Information Technology Reference
In-Depth Information
Fig. 22.
Output waveforms example.
As well as the circuit timing behavior, additional information can be obtained,
like circuit area, number of magnets, wasted area. Moreover a power estimation
is also possible, up to now only considering the magnetic field clock.
8 Performance
One of our main targets during the development of ToPoliNano was to design a
tool capable to handle high complexity circuits with reasonable execution times.
While the software is still in development we were able to make some preliminary
evaluations on performance. The tests are based on currently available machines,
with
I-3
,
I-5
and
I-7
Intel processors, running both Linux and Mac-OS operat-
ing systems. As a benchmark we have used a simple Ripple Carry Adder, made
by N full adders. Just for test purpose we have instantiated up to 10000 full
adders. The placement of all magnets (around 2000000) took only 30 s. The in-
memory occupation of such a circuit was just about 1.5 Gb. The simulation of
one full adder for a time period of 80
s with a simulation step of 1 ps, required
just 0.3 s. To compare the performance of ToPoliNano with existing tools we
have performed some simulations with two widely used micromagnetic simula-
tors, NMAG [
30
] and OOMMF [
29
] on the same machine used for the testing
of ToPoliNano. The simulated structure was a simple NML wire in three differ-
ent cases, changing the length from 4 magnets, to 8 magnets and finally to 12
magnets. Results in terms of simulation time and memory usage are reported in
Table
1
. The time indicated is the machine time required to advance the state of
the circuit of 1 ns. For example, in case of the 4 magnets wire simulated through
NMAG, 43 s of machine-time are required to advance the state of the circuit
of 1 ns. The memory usage and the simulation time increases with the circuit
complexity. Starting from these values it is possible to get a rough estimation
ยต
Search WWH ::
Custom Search