at 225 ◦ C with the
field parallel to the sensing axis to rotate the easy axis of pinned layer by 90 degrees. By
orthogonalizing the magnetic easy axes of the pinned and free layers, the transfer curve
with the magnetic field along the hard axis of the free layers reflects the M-H curve for
the hard direction of the free layers [23–26]. The first magnetic flux concentrators (MFCs)
of permalloy were deposited to be 2.0 mm × 1.6 mm × 10 µm with the magnetic field of
100 mT vertical to the sensing axis. The chip size was 4.2 mm × 2.2 mm × 625 µm. As
shown in Figure 1a,b, the second MFCs of permalloy were fixed to overlap the first MFCs
with thermosetting resin. The sizes of the second MFCs were 2 mm × 20 mm × 800 µm.
The transfer curve and the detectivity of the sensors were measured with the open-loop
system in a triple magnetically shielded box. When constructing the closed-loop systems
for magnetocardiography measurements, feedback coils were placed to cover each set of a
TMR sensor and MFCs bonded on a printed circuit board, and they were mechanically fixed
on the printed circuit board to avoid the physical interference between feedback coils, TMR
sensors and MFCs. The sensor array consisted of using an acrylic frame with 96 sensor cells
in which three sensors were vertically arranged. The subject of the magnetocardiography
measurements was a male in his twenties with no history of heart disease in the past.
During the measurements, the subject laid on his back to be covered with the sensor array
on his chest on a table in a laboratory on the 18th floor of the 26-storied office building.
After obtaining the data with 48 channels (288 sensors), the digital processing of signal
space separation (SSS) (explained later), the common digital filter (the 50 Hz notch filter
and the 100 Hz low-pass filter) and the signal average for one minute referencing the
electrocardiograph signals took place.
 history of heart disease in the past. During the measurements, the subject laid on his back
to be covered with the sensor array on his chest on a table in a laboratory on the 18th floor
of the 26-storied office building. After obtaining the data with 48 channels (288 sensors),
the digital processing of signal space separation (SSS) (explained later), the common dig-
Sensors 2023, 23, 646 ital filter (the 50 Hz notch filter and the 100 Hz low-pass filter) and the signal average4 for
of 18
one minute referencing the electrocardiograph signals took place.

Figure 1. (a) Diagram of sensors with MFCs bonded on a print circuit board. The Si spacers and the
Figure 1. (a) Diagram of sensors with MFCs bonded on a print circuit board. The Si spacers and the
sensorchip
sensor chipon onthe
theboard
boardand
andthe
thesecond
secondMFCs
MFCson onthe
theSiSispaces
spacesare
arefixed
fixedwith
withthermosetting
thermosettingresin.
resin.
(b) Enlarged diagram of the sensor chip in (a). The wiring between the sensors and electrodes isis
(b) Enlarged diagram of the sensor chip in (a). The wiring between the sensors and electrodes
omittedfor
omitted forsimplicity.
simplicity.(c)
(c)Enlarged
Enlargeddiagram
diagramofofTMRTMRsensors.
sensors.(d)(d)Transfer
Transfercurve
curveofofaaTMR TMRsensor
sensor
withthe
with thefirst
firstand
andsecond
secondMFCs.
MFCs.(e)(e)Detectivity
Detectivityspectrum
spectrumof ofTMR
TMRsensors
sensorswith/without
with/withoutthe thefirst
firstand
and
second
secondMFCs
MFCsmeasured
measuredin inan
anopen-loop
open-loopsystem
systemwithwiththe
theACACmagnetic
magnetic field
field of
of 11 nT
nTp-p
p-pat
at10
10HzHz along
along
the
thesensing
sensingaxis.
axis.(f)
(f)Detectivity
Detectivity histogram
histogram measured with 108
measured with 108 sensors
sensorsof ofSample-A
Sample-Aand and124
124sensors
sensorsof
ofSample-B.
Sample-B.RedRedand
andgray
graydata
datacorrespond
correspondtoto Sample-A and Sample-B, respectively.
Sample-A and Sample-B, respectively.

3.3.Results
Results
3.1. TMRSensors
3.1. TMR Sensors
Thestacking
The stacking structure
structure of ofSample-A
Sample-Awas wasSub.
Sub.Si,Si,SiO 2 /Ta
SiO 5/Ru 20/Ta 5/Co 70.5Fe
2/Ta 5/Ru 20/Ta 5/Co70.5
Fe4.5 Si15BB1010
4.5Si15
100/Ru 10/Co
100/Ru 10/Co70.5 Fe
70.5Fe4.5 Si
Si1515
4.5 B 100/Ru 0.4/Co
B1010100/Ru 0.4/Co40 Fe
40Fe40 B 3/MgO/Co
40B2020 3/MgO/Co4040 Fe
Fe4040 B 3/Ru 0.9/Co
B20203/Ru 0.9/Co7575Fe Fe2525
2/Ir
2/Ir2222 Mn
Mn78 7810/Ta 5 / Ru 10 (in nm). In order to reduce the variation in detectivity, we aimedto
10/Ta 5 / Ru 10 (in nm). In order to reduce the variation in detectivity, we aimed
touniformize
uniformize thethe
filmfilm
thickness of each
thickness layer layer
of each by oblique sputtering.
by oblique The material
sputtering. for the free
The material forlayers
the
was Co 70.5 Fe 4.5 Si 15 B10 with excellent flatness and a soft magnetic property
free layers was Co70.5Fe4.5Si15B10 with excellent flatness and a soft magnetic property [23,25,26], which
has been reported to lead the small anisotropy field and high sensitivity. The free layers were
microfabricated to be 600 × 82 µm with the longitudinal side along the sensing axis. The stacking
structure of Sample-B prepared as a reference was Sub. Si, SiO2/Ta 5/Ru 20/Ta 5/Ni80Fe20
130/Ru 0.9/Co40Fe40B20 3/MgO/Co40Fe40B20 3/Ru 0.85/Co75Fe25 2/Ir22Mn78 10/Ta 5/Ru 10.
The films were deposited to have the incident sputtered particles perpendicular to the wafer.
The material of the free layers was the most commonly used crystalline ferromagnetic material,
Ni80Fe20 [24]. The free layers were microfabricated to be 100 × 140 µm with the short side along
Sensors 2023, 23, 646 5 of 18

the sensing axis. Both Sample-A and Sample-B were set on the print circuit boards as shown
in Figure 1a–c. Sample-A with the first and second MFCs showed transfer curves as shown
in Figure 1d, reflecting the typical M-H curve for the hard direction of the free layers. High
magnetic gain can be attained by placing the first MFCs to overlap the free layers but not the
pinned layer nor the tunnel barrier layer. Figure 1e shows the detectivity spectra of Sample-A
without MFCs, only with first MFCs and with the first and second MFCs measured with AC
magnetic fields at 10 Hz. The amplitudes of the magnetic fields were 50 nTp-p for the sample
without MFCs and 1 nTp-p with MFCs. One can see in the figure that two types of MFCs clearly
improve the detectivity. In the sample with the first and second MFCs, the detectivity was
4.8 pT/Hz1/2 at 1 Hz and 11.8 pTrms in the frequency band between 0.2 and 100 Hz, and the
sensor noise at 1 Hz was 1714 nV/Hz1/2/V. The power line noise at 50 Hz and 1/f noise in
a wide frequency range was reduced using the digital processing as shown below. Figure 1f
shows the histogram of the detectivities of 108 sensors of Sample-A and 124 sensors of Sample-B.
We obtained the average detectivity of 14.1 pTrms and the standard deviation, σ, of 1.9 pTrms,
which are much improved compared to Sample-B, whose averaged detectivity was 23.2 pTrms
and standard deviation was 16.4 pTrms.

