CMOS Technology

Only 15,432,758 more

mosfets to do...

metal
ndiff
poly pdiff

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 1

 Basic Fabrication Steps
Growing silicon dioxide to serve as an insulator
between layers deposited on the surface of the
silicon wafer.

Doping the silicon substrate with acceptor and

donor atoms to create p- and n-type diffusions
that form isolating PN junctions and one plate of
the MOS capacitor.

Depositing material on the wafer to create

masks, wires and the other plate of the MOS
capacitor.

Etching deposited materials to create the

appropriate geometric patterns.

Figures
Figuresare
arefrom
fromW. W.Maly,
Maly,Atlas
AtlasofofICICTechnologies:
Technologies:AnAnIntroduction
IntroductiontotoVLSI
VLSIProcesses.
Processes.
(ignore dimensions in figures – they are quite out-of-date!)
(ignore dimensions in figures – they are quite out-of-date!)
6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 2
 Growing Silicon Dioxide
fast

O2 or H2O
+ Thermal oxidation creates
900o to 1100o high quality film used as
Oxygen diffuses thru SiO2 mask during diffusion,
then oxidizes Si surface
insulator and gate
dielectric. Local oxidation is
accomplished using a Si3N4
mask.

Surface is consumed
Bird’s beak reduces
size of unoxidized area

Selective growth by using Si3N4 to

prevent O2 from reaching Si surface

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 3

 Doping by Diffusion
Two-step:
predeposition, drive-in

Constant Source

Two-step process
results in more uniform
concentrations

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 4

 Doping by Implantation

Ion implantation involves much lower process temperatures, much decreased

lateral spreading and better control over dopant profile. But surface of wafer
is damaged and must be repaired by subsequent thermal annealing step which
will redistribute the dopants. Redistribution is minimized with special heating
techniques that minimizes exposure of implanted regions.

Diffusion still used when dopant profile isn’t critical.

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 5

 Is there a

Deposition lawsuit?

Chemical vapor deposition to deposit

Physical vapor deposition to SiO2, Si3N4, single-crystal (epitaxial)
deposit metals (Al, Cu). and polycrystalline (poly) Si.

Nonconformal coverage of steps

leads to non-uniform thickness.
In metals this can lead to higher
current densities in thinner spots
which causes current-induced
metal migration. Modern
approach: planarization.

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 6

 Etching
Photoresist is Develop to
spun onto remove
wafer then exposed
exposed resist.
with UV light, X-
rays or electron
beam (no
mask). Performance note: minimum feature size often
determined by photoresist and etching process.

Wet
etching

isotropic
Remove photoresist mask
Dry
etching

anisotropic

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 7

 Sources of manufacturing problems
Line registration errors Other fab difficulties
resist exposure and development Ö contacts and vias only on “flat”
over/under etching, lateral diffusion surfaces
uneven topography Ö no devices near boundaries of well
Ö no poly contacts over diffusion
Ö systematic errors: corrected by Ö “gate” metal must connect to
bloating/ shrinking mask diffusion
Ö random errors: increase mininum Ö minimum metal coverage
widths and spacings requirements
Electrical properties
Mask misalignment Ö current density limitations
Ö random errors: increase Ö latch-up prevention
extensions and surrounds Process instabilities
mobility variations (why?)
thin-oxide thickness variations
sheet resistances
Ö use of “process corners” in
analysis

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 8

 Design Rules
Exclusion rule
Surround rule Extension
rules

Width
rules

Spacing rules

We can specify the design rules using some convenient units, e.g., microns
but what happens if we want to manufacture the chip using different
manufacturers? One suggestion: use an abstract unit, the lambda, and
scale the design to the appropriate actual dimensions when the chip is to be
manufactured.
6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 9
 Lambda-based design rules
One lambda = one half of the “minimum” mask dimension, typically the length
of a transistor channel. Usually all edges must be “on grid”, e.g., in the
MOSIS scalable rules, all edges must be on a lambda grid.
2x2

1 3
2
2
2 2

3 3 2
1
1
3
diffusion (active)
2x2 3
poly
metal1
contact

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 10

 Let’s build a CMOS inverter
There are lots of different recipes to choose from. Like most
things in life, you get what you pay for: the ability to have
good bipolar devices, radiation hardness, reduced latch-up
and substrate noise, … are all extra cost options. We’ll
consider a “Chevy” process: bulk CMOS with a p-type
substrate:
500µ slice of a silicon ingot that has
been doped with an acceptor (typically
Use <100> surface boron) to increase the concentration of
to minimize surface holes to 1014/cm3 - 1018/cm3.
charge
Good for n-channel
fets, but p-channel
p-type
fets will need a n-
type “well” (or tub)
Back is metalized to provide FETs will be embedded in the to live in!
a good ground connection. substrate, wiring goes on top

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 11

 N-Well implant and drive-in
The black areas of the mask
show where the photoresist
will be etched away, exposing
the underlying material to
implants or further etching.

