This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 129.31.240.94

This content was downloaded on 20/06/2014 at 13:17

Please note that terms and conditions apply.

The spin-valve transistor: a review and outlook

View the table of contents for this issue, or go to the journal homepage for more

2003 J. Phys. D: Appl. Phys. 36 R289

(http://iopscience.iop.org/0022-3727/36/19/R01)

Home Search Collections Journals About Contact us My IOPscience

INSTITUTE OF PHYSICS PUBLISHING JOURNAL OF PHYSICS D: APPLIED PHYSICS

J. Phys. D: Appl. Phys. 36 (2003) R289–R308 PII: S0022-3727(03)38950-8

TOPICAL REVIEW

The spin-valve transistor: a review andoutlookR Jansen

MESA+ Research Institute, SMI, University of Twente, 7500 AE Enschede, The Netherlands

Received 4 June 2003Published 17 September 2003Online at stacks.iop.org/JPhysD/36/R289

AbstractCombining ferromagnetic and semiconductor materials is a challengingroute to create new options for electronic devices in which the spin of theelectron is employed. The spin-valve transistor (SVT) is the first of suchhybrid devices shown to work successfully. This review describes the basicscience and technology of the SVT and derived devices, such as themagnetic tunnel transistor.

It seems almost inevitable that electronics in the near futurewill employ the electron spin degree of freedom. Twoexciting scientific breakthroughs within the last 15 years andtheir exceptionally rapid implementation into products alreadydemonstrates the impact of using the electron spin. In 1988,the giant magnetoresistance (GMR) effect [1] was discoveredin multilayer structures that contain layers of ferromagneticmetals separated by a thin spacer of normal metal. Theresistance of such structures was found to depend greatly on therelative magnetic orientation of neighbouring magnetic layers,making it attractive for application in highly sensitive magneticfield sensors [2–4]. As early as 1997, GMR was incorporatedinto the read-heads of magnetic hard disc recording systems,where it has been one of the main factors enabling thetremendous increase in storage density over the past decade.The second important breakthrough [5] was the demonstrationin 1995 of large tunnel magnetoresistance (TMR) in tunneljunctions in which two ferromagnetic electrodes are separatedby an ultrathin insulator. Key features [6] are the largespin dependence of the tunnelling process, even at roomtemperature, and the reproducible fabrication of reliable tunnelbarriers typically using 1–2 nm of Al2O3. These featuresfacilitated the development of a high performance magneticrandom access memory (MRAM). It is non-volatile, has lowpower consumption and fast switching speed [7–9], and hasthe potential of becoming a universal memory. The firstintroduction of tunnelling based MRAM into the market isimminent.

While GMR and tunnelling magnetoresistance demon-strate the opportunities of using electron spin, it is still anopen question how pervasive the use of spin will be, andwhat classes of material it will involve. In GMR and TMR

structures, ferromagnets are combined with thin layers of nor-mal metals and insulators, respectively. We are thus miss-ing the most essential material in contemporary electronics,namely, semiconductors. The core element of electronics isthe transistor with its amplification, which relies on extremelypure semiconductor materials in which electrical conductioncan be controlled by manipulating the carrier density usingelectric fields supplied by a gate. Semiconductors allow pre-cise tuning of carrier concentrations, band gap engineering,and, interestingly, also exhibit extremely long electron spinlifetimes [10, 11]. Spin-electronics (or spintronics) based onsemiconductors is therefore an active and promising researchfield [3], with several approaches being explored in parallel.These are briefly reviewed below.

Spin-transport in (quantum-) structures built exclusivelyfrom non-magnetic or paramagnetic semiconductors isexplored actively, but a concern is that the creation of a spinpolarization at present involves optical techniques or largemagnetic fields to induce a sizable Zeeman splitting of the

Figure 1. Schematic representation of the different ways in whichsemiconductors and ferromagnetic materials can be combined innovel electronics.

0022-3727/03/190289+20$30.00 © 2003 IOP Publishing Ltd Printed in the UK R289

Topical Review

spin states. It would be advantageous to use conventionalmagnetic materials such as Fe, Ni, Co and their alloys, in whichferromagnetism is robust and Curie temperatures are high.This supports a stable, virtually ever lasting memory function,while switching between different magnetization states isextremely rapid, typically at nanosecond or even picosecondtimescales. It is, thus, attractive to examine different ways(figure 1) to combine ferromagnets and semiconductors.

The most straightforward approach is the one employedin MRAM, where an array of magnetic memory elementsis placed on top of a semiconductor wafer containingtransistors and other circuitry required to drive the memory.This requires a modest, yet not trivial integration at the systemlevel, but does not take advantage of the unique properties ofsemiconductors in manipulating spin. The most intimate formof integration is to put magnetic properties into semiconductormaterials, thus creating ferromagnetic semiconductors [12].Such materials can be obtained by doping with a certain amountof magnetic atoms, as in the case of the notable exampleof GaMnAs. Interplay between ferromagnetism and carrierdensities via electrical gating has been achieved in some ofthe materials [13,14]. The hunt is now on for compounds thatexhibit both semiconducting and ferromagnetic properties attemperatures well above room temperature.

An intermediate option is to create hybrid devicestructures in which ferromagnets and semiconductors arecombined. One could think of many different geometriesof such hybrid structures. Two main categories will bedistinguished, based on whether the control and manipulationof the spins occurs in the semiconductor or in the ferromagneticmaterial. In the first category, electron spins that originatefrom a ferromagnetic source material are injected into asemiconductor, in which they are transported and manipulated,followed by some means of spin detection at the ‘otherend’ of the device [15]. Much progress has recently beenmade on the first important step of spin-injection into thesemiconductor [16–20], but implementation into workingdevices that operate at room temperature remains to bedemonstrated. The opposite is true for the second class ofhybrid devices, where a device concept has been successfullydemonstrated by Monsma et al [21] with the introduction ofthe spin-valve transistor (SVT) in 1995, and the subsequentobservation of huge magnetic response at room temperature[22, 23] a few years later. While the spin dependence of thetransport is in the ferromagnetic materials, the semiconductorsare used to create energy barriers in the electron’s potentiallandscape that are essential to the operation of thedevice.

This review will cover this second category ofsemiconductor/ferromagnet hybrid structures, including theSVT and derived hot-electron devices such as the magnetictunnel transistor (MTT). Hopefully, this review presents abasic understanding of such devices and provides insightinto their strengths and weaknesses, which is also relevantfor spin-electronics in general. Other spintronic devices,such as those based on spin-injection into semiconductors, inthe end also have to address issues such as signal levels,noise, room temperature operation, device scaling, etc. It isimportant to realize the magnitude of the challenge this poses,as contemporary semiconductor electronics is a highly mature

technology with exceptional performance. Spin-electronicswill be successful only if it results in greatly enhancedperformance, or, in a completely novel electronic functionality.Hence, merely creating a transistor based on spin (or carbonnanotubes or molecules) is not sufficient as silicon transistorswith on-off ratios of more than 108 are readily available.

This review is organized in three sections as follows. Insection 1, the state-of-the-art of the SVT is presented. Itstarts with a description of the device structure, the principleof operation, as well as essential fabrication technology. Itthen continues to cover two main topics in more detail: (i) thefundamental physics of the spin-dependent electronic transportand (ii) important device characteristics and performance.In section 2, the MTT is discussed. Since the principleof operation and the underlying physics is essentially thesame as for the SVT, the focus is on those aspects of theMTT that are different from the SVT. In the third section,the use of hot-electron spin-filtering for spin-injection intoa semiconductor is briefly discussed. Finally, a summary isgiven and some promising avenues for further research areidentified.

1. Spin-valve transistor

1.1. Basic device structure and properties

The SVT was introduced in 1995 and is the first workinghybrid device in which ferromagnets and semiconductors havebeen closely integrated, and both materials are essential incontrolling the electrical transport through the device. Thethree-terminal device has the typical emitter/base/collectorstructure of a (bipolar) transistor, but is different in that the baseregion is metallic and contains at least two magnetic layersseparated by a normal metal spacer (see figure 2). The twomagnetic layers act as polarizer and analyser of electron spins,such that the relative orientation of the magnetization of the twolayers determines the transmission of the base. The resultingsalient feature of the SVT is that the collector current dependson the magnetic state of the base. This was first demonstratedby Monsma et al [21] in a SVT showing a huge change of390% of the collector current in an applied magnetic field atlow temperature. In 1998, the first device operating at roomtemperature was achieved, having a 15% effect in fields of a

Figure 2. Basic layout of the SVT, showing the three terminalarrangement with semiconductor emitter (top), semiconductorcollector (bottom), and the metallic base comprising twoferromagnetic thin layers separated by normal metals (middle).

R290

Topical Review

Figure 3. Schematic layout and energy band diagram of a SVT,showing the semiconductor emitter (left) and collector (right), andthe metallic base comprising a spin valve (middle). Also depictedis the stream of electrons that is injected into the spin valve baseabove the Fermi energy.

few kOe [24]. More recently, we succeeded in the reproduciblefabrication of SVTs that exhibit magnetocurrent effects up to400% at room temperature, and in small magnetic fields ofonly a few Oe [22, 23].

Unlike other spintronic devices, the SVT is based onthe spin-dependent transport of non-equilibrium, so-called hotelectrons, rather than Fermi electrons. In order to illustratethis and explain the principle of operation, let us consider thespecific structure that was used in [22]. That particular SVTuses silicon as the semiconductor for the emitter and collector,and has a metallic base that contains a Ni80Fe20/Au/Co spinvalve (see figure 3). At the interfaces between the metal baseand the semiconductors, energy barriers (Schottky barriers) areformed [25]. These energy barriers prevent electrons with theFermi energy from travelling through the structure. To obtainthe desired high quality Schottky barrier with good rectifyingbehaviour and thermionic emission dominating, low doped Si(1–10 � cm) is used, and thin layers of, e.g. Pt and Au areincorporated at the emitter and collector side, respectively.These also serve to separate the magnetic layers from directcontact with Si.

The operation of the SVT is as follows. A current isestablished between the emitter and the base (the emittercurrent IE), such that electrons are injected into the base,perpendicular to the layers of the spin valve. Since the injectedelectrons have to go over the Si/Pt Schottky barrier, theyenter the base as non-equilibrium, hot electrons. The hot-electron energy is determined by the emitter Schottky barrierheight, which is typically between 0.5 and 1 eV, dependingon the metal/semiconductor combination [25]. As the hotelectrons traverse the base, they are subjected to inelasticand elastic scattering, which changes their energy as wellas their momentum distribution. Electrons are only able toenter the collector if they have retained sufficient energy toovercome the energy barrier at the collector side, which ischosen to be somewhat lower than the emitter barrier. Equallyimportant, a hot electron can only enter the collector if itsmomentum matches with that of one of the available statesin the collector semiconductor. The fraction of electronsthat is collected, and thus the collector current IC, dependssensitively on the scattering in the base, which is spindependent when the base contains magnetic materials. Thetotal scattering rate is controlled with an external applied

-40 -20 0 20 400

4

8

12

T = 295 K MC = 238 %

magnetic field (Oe)

colle

ctor

cur

rent

(nA

)

Figure 4. Collector current versus applied magnetic field for a SVTwith Si(100) emitter, Si(111) collector, and the following base:Pt(20 Å)/NiFe(30 Å)/Au(35 Å)/Co(30 Å)/Au(20 + 20 Å).IE = 2 mA, VBC = 0 and T = 295 K.

magnetic field, which changes the relative magnetic alignmentof the ferromagnetic Ni80Fe20 and Co layers in the base.This is illustrated in figure 4, where IC at room temperatureis plotted as a function of magnetic field, for a transistorwith a Si(100) emitter, a Si(111) collector and the followingbase: Pt(20 Å)/NiFe(30 Å)/Au(35 Å)/Co(30 Å)/Au(20+20 Å).At large applied fields, the two magnetic layers have theirmagnetization directions aligned parallel. This gives thelargest collector current (I P

C = 11.2 nA). When the magneticfield is reversed, the difference in switching fields of Co (22 Oe)and NiFe (5 Oe) creates a field region where the NiFe andCo magnetizations are antiparallel. In this state, the collectorcurrent is drastically reduced (IAP

C = 3.3 nA). The magneticresponse of the SVT, called the magnetocurrent (MC), isdefined as the change in collector current normalized to theminimum value, i.e.