3.2. Closed-Loop Systems

While the TMR sensors explained above have high sensitivities, the working range
is as small as ten nano teslas, which leads to a saturation of output in the terrestrial
magnetic field of about 50 micro teslas. Therefore, we constructed the closed-loop systems
using feedback coils as shown in Figure 2a–c. By producing the magnetic field canceling
the external field by introducing the current corresponding to the output of the bridge
circuit with TMR sensors enhanced with the driving amplifier (OP295, Analog Devices Inc.,
Wilmington, MA, USA), the operation point can be fixed. The magnitude of the magnetic
field applied to the sensor can be measured as the voltage of the resistance Rfb , Vout . One
can also obtain a better linearity to the external field by using closed-loop systems with
feedback coils fully surrounding the TMR sensors and the MFCs. With our configuration
composed of TMR sensors and the closed-loop system, we obtained the working range of
±80 µT and the non-linearity in the magnetic field range of ±50 µT (roughly corresponding
Sensors 2023, 23, x FOR PEER REVIEW
to the terrestrial field) of ±0.03%. By using this configuration, the ideal linear algebra
6 of 20
was
established among the magnetic vector data measured by the sensor array, which allowed
us to apply the advanced digital processing such as SSS shown in the next section.

Figure
Figure2.2.(a)(a)
Diagram
Diagram of measurement
of measurementusingusing
the closed-loop system.system.
the closed-loop The amplified differential
The amplified out-
differential
put from the bridge circuit with a TMR sensor used as a resistive element was induced to the feed-
output from the bridge circuit with a TMR sensor used as a resistive element was induced to the
back coil. (b) Photograph of TMR sensor covered by feedback coil. (c) Sensor output and non-line-
feedback
arity in thecoil. (b)ofPhotograph
range of TMR sensor
+/−50 μT. Non-linearity wascovered by from
calculated feedback coil. AD
Vout after (c) Sensor output
conversion, and non-
which
linearity in the range of +/ − 50 µT.
was settled within the error of +/−0.1%. Non-linearity was calculated from Vout after AD conversion,
which was settled within the error of +/−0.1%.
3.3. SSS
Next, we verified a reduction in the environmental fields by SSS using the TMR sen-
sor array configured as a closed-loop system. SSS is the method to divide the signal and
the external environmental noises by the least-squares method using multiple sensor data
Sensors 2023, 23, 646 6 of 18

3.3. SSS
Next, we verified a reduction in the environmental fields by SSS using the TMR sensor
array configured as a closed-loop system. SSS is the method to divide the signal and the
external environmental noises by the least-squares method using multiple sensor data [27].
As shown in Equation (1), the magnetic field vector, B(r), in the spherical coordinate can be
expressed with functions derived from the internal space of the sensor array (the first term
of the right side) and from the external space (the second term of the right side) with the
spherical harmonics Yl,m (θ, ϕ).

Lin l  
1
B(r) = −µ ∑ ∑ αl,m ·∇ ·Y (θ, ϕ)
|r |l +1 l,m
l =1 m=−l
(1)
Lout l  
−µ ∑ ∑ β l,m ·∇ |r |l ·Yl,m (θ, ϕ)
l =1 m=−l

By putting Φ as the magnetic field measured by the sensor array, Φ can be written as
a formula of matrix using Sin xin and Sout xout as follows:
 T
Φ = Φ(r1 ) Φ(r2 ) · · · Φ(rN ) = Φin + Φout
(2)
 
xin
Φ = Sin xin + Sout xout = [Sin Sout ]
xout

where Sin corresponds to the spherical harmonics of the internal signal in Equation (1),
Sout is the spherical harmonics of the external signal and xin and xout are the coefficients
of α and β, respectively. Sin and Sout are the base vectors that can be calculated with the
positions and sensitivities of the sensors. Here, by estimating xin and xout as xcin and x
d out
using the least-squares method using measured Φ and base vectors Sin and Sout , Φ can be
divided into the estimated value of the internal signal Φ
d in and the external signal Φout .
d
 
xcin
−1
x̂ = = St S St Φ (3)
xdout
 
h
b= Φ
i xcin
Φ din + Φ
dout = Sx̂ = [ Sin Sout ] (4)
xdout

Φin = Sin xin (5)

d c
Φ
dout = Sout x
dout (6)
where Φ b is used as the heart signal in the magnetocardiography measurements.
The experimental result of the environmental noise reduction by the SSS is shown
below. The sensor array was constructed using 96 sensor cells containing three TMR
sensors in each, and the TMR sensors were arranged perpendicularly, as shown in Figure 3a.
As can be seen from the upper side, 48 channels comprised the double layer. By three-
dimensionally surrounding the subject’s chest with the arched array through the underarms,
the noise reduction effect by SSS can be increased. Figure 3b shows the block diagram of
the measuring system. After detecting the Vout , as explained above, the computer received
the data through an RC high-pass filter (0.2 Hz) to eliminate the terrestrial magnetic
noises, amplifier (AD8676, Analog Devices Inc., Wilmington, MA, USA), RC low-pass
filter (194 kHz) for anti-aliasing, A/D converter (ADS1296, Texas Instruments Inc., Dallas,
TX, USA), the FPGA and the memories, followed by the signal processing, such as SSS.
The measured result of the noise reduction with this array configuration is shown in
Figure 3c. In a regular laboratory in a 26-storied office building, we observed the magnetic
field of 100 nTp-p at 7 Hz produced by a Helmholtz coil three meters away from the
sensor array, simulating the environmental magnetic field. By inputting the simulated
magnetic field with the known frequency, amplitude and direction, the reducing effect of
the environmental magnetic field by SSS could be evaluated. In the spectra before reducing
Sensors 2023, 23, 646 7 of 18

the noise (Raw data), one can see peaks corresponding to the environmental noises in the
low frequency domain due to elevators and so on, as well as ones around 20 Hz and 50 Hz
due to the experimental equipment and the air conditioning in the laboratory, other than
the input signal from the coil at 7 Hz. In the spectra after applying SSS (After SSS), both
the signal at 7 Hz and the environmental noises around 20 Hz and 50 Hz are decreased.
Comparing the mean values at 7 Hz along the Y-axis before and after applying SSS, the
Sensors 2023, 23, x FOR PEER REVIEW
reduction effect of the environmental noises was estimated to be −73 dB, achieving the
8 of 20
target value of −60 dB. Here, the calculation condition of the SSS was set to be Lin = 3 and
Lout = 4 in Equation (1).

