Information Technology Reference
In-Depth Information
(a)
(b)
Fig. 8. (a) Definition of logic density in QCA, the average gate area expressed as a multiple of
molecular cell footprints. (b) Maximum operating frequency of QCA circuits as a function of
cell width, for A = 300w
2
, 200w
2
, and 100w
2
dissipation area per majority gate. High-density
computer arithmetic designs found in the literature typically use about 200 cell footprints of
circuit area per each logic gate (the middle curve) [
8
].
in standard CMOS layout rules, enabling us to express scalable QCA layout
lengths/distances as a multiple of w. Based on the selection of arithmetic units found in
literature, we calculated the average area A available for each standard gate, which in
QCA is a 3-input Majority Voter Gate (MG) residing in an imaginary layout rectangle
of area A, as illustrated in Fig.
8
(a). We found that the average area per gate inside the
densest core logic (e.g. a full adder) was around A = 100w
2
, while the wiring overhead
of an optimized multi-bit arithmetic unit increased the area per gate typically to
A = 200w
2
. In random logic, the area per gate was typically to A = 300w
2
[
8
].
Figure
8
(b) shows the resulting maximum operating frequency vs. QCA cell width
w, when the heat generation is limited to 100 W/cm
2
. In this comparison, the nominal
area per logic gate was considered to be A = n
fp
w
2
, where the gate span n
fp
was 300,
200, or 100 cell footprints, corresponding to the studied relative gate densities.
Table 3. Estimates for the worst-case bit erasures in QCA arithmetic units, vs. operand word
length n. The logical operations must perform only a linear number of erasures, but the adders
discard 2-6 times that much and the multipliers a square-law amount of information [
8
].
Binary addition with unsigned n-bit operands (typically linear)
Theoretical addition operation
n
Ripple carry adder, lower bound
2n
Ripple carry adder, upper bound
6n
Binary multiplication with unsigned n-bit operands (typically square-law)
Theoretical complete multiplication
n + 1
Theoretical non-trivial multiplication
123 9 log(n + 241) - 673
8n
2
Array multiplier
16n
2
Serial-parallel multiplier
16n
2
Serial-parallel optimized multiplier
- 12n
26n
2
+ 86n - 2
Radix-4 recoded multiplier
Search WWH ::
Custom Search