MC = I PC − IAP

C

IAPC

, (1)

where P and AP refer to the parallel and antiparallel magneticstate of the base spin valve, respectively. Thus, the relativemagnetic response is MC = 240%, which is huge indeed.

It was shown [22, 26], that the collector current andthe MC are virtually independent of a reverse bias voltageapplied across the collector Schottky barrier, provided that theintrinsic ‘leakage’ current this induces in the collector diode isnegligible compared to the hot-electron current. This is indeedthe case, as can be seen in figure 5. A voltage between base andcollector does not affect the hot-electron current because it doesnot significantly change the maximum of the Schottky barrierwhen measured with respect to the Fermi energy in the metal[25]. In other words, the energy barrier seen by hot electronscoming from the base is hardly changed. Similarly, a change ofthe emitter to base voltage, or equivalently the emitter current,does not affect the energy at which hot electrons are injectedinto the base. Also here, the applied voltage hardly modifiesthe maximum of the emitter potential barrier. The result isthat the collector current is simply linearly proportional to theemitter current, as shown in figure 6.

It should be noted here that huge or even colossal magneticresponse has been observed in other materials and devices. Theuniqueness of the SVT is that the huge relative magnetic effectis obtained at room temperature, and that only small magneticfields of a few Oe are required. The combination of these

R291

Topical Review

three features is what makes it attractive. Furthermore, theresults are reproducible and the properties of the device canbe manipulated to a certain extent by controlling the thicknessof layers, the type of materials, etc. Nevertheless, a particularpoint that needs attention is the absolute value of the outputcurrent. In the above example, an emitter current of 2 mAwas used and the corresponding transfer ratio, defined asα = IC/IE, is thus of the order of 10−6. In this context, itis important to remember that the term ‘transistor’ should notbe confused with ‘amplifier’. The word transistor was chosenin analogy with the metal base transistor (MBT) [27–29], adevice that is similar to the SVT except that the base containsonly non-magnetic metals. These structures were studiedextensively in the 1960s and 1970s, from which it becameclear that a metallic base has too little transmission to supportamplification [27–29]. Hence, the SVT should not be judgedon its (lack of) amplification. Rather, it should be viewed as adevice with a magnetic field dependent electrical output, whichis the basic functionality one needs for a magnetic field sensoror a magnetic memory element. Although current gain is thusnot required for most applications, a small absolute currentis a disadvantage. This issue and the progress made so faris addressed in one of the subsequent sections.

Before discussing the device in more detail, the essentialrole of the semiconductors in providing potential energy

0.0 0.5 1.0-10

0

10

20

30

40

IE = 0

IE = 1 mA

IE = 2 mA

Col

lect

or c

urre

nt (

nA)

Collector-base voltage (V)

Figure 5. Transistor characteristics: collector current as a functionof collector base voltage, for different values of the emitter current,at T = 295 K.

0.0 0.5 1.0 1.5 2.00

10

20

30

IE

IC = constant

Col

lect

or c

urre

nt (

nA)

Emitter Current (mA)

Figure 6. Collector current as a function of emitter current for thesame SVT as in figure 5. VBC = 0 and T = 295 K.

barriers must be emphasized. The energy barrier at the emitterside is needed to create injection of hot electrons, such thattransport is governed by non-equilibrium processes. Thecollector energy barrier acts as an energy and momentum filter,allowing only a fraction of the hot electrons to pass into thecollector, and reflecting the rest. The collector Schottky barrier(Au/Si in the above example) selects only on the basis ofenergy and momentum, but not on the spin of the incominghot electrons. In some sense, the role of spin is thus indirectas it is merely used to manipulate the energy and momentumdistribution of the hot electrons during their motion throughthe base. This is essential though, as spin couples to anexternal applied magnetic field and is our handle to the outsideworld. As will be discussed in detail below, the origin of thelarge magnetic sensitivity of the SVT is the non-equilibriumnature of the transport that gives an exponential decay of basetransmission, together with the strong spin dependence of theelastic and inelastic scattering parameters in ferromagneticmaterials.

1.2. Fabrication: vacuum metal bonding

For the correct operation of the SVT, it is crucial to havehigh-quality Schottky contacts that form defect-free energybarriers. This requires device quality semiconductor material.Since there is no known method of growing crystalline Si withlow defect density on top of a thin metal film, an alternativefabrication technology based on metal bonding was developed[30]. The basic idea, illustrated in figure 7, is as amazingas it is essential for the SVT. In ultrahigh vacuum, we firstdeposit part of the base metal layers onto the emitter Si wafer,using a shutter to prevent deposition on the collector Si wafer.Then the shutter is opened, metal is deposited onto both waferssimultaneously, during which the surfaces of the metal coatedwafers are brought into contact. A metal–metal bond is formedthat bonds the two wafers together. The whole process of metaldeposition and bonding is done in situ under ultrahigh vacuumconditions in a molecular beam epitaxy system. This yieldsclean and incredibly strong metallic bonding.

As described previously [30, 31], a kind of recrystalliza-tion occurs when the two metal surfaces are brought into con-tact. This can repair structural defects at the bonding interface,provided the two surfaces are not too rough and atomic diffu-sion is sufficiently large. The driving force for the formation

Figure 7. Concept of vacuum metal bonding, in which two Si wafersare joined together using freshly evaporated metal as the ‘glue’.

R292

Topical Review

Figure 8. Cross-sectional TEM images of two Si wafers aftervacuum metal bonding by Co (left) and Au (right).

of the chemical bonds is the gain in free energy when twosurfaces are combined to form a bulk crystal structure. Thequality, strength and reliability of the metal bond thereforedepends on the type of metal used, as well as on the particularfilm topography (roughness, etc).

To study the bonding interface, we performed metalbonding in structures that are simpler than the SVT in thatthey contain only one type of metal, instead of the completespin valve. The bonding interfaces were then investigated bytransmission electron microscopy (TEM). Two examples areshown in figure 8, where two Si substrates are bonded togetherby two 5 nm thick Co layers (left panel) or two Au layers (rightpanel). For the Co structure, one can clearly recognize a lighterband that indicates a high density of structural defects at thebonded interface (denoted by the dashed arrow). In contrast,for the case of Au one recognizes parallel atomic (111) planesrunning across the complete Au layer and no clear bondinginterface is visible. Although we have mostly used Au, goodresults have also been obtained with other metals such as Cuand Ti [30,31]. For practical reasons related to the mechanicaldesign of the ultrahigh vacuum bond robot, the bonding istypically done at room temperature with 1 cm2 pieces of Si.Special care has to be taken to work in dust-free conditions andremove edge particulates after sawing of the wafers. We notethat bonding is not limited to Si, but is a versatile technique thatallows one to join many kinds of semiconductors. For example,SVTs with GaAs as the emitter and Si as collector have beensuccessfully fabricated using vacuum metal bonding [32].

After metal bonding, the structures are further processedinto smaller devices using standard photolithography and aseries of dry and wet etching steps. The first step is to patternthe emitters into squares ranging from 350 × 350 µm2 to1 × 1 mm2, using lithography and wet etching. The mostconvenient and reproducible way to do this uses a silicon-on-insulator (SOI) wafer for the emitter. After bonding, thehandle wafer of the SOI wafer is first etched away, usingthe buried oxide as the etch stop. Subsequently, the oxide isetched away, leaving a thin Si layer of homogeneous and well-defined thickness. This layer, 3 µm thick, has a doping profilealready built in. The front side is low doped (concentration1015 cm−3, n-type), to form a proper Schottky barrier with

Figure 9. The final test chip with 52 SVTs of different sizes.

the metal base. The back side is highly doped such thata good Ohmic contact is obtained by deposition of Cr andAu. The next step is patterning of the metal base intoslightly larger rectangles using ion milling. A drawback isthat this creates damage in the Si collector, which results inlarge leakage currents along the edges of the metal/collectorSchottky contacts. This is detrimental as it reduces the MCby adding only to the total collector current (denominator inequation (1)). The edge leakage currents are found to dependstrongly on temperature, and, as shown previously [22, 26],may prevent a large magnetic response being obtained at roomtemperature. We therefore introduced an additional processstep to remove the damaged collector Si with a last wetchemical etch. Collector Schottky diodes with reverse biascurrents of the order of 0.1 nA can thus be obtained.

The final standard chip consists of a total of 52 SVTswith different sizes (see figure 9). Electrical connectionsto these relatively large SVTs are made by ultrasonic wirebonding. This is however no longer feasible if smallertransistors are to be made. A modified fabrication processwas therefore developed for the miniaturization of the device[33, 34], for which the use of SOI emitter wafers is essential.Devices with interconnect metallization and bondpads havebeen made so far with sizes down to 10 × 10 µm2, and withproperties comparable to the larger SVTs for which results arepresented here.

1.3. Physical basis: hot-electron spin-transport

Unlike other spintronic devices, the SVT is based on thespin-dependent transport of hot electrons, rather than Fermielectrons. Transport at the Fermi energy has been widelystudied in connection with (giant-) magnetoresistance effectsand it is well-established that conduction in ferromagnets andtheir multilayers is dependent on the spin of the electrons[35]. However, hot-electron transport is distinctly differentfrom ordinary transport at the Fermi energy. Therefore, tounderstand the operation of the SVT and avoid erroneousinterpretation of results, it is best to ‘forget’ about what weknow about GMR and spin-dependent scattering in spin valves.

It has been known for quite some time that scattering ratesof hot electrons in ferromagnetic materials are spin dependent.

R293

Topical Review

Early work on hot-electron scattering at energies of about5 eV employed spin-polarized photoemission from overlayerstructures [36, 37]. This and also later experiments [38, 39]at energies as low as 1.5 eV showed unambiguously that theinelastic mean free path of hot electrons is spin dependentin ferromagnets. More precisely, experiments carried outthus far have always found that the inelastic mean free pathof hot electrons is shorter for the minority spin electrons[40]. A common interpretation is that this originates fromthe difference in the number of unoccupied states for the hotelectron to scatter into, assuming electron–hole pair excitationsto be the dominant scattering mechanism. Scattering processeshave also been investigated using transmission through free-standing magnetic thin film foils [41–44], as well as by a time-resolved, two-photon photoemission (2PPE) experiment [45]that directly probed the inelastic lifetime for majority andminority spin.

Spin-dependent scattering of hot electrons is essential ina variety of spin-polarized electron spectroscopies that arewidely used to examine magnetic materials. It leads to spinfiltering and in magnetic multilayers it gives rise to phenomenasuch as the hot-electron spin-valve effect, first observed forsecondary electrons [46]. This is employed in the magneticversion of ballistic electron emission microscopy (BEEM)[47], which enables magnetic imaging with nanometerresolution [48,49]. With the introduction of the SVT [21,24],hot-electron spin-transport was implemented in a solid-stateelectronic device. A thorough understanding of the spin-dependent scattering mechanisms of hot electrons is paramountto the further development of this type of device. Interestingly,the SVT itself has opened up a new route to study spin-dependent scattering processes of hot electrons, extendingexperiments to lower energy in the range between 0.5 and1.5 eV. Moreover, for a solid-state device like the SVT it israther easy to vary experimental parameters like temperature orapply large magnetic fields. This has led to some new insightsinto the origin of the spin-dependent scattering processes, asdiscussed below.

