EMT

395 Photonic Fabrica3on

Engineering
E-beam lithography (EBL)
Dr. Alaeddin A. Saif

Semester 2
Academic session 2015/2016
1

E-beam lithography (EBL) overview

(direct wri:ng with a focused e-beam)

Use resist like op:cal lithography, but resist exposed by electrons.

Posi:ve resist by polymer chain cuEng, nega:ve by cross-linking or polymeriza:on.
Electron beam is focused to spot size <5nm using electron op:cs.
Very small wavelength: resolu:on less limited by dirac:on.
Generate paLern by direct wri:ng: ne need of mask.
Sequen:al pixel-by-pixel wri:ng: low throughput , unsuitable for mass produc:on.

1.226
For electron:
=
(nm) For EBL at 30kV accelera:on voltage
=0.007nm
(V is electron kine:c
V
energy in eV)

hc 1.24
=
=
( m)
For light:

eV

For an electron with kine:c energy of 1eV, the associated DeBroglie wavelength is 1.23nm,
about a thousand :mes smaller than a 1eV photon.
(Note: electron rest mass energy is mc2=511keV, so rela:vity is unimportant for <50kV)
2

Exposure of resist

Typical energy for breaking a bond: 10eV

But typical energy of the beam: 10-100kV
(problems of aberra:on at low energy that leads to large beam spot size
and low resolu:on, so use high energy for EBL)
Bond is broken by secondary (including Auger) electrons with low energy.
3

E-beam lithography facts

Developed in 1960s along with scanning electron microscope (SEM).
Breakthrough made in 1968 when a polymer called PMMA (poly
methyl meth acrylate) was discovered to have high resolu:on.
Fast growth in 1990s when nano began to become hot and
computer became more available for automa:c lithography control.
Since around 2000, focused ion beam (FIB) paLerning began to
compete with EBL in some applica:ons.
Today EBL is s:ll the most popular nano-paLerning techniques for
academic research and prototyping.

SEM/EBL system components

An electron gun or electron source
that supplies the electrons.
An electron column that 'shapes'
and focuses the electron beam.
A mechanical stage that posi:ons
the wafer under the electron
beam.
(op:onal) A wafer handling system
that automa:cally feeds wafers to
the system and unloads them acer
processing.
A computer system that controls
the equipment.

EBL systems: most research tools are based on SEM

SEM conversion
Conven:onal SEM (30kV)
Almost no SEM
modica:on
Add beam blanker
Add hardware controller
Low cost: <$100K

Dedicated EBL system

Based on SEM system
With perfect integra:on
Interferometer stage
Focus correc:on (laser
sample height control)
Cost $1-2M

E-beam writer
High energy column (100kV)
Dedicated electron op:cs
High reproducibility
Automa:c and con:nuous
(over few days) wri:ng
High cost (>$5M)

NPGS system Raith system Vistec system

Electron guns/source
Schema:c structure of electron gun
Electrons can be emiLed from
a lament (emiLer or cathode)
by gaining addi:onal energy
from heat or electric eld.
C: cathode for emiEng electrons
E: extrac:on electrode
A1, A2: cathode lens electrode to focus the emiLed electrons
Three types of electron guns:
Thermionic emission gun (W, LaB6, not-sharp :p).
Field emission gun (cold, very sharp W :p, tunneling current).
SchoLky gun (eld assisted thermionic emission, sharp :p).
Whether it is eld emission or not depends on the electric eld near the :p apex,
which determines whether tunneling is important or not.
Sharper :p leads to higher electric eld near :p apex, so eld emission (by tunneling)
plays a major role, it is thus called eld emission gun (FEG).
Even thermionic emission relies on the electric eld from the extrac:on electrode, but
7
here thermionic emission plays a major role.

