The main challenge in the pinless technology is how to supply power to the computational elements. We have examined several alternative approaches, including photodiodes, acoustic transducers, microwaves, magnetic coupling, and chemical methods. Each of these raises difficulties, but the difficulties appear to be technical rather than fundamental.
To obtain sufficient voltage to drive a PPP with photodiodes we need
to place several diodes in series. In bulk CMOS processes, this
creates parasitic transistors that short the power supply. Another
problem with photodiodes is that, even in full sunlight, the maximum
power obtainable is 1 kW/m
or 1 mW/mm
. Computationally
demanding applications, however, require more than a few milliwatts.
It may be possible to use light as a power source for a PPP
fabricated with silicon-on-insulator (SOI) processing, or with
multichip modules.
With acoustic transducers such as piezoelectric crystals, there are significant impedance mismatches with both the mechanical environment (the material is too stiff) and with the electrical requirements (the voltage is too high). Microwaves might work, but that would require an environment with high levels of ambient microwave radiation. Malamy, Glasser, and Selvidge [9] experimented with magnetic induction as a method for both wireless power and communcation for chips. They successfully demonstrated how to supply power using strong magnets and pole pieces that were closely coupled to individual chips. This does not seem feasible for a large system, however.
Chemical methods provide sufficient energy density. If the system needs to work for only a short time, it would possible to build batteries into the particles. It is also possible, but beyond the current state of the art, to build fuel-cell like systems that combine hydrogen and oxygen to form water. Such a PPP would have an extended lifetime.
In summary, we expect that the challenge of power distribution will be met by one or more emerging technologies. However, since our research focuses on the organization of the computation, we think it is currently more productive for us to avoid this issue, and use explicit power distribution.