It is now well established that the difference between I PC

and IAPC originates from the spin asymmetry in hot-electron

attenuation length, leading to the dominant transmission ofmajority spins in each magnetic layer. This is schematicallyillustrated in figure 10. We write:

I PC ∝ T M

NiFeTM

Co + T mNiFeT

mCo, (2)

IAPC ∝ T M

NiFeTm

Co + T mNiFeT

MCo, (3)

where T M and T m are the transmission of majority (M) andminority (m) spin hot electrons, given by

T Mfm = �M

in exp

(− tfm

λMfm

)�M

out, (4)

T mfm = �m

in exp

(− tfm

λmfm

)�m

out. (5)

Here, tfm is the film thickness, λMfm and λm

fm the hot-electronattenuation lengths for majority and minority spin, and fmdenotes the ferromagnet. The factors �in, �out denotethe transmission across the interface of the ferromagnet withthe normal metal at each side, which can be spin dependent

Figure 10. Schematic view on spin-dependent attenuation of hotelectrons in a ferromagnetic thin film, causing preferentialtransmission of majority spin hot electrons.

due to the mismatch in the bandstructure of the magneticand non-magnetic metals. Thus, the interfacial attenuationis generally dependent on the combination of ferromagnet andnormal metal. Now, an initially unpolarized current acquires,after transmission of a film fm, a spin polarization P given by

Pfm = T Mfm − T m

fm

T Mfm + T m

fm

. (6)

This is related directly to the magnetocurrent MC via:

MC = I PC − IAP

C

IAPC

= 2PNiFePCo

1 − PNiFePCo. (7)

In general, the attenuation is caused by a combinationof inelastic and elastic scattering processes. It is importantto realize that while the scattering parameters are uniquelydefined for a given ferromagnet and hot-electron energy, theattenuation length is not [50]. The attenuation length dependsnot only on the scattering parameters, but to some extent also onthe device geometry and the energy and momentum selectionapplied by the collector Schottky barrier. For example, in thehypothetical case that a collector is used which passes hotelectrons regardless of their momentum, elastic scattering inthe base metals would not directly cause attenuation, andthe attenuation length would be governed mainly by inelasticprocesses1.

Let us address the origin of the large magnetocurrenteffect and consider the factors that distinguish the SVT frommagnetoresistive structures such as a magnetic multilayer.First of all, transport is by hot electrons, which probe a differentportion of the band structure than Fermi electrons, while thedominant scattering mechanisms are different and includeinelastic scattering such as by electron–hole pair excitations.Second, the resistance of metals is governed by the mean freepath, and the resistivity is inversely proportional to the meanfree path. In contrast, the hot-electron transport in the SVT is a

1 Some indirect effect of elastic scattering still occurs because it changes thepath length in the base and thereby increase the probability for an inelasticscattering event.

R294

Topical Review

non-equilibrium phenomenon where the current transmissiondepends exponentially on the base thickness, and the parameterthat controls the decay is the attenuation length. As statedabove, this is related to the mean free path, but also dependson the energy and momentum selection at the collector barrier.Furthermore, the relative magnetic effect (the MC) of theSVT is rather insensitive to scattering that carries no spindependence, as it attenuates the current for both spins withan equal factor, leaving the ratio unchanged. A final point isthat transport is largely perpendicular to the magnetic layers,such that the electrons have to cross all the interfaces. All thesefactors together facilitate the large effects.

1.4. Interface versus volume scattering

The reproducible fabrication of the SVT permitted a detailedstudy of the spin-dependent transmission of low energy hotelectrons through ferromagnetic films and trilayer structures.A first question that was addressed is, what are the typicalattenuation lengths λM

fm and λmfm for majority and minority spin

hot electrons in a ferromagnetic metal? Another interestingquestion is whether the dominant scattering occurs withinthe volume of the magnetic layers or at the interfaces withthe non-magnetic layers. With respect to spin-dependentscattering, one may expect volume scattering to dominate asits contribution grows exponentially with ferromagnetic filmthickness and will quickly outweigh the (constant) interfacialcontribution (see equations (4) and (5)). This is indeed whatexperiments with the SVT suggest [51,52]. Such experimentsinvolve fabrication of a series of SVTs that are identical exceptfor the thickness of one of the ferromagnetic layers of the base.

Let us discuss the particular example of a transistor withthe following base structure Pt(30 Å)/Ni80Fe20(x)/Au(44 Å)/Co(30 Å)/Au(44 Å), where the thickness x of the Ni80Fe20

layer was varied. Figure 11 shows I PC and IAP

C (top panel) andMC (bottom panel) at T = 100 K, as a function of the thicknessx of the Ni80Fe20 layer, keeping the Co layer thicknessconstant at 30 Å. For the parallel magnetic state of the spinvalve, minority spins are strongly attenuated in both magneticlayers, and only majority spins contribute to I P

C . From theexponential decay of I P

C with thickness, we thus deduce avolume attenuation length of 43 ± 3 Å for majority hot spinsin Ni80Fe20.

The interface attenuation for majority spins is extracted bycomparing the IC value of 169 nA for the device with x = 0, tothe value of 78 nA obtained by extrapolation of the I P

C data tozero thickness. The latter value is a factor of 2.2 lower, whichis due to attenuation by the extra interface that is created whena Ni80Fe20 layer is inserted between the Pt and the Au layerof the base (more precisely, the factor of 2.2 represents thedifference between the original Pt/Au interface, and the twonew Pt/Ni80Fe20 and Ni80Fe20/Au interfaces). Note that this isthe attenuation for the majority spins, as the minority spins arefiltered out by the 30 Å Co layer, irrespective of their scatteringat the Ni80Fe20 interfaces. The interfacial attenuation is acombination of the mismatch of the electronic states at bothsides of the interface, and elastic scattering due to interfacedisorder, defects, etc.

Information on the spin dependence of the volume andinterface attenuation is contained in the difference between I P

C

1

10

100 interface} T = 100 K

AP

P

colle

ctor

cur

rent

(nA

)0 20 40 60 80 100

0

100

200

300

400

500

volume only interface only data

MC

(%

)

NiFe thickness (Å)

Figure 11. I PC and IAP

C (top panel, labels P and AP) and MC (bottompanel) at 100 K, for SVTs with various Ni80Fe20 thickness. Solidand dashed lines, respectively, represent the extreme cases with onlyspin-dependent volume scattering or only spin-dependent interfacescattering (see text).

and IAPC . Let us first compare the data with what is expected

for two extreme cases, using equations (2)–(5) for the collectorcurrent. In the first case, represented by the dashed lines infigure 11, all the spin dependence is assumed to arise fromthe interfacial scattering, while identical volume attenuationlengths are used for both spins. In that case the MC would beindependent of Ni80Fe20 thickness, and I P

C and IAPC decay with

x with the same slope. The experimental data clearly deviatesfrom this behaviour at small thickness. In contrast, the dataagrees very well with the other extreme case (solid lines), inwhich the interface scattering is not spin dependent, and theonly spin-asymmetry is that of the volume attenuation lengths.In this situation we expect that I P

C and IAPC approach each other

as x goes to zero, such that the MC tends to zero. Note that forlarge x, filtering of minority spins is complete and the decay ofboth I P

C and IAPC is determined by the majority spin attenuation

length.More precise analysis, taking into account the experimen-

tal error margins, shows that some weak spin dependence ofthe interface scattering cannot be excluded [51]. For the ratioof the interface attenuation factor � for minority and major-ity spin we obtain �↓/�↑ = 0.8 ± 0.2, while volume atten-uation lengths are λ

↓NiFe = 10 ± 2 Å for minority spin and

λ↑NiFe = 43 ± 3 Å for the majority spin (at 0.9 eV). With these

parameters, a 30 Å thick Ni80Fe20 film attenuates minorityspins a factor of 10 more strongly than the majority spins.Volume scattering is thus by far the dominant contribution tothe spin-dependence of the hot-electron attenuation, mainlybecause it depends exponentially on the layer thickness.

R295

Topical Review

1.5. Elastic versus inelastic

In hot-electron scattering, it is obvious that inelastic scatteringplays a role, as it leads to energy loss and thus makes theelectrons ‘less hot’. However, elastic scattering cannot beneglected a priori. The point is illustrated best by asking whatwe would expect if there were no elastic scattering whatsoever,neither in the volume of the layers, nor at interfaces. Canwe explain the basic electrical characteristics of the SVTusing only inelastic scattering? A simple estimate basedon exponential decay of the transmission of a particularlayer is illuminating. It is well-known from MBTs [27–29]and BEEM [53, 54] that scattering lengths for non-magneticmaterials such as Au or Cu are quite long, at least 100 Å.In the SVT, the thickness of these normal metal layers istypically only 30–60 Å such that inelastic attenuation in theselayers can be neglected. In order to satisfy the experimentalobservations (a typical MC of 400% at low T and an overalltransmission of 10−4 for SVTs with ferromagnetic layers of30 Å (see section 1.7)), one would require inelastic attenuationlengths of 7 Å and 4.5 Å for majority and minority spins,respectively.

Such short inelastic scattering lengths appear unrealistic.Moreover, they are at odds with a recent direct measurement[45] of the inelastic lifetime of hot electrons by 2PPE, whichyielded a lifetime of 8 fs and 4 fs for majority and minorityspin hot electrons, respectively, in Co at an energy of about1 eV. With a typical electron velocity of 2 × 106 m s−1, thistranslates into inelastic scattering lengths of about 160 and80 Å for majority and minority spins in Co. These values aremuch larger than those quoted above. For minority spins, asignificant density of d-states is still present at energies of 1 eVabove the Fermi energy, and one could argue that the electronvelocity is much smaller than assumed above. Yet, such anargument cannot be made for majority spins, for which onlyhighly mobile states with s or p character are available. It ispossible that there is a principle reason why inelastic lifetimesdetermined by the optical technique of 2PPE are much largerthan those observed in electronic transport. This may be [55]due to diffusion of electrons away from the optical spot, dueto matrix elements for optical transitions, or due to a differentenergy resolution of the techniques.

Attenuation lengths for several ferromagnetic materialsin the SVT [51], the MTT [56], and in magnetic BEEM [49]all consistently give minority spin attenuation lengths around10 Å, and majority scattering lengths in the range of 25–60 Å.These measured values are significantly larger than values of4.5 and 7 Å estimated above for the case in which only inelasticscattering is present, but smaller than what is derived fromthe 2PPE results. A consistent picture can only be madeif elastic scattering is included. First, transport attenuationlengths that are determined by elastic and inelastic processesare naturally shorter than predicted from 2PPE, which probesonly the inelastic scattering lifetime. Second, when measuredattenuation lengths are used to calculate the total volumeattenuation of the base layers, one arrives at a transmissionof about 0.01. Since observed base transfer ratios are at least2 orders of magnitude smaller, a significant fraction of theattenuation must be attributed to scattering at interfaces, whichis primarily elastic.

Finally, we mention that attributing the measured transportattenuation length solely to inelastic scattering is inconsistentwith the general trend observed in SVT [57], MTT [56] andBEEM data [49] that the overall transmission increases withincreasing hot-electron energy (see also section 2 on the MTT).For inelastic scattering, the opposite should happen, sinceaccording to Fermi’s golden rule the lifetime is reduced whenat higher energy more final states become available to scatterinto. This is a most compelling argument and we concludethat the experimental data cannot be explained if only inelasticprocesses are considered. The picture of purely ballistictransmission, thus, seems rather inappropriate.

Experimental evidence for the importance of elasticscattering comes from the explicit observation of attenuationdue to interfaces (figure 11 in the previous section) andfrom the direct observation of spin-orbit scattering [58]. Inaddition, we have found that the magnitude of the collectorcurrent is insensitive to crystal orientation (100) or (111)of the Si collector (see data in table 1). Yet, a mostinstructive way to examine the role of elastic scattering is tomake use of phenomenological model calculations, in whichscattering parameters can be systematically varied and theeffect on magnetotransport can be studied. The model and thecalculation procedure are described in detail in [50]. Here,we merely discuss an example that illustrates the role ofinterfacial elastic scattering.