Figure 3.
Figure (a) Diagram
3. (a) Diagram of of sensor
sensorarray.
array.(b)
(b)Diagram
Diagramofof magnetocardiography
magnetocardiography measurement.
measurement. Measured
Meas-
information
ured information of three-dimensional
of three-dimensional magnetic fieldfield
magnetic distribution
distributionis stored in the
is stored memory
in the memory via via
0.2 0.2
Hz
high-pass
Hz high-pass filter, low-noise
filter, amplifier,
low-noise 194 kHz
amplifier, low-pass
194 kHz filterfilter
low-pass and and
24-bit A/DA/D
24-bit converter. Collected
converter. data
Collected
in memory
data in memory wereweresent to
senta PC
to athrough FPGAFPGA
PC through after being storedstored
after being for a definite periodperiod
for a definite of time,
of where
time,
where
digitaldigital processing
processing took place.
took place. (c) Power
(c) Power spectrum
spectrum density
density beforebefore
and and
afterafter applying
applying SSSSSS meas-
measured
ured
with with the environmental
the environmental magnetic
magnetic field field at 7along
at 7 Hz Hz along the Y-axis.
the Y-axis.

The
The magnetocardiography measurementwith
magnetocardiography measurement withthe
thesensor
sensorarray
arrayisisdepicted
depicted inin Figure
Figure 4a.
4a.
TheTheSSSSSS calculation
calculation waswas performed
performed with with = 3 and
Lin = 3 and Lout = 4 in= Equation
4 in Equation (1), and
(1), and the
the data
for one
data forminute (50 heart
one minute (50 beats)
heart were
beats)used
wereforused
the signal average.
for the signal The waveforms
average. before and
The waveforms
after each digital processing are shown in Figure 4b: (i) shows the time series
before and after each digital processing are shown in Figure 4b: (i) shows the time series waveform of
the Z-sensor
waveform in the
of the lower in
Z-sensor layer of Channel
the lower layer 27 (ch27), where
of Channel the environmental
27 (ch27), where the environmen-noises of
sub-nano
tal noises oftesla aroundtesla
sub-nano 20 Hz due to20
around theHzairdue
conditioning and ones of several
to the air conditioning and onesnano of teslas at
several
50 Hzteslas
nano due toat the commercial
50 Hz due to theelectric powerelectric
commercial are super-positioned; (ii) shows the time
power are super-positioned; series
(ii) shows
data
the after
time applying
series data the SSS
after and thethe
applying digital
SSS filter,
and thewhere the filter,
digital environmental
where thenoises are almost
environmental
removed and R waves buried in the environmental noises are observed;
noises are almost removed and R waves buried in the environmental noises are observed; and (iii) shows the
and (iii) shows the smoothened waveform of 50 heart beats by the signal average for one
minute, where the QRS and T waves are clearly observed.
Sensors 2023, 23, 646 8 of 18

Sensors 2023, 23, x FOR PEER REVIEW 9 of 20

smoothened waveform of 50 heart beats by the signal average for one minute, where the
QRS and T waves are clearly observed.

Figure 4. (a) Photograph of magnetocardiography measurement. Cells underneath show the channel
Figure 4. (a) Photograph of magnetocardiography measurement. Cells underneath show the chan-
number
nel numberat the
at corresponding position.
the corresponding (b) Time
position. seriesseries
(b) Time of magnetocardiogram
of magnetocardiogramin thein
digital processing
the digital pro-
measured by Z-sensors in the lower layer in ch27: (i) Raw data (Φ), (ii) After SSS and Digital
cessing measured by Z-sensors in the lower layer in ch27: (i) Raw data ( ), (ii) After SSS and Digital Filter
and (iii)
Filter andAfter signalsignal
(iii) After average for one
average forminute (50 heart
one minute (50 beats).
heart beats).

3.4. Projection operation

3.4. Projection operation
Further reduction in the sensor noise was performed using the projection operation.
Further reduction in the sensor noise was performed using the projection operation.
The projection matrix used for the projection operation is calculated based on the method
The projection matrix used for the projection operation is calculated based on the method
of cross validation. Cross validation is a calculation algorithm with the repetition of the
of cross validation. Cross validation is a calculation algorithm with the repetition of the
analysis with a part of the dataset divided into k groups and a validity check with the rest.
analysis with a part of the dataset divided into k groups and a validity check with the rest.
In this work, we divided the data from the sensor array into two parts (k = 2) and calculated
In this work, we divided the data from the sensor array into two parts (k = 2) and calcu-
the projection matrix by computing the base vector for each group. Figure 5a shows the
lated the projection matrix by computing the base vector for each group. Figure 5a shows
flowchart of the projection operation. Adding in calculating Φ with all the data from the
the flowchart of the projection operation. Adding in calculating with all the data from
whole sensor array, the digital processing from the SSS to the signal average was separately
the whole sensor array, the digital processing from the
performed with data from channels with even numbers, ΦA , and with odd SSS to the signal average was sep-
numbers, ΦB ,
arately performed with data from channels with even numbers, , and with odd num-
to obtain Φ \ and Φ [ in_B . In addition, the data of unused channels in the calculations
bers, , toin_A
obtain _ and _ . In addition, the data of unused channels in the cal-
of Φin_A and were
and estimated werebyestimated
computing in xˆin corresponding
by Scomputing to each channel
\ [
culations of Φin_B _ _ corresponding to
using xc in obtained in Equation (3). Here, we performed a singular value decomposition
each channel using obtained in Equation (3). Here, we performed a singular value
on Φ\ in_A and Φ
decomposition
[ in_B to obtain
on the base vectors L1 and L2 in the time and
domain (ortime
spatial
_ and _ to obtain the base vectors in the
T and P = L ·LT .
do-

domain), followed by calculating the projection
main (or spatial domain), followed by calculating the projection matrices P 1 = L · L
1 matrices
1 2 (= 2 ∙ 2 )
Here, P(=
and 1 and ∙P2 are projection
). Here, andmatrices aretoprojection
the spacematrices
consisting to of
thethespacesignals and noises
consisting of
of the
selected sensors. Comparing the Φ in calculated
signals and noises of selected sensors. Comparing the
d using data from all channels, the
calculated using data from allprojection
operation the
channels, to Pprojection
1 and P2 enabled
operationus totoreduce andthe sensor noises
enabled usbytoretaining
reduce the thesensor
amplitude
noisesof
by retaining the amplitude of the signals (i.e., the projection to reduced the noises de-
rived from the channels with odd numbers and did the same from the channels with
even numbers compared to ).
Sensors 2023, 23, 646 9 of 18

Sensors 2023, 23, x FOR PEER REVIEW 10 of 20

the signals (i.e., the projection to P1 reduced the noises derived from the channels with odd
numbers and P2 did the same from the channels with even numbers compared to Φ d in ).