Electron gun: thermionic emission

(tungsten hairpin laments)
The long :me source of choice has been the W
hairpin source
Working at high temperature, some electrons have
W lament
thermal kine:c energy high enough to overcome the
energy barrier (work func:on, but kT s:ll << work
Vacuum level
vacuum level
func:on).
Escaped electron is then extracted by the electric
eld generated by the nearby extrac:on electrode.
work
Thermionic
Work
thermionic
Current density Jc depends on the temperature and function
electrons
electronic
func:on
cathode work func:on .
eV
(eV)
Cheap to make and use ($12.58 ea) and only a
modest vacuum is required. Last tens of hours.
conduction band

Besides W, single crystal LaB6 is another popular :p

material for thermionic emission guns. About 5-10 more
expensive than W, but last 5-10 longer and is brighter,
but higher vacuum is required (since LaB6 is very reac:ve).

Schema:c model
of thermionic
emission

Field emission guns (FEGs)

Field emiLer
tunneling

(But FEG :p material is NOT semiconductor)

Field emission (i.e. tunneling) becomes important for electric eld F>107V/cm.
Need very high vacuum to prevent arc-over at :p apex.
Strong nonlinear current-voltage characteris:c.
Very short switching :me (t<ns).
Small beam spot size, since eld is high enough for tunneling only near :p apex.

Cold eld emission gun (FEG) behavior

The :p must be very clean to perform properly as a eld emiLer.
Even at 10-6Torr, a monolayer of gas is deposited in just 1 sec.
So :p needs higher vacuum, 10-10Torr vacuum.
At this vacuum, the :p is usually covered with a mono- layer of gas in 5-10 minutes.
Cleaning is performed by ashing - hea:ng the :p for a few seconds to desorbs gas.
The emission then stabilizes for a period of 2-5 hours.
On the stable region (hour 4 to hour 6), total noise + dric is a few percent over a few
minutes, s:ll not stable. (Right acer ashing, current may drop 50% within a hour)
Flash is typically done automa:cally every morning, and SEM is good for 8-10 hours.
For e-beam lithography that need more stable current, good only during hour 4 to hour 8.
Because of the current instability, cold FEG is not good choice for e-beam lithography,
though it is the best for SEM imaging applica:ons.
Cold FEG is more expensive than SchoLky emission guns, but last longer, up to 5 years.

10

SchoLky emiLers: eld assisted thermionic source

It is usually misleadingly called thermal or
SchoLky eld emission guns.
But it is not a truly eld emission gun,
because the :p is blunt and if the heat is
turned o there is no emission (tunneling)
current.
A SchoLky source is actually a eld assisted
(to lower ) thermionic source.
SchoLky emiLers can produce larger amounts
of current compared to cold FEG systems, so
more useful for e-beam lithography.
Because they are always on (hot), organic
contamina:on is not an issue, hence they are
very stable (few % per week change in current)
They eventually fail when the Zirconia
reservoir is depleted, within 1-2 years.

Hitachi Schottky Emitter Tip

11

Source size
The cross-over is an eec:ve real or virtual source
for the downstream electron op:cal system.
(real source)

Cold eld
emission gun

(cold and thermal FEG)

The source size is the apparent width of the disc from which the
electrons appear to come.
The :p physical size does NOT determine the source size.
Small is good for high resolu:on SEM, because less
demagnica:on is needed to aLain a given probe size.
But too small is not necessary, because anyway demagnica:on
is needed to minimize eects of vibra:on and stray elds.

12

Source brightness

4Ib
= 2 2 2
d

Spot
Diameter
d

Brightness is dened as current per unit area per

solid angle, with unit amp/cm2/steradian.
Brightness is the most useful measure of gun
performance.
Brightness varies linearly with energy, so one
must compare dierent guns at the same beam
energy (accelera:on voltage).
Beam
High brightness is not the same as high current.
current
E.g. thermionic emission can have very high beam
Ib
current, but low brightness (due to large d).
Most current will then be blocked by an aperture
Convergence (to limit ) in order to have an acceptable small
angle
beam spot onto the specimen for high resolu:on

imaging.

Measuring at the specimen

13

Comparison of electron emission sources

Key parameters of electron sources:
virtual source size, brightness, energy spread of emiLed electron

(ashing)

*Hitachi cold FEG SEM can go to 2nA.

14

Raster scan vs. vector scan

Raster scan:
The e-beam is scanned in only
one direc:on, and the stage is
mechanically translated in the
perpendicular direc:on.
Beam blanker: parallel plate
with voltage 42V that can
deect (turn o) the beam.

