Understanding Atomic Layer Deposition:

Successes & Limitations and Recipe Development

J Provine & Michelle Rincon November 1, 2012 Stanford University

NNIN ALD Roadshow (special thanks to Ganesh Sundaram, Eric Deguns,

Prof. H. Brongersma, and Prof. Fritz Prinz)

What is ALD? aka How does it work?

• First let’s look atomistically. – A rose by any other name…

• Then we’ll look at what that means in ways we can measure. – Microscopically? Nanoscopically? Angstroscopically?

• Then briefly back at the atomic level to see behind the curtain.

In air H2O vapor is adsorbed on most surfaces, forming a hydroxyl group. With silicon this forms: Si-O-H (s)

After placing the substrate in the reactor, Trimethyl Aluminum (TMA) is pulsed into the reaction chamber.

Tri-methyl aluminum Al(CH3)3(g)

C H

H H

H

Al

O

Methyl group (CH3)

Substrate surface (e.g. Si)

ALD Example Cycle for Al2O

3 Deposition

Al(CH3)3 (g) + : Si-O-H (s) :Si-O-Al(CH3)2 (s) + CH4

Trimethylaluminum (TMA) reacts with the adsorbed hydroxyl groups, producing methane as the reaction product

C

H

H

H

H

Al

O

Reaction of TMA with OH

Methane reaction product CH4

H

H H

H H C

C

Substrate surface (e.g. Si)

ALD Cycle for Al2O3

C H H

Al

O

Excess TMA Methane reaction product CH4

H H C

Trimethyl Aluminum (TMA) reacts with the adsorbed hydroxyl groups, until the surface is passivated. TMA does not react with itself, terminating the

reaction to one layer. This causes the perfect uniformity of ALD. The excess TMA is pumped away with the methane reaction product.

Substrate surface (e.g. Si)

ALD Cycle for Al2O

3

C H H

Al

O

H2O

H H C

O H H

After the TMA and methane reaction product is pumped away, water vapor (H2O) is pulsed into the reaction chamber.

ALD Cycle for Al2O

3

2 H2O (g) + :Si-O-Al(CH3)2 (s) :Si-O-Al(OH)2 (s) + 2 CH4

H

Al

O

O

H2O reacts with the dangling methyl groups on the new surface forming aluminum-oxygen (Al-O) bridges and hydroxyl surface groups, waiting for a new TMA pulse.

Again metane is the reaction product.

O Al Al

New hydroxyl group

Oxygen bridges

Methane reaction product

Methane reaction product

ALD Cycle for Al2O

3

H

Al

O

O

The reaction product methane is pumped away. Excess H2O vapor does not react with the hydroxyl surface groups, again causing perfect passivation to one atomic layer.

O O Al Al

ALD Cycle for Al2O

3

One TMA and one H2O vapor pulse form one cycle. Here three cycles are shown, with approximately 1 Angstrom per cycle. Each cycle including pulsing and pumping takes e.g. 3 sec.

O

H

Al Al Al

H H

O O

O O O O O

Al Al Al O O

O O O

Al Al Al O O

O O O

Al(CH3)3 (g) + :Al-O-H (s) :Al-O-Al(CH3)2 (s) + CH4

2 H2O (g) + :O-Al(CH3)2 (s) :Al-O-Al(OH)2 (s) + 2 CH4

Two reaction steps in each cycle:

ALD Cycle for Al2O

3

Measurement Technique: Ellipsometry

• Measure change in reflected polarization

• Fast and extremely accurate (<1Å) if material model is known

• Stacks of varying films and absorptive films can cause issues

• Average over a large area

Microscopically (Nanoscopically?) What makes a deposition ALD?

• Step One: Linear growth rate

• Is anything wrong here?

y = 1.259x

0

200

400

600

800

1000

1200

1400

0 200 400 600 800 1000 1200

Th

ick

ne

ss (

Å)

Number of Cycles

• Step Two: Self-limiting deposition/cycle

Microscopically (Nanoscopically?) What makes a deposition ALD?

Saturation Curve at 250°C

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

0.0 0.5 1.0 1.5 2.0 2.5

Gro

wth

Pe

r C

ycle

Precursor Dose (seconds)

ALD “Window”

ALD

Window

Temperature

Desorption limited

Condensation limited

Activation energy limited

Decomposition limited

Growth Rate

Å/cycle

Saturation

Level

Each ALD process has an ideal process “window” in which growth is saturated at a monolayer of film.

Outside the ALD Window

0.9

0.95

1

1.05

1.1

1.15

1.2

1.25

1.3

125 175 225 275

Growth per cycle (A)

Temperature (qC)

Thermal Alumina Plasma Hafnia

Outside the ALD Window

• If not all of the reactants are removed from the chamber during the purge…

• An easy indication of this is a loss in film uniformity.

ZrO2 Uniformity Optimization Normal Chamber Thickness

Map

Average Thickness vs. Pulse Time

16

0102030405060708090

100

0 0.1 0.2 0.3 0.4

Th

ick

ne

ss (

Å)

Precursor Pulse Time (s)

Non-Uniformity vs. Pulse Time

Silicon Wafers

0

20

40

60

80

100

120

140

0 0.1 0.2 0.3 0.4

No

n-u

nif

orm

ity

(% o

f a

ve

ra

ge

th

ick

ne

ss)

Precursor Pulse Time (s)

Measurements in Angstroms

ZrO2 Uniformity Optimization Under-Dosed

Chamber Thickness Map

Average Thickness vs. Pulse Time

17

0102030405060708090

100

0 0.1 0.2 0.3 0.4

Th

ick

ne

ss (

Å)

Precursor Pulse Time (s)

