Spin Seebeck devices using local on-chip heating Stephen M. Wu,1,a) Frank Y. Fradin,1 Jason Hoffman,1 Axel Hoffmann,1 and Anand Bhattacharya1 Materials Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA (Dated: 2 February 2015) A micro-patterned spin Seebeck device is fabricated using an on-chip heater. Current is driven through a Au heater layer electrically isolated from a bilayer consisting of Fe O (insulating ferrimagnet) and a spin 3 4 5 detector layer. It is shown that through this method it is possible to measure the longitudinal spin Seebeck 1 effect (SSE) for small area magnetic devices, equivalent to traditional macroscopic SSE experiments. Using 0 a lock-in detection technique it is possible to more sensitively characterize both the SSE and the anomalous 2 Nernst effect (ANE), as well as the inverse spin Hall effect in various spin detector materials. By using the n spin detector layer as a thermometer, we can obtain a value for the temperature gradient across the device. a These results are well matched to values obtained through electromagnetic/thermal modeling of the device J structure and with large area spin Seebeck measurements. 8 2 The spin Seebeck effect (SSE) has been widely stud- current source, while X serves as a spin detector layer ] ied due to the implications it has on the generation of that responds to spin current through the inverse spin l l pure spin currents1,2. In the SSE, applying a thermal Hall effect (ISHE). a h gradient across a magnetic insulator generates a pure Fe O is grown on MgAl O (100) substrates using 3 4 2 4 - spin current that flows into an adjacent material with- ozoneassistedoxidemolecularbeamepitaxy(MBE),de- s e out any chargecurrent3. Experiments involvingthe SSE scribed elsewhere6. The film is then patterned into 10 m have taken many forms, but the canonical method has µm x 800 µm strips through standard photolithography . been to use large area ferromagnetic insulators on the and liquid nitrogen cooled Ar+ ion milling. Liquid ni- at order of several millimeters combined with Peltier ele- trogen cooling is used to limit defects introduced by the m ments to generate a thermal gradient. There are several ionmillingprocess. Usingalift-offprocesswithstandard - disadvantagesto this technique including being sensitive photolithographyandelectronbeamevaporation,alayer d to materialnon-uniformity,being confinedto using large n samples, and being limited by complicated experimental o setups. c [ Here we introduce a micro-patterned spin Seebeck de- 6# 7489:# vice to solve many of these issues. We measure a mi- 1 croscale SSE using confined local on-chip heating, which v iselectricallyseparatedfromboththespincurrentsource 4 # 5/6+ 9 9 and spin current detector. Using this method it is possi- !"# !’# ble tosensitivelyexaminelocalmagnetization,localspin $%&# 5 7 current, and local spin to charge conversion in a sim- ’()*+#,-#’(# 0 ple and scalable device structure. Since the devices are ./&# 0 1 1. small, it is possible to easily integrate them into con- $%!23&1# 0 ventional cryostat systems. While there are alternative 9# 5 small-areaspinSeebecktechniquessuchaslaserheating4 1 orcurrentinducedheatingusingthespindetectorlayer5, ;# : each has its disadvantages. Laser heating involves addi- v i tional experimental setup, making it harder to integrate X into existing device measurementsetups, andcurrentin- r ducedheatingusingthespindetectorlayergeneratessev- a eral undesired conventionalcharge transporteffects that must be separated to extract the SSE. In this work, our devices consist of Fe O (60 3 4 nm)/X/MgO (100 nm)/Au (100 nm) on MgAl O sub- 2 4 strates, where X is either Ti (15 nm), Ti (1.5 nm)/Pt (5 nm) or absent (Fig. 1a). On-chip heating is provided by the Au layer, which is electrically isolated from the rest ofthedevicebyalayerofMgO.Fe O servesasthespin FIG. 1. A schematic of a typical spin Seebeck device is pre- 3 4 sented in (a). A model of theactual device structureused in experiments and simulations is presented in (b). The pillars representwirebondsmadetothedevice,thewirebondsonthe a)Electronicmail: [email protected] second set of pads are hidden for viewing clarity. 