~ Journal of magnellsm

, ~ and magnetic

,444 materials ELSEVIER Journal of Magnetism and Magnetic Materials 175 (1997) 1-9

The art of spl'n electron+cs

John Gregg a'*, Will Allen a, Nathalie Viart a, Randall Kirschman a, Chitnarong Sirisathitkul a, Jean-Philippe Schille a, Mathias Gester b, Sarah Thompson b,

Patti Sparks c, Victor Da Costa d, Kamel Ounadjela d, Mike Skvarla e Clarendon Laboratory, Parks Road, OxJbrd OXI 3PU, UK

bDepartment of Physics, University of York, Heslington, York YOI 5DD, UK CHarvey Mudd College, Claremont, CA 9171l, USA

d IPCMS/GEMME, 67037 Strasbourg, France e Cornell Nanofabrication Facility, Cornell University, Knight Lab., lthaca, NY 14853, USA

Abstract

A brief history is given of the nascent field of spin electronics in which the ability to differentially manipulate up- and down-spin current carriers is exploited. We discuss the impending marriage of spin-dependent effects with semiconductor technology and, in particular, the exploitation of spin-dependent transport in the semiconductors themselves. In this connection, preliminary experiments are described which explore spin transport in ion-implanted silicon. We conclude by evaluating various potential applications of the devices made possible by this exciting new development in electronic technology.

Keywords: Spin electronic; Spin accumulation; Spin transistor; SPICE; Spin diffusion length; Magnetic field sensor

Hitherto, conventional electronics has ignored the spin of the electron. Indeed, with the exception of Hall sensors, solenoids, relays and the occasional specialised microwave device, magnetism has tradi- tionally been the 'poor relation' in the world of electronic circuitry. In the everyday silicon transis- tor, the different families of electrical carrier are distinguished by their different effective charges: however, no practical use is made of the fact that some are spin-up and others are spin-down.

*Corresponding author. Tel.: +44 1865 272 31 l; fax: +44 1865 272 400; e-mail: gregg@ermine.ox.ac.uk.

The recognition of this distinction is the key which promises to unlock a whole new generation of spin electronic devices [1] whose operation relies on differential manipulation of independent fami- lies of current carriers with opposite spin polarisa- tion.

The technical basis for this rests on Mott's obser- vation [2] that electrical conduction in bulk fer- romagnets may be aptly modelled by supposing that the current is carried by two distinct 'spin channels', one consisting of electrons with spin parallel to the magnetisation and the other with spins antiparallel. The relative independence of these two families of carriers arises since, although

0304-8853/97/$17.00 1997 Elsevier Science B.V. All rights reserved PII S 0 3 0 4 - 8 8 53197)00 1 5 5-8

2 J. Gregg et al./Journal of Magnetism andMagnetic Materials 175 (1997) 1 9

spin flip processes are possible which move elec- trons from one channel to the other, they occur on a time scale which is long compared with those of other processes. This two channel model was fur- ther probed and substantiated [3, 4] by measuring the electrical transport of selectively doped fer- romagnets. This confirmed the validity of a crude equivalent circuit for the ferromagnet consisting of two parallel resistors, one large and the other small, thus reflecting the fact that electrons of one-spin polarisation are more heavily scattered than the others. This difference in scattering times for elec-

trons in the two channels is at the heart of the operation of spin electronic devices.

The first spin electronic components were pas- sive two-terminal devices (Fig. 1), namely the giant magnetoresistive multilayers studied independently by the groups of Fert and Griinberg in 1988 [5, 6]. In their simplest form, they consist of magnetic trilayers in which a layer of nonmagnetic metal is sandwiched between two layers of ferromagnetic metal. The structure presents a simple ohmic resis- tance which varies depending on the relative ori- entations of the magnetisations in the two layers.

