NanoSQUIDs: Basics & recent advances

M. J. Mart́ınez-Pérez1 and D. Koelle1

1Physikalisches Institut – Experimentalphysik II and Center for Quantum Science in LISA+,Universität Tübingen, Auf der Morgenstelle 14, D-72076 Tübingen, Germany

(Dated: May 21, 2018)

Superconducting Quantum Interference Devices (SQUIDs) are one of the most popular devices insuperconducting electronics. They combine the Josephson effect with the quantization of magneticflux in superconductors. This gives rise to one of the most beautiful manifestations of macroscopicquantum coherence in the solid state. In addition, SQUIDs are extremely sensitive sensors allow-ing to transduce magnetic flux into measurable electric signals. As a consequence, any physicalobservable that can be converted into magnetic flux, e.g., current, magnetization, magnetic field orposition, becomes easily accessible to SQUID sensors. In the late 1980’s it became clear that down-sizing the dimensions of SQUIDs to the nanometric scale would encompass an enormous increaseof their sensitivity to localized tiny magnetic signals. Indeed, nanoSQUIDs opened the way to theinvestigation of, e.g., individual magnetic nanoparticles or surface magnetic states with unprece-dented sensitivities. The purpose of this review is to present a detailed survey of microscopic andnanoscopic SQUID sensors. We will start by discussing the principle of operation of SQUIDs, plac-ing the emphasis on their application as ultrasensitive detectors for small localized magnetic signals.We will continue by reviewing a number of existing devices based on different kinds of Josephsonjunctions and materials, focusing on their advantages and drawbacks. The last sections are leftfor applications of nanoSQUIDs in the fields of scanning SQUID microscopy and magnetic particlecharacterization, putting special stress on the investigation of individual magnetic nanoparticles.

CONTENTS

I. Introduction 1

II. SQUIDs: some basic considerations 2A. Resistively and capacitively shunted junction

model 2B. dc SQUID basics 3C. SQUID readout 4

1. Flux-locked loop 42. Voltage readout 43. Critical current readout & threshold

detection 54. Dispersive read out 5

III. nanoSQUIDs: design, fabrication &performance 5A. nanoSQUIDs: design considerations 6B. nanoSQUIDs based on metallic

superconductors 71. Sandwich-type SIS junctions 72. Sandwich-type SNS junctions 83. Constriction junctions 94. Proximized structures 11

C. NanoSQUIDs based on cupratesuperconductors 12

IV. nanoSQUIDs for magnetic particle detection 13A. Nanoparticle positioning 13

1. In-situ nanoparticle growth 142. Scanning probe-based techniques 143. Techniques based on chemical

functionalization 15B. Magnetization measurements 15

C. Susceptibility measurements 16

V. nanoSQUIDs for scanning SQUID microscopy 18A. SQUID microscopes using devices on planar

substrates 18B. SQUID-on-tip (SOT) microscope 19

VI. Summary and outlook 20

Acknowledgments 20

References 20

I. INTRODUCTION

The superconducting quantum interference device(SQUID) consists of a superconducting ring intersectedby one (rf SQUID) or two (dc SQUID) Josephson junc-tions. SQUIDs constitute, still at present, the most sen-sitive sensors for magnetic flux in the solid state1,2. Formore than 50 years, a plethora of devices exploiting thisproperty have been envisioned, fabricated and used inmany fields of applications3. These devices include volt-meters, current amplifiers, metrology standards, motionsensors and magnetometers. One of the key applicationsof SQUIDs is in magnetometry. Here, a superconduct-ing input circuit (flux transformer) picks up the mag-netic flux density B, captured by superconducting pick-up loops of some mm2 or cm2 area, and the induced cur-rent is then (typically inductively) coupled to a SQUID.The figure of merit of SQUID magnetometers is the fieldresolution

√SB =

√SΦ/Aeff , which can reach values

down to about 1fT/√

Hz. Here, SΦ is the spectral den-

arX

iv:1

609.

0618

2v4

[co

nd-m

at.s

upr-

con]

18

May

201

8

2

sity of flux noise of the SQUID and Aeff is the effectivearea of the magnetometer.

To ensure good coupling from an input circuit to aSQUID, typically thin film multiturn input coils are in-tegrated on top of a washer-type SQUID loop. Typicalthin film washer SQUIDs have lateral outer dimensionsof several 100µm, the inner hole size is several tens ofµm and the lateral size of the Josephson junctions is sev-eral µm. Such devices are fabricated by conventional thinfilm technology, including micropattering by photolithog-raphy. With the development of a mature junction tech-nology, based on sandwich-type Nb/Al-AlOx/Nb junc-tions in the 1980s4, Nb based dc SQUIDs became themost commonly used type of devices for various appli-cations. At the same time, first attempts were startedto further miniaturize the lateral dimensions of SQUIDs,including the Josephson junctions5. This was made pos-sible by advances in nanolithography6 and was motivatedby the development of the theory for thermal noise inthe dc SQUID7, which showed that the energy resolutionε = SΦ/(2L) of dc SQUIDs can be improved by reducingthe SQUID loop inductance L and junction capacitanceC, to eventually reach and explore quantum limited res-olution of such devices8. These developments have trig-gered the realization of miniaturized dc SQUIDs for theinvestigating of small magnetic particles and for imag-ing of magnetic field distributions by scanning SQUIDmicroscopy to combine high sensitivity to magnetic fluxwith high spatial resolution. In 1984, Ketchen et al.9

presented the first SQUID microsusceptometer devotedto detect the tiny signal produced by micron-sized mag-netic objects, and in 1983 Rogers and Bermon developedthe first system to produce 2-dimensional scans of mag-netic flux structures in superconductors10. Both devel-opments were pushed further in the 1990s. Wernsdorferet al.11,12 used micron-sized SQUIDs to perform experi-ments on the magnetization reversal of nanometric par-ticles, which were placed directly on top of the SQUIDs.At the same time, scanning SQUID microscopes withminiaturized SQUIDs and/or pickup loop structures havebeen developed, at that time with focus on studies onthe pairing symmetry in the high transition temperature(high-Tc) cuprate superconductors

13. Since then mucheffort has been dedicated to the further miniaturizationof SQUID devices and to the optimization of their noisecharacteristics14.

Studies on the properties of small spin systems, suchas magnetic nanoparticles (MNPs) and single moleculemagnets (SMMs), have fueled the development of newmagnetic sensors for single particle detection and imag-ing with improved performance. Many of the recent ad-vances in this field include the development of magneto-optical techniques based on nitrogen vacancy centers indiamond15,16 or the use of carbon nanotubes (CNTs) asspin detectors17. Alternatively, miniature magnetome-ters, based on either microHall bars18 or micro- andnanoSQUIDs, provide direct measurement of the straymagnetic fields generated by the particle under study,

making the interpretation of the results much more di-rect and simple. While their sensitivity deterioratesrapidly when Hall sensors are reduced to the submicronsize, miniaturized SQUID-based sensors can theoreticallyreach quantum limited resolution.

In this review, we give an overview on some basics ofnanoSQUIDs19 and recent advances in the field. After abrief description of some SQUID basics in section II, wewill review in section III important design considerationsfor optimizing nanoSQUID performance and the state ofthe art in fabrication and performance of nanoSQUIDsbased on low-Tc and high-Tc superconductors, with em-phasis on the various types of Josephson junctions used.Subsequently, we will review important applications ofnanoSQUIDs, divided into two sections: Section IV givesan overview on applications of nanoSQUIDs for magneticparticle detection, and section V addresses nanoSQUIDsfor scanning SQUID microscopy. We will conclude witha short section VI, which gives a summary and outlook.

II. SQUIDS: SOME BASIC CONSIDERATIONS

The working principle of a SQUID is based on twofundamental phenomena in superconductors, the flux-oid quantization and the Josephson effect. The fluxoidquantization arises from the quantum nature of super-conductivity, as the macroscopic wave function describ-ing the whole ensemble of Cooper pairs shall not in-terfere destructively. This leads to the quantization ofthe magnetic flux Φ threading a superconducting loop20,in units of the magnetic flux quantum Φ0 = h/2e ≈2.07× 10−15 Vs.

The Josephson effect21,22 results from the overlap ofthe macroscopic wave functions between two supercon-ducting electrodes at a weak link forming the Josephsonjunction (JJ). The supercurrent Is through the weak linkand the voltage drop U across it satisfy the Josephsonrelations

Is(t) = I0 sin δ(t) (a) U(t) =Φ02π

δ̇ (b) , (1)

with the gauge-invariant phase difference δ between themacroscopic wave functions of both superconductors andthe maximum attainable supercurrent I0; the dot refersto the time derivative. The simple sinusoidal current-phase relation (CPR), Eq. (1(a)), is found for many kindsof JJs. However, some JJ types exhibit a non-sinusoidalCPR, which can even be multivalued23.

A. Resistively and capacitively shunted junctionmodel

A very useful approach to describe the phase dynamicsof a JJ is the resistively and capacitively shunted junction(RCSJ) model24–26. Within this model, the current flow

3

(a)

(b)

Fig. 1

R I0 IN U

I

C

Is Iqp Id

0 1 2

UJ

δ/2π

2EJ i=

1

1/2

0

FIG. 1. RCSJ model: (a) Equivalent circuit. (b) Tilted wash-board potential for different normalized bias currents i.

is split into three parallel channels [Fig. 1(a)]: (i) a su-percurrent Is [Eq. (1(a))], (ii) a dissipative quasiparticlecurrent Iqp = U/R across an ohmic resistor R and (iii) adisplacement current Id = C ∂U/∂t across the junctioncapacitance C. A finite temperature T is included as athermal current noise source IN from the resistor. WithKirchhoff’s law and Eq. (1(b)), one obtains the equationof motion for the phase difference δ

I + IN = I0 sin δ +Φ0

2πRδ̇ +

Φ0C

2πδ̈ . (2)

This is equivalent to the equation of motion of a point-likeparticle moving in a tilted washboard potential [Fig. 1(b)]

UJ = EJ(1− cos δ)− (i+ iN)δ , (3)

with normalized currents i = I/I0, iN = IN/I0 and theJosephson coupling energy EJ = I0Φ0/(2π). In this anal-ogy, the mass, friction coefficient, driving force (tiltingthe potential) and velocity correspond to C, 1/R, I andU , respectively. Hysteresis in the current voltage char-acteristics (IVC), i.e. bias current I vs time averagedvoltage V = 〈U〉, can be understood as a consequence ofthe particle’s inertia: the dissipative state 〈δ̇〉 ∝ V 6= 0 isachieved once the metastable minima of the washboardpotential disappear at I ≥ I0. If I is decreased fromI > I0, the particle becomes retrapped at Ir < I0, lead-ing to a hysteretic IVC. This behavior can be quantifiedby the Stewart-McCumber parameter

βC ≡2π

Φ0I0R

2C . (4)

In order to obtain a non-hysteretic IVC, βC must be keptbelow ∼ 1. This can be e.g. achieved by means of anadditional shunt resistor, parallel to the JJ.

