Transcript MOSFET Scaling - Northeastern University

MOSFET Scaling
ECE G201
Most Simple Model: Constant Field Scaling
E = VDD/L
after scaling becomes
E = (VDD/a)/(L/a)
…where a>1
next
Impurity Concentration Scaling
must also follow length scaling for depletion widths
Recall, that the source and drain are heavily doped and
therefore the junctions are one-sided (n+p for NMOS):
W = (2eVDD/qNA)1/2 …unscaled FET
W/a = (2eVDD / a2qNA)1/2
= [2e(VDD/a)/qaNA]1/2
Therefore, the doping levels must increase by a factor a if
the depletion widths are to scale down.
Historical Scaling
“Moore’s Law:” number of transistors/chip doubles every 18 mo.
1 generation: ~18 mo.
L decreases by
0.65/generation
(a = 1/0.65 = 1.5)
VDD decreases by
0.85/generation
Therefore, constant
field scaling (VDD/L) is
not strictly followed.
Generalized Scaling
Length: a = 1/0.65 = 1.5
Voltage: b = 1/0.85 = 1.2  Electric field: E increases
x1.25
Doping: ba = x1.8 (!)
note: not strictly followed
Junction Leakage Current
Tunneling current due to highly doped Drain-Body junctions
B
EC
EV
D
IJE
W
Recall: tunneling
T = Kexp(-2kW)
Gate Leakage Current
tox 0 means large tunneling current
A large oxide capacitance
is needed to control the channel
charge and subthreshold
current:
Vch = VGS(Cox’+CB’)/Cox’
…where Cox’ = eox/tox
since tox is limited by tunneling,
research is focused on alternate
gate dielectric materials with
larger permittivity (“high-K”).
High-K gate insulator reduces tunneling
current by allowing a thicker insulator
0.8 nm
High-K Issues
• Large number of interface traps, Qit
– impacts VT control and repeatability
• Process integration
– SiO2 is relatively easy (thermal oxidation of Si)
• Potential materials:
– HfO2, ZrO2, TiO2, BST….?
Subthreshold Current (revisited)
VDD scaling  VT scaling
Total Stand-by Power
Poff = VDD(Ig + IJE + Ioff)
Scaling Directions (I)
SOI (DST, depleted substrate transistor)
Very thin body region (Tsi = L/3)
makes the source and drain
spreading resistance (RS) large.
Improves subthreshold slope, S
and decreases Ioff
Raised S/D improves ID (next)
Also decreases CjE
…and IJE
Raised S/D
(i.e., decreased RD, RS)
Switching Speed: High current (ION)
but low voltage and low IOFF
Scaling Directions (II)
The “FinFET” moves from a single gate to
double and triple gate structures and also
multiple channels.
Advantages:
Control of the channel: must be fully depleted!
Improved RS, RD due to thicker Si body
Gate
prevents “top” gate
Fin (30nm)
BOX
MOSFET Future (One Part of)
• International Technology Roadmap for
Semiconductors, 2007 update.
• Look at size, manufacturing technique,
future devices.
THE INTERNATIONAL ROADMAP COMMITTEE POSITION ON TECHNOLOGY PACING
In previous editions of the ITRS, the term “technology node” (or “hpXX node”) was used in an attempt to provide a single,
simple indicator of overall industry progress in integrated circuit (IC) feature scaling. It was specifically defined as the
smallest half-pitch of contacted metal lines on any product. Historically, DRAM has been the product which, at a given time,
exhibited the tightest contacted metal pitch and, thus, it “set the pace” for the ITRS technology nodes.
However, we are now in an era in which there are multiple significant drivers of scaling and believe that it would be
misleading to continue with a single highlighted driver, including DRAM.
For example, along with half-pitch advancements, design factors have also rapidly advanced in Flash memory cell design,
enabling additional acceleration of functional density. Flash technology has also advanced the application of electrical
doubling of density of bits, enabling increased functional density independent of lithography half-pitch drivers. A second
example is given by the MPU/ASIC products, for which the speed performance driver continues to be the gate-length isolated
feature size, which requires the use of leading-edge lithography and also additional etch technology to create the final
physical dimension.
Significant confusion relative to the historical ITRS node definition continues to be an issue in many press releases and other
documents that have referred to “node acceleration” based on other, frequently undefined, criteria. Of course, we now expect
different IC parameters to scale at different rates, and it is certainly legitimate to recognize that many of these have productspecific implications.
In the 2007 ITRS, we will continue the practice of eliminating references to the term “technology node.” As mentioned
above, the IRC has recommended that the only standard header will be year of first production, and DRAM M1 half-pitch is
just one among several historical indicators of IC scaling.
With this latest change to standard ITRS table format policy, it is hoped that the ITRS will not contribute to industry
confusion related to the concept of “technology node.” Of course, “node” terminology will continue to be used by others.
Hopefully, they will define their usage within the context of the application to the technology of a specific product.
Questions?
Scaling (a, e)
Tunneling
Subtheshold Current
High-K gate dielectric
Spreading Resistance (Raised S/D)
FinFETs