For the case in which elastic interface scattering was setto zero, the top panel of figure 12 shows how the calculatedcollectable current decays as the hot electrons traverse themetal base. Here, ‘collectable’ refers to those electrons thatstill satisfy the energy and momentum constraints requiredfor transmission into the Si collector [50]. We observe anexponential attenuation of the current in the volume of eachlayer, with the strongest decay in each of the ferromagneticlayers. The bottom panel of figure 12 shows the results if theinterface diffusivity is set to 0.9 for each interface in the base(including the metal semiconductor interfaces at the emitterand collector side). This corresponds to 90% of the electronsincident on an interface being scattered elastically, withscattered electrons distributed isotropically over all angulardirections. One notes a pronounced drop in the calculatedcollectable current at each interface. This adds up to a largereduction of the overall base transmission. For the emittercurrent of 2 mA used, the total transmitted current I P

C is only8 × 10−9 A for the strong interface scattering case, as opposedto 8×10−6 A without interface scattering. The difference is asmuch as three orders of magnitude. Thus, elastic scattering atinterfaces, for example, due to disorder, can cause a significantreduction of the collector current. Elastic scattering thereforeplays a perhaps surprisingly important role in hot-electrontransport.

Table 1. Transfer ratio α for several Si/Pt/Au/Si transistors withdifferent crystal orientation of the collector. The Schottky barrierheight � and ideality factor n are also given.

�e Base �c α

Emitter (eV) n (Å) Collector (eV) n 10−4

Si(100) 0.91 1.01 40Pt/40Au Si(111) 0.80 1.20 8.5SOI(100) 0.85 1.07 40Pt/40Au Si(111) 0.76 1.20 8.0SOI(100) 0.88 1.02 40Pt/40Au Si(100) 0.77 1.15 8.2

R296

Topical Review

0 4 8 12 16 20

10-9

10-8

10-7

10-6

10-5

10-4

10-3

AuCoAuNiFe

90% interfacescattering

no interfacescattering

P

AP

P

spin down

spin up

AP

PtC

olle

ctab

le c

urre

nt (

A)

0 4 8 12 16 20

10-9

10-8

10-7

10-6

10-5

10-4

10-3

P

Pspin down

spin up

position in base (nm)

Col

lect

able

cur

rent

(A

)

Figure 12. Calculated collectable current versus position in themetal base, for zero and 90% interfacial scattering in the top andbottom panel, respectively.

1.6. Temperature and spin waves

For ferromagnets, experiments carried out this far have alwaysfound that the inelastic mean free path of hot electrons isshorter for the minority spin electrons [40]. A commoninterpretation is that this originates from the difference in thenumber of unoccupied states for the hot electron to scatter into,assuming excitation of electron–hole pairs to be the dominantscattering mechanism. However, recent experimental [59] andtheoretical work [60,61] suggests that scattering by spin-waveexcitations may also contribute to the spin dependence of themean free path. The spin asymmetry is created by spontaneousspin-wave emission, which, due to conservation of angularmomentum, is allowed only for minority spins. Spontaneousspin-wave emission is not dependent on temperature (T ) andoccurs even at T = 0. It is calculated [60, 61] to dominate atelectron energies below ≈3 eV, and is found to be the majorsource of small energy losses [59]. Besides spontaneousemission, there is also a thermal component which involvesspin-wave absorption as well as emission. Information onthese contributions is obtained most directly by varying T .While spectroscopic measurements have been performed, theT -dependence had so far not been explored. Using the SVT,we have obtained the first measurements of spin-dependenthot-electron scattering as a function of temperature [23]. Wefound that thermal spin waves have a noticeable effect onspin-transport of hot electrons, in particular by introducingspin-mixing.

Figure 13 shows the typical variation of the collectorcurrent with applied magnetic field at two temperatures (80 and290 K). At high fields the magnetizations of the Co andNi80Fe20 films in the spin-valve base are parallel. Whenone of the magnetizations is reversed, a clear antiparallel(AP) magnetization alignment is obtained. At T = 80 K,the magnetocurrent MC = 560%, while a huge effect of

0

4

8

12MC = 560 %

T = 80 Kcolle

ctor

cur

rent

(nA

)

-100 -50 0 50 1000

4

8

12

magnetic field (Oe)

MC = 350 %

T = 290 Kcolle

ctor

cur

rent

(nA

)

Figure 13. Collector current versus magnetic field at T = 80 K(top) and T = 290 K (bottom). The SVT structure is: n-typeSi(100)/Pt(20 Å)/NiFe(60 Å)/Au(35 Å)/Co(30 Å)/Au(20 + 20 Å)/n-type Si(111). IE = 2 mA and VBC = 0.

0

4

8

12

antiparallel

parallel

colle

ctor

cur

rent

(nA

)

0 100 200 3000

300

600

0.0

0.5

1.0

CIP

-MR

(%) spin-valve transistor

CIP-MR

temperature (K)

MC

(%

)

Figure 14. Temperature variation of I PC and IAP

C (top panel) and themagnetocurrent (bottom panel) for the same SVT as in figure 13.

350% still remains at room temperature. Similar results werereproducibly obtained.

In figure 14, the temperature variation of I PC and IAP

C(top panel) and the magnetocurrent MC (bottom panel) aredisplayed. For the parallel case, the collector current firstincreases with T , but decreases above 200 K. In contrast,IAP

C increases over the whole temperature range. Theresulting decay of the MC is relatively weak, especially below200 K. The conventional current-in-plane magnetoresistance(CIP-MR) measured on an identical spin valve is also included.The CIP-MR is not only orders of magnitude smaller, but alsoexhibits a stronger thermal decrease that is linear, as also foundby others [62].

R297

Topical Review

To understand the data in figure 14 we consider thatthermal scattering may change the MC in two ways. First,scattering may attenuate, i.e. remove electrons from thecollected current in a spin-dependent fashion. The secondpossibility is spin-mixing, in which electrons are scattered intothe other spin channel by a spin-flip process, without beingremoved from the collected current. We have shown [23] thatwhen we neglect possible thermally-induced spin-mixing, theMC does not decay with T . Thus, the decay of MC is attributedto thermally-induced spin-flip scattering, causing mixing ofthe two spin channels. This is consistent with the observation(figure 2) that the current for the parallel state goes down above200 K, while the current for the antiparallel case continues togo up. Since exchange scattering by paramagnetic impuritiesand spin-orbit scattering have negligible T -dependence in ourexperimental range [63], we consider spin-flip scattering bythermal spin waves.

We stress that spin-mixing only results for those scatteringevents in which the energy and/or momentum transferred tothe spin-wave is such that after scattering, the hot electronis still able to enter the collector (recall that the electronneeds a minimum energy to surmount the collector barrier,and has to be incident at an angle smaller than the criticalangle of acceptance [24,47]). We therefore consider the Bose–Einstein distribution function Nq = 1/[exp(Dq2/kT ) − 1],which determines the typical wave vector q and energyDq2 of thermal spin waves (D the spin-wave stiffness,typically 400 meV Å2). We see that even at 300 K, virtuallyall thermal spin-waves have low energies, that are smallerthan the maximum allowed energy loss (≈60 meV) given bythe difference between emitter and collector barrier. Thus, thechange of hot-electron energy will generally not remove theelectron from the collected current. The situation is differentwhen one considers the wave vector q, which has magnitudesup to roughly 0.3 Å−1 at room temperature. For the largestq, spin-wave scattering may deflect the hot electron enoughto prevent collection, thus causing attenuation. However, forsmall q or when the q-component perpendicular to the hot-electron direction of motion is small, the deflection is weak.In that case, the process contributes to spin-mixing.

In a subsequent experimental study [51], we havemeasured the actual attenuation length associated with thermalscattering by spin waves. Using Matthiessens’ rule, we canwrite λM

fm and λmfm in terms of the attenuation lengths for all the

distinctive scattering processes:

1

λMfm

= 1

λMe–h

+1

λMel

+1

λph+

1

λMTSW abs

, (8)

1

λmfm

= 1

λme–h

+1

λmel

+1

λph+

1

λmTSW emis

+1

λmSSW emis

, (9)

where λe–h, λel and λph are the attenuation lengths associatedwith electron–hole pair excitations, elastic scattering andphonon scattering, respectively. Also included are attenuationlengths λM

TSW abs and λmTSW emis due to absorption and emission

of thermal spin waves, respectively, as well as a term λmSSW emis

due to spontaneous emission of spin waves. Due to theconservation of angular momentum, only majority-spin hotelectrons can absorb spin waves, whereas (spontaneous andthermal) emission is allowed only for minority spins. Thus,

the overall rate of spin wave scattering has a spin asymmetrydue to λm

SSW emis.In order to extract the thermal component of the hot-

electron attenuation, we have measured I PC as a function of

temperature [51]. We divide out all temperature independentattenuation factors by plotting the normalized collector currentIN

C = I PC(T )/I P

C(100 K) (see figure 15). The top panelshows data for transistors with different Ni80Fe20 thicknessand constant Co thickness. Starting at 100 K the collectorcurrent first goes up slightly as seen before, and then isreduced significantly towards room temperature. Moreover,the variation with T is more pronounced at higher Ni80Fe20

thickness, implying that the additional attenuation at higherT is due to a thermal volume scattering process. Thesame behaviour is observed when the thickness of the Colayer is varied (bottom panel of figure 15). However, theT -variation for Co has weaker dependence on layer thicknessthan for Ni80Fe20, which shows that the thermal attenuation isstronger in Ni80Fe20 than it is in Co. This is consistent withattenuation due to scattering on thermal spin-waves [23], asthe Curie temperature of Co (1388 K) is larger than that ofNi80Fe20 (873 K).

Using the procedure described in [51] one can extractthe attenuation length λTSW(T ) due to thermal spin waves.The extracted thermal attenuation lengths for Ni80Fe20 andCo versus T are shown in figure 16. For Ni80Fe20 twosets of data are shown, one for the SVT, the other labelledAMT is extracted from another set of transistors with

0.6

0.7

0.8

0.9

1.0

NiFe20 Co30 NiFe30 Co30 NiFe50 Co30 NiFe70 Co30 NiFe100 Co30

I C (

norm

aliz

ed)

100 150 200 250 300

0.6

0.7

0.8

0.9

1.0

NiFe30 Co20 NiFe30 Co30 NiFe30 Co40 NiFe30 Co50

temperature (K)

I C (

norm

aliz

ed)

Figure 15. Top panel: normalized I PC versus T for SVTs with a

Ni80Fe20/Au/Co spin valve with different Ni80Fe20 thickness and aconstant Co thickness of 30 Å. Labels indicate the thicknesses in Å.Bottom panel: similar for varying Co thickness and a constantNi80Fe20 thickness of 30 Å.

R298

Topical Review

100 150 200 250 3000

200

400

600

800

1000

1200

1400

Co NiFe (SVT) NiFe (AMT)

ther

mal

att

enua

tion

leng

th (

Å)

temperature (K)

100 200 3000.92

0.96

1.00

1.04 NiFe (AMT)

T (K)

Γ(T

) / Γ

(100

K)

Figure 16. Thermal spin-wave attenuation lengths versus T for Coand Ni80Fe20 (two data sets). The inset shows the thermal variationof the interface attenuation for Ni80Fe20.

Pt(20 Å)/Ni80Fe20/Au(44 Å) base. As noted before, λTSW(T )

is shorter for Ni80Fe20 than for Co. Interestingly, λTSW(T ) ismuch shorter than hitherto assumed, with room temperaturevalues of 130 ± 20 Å for Ni80Fe20 and 270 ± 40 Å for Co.Especially for Ni80Fe20 this is only three times larger thanthe majority spin attenuation length at low T . Hence, thisshows convincingly that hot-electron attenuation lengths aredependent on T . For instance, the addition of thermal spinwave scattering with a length scale of 130 Å reduces theattenuation length from 43 Å at 100 K, to a significantly lowervalue of (1/43 + 1/130)−1 = 32 Å at room temperature.

