Actually, statistical control is a reasonably strong approach. I take ethernet as an example.
10/100 Mbit ethernet was once dominated by National Semiconductor, heavily relying on their analog background to control tightly the parameters involved. They were overtaken by a disruptive technology using DSP and statistical "control". It turned out that it made the analog simpler, and the digital side of the issue meant that die shrinking took place much faster. By the time National spent $120M buying Comcore to play catchup, their die size was 60% larger than Broadcom. The next generation was gigabit ethernet, where the vast majority of the game was with DSP and Marvell entered the picture. As each generation of ethernet came out, it was more digital, more millions of transistors doing DSP where analog used to be, and eventually it was so cheap that we now buy those chips for $2 at 1Gig/s when they were originally $45 at 0.1Gig/s By using a statistical approach, Rossi puts himself on the digital scaling roadmap rather than the analog scaling roadmap. It has tremendous merits. What is the danger? If an air conditioner goes on during August when it ain't hot, what's the harm? If Rossi's device goes kaflooiee in the first generation, it will just stop working. By the time the 3rd generation rolls out, it will no longer go kaflooiee, and it will be under far tighter control than if he had taken the "analog" route. On Tue, Aug 19, 2014 at 8:45 PM, Axil Axil <janap...@gmail.com> wrote: > > Statistical control is like saying that most of the time it is hot in > august so turn on the air conditioners in august. Most of the time you are > correct, but sometimes a bad thing happens. > >