2 ofTi/PtorPtisthendepositedontotheFe O strip. Fi- 3 4 1.5 25 nally, using the same lift-off processthe MgO/Auheater 0.4 Fe3O4/Ti/Pt layerispatternedontothedevice,withseparatecontacts V0.2 Fe3O4/Ti to the side. The finished device is shown in a 3D model 1.2 V (0.0 Fe3O4 20 in Fig. 1b. -0.2 The response in the spin current detector layer is V) 0.9 -0.4 -2 0 2 15 E~ISHE ∝ J~S × σˆ, where J~S is the injected spin cur- V ( 0.6 Magnetic Field (kOe) 10 rent, and σˆ is the unit vector of the spin. Because the spincurrentinjectedintotheadjacentspindetectorlayer 0.3 5 is directly proportional to the thermal gradient and the magnetization, the total response is E~ ∝ ∇T ×M~. SSE 0.0 0 We are not limited to using DC techniques to detect the ISHE signal from the spin detector layer. By using AC 0 50 100 150 200 250 300 lock-indetection, the sensitivity ofthe measurementcan be greatlyincreased. Since the power generatedthrough Temperature (K) Joule heating P =I2 R∝∇T, the voltagemeasured FIG.2. ThemagnitudeofthespinSeebeck/anomalousNernst heater will also be ∝ I2 . If an AC signal is used to excite response with respect to temperature. Devices with spin de- heater the heater I ∝ sin(ωt), then the measured voltage tector layers of Ti (15 nm), Ti(1.5 nm)/Pt (5 nm), and a heater V ∝ 1(1−cos(2ωt)). By lock-in detecting the 90◦ out- control device with Fe3O4 are presented. The inset curve is 2 anexampleofaspinSeebecksignalwithrespecttomagnetic of-phase component at the 2ω frequency it is possible to field for a Ti/Pt device at 15 K. The linear contribution to detect the spin Seebeck effect withmuchhigher sensitiv- thisvoltageresponseislikelyduetotheordinaryNernsteffect itybecauseparasiticeffectsoccurattheω frequency. By from the spin detector materials. usingthis techniqueandignoringthe constantterm,itis possible to detect signals as small as 2 nV depending on the integration period of the lock-in amplifier. deviceanddeviceswithspindetectorlayersareduetothe The sample is mounted onto a standard circuit board largedifferenceinresistivitiesbetweenthedifferentfilms. using silver paint, and contact to the device is made us- Thelowresistancespindetectorlayersessentiallyactasa ing wirebonds before it is loaded into a Quantum De- shunt that reduces the total voltageresponse7,11. At the sign Physical Properties Measurement System (PPMS) same time, the difference in the magnitudes between the cryostatfor temperature and magnetic fieldcontrol. Ex- Ti and the Ti/Pt devices is explained by the individual periments performed on three different device stacks are spin Hall angles of the two materials. The spin Hall an- presentedinFig. 2,eachwithadifferentspincurrentde- gleinTiis smallandnegative,aspredictedtheoretically tector layer (Ti, Ti/Pt, or Fe3O4 alone). In each exper- through tight binding calculations12 and experimentally iment a constant 5 Vpp signal at 99 Hz is applied across observed13, while Pt has been the long standing canoni- the Au heater and a 60 ohm load resistor. Since the re- calstandardfor ISHE experiments due to its largeISHE sistance of the heater layer changes with temperature, response14. Our measurements reproduce these results the power applied using this measurement also changes well,showingthatourmethodisequivalenttolargearea from 17.2 mWrms at 300 K to 21.9 mWrms at 15 K. At