The inset of figure 16 shows the thermal variation of theinterface attenuation �(T ) for Ni80Fe20. Data is obtained bydividing the IN

C curve for certain Ni80Fe20 thickness by thecurve for x = 0, and using the attenuation lengths of figure 16to remove the volume scattering part. We find only a slightchange of about 5% in the interface attenuation between 100and 300 K. This shows that thermal spin wave scattering isprimarily a volume scattering process.

The above results suggest the importance of spin waves forthe spin asymmetry of hot-electron transmission. As shownin [60, 61], the asymmetry is created by the T -independentcontribution of spontaneous spin wave emission, which isonly allowed for minority spins. It has so far not beenpossible to isolate spontaneous spin wave emission andmeasure the corresponding scattering length. However, theabove quantification of the thermal spin wave contributionallows us to estimate the attenuation due to spontaneousemission, which is expected to be significantly strongerthan the thermal scattering rate. This is because thermalspin waves occupy only a small fraction of the spin-wavephase space up to energies of the order of kT, while forspontaneous spin wave emission, the complete phase spaceup to the hot-electron energy (≈0.9 eV � kT) is availableat all temperatures. The strength of the observed attenuationdue to thermal spin waves is thus indirect evidence for theimportance of spontaneous spin-wave emission. If we crudely

estimate the spontaneous emission rate to be about an orderof magnitude larger than the thermal emission rate, we obtainan attenuation length for spontaneous emission that is closeto the measured minority-spin attenuation length (10 Å forNi80Fe20). This strongly suggests that the minority spinattenuation is dominated by spontaneous spin-wave emission,as theory predicts [60,61]. Since the process cannot contributeto attenuation of majority hot spins, we conclude that thespin asymmetry of the attenuation length may well be dueto spontaneous spin-wave emission, instead of the spin-dependent rate of electron–hole pair generation that arises fromthe exchange split bandstructure.

The notion of spontaneous spin-wave emission as thedominant source of spin-asymmetry explains the absence ofbandstructure features in spectroscopic data. In particular, itexplains the puzzling observation [55] by time-resolved 2PPE,that the inelastic lifetime for minority spins is shorter than themajority spin lifetime, not only in Co and Ni, but also in Fe.For Fe, a reversed spin-asymmetry at low energy (<1 eV) isexpected from the bandstructure [55]. However, spontaneousspin wave emission would, as it is forbidden for majority spins,always result in a larger lifetime for majority spins, as observedin the 2PPE experiment.

1.7. Device output and transfer ratio

In close harmony with the more fundamental researchdescribed in the preceding sections, quite some effort hasbeen devoted to improve one of the most important deviceparameters, namely, the output current level. Surely, theprospects of application of the SVT will be greatly enhancedif the collector currents can be raised to well above thenanoampere range, such that one can capitalize on the hugemagnetic response of the device. Fortunately, significantprogress has been made in recent years [57], and there is stillplenty of scope for further improvement.

From the basic relation IC = αIE we see that we canincrease IC by either increasing the emitter current (IE), or thetransfer ratio (α = IC/IE) of the transistor base. However, weknow that the emitter current has an upper limit imposed by thebreakdown of the device [34]. Therefore, we should optimizeα. Parenthetically, the most fruitful way to achieve this is tochange the non-magnetic layers of the base. In a typical SVTstructure of Si(100)/NME/Ni82Fe18/Au/Co/ NMC/Si(100), thinnormal metal layers NME and NMC are incorporated at theemitter and collector side of the structure, respectively. Theprimary function of these layers is to create high qualitySchottky barrier diodes with thermionic emission dominating,to ensure efficient hot-electron injection from the emitter andlow reverse bias leakage of the collector diode. From amagnetic point of view these layers are not active, providingsome flexibility in device design. Below we show that this canbe exploited [57] to enhance the transfer ratio in two ways: (i)controlling the difference in height of the emitter and collectorSchottky barrier, and (ii) by carefully selecting materials withlarge scattering lengths for hot electrons.

We start by investigating the role of the Schottky barrierheight of emitter (�e) and collector (�c) on the transfer ratio.The Schottky barrier heights in the SVT are determined bythe choice of semiconductor, silicon in our case, and the non-magnetic metal. We have systematically varied the Schottky

R299

Topical Review

barrier metal in two series of SVTs [57]. In the first series,denoted as Pt series, NME is Pt and NMC is either Pt, Au orCu. In the second series, denoted as Au series, NME is Au andNMC is either Au or Cu. The properties of the two series oftransistors are summarized in table 2.

In figure 17, we show the parallel collector current versusSchottky barrier height difference (��B = �e − �c) of thetwo series. Both the Pt and Au series show an increase incollector current for larger ��B. The increase is not relatedto a change in hot-electron energy, since in each series theemitter Schottky barrier is fixed, and only the collector barrieris lowered as one goes from Pt to Au to Cu. The samestudies have been performed on MBTs in which the spinvalve has been omitted from the base. The structure is thusSi(100)/NME/NMC/Si(100), where NME and NMC performthe same function and are made of the same materials as usedfor the SVTs in table 1. The results (not shown) yield the sametrend for the transfer ratio versus ��B and material as for theSVT series.

The enhancement of IC is due to the larger number ofstates available in the collector semiconductor. Electronsthat are injected at an energy just above the collector barrier(small ��B) arrive at the base/collector interface at energiesjust above the conduction band minimum of the collector Si.Near the bottom of the band the density of states is small andonly few electrons can enter the semiconductor. When thecollector barrier is lowered or the emitter barrier is increased(large ��B), the injected electrons can be transmitted tostates higher up in the collector conduction band where theavailable density of states is larger. Thus, a larger fractionof the electrons can be collected. This is also seen in BEEM

Table 2. Properties of SVTs with different combinations of emitterand collector barriers, where sv denotes a NiFe/Au/Co spin valve.T = 290 K.

�e �c ��B α MCBase (eV) (eV) (eV) (×10−4) (%)

Pt/sv/Pt 0.88 0.86 0.02 0.01 213Pt/sv/Au 0.88 0.83 0.05 0.07 260Pt/sv/Cu 0.87 0.61 0.26 1.06 218Au/sv/Au 0.81 0.80 0.01 0.10 204Au/sv/Cu 0.82 0.69 0.13 1.18 230

0.00 0.05 0.10 0.15 0.20 0.25 0.30

0

50

100

150

200

250

Au / sv / Au

Pt / sv / Pt

Pt / sv / Au

Pt / sv / Cu

Au / sv / Cu

Si / Au emitter Si / Pt emitterC

olle

ctor

cur

rent

(nA

)

Schottky barrier height difference (eV)

Figure 17. I PC for five types of SVTs versus ��B. The circles are

the Pt series and the squares are the Au series. IE = 2 mA,T = 290 K and sv denotes a NiFe/Au/Co spin valve.

experiments [47], as well as in transistors with a tunnel barrieremitter (as described in section 2).

While the transfer ratio improved a factor of 118 fromthe Pt/sv/Pt base to the Au/sv/Cu base, the variation in MC issmall and non-systematic (see table 1). This implies that therelatively small changes in the energy of the injected electronshave no large impact on the spin-dependent attenuationprocesses. Also the change of the collector barrier heightenhances the collection of both the spin up and spin downelectrons equally. We can thus enhance the transfer ratio andat the same time preserve the huge magnetic response. Themaximum transfer ratio obtained so far was 1.2 × 10−4 forthe Si/Au(20 Å)/NiFe(30 Å)/Au(70 Å)/Co(30 Å)/Cu(40 Å)/Sitransistor. The corresponding MC at room temperature was230% (see figure 18) and nearly 400% at 100 K.

Besides the increase in transfer ratio with ��B, we alsoobserve that the transfer ratio is larger for the Au series thanfor the Pt series. The difference can be explained by the muchshorter hot-electron attenuation length of Pt as compared toAu. BEEM studies yield an attenuation length (λ) of about200 Å for Au [53, 54]. Although no data is available formetallic Pt, a λ of about 40 Å is obtained in a BEEM studyon PtSi [64]. Since the transfer ratio depends exponentially onthe attenuation length, one expects a larger transfer ratio for theAu series than for the Pt series at comparable ��B. Thus, non-magnetic materials with large hot-electron scattering lengths,such as Au and also Cu, are preferred. Materials such as Ptshould be avoided, if possible, despite the fact that it producesthe highest possible Schottky barrier on Si (which was theoriginal motivation for its use in early devices).

Note that by changing the Schottky barrier materialsNME and NMC one may introduce changes in the structuralproperties of the SVT. For example, the quality of the metalbond in NMC may depend on the material used. With respectto layer NME we have specifically examined its influence onthe transfer ratio [57]. For that purpose, a series of transistorswith a Pt/NiFe(30 Å)/Au(40 Å)/Co(30 Å)/Au(40 Å) base layerwere prepared, with the Pt thickness varied from 20 to 60 Å.One would expect an exponential decay with layer thickness.Surprisingly, however, an initial increase in transfer ratio withPt layer thickness was reproducibly obtained (see figure 19).This is attributed to a structural change of the SVT with

-40 -20 0 20 400.00

0.05

0.10

0.15

0.20

0.25

MC = 230%

IE = 2mA, T = 290 K

Col

lect

or c

urre

nt (

µA)

Magnetic field (Oe)

Figure 18. Collector current versus applied magnetic field for aSi/Au(20 Å)/NiFe(30 Å)/Au(70 Å)/Co(30 Å)/Cu(40 Å)/Si transistor.IE = 2 mA and T = 290 K.

R300

Topical Review

0 10 20 30 40 50 60 701

10

100Exponential decayexp(-t

pt /15Å)

Col

lect

or c

urre

nt (

nA)

Pt thickness (Å)

Figure 19. Parallel collector current versus Pt thickness tpt for aSi/Pt/NiFe(30 Å)/Au(40 Å)/Co(30 Å)/Au(40 Å)/Si transistor.IE = 2 mA and T = 100 K.

Pt thickness, although the precise cause has not yet beenidentified.

The above example nicely illustrates the effect of thestructural quality of the SVT on the transfer ratio, andthereby points to a promising route for further improvement.As shown in section 1.5, a significant part of the hot-electron attenuation is due to elastic scattering processeswhich likely originate from structural imperfections (defects,grain boundaries, stacking faults, disorder at interfaces, etc).These are surely present in the polycrystalline, non-epitaxialstructures we have studied so far. Going to epitaxial transistorstructures is an approach we are currently exploring. Anotheroption is to reduce the number of layers (and interfaces) inthe base, and with it, the elastic scattering. This can be donein the slightly modified design of a MTT (see section 2 onthe MTT). Finally, we note that some optimization of thethickness of the ferromagnetic layers is needed, since there is atrade-off between transfer ratio and MC that yields a maximumin absolute change of collector current (�IC) with magneticfield [24, 57].

In the preceding paragraphs we have focused on how toimprove the transfer ratio, keeping the emitter current fixedat a somewhat arbitrary value of 2 mA. However, this is farbelow the breakdown current of the SVT [34]. Thus, theabsolute value of the output collector current is by no meanslimited to the value of 0.2 µA shown in figure 18. To illustratethis, figure 20 shows how the collector current, transfer ratioand MC vary with emitter current, for a SVT with similarstructure as in figure 18. We find that the output currentof the SVT increases approximately linearly with emittercurrent, and large collector currents up to 50 µA are obtained.Interestingly, the increase of output current by several ordersof magnitude is accompanied by a rather weak reduction ofthe MC, and even a slight increase of the transfer ratio. Thisconvincingly demonstrates that the collector current is notintrinsically limited to small values in the nA regime, but canapproach 0.1 mA. In the end, the output current that can beachieved depends on the maximum emitter current that canbe tolerated for a given application. This is determined byextrinsic requirements such as power consumption, devicedimensions since the breakdown emitter current depends ondevice size [34], and input impedance (for the maximum

0

10

20

30

40

50

AP

P

I C (

µA)

0

100

200

300

400

MC

(%

)

0 50 100 150 200 2500.0

0.5

1.0

1.5

2.0

emitter current (mA)

α (1

0-4)

Figure 20. The dependence on emitter current of collector current(top panel); MC (middle) and transfer ratio (bottom) for aSi/Au (20 Å)/NiFe (30 Å)/Au(70 Å)/Co(30 Å)/Cu(40 Å)/Sitransistor.

emitter current in figure 20, the applied emitter voltage wasabout 4 V, corresponding to an input impedance of 16 �). Thisalso implies that care has to be taken when comparing theproperties of the SVT with similar devices such as the MTT,and correct conclusions can only be drawn if differences in theapplied emitter current are properly considered [65].

