lable at ScienceDirect

Vacuum 83 (2009) 1007–1013

Contents lists avai

Vacuum

journal homepage: www.elsevier .com/locate/vacuum

Towards the sub-50 nm magnetic device definition: Ion beam etching(IBE) vs plasma-based etching

Xilin Peng a,*, Stacey Wakeham a, Augusto Morrone a, Steven Axdal a, Michael Feldbaum b,Justin Hwu b, Tom Boonstra a, Yonghua Chen a, Juren Ding a

a RHO, Seagate Technology, 7801 Computer Avenue South, Bloomington, MN 55435, USAb RMO, Seagate Technology, 47050 Kato Road, Fremont, CA 94538, USA

a r t i c l e i n f o

Article history:Received 8 July 2008Received in revised form8 December 2008Accepted 11 December 2008

Keywords:Tunneling magneto resistive (TMR)Ion beam etching (IBE)Reactive ion etching (RIE)XPSMagnetic materials

* Corresponding author. Tel.: þ1 952 402 8524; faxE-mail address: xilin.peng@seagate.com (X. Peng).

0042-207X/$ – see front matter � 2008 Elsevier Ltd.doi:10.1016/j.vacuum.2008.12.003

a b s t r a c t

Conventionally, the tunneling magneto resistive (TMR) devices for both hard drive and magnetic randomaccess memory (MRAM) are defined via photolithography and subsequent ion mill processes. Due tonon-volatility of ion milling byproducts, re-deposition of device materials across the tunneling barrierwill increase the critical dimension (CD) and reduce the pattern transfer fidelity; moreover, it causeselectrical shunting and TMR ratio drop. Therefore, either relatively large angle primary or two-step millwith a subsequent large angle side mill is required to clean-up such re-deposition across the barrier. Suchprimary milling angle and side milling time at a fixed primary mill angle have been determinedexperimentally to be w20–30� and above 30 s, respectively, in this study. However, it was found thatextended side milling can cause substantial damage for sub-w30 nm. We also investigated the plasma-based etching of such TMR devices using various chemistries and presented optical emission spectrum ofsuch chemistries. The plasma etched TMR device profile and the possible interaction between thechemistry with the MgO barrier was also discussed.

� 2008 Elsevier Ltd. All rights reserved.

1. Introduction

With the ever-increasing requirements to handle huge amount ofdata in modern information/communication systems, there is anincreasing demand for higher hard disk drive capacity and betterperformance. For the areal density approaching w1 TB/in2, themagnetic reader sensor size must be decreased to read narrowerrecorded bits with minimal side reading. From recording aspect, oneeffective way to increase the recording density is to pre-define smallmagnetic dot-arrays as the media, i.e. bit patterned media (BPM). Forthe magnetic random accessmemory (MRAM) applications, magneticdevices are packed more closely than for the hard drive applicationdue to package density requirements. This creates huge challenges fora definition process for magnetic devices (tunneling magneto resis-tive (TMR) sensors or BPM) that does not result in re-deposition ofdevice materials. Re-deposited material tends to grow on the sidewallof the initial photo resist and thus inflates the final device physicalcritical dimension (CD) and causes the pattern fidelity issue. Moreimportantly, re-deposited material results in electrical shuntingacross the barrier (for TMR devices) if not fully cleaned-up. A side mill

: þ1 952 402 8349.

All rights reserved.

process at large angle is normally added after the primary ion beammill definition to remove re-deposited materials and to providea better CD control. Unfortunately, large angle side mill has draw-backs: (1) it creates damages at the edges of devices and degradesdevice performance; (2) at high package density, side milling issometimes impossible due to shadowing effects; and (3) the sameshadowing effect can cause poor process uniformityat large angle dueto clamps used at wafer edges. To minimize edge damage of devices,a low beam energy can be used, however, the beam divergence angleincreases at lower beam energy and compromises milling uniformity[1–4]. The shadowing effect due to device package density is verydifficult to solve; it can be minimized by optimizing clamp design atedges, though it cannot be fully avoided since electro static chucks(ESCs) are very difficult to implement for rotating wafer stages.

