I was fascinated by Lou's talk at the Symposium. His idea of a
"parallel" power system is an excellent one if a few potential problems
can be addressed.
The first and most obvious one is efficiency. Switching converters tend
to be much less efficient at low voltages and high currents, though
power MOSFETs have gotten very good of late. Every Pentium 4 motherboard
has had a DC-DC converter that takes 12V and produces ~1V at up to
100A to power the CPU. The fact that there isn't a massive heat sink and
fan for the supply as well as the CPU says that these converters can't
be too terribly inefficient. I'll be interested in some actual figures
for the "virtual battery" converters.
The second issue is robustness. Just as serial architectures are
vulnerable to single-point open faults, parallel architectures are
potentially vulnerable to single-point shorting faults, e.g., of the
power transistors and diodes in the converter. I'm hoping a judicious
application of fuses can take care of this.
Super caps look very promising thanks to their claimed long cycle lives,
but I wonder about their radiation robustness. Imagine what a highly
energetic, totally ionized particle might do as it rips through the
Aerogel holding large numbers of electrons a few microns apart from
large numbers of holes. Have any radiation tests been performed on these
things?
Also, I would not take their cycle life claims as gospel without some
tests to back them up. Electrolytic caps have become some of the most
unreliable components in electronics today. They fail on a regular basis
in computer power supplies and motherboards, often spectacularly.
While super caps aren't built quite the same way, both kinds of caps
achieve their high capacities with lots of extremely thin dielectric
material. Cycling the cap must place some pretty serious electrostatic
compression stresses across this material. What happens, e.g., after
many cycles at very low temperatures?
At a system level, I see a potential "gotcha" with multiple paralleled
virtual battery units. You can never get multiple voltage references to
exactly agree, so if the control loop gains are too high, then one
virtual battery might think the bus voltage is a little too low when
another might think it's too high. Then you'll have large amounts of
current flowing from one virtual battery into another, obviously an
undesirable situation.
The simplest fix I can think of is to program the control logic in each
virtual battery to follow a specified current vs voltage characteristic.
The slope of this curve must be gentle enough so that small differences
in reference voltages across the units will not produce significantly
different I/V curves. Bus regulation won't be quite as tight, but
hopefully the loads won't mind.
Overall I think Lou's proposed architecture is an excellent idea, and
with attention to a few details it should be a lot more reliable than
the conventional power systems we've had in the past.
--Phil