Donor atoms (e.g, P) are

implanted through a window
in the oxide mask and then
driven-in by the next high-
temp operation. The
concentration is around
1016cm-3 and the doping
profile is relatively flat.

Performance note: reversed-

biased PN junctions have
lots of capacitance!

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 12

 PN Junctions as Insulation
n Once the two materials are in contact, the
mobile carriers move:

diffusion of holes from P to N and

electrons from N to P ⇒ depletion of
majority carriers in boundary region.

drift of majority carriers due to E field

p formed by fixed ions ⇒ acts in opposite
direction of diffusion

n
At equilibrium, the sum of the drift currents =
sum of the diffusion currents. A depletion
depletion region is formed with a voltage across it due to
region induced field. At room temp, with doping
concentrations of 1015/cm3, this voltage is 0.6v.
The net result is a diode: Ipn
n
p If VPN ≤ 0, the two
regions are electrically
isolated p Vpn
6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 13
 Channel-stop implant

Si3N4 is used to mask-off

active regions (where the
FETs will be built). Then a
channel-stop boron implant
is performed which increases
acceptor concentration
outside of the active regions
and N-wells.

Performance note: this

implant leads to significant
sidewall capacitance for
mosfet source/drain regions.

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 14

 Grow field oxide

A “thick” layer of SiO2 is

formed by oxidizing the
unmasked portions of the
wafer with wet oxygen. This
field oxide, along with the
channel-stop implant, will
isolate the N- and P-fets. The
high temperatures used to
grow the oxide also
redistribute the dopants in
the well, but this is usually the
last high-temp operation in
the fabrication process.

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 15

 Grow gate oxide

Now grow a “thin” (10’s of

Angstroms) layer of SiO2, called
gate oxide, on the surface --
effect on field oxide is negligible.
The gate oxide needs to be of
high quality: uniform thickness,
no defects! The thinner the
oxide, the more IDS the FET will
have (we’ll see why soon) but the
harder it is to make it defect-
free.

Performance note: thin oxides →

punch-through issues → lower
operating voltages → lower
power dissipation but less IDS.

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 16

 Deposit polysilicon
On top of the thin oxide a thick layer
of polycrystalline silicon, called
polysilicon or poly for short, is
deposited by CVD. The poly layer is
patterned and plasma etched (thin ox
not covered by poly is etched away
too!) exposing the surface where the
source and drain junctions will be
formed. Poly has a high sheet
resistance of 20Ω/sq which can be
reduced by adding a layer of a silicided
refractory metal such titanium (TiSi2),
tantalum (TaSi2) or molybdenum
(MoSi2) => 1, 3 or 5 Ω/sq.

Gate oxide Performance note: modern wire aspect

ratio is more like H = 2W, so fringing
fields important when calculating
capacitance.

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 17

 N+ source/drain implant

Implant happens everywhere

but source/drain region of
pfets

Donor implant is used to

create N-fet source/drain
diffusions and an ohmic N-well
contact. Usually As is
preferred to obtain shallow
junctions and minimal lateral
diffusion. High doses are
N-fet source/drains needed to make low resistance
are “self-aligned” with N+ well contact (25 Ω/sq) diffusion wires.
poly

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 18

 P+ source/drain implant

The negative (or

complement) of the previous
mask is used to define the
p+ source/drain regions of P-
fets. Boron is used as the
dopant in this step. Then a
short thermal annealing step
is performed to repair
surface damage caused by
the implantation.

P-fet source/drains
are “self-aligned” with
poly

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 19

 Grow intermediate oxide

Next an intermediate oxide layer

is deposited over the entire
wafer using CVD (no more
thermal steps please!). In
modern processes, this layer is
planarized using a polishing
process so that the subsequent
metal layers will be flat and
hence have a uniform thickness.