Vector scan:
The e-beam is scanned in both x- and ydirec:ons with beam blanking, wri:ng the
paLern pixel-by-pixel.
No stage movement within each wri:ng eld.
Acer each wri:ng eld, the substrate/stage
moves to the next loca:on.

42V

15

Raster scan versus vector scan

Raster scan:
Very simple and fast.
Very repeatable
But sparse paLerns take as long as dense paLerns.
Dicult to adjust dose during wri:ng.
For photo-mask making.

Raster scan

Vector scan:
Fast wri:ng of sparse paLerns (unwriLen
areas skipped).
Easy dose varia:on from shape to shape.
SeLling :me & hysteresis, need to wait at
beginning of each paLern.
For nanolithography and R&D.

SeLling :me:
Wai:ng period at
beginning of each element

Vector scan
16

Interac:ons of electrons with maLer

Unlike X-rays, electrons interact very strongly with maLer
Elas:c scaLering of electrons can be treated classically as a form of
Rutherford backscaLering
Inelas:c scaLering arises from numerous interac:ons:
o Low energy secondary electrons genera:on.
o Inner shell excita:ons (x-ray uorescence and Auger electrons)
o Electron-hole pair crea:on and recombina:on
o Phonon excita:on (heat)

17

Informa:on from electron beam-specimen interac:ons

When electron beam strikes the sample, both photon and electron signals are emiLed.

18

Excita:on depth prole and its relevance for SEM

Secondary
electron
(detector)

SED

Primary
electron
beam

BackscaLered
electron

BSE

CL
Cathode luminescence

Dierent emission mechanisms

arise from dierent depths.
For SEM imaging, both SE and BSE can be used, but with dierent informa:on.
BackscaLered electrons (BSE):
Specular reec:on (of primary electron)
Higher energy (so travels further)
Low resolu:on (since travels further)
Encode some chemical informa:on
(material contrast, high Z high signal)

Secondary electrons (SE):

Isotropic emission
Low energy
High resolu:on
BeLer structural contrast
19

Energy-dispersive x-ray
spectroscopy (EDS, EDX)
Energy levels: En

2
(
Z )
=C

n2
C is a constant depending on e, h, 0.
1 is due to screening (=0 for ideal model).
n=1 (K-shell), 2 (L-shell), 3 (M-shell).

SEM is ocen equipped with EDX that detect x-ray emiLed (needs only extra $50K).
Each element has its characteris:c x-ray, so EDX is for elemental analysis.
Since both primary electrons and emiLed x-ray can penetrate certain depth (up to
many m), EDX is not for top surface analysis.
The principle is the same as electron impact x-ray source, but here a focused electron
beam is used to impact or bombard the sample, thus capable of mapping surface
element with high spa:al resolu:on.
Light element (Z4) usually cannot be detected, since their characteris:c energy is
lower and thus cannot go through the x-ray detector window.

20

EDX analysis of a material

The peak for W could be L (K energy is much higher)

21

Forward/back scaLering events

Scattering: spreading of the beam, lost of resolution

Resist

Substrate
Forward scaLering
Proper:es:
Very ocen
Small angle
Very inelas:c (i.e. lose energy)
Genera:on of SE with low energy.

Back scaLering
Proper:es:
Occasionally (collision with nucleus)
Large angles, thus mainly elas:c
High energy, same range as primary electrons.
Large travel length, cause proximity eect.

SE with few eV are responsible for most resist exposure. Such SE diuses laterally few nm.
22
BackscaLering is responsible for resist exposure far from incidence (proximity eect).

Monte-Carlo simula:ons of electron trajectory

ScaLering probability varies as square of atomic number Z, and inversely as the incident kine:c energy.
Greater penetra:on
depth for low Z
materials.
BSE emission branch
increases with Z.