(a)

Fig. 1

i iiii!!i !!iiii

(b)

N k = vv "t%

f f - N k = lsd

N steps

So Spin Diffusion Length = l sa =V Vv T~+~,

(

Circuit 7 ,,

Fig. 2 Fig. 3

Fig. 1. The giant magnetoresistor. Plots of the Y (pink) and ~ (blue) electrochemical potentials in (a) ferro-aligned and (b) antiferro- aligned trilayers. The separation of the curves is proportional to the spin accumulation. The effective interface resistances are shown by the discontinuous green lines: they have different values for ferromagnetic and antiferromagnetic alignment provided that the spin accumulation from adjacent interfaces overlap.

Fig, 2. The progress of a drunken spin.

Fig. 3. The Johnson spin transistor.

,Z Gregg et al. / Journal of Magnetism and Magnetic Materials 175 (1997) l 9 3

Although they were initially characterised by pas- sing current in the plane of the device (CIP), it is perhaps more instructive to consider the case where the current is oriented perpendicular to the film plane (the CPP configuration) when a similar vari- ation in device resistance is observed. On applica- tion of the driving voltage, there is differential displacement of the spin-up and spin-down Fermi surfaces in the ferromagnetic metal layers; the majority spin surface (which has the longer momentum relaxation time) has the greater dis- placement. The result is that, associated with the electron current which is supplied to the intermedi- ate nonmagnetic layer, is a net flow of magneti- sation which diffuses a certain distance before decaying owing to spin flipping of the electrons which carry it. If this so-called 'spin accumulation' [7] extends across the sandwich layer to reach the other ferromagnetic interface before decaying, then the resistance of the whole device is sensitive to the magnetic configuration of the layers. Parallel align- ment gives low resistance and antiparallel align- ment high resistance. To view the problem in a different way, magnetic information is encoded onto the carrier population by the first ferromag- netic layer (the polariser) and transferred to the second ferromagnetic layer (the analyser) whose magnetic orientation determines how the informa- tion gets processed.

The behaviour of this simple giant magnetore- sistor illustrates an important basic point: spin electronic devices work by transferring spin in- formation from one part of the device to another. This information is mediated by the electrical car- riers and it decays on a characteristic length scale (the spin diffusion length) which is the average distance diffused by a carrier spin before flipping. An essential criterion for the creation of spin elec- tronics is, thus, the ability to engineer structures whose physical dimensions are of this order or smaller. With the advent of modern thin-film and nanofabrication technologies, this dream is now a reality.

A simple understanding of the physics underly- ing this length scale may be gleaned from the fol- lowing consideration. A spin-polarised electron injected into a paramagnetic medium undergoes many collisions which modify its momentum before

it eventually spin-flips (Fig. 2). By analogy with the progress of a drunken sailor after closing-time, the lateral displacement of the spin is roughly

Isd = x ~ 2, if it experiences N momentum scatter- ing events each with characteristic length 2. The actual path travelled by the spin is N2 = VFZ~ (where vv is the Fermi velocity and z~+ is the spin- relaxation time) from which it follows that

l~d = )x/~vv~T~. This quantity, thus, depends both on the mean free path and on the spin-flip time and is wildly sensitive to the presence of impurities. For example, it reduces from about 1 ~tm for pure silver to roughly 100 ~, for silver with 3% of added gold impurity [8]. Evidently, materials composition and purity is another crucial engineering parameter in the realisation of spin electronics.

So far, we have considered only the operation of a two-terminal spin electronic device in which wires are attached to the two outer ferromagnetic layers of a trilayer sandwich. The logical continuation is to create a three-terminal device by making an additional contact to the central layer [-9]. This, the first all-metal spin transistor, works by pumping current from ferromagnetic emitter to paramag- netic base to create a spin accumulation in the base layer (Fig. 3). A spin accumulation is simply an excess of number density of up-spins over down- spins and can, thus, be viewed as a divergence in the electrochemical potentials of the two spin channels. So the potential at which the (ferromagnetic) collec- tor sits is dependent on its magnetic orientation since this essentially determines which electro- chemical potential it is sampling.