B. dc SQUID basics

The dc SQUID27 is a superconducting loop (with in-ductance L) intersected by two JJs [Fig. 2(a)]. With anexternally applied magnetic flux Φ through the loop, thefluxoid quantization links the phase differences δ1 and δ2of the two JJs to the total flux in the SQUID ΦT = Φ+LJvia

δ1 − δ2 + 2πn =2π

Φ0(Φ + LJ) . (5)

Here, J is the current circulating in the SQUID loop andn is an integer28. Defining the screening parameter as

βL ≡2LI0Φ0

, (6)

one finds in the limit βL � 1 a negligible contributionof LJ to ΦT in Eq. (5), and by assuming for simplic-ity identical values for I0 in the two JJs, the maximumsupercurrent (critical current) Ic of the SQUID can beeasily obtained as

Ic = 2I0

∣∣∣∣cos(πΦΦ0)∣∣∣∣ . (7)

The pronounced Ic(Φ) dependence [Fig. 2(b) for βL � 1]can be used to probe tiny changes in applied magneticflux. No analytical expression for Ic(Φ) can be obtainedwhen a finite βL and hence a finite L is included, unlessrestrictions are imposed to some of the important SQUIDparameters26,29. An increasing βL leads to a monotonicdecrease of the critical current modulation ∆Ic/2I0 with

(b)

0 0.5 1.0 1.5 2.00

1

25

0.01

1

βL

Ic/I0

Φ/Φ0

(a)

δ1 Φ

J

L

I

δ2

(c)

10-2 10-1 100 10110-2

10-1

100

∆Ic/2I0

βL

∝1/βL

Fig. 2

FIG. 2. The dc SQUID: (a) Schematic view. (b) Criticalcurrent vs applied magnetic flux for different βL and (c) Icmodulation vs βL, both calculated for T = 0 and identicalJJs.

4

increasing βL [Fig. 2(b,c)]. This effect allows to estimateL from the measured Ic(Φ).

We note that the inductance L = Lg + Lk has twocontributions28: The geometric inductance Lg relatesthe induced flux LgJ to the current J circulating in theSQUID loop. The kinetic inductance Lk is due to thekinetic energy of J and can often be neglected; however,it becomes significant when the width and/or thicknessof the SQUID ring are comparable to or smaller than theLondon penetration depth λL.

For most applications, the dc SQUID is operated inthe dissipative state as a flux-to-voltage transducer. Inthis case, the SQUID is current biased slightly above Ic,leading to a Φ0-periodic modulation of V (Φ), which isoften sinusoidal. This mode of operation requires non-hysteretic IVCs, i.e., βC

5

above, the IVCs should be non-hysteretic in this case.As the transfer function VΦ is typically small (several10−100µV/Φ0), the voltage noise at the output can eas-ily be dominated by room-temperature amplifier noise.To circumvent this problem, one can use a flux modula-tion scheme37. Here, the SQUID is flux-modulated by anac signal (amplitude Φ0/4, frequency fm ∼ 100 kHz), andthe resulting ac voltage across the SQUID is amplifiedwith a (cold) step-up transformer to increase the SQUIDsignal and noise. The modulated SQUID response isfurther amplified at room temperature and lock-in de-tected. Suitable electronics achieve a bandwidth of up to100 kHz.

In a different approach, one can increase VΦ by addi-tional positive feedback (APF), which distorts the V (Φ)characteristics and increases VΦ at the positive slope.This enables simple direct readout of the SQUID signal37.Alternatively, a low-noise SQUID or serial SQUID array(SSA) amplifier can be used to amplify the SQUID volt-age at low T in a two-stage readout configuration.

3. Critical current readout & threshold detection

For SQUIDs with hysteretic IVCs one can exploit theIc(Φ) modulation directly. In this case one ramps thebias current until the SQUID switches to the dissipativestate, producing a voltage drop. At this point the currentis switched off, and Ic is calculated from the durationof the ramp38. This technique can also be used with aFLL scheme38–40. Sensitivity is limited by the accuracyin determining Ic, which is described by the escape of aparticle from a potential minimum. Such a process canbe thermally activated or quantum driven and is stronglyinfluenced by electronic noise. Hence, a large number ofswitching events is needed to obtain sufficient statistics.

To minimize Joule heating, the SQUID can be oper-ated as a threshold sensor. Here, the SQUID is current-biased very close to the switching point. If the magneticflux threading the loop changes abruptly, the SQUID istriggered to the dissipative state and a voltage drop willbe measured38.

Both techniques were applied to magnetization rever-sal measurements on MNPs in sweeping magnetic fieldsH38. For measurements up to large H, applied alongany direction, the measurement procedure is divided intothree steps. First, H is applied to saturate the particle’smagnetization along any direction. Second, H is sweptalong the opposite direction to a value Htest and back tozero. To check whether this reversed the particle’s mag-netization, an in-plane field sweep is done as a third step.If the particle’s magnetization reversal is (not) detectedin the third step one can conclude that Htest was above(below) the switching field Hsw. These steps can be re-peated several times to determine Hsw precisely. Notethat the second step can be performed above Tc of theSQUID. Rather than tracing out full M(H) loops, thistechnique can be used to trace out the dependence of Hsw

on the field direction and temperature41.

4. Dispersive read out

So far, we discussed SQUID operation in the volt-age state or close to it. Such schemes entail dissipa-tion of Joule power that might affect the state of themagnetic system under study. An elegant way to cir-cumvent this problem is the operation of the SQUIDas a flux-dependent resonator; this has also the ad-vantage of increasing enormously the bandwidth up to∼ 100 MHz42,43. The SQUID is always in the supercon-ducting state and acts as a flux-dependent inductanceconnected in parallel to a capacitor. The resonance fre-quency of the circuit depends on the total flux threadingthe SQUID loop. This can be read out by conventionalmicrowave reflectometry giving a direct flux-to-reflectedphase conversion. The devices are operated in the linearregime, i.e., using low-power driving signals. To deter-mine the spectral density of flux noise, the overall volt-age noise of the circuit is estimated and scaled with thetransduction factor dV/dΦ. The noise performance canbe boosted considerably by taking advantage of the CPRnon-linearity, i.e., operating the nanoSQUID as a para-metric amplifier. For this purpose, the driving power isincreased so that the resonance peak is distorted, givinga much sharper dependence of the reflected phase on Φ.

III. NANOSQUIDS: DESIGN, FABRICATION &PERFORMANCE

NanoSQUIDs are developed for detecting small spinsystems, such as MNPs or SMMs, or for high-resolutionimaging of magnetic field structures by SQUID mi-croscopy. For such applications, the figure of merit isthe spin sensitivity, which can be boosted down to thelevel of a single electron spin. The use of strongly minia-turized SQUID loops and JJs is based on the followingideas:

• Stronlgy localized magnetic field sources(e.g. MNPs) are placed in close vicinity to theSQUID, instead of using pickup coils [Fig. 4(a)]which degrade the overall coupling. A singleSQUID loop [Fig. 4(b)] can be used to detect themagnetic moment µ of a MNP, or gradiometricconfigurations [Fig. 4(c,d)] enable measurementsof the magnetic ac susceptibility χac.

• The coupling of the stray field from local fieldsources to the nearby SQUID can be improved byreducing the cross section (width and thickness) ofthe superconducting thin film forming the SQUIDloop (see section III A).

• The sensitivity of the SQUID to magnetic flux(magnetic flux noise in the thermal white noise

6

(a)

Bext

nanoSQUID magnetometer

µ

Bac

series-gradiometer

µ Bac

parallel-gradiometer

µ

SQUID magnetometer

M

pickup coils

(b)

(c) (d)

µ

Fig. 4

FIG. 4. Layouts of various SQUID sensors. (a) SQUIDmagnetometer based on gradiometric pickup coils coupled in-ductively (via mutual inductance M) to a SQUID. (b)-(d)NanoSQUIDs without intermediate pick-up coils; the strayfield created by a MNP with magnetic moment µ is directlysensed by the SQUID loop. Magnetization measurements canbe performed by applying an external magnetic field Bext inthe nanoloop plane (b). The frequency-dependent magneticac susceptibility χac can be sensed by using series (c) or par-allel (d) planar gradiometers; a homogeneous ac excitationmagnetic field Bac is applied perpendicular to the gradiome-ter’s plane through on-chip excitation coils.

limit) can be improved by reducing the loop in-ductance, i.e. by shrinking the lateral size of theSQUID loop (see section III A).

• For magnetization reversal measurements onMNPs, an external field Bext is applied ideally ex-actly in the plane of the SQUID loop to switchthe MNP’s magnetization (see section IV B), albeitwithout coupling flux directly to the loop. By re-ducing the dimensions of the JJs and the loop, thenanoSQUID can be made less sensitive to Bext forsmall misalignment of Bext.

• Reducing the loop size together with the SQUID-to-sample distance can significantly boost the spa-tial resolution for scanning SQUID microscopy ap-plications (see section V).

A. nanoSQUIDs: design considerations

The ability of a nanoSQUID to resolve tiny signalsfrom the magnetic moments of small spin systems de-pends (i) on the intrinsic flux noise SΦ of the SQUIDand (ii) on the amount of flux Φ which a particle withmagnetic moment µ couples to the SQUID loop. The lat-ter can be quantified by the coupling factor φµ ≡ Φ/µ,with µ ≡ |µ|. As a result, one can define the spinsensitivity

√Sµ =

√SΦ/φµ, with units µB/

√Hz; µB is

the Bohr magneton.√Sµ expresses the minimum mag-

netic moment that can be resolved per unit bandwidth.

Hence, optimizing nanoSQUID performance requires tominimize SΦ while maximizing φµ.

As mentioned in section II B, SΦ has typically a low-frequency 1/f -like contribution and a thermal whitenoise part SΦ,w. The 1/f contribution is hard to op-timize by design; however, SΦ,w depends on geometricalparameters through the loop inductance L, but also onjunction parameters such as I0, R and C. The SΦ(L)dependence (Eq.(8)) implies that SΦ can be improved bydecreasing L via the loop dimensions, while consideringthe constraints on βC and βL, which will affect the choiceof junction parameters. Such an optimization procedurecan be tested experimentally by performing flux noisemeasurements of the SQUIDs.

