Barriers to Scaling and the
Corresponding Breakthroughs |
| . |
. |
Since the CMOS LSI first appeared, its performance
has been constantly improved by
scaling down to smaller sizes and reducing
the power supply voltage. Along with cost
reductions due to reduced chip feature sizes,
these scaling rules have become the foundation
that supports both the evolution of logic
LSIs, such as CPUs, and the electronics industry.
Recently, three enormous barriers to
the progress of this scaling have arisen. Thus
it became necessary for us to overcome these
barriers with new technologies during development
of the 45 nm generation.
|
|
|
Barriers and Breakthroughs 1
Increased leakage current
(1) Increased off current
Although the scaling rules call for the
voltages that become the on/off thresholds
to fall as the supply voltage falls, in
the 45 nm generation, if the threshold
voltage were reduced according to the
rules, the transistor off state would become
inadequate and a current comparable
to the operating current would flow
in standby mode.
(2) Limits to thinning of the gate insulator
The success of today’s CMOS technology
is due to the fortunate fact that silicon
oxide can form a superlative insulating
film. In the 50 years since the birth
of the CMOS LSI, performance has been
improved by making this gate insulator
thinner and thinner. However, the gate
insulator film, which reached the 1 nm
level in the 65 nm generation, already
has a thickness of only about 5 atomic
layers. If this film is made even thinner,
quantum effects result in electric charge
passing through the insulator causing the
standby current to rise rapidly.
To achieve the targeted performance improvements
while avoiding these issues,
we developed a strained silicon technology
that achieves performance improvements
without depending on threshold
voltage reduction and gate insulator thinning
that follows earlier scaling rules. |
Barriers and Breakthroughs 2
Increased wiring capacitance
due to smaller design rules
The scale down in design rules increases the
wiring capacitance between adjacent lines.
This cancels the effects of reduced load capacitance
and gets in the way of improving
the characteristics. To reduce the wiring capacitance,
low-k materials such as fluorine or
carbon doped oxide have been used as inter-/intra layer dielectrics. However, it is difficult
to continue to reduce the dielectric constant
in a stable manner in an insulating material.
Therefore, for the 45 nm generation, we introduced
pores in the low-k film to reduce the
wiring capacitance. |
Barriers and Breakthroughs 3
Limits to scaling
from optical lithography
While reduced feature sizes in CMOS LSIs
has been achieved by developments in lithography
technology based on using shorter light
wavelengths, development of lithography
technologies using wavelengths shorter than
that of the ArF laser (193 nm) introduced at
the 90 nm generation have been delayed.
In this work, we inserted water between the
pattern exposure lens and the silicon wafer and
used the index of refraction of water to improve
the resolution for the circuit pattern. As
a result of this, we achieved an SRAM cell
size with an area reduced by 50% from that
of the 65 nm generation. |
|
The introduction of these technologies has
made it possible for the high-speed technology
version of the 45 nm generation to
achieve, at 0.8 V, speed equivalent to that of
the 65 nm generation at 1.0 V. This fact alone
means that the power becomes 64% (since
0.8 × 0.8 = 0.64) of the earlier generation, a
reduction of 36%. Added to this, the finer
feature sizes mean that the device sizes are
reduced and this reduces load capacitances.
These power consumption reduction effects
make it possible to adjust the trade off between
power consumption reduction and
computational speed improvements freely.
|