I'm not against it at all, but II really would like to understand how Stallman's proposal would apply to, say, the following example (one I've used before):
Suppose we take something like the FM demodulator in a radio. When Edwin Armstrong invented it, back in the stone age of the 1930s, I think we can all agree that (a) it was an "actual physical device," and (b) that it met all the other criteria (novelty, non-obviousness, utility, etc.) needed for a patent. It was implemented with the technology available at the time -- stone knives, bear skins, vacuum tubes (valves), and a transformer.
Skipping over details like the invention of ratio detectors, phase-locked loops, etc., the next change in implementation of FM detectors came when the tubes were replaced with discrete transistors. This required some change in bias methods, impedance levels, etc., but no major redesign. It did save cost, size, and power, though.
The next change was integration. At first, the transformer was still needed for the demodulator, and so it was pinned out of the ICs, which were still analog. This saved cost, size, and power still further.
Later, schemes were found to integrate the function of the transformer, fully integrating the (still analog) demodulator. This saved cost and size still further.
Still later, improvements in integration processes enabled the function of the FM demodulator to be performed digitally, using an analog-to-digital converter (ADC) and a bunch of hard-wired logic gates, emulating the mathematical function performed by the analog demodulator. This saved cost, size, and power still further.
Demodulator designs were next ported into programmable hardware dedicated to signal-processing applications (digital signal processors); this required the ADC, plus the algorithm converted to the DSP's assembly language. This saved cost and size.
After that, demodulator designs were moved into hardware register-transfer languages, like Verilog, providing portability from chip to chip using standard-cell logic families. This saved cost.
Later, the Verilog designs were ported into field-programmable gate arrays (FPGAs), enabling one to program the hardware in the chip to become, when preceded by the ADC, an FM demodulator. This saved cost.
Finally, technology improved to the point that the FM demodulator could be made by an ADC followed by a microcomputer, programmed with software in a high-level language as part of a much larger system. This saved cost.
At what point in this development do we draw the line and say, "Below this, it's not patentable (or patent infringement)?" Where is "software"?