Figure
Figure5.5.(a)
(a)Processing
Processingflow
flowofofprojection
projectionoperation.
operation.(b)
(b)Time
Timeseries
seriesof
ofZ-sensors
Z-sensorsin
inthe
thelower
lowerlayer
layer
in
inch2
ch2measured
measuredwithout
withoutthe
thesubject
subjectto
toevaluate
evaluatethe
thechange
changeof ofnoises
noisesbefore
beforeand
andafter
afterthe
theprojection
projection
operation.
operation.TheThestandard deviation
standard before
deviation and
before after
and thethe
after projection operation
projection was was
operation calculated to beto
calculated 1.77
be
pTrms and 0.99 pTrms, respectively. (c) Magnetocardiogram map obtained with the Z-sensors in the
1.77 pTrms and 0.99 pTrms , respectively. (c) Magnetocardiogram map obtained with the Z-sensors in
lower layers in 48 channels after SSS, digital filter, signal average and projection operation.
the lower layers in 48 channels after SSS, digital filter, signal average and projection operation.

Figure
Figure 5b showsthe
5b shows thetime
timedomain
domain data
data of the
of the Z-sensor
Z-sensor in lower
in the the lower
layer layer of ch27.
of ch27. These
These measurements were carried out without the subject to compare the sensor
measurements were carried out without the subject to compare the sensor noises before noises
before andthe
and after after the projection
projection operation.
operation. While
While σ = 1.77σ pT
= 1.77 pTrms before the projection oper-
rms before the projection operation, it
ation, it decreased to 0.99 pTrms after the projection operation, which is below the target
value of the sensor detectivity 1.0 pTrms. It was also confirmed that the noise reduction
effect was −4.5 dB by comparing the second norms of all the sensors.
Sensors 2023, 23, 646 10 of 18

decreased to 0.99 pTrms after the projection operation, which is below the target value of
the sensor detectivity 1.0 pTrms . It was also confirmed that the noise reduction effect was
−4.5 dB by comparing the second norms of all the sensors.
Figure 5c shows the heart magnetic field maps obtained with the Z-sensors of the
lower layer close to the body surface in 8 × 6 channels measured through the flowchart
shown in Figure 5a. It can be seen that the polarity of the heart signal reverses near the
heart, suggesting that the heart magnetic field is observed. The signals of R waves can be
observed with the sensors away from the heart, including ones placed on the side of the
body. These results mean that heart magnetic fields are observed from three-dimensional
directions, which is expected to improve the accuracy of the current source estimation
compared to measurements only from the front.

3.5. Preparing the Magnetocardiogram

To create detailed current arrow maps upon the heart, the dataset was reconstructed
onto the divided area of 200 × 200 mm at 40 mm above the body surface with 9 × 9 segments
at 25 mm intervals. Once xc in was calculated by SSS, the magnetic field distribution on
an imaginary sensor array surface, different from the real positions and angles of the
sensors, could be easily estimated by preparing vector Sin for xc in relating to the positions
and angles to be calculated. Therefore, the magnetocardiogram can be optimized for
diagnostic application as required. Figure 6a shows the magnetocardiogram reconstructed
by estimating the Z-axis magnetic field, which was processed through the same procedure
shown in Figure 5a except that Sin was adjusted in the SSS calculation.
Figure 6b shows the magnetic field strength maps in contour lines at peaks of R
and T waves along the Z-axis and the current arrow map calculated from the positional
differentiation along the Z-axis. The current arrow maps were calculated by Ix = ∆Bz /∆y
and Iy = −∆Bz /∆x. At the R and T waves, it can be seen that the current passes to the lower
left of the heart, which is similar to the pattern observed in a healthy subject [10]. Figure 7a
shows the time variation of the current arrow map in the QRS domain. The calculation
conditions of SSS were Lin = 3 and Lout = 4 and Lin = 6 and Lout = 4. Although Lin = 6 gave
slightly more detailed information in the domain near the S wave, both conditions resulted
in an extremely clear magnetocardiogram comparing favorably with a measurement by a
SQUID performed in magnetically shielded rooms (Figure 7b) in the QRS and ST domains.
In the time variation during the QRS domain, the magnetocardiogram is showing the
excitation of the ventricular septum (17 ms), leading to the excitation of the left ventricle
(40 ms) and current flow to the right ventricle (65 ms). In these QRS and ST domains,
relationships have been reported between the abnormality in the magnetocardiogram maps
and heart disease, such as ischemic heart disease [6,7] and ventricular arrhythmia [8–10].
The spatial resolution can be improved by increasing Lin , which is important for
pattern analysis on unhealthy subjects. In this work, the measurements up to Lin = 6
allowed us to observe the detailed distribution on a healthy subject. However, we still need
to collect further experimental data to distinguish how large Lin needs to be, but we believe
this system can be applied to disease determination, such as in a previous report [28], by
setting an appropriate Lin .
Sensors2023,
Sensors 23, x646
2023,23, FOR PEER REVIEW 1211ofof20
18

Figure
Figure 6.
6. (a)
(a) Top:
Top: magnetocardiogram
magnetocardiogrammapmap reconstructed
reconstructedin in the
the area
area of
of 200
200 ××200
200mm mmat at 40
40 mm
mm
above
abovethe
thebody
body surface
surfacewith
with 25
25mm
mm intervals;
intervals; lower
lower left:
left: reconstructed
reconstructedarea;
area; lower
lower right:
right: expanded
expanded
magnetocardiograph
magnetocardiograph ofofdata
datasurrounded
surroundedwith
with thethe
redred
square on the
square magnetocardiogram
on the magnetocardiogram map.map.
(b)
Magnetic field strength maps in contour lines and current arrow maps at peaks of R and T waves.
 = 6 gave slightly more detailed information in the domain near the S wave, both conditions
resulted in an extremely clear magnetocardiogram comparing favorably with a measure-
ment by a SQUID performed in magnetically shielded rooms (Figure 7b) in the QRS and
ST domains. In the time variation during the QRS domain, the magnetocardiogram is
showing the excitation of the ventricular septum (17 ms), leading to the excitation of the
Sensors 2023, 23, 646 12 of 18
left ventricle (40 ms) and current flow to the right ventricle (65 ms). In these QRS and ST
domains, relationships have been reported between the abnormality in the magnetocar-
diogram maps and heart disease, such as ischemic heart disease [6,7] and ventricular ar-
rhythmia [8–10].
(b) Magnetic field strength maps in contour lines and current arrow maps at peaks of R and T waves.

Figure 7. (a) Time series of current arrow maps in the QRS domain (300~390 msec in the lower left
Figure 7. (a) Time series of current arrow maps in the QRS domain (300~390 msec in the lower left
image in Figure 6a) at each time point referencing the beginning of the QRS wave (300 msec). SSS was
image in Figure 6a) at each time point referencing the beginning of the QRS wave (300 msec). SSS
performed
was performed Lin =L6inand
with with LoutL=out4=using
= 6 and sensors
4 using in the
sensors lower
in the layer
lower of of
layer each channel.
each (b)(b)
channel. Reference of
Reference
current
of arrow
current maps
arrow measured
maps measuredby 64bychannels of SQUID
64 channels sensors
of SQUID with magnetically
sensors shielded
with magnetically rooms
shielded
(Kimura et al., Circ.
rooms (Kimura J. 2018).
et al., Circ. J. 2018).