As a result, plasma-based etching has been very activelyresearched [5–9] by both industry and academia to supplementconventional ion beam etching (IBE). Plasma-based etching shouldbe easier to scale up to large wafer dimension, compared with IBE.Plasma-based etching tools do not require grids, hence less grids-related contaminations, and the operation/maintenance cost ischeaper. However, there are challenges in various aspects ofplasma-based etching: (1) selection of the right etching chemistrywith sufficient byproduct volatility; (2) reduction of chemicalreactions or damage to the device edges; (3) choice of hard mask

0

0.2

0.4

0.6

0.8

1

1.2

-5 5 15 25 35 45 55 65 75

-5 5 15 25 35 45 55 65 75

Milling angle (degrees)

No

rm

alized

T

MR

0

0.2

0.4

0.6

0.8

1

1.2

Milling angle (degrees)

No

rm

alized

(R

A) p

ro

du

ct

a

b

Fig. 1. Normalized (a) TMR and (b) resistance� area (RA) product for a various TMRwafers as a function of primary mill angles. No side mill was applied here post-primarymill. It is obvious that a primary milling angle of 25–35� is required to avoid re-depacross the barrier if no 2nd side mill is applied.

X. Peng et al. / Vacuum 83 (2009) 1007–10131008

which enables a high etching selectivity; and (4) subsequentprocess considerations. For example, if a hard mask has to be used,how can it be removed after device definition.

In this paper, we present results and challenges for sub-50 nmmagnetic device definition via both conventional IBE and plasma-based etching.

2. Experiments

2.1. Device stack

2.1.1. TMRWe have used a conventional TMR device with the following

structure layout in this study: Seed layer/AFM/SAF/MgO/free layer/Cap layer (AFM is the abbreviation of ‘‘Antiferromagnet’’, while SAFis the abbreviation of ‘‘Synthetic Antiferromagnet’’).

2.1.2. Patterned mediaFor patterned media study, the following film stack structure

was used: Seed layer/CoPtCr/Cap layer, which was deposited witha magnetron physical vapor deposition (PVD) cluster tool.

2.2. Device definition

2.2.1. TMR deviceAfter the TMR stack film deposition, photolithography and ion

beam milling or plasma-based etching were used to define the finaldevice dimension and profile. For conventional ion beam milling,commercially available radio frequency (RF) ion sources witha three-grid design were used. A typical beam is operated at1.33�10�2 Pa using Ar. Beam current and voltages can be varied forbetter process uniformity. The incident beam angle can be changedfrom 0 to 80� to control the junction profile and re-depositedmaterial. For plasma-based etching, inductively coupled top RFsource and a capacitively coupled bottom RF source, both at13.56 MHz, were used. The chuck temperature was set at 353 K andthe shielding was set at 473 K to facilitate the vaporization ofetching byproduct. Available etching chemistries in the systeminclude Ar, CF4, O2, NH3, and CH3OH.

2.2.2. Patterned mediaImaging resist was applied using a propriety mask imaging

technique and was patterned using ion mill technique to define themedia pattern. After ion milling, the resist residual was strippedusing an inductive coupled plasma (ICP) source with CF4/O2

chemistry.

2.3. Device isolation and hard bias formation

After device definition by either IBE or plasma-based etching, anelectrical isolation layer (such as Al2O3) is deposited by sputtering,chemical vapor deposition or atomic layer deposition. To reduce thethermal budget during the isolation process, the substratetemperature is limited to below 473 K. The isolation layer isgenerally less than 100 Å to balance the good electrical isolationproperty and good permanent magnet hard bias strength to themagnetic free layer.

Following isolation formation the hard bias and top electrodeare deposited.