The optimization of the coupling factor φµ = Φ/µ ismore difficult. It is defined as the magnetic flux Φ cou-pled to the SQUID loop by the magnetic dipole field ofa point-like particle, divided by its magnetic moment µ.The magnitude of φµ depends on SQUID geometry, par-ticle position rµ (relative to the SQUID) and orientationêµ = µ/µ of its magnetic moment. This quantity is notdirectly accessible by experiments, and one has to relyon estimates, analytic approximations or numerical cal-culations for determining φµ and optimizing it.

To the best of our knowledge, Ketchen et al.44 werethe first to give an estimate of φµ. For a magnetic dipoleat the center of an infinitely thin loop with radius a, withêµ along the loop normal

φµ =µ02a

= (re/a)·(Φ0/µB) ≈ (2.8µm/a)·(nΦ0/µB) (9)

was found.45 The r.h.s. of Eq. (9) is obtained with the

definition of the classical electron radius re =µ0e

2

4πme, Φ0 =

h2e and µB =

eh4πme

, which yields µBΦ0 =2reµ0

.

The coupling improves if the particle is moved closeto the loop’s banks46. However, a quantitative estimateof φµ is more difficult in this near-field regime

47, as thecross-section of the SQUID banks and the flux focusingeffect caused by the superconductor must be taken intoaccount. The calculation of φµ requires calculating themagnetic field distribution at the position of the SQUID,originating from a magnetic moment µ at position rµ,and from this the magnetic flux coupled to the SQUID.This problem can be simplified by exploiting the fact thatsources and fields can be interchanged, i.e., one evaluatesthe magnetic field BJ(rµ), created by a circulating su-percurrent J through the SQUID loop, at the positionrµ of the magnetic dipole. With the normalized quantitybJ = BJ/J , which does not depend on J , one finds

46,48

φµ(rµ, êµ) = êµ · bJ(rµ) . (10)

This allows to calculate φµ for any position and orien-tation of the magnetic dipole in 3D space once bJ isknown.49

The normalized field bJ has to be determined fromthe spatial distribution of the supercurrent density jscirculating in the SQUID loop, which depends only on

7

Fig. 5

FIG. 5. Calculated coupling factor φµ vs position of a mag-netic dipole pointing in x-direction on top of Nb nanoSQUIDs.Main graphs show contour plots φµ(x, z) for (a) a magne-tometer and (b) a gradiometer. Nb structures are indicatedby black rectangles; dashed lines indicate position of lines-cans φµ(x) [above (a)] and φµ(z) [right graphs]. Insets showscanning electron microscopy (SEM) images. Reprinted withpermission from Nagel et al.50. Copyright [2011], AIP Pub-lishing LLC.

the SQUID geometry and on λL. This has been donefor various types of nanoSQUIDs by numerically solv-ing the London equations48,50–55. Numerical simula-tions of φµ reveal that the coupling can be increased inthe near-field regime if the magnetic dipole is placed asclose as possible on top of a constriction in the SQUIDloop, which is as thin and narrow as possible55. Typi-cal φµ = 10 − 20 nΦ0/µB have been obtained for mag-netic dipoles at 10 nm distance from a constriction (∼100 − 200 nm wide and thick) in YBa2Cu3O7 (YBCO)nanoSQUIDs.56 Simulation results for two types of NbnanoSQUIDs [Fig. 5] show that the dipole has to ap-proach the SQUID surface closely to reach values abovea few nΦ0/µB (see φµ(z) linescans in the right graphsin Fig. 5). The φµ(x) linescans (top graph in Fig. 5)show that the coupling is maximum right above the loopstructures50.

Measurements on spatially extended magnetic sys-tems, such as a Ni nanotube51 or a Fe nanowire53, werefound to be consistent with the numerical approach de-scribed above. This was done by comparing the measuredflux coupled to nanoSQUIDs from fully saturated tubesor wires with the calculated flux signals, obtained by in-tegrating φµ over the finite volume of the sample. Firstmeasurements on the SQUID response as a function ofthe position of a magnetic sample have been reportedearlier. In those experiments, small SQUID sensors were

coupled to a ferromagnetic Fe tip, which was scannedover the sensor’s surface while recording the SQUID out-put in open-loop configuration57.

The optimization of the spin sensitivity in the thermalwhite noise limit requires the knowledge of the depen-dence of φµ and SΦ,w on SQUID geometry, as this affectsboth the SQUID inductance and the coupling. A de-tailed investigation of this problem was done for YBCOnanoSQUIDs55 (see section III C). This study shows thatit is essential to consider the increase in kinetic induc-tance Lk when the thickness and width of the loop isreduced to a length scale comparable to or even smallerthan λL. Hence, to improve the Sµ one has to find a com-promise between improved coupling and deterioration offlux noise (via an increased Lk) upon shrinking the crosssection of the SQUID loop.

B. nanoSQUIDs based on metallic superconductors

1. Sandwich-type SIS junctions

The SIS junction technology (S: superconductor, I:thin insulating barrier), typically producing JJs in aNb/Al-AlOx/Nb trilayer geometry, is the most com-monly used approach to fabricate conventional SQUID-based devices. This technology is highly developed andreproducible, yielding high-quality JJs with controllablecritical current densities jc from ∼ 0.1 up to a fewkA/cm2 at 4.2 K. However, a major disadvantage is thelow jc, which results in too small values for the criticalcurrent if submicron JJs are used. As a consequence,even if the SQUID loops are miniaturized, the operationof micron-sized JJs in large magnetic fields is only pos-sible with careful alignment of the field perpendicular tothe junction plane, as an in-plane field in the 1-10 mTrange can easily suppress the critical current due to theFraunhofer-like modulation of Ic(B). Frequently usedwindow-type JJs come with a large parasitic capacitancedue to the large area of surrounding superconducting lay-ers. A commonly used approach is therefore to use nor-mal metal layers to shunt these junctions, for loweringβC to yield non-hysteretic IVCs, albeit at the cost ofalso lowering the characteristic voltage Vc = I0R. Theabsence of hysteresis offers the advantage to operate theSQUID as a flux-to-voltage converter, using conventionalreadout techniques.

As a key advantage, the SIS technology offers a welldeveloped multilayer process, allowing for the realiza-tion of more complex designs, as compared to a sin-gle layer technology. This allows for the fabrication ofsuperconducting on-chip input circuits such as couplingtransformers, susceptometers or advanced gradiometers.This approach has been taken very successfully to real-ize miniaturized structures for applications in magneticparticle measurements and scanning SQUID microscopy,although those did not really involve SQUIDs with (lat-eral outer) dimensions in the submicrometer range.

8

100 µm

field coils

Fig. 6

FIG. 6. SEM im-age of a SQUID micro-susceptometer which ananoloop patterned inthe pickup coil (inset).Images courtesy of J.Sesé.

The first SQUID device designed to measure mag-netic signals from MNPs was based on micrometricNb/NbOx/Pb edge junctions, which were connected inparallel to two oppositely wound loops to form a micro-susceptometer9. The white flux noise at 4.2 K was0.84µΦ0/

√Hz. This susceptometer was operated in a

dilution refrigerator, and the output signal was mea-sured in open-loop configuration and amplified by an rfSQUID preamplifier. Magnetic susceptibility measure-ments performed with this system will be reviewed in sec-tion IV C. Very similar devices based on Nb/Al-AlOx/Nb

JJs with√SΦ = 0.8µΦ0/

√Hz at 4 K and 0.25µΦ0/

√Hz

below 0.5 K were adapted to the use in scanning SQUIDmicroscopes58,59; see section V.

Broad-band SQUID microsusceptometers have beenrealized by locally modifying SQUID current sen-sors based on Nb/Al-AlOx/Nb JJ technology. Thosesensors60 come in two types: (i) high-input inductance(∼ 1µH) sensors incorporate an intermediate trans-former loop with gradiometric design; (ii) low-input in-ductance (2 nH) devices without intermediate loop; herethe input signal is directly coupled to the SQUID viafour single-turn gradiometric coils connected in parallel.These SQUIDs are non-hysteretic down to sub-K tem-peratures with

√SΦ,w = 800 nΦ0/

√Hz at T = 4.2 K.

Modification of these sensors was done by FIB millingand FIB-induced deposition (FIBID) of superconduct-ing material with W(CO)6 as precursor gas

35,61. Thisallowed converting the intermediate transformer loopinto a susceptometer inductively coupled to the SQUID[Fig. 4(a)]. By modifying the gradiometric microSQUIDitself it is possible to directly couple an MNP to theSQUID loop34 [Fig. 4(d)]. Later, SQUID-based micro-susceptometers with improved reflection symmetry wereproduced62,63. The sensitivity was boosted by defining ananoloop (450 nm inner diameter, 250 nm linewidth) byFIB milling in one of the pickup coils [Fig. 6]. These sen-sors offer an extremely wide bandwidth (1 mHz - 1 MHz)and can be operated at T = 0.013 − 5 K for the inves-tigation of microscopic crystals of SMMs and magneticproteins; such measurements will be reviewed in sectionIV C.

Submicrometric Nb/AlOx/Nb JJs in a cross-type de-sign were recently used for fabricating miniaturizedSQUIDs64. The key advantage of cross-type JJs overconventional window-type JJs is the elimination of theparasitic capacitance surrounding the JJ, which becomesincreasingly important upon reducing the JJ size. AtT = 4.2 K, 0.8 × 0.8µm2 JJs show non-hysteretic IVCs,if they are shunted with a AuPd layer. Sensors arealso produced with an integrated Nb modulation coil.Square-shaped washer SQUIDs with minimum inner sizeof 0.5µm have an inductance of a few pH. SQUIDs oper-ated in liquid He and read out with a low-noise SQUIDpreamplifier yield

√SΦ,w = 66 nΦ0/

√Hz65.

2. Sandwich-type SNS junctions

SNS junctions (N: normal conductor) offer the advan-

tage of large critical current densities >∼105 A/cm2

at4.2 K and non-hysteretic IVCs, albeit at the cost of some-what reduced I0R values. Hence, this type of JJs is verywell suited for fabricating nanoSQUIDs with junction sizein the deep submicron range.