4. Discussion
The spatial resolution can be improved by increasing , which is important for pat-
4.1. TMR
tern Sensors
analysis on unhealthy subjects. In this work, the measurements up to = 6 allowed
us toDuring
observe thethe
thin film deposition
detailed in this
distribution on work, the incident
a healthy subject. angle of sputtering
However, we still particles
need to
was set to be about 39 degrees onto the wafer surface
collect further experimental data to distinguish how large to minimize needs to be, butdistribution
the in-plane we believe
of film
this thickness
system can be to applied
reduce the variation
to disease of resistivities such
determination, and the sensitivities
as in a previousofreport
sensors taken
[28], by
out from
setting anaappropriate
wafer. In this configuration,
. the variation of the detectivities of Sample-A was
greatly reduced compared to that of Sample-B deposited by facing target type sputtering. In
general, oblique sputtering deposition with crystalline targets is known to lead the growth
of columnar crystals on wafers, leading to the deterioration of the flatness. When the flatness
of the tunnel barrier layer in the TMR thin films is worsened, the electronic 1/f noise is
increased due to the charge trapping in the tunnel barrier layer and at the interfaces of
the tunnel junctions [29]. Moreover, the degrading of the flatness leads to magnetization
domain hopping between the metastable ripple states, causing the enhanced magnetic 1/f
noise [30]. Figure A1a,b in Appendix A shows cross-sectional images observed by scanning
transmission electron microscope (STEM) on the TMR thin films with Ni80 Fe20 free layers
deposited by facing target type sputtering and oblique sputtering, respectively. In both
images, characteristic shadows of columnar crystals in Ni80 Fe20 can be seen. The grain sizes
of those crystals are increased in the films deposited by oblique sputtering, which causes the
deteriorated flatness of the MgO tunnel barrier layer. Here, to inhibit the development of
crystallization in the free layers, amorphous Co70.5 Fe4.5 Si15 B10 was selected as the material
for the free layers. As a result, the columnar crystals were significantly reduced, and the
flatness of MgO was greatly improved. A previous work [26] reported that the flatness
of Co70.5 Fe4.5 Si15 B10 with a thickness of 30 nm showed a similar value to that of Ni80 Fe20 .
The difference between the previous report [26] and our work may be explained by the
Sensors 2023, 23, 646 13 of 18

presence or absence of the development of columnar crystals being emphasized in our

work because the free layer was as thick as 130 nm. The relationships between the average
surface roughness, Ra , measured by the atomic force microscope (AFM) and the magnetic
and electronic noises fabricated in this work are shown in Figure A1d,e, respectively. The
shape of the free layers in each sample is the same as in Sample-B. Here, the electronic
1/f noise was measured with the external magnetic field along the sensing axis to saturate
the sensors, and the magnetic 1/f noise showed the value obtained by eliminating the
effect of electronic noise from that measured at the center of the operation range in the
magneto-resistive curve. As expected, the magnetic noise decreased with the Ra . On the
other hand, the electronic noise did not show the simple behavior against Ra . In samples
deposited by oblique sputtering, the use of Co70.5 Fe4.5 Si15 B10 resulted in a lower Ra and
electronic noise. However, in the samples with Ni80 Fe20 free layers the facing target type
sputtering induced improved electronic noise compared to oblique sputtering. This may be
caused by the composition ratio, the crystal orientation or the defect density in the MgO
layers, although the cause is not clear at present.
Next, we discuss the shape of the free layers to obtain higher sensitivity. In general,
the diamagnetic field, which occurs with the magnetic field in ferromagnetic materials,
is decreased as the material is longer along the field [31]. Therefore, it was expected
that a lower diamagnetic field and a higher sensitivity would be achieved by stretching
the free layer along the sensing axis. However, when the free layer contained a single
Co70.5 Fe4.5 Si15 B10 sheet, the magnetic domain structure in the Co70.5 Fe4.5 Si15 B10 layer
appeared to set the vector sum of the magnetic moment in the free layer to zero. Figure A2a
shows the image observed with a Kerr microscope in a multilayer structure of Sub. Si,
SiO2 /Ta 5/Ru 20/Ta 5/Co70.5 Fe4.5 Si15 B10 170/Ru 0.4/Co40 Fe40 B20 3/Ta 2.5 patterned to
be 100 × 140 µm. The triangle magnetic domain structure can be seen at the bottom of
the pattern, which shows the development of a closure domain. As the direction of the
magnetic easy axis was parallel or antiparallel to the sensing axis in the triangle domain,
the hysteresis of the magneto-resistive curve was developed when this domain was directly
under the magnetic tunnel junctions, decreasing the sensitivity near the zero-field state.
When stretching the free layer along the sensing axis, the triangle domain expanded in the
free layer, leading to a decrease in the sensitivity. Therefore, the magnetic domains must be
unified to obtain long free layers along the sensing axis. In this work, we aimed to eliminate
the magnetic domain structure by sandwiching the non-magnetic Ru film with double
Co70.5 Fe4.5 Si15 B10 sheets in the free layer to magnetostatically couple the Co70.5 Fe4.5 Si15 B10
sheets on the side of the patterns. The result of magnetic domain observation on a pattern
with a multilayer structure of Sub. Si, SiO2 /Ta 5/Ru 20/Ta 5/Co70.5 Fe4.5 Si15 B10 100/Ru
10/Co70.5 Fe4.5 Si15 B10 100/Ru 0.4/Co40 Fe40 B20 /Ta 2.5 is shown in Figure A2b. In this case,
the whole pattern had a single magnetic domain. By using this stacking structure and
microfabricating the free layer to be very long along the sensing axis, the sensitivity of
Sample-A in the frequency band between 0.2 and 100 Hz reached 327 µV/nT/V on average,
which is improved compared to 116 µV/nT/V in Sample-B.

4.2. Environmental Noise

For the magnetocardiography measurements without magnetically shielded rooms,
the largest problem is the noises, such as the terrestrial magnetic field and city noise.
Particularly, city noise drastically interrupts the heart field as its amplitude is in the or-
der of nano teslas in the same frequency band as heart beats. Therefore, magnetic field
shielding by software is crucially important. Gradiometers are widely used to reduce
noise disturbance [17,20]. While SSS requires a large number of sensors for the calculation
process according to the degree of calculation (Lin , Lout ), a gradiometer has advantages
in that it needs only two sensors, and the configuration of the system is simple and easy
to handle. However, a gradiometer is difficult to use with environmental noise where a
spatial gradient exists. In addition, as the gradiometer is a method to obtain the difference
in outputs between different sensors, it decreases not only the environmental noise but
Sensors 2023, 23, 646 14 of 18