2.4. Device testing

2.4.1. TMR deviceFinished devices were tested on a 4-probe tester with the

external field sweeping from �1000 to þ1000 Oe to gauge their

electrical and magnetic responses. For each device category, at least64 sites on a 150 mm diameter wafer were tested. Device perfor-mance was then analyzed statistically. Transmission electronmicroscopy (TEM) was used for selected samples to provide detailsof the junction profile, the reaction layer around the junction, andfailure mechanism.

2.4.2. Patterned mediaPatterned media were tested using magneto-optical Kerr effect

(MOKE) technique. The parameters tested included Hc (coercivity),thermal stability (KuV), nucleation field (Hn), and switching fielddistribution (SFD). Measurements were conducted using multiplelocations at each stage of the process.

3. Results and discussions

3.1. Conventional IBE definition process

3.1.1. Re-depositionFig. 1 summarizes the trend of both the TMR (a) and the resis-

tance� area product (RA) (b) as a function of the primary millingangle. No side clean-up mill was applied in this case, simplifyingdata interpretation. It is evident that both TMR and RA reacha stable level when the primary milling angle is about 25–35� fromwafer surface normal. Generally, the rejected atoms from a surfacesubjected to primary ion beam irradiation will follow a cosine

barrier

Re-dep

20 nm

isolation

Edge of milled device

Fig. 3. A portion of a cross-section TEM image for a TMR device with re-depositedmaterial across the barrier clearly visible. This wafer was completely shorted because

X. Peng et al. / Vacuum 83 (2009) 1007–1013 1009

distribution. As a result, sidewall re-deposition is created. Themoment of the primary beam at any angle can be decomposed intoboth vertical and horizontal components. The horizontal compo-nent of the Ar ion beam will provide a clean-up effect in compe-tition with re-deposition. Once re-deposited material is removed bythe horizontal beam component, the magnetic device should befree from any metal layer across the barrier and the TMR and RAvalues will approach a stable level.

Due to profile control requirements, a large angle primarymilling is sometimes not practical. Instead, a 0� (vertical incidentbeam) mill is used, followed by a large angle clean-up. The largeangle milling is necessary to remove the re-deposited metal layeracross the oxide barrier on TMR devices. Generally, a critical time ata fixed side mill angle is required to achieve a clean and ‘‘re-dep-free’’ TMR device junction. Fig. 2 illustrates how the clean-up sidemilling time at 65� impacts the device TMR and RA. Thirty secondsof side clean-up as necessary to remove re-dep. Fig. 2 also clearlyshows that when there is a sidewall metal deposition across thebarrier, the device TMR and RA sigma are also greater, indicatingpoor process controllability.

The TEM image in Fig. 3 highlights the existence of such re-deposition post-primary mill, if there is no 2nd side mill or the 2ndside mill time is not long enough. A black line on the right side ofthe isolation around junction is visible and indicated by an arrowon the figure. The black line crosses the barrier layer, causingshorting. For wafers with sufficient side mill clean-up, no evidence

No

rm

alized

(R

A) p

ro

du

ct

0

0.2

0.4

0.6

0.8

1

1.2

0

0.2

0.4

0.6

0.8

1

1.2

-10 40 90 140Side milling time, sec (at 65 degrees)

-10 40 90 140Side milling time, sec (at 65 degrees)

No

rm

alized

T

MR

a

b

Fig. 2. Normalized (a) TMR and (b) resistance� area (RA) product for a various TMRwafers as a function of 2nd side mill time at a fixed angle of 65� . Primary milling wascarried out using end point (detect the signal rising of bottom shield NiFe material) ata fixed angle of 0� . It is obvious that a 30 s side mill is required to clean-up the re-depacross the barrier and get RA and TMR to increase to a normal level.

2nd cleaning-up step was not carried out due to machine failure and consequently, andthe TMR is only 1/10 the value for the non-shorted devices.

of re-deposition is seen and no shorting is observed in electricaltesting data.