In a Nb/HfTi/Nb trilayer process, originally devel-oped for Josephson arbitrary waveform synthesizers66,JJs with 200× 200 nm2 area or even below are obtainedby e-beam lithography and chemical-mechanical polish-ing, producing nanoSQUIDs50,54 with 24 nm thick HfTibarriers; the latter can be varied to modify jc. As for theSIS JJ technology, the fabrication process offers muchflexibility for realizing complex designs. Both series-and parallel-gradiometers and single SQUID loops wererealized50,54,67. Devices were patterned in a washer- ormicrostrip-type geometry, with the loop plane parallelor perpendicular to the junction’s (substrate) plane, re-spectively. A key advantage of the microstrip-type ge-ometry [Fig. 7] is the possibility to realize very smallloop areas, defined by the thickness of the insulating in-terlayer between the top an bottom Nb lines times thelateral separation of the two JJs. This results in verysmall SQUID inductances, typically a few pH. Moreover,a magnetic field applied in the plane of the SQUID loop

9

Fig. 7

I z

x y B

Nb

Imod Φmod

HfTi

600 nm

HfTi Nb

loop size: 600 × 220 nm2

FIG. 7. Layout of Nb/HfTi/Nb nanoSQUID in microstripgeometry. Arrows indicate flow of bias current I, modulationcurrent Imod and direction of external field B. Inset showsSEM image with JJs (200 × 200 nm2) indicated by dashedsquares. SEM image courtesy of B. Müller.

can be perpendicular to the JJ (and substrate) plane;in this way the field-induced suppression of Ic can beavoided. It has been shown that magnetic fields up to0.5 T can be applied while degrading only marginally theperformance54. On-chip flux biasing is easily possible foroperation in FLL. White flux noise ∼ 110 nΦ0/

√Hz has

been obtained. Based on numerical solutions of the Lon-don equations for φµ, this yields a spin sensitivity of just

∼ 10µB/√

Hz for a magnetic dipole 10 nm away from theSQUID loop. Magnetization measurements on magneticnanotubes have been performed successfully and will besummarized in section IV B.

By combining three mutually orthogonal nanoSQUIDloops, a 3-axis vector magnetometer has been realizedvery recently68. Here, the idea is to distinguish thethree components of the vector magnetic moment µ ofa MNP placed at a specific position, and subjected to anapplied magnetic field along z-direction for magnetiza-tion reversal measurements. The layout of the device is

Fig. 8

500 nm

SQz

SQy SQx

FIG. 8. SEM image of a 3-axis vector magnetometer, con-sisting of two orthogonal nanoSQUIDs (SQx, SQy) and anorthogonal gradiometric nanoSQUID (SQz). Black dottedsquares indicate positions of Josephson junctions.

Fig. 9 FIG. 9. SEM image of a 3-dimensional nanoSQUID fabricatedusing FIB sculpting and all Nb technology. The flux capturearea of the nanosensor is 1× 0.2µm2, and the two Josephsontunnel junctions have an area of about 0.3×0.3µm2. The insetis a sketch of the device, showing the current paths throughthe device. Reprinted with permission from Granata et al.69.Copyright [2013], AIP Publishing LLC.

shown in Fig. 8. Two microstrip-type Nb nanoSQUIDsSQx and SQy, as described above, with perpendicularloops are sensitive to fields in x- and y-direction, respec-tively. A third SQUID, SQz has a gradiometric layout,in order to strongly reduce its sensitivity to the appliedhomogeneous magnetic field. Simultaneous operation ofall three nanoSQUIDs in such devices in FLL has beendemonstrated at 4.2 K in fields up to 50 mT, with a flux

noise S1/2Φ,w

10

Fig. 8

(b) µ Bext

(a)

FIG. 10. cJJ-basednanoSQUIDs: (a) schematicview with a MNP (magneticmoment µ) close to one con-striction where coupling ismaximum. (b) SEM imagesof Nb microSQUID with Niwire on top (left) and NbnanoSQUID (right), drawn toscale in left graph. Graph (b)Reproduced with permissionfrom38. All rights reservedc© IOP Publishing [2009].

of such constriction-type Josephson junctions (cJJs) areoften hysteretic, due to the heat dissipated above Ic.Short enough cJJs show a sinusoidal CPR; however, asignificant deviation occurs if the constriction length islarger than ξ, which can even lead to multivalued CPRs.Hence, optimization of SQUID performance based on anRCSJ analysis is difficult, and hysteretic IVCs preventsconventional SQUID operation with current bias. Still,non-hysteretic IVCs can be achieved by operation closeenough to Tc, were Ic is reduced, or by adding a metal-lic overlayer as a resistive shunt. Another drawback isthe large kinetic inductance Lkin of the constriction, thatcan dominate the total SQUID inductance L and preventimproving the flux noise by shrinking the loop size. Onthe other hand, cJJ-based nanoSQUIDs in a simple pla-nar configuration can be fabricated relatively easily fromthin film superconductors, e.g., Al, Nb or Pb, throughone-step electron-beam (e-beam) or FIB nanopattern-ing. Moreover, the use of nanometric-thick films and thesmallness of the constriction makes these SQUIDs quiteinsensitive to in-plane magnetic fields and yields largecoupling factors if MNPs are placed close to the constric-tion [Fig.10(a)]. The small size of cJJs is a key advantagefor fabricating nanoSQUIDs with high spin sensitivity.

First thin film Nb dc SQUIDs based on cJJs withlinewidths down to 30 nm, patterned by e-beam lithog-raphy, were reported in 19805. Despite their large L =150 pH, miniaturized SQUIDs, with loop size ∼ 1µm2,exhibited low flux noise ∼ 370 nΦ0/

√Hz at 4.2 K. Dur-

ing the 1990s, the use of cJJ nanoSQUIDs for the in-vestigation of small magnetic systems was pioneered byWernsdorfer et al.11,12,38. Figure 10(b) shows examplesof such devices, which were patterned by e-beam lithog-raphy from Nb and Al films71. Typical geometric pa-rameters were 1µm2 inner loop area, 200 nm minimumlinewidth and 30 nm film thickness. The size of the con-strictions (∼ 30 nm wide, ∼ 300 nm long) was signifi-cantly larger than ξ for Nb. This lead to a highly non-ideal CPR23,72 and hence non-ideal Ic(Φ) dependencewith strongly suppressed Ic modulation depth for Nb cJJSQUIDs. Furthermore, Lkin of the constrictions can bea few 100 pH, dominating the overall inductance of thedevices72. Impressively large magnetic fields could beapplied parallel to the nanoSQUID loops up to 0.5 T for

Al and 1 T for Nb. From the measured critical currentnoise, the flux noise was calculated as ∼ 40µΦ0/

√Hz for

Al and ∼ 100µΦ0/√

Hz for Nb71. Due to hysteretic IVCsthese nanoSQUIDs were operated in Ic readout mode oras threshold detectors (see section II C 3). These sensorsallowed the vastest realization of true magnetization mea-surements (section IV C) and were also implemented intoprobe tips to perform scanning SQUID microscopy71,73.

For similar Nb cJJ-based nanoSQUIDs (30 nm thick,∼ 200 nm inner loop size, cJJs down to 280 nm long and120 nm wide) switching current distributions were mea-sured from 4.2 down to 2.8 K74. A detailed analysis of thenoise performance for Ic readout revealed a flux sensitiv-ity of a few mΦ0, which was shown to arise from ther-mally induced Ic fluctuations in the nanobridges. Morerecently, hysteretic nanoSQUIDs made of Al-Nb-W lay-ers (2.5µm inner loop size; 40 nm wide, 180 nm long cJJs)could be operated with oscillating current-bias and lock-in read-out at T < 1.5 K75. In this configuration Ic isconsiderably reduced due to the inverse proximity effectof W on Nb.

Nanometric Nb SQUIDs (50 nm thick, down to 150 nminner hole size) were also fabricated by FIB milling toproduce cJJs (80 nm wide, 150 nm long)76. It was ob-served that Ga implantation depth can reach values of30 nm, suppressing the superconducting properties of Nb.At T = 4.2 K, devices with relatively small Ic < 25µAshowed nonhysteretic IVCs and could be operated ina conventional current-bias mode, yielding

√SΦ,w ∼

1.5µΦ0/√

Hz.

A possible way to approach the sinusoidal CPR ofideal point contacts is the use of variable thicknessnanobridges. Here, the thicker superconducting bankscan serve as phase reservoirs, while the variation in thesuperconducting order parameter should be confined tothe thin part of the bridges77. cJJ-based nanoSQUIDswere realized by local anodization of ultrathin (3−6.5 nm-thick) Nb films using a voltage-biased atomic force micro-scope (AFM) tip78. This technique produced constric-tions (30 − 100 nm wide and 200 − 1000 nm long) andvariable thickness nanobridges by further reducing theconstriction thickness down to few nm (within a ∼ 15 nmlong section). The latter exhibited ∆Ic/Ic twice as large

11

as the former, indicating an improved CPR.Vijay et al.79 produced Al nanoSQUIDs based on

cJJs (8 nm thick, 30 nm wide) with variable length (l =75 − 400 nm). The cJJs were either connected to super-conducting banks of the same thickness (‘’2D devices”)or to much thicker (80 nm) banks (‘’3D devices”). For 3Ddevices with l ≤ 150 nm ≈ 4ξ, the measured Ic(Φ) curvesindicate a CPR which is close to the one for an idealshort metallic weak link. Both 2D and 3D devices werefully operative up to in-plane magnetic fields of 60 mT80.Such nanoSQUIDs were operated with dispersive readout(see section II C 4) yielding impressive flux noise values

of 30 nΦ0/√

Hz for a 20 MHz bandwidth43.Variable thickness bridges have recently also been re-

alized by connecting suspended Al nanobridges (25 nmthick, 233 nm long, 60 nm wide) to Nb(30 nm)/Al(25 nm)bilayer banks to form a nanoSQUID (2.5µm-in-diameterloop)81. These devices have the advantage of using cJJsfrom a material (Al) with relatively large ξ, while main-taining relatively high Tc and critical magnetic field inthe superconducting banks forming the SQUID loop.

Thermal hysteresis in the IVCs of cJJs can be sup-pressed by covering the devices with a normal metal-lic layer, which provides resistive shunting and acts asa heat sink. cJJ-based nanoSQUIDs from 20 nm-thickNb films covered by 25 nm-thick Au have been patternedby e-beam lithography to realize 200 nm inner loop sizeand constriction widths in the range 70− 200 nm, yield-ing L ∼ 15 pH82 . The Au layer prevented hysteresisin the IVCs at temperatures above 1 K, allowing con-ventional SQUID readout in the voltage state, yielding√SΦ,w ∼ 5µΦ0/

√Hz at 4.2 K, increasing by about 15 %

when operating in a magnetic field of 2 mT83. Field op-eration up to few 100 mT was improved by reducing thehole size down to 100 nm and the largest linewidths downto 250 nm84. Preliminary experiments were performed onferritin nanoparticles attached to the cJJs85. However,the magnitude of the flux change observed in some cases(up to 440µΦ0) was larger than the expected one for aferritin NP located at optimum position (up to 100µΦ0).