Low Z

High Z

Number of backscaLered electrons is not

dependent on energy, but its spa:al distribu:on is.
Proximity eects are diluted (spread over larger
area) at high energies.
23

(Sensi:vity)

(penetra:on depth)

Eect of voltage on dose

At small kV, penetra:on depth is low, so cannot expose thick resist. (e.g. at 0.8kV,
penetra:on depth only 40nm in PMMA).
At >2kV, resist sensi:vity is higher for lower kV, so faster wri:ng.
But lower kV has larger beam spot size and more serious forward scaLering, both of
which reduces resolu:on.
In addi:on, lower kV has lower aLainable beam current that reduces wri:ng speed.
Therefore, typical EBL is done at 30kV or higher.
24

Proximity eect

Real dose D=E(1+b)

E is as-exposed dose
b is due to proximity eect
b2 (not small) for large area exposure
Proximity eect is negligible for isolated/sparse ne features.
It is good for areal exposure (e.g. pad >>1m), since pixel can be much larger than beam
spot size (right gure). E.g., beam step size (pixel) of 50nm is usually enough even with a
beam spot size only 5nm.
25
Proximity eect is worst for dense and ne paLerns, such as gra:ng with sub-50nm pitch.

Resist prole
A thin layer may be developed due
to exposure by proximity eect
Due to forward scaLering and (to a less
degree) proximity eect, posi:ve resist has
always an undercut prole, good for lico.
Resist
Nega:ve resist always has a tapered
Substrate
prole, bad for lico.
For paLerning dense ne features, an
undercut prole ocen causes resist
structure to collapse due to capillary force
when developer is dried.

Original thickness

Developed prole

Posi:ve resist

That is, proximity eect makes paLerning

dense ne features dicult.
Dense AND ne structures
Resist (posi:ve)
prole, not stable

substrate

Nega:ve resist

26

EBL resist: sensi:vity and contrast

Resist development curves:
(a) Resist A is of higher
sensi:vity than B.
(b) A is of higher contrast than
B; C is nega:ve resist.

Here D1 and D0 is dened for B

nega:ve

posi:ve

thickness
substrate
Sensi:vity:
For posi:ve resist: D1 value, or dose required to fully develop the resist to boLom, close
to D1 value.
For nega:ve resist: dose that results in half resist thickness remaining acer development.
Contrast : dened as slope of the development curve.

log10 (D1 / D0 )

For perfect resist, D1=D0, so =.

Usually >2.0 is good
For PMMA, =5-10.

27

Posi:ve vs. nega:ve resist

Posi:ve resist (trench)

Nega:ve resist (line)

Posi:ve: long molecular chains are broken by energized electrons into short chain
(chain secession).
Nega:ve: ini:al short chain molecules are joined upon exposure to form long chains
(cross-linking), so become insoluble in developer.

28

Which resist to choose

Which resist to choose depends on which will give the minimum exposure area.
For isolated sparse features, posi:ve resist is suitable for lico process, while nega:ve
for direct etch process.
To paLern Si (or SiO2), either Cr lico then RIE using Cr as mask, or direct etch of Si
using resist as mask will work, though Si can be etched deeper with more ver:cal
sidewall using Cr as mask.
Lico process using posi:ve resist
resist
(polymer)

Direct etch process using nega:ve resist

resist
(polymer)

1. Spin on posi:ve resist

1. Spin on nega:ve resist

2. EBL

2. EBL

3. Cr deposi:on

3. RIE substrate

4. Lico Cr
5. RIE substrate using Cr as mask (not shown)

29

Resist tone reversal process

Very useful if one has only one cheap posi:ve resist: PMMA
Resist
Under-layer (polymer)
substrate
Prole
in resist

substrate
substrate

Prole in
underlayer

substrate

substrate
Acer metal lico

For posi:ve resist, trench (or

hole) formed acer EBL.
Lico a hard material (Cr, SiO2)
RIE under-layer.
The resulted polymer structure is a line
(or pillar), as if nega:ve resist is used.