The impact of the re-deposition on device performance (for bothRA and TMR) can be estimated by a two-resistor-in-parallel model.Assuming the device size as a 45 nm by 500 nm rectangle witha nominal RA product of 0.6 U mm2 and a nominal TMR ratio of 50%,we have modeled how the shunting layer RA product (at a fixedshunting layer thickness of 10 nm) and the shunting layer thickness(at a fixed RA product) affect the final device TMR and RA. Resultsare shown in Fig. 4. Basically, the lower the shorting layer RA, thelarger the TMR and RA drops that are expected for the final device.

3.1.2. Edge damageOnce the device size becomes small enough, the edge surface

area becomes significant. Damage due to the side milling orsubsequent processes can have substantial impact of the deviceperformance. For instance, in a NiFe/Cu/Co/Cu Pseudo-Spin-Valve(PSV) system, Castano et al. [10] found that the Giant MagnetoResistance (GMR) ratio and film saturation magnetization of suchPSV dropped as the device wire width was reduced from 150 to 80and to 60 nm. This was attributed to the formation of a disorderzone during ion milling on the edges of such devices. Similarly,Katine [11] observed that as spin valve device line width wastrimmed below 50 nm by focused ion beam (FIB), its GMR ratiodropped from 11.8% at sheet film level to w7%, due to edge damagecaused by FIB trimming. The FIB-induced dead layer was estimatedat w6 nm on each side in Katine’s report. Our results shown inFig. 5 also indicate that the TMR ratio will drop as device size getssmaller. The smallest device for this wafer, i.e. those with the largestresistance in the graph, was found to be w15 nm from independentTEM measurement. It is evident that TMR for such small devicesdrops by 50%, compared with large devices (with low resistance fora fixed RA product). Interestingly, for the patterned media we didnot observe significant property changes post ion mill definition.Perhaps the size (w40 nm) was not small enough. Another

Device RA = 0.6 ohm.µm2

Device RA = 0.6 ohm.µm2

size = 45nm x 500nm shunting layer = 10 nm

0

0.2

0.4

0.6

0.8

1

1.2

0 10 20 30 40

0 10 20 30 40

Shunting layer RA, ohm. m2

No

rm

alize

d a

ctu

al

TM

R

Device size = 45nm x 500nm

shunting layer RA = 0.2 ohm.µm2

0

0.2

0.4

0.6

0.8

1

1.2

Shunting layer thickness, nm

No

rm

alize

d a

ctu

al

TM

R

b

a

µ

Fig. 4. A simple two-resistor-in-parallel model predication of the TMR drop if shuntingexists due to re-dep. (a) Normalized TMR as a function of the shunting layer RAproduct at a fixed shunting layer thickness of 10 nm; (b) normalized TMR as a functionof the shunting layer thickness, assuming a shunting layer RA of 0.2 U mm2.

X. Peng et al. / Vacuum 83 (2009) 1007–10131010

possibility is that no clean-up side milling was introduced for BPM,which left the crystalline structure of magnetic domains un-damaged.

Impurity incorporation and diffusion during milling are anothersource of magnetic saturation moment loss and result degradation ofdevice performance. For example, photo resist can be decomposed

0

0.2

0.4

0.6

0.8

1

1.2

0 0.4 0.8 1.2

Normalized device resistance

No

rm

alized

T

MR

Fig. 5. Normalized TMR as a function of device resistance for a typical TMR wafer withvarious device sizes, defined by conventional IBE (400 W RF power and 600 mA beamcurrent) with extended side milling. At large resistance (or narrow device), TMR dropsoff and sigma for both resistance and TMR increases.

under ion bombardment. Any elements inphoto resist can intermix atthe edges and impact device performance, particularly for very smalldevices. Most of such intermixed layer is expected to be removed bysubsequent clean-up milling, but full removal of such a layer seemsunlikely.

The subsequent isolation deposition can further increasedevice edge damage and magnetic dead layer thickness. Forexample, if oxygen/water-containing chemistries are employed foralumina isolation process, a reaction layer between the aluminaand the device junction wall and bottom electrode can readilyform, particularly if the isolation process occurs at elevatedtemperature. Fig. 6 shows such a reaction zone around the junc-tion wall and along the bottom electrode. This reaction layer wasmeasured at w4 nm in this graph, which is significant for sub-30 nm devices.