macroscopic measurements. each temperature the voltage across the device is mea- To look further in detail into the heat flow in our de- sured with respect to an in-plane magnetic field applied vices, thermal simulations were performed on the device transverse to E~SSE. A typical curve is presented in the model presented in Fig. 1b, using the Computer Simu- inset of Fig. 2. The magnitude of the voltage response, lationTechnology(CST)Studio Suite softwaremodeling as defined as the voltage difference at zero applied mag- package. Values for material resistivity were taken from netic field for the two different magnetization states of measured values, while values for thermal conductivity Fe3O4,is shown against temperature in Fig. 2. and heat capacity were taken from standard sources in Fe O is only insulating below a well known metal- literature15,16. Since the device sits on a large substrate 3 4 insulatortransitionat120KknownastheVerweytransi- thatismountedusingsilverpaintontoastandardPPMS tion. Attemperaturesabovethe Verweytransitionthere rotatorcircuitboardsolidlyheatsinkedtothePPMS,the is a contribution from both the anomalous Nernst effect back side of the substrate is assumed to be held at the from conducting Fe O , and the ISHE due to the SSE bathtemperature. Thesameconditionisassumedofthe 3 4 in Ti or Ti/Pt. At these temperatures it is not possible fourwirebondcontactstothedevice,sinceeachwirebond to separate the two effects7. However,below the Verwey is also connected directly to the circuit board. The DC transition ,there is a recoveryin both the Ti/Pt and the heater current chosen for this simulation was equivalent Ti devices due to the SSE, while the signalin the Fe O to the maximum peak current in our AC measurements, 3 4 controldevicegoestozero. The1.5nmTispacerlayerin 24.5 mA at 300 K and 30.1 mA at 15 K. The resulting betweentheFe O andthe PtintheTi/Ptdeviceserves heater power loss density is presented in Fig. 3a for a 3 4 toremoveanyeffectsfromproximitymagnetism8–10. The Ti/Pt device at 300 K. This shows that all the heater largedifferenceinmagnitudesbetweentheFe O control power is constrainedto the narrowestand most resistive 3 4 3 a Using thermal simulation it is possible to model the out-of-planethermalgradientacrosseachthinfilminthe centerofourdevice(Fig. 4). Thethermalgradientacross just the Fe O layer is 0.298 K/µm at 300 K and 0.014 3 4 K/µmat15 K.This difference is consistentwith the rel- ativemagnitudesofthethermalconductivityofFe O at 3 4 300 and 15 K. The inset of Fig. 4 shows that the much larger temperature drop is across the substrate due to its size relative to the thin film (550 µm vs. 0.06 µm). b It also shows that the temperature drop across the sub- strate is non-uniform due to the heating being localized to the device and not the entire top surface of the chip. Locally at the surface, on the scale of our thin films, the temperature distribution is highly linear. 302.23 FIG. 3. Electromagnetic and thermal simulation results for MgAl2O4 Fe3O4 MgO Au a Ti/Pt spin Seebeck device. (a) shows the calculated power Pt l(obs)ssdheonwssittyheducealctuolaJtoeudlelohceaalttienmgpinerathtuerehewaittehrinlatyheer,dwevhiiclee K)302.22 Ti that results from heating. ure ( at er302.21 p m partofthedevicewheretheFe O /Ti/Ptdevicestackis. e 302 3 4 T Theresultingtemperaturewithinthe deviceispresented 302.20 301 in Fig 3b. The ∆T fromthe back ofthe substrate to the 300 top of the heater is 2.226 K at 300 K, and 0.565 K at -500 -250 0 15 K. Given this ∆T, the voltage response measured in 302.19 -0.1 0.0 0.1 0.2 our devices match well with large area SSE experiments on both thin film and bulk ferromagnets11,17. Since the Distance from substrate surface ( m) largeareaexperimentsshowthatthereisalinearvoltage