also the signal itself. As the noises have no correlation between sensors, the noise of the
gradiometer is enhanced after the subtraction.
In this work, we succeeded in decreasing the environmental noise by −73 dB using the
software shield of SSS and without sacrificing the detectivity. General single magnetically
shielded rooms decrease the environmental noise by about −20 dB, which means our SSS
has the effect of more than a triple magnetically shielded room.
Performing environmental noise reduction in SSS depends on the configuration and
error of sensors. We assume there are two reasons why we could obtain the large noise
reduction effect. The first is the use of three-axis sensors that three-dimensionally sur-
rounded the body through the underarms, leading to an increase in the information of the
spatial magnetic field distribution and to the improvement of the calculation accuracy with
spherical harmonics used for the base vectors. Let us compare the condition number that
indicates the accuracy of the estimated solution by computing the ratio of the maximum
and minimum singular values under a condition of Lin = 6 and Lout = 4. In the configuration
where the three-axis sensors were three-dimensionally placed to the side of the body, the
condition number was calculated to be 1226. When the sensors on the side of the body
were lifted upward to form the flattened array shape, the condition number increased to
9849, which is worse than that of the original configuration. In addition, when all sensors
in the array were replaced with Z-axis sensors, the condition number was 2305, worse than
that with three-axis sensors.
The second reason is the adoption of a closed-loop system with feedback coils sur-
rounding the whole TMR sensor configuration to increase the linearity against the input
magnetic field, which improved the accuracy of the sensors enough to apply the advanced
digital signal processing. It is known that the sensor errors affect the reduction in envi-
ronmental noise [27]. In the array we constructed in this work, we estimated that the
environmental noise reduction effect would be −66.8 dB and −61.3 dB when the non-
linearity was 0.1% and 0.3%, respectively. We presumed that we could obtain the almost
expected noise reduction effect by improving the sensor non-linearity down to about 0.03%.
Referencing the power spectral density in Figure 3c, the environmental noise of about
10 nT at 50 Hz derived from the commercial electric power will not be problematic as it
can be eliminated with the notch filter. The purpose of SSS is to reduce the noises around
20 Hz and the environmental 1/f noise. These noises are originally less than 1 nT and
decrease below 1 pT after reducing by −73 dB using SSS. Moreover, they decrease to
about 0.1 pT after the signal average for one minute, which is one digit smaller than the
sensor noise. Although there were a number of noise sources, such as elevators and air
conditioning around the measurement location, we used no countermeasures, such as
placing the measurement system away from the noise sources. We think that there is
almost no limitation to the installation location even without magnetically shielded rooms.
Additionally, the heart magnetic signal and the environmental noises from the external
space were separately obtained by SSS. Therefore, it is possible to avoid the disturbance by
suspending the measurement when sudden environmental noises occur or eliminating the
measurement data while the environmental noises exceed the default value by monitoring
the external field.

4.3. Sensor Noise

While TMR sensors were used in this work, GMR sensors are also known as magnetic
sensors with high detectivities operatable at room temperatures. In general, while TMR
sensors have a higher magneto-resistance ratio and sensitivity than GMR sensors, the
electronic 1/f noise that occurs in the insulating MgO barrier in the low frequency band is
larger than with GMR sensors [21]. Nevertheless, sensors with higher sensitivities hide the
noises of the following circuit such as amplifiers and resistive elements. Therefore, even if
the detectivities are the same in TMR and GMR sensors, the use of TMR sensors is expected
to result in a lower final noise as a module after digital processing than when using GMR
sensors. In the magnetocardiograph fabricated in this work, we adopted electronic devices
Sensors 2023, 23, 646 15 of 18

with low noises, and the system noise of the circuit was estimated to be 1.5 pTrms between
0.2 and 100 Hz as a designed value before the signal average, which is thought to hardly
have any impact compared to the noise of TMR sensors (14.1 pTrms ).
Finally, we discuss the noise reduction in SSS based on the detectivity of the TMR
sensors and the final detectivity after digital processing, even though the direct estimation
was difficult as the environmental noise was dominant before SSS. The original detectivity
of the TMR sensors, 14.1 pTrms , was decreased to 0.99 pTrms , by −23.0 dB, through the
digital processing of the SSS, signal average and projection operation. The noise reduction
effect by the signal average for one minute (50 heart beats) was estimated to be −17.0 dB,
and the SSS and projection operation were estimated to reduce the noise by −6.0 dB.
Adding to the −4.5 dB at the projection operation part, the SSS also decreased the noise by
about −1.5 dB. A part of the sensor noise, as well as the environmental noise, was thought
to be cut away to the external space, which helped to achieve the target detectivity of 1.0 pT.

5. Conclusions
In this work, we succeeded in fabricating a high-accuracy magnetocardiograph with-
out magnetically shielded rooms by improving the TMR sensors and decreasing envi-
ronmental noises and sensor noises with digital processing. TMR sensors with a high
detectivity of 14.1 pTrms were prepared by applying a new multilayer structure and the
shape of free layers. The large 1/f noise in TMR sensors was reduced by selecting the
amorphous Co70.5 Fe4.5 Si15 B10 known to have the excellent flatness. The variation of the
detectivities was suppressed by oblique sputtering. By adopting SSS, the reduction effect
of the environmental noise attained −73 dB, which overtakes triple magnetically shielded
rooms. Moreover, the digital processing combining SSS and the projection operation led to
a reduction in the sensor noise by −6.0 dB. As the result, the detectivity of the magneto-
cardiograph without a magnetically shielded room reached 1.0 pTrms with the integration
for one minute in the frequency band of 0.2–100 Hz. By using this system, we obtained a
remarkably fine magnetocardiogram and current arrow maps in the QRS and ST segments,
comparing favorably with a measurement by a conventional SQUID performed in magneti-
cally shielded rooms. As the clear magnetocardiogram was obtained without magnetically
shielded rooms at room temperature, the future task is to examine the availability of this
magnetocardiograph through clinical trials. Although the P wave due to atrial depolar-
ization was not clearly observed because of the shortage of the detectivity with the signal
average for one minute, it may be possible to detect it by further improvements of the TMR
sensors and digital processing. The spatial resolution and heart field detectivity can be
improved by placing sensors closer to the heart by reforming the shape of the frame of the
array and the setting of the sensors, as the distance between the sensors in the sensor array
shown in this work and the body surface was approximately 40 mm.

Author Contributions: Conceptualization, all authors; methodology, K.K., M.K., T.N. (Takafumi Nakano),
K.F., M.Y., M.M. and K.S.; software, M.K., T.N. (Takenobu Nakamura) and K.I.; validation, S.K.
(Seiichi Kato), S.O., Y.M., M.O., Y.A., S.K. (Seiji Kumagai) and H.M. (Hidenori Mochizuki); formal analysis,
K.K., M.K. and M.M.; investigation, K.K., M.K., T.N. (Takafumi Nakano), K.F., T.N. (Takenobu Nakamura),
K.I., M.Y., M.M. and K.S.; resources, M.O., Y.A., S.K. (Seiji Kumagai) and H.M. (Hidenori Mochizuki);
data curation, all authors; writing—original draft preparation, K.K. and M.K.; writing—review and
editing, M.O., Y.A., S.K. (Seiji Kumagai) and H.M. (Hidenori Mochizuki); visualization, M.K. and
T.N. (Takenobu Nakamura); supervision, M.O., Y.A., S.K. (Seiji Kumagai), H.M. (Hitoshi Matsuzaki)
and H.M. (Hidenori Mochizuki); project, M.O., Y.A., S.K. (Seiji Kumagai), H.M. (Hitoshi Matsuzaki) and
H.M. (Hidenori Mochizuki); funding acquisition, H.M. (Hidenori Mochizuki). All authors have read and
agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: The study was conducted in accordance with the Declaration
of Helsinki and approved by the Research Ethic Committee of Asahi Kasei for studies involving
humans (approval number: 2110).
 Sensors 2023, 23, 646 16 of 18

