The new semiconductor technology angle in the article seem highly fishy to me. Apart from the fact that the statement felt like it may have said "In 10 years we will all be living in colonies on the moon", III-V materials have been losing market share to silicon for decades.

The article mentions that great electron mobility of the III-V materials, which is true, but forgets to mention that they had poor hole mobility. Now I am not a process expert, so maybe there are new techniques to address this. However, over the past 20 years or so this meant that you couldn't make very good CMOS logic and had to use NMOS only architectures. This and the poor scaling has kept the III-Vs away from large scale integrated logic chips.

The III-V devices were used in RF circuits, but they were replaced by Si-Ge and now many RF circuits use regular silicon processes. The III-Vs are still useful for optics.

The truth is that silicon has many problems that may prevent the industry from continuing to scale circuits to smaller geometries and the available workarounds are generally painful. But, the other options are worse.

Maybe in 10 years we will all be using cell phones that use carbon nanotubes... in our colonies on the moon.

In reality, sound is all analog. Those vibrating strings on that guitar... analog. The vocal cords in the singers throat... analog. The vibrating membrane on those drums... you get the idea. The challenge comes in when you want to store that information so that you can play it back later (by creating vibrations in someone's eardrum most likely). In studio recordings, the limits to the noise floor, distortion, and frequency response is set by the analog circuits unless it is a really crappy system.

Before digital computers were available, the only options were to create static variations in physical media, i.e wax cylinders, vinyl records, magnetic tape, etc. The variations were analogous to the sound waves in the air (hence calling them analog).

Digital sound samples the sound in time and quantizes them so that the can be represented numerically. The beauty here is that the physical medium no longer matters. Once you have the numbers, you could store them on spinning magnetic disks or marbles in shot glasses. The difference is cost and practicality.

There is a lot of information theory to cover here, but the relevant basics are that the quality of the stored digital data (talking about PCM here, compression is an other layer entirely) is how finely you quantize each sample (e.g. bit depth) and how often you take samples (e.g. sample rate). In a well designed digital audio system, these factors will not be the limiting factor of your performance. This was true in the early days of CD audio. The dynamic range of the ADCs and DACs was less than what 16-bit quantization could achieve. Also, the analog anti-aliasing filters of the day could not handle the 44.1kHz sample rate well as they had to have very steep rolloff.

Nowadays, the studio ADCs are capable of greater than 120dB dynamic range (the best datasheet I've seen is 127dB) and oversampling techniques like delta-sigma modulation have made the analog filters much simpler. 24-bit resolution is more than enough to handle this. Higher sample rates were initially to help with the analog filtering, but that does not matter today since almost all audio DACs actually run at several MHz internally and use digital interpolation filters to generate the oversampled data.

So, the theoretical 144dB dynamic range of 24-bit audio is not achievable today and will likely not be for the foreseeable future. Going to 32-bit only makes sense if you already have 32-bit hardware and you don't save any resources by going to 24-bit. There is a slim case to make if you are doing lots of processing, but the advantage over 24 bit is just a practical one in most cases.

This kind of turned into a rant, but there seemed to be a lot of analog vs digital comments and I wanted to try to provide some perspective.

From "Bart the Fink" in season 7...

Krusty (to Bart): Bah. What good is respect without the moolah to back it up. Everywhere I go I see teachers driving Ferraris, research scientists drinking champagne.

This is a research idea that MAY be useful, the demise of CMOS silicon has been highly exaggerated.

From the summary:

"an inverter, which was able to switch on and off 500,000 times per second" -> 500kHz is not so great

"however, began to break down after 2 billion cycles" or about 1 second at current processor speeds. That increases to 4000 seconds at 500kHz, or a little more than an hour.

Also, we can put billions of error free transistors on a chip for a few dollars. THAT is the real hurdle that nothing else has been able to clear yet. We will likely be with silicon for a while after it stops shrinking for this reason.
   

not sure about "eFUSE" specifically, but the fuses on chips are used to write permanent data to a chip after manufacturing, typically during testing. If this is being done in the field and is un-doable, then it is most likely some non-volatile memory that they are calling "eFUSE". I'd bet that the firmware makes it look like a fuse except under certain circumstances.

Actual fuses in chips are thin pieces of metal wire that connect to the power supply voltage. When you try to read the voltage, that connection gives you a logic "1". To blow the fuse, you use a higher power supply than you would normally use and run enough current through the thin wire so that it melts and breaks the connection. Then, when you try to read back that bit it will read a logic "0" for ever. This is very handy for encryption, calibration data, and manufacturing information such as lot# and chip location. Many years back, IIRC, Intel tried to put serial numbers on their CPUs that could be read back by the software. They backed off after some public outcry. This used the fuses described above.