Hole array in
100nm-thick Au
with pitch 600nm,
by lico using
PMMA resist. 30

Sensi:vity

Sensi:vity depends on:

1. Electron energy keV (or accelera:on voltage kV). Higher energy requires higher
exposure dose, so lower sensi:vity.
2. For example, sensi:vity for PMMA is 250C/cm2 with 30kV, but 600 C/cm2 with
100kV (high energy electrons pass fast through the resist, genera:ng fewer low
energy secondary electron (SE) to expose the resist, or they are just not ecient to
generate SE due to the large energy dierece.
3. Substrate material: high density substrate (high Z) material results in more back
scaLering of electrons into resist layers, leading to higher sensi:vity (e.g. PMMA on
Au sensi:vity is 2 that on Si). But of course more severe proximity eect.
4. Process condi:ons: post exposure baking condi:ons for chemically amplied resist;
strength of developer, development temperature and :me.
Note: higher sensi:vity means one needs lower dose to fully expose the resist.
31

Contrast
Low contrast resist prole

Contrast curve of SU-8

Contrast only 0.92

High contrast resist prole

High contrast:
Steeper sidewalls
Greater process la:tude
BeLer resolu:on
Higher aspect ra:o structure
Less sensi:vity to proximity eect, high density paLern.
Low contrast: good only for 3D gray scale lithography

32

The standard EBL resist: PMMA (posi:ve)

The most popular e-beam resist, very cheap and last forever, easy handling.
Dont buy it (expensive), mix them by yourself.
Very high-resolu:on and contrast.
Typical molecular weight is 950kg/mol. Lower Mw (e.g. 15kg/mol) leads to higher
sensi:vity and lower contrast.
Usually dissolved in a solvent: chlorobenzene or anisole (less toxic, 2-4%).
Developer mixtures can be adjusted to control contrast and sensi:vity.
The downside: low sensi:vity, poor dry etch resistance (good for lico, not for direct
etch paLern transfer).

33

PMMA developer

(Methyl isobutyl ketone)

1. MIBK : IPA (isopropanol)=1:3 for typically 60sec, most popular developer.

2. Cellosolve (2-ethoxyethanol): methanol=3:7 for 7-10sec, claimed by some to
have slightly higher contrast than MIBK.
3. MEK (methyl ethyl ketone) : ethanol=26.5:73.5 for 2-5 second.
4. IPA : H2O=7:3, co-solvent system, i.e. neither IPA nor water alone dissolves
exposed PMMA. Claimed by some to have beLer performance than MIBK.

34

PMMA dose table

10kV

20kV

30kV

Area dose (C/cm2)

100

180

250

Line dose (nC/cm)

0.5

0.9

1.3

Dot dose (fC)

1.5

Doses for MIBK:IPA=1:3 developer 60second.

All values are good star:ng points, need dose test before each wri:ng.
Use above area dose only for large features (>proximity eect range).
Otherwise, e.g. when wri:ng 1m stripes, 250C/cm2 is not enough.

Proximity eect factor b can be es:mated roughly from the above table:
Real dose D=E(1+b)
E is as-exposed dose
b is due to proximity eect
b2 (not small) for areal exposure

For line dose 1.3nC/cm, wriLen line-width is 15nm.

So area dose 1.3nC/(15nm1cm)=867C/cm2.
Therefore, 1+b=867/250, and b=2.5.
B=1.7 if es:mated from the dot dose ((15nm)2 dot).
35

The most popular commercial resist: Zep-520 (posi:ve)

Developed by ZEON in Japan to replace PMMA.
Higher sensi:vity (3-5x faster), and higher etch resistance (3x)
For ultrahigh resolu:on (sub-10nm), PMMA might s:ll be beLer.
Expensive: $1000/100ml.
One-year shelf :me, making it more expensive compared to PMMA.
Composi:on: methyl styrene/chloromethyl acrylate copolymer.

Developer:
ZED-N50 (100% n-Amyl Acetate)
Xylene (o-,m-,p- mixed)

Solvent:
Anisole, for lico or dilu:ng the resist
for thinner lm.
Line in ZEP-520 wriLen at 75keV.
Resist thickness 1.5m, line-width 50nm
36

HSQ (nega:ve)
Silicon dioxide based inorganic material (not polymer).
Sensi:vity and contrast similar to that of PMMA (depends on developer strength).
Very high resolu:on and very dense paLern when using <25nm-thick lm.
Exposed HSQ is in the form of amorphous oxide, good etching mask.
Product of Dow Corning under product code Fox12.
It is an unusual resist: development by chemical reac:on (un-exposed HSQ reacts with
diluted NH4OH or NaOH developer to produce H2), not by dissolu:on; and
development saturates (i.e. no more reac:on) acer a certain :me.
Salty developer (add NaCl to NaOH solu:on) increases contrast.