Taking into consideration the potential impact of such damage,it has never been more challenging than today to make sub-50 oreven 30 nm magnetic devices with minimal performance degra-dation. There have been various efforts made to minimize edgedamage and to avoid ‘‘dead’’ layer formation. One way is to usea lower beam energy ion mill process, another is to employ a novelplasma-based etching method, and a third is to use a no-oxideisolation, which we will discuss in more detail in the later section ofthis paper. There are hardware limitations in conventional IBEsource to lowering the beam energy below 100–150 eV or so, due tothe increased beam divergence and the resulting within wafer non-uniformity. The beam divergence increases with an increase in theratio of the beam voltage to total voltage (beam voltageþ acceler-ation voltage) [1–4]. At lower beam voltage, the accelerationvoltage needs to be reduced accordingly to keep the same beamdivergence angle. Unfortunately, when the accelerating voltage isreduced beyond some threshold, the electrons from the neutral-izers back-stream to the accelerating grid, causing excessive heat-ing and beam instability. Further, the maximum extractable beamcurrent is also significantly reduced. Fig. 7 highlights how theaccelerating grid’s current changes as the accelerating grid’spotential is reduced for one type of IBE source used. It is apparentthat electron back-streaming becomes significant when the accel-erating grid voltage is below 700 V (negative) for an RF sourcepower of 500 W. If the RF source power is increased to 750 W, thethreshold increases to 1200 V (negative). Of course, the criticalaccelerating voltage threshold is dependent on the ion source griddesign.

Reaction layer on sidewall

Isolation layer

Tunneling barrier

Reaction layer onbottom electrode

20 nm

Fig. 6. The w4 nm reaction layer formed during isolation deposition is clearly visible.This could degrade the performance of the device below 30 nm widths, due to thereaction layer consuming a substantial volume of the magnetic layer.

Beam voltage = 200V

0

20

40

60

80

100

120

0 500 1000 1500 2000 2500Acceleration grid potential (negative), V

Acceleratio

n g

rid

cu

rren

t,

mA

RF 500w

RF 750w

Fig. 7. Acceleration grid current as a function of the acceleration (negative) voltage.The increased acceleration grid current at reduced acceleration potential is due to theelectron back-streaming from neutralizers.

Table 1Volatility of possible magnetic material byproducts etched using Halogen and CO-based chemistries.