Low-noise nanoSQUIDs from a Nb/amorphous W bi-layer (200 and 150 nm thick, respectively) have been pro-duced by FIB milling86. The SQUID loop (370 nm in-ner diameter) was intersected by two nanobridges (65 nmwide and 60 − 80 nm long) which showed non-hystereticIVCs at 5 − 9 K. Readout in the voltage state gave√SΦ,w = 200 nΦ0/

√Hz at 6.8 K. Recently, the same

group extended the operation temperatures down to <1 K by using superconducting Ti films, inversely prox-imized by Au layers to reduce Tc

87. These SQUIDs(with 40 nm wide and 120 nm long constrictions) exhib-ited no hysteresis within 60 mK < T < 600 mK and had√SΦ,w = 1.1µΦ0/

√Hz. These devices allowed the de-

tection of the magnetic signal produced by a 150 nm di-ameter FePt nanobead having 107 µB at 8 K in fields upto 10 mT88.

As mentioned earlier, cJJ-based nanoSQUIDs can beoperated in strong magnetic fields applied in the plane

of the loop, which is limited by the upper critical field ofthe superconductors. The use of very thin superconduct-ing layers can increase the effective critical field. Follow-ing this idea, 3 − 5 nm-thick cJJ Nb nanoSQUIDs werefabricated, supporting in-plane fields up to 10 T. Thesesensors proved to be well suited for measuring magnetiza-tion curves of microcrystals of Mn12 SMMs

89. However,their large kinetic inductances lead to large flux noise(∼ 100µΦ0/

√Hz). More promising is the use of materi-

als with larger upper critical fields, such as boron-dopeddiamond90. Micrometric SQUIDs based on 100 nm-wideconstrictions in 300 nm thick films were demonstrated tooperate up to impressive fields of 4 T applied along anydirection. These devices were, however, hysteretic due toheat dissipation. Flux sensitivity was determined fromthe critical current uncertainty giving 40µΦ0/

√Hz.

Finally we note that the smallest nanoSQUIDs real-ized so far, which also include cJJs, are the SQUIDs-on-tip (SOTs)91,92. These devices will be discussed in moredetail in section V.

4. Proximized structures

A normal metal in good contact between supercon-ducting electrodes acquires some of their properties dueto the proximity effect, inducing a mini-gap in the densityof states of the normal metal. Andreev pairs can prop-agate along relatively long distances at low T , carryinginformation on the macroscopic phase of the supercon-ductor. In the long (short)-junction regime, when theThouless energy of the metal is larger (smaller) than thesuperconducting energy gap, the junction properties willbe governed by the normal metal (superconductor).

The first dc SQUID built with long proximized JJswas based on a CNT intersecting an Al ring93. Agate-modulated supercurrent was demonstrated and flux-induced modulation of the critical current (few nA) wasobserved at mK temperatures. The goal was to exploitthe small cross section of the CNT (∼ 1 nm2) to pro-vide optimum coupling for molecular nanomagnets at-tached to it. An experimental proof-of-principle of sucha CNT-based magnetometer is, however, still missing. Amicrometric dc SQUID with graphene proximized junc-tions (50 nm long, 4µm wide) was also reported94. Flux-induced Ic modulation was observed, however, no noiseperformance of the device was reported.

Micrometric dc SQUIDs containing normal metalbridges as weak links have also been reported.Nb/Au/Nb and Al/Au/Al-based devices showed IVCswith pronounced hysteresis, due to heat dissipated in thenormal metal after switching95. SQUIDs with shorterCu nanowires (280 − 370 nm long, 60 − 150 nm wide,20 nm thick) enclosed in a V ring were non-hysteretic.NanoSQUIDs based on proximized InAs nanowires (∼90 nm diameter, 20 or 50 nm long) were also reported96

with JJs in the intermediate length regime [Fig.11].A different kind of interferometer consists of a su-

12

Fig. 11

1 µm

FIG. 11. SEM image of a SQUID sensor consisting of a prox-imized highly doped InAs nanowire enclosed within a V ring[after Spathis et al.96]. SEM image courtesy of F. Giazottoand S. D’Ambrosio.

perconducting loop interrupted by a normal metal is-land. A magnetic field applied to the loop varies thephase difference across the normal metal wire, allowingflux-modulation of the minigap. This behavior can beprobed by an electrode tunnel-coupled to the normalmetal island [Fig.12], providing a flux-modulated elec-tric response similar to conventional dc SQUIDs. Thisdevice received the name Superconducting Quantum In-terference Proximity Transistor (SQUIPT), for being themagnetic analog to the semiconductor field-effect transis-tor. SQUIPTs were pioneered by Giazotto et al.97 usingAl loops and Cu wires (∼ 1.5µm long, ∼ 240 nm wide).These magnetometers were further improved by reducingthe length of the normal metal island down to the short-junction limit, leading to a much larger mini-gap opening.By choosing proper dimensions of the normal metal is-land, such sensors do not exhibit any hysteresis down tomK temperatures98,99 and can be voltage or current bi-ased, providing impressive values of VΦ of a few mV/Φ0.

Fig. 12

200 nm

tunnel barrier

clean barriers

N

S

FIG. 12. Scheme of a SQUIPT. The inset shows a SEM im-age of the SQUIPT core; a normal metal probe is tunnel-connected to a proximized Cu island enclosed within an Alring. SEM image courtesy of F. Giazotto and S. D’Ambrosio.

SQUIPTs are in their early stage of development100, stillshowing a very narrow temperature range of operationlimited to sub-kelvin. On the other hand, they exhibitrecord low dissipation power of just ∼ 100 fW (Ic ∼pA,Vout ∼ 100 mV) and should achieve flux noise levels ofjust a few nΦ0/

√Hz. The latter has not been determined

experimentally yet due to limitations from the voltagenoise of the room temperature amplifiers.

C. NanoSQUIDs based on cuprate superconductors

High-Tc cuprate superconductors such as YBCO havevery small and anisotropic values of ξ, reaching ∼ 1 nmfor the a− b plane and a minute ∼ 0.1 nm for the c axis,making the fabrication of cJJs extremely challenging.Still, the fabrication of YBCO cJJs with 50 nm × 50 nmcross-section and 100− 200 nm length has been reportedrecently101. These JJs exhibit large Ic of a few mA at300 mK. NanoSQUIDs based on this technology were fab-ricated and preliminary measurements showed low fluxnoise

√SΦ,w = 700 nΦ0/

√Hz at 8 K.

Probably the most mature JJs from cuprate supercon-ductors are based on Josephson coupling across grainboundaries (GBs). Grain boundary junctions (GBJs)can be fabricated, e.g., by epitaxial growth of cuprate su-perconductors on bicrystal substrates or biepitaxial seedlayers102–104. Although micrometric SQUIDs based onGBJs have been produced30, the miniaturization of high-quality GBJs is challenging, because of degradation ofthe material due to oxygen loss during nanopatterning.NanoSQUIDs made of high-Tc GBJs are, on the otherhand, very attractive due to their large critical currentdensities (∼ 105 A/cm2 at 4.2 K) and huge upper criticalfields (several tens of T).

YBCO GBJ nanoSQUIDs were fabricated by FIBmilling48,52,53. Devices consist of 50 − 300 nm thickYBCO epitaxially grown on bicrystal SrTiO3 substrates(24◦ misorientation angle) and covered by typically 60 nm

Fig. 13

GB

FIG. 13. SEM image of YBCO nanoSQUID loop (400 ×300 nm2), intersected by 130 nm wide GBJs; the GB is indi-cated by the vertical dashed line. The loop contains a 90 nmwide constriction for flux biasing and optimum coupling. [af-ter Schwarz et al.52]

13

thick Au serving as resistive shunt and to protect theYBCO during FIB milling. Typical inner hole size is200− 500 nm and GBJs are 100− 300 nm wide [Fig. 13].Devices are non-hysteretic and work from < 1 K up to∼ 80 K. Large magnetic fields can be applied perpendic-ular to the GBJs in the substrate plane, without severedegradation of the Ic modulation for fields up to 3 T

52.Via a modulation current Imod through a constriction(down to ∼ 50 nm wide) in the loop, the devices can beflux-biased at their optimum working point, without ex-ceeding the critical current, i.e. the constriction is notacting as a weak link. The constriction is also the posi-tion of optimum coupling of a MNP to the SQUID.

Numerical simulations based on London equations forvariable SQUID geometry provided expressions for L andφµ [via Eq. (10)] for a magnetic dipole 10 nm above theconstriction, as a function of all relevant geometric pa-rameters. Together with RCSJ model predictions forSΦ,w at 4.2 K, an optimization study for the spin sen-sitivity has been performed. An optimum film thick-ness dopt = 120 nm was found (for λL = 250 nm). Forsmaller d, the increasing contribution of Lkin to the fluxnoise dominates over the improvement in coupling. Foroptimum βL ∼ 0.5 and d = dopt, the spin sensitivity de-creases monotonically with decreasing constriction lengthlc (which fixes the optimum constriction width wc). Forlc and wc of several tens of nm, an optimum spin sensi-tivity of a few µB/

√Hz was predicted in the white noise

limit55.

For an optimized device with small inductance L ∼4 pH (d = 120 nm, lc = 190 nm, wc = 85 nm), directreadout measurements of the magnetic flux noise at 4.2 Kgave 50 nΦ0/

√Hz at 7 MHz (close to the intrinsic thermal

noise floor), which is amongst the lowest values reportedfor dc SQUIDs so far [Fig. 14]. With a calculated cou-pling factor φµ = 13 nΦ0/µB, this device yields a spin

sensitivity of 3.7µB/√

Hz at 7 MHz and 4.2 K53. Due to

Fig. 14

100 101 102 103 104 105 106 1070.02

0.1

1

10

f (Hz)

experiment fit

S1/2 Φ (µΦ

0/Hz1

/2)

45 nΦ0/Hz1/2

T=4.2 K

FIG. 14. (b) Rms flux noise of optimized YBCO nanoSQUID,measured in open-loop mode. Dashed line is a fit to the mea-sured spectrum; horizontal line indicates fitted white noise.[after Schwarz et al.53]

the extremely low white noise level, 1/f -like excess noisedominates the noise spectrum within the entire band-width of the readout electronics. Bias reversal can onlypartially eliminate this excess noise, which deserves fur-ther investigation.

Finally, an encouraging step towards the controlled for-mation and further miniaturization of high-Tc JJs hasbeen made recently by105. For this purpose a 0.5-nm-diameter He+-beam was used to fabricate ∼ 1 nm-narrowion-irradiated barriers on 4µm wide and 30 nm thickYBCO bridges. The key point is the smallness of the ionbeam diameter, which allows the introduction of pointlike defects. By varying the irradiation dose between1014 − 1018 He+/cm2 the authors showed the successfulrealization of JJs exhibiting SNS-like or tunnel-like be-havior. This technique has been applied to the fabrica-tion of SQUID devices106, but their downsizing to thenanoscale still needs to be realized.