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 20

 Cut poly/diff and substrate contacts

Holes (aka vias) are etched

in the oxide where contacts
to poly or diff are wanted
(sorry, no poly contacts over
gate region). The holes are
filled with tungsten plugs to
ensure good electrical
connections. Contacts vary
in resistance from .25Ω to
10Ω. Note that vias are
usually constrained to be a
particular size, so an array
of vias is used when making
a large “contact”.

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 21

 Deposit and etch metal layer

Aluminum is deposited using

PVD, patterned, then etched to
form low-resistance (.07 Ω/sq)
interconnect. With
planarization, multiple levels of
metal interconnect are
possible -- 3 to 5 layers are
common in today’s processes.
Each additional level of
interconnect requires two
masks: one for vias and one for
forming the wires.

Performance note: more layers

reduces routing congestion
leading to a more compact
design. Sometimes layers are
devoted completely to power
distribution.
N-channel MOSFET P-channel MOSFET

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 22

 Multiple levels of interconnect
IBM photomicrograph (Si has been removed!)

Metal 2

M1/M2 via

Metal 1

Polysilicon

Diffusion

Mosfet (under polysilicon gate)

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 23

 FET = field-effect transistor

The four terminals of a fet (gate, source, drain and bulk) connect to
conducting surfaces that generate a complicated set of electric fields in the
channel region which depend on the relative voltages of each terminal.
gate
inversion
happens here
Eh
source drain
Ev

bulk

INVERSION: CONDUCTION:
A sufficiently strong vertical field will If a channel exists, a horizontal field will
attract enough electrons to the surface cause a drift current from the drain to
to create a conducting n-type channel the source.
between the source and drain.

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 24

 “Off” operating region
VS VGS < VTH VD

Figure
Figureisisfrom
fromJ.J.
Rabaey, Digital
Rabaey, Digital
Integrated
IntegratedCircuits
Circuits

The gate voltage with respect to the source (VGS)

required to form the channel is called the threshold
voltage. The process engineers implant ions in the
channel to achieve the chosen threshold voltage.

Performance note: in small devices, there is

significant IDS even when VGS < VTH (subthreshold
conduction).
6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 25
 “Linear” operating region
VS VGS > VTH 0 < VDS < VDsat

IDS

Why is this bigger here than

on other side?

Larger VGS creates deeper channel Larger VDS increases drift current but
which increases IDS also reduces vertical field component
Knee due to velocity saturation and which in turn makes channel less deep.
At some point, electrons are traveling
IDS mobility degradation in small devices
as fast as possible through the
channel (“velocity saturation”) and the
Increasing current stops growing linearly.
VGS
IDS ∝ (µnεox/tox)(W/L)
VDS
6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 26
 Saturated operating region
VS VGS > VTH VDsat < VDS

IDS

L’ = L - δL
VDsat ≡ VGS-VTH
δL

This looks just like a fet with a channel When VDS = VGS-VTH the vertical field
length of L’ < L. Shorter L’ implies component is reduced and the channel is
greater IDS. As VDS increases, δL gets pinched-off. Electrons just keep traveling
larger. across depletion region…
IDS
Increasing VGS

9/18/02
VDS
6.371 – Fall 2002 L05 – CMOS Technology 27
 NFET IDS curves: then and now

IDS vs. VGS

Figures
Figuresare
arefrom
fromJ.J.
Rabaey, Digital
Rabaey, Digital
Integrated
IntegratedCircuits
Circuits Long channel Short channel

IDS vs. VDS

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 28

 G
NFET Summary S D
n n
p
D D +

G G VDS ≥ 0
+

S - S -
Operating regions: VGS
0.5V
cut-off: IDS
VGS < VTH S D
linear saturation
linear:
VGS ≥ VTH S “ “ D VGS
VDS < VDsat
VGS - VTH

saturation:
VGS ≥ VTH VDS
S D
VDS ≥ VDsat

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 29

 G
PFET Summary S D
p p
n
D D -

G G VDS ≤ 0
+

S - S +
Operating regions: VGS
–0.5V
cut-off:
VGS > VTH S D -VDS

linear:
VGS ≤ VTH S “ “ D
-VGS
VDS > VDsat
VGS - VTH
saturation linear
saturation:
VGS ≤ VTH
S D
VDS ≤ VDsat -IDS

6.371 – Fall 2002 9/18/02 L05 – CMOS Technology 30