I don't think we're disagreeing here... ;-)
One problem is that "PPA" is too simplistic -- if one process is better on all three counts (power *and* performance *and* area) than another then it's obviously "the best", but that's not the issue here -- it's not even the case with the different options within one process, for example comparing the different N3 FlexFin libraries (3-fin, hybrid 3/2 fin, 2 fin, hybrid 2/1 fin, 1-fin) they all have different power/area vs. frequency curves, and are different at different voltages.
Even your last point isn't strictly true -- yes if Intel can get a core to 3GHz with less power than TSMC then that's a winner, but what if the cost is bigger die area so you can't fit as many cores on a chip or the cost is higher (and maybe yield is lower)? It all depends what your priorities are... ;-)
(also GPUs are generally power-limited so don't push clock rates anywhere near CPU ones, it's more efficient to run slower and at a lower non-boosted voltage)
Backside power is another example of this, which Intel are promoting as "the bees' knees" because it helps keep clock rate up, especially at higher supply voltages like CPUs use on boost settings. But the gains at lower clock rates and voltages are a lot smaller (yes I've looked at this in detail), and there's a cost premium to BSP as well as some risk until it's all been pushed out into mass production successfully. From what I've seen it might well be appropriate for Intel and their product mix, but it's less obviously the best choice for TSMC even at N2 for many (most?) customers -- which is presumably why they're introducing it as a second (optional) N2 step a year or so later.
Intel are coming into this battle from a single-customer x86 CPU-centric viewpoint, TSMC are coming from a widely-spread customer base in many areas with CPUs -- x86 or others -- only being a relatively small part of their business. Since the current view is that the massive expansion in ICs over the next few years will be AI-driven not x86-driven, I think Intel are going to have a hard time muscling in on TSMCs market dominance -- but good luck to them for trying, a monopoly is rarely a good thing.
One problem is that "PPA" is too simplistic -- if one process is better on all three counts (power *and* performance *and* area) than another then it's obviously "the best", but that's not the issue here -- it's not even the case with the different options within one process, for example comparing the different N3 FlexFin libraries (3-fin, hybrid 3/2 fin, 2 fin, hybrid 2/1 fin, 1-fin) they all have different power/area vs. frequency curves, and are different at different voltages.
Even your last point isn't strictly true -- yes if Intel can get a core to 3GHz with less power than TSMC then that's a winner, but what if the cost is bigger die area so you can't fit as many cores on a chip or the cost is higher (and maybe yield is lower)? It all depends what your priorities are... ;-)
(also GPUs are generally power-limited so don't push clock rates anywhere near CPU ones, it's more efficient to run slower and at a lower non-boosted voltage)
Backside power is another example of this, which Intel are promoting as "the bees' knees" because it helps keep clock rate up, especially at higher supply voltages like CPUs use on boost settings. But the gains at lower clock rates and voltages are a lot smaller (yes I've looked at this in detail), and there's a cost premium to BSP as well as some risk until it's all been pushed out into mass production successfully. From what I've seen it might well be appropriate for Intel and their product mix, but it's less obviously the best choice for TSMC even at N2 for many (most?) customers -- which is presumably why they're introducing it as a second (optional) N2 step a year or so later.
Intel are coming into this battle from a single-customer x86 CPU-centric viewpoint, TSMC are coming from a widely-spread customer base in many areas with CPUs -- x86 or others -- only being a relatively small part of their business. Since the current view is that the massive expansion in ICs over the next few years will be AI-driven not x86-driven, I think Intel are going to have a hard time muscling in on TSMCs market dominance -- but good luck to them for trying, a monopoly is rarely a good thing.