IV. NANOSQUIDS FOR MAGNETIC PARTICLEDETECTION

Originally, nanoSQUIDs were conceived for the inves-tigation of individual MNPs and SMMs. These sys-tems are of key technological importance with appli-cations ranging from electronics, including hard discs,magnetic random access memories, giant magneto re-sistance devices, and spin valves, through on-chip adia-batic magnetic coolers, and up to biotechnology appli-cations including enhanced imaging of tissues and or-gans, virus-detecting magnetic resonance imaging, andcancer therapy (see, e.g., Ref.107). Moreover, magneticmolecules appear as an attractive playground to studyquantum phenomena108 and could eventually find appli-cation in emerging fields of quantum science such as solid-state quantum information technologies109 and molecularspintronics110.

In this section we will review, as an important applica-tion of nanoSQUIDs, the investigation of small magneticparticles. We will first address challenges and approachesregarding positioning of MNPs close to the SQUIDs andthen discuss measurements of magnetization reversal andof ac susceptibility of MNPs.

A. Nanoparticle positioning

The manipulation and positioning of MNPs close tothe nanoSQUIDs is particularly important since themagnetic signal coupled to any form of magnetometerstrongly depends on the particle location with respect tothe sensor. Although conceptually very simple, this prob-lem has hampered the realization of true single-particlemagnetic measurements so far. Many strategies havebeen developed to improve the control on the position-ing of MNPs or SMMs on specific areas of nanoSQUIDsensors.

14

Fig. 13

1 µm 200 nm

(a) (b)

FIG. 15. SEM images of (a) Conanoparticle deposited by FEBIDon the constriction of a YBCOnanoSQUID and (b) nanodot de-posited by FIBID on a SiNi can-tilever above a Nb nanoloop. Par-ticles are highlighted by dashed cir-cles SEM images courtesy of J. Sesé.

1. In-situ nanoparticle growth

In an early approach, called the drop-casting method,small droplets with suspended MNPs were deposited ona substrate containing many nanoSQUIDs. After sol-vent evaporation some of the MNPs happened to oc-cupy positions of maximum coupling. This method wassuccessfully applied to investigate 15− 30 nm individualCo MNPs111 . In a similar approach, MNPs based onCo, Fe or Ni were sputtered using low-energy clusterbeam deposition techniques onto substrates containinga large amount of microSQUIDs112. Alternatively, MNPand Nb deposition was realized simultaneously to em-bed nanometric clusters into the superconducting films,which were subsequently patterned to form nano- ormicroSQUIDs113. The drawback of these techniques isthe lack of precise control of the MNP positions relativeto the SQUIDs, which requires the use and characteriza-tion of many tens or even hundreds of SQUIDs.

Improved nanometric control over the particle po-sition can be achieved by nanolithography methods.This has been used to define Co, Ni, TbFe3 andCo81Zr9Mo8Ni2 MNPs with smallest dimension of 100×50 × 8 nm311. Alternatively, focused e-beam induceddeposition (FEBID) of high-purity cobalt (from a pre-cursor gas, e.g., Co2(CO)8

114) allows the definition ofmuch smaller particles (down to ∼ 10 nm) and arbitraryshape located at precise positions with nanometric reso-lution. This technique has been successfully applied tothe integration of amorphous Co nanodots onto YBCOnanoSQUIDs [Fig. 15(a)]115.

2. Scanning probe-based techniques

The use of a scanning probe, e.g., the tip of an AFMcan be used for precise manipulation of the position of aMNP. AFM imaging in non-contact mode is first used tolocate MNPs dispersed over a surface Then, using con-tact mode, the tip is used to literally ‘’push” the MNP tothe desired position116,117. This technique was applied toimprove the coupling between a nanoSQUID and Fe3O4NPs (15 nm diameter) deposited via the drop-castingmethod38. Micro- and nanomanipulators installed inside

SEMs have also been used for this purpose. For instance,a sharpened carbon fiber mounted on a micromanipulatorin a SEM has been used to pick up a ∼ 0.15µm diametersingle FePt particle an deposit it onto a nanoSQUID88.

Alternatively, larger carriers that are more easily visi-ble and manipulated can be used to manipulate the posi-tion of MNPs. For example, microscopic SiNi cantileverscontaining the MNP of interest can be moved using amicromanipulator118 [Fig. 15(b)]. In particular, CNTsappear as promising tools for this purpose. SMMs haveindeed been successfully grafted over or encapsulated in-side CNTs, which were later used to infer their magneticproperties17. Similarly, an Fe nanowire encapsulated ina CNT has been mounted by micromanipulators on topof YBCO nanoSQUIDs for magnetization reversal mea-surements (see section IV B)53.

Another promising approach is dip pen nanolithogra-phy (DPN). Here, an AFM tip is first coated with a solu-tion containg MNPs and then brought into contact witha surface at the desired location. Capillarity transportof the MNPs from the tip to the surface via a watermeniscus enables the successful deposition of small collec-tions of molecules in submicrometer dimensions119. Bel-lido et al.120 showed that this technique can be appliedto the deposition of dot-like features containing mono-layer arrangements of ferritin-based molecules onto mi-croSQUID sensors [Fig. 16(a)] for magnetic susceptibilitymeasurements121 (section IV C). The number of MNPsdeposited per dot can be controlled (via the concentra-tion of the ferritin solution and dot size) from several hun-dred of proteins down to individual ones120. Recently,DPN has also been applied to the deposition of dot-likefeatures containing just 3−5 molecular layers of Mn12 andDy2 SMMs onto the active areas of microSQUID-basedsusceptometers, enabling the detection of their magneticsusceptibility122,123 [Fig. 16(b)].

Recently, individual magnetic nanotubes, attached toan ultrasoft cantilever were brought in close vicinity toa nanoSQUID at low T 51,124,125. This technique allowedthe authors to investigate magnetization reversal of thenanotubes by combining torque and SQUID magnetom-etry (see section IV B).

We note that scanning SQUID microscopy could alsobe applied to the study of MNPs deposited randomly on

15

5 µm

Fig. 14

(a) (b)

7 µm

FIG. 16. (a) Ferritin nanodots (dashed circles) deposited by DPN on top of the pickup coil of a SQUID-based microsusceptome-ter. Each dot contains 104 proteins approximately arranged as a monolayer. Scheme of the DPN nanopatterning technique;a conventional AFM probe delivers dot-like features containing monolayer arrangements of ferritin over the surface [afterMart́ınez-Pérez et al.121]. (b) Optical microscope image taken during the DPN patterning process showing the AFM probeover a microsusceptometer’s pickup coil. The blow-up shows an AFM image of the resulting sample containing 5 molecularlayers of Dy2 SMMs. Images courtesy of F. Luis.

surfaces126. This would provide an elegant way of lo-cating magnetic systems close enough to the sensor andwould also enable in-situ reference measurements. How-ever, their use for the investigation of magnetic moleculesor nanoparticles arranged on surfaces is still awaiting.

3. Techniques based on chemical functionalization

The above mentioned techniques can be further im-proved by chemically functionalizing the sensor’s surfaceor the MNPs or both of them127. This usually pro-vides high quality monolayers or even individual mag-netic molecules at specific positions. For instance, Mn12SMMs could be successfully grafted on Au, the preferredsubstrate for chemical binding, by introducing thiolgroups in the clusters128. In a further step, such Mn12molecules could be individually isolated by a combinationof molecule and Au substrate functionalization129.

This technique has also been applied to the de-position of ferritin-based MNPs onto Au-shuntednanoSQUIDs130. For this purpose, a 200 × 200 nm2window was opened through e-beam lithography onto aPMMA layer deposited on top of the nanoSQUID. Thiswindow was then covered with organic linkers that werelater used to attach the ferritin MNPs. The success ofthis process was finally determined by AFM, showing ev-idence that few proteins were attached.

B. Magnetization measurements

NanoSQUIDs can be applied to study the reversalof magnetization M of MNPs placed nearby. For thispurpose an external magnetic field Bext is swept while

recording changes in the magnetic moment µ of the sam-ple coupled as a change of magnetic flux to the SQUID[Fig. 4(b)]. Usually, M(Bext) is hysteretic, due to an en-ergy barrier created by magnetic anisotropy. Such hys-teresis loops reveal information on the reversal mecha-nisms, e.g. domain wall nucleation and propagation or theformation of topological magnetic states like vortices, co-herent rotation, or quantum tunneling of magnetization.Depending on the particle’s anisotropy, this requires theapplication of relatively large Bext, a difficult task whendealing with superconducting materials. Measurementsare usually done by careful alignment of Bext with respectto the nanoSQUID, to minimize the magnetic flux cou-pled to the loop and the JJs by Bext directly. The maxi-mum Bext will be limited by the upper critical field of thesuperconducting material, e.g. ∼ 1 T for Nb films, unlessultrathin films are used, which however increases signifi-cantly Lk and hence the flux noise (see section III B 3).

The vastest amount of dc magnetization studies per-formed on individual MNPs was provided by the pioneer-ing work of Wernsdorfer and co-workers. They were ableto measure magnetization curves of a number of MNPsmade of Ni, Co, TbFe3 and Co81Zr9Mo8Ni2 with sizesdown to 100× 50× 8 nm2. Furthermore, they succeededin measuring the dc magnetization of the smallest MNPsever detected to date. These are 3 nm diameter crys-talline Co MNPs (103 µB each) directly embedded intothe Nb film forming the nanoSQUID41. The detectedmagnetization switching process was attributed to an in-dividual MNP located precisely at the cJJ, where thecoupling factor is maximized. These studies also enabledthe determination of the 2nd and 4th order anisotropyterms in the magnetic anisotropy of the Co MNPs. Ad-ditionally, many exciting phenomena were studied withthis technique. These include, e.g., the observation of

16

Fig. 15

Si cantilever Laser

beam Ni tube

Ib

Imod

Ib Imod

JJs

ΦN

N/Φ

0 ∆

f (H

z)

µ0H (mT)

z

x

y

(a) (b)

(c)

FIG. 17. (a) Sketch of combined torque andnanoSQUID magnetometry on a Ni nanotube.(b,c) Simultaneously measured hysteresis loops(b) ΦNN(H), (c) ∆f(H). Arrows indicate Hsweep direction. Dashed lines indicate disconti-nuities appearing in both ΦNN(H) and ∆f(H).[after51 and124]

Stoner-Wohlfarth and Néel-Brown type of thermally as-sisted magnetization reversal in individual Co clusters(25 nm, 106 µB)

111 or the observation of macroscopicquantum tunneling of magnetization in BaFeCoTiO sin-gle particles (10 − 20 nm, 105 µB)131. Magnetization re-versal triggered by rf field pulses on a 20 nm diameter CoNP was also reported132 and, recently, the effects of theantiferromagnetic-ferromagnetic exchange bias between aCo nanocluster and a CoO layer were revealed133. Micro-metric SMM crystals were also investigated with an arraycontaining four microSQUIDs134. These experiments al-lowed observing the modulation of the small (10−7 K)tunnel splitting in Fe8 molecular clusters under the ap-plication of a transverse magnetic field135.

