From microelectronics down to nanotechnology

Download Report

Transcript From microelectronics down to nanotechnology

From microelectronics down
to nanotechnology
[email protected]
Contents
•
•
•
•
•
•
•
Lithography: scaling x- and y-dimensions
MOS transistor physics
Scaling oxide thickness (z-dimension)
CNT transistors
Conducting wires (metals and CNTs)
Memories
Moore’s law and fabrication economics
Top down nanotechnology
Scaling from micrometer down
• Production-proven techniques
• Laboratory techniques
• Laboratory tricks
Issues to be tackled:
• Cost
• Area
• Speed
• Defect density
Writing patterns
By pen
• easy to write anything
• slow to write long works
• easy to change your mind in the middle
By printing press
• very fast for large number of copies
• expensive to make the first copy
• expensive to make changes
Pen vs. printing press
• AFM needle
• electron beam
• optical projection
• nanoimprint
Lithography
Printing press style:
• optical: mainstream
• EUV/X-ray: small linewidth but otherwise problematic
• imprint: partial solution at the moment; only certain
aspects proven
Pen-like:
• electron beam: slow writing speed
• ion beam: very slow writing speed, ultimate resolution
• AFM, dip-pen,...
The goal of lithography is to make lines
and spaces small (only this will
increase device packing density)
Contact/proximity lithography
light from
light source
quartz mask with
chromium pattern
gap
photoresist covered
silicon wafer
Resolution= line + space
Resolution =3 ( /n)* (gap + ½ thickness )
•  (Hg-lamp line)
436 nm
• gap between mask and resist
g  5 µm
• d resist thickness
d  1 µm
• n resist refractive index
n  1.6
• 1 µm in production
• 100 nm in research
Projection lithography
Sources of radiation
(UV 365 nm-436 nm,
DUV 193 nm-248 nm,
EUV, X-rays, electrons,
ions)
Optical system I
(lenses, mirrors)
Mask (pattern)
Optical system II
(lenses, mirrors)
Numerical aperture NA=sin

Imaging medium (resist)
Wafer (with patterns)
Wafer stage
(alignment mechanism)

