1. Clock rate – the logic speed of advanced
technologies will be significantly faster than
current generation,
2. Device density – linear feature size
reductions of close to two orders of
magnitude resulting in three to four orders
of magnitude gates per unit area,
3. Parallelism – systems implemented in
nanoscale technology will comprise orders
of magnitude greater number of local
execution sites and therefore will need to
exploit much higher degrees of algorithm
concurrency.
4. Latency – time of signal propagation
(measured in clock cycle time) through a
sequence of gates or equivalent physical
constructs across an entire component is
dramatically greater than that of
conventional system devices,
5. Reliability – much smaller devices may
break more easily; moreover, the probability
of single-point failures grows with the scale
of the system
The consequence of these interrelated factors
alone and in combination is that:
x
Advanced components approaching the
nanoscale will exhibit strong bias towards
locality of action far more extreme than
today, and
x
Architecture defined to be implemented
using nano-scale technologies will have to
respond to this.
At one time, processors could employ a single
clock such that any logic within the processor was
accessible in a single clock cycle for a round trip
signal. Even today, this is not really feasible and a
combination of slower global clocks with faster local
clocks is used while super-pipelined logic structures
make the on-chip latency problem manageable. But
with nanoscale devices at super high clock speeds
and many more gates of propagation possible per unit
distance, only a tiny fraction of the total on-chip
1. Clock rate – the logic speed of advanced
technologies will be significantly faster than
current generation,
2. Device density – linear feature size
reductions of close to two orders of
magnitude resulting in three to four orders
of magnitude gates per unit area,
3. Parallelism – systems implemented in
nanoscale technology will comprise orders
of magnitude greater number of local
execution sites and therefore will need to
exploit much higher degrees of algorithm
concurrency.
4. Latency – time of signal propagation
(measured in clock cycle time) through a
sequence of gates or equivalent physical
constructs across an entire component is
dramatically greater than that of
conventional system devices,
5. Reliability – much smaller devices may
break more easily; moreover, the probability
of single-point failures grows with the scale
of the system
The consequence of these interrelated factors
alone and in combination is that:
x
Advanced components approaching the
nanoscale will exhibit strong bias towards
locality of action far more extreme than
today, and
x
Architecture defined to be implemented
using nano-scale technologies will have to
respond to this.
At one time, processors could employ a single
clock such that any logic within the processor was
accessible in a single clock cycle for a round trip
signal. Even today, this is not really feasible and a
combination of slower global clocks with faster local
clocks is used while super-pipelined logic structures
make the on-chip latency problem manageable. But
with nanoscale devices at super high clock speeds
and many more gates of propagation possible per unit
distance, only a tiny fraction of the total on-chip
การแปล กรุณารอสักครู่..