Magnetization reversal mechanisms in single Niand permalloy nanotubes were investigated usingNb/HfTi/Nb-based nanoSQUIDs51,124,125. Experimentswere performed at 4.2 K with Bext = µ0H applied alongthe nanotube axis (z-axis), with the SQUID loop in thex-z plane. The nanoSQUID was mounted on an x-y-zstage below the bottom end of the nanotube which isaffixed to an ultrasoft Si cantilever [Fig.17(a)]. The nan-otube was positioned to maximize the flux ΦNN coupledto the nanoSQUID. While recording the SQUID outputoperated in FLL, simultaneously the magnetic torque ex-erted on the nanotube was detected, by recording thefrequency shift ∆f on the cantilever resonance frequencyas a function of H. Measurements on a Ni nanotubeshowed discontinuities at the same values of H that wereascribed to switching of the magnetization along the nan-otube [Fig. 17(b)]. These experiments provided, on theone hand, the magnetic field stray produced by the nan-otube’s end and, on the other, the volume magnetization,giving evidence for the formation of a magnetic vortex-like configuration in the nanotube. Measurements on anindividual permalloy nanotube evidenced the nucleationof magnetic vortices at the nanotube’s end before prop-agating through its whole length, leading to the com-plete switching of the magnetization. Furthermore, ithas been shown that a thin exchange-coupled antiferro-

magnetic native-oxide layer on the nanotube modifies themagnetization reversal process at low temperatures125.

YBCO nanoSQUIDs were used for the investigationof magnetization reversal in a Fe nanowire grown in-side a CNT attached on top of the SQUID53 [Fig.18(a)].Magnetization measurements were performed at 4.2 Kin FLL by continuously sweeping H in the plane ofthe SQUID loop, along the Fe wire axis. Rectangu-lar shaped hysteresis loops [Fig.18(b)] indicate a singledomain state for the nanowire. The magnitude of theswitching field suggests that magnetization reversal takesplace non-uniformly, e.g., by curling. These results agreevery well with previous measurements on an individualnanowire using a micro-Hall bar18, albeit with a signif-icantly improved signal-to-noise ratio. Similarly, YBCOnanoSQUIDs were used to detect the magnetization re-versal of individual Co MNPs with magnetic moments(1− 30)× 106 µB at different temperatures ranging from300 mK up to 80 K. These studies allowed the identi-fication of two different reversal mechanisms which de-pend on the dimensions and shape of the Co particles.The different reversal mechanisms are linked to the sta-bilization two different magnetic states, i.e., the (quasi)single-domain and the vortex state115.

C. Susceptibility measurements

Even more demanding, nanoSQUIDs can also be usedto quantify the response of a MNP to an oscillating mag-netic field Bac = B0 cos(ωt), i.e., its frequency-dependentmagnetic susceptibility χac = χre + iχim, where χre is thepart going in-phase with Bac and χim is the out-of-phasepart. These quantities bear much information on the dy-namic behavior of spins and the relaxation processes tothermal equilibrium, the interaction between spins, andthe ensuing magnetic phase transitions. These measure-ments can be performed using SQUID-based susceptome-ters, usually in a gradiometric design to be insensitive tohomogeneous external magnetic fields, but sensitive to

17

the imbalance produced by a MNP located in one of thecoils [Fig. 4(c,d)]. χre and χim are directly accessible byapplying a homogeneous Bac via on-chip excitation coilsand lock-in detecting the nanoSQUID output. Alterna-tively,

√SΦ can be measured, as it is directly related to

χim through the fluctuation-dissipation theorem136. The

detection of χac demands high sensitivity, as the net os-cillating polarization induced in the sample is, by far,smaller than the total saturation magnetization. At best,broad-band frequency measurements must be performedwhich also provide an easy way to filter out the 1/f noiseof the SQUIDs, therefore improving the effective sensi-tivity of the sensor. Frequencies are usually restrictedto ∼ 1 MHz, mainly limited by the room temperatureamplifiers and the FLL circuit.

One of the most controversial observations of quan-tum coherence in nanoscopic magnets was realized us-ing the SQUID-based microsusceptometer developed byKetchen et al.9. This device allowed the detection ofthe magnetic susceptibility of small spin populations ofnatural horse-spleen ferritin137. For a sample with just4 × 104 proteins (∼ 200µB/protein), a resonance peakin both the out-of-phase component of χac and

√SΦ

has been observed and was attributed to the zero-fieldsplitting energy137,138. This is the energy separating thetwo non-degenerated low-energy quantum states, i.e., the(anti-)symmetric combination of the classical states cor-responding to magnetization pointing (down) up. Thisinterpretation and the magnitude of this zero-field split-ting (900 kHz) is still an object of debate.

MNPs artificially grown inside ferritin were also stud-ied using a SQUID-based microsusceptometer121. Themagnetic core with diameter of just a few nm was com-

Fig. 18

Fe nanowire

CNT

SQUID loop

300 nm

(a)

-150 -100 -50 0 50 100 150-2

-1

0

1

2

Φ (m

Φ0)

M (1

06A/

m)

µ0H (mT)

T=4.2 K

Ms

-Ms -100

-50

0

50

100(b)

FIG. 18.(a) SEM imageof Fe nanowireencapsulatedin a CNT ontop of a YBCOnanoSQUID.(b) Hysteresisloop Φ(H) ofthe Fe nanowire,detected by theSQUID. Leftaxis correspondsto magnetiza-tion signal M ;the literaturevalue for thesaturation mag-netization Ms =1710 kA/m ofFe is indicatedas dashed lines.[after Schwarz etal.53]

Fig. 17

0.0 0.2 0.40

40

80

0.0 0.2 0.4

10-2 100 102 104 10-2 100 102 104

T (K)

Bulk-like

χ re (µ

B/kO

e)

Monolayer dots f (kHz)

0.0212.179

T (K)

T (mK) 15 48 78 125184 230 289 401

χre

χim

χ (a

.u.)

f (Hz)

f (Hz)

(a)

(b)

FIG. 19. Magnetic susceptibility χ measured with SQUID-based microsusceptometers. (a) Ferritin monolayer dots andbulk sample: χre(T ) obtained at three different frequencies.The superparamagnetic blocking of the susceptibility is visiblebelow 50 mK in both cases [after Mart́ınez-Pérez et al.121].(b) HoW10 SMM crystal: χre(f) (left) and χim(f) (right)measured at different T .

posed of antiferromagnetic CoO leading to a tiny mag-netic moment of ∼ 10µB per protein. Monolayer ar-rangements of ferritin MNPs (total amount ∼ 107 pro-teins) were deposited by DPN directly onto the SQUID,maximizing the coupling between the samples and thesensor’s pickup coils120 (see section IV A 2). Using Bac ∼0.1 mT, these experiments showed that ferritin-basedMNPs arranged on surfaces retain their properties, stillexhibiting superparamagnetic blocking of the magneticsusceptibility [Fig. 19(a)]. Furthermore, these resultsallowed to determine experimentally the spin sensitiv-ity. This was done by determining the coupling, i.e., themeasured flux signal coupled to the microsusceptome-ter divided by the total magnetic moment of the parti-cle, which was located at an optimum position on topof the field coil or close to the edge of the pickup-loop.Together with the measured flux noise of the SQUID,

this yielded S1/2µ ∼ 300µB/

√Hz. Additionally, a large

amount of measurements on SMM micron-sized crystalsor powder at very low T were reported [Fig. 19(b)]. Thelarge bandwidth of these susceptometers (1 mHz–1 MHz)enabled, e.g., the investigation of the relationship be-tween quantum tunneling and spin-phonon interactionand to point out novel and reliable molecular candidatesfor quantum computing and low-temperature magneticrefrigerants (e.g., Refs.34,139–141).

Microsusceptometers were also used to detect the acmagnetic susceptibility of just ∼ 9 × 107 Mn12 SMMsarranged as dot-like features containing 3–5 molecu-lar layers122. Measurements showed an evident de-crease of the magnetic relaxation time compared to that

18

observed in crystalline Mn12. This phenomenon wasattributed to structural modifications of the surface-arranged molecules leading to an effective decrease oftheir activation energy. These sensors have been alsoapplied to the investigation of quantum spin dynamics ofFe4 SMMs grafted onto graphene flakes

142.

V. NANOSQUIDS FOR SCANNING SQUIDMICROSCOPY

In scanning SQUID microscopy (SSM) the high sensi-tivity of SQUIDs to magnetic flux is combined with highspatial resolution by scanning a sample under investi-gation relative to a miniaturized SQUID sensor, or viceversa. A variety of SSM systems has been developed inthe 1990s and refined since then. Those were based onboth, metallic low-Tc and high-Tc cuprate superconduc-tors, although the majority of work focused on the low-Tcdevices. For a review on the developments of SSM in the1990s see Ref.143.

Obviously, miniaturized SQUID structures can signifi-cantly improve the spatial resolution and sensitivity to lo-cal magnetic field sources. A key issue is the requirementto approach the surface of the samples under investiga-tion to a distance which is of the order of or even smallerthan the SQUID size or pickup loop, in order to gain inspatial resolution by shrinking the lateral dimensions ofthe structures. Several strategies for improving the spa-tial resolution in SSM have been followed, which can bedivided into three approaches. The two conventional ap-proaches, developed in the 1990s use SQUID structureson planar substrates. One is based on the sensing of localfields by a miniaturized pickup loop, coupled to a SQUIDsensor; the other is based on using miniaturized SQUIDloops to which local magnetic signals are coupled directly(section V A). A very recently developed third approachuses the SQUID-on-tip (SOT), i.e. a SQUID depositeddirectly on top of a nanotip (section V B).