1.8. Noise

An important parameter for any electronic device is the signal-to-noise ratio (SNR). Let us first examine the origin of thenoise in the SVT. To characterize the noise behaviour overa wide range of collector current values, measurements havebeen performed on three different types of transistors [66].Besides the SVT with Pt/NiFe/Au/Co/Au spin valve base, wealso used a Pt/NiFe/Au MBT (with only a single magneticlayer), as well as the non-magnetic MBT with a Pt/Au base.The collector current noise spectrum showed only white noisein the frequency range of the measurement (10 Hz–1 kHz).The noise current spectral density is plotted as function ofIC in the top panel of figure 21, where the IC for each ofthe three transistors was varied by changing IE. The log–log plot yields a straight line with a slope equal to 1, whichmeans a linear relationship between noise spectral density andcollector current. This is expected when shot noise dominates.In fact, the experimental data are in excellent agreement withthe calculated shot noise of 2qIC, as indicated by the solid line.

R301

Topical Review

10-8

10-7

10-6

10-5

10-4

10-27

10-26

10-25

10-24

10-23

Si / Pt / Au / Si Si / Pt / NiFe / Au / Si Si / Pt / NiFe / Au / Co / Au / Si calculated shot noise 2q IC

curr

ent

spe

ctra

l den

sity

(A

2 /Hz)

collector current (A)

-40 -20 0 20 400

1x10-26

2x10-26

curr

ent

spe

ctra

l den

sity

(A

2 /Hz)

Magnetic field (Oe)

measured noise in SVT calculated shot noise 2q IC

Figure 21. Top panel: Noise current spectral density versus IC fortransistors with three different base structures: Pt/NiFe/Au/Co/Au(��), Pt/NiFe/Au (•), Pt/Au (◦). Bottom panel: noise versusmagnetic field for the SVT. Solid lines in both panels are the shotnoise calculated from the measured IC.

For the SVT, the relation between noise and the magneticstate of the base was studied. As shown in figure 21, bottompanel, the noise varies with applied magnetic field very muchlike the collector current itself, and no extra noise appears inthe field regions where the magnetization reversal occurs. Thechange of the noise with field is entirely due to the change of thecollector current. The solid curve is the shot noise calculatedfrom 2qIC, using the measured collector current versus fieldcurve. Again, good agreement is obtained with the data. Weconclude that shot noise dominates in the investigated currentand frequency regime.

We can now evaluate the SNR of the device, using asthe signal the full difference between I P

C and IAPC . The result is

shown in figure 22, where SNR is plotted against I PC for devices

with different values of the MC percentage. It is evident thatthe SNR improves significantly at higher collector current,emphasizing the need for further increasing the transfer ratio.One also notes that there is a significant gain in SNR betweenan MC of 10% and 100%, however, beyond 100% the gainlevels off and only a marginal SNR increase is observed whenthe MC is further raised to 1000%. This may seem counterintuitive, but is simply because the absolute value of the signalI P

C − IAPC is already close to the maximum at an MC of several

100%. It is for this reason that we have never attempted tocreate an SVT with MC values above 1000%, although it isstraightforward to do so, one simply increases the thicknessof the ferromagnetic layers (to 50 Å or more). In fact, the netresult would be a loss in SNR, because the collector current ofsuch a structure would drop significantly.

10-10

10-9

10-8

10-7

10-6

10-5

10-4

10-3

80

100

120

140 Calculated

MC = 1000 % MC = 100 % MC = 10 %

SN

R (

dB)

Collector output current (A)

Figure 22. Calculated SNR in a 1 Hz bandwidth based on shotnoise, as a function of collector current and for different values ofthe magnetic sensitivity.

2. Magnetic tunnel transistor

The MTT is derived from the SVT and is so similar thatit is often referred to as a tunnel SVT. The MTT differs inthe structure of the emitter used to inject the hot electronsinto the transistor base. Whereas the SVT uses a Schottkybarrier, the MTT has a tunnel barrier as the emitter. In the firstpublication on the SVT in 1995 by Monsma et al [21], thisalternative device geometry using a tunnel barrier structure asthe emitter was already proposed. In 1997, Mizushima et alwere the first to actually fabricate a MTT [67], and they ownthe patent [68].

The MTT device exists in two slightly different variations,as shown in figure 23. The design on the left has a thininsulating layer to separate the metal base from a non-magneticmetal emitter electrode. When a voltage VEB is appliedacross the tunnel barrier between emitter and base, unpolarizedelectrons are injected by tunnelling into the metal base, arrivingat an energy eVEB above the Fermi level in the metal base. Therest is the same as in the SVT; spin-dependent transmission ofthe hot electrons through the two ferromagnetic thin films inthe metal base, and collection across a Schottky barrier withenergy and momentum selection.

The design on the right in figure 23 also has atunnel barrier, but uses a ferromagnetic emitter electrode.Because the tunnel probability is spin-dependent in such astructure, the injected hot electrons are already spin polarizedas they enter the transistor base. Therefore, the firstpolarizing magnetic layer in the base can now be omittedand only the second analyser magnetic layer needs to beretained. Again, the collector current is determined by spin-dependent transmission of the hot electrons through theferromagnetic base and collection across a Schottky barrierwith energy and momentum selection. Because only theemitter part is different, we can directly apply much of theknowledge obtained from studies of the SVT described insections 1.3–1.8 of section 1. The exponential variation ofthe base transmission, the importance of elastic, inelastic andthermal scattering processes in the layers and at interfaces,and the conditions for transmission into the collector areessentially the same for MTT and SVT. As an example, hot-electron attenuation lengths measured with the MTT [56] are

R302

Topical Review

Figure 23. Energy diagrams of the MTT with a non-magnetictunnel injector (left) or a ferromagnetic tunnel injector (right) togenerate the hot-electron current.

comparable to that previously found for the SVT [51] andmagnetic BEEM [49]. Therefore, we will focus here onlyon those aspects that are different, referring the reader to theliterature [56, 67, 69–74] on the MTT for further information.

A first difference is that vacuum metal bonding, as used forthe SVT, is not required. For the MTT, the base as well as theemitter structure can be created by vacuum deposition onto thecollector semiconductor. For the fabrication of the thin tunnelbarrier, methods well-established for fabricating magnetictunnel junctions can be used [6]. Typically, this involvesdeposition of a 1–2 nm thin Al layer, and subsequent oxidation.The emitter is then deposited on top, and can be magneticallypinned using exchange bias from an antiferromagnet if desired.

A second notable difference is that in the MTT, the energyof the injected hot electrons is given by the voltage VEB

applied across the tunnel barrier. The hot-electron energy isthus tunable over a certain range of energies, typically up to3 eV. This allows one to do spectroscopic studies. Also, aswe have already seen in section 1.7, raising the hot-electronenergy is beneficial for the transfer ratio of the device, sincemore electrons can be transmitted into the larger densityof states higher up in the conduction band of the collectorsemiconductor. For the MTT, this is illustrated in the exampleof figure 24, which shows that above a certain threshold voltageset by the collector Schottky barrier, the transfer ratio growsmore than linearly with VEB. The same behaviour is well-known in BEEM. Thus, in the MTT an increase of VEB resultsin a larger collector current for two reasons, obviously theemitter current is increased, but at the same time the transferratio of the base is enhanced.

A third important aspect is that the MTT with aferromagnetic emitter has only a single magnetic layer inthe base, as opposed to two magnetic layers and the non-magnetic spacer for the SVT (and also for MTT with non-magnetic emitter). Since any base layer and interface is asource of scattering for the hot electrons, we anticipate thatthe transfer ratio of a MTT with magnetic emitter should besignificantly higher. While the MTT was proposed as analternative configuration, it is only after the understandingcreated by experiments with the SVT that it became clearthat the MTT may allow larger output currents to be achieved.A drawback of the MTT with a ferromagnetic emitter is that therelative magnetic response (MC) is limited by the tunnelling

1E-13

1E-12

1E-11

1E-10

1E-9

AP

P

colle

ctor

cur

rent

(A

)

1E-6

1E-5

1E-4

emitt

er c

urre

nt (

A)

-1.25 -1.00 -0.75 -0.50 -0.25 0.00

1E-8

1E-7

1E-6

1E-5

AP

P

tran

sfer

rat

io

emitter voltage (V)

Figure 24. Emitter current, collector current and transfer ratio of aMTT versus emitter voltage VEB applied across the tunnel barrier.Data by O M J van ’t Erve, MESA+, in collaboration withS S P Parkin, IBM Almaden (unpublished).

spin polarization. This issue is discussed in the followingparagraph.

The expressions for the transmission and magnetocurrentof the MTT with non-magnetic tunnel emitter (left designof figure 23) are the same as equations (2)–(7) for theSVT. However, the situation is different for the MTT witha ferromagnetic tunnel emitter, since the emitter current nowbecomes spin polarized, with the polarization determined bythe tunnelling process. For the parallel magnetization state,the emitter tunnel current is a sum of a majority and a minorityspin contribution according to [6]:

I PE ∝ δM

E δMB + δm

E δmB . (10)

Here δME and δm

E are the fraction of majority and minoritytunnelling electrons associated with the emitter/insulatorinterface (label E), and δM

B and δmB are the fraction of tunnel

electrons associated with the insulator/base interface (label B).These are defined in the usual way with help of the tunnellingspin polarization Pt :

Pt,E = δME − δm

E

δME + δm

E

(11)

and

Pt,B = δMB − δm

B

δMB + δm

B

. (12)

R303

Topical Review

To obtain the collector current, we need to multiply eachof the two emitter current components in equation (10) withthe majority or minority spin hot-electron transmission factorappropriate for the ferromagnetic layer in the base (M2 infigure 23). We then have:

I PC ∝ δM

E δMB T M

B + δmE δm

B T mB , (13)

where T M and T m are defined as before in equations (4) and (5).If we assume that the emitter is magnetically pinned, thenfor the antiparallel case the magnetization of the base will bereversed and the resulting IAP

C can be written as:

IAPC ∝ δM

E δmB T m

B + δmE δM

B T MB . (14)

Using these equations, the magnetocurrent can be cast intothe familiar form:

MC = I PC − IAP

C

IAPC

= 2Pt,EP ∗B

1 − Pt,EP ∗B

, (15)

where we have defined a renormalized polarization P ∗B of the

base as:

P ∗B = δM

B T MB − δm

B T mB

δMB T M

B + δmB T m

B

. (16)

Note that this renormalized polarization is a combinationof tunnelling factors and hot-electron transmission factors, andas such should not be confused with the regular tunnellingspin polarization as defined in equation (12). Thus, wesee that the MC of a MTT with a ferromagnetic tunnelinjector is determined by two completely unrelated physicalprocesses. The polarization of the emitter current is due tospin-dependent tunnelling, while the subsequent transmissionof the hot electrons in the base of the MTT is given by thespin-dependent scattering of hot electrons.