Informed Consent Statement: Informed consent was obtained from all the subjects involved in
the study.
Data Availability Statement: Data are available from the corresponding author upon request.
Sensors 2023, 23, x FOR PEER REVIEW 18 of 20
Acknowledgments: This work was partly supported by the CSIS and CIES, Tohoku University.
Conflicts of Interest: The authors declare no conflict of interest.
Appendix A
Appendix A

FigureA1.
Figure (a–c)
A1.(a–c) Cross-sectional
Cross-sectional STEMSTEMimageimage of TMR
of TMR multilayers.
multilayers. The stacking
The stacking structurestructure
of (a,b) isof
(a,b)Si,isSiO
Sub. Sub.2/TaSi,5/Ru
SiO20/Ta
2 /Ta 5/Ru 20/Ta130/Ru
5/free-layer 5/free-layer
0.9/Co130/Ru
40Fe40B20 0.9/Co 40 Fe40
3/MgO/Co B20
40Fe40B3/MgO/Co
20 3/Ru 0.85/Co 40 Fe FeB2520
7540

2/3/Ru
Ir22Mn 78 10/ Ta
0.85/Co 5 25
75 Fe / Ru
2/10
Ir22inMnnm. The Ta
78 10/ table
5 /shows
Ru 10 the configuration
in nm. of the the
The table shows sputtering systemof
configuration andthe
the materialsystem
sputtering of theandfreethe
layer. (d,e) of
material Average
the free surface roughness,
layer. (d,e) Average R a, dependence
surface roughness, of Rmagnetic
a , and
dependence
electronic
of magnetic noises,
andrespectively. The longitudinal
electronic noises, respectively.error
Thebars show theerror
longitudinal difference between
bars show the the maxi-
difference
mum or minimum values and the average value in three samples prepared
between the maximum or minimum values and the average value in three samples prepared in the in the same condition.
The transverse error bars show the empirical measurement errors.
same condition. The transverse error bars show the empirical measurement errors.
 Figure A1. (a–c) Cross-sectional STEM image of TMR multilayers. The stacking structure of (a,b) is
Sub. Si, SiO2/Ta 5/Ru 20/Ta 5/free-layer 130/Ru 0.9/Co40Fe40B20 3/MgO/Co40Fe40B20 3/Ru 0.85/Co75Fe25
2/ Ir22Mn78 10/ Ta 5 / Ru 10 in nm. The table shows the configuration of the sputtering system and
the material of the free layer. (d,e) Average surface roughness, Ra, dependence of magnetic and
Sensors 2023, 23, 646
electronic noises, respectively. The longitudinal error bars show the difference between the 17
maxi-
of 18
mum or minimum values and the average value in three samples prepared in the same condition.
The transverse error bars show the empirical measurement errors.

Figure A2. Magnetic domain structure observed in patterned deposited wafer of (a) Sub. Si, SiO2/Ta
Figure A2. Magnetic domain structure observed in patterned deposited wafer of (a) Sub. Si, SiO2 /Ta
5/Ru 20/Ta 5/Co70.5Fe4.5Si15B10 170/Ru 0.4/Co40Fe40B20 3/Ta 2.5 and (b) Sub. Si, SiO2/Ta 5/Ru 20/Ta
5/Ru 20/Ta 5/Co70.5 Fe4.5 Si15 B10 170/Ru 0.4/Co40 Fe40 B20 3/Ta 2.5 and (b) Sub. Si, SiO2 /Ta 5/Ru
20/Ta 5/Co70.5 Fe4.5 Si15 B10 100/Ru 10/Co70.5 Fe4.5 Si15 B10 100/Ru 0.4/Co40 Fe40 B20 3/Ta 2.5. The
pattern sizes are 100 × 140 µm in both. The broken lines on the right-hand diagram show the
observed area.