Despite being a three-terminal device, the metal- spin transistor, unfortunately, affords no power gain. However, by the expedient of sandwiching a metal giant magnetoresistor between two Schottky barriers, a different type of three-terminal device may be constructed which, at least in prin- ciple, overcomes this objection [10]. The silicon layers which form the top and bottom of the device function as emitter and collector, respectively, while the magnetoresistor is now the base (Fig. 4). When correctly biased, the emitter/base junction injects hot carriers into the base magnetic multilayer which lose energy in transit due to scattering. Only a fraction of these hot carriers arrive at the

4 J. Gregg et al./rjourrtal of Magnetism and Magnetic Materials 175 (1997) 1 9

Si GMR Multilayer I Si

I Emitter Base Collector

t '1 I ,i I

Sehottky Barriers

Fig. 4 Fig. 5

COI

BASE / -

EMITTER

Magnetic Analyser

Magnetic - - Polarizer

0 35.0 w~ 0 35 .0 h t ; tslPe Heltiht |~ t~ L v . 6~ , l l t~ l l~ Z rant~ 3(]0 r,~ t ra~t~ 1,53 na

Fig. 6 Fig. 4. The M o n s m a transistor. Represents the first at tempt to integrate ferromagnetic metals with silicon to produce a spin electronic device, lts performance derives from the ablity to magnetically tune the hot spin energy relaxation rate in the multilayer metal base region.

Fig. 5. The SPICE transistor. Current injected into the base region is spin polarised by the emitter. The analyser acts as a guard rail to regulate incidence of the base diffusion current drop. By varying the relative magnetic orientation of polariser and analyser the current gain of the device may be magnetically modulated.

Fig. 6, A minor-scale hybrid cobalt silicon structure used to test the spin transport in ion-implanted silicon. The left-hand scan shows the topography of the cobalt islands which are grown on an intrinsic silicon wafer. The right-hand scan illustrates the low-field domain structure.

J. Gregg et al. / Journal of Magnetism and Magnetic Materials 175 (1997) 1 9 5

base/collector interface with sufficient energy to clear the second Schottky barrier and continue into the collector silicon layer. Changing the magnetic configuration of the base multilayer changes the scattering length for energy relaxation and, hence, the fraction of the emitter current which arrives at the collector. The result is a three-terminal device not unlike the conventional bipolar transistor in behaviour in which the current gain, hfe, is a func- tion of externally applied magnetic field. The actual value of hfe is less than unity, so, once again, realis- ing positive power gain is difficult, but, even so, the device promises to make an excellent magnetic field sensor. Variations in hf, of 100% have been ob- served in fields of order millitesla.

The advent of combining silicon and ferromag- netic metals in a single-spin electronic device is a highly significant step forward for two reasons. Firstly, unlike metals, the implemantation and use of depletion layers in doped semiconductor is prac- tical. Moreover, semiconducto~metal junctions offer non-ohmic characteristics which may be ex- ploited for rectification and for absorbing the de- vice bias voltage, thereby enabling regions of the device to be electric-field free. Furthermore, and perhaps more importantly, this voltage shielding property of such junctions together with the dual carrier nature of the semiconductor allows the use of diffusion currents for electrical transport, there- by enabling different classes of spin electronic de- vice to be made. With a view to the ultimate integration of spin electronics with conventional semiconductor electronics, silicon is probably the most judicious choice of semiconductor to develop for spin electronic applications on the grounds that it is the material into which most research and development funding has so far been invested.

Thus far, silicon has been conventionally em- ployed in spin electronics. The logical progression is to exploit the behaviour of spin-polarised cur- rents in the silicon itself. Thanks to the support of EPSRC and the European Commission, work is in progress on novel three- and four-terminal spin electronic devices which attempt to do this. Hybrid designs using semiconductor/ferromagnetic metals are being developed with a view to achieving en- hanced gain, signal to noise and field sensitivity. One such device under joint development by the

Oxford Spin Electronics Group and Sarah Thomp- son's research team in York is the SPICE (Spin Polarised Injection Current Emitter) Transistor (Fig. 5). This design uses the concept of a spin- polarised diffusion current to realise a performance which combines the magnetic sensitivity of the Monsma transistor with the high current gain of a conventional bipolar transistor.