Non-Uniformity vs. Pulse Time

0

20

40

60

80

100

120

140

0 0.1 0.2 0.3 0.4

No

n-u

nif

orm

ity

(% o

f a

ve

ra

ge

th

ick

ne

ss)

Precursor Pulse Time (s)

Silicon Wafers

Thickness drop-off

High non-uniformity

Measurements in Angstroms

ZrO2 Uniformity Optimization Chemical Vapor Deposition

(CVD) Chamber Thickness Map

Average Thickness vs. Purge

Time

18

0

20

40

60

80

100

0 10 20 30 40

Th

ick

ne

ss (

Å)

Purge time after

precursor pulse (s)

Non-Uniformity vs. Purge Time

0

0.5

1

1.5

2

2.5

3

3.5

0 10 20 30 40

No

n-u

nif

orm

ity

(% o

f a

ve

ra

ge

th

ick

ne

ss)

Purge time after

precursor pulse (s)

Silicon Wafers

Measurements in Angstroms

HfO2 Uniformity Optimization

Normal ALD Process Non-Uniformity vs. Pulse Time

19

Substrate

Precursor-saturated surface

00.20.40.60.8

11.21.41.6

0 10 20 30 40

No

n-u

nif

orm

ity

(% o

f a

vg

th

ick

ne

ss)

Purge time after

precursor pulse (s)

Film

Purge time is too long

Precursor-saturated surface Precursor-saturated surface Precursor-saturated surface Precursor-saturated surface

Desorption! Average Thickness vs. Purge Time

0

20

40

60

80

100

120

0 10 20 30 40T

hic

kn

ess (

Å)

Purge time after

precursor pulse (s)

Lower thickness

Higher Non-uniformity

What is going on atomistically?

• Compositional analysis – Very complete material map – Accuracy .1-10% depending

on material • Weighted average over a

depth of ~10nm • Lateral spot size is typically

>10µm (system dependent)

• Secondary ions emitted from the surface (and sub-surface)

• Extremely accurate trace contamination

• Depth profiling • Significant damage to surface • Interface mixing

What is going on atomistically? We need a technique to look 1 atom deep.

LEIS and The Hypocratic Oath

LEIS on an ALD film of ZrO2

LEIS on an ALD film of ZrO2

The little secret of ALD…if you know this, you know more than 90% of the people I’ve talked to about ALD.

Platinum Nucleation

0

50

100

150

200

250

0 200 400 600 800 1000

Film

Thi

ckne

ss (Å

)

# of Cycles

Thermal 270°C

Thermal 100°C

Pt from MeCpPtMe3 and O

2

Plasma Enabled (PE)ALD

ͻ Remote Plasma as a reactant – Widens ALD window for materials by decreasing

activation energy – Avoids precursor decomposition or damaging

substrates with limited thermal budget – Remote ICP source prevents substrate damage from

ions – Faster deposition cycle times – Fewer contaminates in films – Smaller nucleation delay

ͻ Film Examples

– Low temperature oxides – Metal nitrides – Metallic films

• High-Aspect Ratio Structures

– Radical recombination prevents greater than ~20:1

Fiji PE-ALD chamber

Benefits of Plasma

• Precursor temp 90°C

• Nucleation is eliminated when using O2 plasma as reactant

• Constant growth rate per cycle even at low process temp

Decreased nucleation for metallic films

0

50

100

150

200

250

300

350

400

0 200 400 600 800 1000

Film

Thi

ckne

ss (Å

)

# of Cycles

Thermal 270°C

Plasma 270°C

Thermal 100°C

Plasma 100°C

Pt from MeCpPtMe3 and O2

Pt Nucleation Pt from MeCpPtMe

3 and O

2

Thermal vs Plasma ALD

Plasma ALD Pt on Al2O

3 Plasma ALD Pt on thermal SiO

2

Nanolaminate Formation / Doping

� “Doping” in ALD � Doping is carried out by substituting a pulse of

precursor M1 with dopant precursor M2

� This allows exploration of a wide concentration window without having to prepare new targets for each concentration (as in sputtered depositions)

� Grated films with uniform properties possible

� Annealing films to “activate” dopants is not required

Substrate

M1

M2

M1

Dopant cycle M2

Summary • The Canonical view of ALD is valuable for

understanding, but it does not provide full understanding

• A more complete understanding is useful – Selective deposition – Surface preparation – Troubleshooting

• For analysis there is no magic bullet – Best to heavily armed – The right analysis for your interest

Questions?

On to Process Development

High quality recipe example: MgO

High quality recipe example: MgO

High quality recipe example: MgO

Guard against “unstable” precursors

Sufficient dose is needed

System Purge

• Reactant X must be fully cleared before reactant Y is introduced to the system – Otherwise CVD-like growth

• Purge time depends on temperature (and chemical)

Exposure Mode • For high aspect ratios or to improve deposition

uniformity/consistency • Close stop valve before pulse and allow

saturation time before purge

Boost for Low Volatility Precursors

Summary • ALD can be an extremely stable, repeatable,

controllable process • However, good recipe characterization and

precursor selection is essential • Resources

– Academic literature – Cambridge Nanotech (many well established recipes) – ALD work at stanford:

snf.stanford.edu/SNF/equipment/chemical-vapor-deposition/ald/

– jprovine@stanford.edu; mmrincon@stanford.edu

Finally: Some Things Learned During Roadshow

• Equipment – University of Maryland:

Beneq – UC Berkeley: Fiji and

Picosun – UT Austin: Savannah and

Fiji (w/ turbo pump) – Arizona State: Savannah – Cornell: Oxford – Washington (Seattle):

Oxford – GA Tech: Fiji – Harvard: Fiji and Savannah – Utah: Fiji

• Processes – Maryland: thermal

vanadium oxide – UC Berkeley: thermal

Ru – UT Austin: thermal

BeO2