A. SQUID microscopes using devices on planarsubstrates

SQUID microscopes developed at IMB research byKirtley et al.144 are based on Nb/Al-AlOx/Nb technol-ogy. The sensors are based on a single SQUID loopwith an integrated pickup loop44. The pickup loops havediameters down to ∼ 4µm and are connected via wellshielded superconducting thin film leads to the SQUIDloop at typically ∼ 1 mm distance on the same chip145.This technology has also been used to realize a miniaturevector magnetometer for SSM by using three SQUIDswith orthogonal pickup loops on a single chip146. Asa key advantage, the IBM designs are based on thevery mature Nb multilayer SIS technology, including pat-terning by photolithography, that allows e.g. using theHYPRES147 process for sensor fabrication. Moreover,

this allows to integrate field coils around the pickup loopfor susceptibility measurements and modulation coils in-ductively coupled to the SQUID loop for separate fluxmodulation of the SQUID, i.e. without disturbing thesignals to be detected by the pickup loop. The Si sub-strate is polished to form a corner, typically at a distancedcorner of a few tens of µm away from the center of thepickup loop. SQUID microscopes based on such sensorsuse a mechanical lever for scanning. The SQUID chip ismounted on a cantilever with a small inclination angle αto the plane of the sample. The vertical pickup-loop tosample distance is then given by dcorner sinα

144. If theSQUID is well thermally linked to the liquid He bath foroperation at 4.2 K, the sample mounted in vacuum canbe heated to above ∼ 100 K148.

The most important application of the IBM micro-scope was the pioneering work on the order parametersymmetry of cuprate superconductors. Just to men-tion a few examples, this includes key experiments forproviding clear evidence of dx2−y2-wave pairing in thecuprates by imaging fractional vortices along YBCOGBJs149, the formation of half-integer flux quanta incuprate tricrystals150 and in Nb/cuprate hybrid Joseph-son junctions, forming zigzag-type JJs or huge arrays ofπ-rings151. For more applications, see the review13.

Very similar devices, also based on the Nb multilayertechnology, have been developed and used for SSM bythe Stanford group of Moler and co-workers58,59. Basedon the original microsusceptometer design of Ketchenet al.44, these devices contain two oppositely woundpickup coils, to cancel homogeneous applied fields. Sen-sors with ∼ 4µm pickup-loop diameter achieved

√SΦ =

0.8µΦ0/√

Hz at 4 K and 0.25µΦ0/√

Hz below 0.5 K59,152.The sensor’s substrate was cut by polishing, leading todcorner ∼ 25µ. A capacitive approach control was usedto monitor the probe-to-sample distance. These micro-susceptometers were largely improved by using a terracedcantilever obtained through a multilayer lithography pro-cess. In this way the pickup loop stands above the restof the structure lying at just 300 nm above the samplesurface. Additionally the pickup loop diameters werereduced down to 600 nm using focused ion beam (FIB)milling152. Based on these SQUID sensors, the Stanfordgroup has developed a SQUID microscope operating attemperatures down to 20 mK in a dilution refrigerator153.

The SSM system of the Stanford group has been verysuccessfully applied to a variety of interesting systems.Just to give a few examples, this includes the study ofedge currents in topological insulators154, surface mag-netic states155 and twin walls156 at the LaAlO3/SrTiO3interface, or unpaired spins in metals33.

As an alternative approach, the group of Hasselbachand co-workers at Institut Néel, Grenoble developedan SSM based on miniaturized Nb and Al SQUIDsloops with constriction JJs71, very similar to the onesof the Wernsdorfer group38. This approach allows fora relatively simple single layer fabrication process withprospects of strong miniaturization. To achieve at the

19

same time small probe-to-sample distances, the sensor’ssubstrate was cut using a dicing machine and a mesa wasdefined by means of reactive ion etching so that the dis-tance between the SQUID and apex of the mesa (‘tip’)was only 2 − 3µm. With an inclination angle α ∼ 5◦,this gives a smallest vertical distance to a sample surfaceof ∼ 0.26µm. The SSM setup is combined with forcemicroscopy, based on the use of a mechanically excitedquartz tuning fork and operates in a dilution refrigerator,achieving minimum SQUID and sample temperatures of0.45 K73. Very recently, in a modified setup, a SQUID-to-sample distance of 420 nm has been demonstrated, ina setup with 40 mK base temperature157.

The SSM system of the Grenoble group has been ap-plied to the investigation of basic properties of supercon-ductors. This includes, e.g., studies on the direct observa-tion of the localized superconducting state around holesin perforated Al films158 or on the Meissner-Ochsenfeldeffect and absence of the Meissner state in the ferromag-netic superconductor UCoGe159.

B. SQUID-on-tip (SOT) microscope

An important breakthrough in the field ofnanoSQUIDs applied to SSM was achieved recentlywith the implementation of the SQUID-on-tip (SOT)by the Zeldov group at the Weizman Institute ofScience91,92. This device is based on the deposition ofa nanoSQUID directly on the apex of a sharp quartzpipette [Fig.20]. The fact that the nanoSQUID islocated on a sharp tip reduces the possible minimumprobe-to-sample distances to below 100 nm, boostingenormously the spatial resolution of the microscope. Al,Nb and Pb nanoSQUIDs based on Dayem bridges areshadow-evaporated in a three-angle process, withoutrequiring any lithography or milling steps. For thispurpose, a quartz pipette is first pulled to form a sharphollow tip with 40 − 300 nm inner diameter. By meansof a laser diode parallel to the tip, the latter is alignedpointing down towards the source which defines the0◦ position. Then a thin layer (< 10 nm) of supercon-ducting material is deposited, followed by two thickerleads (> 25 nm) deposited at ±100◦. The resultingweak links formed at the tip apex between these twoleads constitute two Dayem bridges. Special care mustbe taken for fabricating the Nb and Pb sensors. Theformer require the previous deposition of a thin AlOxbuffer layer to prevent contamination from the quartztip. A dedicated ultra-high vacuum e-beam evaporationsystem was used for depositing Nb from a point source.On the other hand, the so far most sensitive Pb sensorsrequire the use of a He cooling system for the tips duringdeposition to prevent the formation of islands due to thelarge surface mobility of these atoms at higher tempera-tures. This procedure lead to the smallest nanoSQUIDsfabricated so far, with effective nanoloop diametersdown to 50 nm. The resulting inductance of the loop

Fig. 19.9 FIG. 20. SQUID-on-tip (SOT): (a) schematic of a sharpquartz pipette with superconducting leads, connecting to theSOT at the bottom end; inset shows magnified view. (b)SEM image of a Nb SOT having a diameter of 238 nm.Reprinted by permission from Macmillan Publishers Ltd: Na-ture Nanotechnology92, copyright (2013).

reaches values below 10 pH, dominated by the kineticinductance of the thin superconducting layer. Althoughthese nanoSQUIDs exhibit hysteretic IVCs, operationwith voltage bias and reading out the resulting currentsignal with an SSA enables the detection of the intrinsicflux noise of the devices. The SOTs can be operated inlarge magnetic fields up to ∼ 1 T (limited by the uppercritical fields of the superconducting materials). So far,flux biasing to maintain the optimum working pointduring continuous external field sweep is not possible.By adjusting the external magnetic field to values whichyield large transfer functions, these devices exhibitextraordinary low flux noise levels down to 50 nΦ0/

√Hz

for the Pb SOTs92. The latter varies, depending onthe biasing external magnetic field. For a magneticdipole located at the center of the loop with orientationperpendicular to the loop plane (assuming an infinitelynarrow width of the loop, i.e. the approximation used byKetchen et al.44), this translates into a spin sensitivity

of 0.38µB/√

Hz, i.e. the best spin sensitivity reported sofar for a nanoSQUID.

A device able of distinguishing in-plane and out-of-plane magnetic signals was also reported160. This isachieved by using a pipette with θ-shaped cross section toform a three JJ SQUID (3JSOT). This tip is later milledby FIB leading to a V shaped apex with two obliquenanoloops connected in parallel. By measuring the de-pendence of the maximum critical current on the exter-nally applied in-plane and out-of-plane magnetic fieldsIc(H||, H⊥), it is possible to determine all the geometricaland electric parameters of the device. Field componentscan be decoupled by biasing the 3JSOT at specific fields(H||, H⊥) in which Ic depends strongly on one of the twoorthogonal components of the magnetic field while beinginsensitive to the other. As a drawback, this device is

20

not able of distinguishing both in-plane and out-of-planecomponents of the magnetic flux simultaneously, but onlywhen operated at different flux biasing points.

For SSM, a system operating in a 3He system with300 mK base temperature has been developed, with theSOT glued on a quartz tuning fork, to operate the systemalso in an magnetic force microscopy mode. This allowsscanning (using piezo-scanners) at extremely small tip-to-sample distances of only a few nm161. A spatial resolu-tion below 120 nm was demonstrated by imaging vorticesin Nb thin films with a 117 nm-diameter Pb SOT92.

The SOT-SSM system has been successfully appliedto the study of vortex trajectories in superconductingthin films, allowing the investigation of the influenceof the pining force landscape162. More recently, thistool was used to observe nanoscopic magnetic struc-tures such as ferromagnetic metallic nanoislands at theLaMnO3/SrTiO3 interface

163 or magnetic nanodomainsin magnetic topological insulators164.

VI. SUMMARY AND OUTLOOK

Significant progress in thin film fabrication and patter-ing technologies has enabled the development of stronglyminiaturized dc SQUIDs with loop sizes on the microm-eter scale (microSQUIDs) or even with sub-micrometerdimensions (nanoSQUIDs), or SQUIDs coupled to minia-turized pickup loops. Such devices are based on a vari-ety of Josephson junctions, intersecting the SQUID loop,many of them also on the sub-micrometer scale. Asa key advantage of such strongly miniaturized SQUID

structures, they can offer significantly reduced flux noise,down to the level of a few tens of nΦ0/

√Hz, cor-

responding to spin sensitivities around 1µB/√

Hz andimproved spatial resolution for scanning SQUID mi-croscopy. Hence, strongly miniaturized SQUIDs are verypromising detectors for investigating tiny and stronglylocalized magnetic signals produced, e.g., by magneticnanoparticles or for high-resolution scanning SQUID mi-croscopy. Very recent advances, including the demon-stration of single spin sensitivity and a breakthrough inspatial resolution of scanning SQUID microscopy openup promising perspectives for applications in nanoscalemagnetism of condensed matter systems.

ACKNOWLEDGMENTS

We gratefully acknowledge valuable contributions bythe nanoSQUID team at Tübingen, M. Kemmler,J. Nagel, R. Wölbing, T. Schwarz, B. Müller, S. Hessand R. Kleiner, and by our collaborators O. Kieler andA. Zorin et al. at PTB Braunschweig, T. Schurig et al. atPTB Berlin, B. Büchner et al. at IFW Dresden, M. Pog-gio et al. at Univ. Basel, D. Grundler et al. at TUMunich and EPFL Lausanne, A. Fontcuberta i Morralat EPFL Lausanne and J. Sesé at INA Zaragoza. Thiswork was funded by the Alexander von Humboldt F