Nanomagnet Logic (NML) - Field-Coupled Nanocomputing: Paradigms, Progress, and Perspectives - page 28

without the need for explicit storage hardware (and the associated area and static/
dynamic power dissipation associated with it).
Architectural-level design techniques such as these should allow us to minimize
the ''cons'' of NML (nearest neighbor dataflow and higher latency devices when
compared to CMOS FETs) and exploit the ''pros'' (inherently pipelined logic with no
overhead). As a representative example [ 43 ], our projections suggest that hardware for
finding specific patterns in incoming data streams could be *60-75X more energy
efficient (at iso-performance) than CMOS hardware equivalents. Moreover, these
projections include clock energy overheads.

5 Summary and Discussion

In this chapter, we have presented an overview of our work over the years on nano-
magnet logic, which can be viewed as a magnetic implementation of the original QCA
field-coupled computing idea. We discussed NML basics, as well as approaches and
issues related to the realization of integrated systems. This review was Notre-
Dame-centric by design, to provide a somewhat historical perspective on the work of
our group.
Finally, we would like to mention a couple of other related research efforts that
also use nanomagnets to represent logic state, but that employ different mechanisms to
couple and switch these magnets. One such effort, the Spin-Wave Bus proposed by a
group at UCLA [ 44 ], is based on spin waves propagating in a layer underneath the
magnets. Since spin waves (plasmons) decay, this scheme requires amplifying ele-
ments to restore the signals. Another scheme, the All-Spin Logic proposed by a group
at Purdue [ 45 ], is based on nanomagnet coupling by spin diffusion in a magnetic layer
underneath the magnets. This scheme requires wires to be connected to the magnetic
dots in order to inject spin-polarized electrons that then diffuse and provide the
coupling mechanism. These approaches are interesting, and further research is war-
ranted. However, in our opinion, it is hard to see how coupling between dots using
either spin waves or spin diffusion can be more efficient or lower power than coupling
by direct magnetic fringing fields.
We end with a historical note. It was recognized in the very early days of digital
computer design that magnetic phenomena are attractive for several reasons [ 46 ]:
They possess an inherent high reliability; They require in most applications no power
other than the power to switch their state; They are potentially able to perform all
required operations, i.e., logic, storage and amplification. In fact, some of the very
early computers used ferrite cores not only for memory, but also for logic. Ferrite
cores were coupled by wires strung in specific ways between them so as to achieve
logic functionality. For example, the Elliott 803 computer used germanium transistors
and ferrite core logic elements. Of course, this kind of magnetic logic technology
based on stringing wires between bulky magnetic cores was not competitive against
emerging semiconductor technology. However, with the advent of modern fabrication
technology, which allows the fabrication of arrays of nanometer-size single-domain
magnets, the old quest for magnetic logic might become a reality.