CNTFET quaternary multipliers are less efficient than the corresponding binary ones

Abstract

We compare N*N quaternary digit and 2N*2N bit CNTFET multipliers in terms of Worst case delay, Chip area, Power and Power Delay Product (PDP) for N=1, N=2 and N=4. Both multipliers use Wallace reduction trees. HSpice simulations with 32-nm CNTFET parameters shows that the binary implementations are always more efficient: the 1*1 quit multiplier is far more complex than a 1*1 bit multiplier (AND gate) and generate both product and a carry terms. Even with half number of terms, the quaternary reduction tree has the same number of terms than the binary one, and uses quaternary adders that are also more complicated than the binary ones. The quaternary multipliers have larger worst case delays, more power dissipation and far more chip areas than the binary ones computing the same amount of information.