37

SU-8 (very high sensi:vity, but low contrast)

Chemically amplied nega:ve tone resist
Epoxy-based, mostly used as photo-resist
Extremely high sensi:vity over 100x that of PMMA
Low contrast (0.9), unsuitable for dense paLerning

10keV

High resolu:on possible for sparse paLerns at high kV

Rough edges and residues due to random exposure from
back scaLering electrons and random photo-acid diusion.
Ideal for low resolu:on wri:ng over large area (fast).
50keV

100keV

24nm line at pitch 300nm in 100nm thick SU-8

38

SU-8 resist formula:on and process

Glass transi:on temperature of SU-8

Un-cross-linked: 50oC.
Fully cross-linked: 230oC

Like all chemically amplied resist, post exposure baking temperature and
:me is very cri:cal.
Typically 90oC for 2-3min on hotplate.
Spin-coa:ng, bake, exposure, post-bake and development need to be done
quickly without much delay.
Dont let SU-8 exposed to room light for long (will be exposed by UV light).
39

Undercut prole for lico

Metal lico process

Process of double layers resist for easy lico

40

Resolu:on enhancement process: ultrasonic development

Ultrasonic helps to remove exposed resist (for posi:ve tone) from inside narrow trenches.

41

Resolu:on enhancement process:

low temperature development (ZEP)
Contrast curve
for ZEP-520

Contrast as a
func:on of
development
temperature.

Low T development increases contrast,

but decreases sensi:vity

Comparison of edge roughness of ZEP-520

resist lines (40nm wide) developed at
(top) room temperature; (boLom) -4o42
C.

Resolu:on enhancement process: low

temperature development (PMMA)
Figure 2-6: Schema:c illustra:on of one possible explana:on
of resolu:on-enhancing mechanism of cold development.
1) When a feature is exposed in PMMA, the soluble resist in
the exposed region is surrounded by a boundary region of
resist that, due to the ini:al polydispersity of the PMMA
and random nature of chain scission, contains both
soluble and insoluble polymer chains.
2) During development, this region phase-separates, with
the soluble chains diusing toward the soluble region and
the insoluble chains diusing toward the insoluble region.
3) The result is a region of soluble PMMA that is larger than
the ini:al exposed feature, resul:ng in a degrada:on in
resolu:on.
Cold development helps prevent this by limi:ng the diusion
that can occur in the boundary region, since diusion is a
thermally-dependent process.

The exact mechanism of cold development is s:ll not clear.

Another possible mechanism: colder developer is weaker
solvent, so less aLack to unexposed or partly exposed resists.
43

Charging during e-beam wri:ng

Even though most resists are insula:ng, charging is not an issue for typical resist
thickness of <500nm, because most electrons penetrate deep into the conduc:ng
substrate at 30kV. (more serious charging for lower kV)
When electron beam lithography must be performed on insula:ng substrates
(quartz, SiO2/Si), nega:ve charge buildup can occur on substrate surface, causing
beam deec:on, and thus paLern distor:on.
Ini:al Condi:on

Future Condi:on
electron beam

electron beam

repulsive electric
poten:al lines

ma-N 2403 nega:ve resist

Glass substrate

nega:ve charge
accumula:on

44

An:-charging technique: coat conduc:ng layer

To eliminate charging eect, one can coat a conduc:ng layer on top of or
beneath the resist.
Typically 10nm metal is enough, such as Al, Ti, Cr or Au; conduc:ng polymer may
also work.
Lighter metal (Al) causes less (forward) scaLering of electron beam than the
heavier one (Au), so is preferred.
Al and Ti can be removed easily by diluted HF (1:100 diluted).
Some resist (such as SU-8) is sensi:ve to UV light generated during e-beam
evapora:on (thermal evapora:on or spuLering is OK).
GOOD
(no charging)
Coat metal

BAD
(moderate charging)
ESpacer > 24 hours
(a conduc:ng polymer?)

BAD
(severe charging)
no an:-charging layer

45