FIG.4. Temperatureintheout-of-planedirectionatthecen- response to thermal gradient up to ∆T= 20 K, by opti- terofaTi/PtspinSeebeckdeviceascalculatedfromthermal mizing the power transferred to our heater our signal to modeling. Theinsetshowsthesametemperaturedistribution noise ratio could be increased even further by increasing at an expanded range through the entiresubstrate. ∆T in our system. The averaged ∆T along the device length, shows the uniformity of the thermal gradient is Using the metallic spin detector layer in our device within 3% according to simulations. as a thermometer, it is possible to compare the results Although there is a large ∆T across the entire sub- of thermal modeling with measured values for ∆T. By strate and device, the temperature drop across just the characterizing the resistance of the Ti (15 nm) strip in Fe3O4 layer is not as large and depends its individual our Fe3O4/Ti spin Seebeck device we have a one-to-one thermalconductivity. Characterizing∆Tacrossthinfilm relationbetweentemperatureandresistance. Byrunning ferromagnets has been an ongoing challenge in SSE ex- an AC signal through our on-chip heater and measuring periments both on the macroscale and the microscale the change in resistance in the Ti strip it is possible to since there is no standard method to probe the tem- directly measure the change in temperature due to the perature at both sides of the thin film7. By applying heater. The resistance of the Ti strip and its derivative aconstant∆Tacrossthesubstrateandthe thinfilmfer- with respect to temperature is measured and shown in romagnetthereisnoguaranteethatthethermalgradient Fig. 5a-b. Next, the same 99 Hz 5 Vpp signal is sent to acrossthe thinfilmwillbe constantwithrespecttotem- the heater through a 60 ohm load like in the experiment peraturesincethethermalconductivitiesofthesubstrate presented in Fig. 2. The peak to peak amplitude of the and the thin film will change relative to each other. By change in resistance of the Ti strip is presented in Fig using an on-chip heater, we effectively send a constant 5c. Finally, the change in temperature is derived from heat current through the device instead of setting up a ∆T = ∆R (Fig. 5d). Measurements of the thermal dR/dT constant ∆T. This is analogous to performing a current time constant of our system revealed two time scales for biased vs. voltage biased electrical measurement. Using changes in the resistance of Ti strip at all temperatures. thismethoditmaybeeasiertoeliminatesubstratebased Oneofwhichoccursfasterthanourdataacquisitionsys- effectssincethethermalgradientacrossthethinfilmfer- tem can resolve (5000 samples/s), and one of which oc- romagnet will only depend on the properties of the thin curs onthe scale of seconds. The longertime constantis filmandnotitsrelativevaluecomparedtothesubstrate. likelyduetoheatingtheentireexperimentalprobewithin 4 tection layer as a thermometer it is possible to extract a 5.5 equivalentthermalgradienttotheconstant∆Tmeasure- a ) 5.0 ments in conventional spin Seebeck experiments. These k (4.5 measurements match well with thermal simulations of R ourdevicestructure. Thistypeofdevicestructureallows 4.0 for easier access to lower temperature, higher magnetic 3.5 ) field, and smaller magnetic materials all of which have K / 9 b been challenging to explore using other methods, which T ( 6 require more custom experimental setup. Through the d exploration in these regimes it may be possible to pro- / R 3 vide further insight into the basic mechanisms behind d the SSE, as well as potentially finding other interesting 0 thermal spin transport phenomenon. ) c 4 ( R 2 ACKNOWLEDGMENTS 0 All authors acknowledges support of the U.S. Depart- K)1.2 d ment of Energy (DOE), Office of Science, Basic Energy ( Sciences (BES), Materials Sciences and Engineering Di- T vision. The use of facilities at the Center for Nanoscale 0.6 Materials, was supported by the U.S. DOE, BES under contract No. DE-AC02-06CH11357. 