In the design and fabrication of such devices, a number of important fundamental questions arise. In particular, comparatively little is known about the spin transport properties of silicon, espe- cially the spin-flip cross-sections of conventional n- and p-silicon dopants. A series of spin transport experiments is being conducted by the Oxford and York groups into the spin transport of ion-im- planted silicon. The low-field magnetic structure of a precursor SPICE emitter structure is illustrated [11] (Fig, 6) in which micron-scale cobalt elements inject polarised currents into n-doped silicon across oxide tunnelling barriers. The evolution of the polarisation across several microns of doped silicon is monitored by similar ferromagnetic analysers with different shape anisotropies to the polarisers, and the transport of the structure is measured as a function of domain structure by applying a mag- netic field in the injector/analysers (Fig. 7). The results of this study are extremely promising: in particular, they demonstrate that spin information transfer across bulk-doped silicon is possible and that conventional silicon dopants are compatible with spin diffusion lengths of the order of microns. Moreover, ion implantation doping seems to be a viable option for spin electronics fabrication and it is possible to implement ferromagnetic/ silicon interface structures with acceptable spin de- polarisation levels for spin electronic device ap- plication.

The study of new potential device materials, with particular focus on oxides for use as spin injectors, is being headed by the Oxide Spin Electronics Net- work (OXSEN), an EC funded project coordinated by Professor Michael Coey (Trinity College Dub- lin) (Fig. 8). The pursuit of novel device structures and implementations is the focus of the EC-funded Brite Euram project on Hot-Spin Electronic Active Magnetic Sensors (HotSEAMS) coordinated by John Gregg (Oxford). The EC has also funded

6 J. Gregg et al. /Journal of Magnetism and Magnetic Materials 175 (1997) 1- 9

%"

-1500

Fig. 7

I I I I 1

-1000 -500 0 500 1000

Field (mT)

1500

Twente Groningen

Thomson CSF I CENG I r e n o b ~ 3 h e f f i e l d

Siemens I Orsay ] Jena

Strasbourg Barcelona

HotSEAMS Hot Spin Electronic Active Magnetic Sensors Coordinator : J.F. Gregg, Oxford

Fig. 8

OXSEN Oxide Spin Electronics Network Coordinator : J .MD. Coey, TCD

Fig. 7. Spin transport measurements in doped silicon. The device being measured consists of alternating 10 x 30 Jam and 100 × 30 lam cobalt islands sputter-deposited on an intrinsic silicon substrate. The gaps between the cobalt islands are 4 lam and have been selectively ion implanted to allow current to flow along the device, which contains some 1000 cobalt islands in series. The figure shows the variation of resistance of the sample with an applied magnetic field. The steep rise in resistance around zero field is believed to be caused by the relaxation of the domains at the edges of tt~e cobalt islands. These initial results serve to put a lower limit on the spin diffusion length in silicon of several microns.

Fig. 8. European Spin Electronics.

L Gregg et al./Journal of Magnetism and Magnetic Materials 175 (1997) 1-9 7

Process Technology

Machine Engineering

Building Technology

Automotive Appl.

World Market of Sensors • 1994

• 2004

0 2 4 6 8 10

,Sales [Bill. ECU]

European Automobile Sensor Market

Position & Motion S.

Pressure Sensors

Temperature Sensors

Level Sensors

Gas Sensors

• 1994

• 2001

0 0.2 0.4 0.6 0.8 1 1.2 Sales [Bill. ECUI

Fig. 9. Sensor markets. By 2004 the world market for automobile sensors will have risen approximately 8 Billions ECU. The lion's share of the European automobile sensor market is in position and motion sensors where spin electronics offers improved performance and reliability.

a Thematic Brite Euram programme (Automag) whose purpose is to address the potential applica- tions and barriers to market entry of the new spin electronic technology in the European automotive industry. Bearing in mind that opera- tion in hostile and dirty environments will be an eventual requirement, particular attention is being paid to corrosion resistance and immunity to high temperatures and radio frequency inter- ference.