Etch Products Melting point Boiling Points Ref

Co2(CO)8 51 �C Decompose at 52 �C b

CoBr2 678 �C n/a d

CoCl2 724 �C 1049 �C d

CoCl2 $ 6H2O 86 �C (dehydrates) n/a d

CoCl3 n/a n/a d

CoF2 1200 �C 1400 �C d

CoF3 927 �C n/a d

CoF4 n/a n/a d

Fe(CO)5 �20 �C 103 �C a

FeCl2 674 �C 1023 �C d

FeCl2 $ 2H2O 120–150 �C (Dehydrates) n/a d

FeCl2 $ 4H2O 150 �C (Dehydrates) n/a d

FeCl3 306 �C n/a d

FeCl3 $ 6H2O 37 �C 280 �C d

FeF2 1100 �C n/a d

FeF3 >1100 �C n/a d

IrBr2 n/a n/a d

IrBr3 n/a n/a d

IrBr4 Decomposes n/a d

IrCl3 763 �C n/a d

IrCl4 Decompose n/a d

IrF3 250 �C n/a d

IrF4 n/a n/a d

IrF6 44 �C 53 �C d

[IrF5]4 104 �C n/a d

MgBr2 711 �C 1158 �C d

MgCl2 721 �C 1510 �C d

MgF2 1150 �C 2230 �C d

MnBr2 698 �C 1027 �C d

MnCl2 654 �C 1225 �C d

MnCl3 n/a n/a d

MnF2 920 �C 1820 �C d

MnF3 >600 �C n/a d

MnF4 >RT n/a d

Ni(CO)4 �19 �C 43 �C c

NiBr2 965 �C Sublimes d

NiCl2 1001 �C 993 �C d

NiF2 1450 �C >1000 �C d

RuBr2 n/a n/a d

RuBr3 400–500 �C–decompose n/a d

RuCl2 n/a n/a d

RuCl3 >500 �C n/a d

RuF3 650 �C n/a d

RuF4 n/a n/a d

RuF6 54 �C n/a d

[RuF5]4 87 �C 227 �C d

TaBr3 200 �C (to TaBr5þ TaBr2) n/a d

TaBr4 400 �C (disproportionates) n/a d

TaBr5 280 �C 345 �C d

TaCl3 n/a n/a d

TaCl4 n/a n/a d

TaCl5 210 �C 233 �C d

TaF3 n/a n/a d

[TaF5]4 97 �C 229 �C d

a http://www.ilo.org/public/english/protection/safework/cis/products/icsc/dtasht/_icsc01/icsc0168.pdf.

b http://www.ilo.org/public/english/protection/safework/cis/products/icsc/dtasht/_icsc09/icsc0976.pdf.

c http://www.inchem.org/documents/icsc/icsc/eics0064.htm.d http://www.webelements.com.

X. Peng et al. / Vacuum 83 (2009) 1007–1013 1011

3.2. Plasma-based etching process

Compared with IBE, plasma-based etching offers some uniquefeatures, such as a choice of various chemistries, tunable ionbombardment energy (for dual frequency plasma reactors), theability to scale up to large wafer size and potentially better processuniformity. Due to the possibility of no sidewall re-deposition, sidemill clean-up can be removed, which could reduce ion-induceddamage. High package density may be more achievable. CD controlmay also be better than with IBE, largely due to less re-deposition.However, the etching chemistry is more reactive than Ar ions andthe potential of chemical reaction with magnetic layers is a concernand needs to be well controlled. With proper plasma chemistryselection and subsequent passivation treatment, the above issues/concerns are not insurmountable.

3.2.1. Chemistry selectionThe low volatility of magnetic material (Co, Fe, Ni, IrMn, Ru, etc)

etching byproducts is one of the biggest challenges for successfulplasma etching processes of magnetic materials. Table 1 listsmelting and boiling points compiled from various sources for somemagnetic material byproducts formed with halogen chemistries,which are widely used by the semiconductor industry. Except forcarbonyls, Co, Fe and Ni form hardly any byproducts which easilyvaporize below 373 K.

Although halogen-based chemistry, in particular Cl2 chemistry[12,13], has also been employed by some research groups formagnetic device etching, a careful post Cl2 etching treatment(water rinsing or H2O or O2/H2-containing plasma passivation[13–15]) is required before exposing the wafers to the ambient dueto the reaction of water moisture in the ambient with the Cl2 andformation of corrosive HCl byproduct. In addition, a high substratetemperature (w473 K) is preferred to facilitate the removal of lowvolatility etching byproducts.

As a result, most commonly reported chemistries for magneticdevice etching in the literatures are: CO, CO/NH3 [7,8,16–18] andmethanol (CH3OH) [9]. The addition of NH3 to CO is believed toprevent dissociation of CO and favor formation of transitional metalcarbonyls. Nakatani [18] observed a w5X etching rate boost witha 50% NH3 addition to the CO.

3.2.2. Mask material and thickness selectionSince both CO/NH3 and CH3OH chemistries react with photo

resist (O and H plasmas consume the photo resist very quickly),a hard mask layer between the photo resist and the underneath

magnetic layer has been required. Nakatani used SiO2 as the hardmask for NiFe permalloy etching in his study. A CF4 plasma was firstemployed to transfer the photo pattern to the SiO2 which thenacted as a hard mask for the subsequent permalloy etching in CO–NH3 plasma. Due to the relatively low selectivity (2–3:1 for CoFe/NiFe etching) of SiO2 hard mask, Matsui [7] suggested using a metal(Ti, Ta, etc) as a hard mask and reported a selectivity of 6–16, whichis much higher than SiO2 hard mask.