References
1. Fenici, R.; Brisinda, D.; Meloni, M. Clinical application ofmagnetocardiography. Expert Rev. Mol. Diagn. 2005, 5, 291–313.
[CrossRef]
2. Park, J.-W.; Hill, P.M.; Chung, N.; Hugenholtz, P.G.; Jung, F. Magnetocardiography Predicts Coronary Artery Disease in Patients
with Acute Chest Pain. Ann. Noninvasive Electrophysiol. 2005, 10, 312–323. [CrossRef] [PubMed]
3. Hailer, B. The Value of MCG in PTS with and Without Relevant Stenoses of the Coronary Arteries Using an Unshielded System.
PACE 2005, 28, 8–16. [CrossRef] [PubMed]
4. Hailer, B.; Leeuwen, P.V.; Chaikovsky, I.; Auth-Eisernitz, S.; Schäfer, H.; Grönemeyer, D. The Value ofmagnetocardiography in the
Course of Coronary Intervention. Ann. Noninvasive Electrophysiol. 2005, 10, 188–196. [CrossRef]
5. On, K.; Watanabe, S.; Yamada, S.; Takeyasu, N.; Nakagawa, Y.; Nishina, H.; Morimoto, T.; Aihara, H.; Kimura, T.; Sato, Y.; et al.
Integral Value of JT Interval inmagnetocardiography in Sensitive to Coronary Stenosis and Improves Soon After Coronary
Revascularization. Circ. J. 2007, 71, 1586–1592. [CrossRef]
6. Zhao, C.; Chen, Y.; Zhao, D.; Wang, J.; Chen, Q. Comparison betweenmagnetocardiography and myocardial perfusion imaging in
the diagnosis of myocardial ischemia in patients with coronary artery disease. Int. J. Clin. Exp. Med. 2019, 12, 13127–13134.
7. Wu, Y.-W.; Lin, L.-C.; Tseng, W.-K.; Liu, Y.-B.; Kao, H.-L.; Lin, M.-S.; Huang, H.-C.; Wang, S.-Y.; Horng, H.-E.; Yang, H.-C.; et al.
QTc Heterogeneity in Restmagnetocardiography in Sensitive to Detet Coronary Artery Disease: In Comparison with Stress
Myocardial Perfusion Imaging. Acta Cardiol. Sin. 2014, 30, 445–454.
8. Joung, B.; Kim, K.; Lee, Y.-H.; Kwon, H.; Lim, H.K.; Kim, T.-U.; Ko, Y.-G.; Lee, M.; Chung, N.; Kim, S. Magnetic Dispersion of the
Later Repolarization in Brugada Syndrome. Circ. J. 2008, 72, 94–101. [CrossRef]
9. Kawakami, S.; Takaki, H.; Hashimoto, S.; Kimura, Y.; Nakashima, T.; Aiba, T.; Kusano, K.F.; Kamakura, S.; Yasuda, S.; Sugimachi,
M. Utility of High-Resolutionmagnetocardiography to Predict Later Cardiac Events in Noniscemic Cardiomyopathy Patients
With Normal QRS Duration. Circ. J. 2017, 81, 44–51. [CrossRef]
10. Kimura, Y.; Takaki, H.; Inoue, Y.Y.; Oguchi, Y.; Nagayama, T.; Nakashima, T.; Kawakami, S.; Nagase, S.; Noda, T.; Aiba, T.; et al.
Isolated Late Activation Detected bymagnetocardiography Predicts Future Lethal Ventricular Arrhythmic Events in Patients With
Arrhythmogenic Right Ventricular Cardiomyopathy. Circ. J. 2018, 82, 78–86. [CrossRef]
11. Hailer, B.; Leeuwen, P.-V. Prediction of malignant arrhythmias after myocardial infarction on the basis of MCG. In Proceedings of
the 12th International Conference on Biomagnetism, Espoo, Finland, 13–17 August 2000; pp. 504–507.
12. Ando, Y.; Nishikawa, T.; Fujiwara, K.; Oogane, M.; Kato, D.; Naganuma, H. Present and perspective of bio-magnetic measurement
using ferromagnetic tunnel junctions. Jpn. Soc. Med. Biol. Eng. 2014, 52, OS-33–OS-34.
13. Boto, E.; Holmes, N.; Leggett, J.; Roberts, G.; Shah, V.; Meyer, S.S.; Muñoz, L.D.; Mullinger, K.; Tierney, T.M.; Bestmann, S.; et al.
Moving magnetoencephalography towards real-world applications with a wearable system. Nature 2018, 555, 657–661. [CrossRef]
14. Knappe, S.; Sander, T.; Trahms, L. Optically pumped magnetometers for MEG. In Magnetoencephalography, 2nd ed.; Supek, S.,
Aine, C.J., Eds.; Springer Nature: Cham, Switzerland, 2019; pp. 1301–1312.
15. Arai, K.; Kuwahata, A.; Nishitaki, D.; Fujisaki, I.; Matsuki, R.; Xin, Z.; Nishio, Y.; Cao, X.; Hatano, Y.; Onoda, S.; et al. Millimetre-
scale magnetocardiography of licing rats using a solid-state quantum sensor. arXiv 2021, arXiv:2105.11676.
16. Hayashi, K.; Matsuzaki, Y.; Ashida, T.; Onoda, S.; Abe, H.; Ohshima, T.; Hatano, M.; Taniguchi, T.; Morishita, H.; Fujiwara,
M.; et al. Experimental and Theoretical Analysis of Noise Strength and Environmental Correlation Time for Ensembles of
Nitrogen-Vacancy Centers in Diamond. J. Phys. Soc. Jpn. 2020, 89, 054708. [CrossRef]
17. Kanno, A.; Nakasato, N.; Oogane, M.; Fujiwara, K.; Nakano, T.; Arimoto, T.; Matsuzaki, H.; Ando, Y. Scalp attached tangential
magnetoencephalography using tunnel magneto-resistive sensors. Sci. Rep. 2022, 12, 6106. [CrossRef] [PubMed]
18. Shirai, Y.; Hirao, K.; Shibuya, T.; Okuwa, S.; Hasegawa, Y.; Adachi, Y.; Sekihara, K.; Kawabata, S. Magnetocardiography Using a
Magnetoresistive Sensor Array. Int. Heart J. 2019, 60, 50–54. [CrossRef]
Sensors 2023, 23, 646 18 of 18

19. Adachi, Y.; Oyama, D.; Terazono, Y.; Hayashi, T.; Shibuya, T.; Kawabata, S. Calibration of Room Temperature Magnetic Sensor
Array for Biomagnetic Measurement. IEEE Trans. Magn. 2019, 55, 1000506. [CrossRef]
20. Zhao, W.; Tao, X.; Ye, C.; Tao, Y. Tunnel Magnetoresistance Sensor with AC Modulation and Impedance Compensation for
Ultra-Weak Magnetic Field Measurement. Sensors 2022, 22, 1021. [CrossRef] [PubMed]
21. Davies, J.E.; Watts, J.D.; Novotny, J.; Huang, D.; Eames, P.G. Magnetoresistive sensor detectivity: A comparative analysis. App.
Phys. Lett. 2021, 118, 062401. [CrossRef]
22. Watanabe, S.; Yamada, S. Magnetocardiography in Early Detection of Electromagnetic Abnormality in Ischemic Heart Disease. J.
Arrhythmia 2008, 24, 4. [CrossRef]
23. Oogane, M.; Fujiwara, K.; Kanno, A.; Nakano, T.; Wagatsuma, H.; Arimoto, T.; Mizukami, S.; Kumagai, S.; Matsuzaki, H.;
Nakasato, N.; et al. Sub-pT magnetic field detection by tunnel magneto-resistive sensors. Appl. Phys. Express 2021, 14, 123002.
[CrossRef]
24. Fujiwara, K.; Oogane, M.; Nishikawa, T.L.; Naganuma, H.; Ando, Y. Fabrication of magnetic tunnel junctions with a bottom
synthetic antiferro-coupled free layers for high sensitive magnetic field sensor device. J. Appl. Phys. 2012, 111, 07C710. [CrossRef]
25. Fujiwara, K.; Oogane, M.; Kanno, A.; Imada, M.; Jono, J.; Terauchi, T.L.; Okuno, T.; Aritomi, Y.; Morikawa, M.; Tsuchida, M.; et al.
Magnetocardiography and magnetoencephalography measurements at room temperature using tunnel magnetoresistance. Appl.
Phys. Express 2018, 11, 023001. [CrossRef]
26. Kato, D.; Oogane, M.; Fujiwara, K.; Nishikawa, T.; Naganuma, H.; Ando, Y. Fabrication of Magnetic Tunnel Junctions with
Amorphus CoFeSiB Ferromagnetic Electrode for Magnetic Field Sensor Devices. Appl. Phys. Exp. 2013, 6, 103004. [CrossRef]
27. Taulu, S.; Kajola, M. Presentation of electromagnetic multichannel data: The signal space separation method. J. Appl. Phys. 2005,
97, 124905. [CrossRef]
28. Stroink, G.; Meeder, R.J.J.; Elliott, P.; Lant, J.; Gardner, M.J. Arrhythmia Vulnerability Assessment Using Magnetic Field Maps and
Body Surface Potential Maps. PACE 1999, 22, 1718–1728. [CrossRef]
29. Welland, M.E.; Koch, R.H. Spatial location of electron trapping defects on silicon by scanning tunneling microscopy. Appl. Phys.
Lett. 1986, 48, 724. [CrossRef]
30. Egelhoff, W.F., Jr.; Pong, P.W.T.; Unguris, J.; McMichael, R.D.; Nowak, E.R.; Edelstein, A.S.; Burnette, J.E.; Fischer, G.A. Critical
challenges for picoTesla magnetic-tunnel-junction sensors. Sens. Actuators A 2009, 115, 217–225. [CrossRef]
31. Aharoni, A. Demagnetizing factors for rectangular ferromagnetic prisms. J. Appl. Phys. 1998, 83, 3432. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.