Potential applications of spin electronics lie in the data-storage industry as might be expected, but also in robotics, and precision engineering where fast accurate position and motion sensing of me-

chanical components is a requirement. In particu- lar, there is a growing awareness of the need for such technology in the development of improved fuel handling systems and electronic engine control in the automotive industry (Fig. 9). The require- ment for robust, temperature stable, high signal- to-noise sensors extend to many other aspects of automobiles including antiskid systems, speed con- trols and crash avoidance. Another potential niche is in the area of programmable gate arrays (Fig. 10) where magnetically programmable hardware links offer the prospect of low power consumption, non- volatility, reduced complexity, reprogrammability and speed of programming.

8 3 Gregg et al./' Journal of Magnetism and Magnetic Materials 175 (1997) 1 -- 9

Current technologies used f o r FPGA implementation.

Anti fuse (permanent link)

low resistance path when eleelaieally programmed

Di®l,eu, ic Poly~ilico n

SRAM

pass transistor controlled by the state of a SRAM bit

Floating Gate Programming

turned off by injecting charge onto the gate

floatin$ gate

- - \ I

I EPROM or EEPROM technology

Spin Electronics speed offers all the advantages o f c o m p a c t n e s s , nonvolatility and low power consumption.

(~) small size immcthately on power-up

( ~ not reprogrammable

( ~ repr ogrammable (~) standard technology

@ external permanent memory @ needed

occupies large area

reprogrammable no external memory required

high static power consumption long processing

EPROM (Ultraviolet erasable) : not in-citcxtit reprogran~nable

EEPROM (F.l~uieally evasabl©) : occul~es large area

A s c h e m a t i c configuration for m a g n e t o - o p t i c programming of Spin E l e c t r o n i c Gate Arrays.

B Fig. 10. Programmable gate arrays. In computat ion tasks involving many similar arithmetic operations, huge savings in time can be made using application specific integrated circuits (ASIC). However, their commercial success suffers from long development lead times, early obsolescence and relatively high cost; hence the rationale behind field programmable gate arrys (FPGAs) which are gate arrays whose hardware connections can be programmed to make ASICs at will by the user. Three types of commercial FPGA exist based on antifuse, SRAM and floating gate technologies, respectively. The first is nonprogrammable, the second occupies large area and requires external memory, and third occupies large area and has high static power consumption. Magneto-optically recordable spin electronic transistors offer the possibilty of making compact, nonvolatile low-power FPGAs which have the added advantage that they are faster and very much simpler to programme.

Acknowledgements

The authors are grateful for the support of the Brite Euram and TMR programs of the European Commission, the EPSRC, the British Council and

Leybold UK. They also wish to acknowledge the late Dr. Neville Robinson whose vitality, ingenuity, keen sense of humour and 'lively awareness of Maxwell's equations' have afforded all of us such enjoyment in the study of spin electronics.

J. Gregg et al. /Journal of Magnetism and Magnetic Materials 175 (1997) 1 9 9

References

[1] J.F. Gregg, S.M. Thompson. Electron. Wireless World (November 1994) 904.

[2] N.F. Molt, Proc. R. Soc. 153 (1936) 699. [3] A. Fert, I.A. Campbell, Phys. Rev. gctt. 21 (1968) 1190. [4] A. Fert, 1.A. Campbell, J. Phys. F: Metal Phys. 6 (1975)

849. [5~] M.N. Baibich, J.M. Broto, A. Fert, F. Nguyen Van Dau, F.

Petrott~ Phys. Rev. Lett. 61 (1988) 2472.

[6] G. Binasch et al., Phys. Rev. B 39 (1989) 4824. [7] T. Valet, A. Fort, Phys. Rev. B 48 (1993) 7099. [8] A. Fert, Private communication, 1993. [9] M. Johnson, Science 260 (1993) 320.

[10] D.J. Monsma, J.C. Lodder, J.A. Popma, B. Dieny, Phys. Rev. Left. 74 (1995) 5260.

[11] British patent, High beta magnetic spin transistor device, 1996.