0.0 0 50 100 150 200 250 300 350 1K. Uchida, S. Takahashi, K. Harii, J. Ieda, W. Koshibae, Temperature (K) K.Ando,S.Maekawa, andE.Saitoh,Nature455,778(2008). 2K.Uchida, H.Adachi, T.Ota,H.Nakayama, S.Maekawa, and FIG. 5. The measured resistance vs. temperature of E.Saitoh,Appl.Phys.Lett.97,172505 (2010). Fe O /Ti strip is presented in (a), along with its derivative 3 4 3K. Uchida, J. Xiao, H. Adachi, J. Ohe, S. Takahashi, J. Ieda, with respect to temperature in (b). (c) shows the change in T. Ota, Y. Kajiwara, H. Umezawa, H. Kawai, et al., Nature resistanceoftheFe3O4/Tistripuponheatingwithaconstant Mater.9,894(2010). 5 Vpp applied across a 60 ohm load resistor and the heater 4M.Weiler,M.Althammer,F.D.Czeschka,H.Huebl,M.S.Wag- layer. Using the results from (b) and (c), the temperature ner,M.Opel,I.-M.Imort,G. Reiss,A.Thomas,R.Gross, and differenceacrossthedeviceiscalculatedwithrespecttotem- S.T.B.Goennenwein, Phys.Rev.Lett.108,106602 (2012). perature and presented in (d). 5M. Schreier, N. Roschewsky, E. Dobler, S. Meyer, H. Huebl, R.Gross, andS.T.Goennenwein,Appl.Phys.Lett.103,242404 (2013). 6M. Liu, J. Hoffman, J. Wang, J. Zhang, B. Nelson-Cheeseman, the cryostat along with the sample. At 99 Hz the only andA.Bhattacharya, Sci.Rep.3(2013). contribution we measure is the short time constant ∆T 7S. M. Wu, J. Hoffman, J. E. Pearson, and A. Bhattacharya, across just the device. The values at both low and high Appl.Phys.Lett. 105,092409(2014). temperatures show values close to the predicted results 8S.-Y. Huang, X. Fan, D. Qu, Y. P. Chen, W. G. Wang, J. Wu, T.Y.Chen,J.Q.Xiao, andC.L.Chien,Phys.Rev.Lett.109, ofthermalmodeling. Theremainingdifferencesarelikely 107204(2012). dueto thedifferencesbetweenthe valuesofthermalcon- 9T. Kikkawa, K. Uchida, Y. Shiomi, Z. Qiu, D. Hou, D. Tian, ductivity taken from literature and our samples. Since H. Nakayama, X.-F. Jin, and E. Saitoh, Phys. Rev. Lett. 110, our measurements are heat current biased, the results 067207(2013). in Fig. 5d scale inversely to measurements of thermal 10S. Gepr¨ags, S. Meyer, S. Altmannshofer, M. Opel, F. Wilhelm, A. Rogalev, R. Gross, and S. T. B. Goennenwein, Appl. Phys. conductivity on MgAl2O4, which seems to be the largest Lett.101,262407 (2012). determining factor of ∆T in the device. There is a local 11R. Ramos, T. Kikkawa, K. Uchida, H. Adachi, I. Lucas, M. H. increase in ∆T near the Verwey transition in Fig. 5d Aguirre,P.Algarabel,L.Morell´on,S.Maekawa,E.Saitoh,etal., that cannot be directly explained by thermal conductiv- Appl.Phys.Lett. 102,072413(2013). itychangesinMgAl O orFe O singlecrystals15 alone, 12T. Tanaka, H. Kontani, M. Naito, T. Naito, D. S. Hirashima, 2 4 3 4 K.Yamada, andJ.Inoue,Phys.Rev.B77,165117 (2008). butdoesresemblethediscontinuityinFe3O4 heatcapac- 13K. Uchida, M. Ishida, T. Kikkawa, A. Kirihara, T. Murakami, ity at this temperature16. andE.Saitoh,J.Phys.: Condens.Matter26,343202 (2014). WehaveintroducedamethodtomeasurethespinSee- 14A.Hoffmann,IEEETrans.Magn.49,5172(2013). beckeffectusingamicropatterneddevicewithanon-chip 15G.A.Slack,Phys.Rev.126,427(1962). 16J.P.Shepherd,J.W.Koenitzer,R.Arago´n,C.J.Sandberg, and heater. Byusingasmallscaledeviceitispossibletosen- J.M.Honig,Phys.Rev.B31,1107(1985). sitively measure local magnetization, local spin current, 17K.Uchida,T.Ota,H.Adachi,J.Xiao,T.Nonaka,Y.Kajiwara, andlocalspintochargeconversiononthemicroscaleand G.Bauer,S.Maekawa, andE.Saitoh,J.Appl.Phys.111,103903 potentially the nanoscale. By using the spin current de- (2012).