Resolution = k1 /NA
Depth of focus = k2 /NA2
Resist trimming trick
• Isotropically etch resist
thinner & narrower lines
line-to-line spacing unchanged
Works best for
narrow initial lines
Used in industry
Phase shift masks (PSM)
binary mask (quartz/chrome)
phase shift mask (PSM)
amplitude
shifter
Conditions for
phase shifter:
= 2L/
= 2nL/.
intensity
resist exposure threshold
L(n-1) = /2
Phase Shift Mask Fabrication
2-resist way
Single resist way
quartz
quartz
quartz
quartz
quartz
quartz
quartz
PSM produces /2 lines !
X-ray lithography
•
•
•
•
•
•
Contact/proximity lithography
= 13 nm, resolution very good
highly penetrating radiation
not sensitive to particles, but
need thick metal to block x-rays
need 1X original because no x-ray mirrors
Optical vs. X-ray masks
-reduction optics
-flat structures
-mask is final size
-highly 3D structures
40 nm of metal stops UV light;
Need >1 µm thick metal to stop X-rays
Electron beam lithography (EBL)
• beam spot size 5 nm easily
• beam scattering in resist (in all solids)
• 10 nm can be made, but not easily
• use higher energy ( heating, charging)
• use thinner resist ( etch resistance down, defects up)
Spot size vs. linewidth
• linewidth typically 3*spot size to ensure
reproducibility and reduce roughness
missing
pixels
Raster vs. vector scanning
Pixel-by-pixel
raster scan;
exposure / no
exposure decision
at each pixel
“Intelligent
skipping of
empty
spaces”
EBL pros and cons
• flexible writing of structures individually
• writing speed very low indeed  small
areas only
• better resolution lower writing speed
• thinner resist better resolution, worse etch
and implant resistance, danger of pin hole
defects
Nano imprint (NIL)
•
•
•
•
•
1X master is pressed against polymer
force is used (pressure, temperature, UV)
release of the master
clearing the bottom residue
feature size limited by master fabrication only
NIL results
NIL problems
Problem 1:
• need 1X original pattern (cf. X-ray lithography)
Problem 2:
• need 3D original master
Problem 3:
• Lifetime of the master ? Does repeated contact
with the polymer damage or contaminate the
master ?
Problem 4:
• Who is the first one to try something really new
which may not work in production ?
MOS transistor
The goal of silicon processing
and thin film technology is to
control diffusion depths, film
thicknesses and interfaces at
atomic precision.
gate length Lg
field oxide
gate polysilicon
gate oxide
source channel drain
MOS gate oxide seem by TEM
“Metal” gate
made of highly
doped
polycrystalline
silicon
Amorphous oxide
Single crystalline
silicon substrate
Scaling of gate oxide
• Gate oxide thickness  gate length/50
Lg
Tox
• 1960’s
30 µm
600 nm
• 1970’s
5 µm
200 nm
• 1980’s
1 µm
20 nm
• 1990’s
0.35 µm 7 nm
• 2000’s
100 nm 2 nm
Oxide thickness limitations
•
•
•
•
•
•
Leakage current (tunneling)
Pinhole defects
Trapped charge
Interface traps
Interface structure (dangling bonds)
Crystallization and grain boundaries
(not in SiO2 !)
Leakage current explodes below 2 nm
High-k dielectrics (e.g. HfO2)
Because most high
dielectric constant
materials (high-k) are
oxides, some oxygen is
present during deposition,
and some SiO2 is formed
at the interface. The
question is: can you
control its formation and
thickness with Ångström
accuracy ?
EOT: Equivalent Oxide Thickness
EOT = (SiO2/ high) * thigh- + tSiO2
where tSiO2 is the interfacial silicon dioxide thickness, if any.
ZrO2 film of 6 nm physical thickness with  23 has EOT  1 nm
Control of oxide layer
Gate First High-k/Metal Gate Stacks With Zero SiOx Interface Achieving EOT=0.59
nm for 16 nm Application, VLSI Technology Symposium 2009
Half time
CNT transistors
Transistor characteristics
CNT network transistors (TKK)
Random network, many current
paths from source to drain.
High performance compared with
polymer transistors
CNTN transistors (TKK)
CNT circuitry by IBM (2006)
The five-stage
CMOS type
nanotube ring
oscillator using
palladium p-type
gates and
aluminum n-type
gates. The upper
right inset shows
the nanotube itself
with a diameter of
~2 nm.
CNT transistor time scales
•
•
•
•
•
•
•
•
1998 first CNT transistor (FET)
2001 logic gate
2002 Schottky switch
2002 top gate FET
2003 ballistic transport demonstrated
2004 AC characterization
2006 circuit demo, 72 MHz ring oscillator
2015? commercial devices (IBM guess)
Metallization
6 levels of metal, cross section
IC complexity increase over time
IC metal wire scaling (by n>1)
C’ =  (W/n)L/(T/n) = C
R’ =L/(H/n)(W/n) = n2R
RC time delay ’ is then given by
’ = R’C’ = n2RC
W
While transistor
performance
improves with
downscaling, scaled
metal wires are
worse !
T
H
L
metal
dielectric
Electromigration
Momentum transfer and displacement of lattice atoms by electrons
Depends on bond strength (which can be gauged by melting point)
Aluminum, low melting point, 650oC, low electromigration resistance
Copper 1083oC
Tungsten
3387oC
improved EM resistance
Hu, C.-K. et al: Electromigration of Al(Cu) two-level structures: effect of Cu kinetics of damage formation,
J.Appl.Phys. 74 (1993), p. 969
Grain size effects in metals
Mechanical properties
scale beneficially with
smaller grain size
Thermal properties
mostly unchanged
Resistivity increases
with decreasing grain
size
Erb et al: in The Nano-Micro Interface, Wiley-VCH 2004
Resistivity depends on patterns!
You cannot calculate
thickness from
resistance
R = ρL/Wt
because thin film
resistivity ρ is
linewidth and
thickness dependent
(use e.g. X-rays to get
an independent
thickness value)
G.B. Alers, J. Sukamto, S. Park, G. Harm and J. Reid, Novellus Systems, San Jose -- Semiconductor International, 5/1/2006
Grain size affected by:
-underlying film (chemistry and texture)
-deposition process (sputtering vs. plating; & plating A vs. plating B)
-material purity
-thermal treatments
-geometry of structures on wafer
G.B. Alers, J. Sukamto, S. Park, G. Harm and J. Reid, Novellus Systems, San Jose -- Semiconductor International, 5/1/2006
Current density barrier
Electromigration
limit of metals
ca. 1 MA/cm2
Vertical CNT connections
Seeded growth in contact holes
CNT’s show ohmic behaviour
at current density 4*108 A/cm2
Flash memory: close to limits
tunnel oxide 10 nm
interpoly oxide/nitride/oxide 50 nm
Gate linewidth 100 nm
Limit: thinner tunnel oxide traps charge and does become leaky
(10 000 to 100 000 rewrites)
PCM: phase change memory
Chalcogenide materials exhibit
• 100X resistivity difference between
crystalline and amorphous states
• factor of 2 difference in reflectivity
Memory programmed and read electrically
and/or optically
GST = Ge2Sb2Te5
Chalcogenide PCM
Programming pulse: 100 ns
GST hot spot heated > 620oC
molten GST
rapid cooling  amorphous GST
MΩ resistance
Reverse programming: also 100 ns
550oC, crystallization
kΩ resistance
Reliability & problems
Actual GST device
Moore’s law: Intel view
•
•
•
•
•
•
•
•
•
•
•
•
Year
Node
2005
2007
2009
2011
2013
65 nm
30 nm
45 nm
20 nm
32 nm
15 nm
22 nm
10 nm
15 nm not clear how
-carbon nanotubes ?
-III-V on Si ?
non-electronic devices ?
-spintronics
-optical devices
2019
Lg
EOT
0.8 nm
0.7 nm
0.6 nm
0.5 nm
Linewidth goes down
Chip size vs. wafer size
Chip size determined by:
Wafer size determined by:
•how many functions needed
•production economics
•how small lines used
Chip size goes up !
Lithography cost up exponentially
Cost of top-down nanofabrication
•
•
•
•
•
•
•
28 mask levels with 32 nm minimum lines
cost of finished chips 10$/cm2
cost per mask level = 30 cents/cm2
lithography equipment cost 35 M$
need to fabricate > 1 000 000 cm2/year
defects: 1 failed devices in 5 million
cf. DNA self-assembly: ppm error rates
correspond to very slow replication
Continued scaling till 2059
•
•
•
•
2.5 Å minimum linewidth
0.04 Å gate oxide thickness
2 mV operating voltage
64 exabit DRAMs (exa = 1018)
• This is not a scaled MOS transistor but
something completely different
• But Moore’s law is general; it is about economics
of device manufacturing; not about transistors