Also, it is critical to use as thin a metal as possible to avoid thesidewall re-deposition. We have observed that if a Ta mask is

X. Peng et al. / Vacuum 83 (2009) 1007–10131012

opened via IBE and the photo resist is removed by ashing, a Ta‘‘fence’’ structure exists. There was significant deposition of etchingbyproducts along the sidewall of the etched magnetic device andthe bottom of the Ta ‘‘fence’’ due to the shadowing effect. Thisindicated that the etching byproduct is not very volatile and itsremoval rate is largely driven by ion bombardment. This is in goodagreement with Matsui’s report [7] that a <0.5 um Ta hard mask ispreferred.

3.2.3. OES of CO/NH3 and CH3OH plasmasTo better understand etching with various chemistries, we

collected Optical Emission Spectra (OES) of various etching chem-istries including NH3, N2, CO, CO/NH3, CH3OH, and Cl2 (shown inFig. 8). Basically, CO and CH3OH have very similar OES (except thatthe CH3OH plasma shows an additional strong H peak). It is notsurprising that CH3OH chemistry works similar to CO/NH3 chem-istry (NH3’s role is still not clear). CH3OH may have more advan-tages than CO/NH3, due to lower toxicity and ease of handling in

200 400 600 800

Wavelength, nm

In

ten

sity, a.u

.

200 300 400 500 600 700 800

Wavelength, nm

In

te

ns

ity

, a

.u

.

NH3_shifted

NH3+CO_Shifted

N2_Shifted

CH3OH_Shifted

CO

a

b

Fig. 8. Optical emission spectra (OES) when using a silicon substrate and the bias areturned off for (a) N2, NH3, CO–75%NH3, CH3OH, and CO plasmas. Note that the intensitywas shifted to provide a better view of each spectrum. (b) Cl2 plasma.

production. Although Nakatani [18] reported that NH3 addition toCO is necessary to facilitate a higher etching rate, NH3 addition onlyadds additional N and H peaks compared with pure CO and CH3OHplasmas. The broad spectra band from 300 to 600 nm is associatedwith CO. Their intensity drop for CO–NH3 chemistry is probablycaused by the reduction of CO concentration, not to suppression ofCO decomposition as Nakatani [18] reported.

For Cl2 plasma, the OES shows both Cl radicals (from 700 to800 nm) and Cl2

þ and Clþ (from w400 to 600 nm broad peak) [19],the peaks from 200 to 300 nm were as yet un-identified from thecurrent experiment. Due to corrosion concerns using Cl2 chemistryin this study, we only tested etching TMR device using CH3OH.

3.2.4. Potential issues with plasma-based etching of magneticdevices3.2.4.1. Chemical reactions for TMR device. Fig. 9 shows a typicalTEM cross-section image of a TMR device etched in CH3OHplasma, followed by CVD SiN encapsulation to protect the side-wall. Potentially, CVD SiN may provide some benefits in terms ofreduced edge reaction layer since it uses non-oxygen containingchemistry and shows reasonably good step coverage. It is evidentfrom Fig. 9 that a clear and symmetrical junction has been ach-ieved by such plasma etching. No edge reaction layer such asthose shown in Fig. 6 for atomic layer deposition (ALD) aluminaisolation was observed in this case. However, to make a fullyfunctional device, additional processes must follow. For example,the top isolation layer (SiN in this case) needs to be removed forgood electrical contact. Also the top Ta hard mask will be partiallyoxidized in the CH3OH and CO-based chemistries and must becleaned-up to make a good electrode contact without attackingthe device’s free layer.