The fact that the MC depends on Pt,E has an importantconsequence, because the tunnel polarization Pt,E of theemitter/insulator interface is basically fixed, i.e. not dependenton the thickness of the emitter electrode. Rather, it isdetermined by the choice of metal and tunnel insulator [6]. Inpractice, this means Pt,E is limited to about 50% for typicalferromagnetic materials and Al2O3 tunnel barriers. Usingthe expressions given above it is straightforward to calculatethe expected MC for a given emitter tunnel polarization (seefigure 25, in which hot-electron transmission parameters fora Co base ferromagnet (λM = 21 Å and λm = 8 Å) wereused). As can be seen, the MC grows with increasing thicknessof the base ferromagnetic layer as the base becomes a betterspin ‘analyser’. However, at large base thickness essentiallyno minority spins are transmitted, and the MC saturates ata value that depends on the tunnel spin polarization. For themaximum tunnel polarization of 50%, the MC cannot be largerthan 200%. This constitutes a fundamental difference with theSVT, where the MC can be made arbitrarily large by increasingthe thickness of both the ferromagnetic layers of the base.

An interesting positive consequence is that the MTT can beused to probe the tunnel spin polarization at large bias. Recallthat the MTT typically operates at a bias voltage of 1 V orlarger. At such a voltage, the regular TMR of a ferromagnetictunnel junction is usually negligibly small [6]. However,experimental MTTs with ferromagnetic emitter clearly display

0 20 40 60 80 1000

50

100

150

200

10%

20%

30%

40%

Pt,E

= 50%

Cal

cula

ted

MC

(%

)

Thickness of FM layer in base (Å)

Figure 25. Calculated magnetic response of an MTT withferromagnetic emitter, for different values of the tunnel spinpolarization Pt,E of the emitter.

-200 -100 0 100 2000

1

2

3

4

5

AP

P

MC = 115 %

VEB

= 1 V

magnetic field (Oe)

colle

ctor

cur

rent

(nA

)

0

10

20

30

40

AP

P

TMR = 3 %

TMR = 40 %

VEB

= 1 V

VEB

= 10 mVtu

nnel

res

ista

nce

(kO

hm)

Figure 26. Bottom panel: collector current versus magnetic field fora Co84Fe16/Al2O3/Co(10 nm)/Si MTT at 1 V bias and T = 77 K.The emitter was pinned by Ir22Mn78. Top panel: the TMR of theemitter-base tunnel junction, measured at 10 mV and 1 V bias,respectively. Data by O M J van ’t Erve, MESA+, in collaborationwith S S P Parkin, IBM Almaden (unpublished).

large magnetocurrent. An example is shown in figure 26for a Co84Fe16/Al2O3/Co(10 nm)/Si transistor, fabricated bysputtering through shadow masks. A large difference incollector current between P and AP states is clearly observed,corresponding to an MC of about 115%. This demonstratesthat highly spin polarized electrons are injected into the base bytunnelling from the Co84Fe16 emitter at an applied bias of 1.0 V.The corresponding tunnel spin polarization is about 36%. Notethat the TMR for the same junction, measured between baseand emitter, is only 3% at 1.0 V, while a TMR of about 40% was

R304

Topical Review

measured at a bias of 10 mV. This unambiguously proves thatthe disappearance of TMR at high bias does not necessarilyimply that the tunnel current is no longer spin polarized.

3. Hot-electron spin injection

One of the crucial ingredients of semiconductor based spinelectronics is the ability to inject a highly spin-polarizedcurrent into a semiconductor. While electrical spin injectionappears conceptually simple, it has turned out to be a majorexperimental challenge. One successful approach is to usea paramagnetic II–VI semiconductor (BexMnyZn1−x−ySe) asspin aligner, yielding injection of highly polarized (90%)carriers [75]. The application potential of this method ishowever hampered by the large magnetic fields required toinduce a sizeable spin splitting in the paramagnetic injector.A second successful demonstration involved the injection froma ferromagnetic semiconductor GaxMn1−xAs into GaAs [76].Both approaches are limited to low temperature, although aroom temperature magnetic semiconductor may change theprospects of this method. Nevertheless, the carrier spinpolarization in any ferromagnet decays with temperature,especially when the Curie temperature is approached, such thatthe use of conventional ferromagnets with Curie temperaturefar exceeding room temperature is still the most promisingroute.

It has been argued [77] that direct spin injection from aferromagnetic metal contact into a semiconductor, via diffusivetransport, is nearly impossible due to the large mismatchof resistance between metal and semiconductor. This is inagreement with the absence of any sizeable magnetotransporteffects in experiments ([77] and references therein). It wassuggested [77–82] that the problem may be overcome eitherby using completely spin-polarized contacts (e.g. half-metallicferromagnets), by using ballistic transport, or by introducingsome kind of spin-dependent ‘interface resistance’. Suchresistance can naturally exist at the metal/semiconductorinterface due to differences in the electronic states, as shownfor the case of Fe/InAs [83]. Alternatively, a tunnel barriercan be incorporated. The latter was invoked to explain theobserved electrical injection of 2% spin-polarization from Feinto GaAs at room temperature [16]. More recently, tunnellinginjection of spins across a Fe/AlGaAs Schottky contact wasreported [17, 18], where a spin-polarization of 32% in theburied GaAs quantum well was obtained at low temperature.Spin injection across a thin AlOx tunnel barrier, from CoFeinto an AlGaAs/GaAs semiconductor was also demonstrated[19, 20], the spin polarization being in excess of 11% at roomtemperature. Note that evidence for room temperature spininjection into GaAs across a vacuum tunnel barrier, from aNi tip in a scanning tunnelling microscope, was reported in1992 [84].

The work on hot-electron spin-transport in the SVT hasled us to consider an alternative method to achieve controllablespin injection into semiconductors. The approach is based onspin-filtering of a current of hot electrons during transmissionthrough a thin film of ferromagnetic metal, just prior toinjection into the semiconductor. Since transport is based onnon-equilibrium carriers, the conductivity mismatch does not

normal metal ferromagnet tunnel barrier

V

EF hot electrons

semi-conductor

normal metal ferromagnet Schottky barrier

hot electrons

semi-conductor

Figure 27. Energy diagrams of HESF contacts with a Schottkybarrier (left) or a tunnel barrier (right) to generate the hot-electroncurrent. A thin ferromagnetic film acts as spin filter.

play a role, and injection can be achieved with conventionalferromagnetic metals.

Figure 27 shows possible arrangements of the hot-electron spin filter (HESF) contacts. First, an unpolarizedhot-electron current is generated using a Schottky barrier(left) or tunnel barrier (right). The hot-electron energy isdetermined by the Schottky barrier height (0.5–1 eV) or by thevoltage (1–3 eV) applied across the tunnel barrier, respectively.Second, the hot-electron current is passed through a thinferromagnetic metal film that acts as a polarizer for the hot-electron spins. The final step is the actual injection of thespin-polarized hot electrons into the semiconductor across aSchottky barrier. In this contact structure, the ferromagneticfilm is generally (but not necessarily) separated from thesemiconductor by a non-magnetic thin film, for example, toavoid magnetically dead layers or to tailor the Schottky barrierproperties.

We emphasize that the conductivity mismatch [77] doesnot apply to hot-electron transport, as it is a non-equilibriumsituation which cannot be described in terms of a series ofresistances. Rather, when a current of hot electrons is injectedinto the ferromagnetic thin film, the energy and momentumrelaxation of the hot electrons results in an exponential decayof the number of hot electrons with distance in the ferromagnet.Subsequent transmission into the semiconductor, across theSchottky barrier, occurs without any electric field or imbalancein electrochemical potential across the metal-semiconductorinterface. In other words, the electron flow across the interfaceis due to the kinetic energy of the hot electrons, and generallyno bias voltage is applied between the ferromagnet and thereceiving semiconductor. This applies not only when transportis ballistic, but even if the hot-electron transport is diffusive,i.e. dominated by elastic scattering.

The two most important characteristics of the contact arethe spin polarization of the injected electrons, and the fractionof the hot electrons that arrive in the semiconductor. We mayassume that the spin polarization of the hot electrons is notaffected by the transport across the non-magnetic metal filmand its interface with the receiving semiconductor. This isreasonable since only non-magnetic materials are involved,and depolarization occurs via the weak spin-orbit scattering.It is then sufficient to determine the spin polarization afterspin filtering by the ferromagnetic film, which has been donefor a number of materials using the SVT and MTT. Note

R305

Topical Review

0

20

40

60

80

100

Co

Ni80

Fe20

T = 100 K

spin

filt

er p

olar

izat

ion

(%)

0 20 40 60 80 100

0.01

0.1

1

Co

Ni80

Fe20

tota

l tra

nsm

issi

on

100 K 300 K

FM thickness (Å)

Figure 28. Spin polarization (top panel) and total transmission(bottom) of a hot-electron current versus thickness of theferromagnetic film (Co or Ni80Fe20) that acts as spin filter.

that the large measured ratio between I PC and IAP

C for theSVT: (i) shows that hot-electron spin filtering is extremelyefficient, and (ii) proves that the spin polarization does notdecay significantly during transport through the non-magneticAu spacer between the Ni80Fe20 and Co layer.

Analysis of data such as shown in section 1.4 usingequations (2)–(7) allows the determination ofλM

fm andλmfm, from

which the polarization of the hot electrons after spin-filteringcan be evaluated. The resulting polarization as a functionof film thickness for Ni80Fe20 or Co are shown in figure 28(top panel). The spin polarization increases with increasingthickness, and rapidly approaches unity (P = 96% at 5 nm).Thus, almost complete spin-polarization is achieved forrelatively thin films. The total transmission of a ferromagneticfilm (bottom panel) is at an acceptable level, close to 0.1 forfilms of a few nanometres (the attenuation due to the interfaceswas also taken into account).

For practical devices based on spin injection theirproperties at room temperature are relevant. Measurementssuch as those in section 1.6 at variable temperature showthat hot-electron attenuation in ferromagnets is somewhatenhanced at room temperature due to scattering by thermalspin waves [51]. However, as can be seen in figure 28 (bottompanel), the net reduction of the transmission is relatively weak,especially for materials with a high Curie temperature suchas Co. Besides the slight reduction of the transmission,thermal spin waves also produce mixing of the two spinchannels [23]. This implies that the spin filter polarization alsodecays at higher temperature. Measurements of the MC versustemperature allow a direct quantification via equation (7).The data for a 6 nm Ni80Fe20 film in figure 29 shows thatP decays from 98% at low temperature, to 85% at roomtemperature. For Co, the decay of P is negligible. Thus,

0 100 200 3000

20

40

60

80

100

P = 85 %

P = 98 %

6 nm Ni80

Fe20

temperature (K)

spin

filt

er p

olar

izat

ion

(%)

Figure 29. Temperature dependence of the spin polarization of ahot-electron current after spin filtering by 6 nm Ni80Fe20.

even at room temperature, a HESF allows the creation of ahighly spin-polarized current, with conventional ferromagneticmetals that can be manipulated with small magnetic fields.Several research groups are now working to directly measurethe spin polarization of the hot electrons after injectioninto the semiconductor collector, using electroluminescencetechniques [16–20, 75, 76]. This will show whether theassumption of negligible spin-depolarization of hot electronsat the metal/semiconductor interface is correct, and whetherthe high energy of the electrons gives rise to additional spinrelaxation mechanisms in the semiconductor.

4. Outlook

Perhaps the most influential aspect of the SVT is that itshowed the value of functional integration of semiconductorand ferromagnetic materials into hybrid electronic devices.Room temperature effects as large as 400% in small magneticfields are now routinely obtained, a feature that has attractedsignificant attention. Much progress has been made inoptimizing device performance, focusing on the output currentlevel and noise sources. Further improvements are stillrequired in order to capitalize on the huge magnetic sensitivityof such structures for application in magnetic field sensors ormagnetic memories. In that context, the properties of SVT andMTT structures when scaled to sub-micron dimensions will berelevant.

At the same time, the SVT opened up a new routeto systematically study the fundamental physics of spin-dependent transport of hot electrons at energies of the orderof 1 eV. New insights into the origin of spin-dependenthot-electron scattering have been obtained, including thedominance of volume effects over interfaces in the spin-dependence of the transmission, the effect of thermal spinwaves, and the surprisingly important role of elastic scatteringprocesses. The latter feature points to the relevance ofestablishing the precise relation between scattering andstructural properties of the devices, an area that has stillremained largely unexplored. It also hints at an interestingroute to the next generation of devices in which epitaxialgrowth techniques will be used to obtain transistors with highstructural quality.