3.2.4.2. Chemical reaction for patterned media. For the patternedmedia process using IBE, a subsequent CF4/O2 plasma treatment atroom temperature and 90 mTorr were employed to strip off resistresidues left by IBE process. A dual frequency ICP reactor was usedwith 700 W source power and 30 W bias power. A 45 s cleaningstep in CF4/O2 plasma treatment drastically affected magneticproperties: coercivity (Hc) dropped 11.5%, thermal stability (KuV)13%, and switching field distribution (SFD) 16%. We are not sure ofthe root cause of the magnetic property degradation, whether it beF or O reaction with the magnetic materials or another cause. If dueto oxygen, the impact of methanol ICP etching on the properties ofpatterned media needs to be further evaluated.

Fig. 9. TEM image for a CH3OH RIE defined TMR sensor x-section.

X. Peng et al. / Vacuum 83 (2009) 1007–1013 1013

In summary, while plasma etching offers advantages whencompared to conventional IBE, it creates issues that need to befully addressed before plasma-based etching can be widelyaccepted as an effective TMR device and patterned media defi-nition method.

4. Conclusions

We have compared conventional IBE and plasma-based etchingmethods for magnetic device patterning. Reduction of sidewall re-deposition and minimization of ion-induced edge damage are keyenablers for sub-50 nm device formation. Non-oxygen basedisolation materials (SiN for example) could offer benefits byavoiding dead layer formation along the junctions of ultra-smalldevices.

For plasma-based etching, there is still a substantial physicalsputtering component in CH3OH or other CO-based chemistry.Effective removal of residues and passivation of the device areimportant to ensure good performance of the devices.

Acknowledgments

The authors would like to thank Yongxiong Lu and ThomasMaclaughlin from Seagate, Ireland for fruitful discussion and

simulation support of junction profile, and Tien Dam and KhoungTran for engineering and testing support.

References

[1] Kaufman HR, Harper JME, Guomo JJ. J Vac Sci Technol 1979;16:899.[2] Kaufman HR. J Vac Sci Technol 1986;A4:764.[3] Huth Ch, Scheer HC, Schneemann B, Stoll HP. J Vac Sci Technol 1990;A8:4001.[4] Korzec D, Engemann J. J Vac Sci Technol 1989;B7:1448.[5] Hwang JH, Jin G, Chen X. US patent # 6821907; November 2004.[6] Jung KB, Marburger J, Sharifi F, Park YD, Lambers ES, Pearton SJ. J Vac Sci

Technol 2000;A18:268.[7] Matsui N, Mashimo K, Egami A, Konishi A, Okada O, Tsukada T. Vacuum

2002;66:479.[8] ANELVA GIHO, vol. 9; March 15, 2002.[9] Kodaira Y, Hiromi T. US patent application number, 0016957 A1; 2005.

[10] Castano FJ, Haratani S, Hao Y, Ross CA, Smith HI. Appl Phys Lett 2002;81:2809.[11] Katine JA, Ho MK, Ju YS, Rettner CT. Appl Phys Lett 2003;83:401.[12] Cho H, Lee K-P, Jung KB, Pearton SJ, Marburger J, Sharifi F, et al. J Appl Phys

2000;87:6397.[13] Jung KB, Cho H, Hanh YB, Hays DC, Lambers ES, Park YD, et al. J Electrochem

Soc 1999;146:1465.[14] Saito M, Touno I, Omiya K, Homma T, Nagatomoa T. J Electrochem Soc

2002;149:G451.[15] Kim YS, Ji HY, Han KS, Song CW. Jpn J Appl Phys 1998;37:L1336.[16] Jung KB, Hong J, Cho H, Onishi S, Johnson D, Park YD, et al. J Electrochem Soc

1999;146:2163.[17] Orland AS, Blmenthal R. J Vac Sci Technol 2005;B23:1589, 1597.[18] Nakatani I. IEEE Trans Magn 1996;32:4448.[19] Woo SG, Kim SH, Ju SY, Son JH, Anh J. Jpn J Appl Phys 2000;39:6996.