The basic concept of the SVT has also led to thedevelopment of a number of related device structures such as

R306

Topical Review

the MTT. The tunability of the hot-electron energy providesunique possibilities for spectroscopic studies, while enhancedbase transmission is anticipated. Moreover, it has been realizedthat hot-electron spin filtering may have some attractivefeatures for spin-injection into a semiconductor, in particularthe ability to reach a spin polarization near 100% withconventional ferromagnets. Work that is in progress will soonprovide more information on the feasibility of this approach.Another interesting research avenue we are currently exploringis the study of hot-electron spin-transport in novel materialssuch as half-metallic ferromagnets and oxides. Such materialsmay also offer new types of hybrid electronic devicescombining ferromagnets and semiconductors.

Acknowledgments

The author is grateful to numerous colleagues for theircontribution to the research described in this topicalreview. We acknowledge financial support from the RoyalNetherlands Academy of Arts and Sciences (KNAW), theDutch Technology Foundation (STW), the Dutch Foundationfor Fundamental Research on Matter (FOM), and the EuropeanCommission.

References

[1] Baibich M N, Broto J M, Fert A, Nguyen Van Dau F, Petroff F,Etienne P, Creuzet G, Friederich A and Chazelas J 1988Phys. Rev. Lett. 61 2472

[2] Prinz G A 1998 Science 282 1660[3] Wolf S A, Awschalom D D, Buhrman R A, Daughton J M,

von Molnar S, Roukes M L, Chtchelkanova A Y andTreger D M 2001 Science 294 1488

[4] Parkin S S P et al 2003 Proc. IEEE 91 661[5] Moodera J S, Kinder L R, Wong T M and Meservey R 1995

Phys. Rev. Lett. 74 3273[6] Moodera J S and Mathon G 1999 J. Magn. Magn. Mater.

200 248[7] Parkin S S P et al 1999 J. Appl. Phys. 85 5828[8] de Boeck J and Borghs G 1999 Phys. World 12 27[9] Tehrani S et al 2003 Proc. IEEE 91 714

[10] Kikkawa J M and Awschalom D D 1998 Phys. Rev. Lett. 804313

[11] Kikkawa J M and Awschalom D D 1999 Nature 397 139[12] Ohno H 1998 Science 281 951[13] Ohno H, Chiba D, Matsukura F, Omiya T, Abe E, Dietl T,

Ohno Y and Ohtani K 2000 Nature 408 944[14] Park Y D, Hanbicki A T, Erwin S C, Hellberg C S, Sullivan M,

Mattson J E, Ambrose T F, Wilson A, Spanos G andJonker B T 2002 Science 295 651

[15] Datta S and Das B 1990 Appl. Phys. Lett. 56 665[16] Zhu H J, Ramsteiner M, Kostial H, Wassermeier M,

Schonherr H-P and Ploog K H 2001 Phys. Rev. Lett. 87016601

[17] Hanbicki A T, Jonker B T, Itskos G, Kioseoglou G andPetrou A 2002 Appl. Phys. Lett. 80 1240

[18] Hanbicki A T, van ’t Erve O M J, Magno R, Kioseoglou G,Li C H, Jonker B T, Itskos G, Mallory R, Yasar M andPetrou A 2003 Appl. Phys. Lett. 82 4092

[19] Motsnyi V F, De Boeck J, Das J, Van Roy W, Borghs G,Goovaerts E and Safarov V I 2002 Appl. Phys. Lett. 81 265

[20] Van Dorpe P, Motsnyi V F, Nijboer M, Goovaerts E,Safarov V I, Das J, Van Roy W, Borghs G and De Boeck J2003 Jpn. J. Appl. Phys. 42 L520

[21] Monsma D J, Lodder J C, Popma Th J A and Dieny B 1995Phys. Rev. Lett. 74 5260

[22] Anil Kumar P S, Jansen R, van ’t Erve O M J, Vlutters R,de Haan P and Lodder J C 2000 J. Magn. Magn. Mater.214 L1

[23] Jansen R, Anil Kumar P S, van ’t Erve O M J, Vlutters R,de Haan P and Lodder J C 2000 Phys. Rev. Lett. 853277

[24] Monsma D J, Vlutters R and Lodder J C 1998 Science 281 407[25] Sze S M 1981 Physics of Semiconductor Devices 2nd edn

(New York: Wiley)[26] Anil Kumar P S, Jansen R, van ’t Erve O M J, Vlutters R,

Kim S D and Lodder J C 2001 Physica C 350 166[27] Sze S M, Crowell C R, Carey G P and LaBate E E 1966

J. Appl. Phys. 37 2690[28] Crowell C R and Sze S M 1966 J. Appl. Phys. 37 2683[29] Sze S M 1969 Physics of Semiconductor Devices 1st edn

(New York: Wiley)[30] Shimatsu T, Mollema R H, Monsma D J, Keim E G and

Lodder J C 1998 J. Vac. Sci. Technol. A 16 2125[31] Lodder J C, Monsma D J, Vlutters R and Shimatsu T 1999

J. Magn. Magn. Mater. 198–199 119[32] Dessein K, Boeve H, Anil Kumar P S, De Boeck J, Lodder J C,

Delaey L and Borghs G 2000 J. Appl. Phys. 87 5155[33] Kim S D, van ’t Erve O M J, Jansen R, Anil Kumar P S,

Vlutters R and Lodder J C 2001 Sensors Actuators A 91 166[34] Kim S D, van ’t Erve O M J, Vlutters R, Jansen R and

Lodder J C 2002 IEEE Trans. Electron Devices 49 847[35] Mott N F 1964 Adv. Phys. 13 325[36] Pierce D T and Siegmann H C 1974 Phys. Rev. B 9 4035[37] Pappas D P, Kamper K-P, Miller B P, Hopster H, Fowler D E,

Brundle C R, Luntz A C and Shen Z-X 1991 Phys. Rev.Lett. 66 504

[38] Grobli J C, Guarisco D, Frank S and Meier F 1995 Phys. Rev.B 51 2945

[39] Grobli J C, Oberli D and Meier F 1995 Phys. Rev. B 52R13095

[40] Schonhense G and Siegmann H C 1993 Ann. Phys. (Leipzig)2 465

[41] Drouhin H-J, Lampel G, Lassailly Y, van der Sluijs A J andMarliere C 1995 J. Magn. Magn. Mater. 151 417

[42] Drouhin H-J, van der Sluijs A J, Lassailly Y and Lampel G1996 J. Appl. Phys. 79 4734

[43] Oberli D, Burgermeister R, Riesen S, Weber W andSiegmann H C 1998 Phys. Rev. Lett. 81 4228

[44] Weber W, Riesen S and Siegmann H C 2001 Science 291 1015[45] Aeschlimann M, Bauer M, Pawlik S, Weber W,

Burgermeister R, Oberli D and Siegmann H C 1997 Phys.Rev. Lett. 79 5158

[46] Celotta R J, Unguris J and Pierce D T 1994 J. Appl. Phys. 756452

[47] Bell L D and Kaiser W J 1988 Phys. Rev. Lett. 61 2368[48] Rippard W H and Buhrman R A 1999 Appl. Phys. Lett. 75 1001[49] Rippard W H and Buhrman R A 2000 Phys. Rev. Lett. 84 971[50] Vlutters R, Jansen R, van ’t Erve O M J, Kim S D, Lodder J C,

Vedyayev A and Dieny B 2002 Phys. Rev. B 65 024416[51] Vlutters R, van ’t Erve O M J, Kim S D, Jansen R and

Lodder J C 2002 Phys. Rev. Lett. 88 027202[52] Vlutters R, Jansen R, van ’t Erve O M J, Kim S D and

Lodder J C 2001 J. Appl. Phys. 89 7305[53] Weilmeier M K, Rippard W H and Buhrman R A 1999 Phys.

Rev. B 59 R2521[54] Bell L D 1996 Phys. Rev. Lett. 77 3893[55] Knorren R, Bennemann K H, Burgermeister R and

Aeschlimann M 2000 Phys. Rev. B 61 9427[56] van Dijken S, Jiang X and Parkin S S P 2002 Phys. Rev. B 66

094417[57] van ’t Erve O M J, Vlutters R, Anil Kumar P S, Kim S D,

Postma F M, Jansen R and Lodder J C 2002 Appl. Phys.Lett. 80 3787

[58] Jansen R, Kim S D, Vlutters R, van ’t Erve O M J andLodder J C 2001 Phys. Rev. Lett. 87 166601

[59] Plihal M, Mills D L and Kirschner J 1999 Phys. Rev. Lett. 822579

R307

Topical Review

[60] Hong J and Mills D L 1999 Phys. Rev. B 59 13840[61] Hong J and Mills D L 2000 Phys. Rev. B 62 5589[62] Dieny B, Speriosu V S and Metin S 1991 Europhys. Lett.

15 227Dieny B, Humbert P, Speriosu V S, Metin S, Gurney B A,

Baumgart P and Lefakis H 1992 Phys. Rev. B 45 806[63] Fert A, Duvail J L and Valet T 1995 Phys. Rev. B 52 6513[64] Niedermann P, Quattropani L, Solt K, Maggio-Aprile I and

Fischer O 1993 Phys. Rev. B 48 8833[65] van Dijken S, Jiang X and Parkin S S P 2003 Appl. Phys. Lett.

83 951[66] van ’t Erve O M J, Anil Kumar P S, Jansen R, Kim S D,

Vlutters R, Lodder J C, Smits A A and de Jonge W J M2001 Sensors Actuators A 91 192

[67] Mizushima K, Kinno T, Tanaka K and Yamauchi T 1997 IEEETrans. Magn. 33 3500

[68] Mizushima K, Kinno T, Yamauchi T and Inomata K 1996US Patent No US5747859

Mizushima K, Kinno T, Yamauchi T and Inomata K 1998US Patent No US5973334

[69] Mizushima K, Kinno T, Tanaka K and Yamauchi T 1998 Phys.Rev. B 58 4660

[70] Yamauchi T and Mizushima K 1998 Phys. Rev. B 58 1934Yamauchi T and Mizushima K 2000 Phys. Rev. B 61

8242

[71] Kinno T, Tanaka K and Mizushima K 1997 Phys. Rev. B 56R4391

[72] Sato R and Mizushima K 2001 Appl. Phys. Lett. 79 1157[73] van Dijken S, Jiang X and Parkin S S P 2002 Appl. Phys. Lett.

80 3364[74] van Dijken S, Jiang X and Parkin S S P 2003 Appl. Phys. Lett.

82 775[75] Fiederling R, Keim M, Reuscher G, Ossau W, Schmidt G,

Waag A and Molenkamp L W 1999 Nature (London)402 787

[76] Ohno Y, Young D K, Beschoten B, Matsukura F, Ohno H andAwschalom D D 1999 Nature (London) 402 790

[77] Schmidt G, Ferrand D, Molenkamp L W, Filip A T andvan Wees B J 2000 Phys. Rev. B 62 R4790

[78] Filip A T, Hoving B H, Jedema F J, van Wees B J, Dutta B andBorghs S 2000 Phys. Rev. B 62 9996

[79] Rashba E I 2000 Phys. Rev. B 62 R16267[80] Grundler D 2001 Phys. Rev. B 63 R161307

Grundler D 2001 Phys. Rev. Lett. 86 1058[81] Kirczenow G 2001 Phys. Rev. B 63 054422[82] Fert A and Jaffres H 2001 Phys. Rev. B 64 184420[83] Zwierzycki M, Xia K, Kelly P J, Bauer G E W and Turek I

2003 Phys. Rev. B 67 092401[84] Alvarado S F and Renaud P 1992 Phys. Rev. Lett. 68 1387

Alvarado S F 1995 Phys. Rev. Lett. 75 513

R308