Its also important to keep in mind that the smaller the chip, the better the yields. Not only will yield % be better, you will get more chips per wafer duo to size of actual chip as well.
No, this isn't necessarily true.
Chip production doesn't follow the linear path which you described i.e. based on size. When you're breaking up a wafer sheet for chips, the resultant chips aren't born equal because they will have variant quality. For example, one chip may have all CUs activated but in order to hit the required clock, draws more power than another chip which will hit the same clocks at lesser power. Then there could be another chip where few CUs are not aligned. There are too many factors at play here. There is a reason why GPU manufacturers prescribe a set of criterias so they can tier their cards.
Why do you think AMD has two different products in form of 5700 and 5700XT when they are made from same wafer? They created two different SKUs because of variance in silicon quality and prescribed few requirements in accordance to which the chips are divided. Chips which are of higher quality are allocated in a 5700XT bin while ones which are of lower quality, go into 5700 bin. Before a chip can be used in 5700XT, it has to tick the requirements of a) 40/40 active CUs b) Reach 1905MHz and c) Draw 225W power at max. The lesser capable chips can be used in 5700 if it a) Has 36/40 active CUs b) Reaches 1725MHz clock speed and c) Draws 180W power at max.
If a chip has 40/40 active CUs but draws 260W of power to reach 1905MHz, it goes in the 5700 bin. Similarly, if a chip has 36/40 active CUs but can reach 1905MHz while drawing 200W of power, it'll still go into 5700 bin. Why? Because the baseline requirements which AMD prescribed for these SKUs are final. The whole idea was conjured up by them in first instance because going narrow and fast significantly reduces the yields. Having a 2nd SKU saves them cost and make this production of narrow and fast chips somewhat feasible.
Now console manufacturers on the other hands, specifically Sony do not have the same luxury of different SKUs. Before they decide on a chip which goes in the final product, they have to decide upon a baseline which is low enough i.e. have a moderate clock speed, hit the required clock speed without any discrepancy between power drawn and thermal targets so majority of chips which are manufactured can be used. The 36CU chip which you are on about is the same chip as used in 5700/5700XT, their requirements would be a) Have 36/40 active CUs, b) Reach 2000MHz clock speed c) Draw ~150W power. Any scenario in testing, when a chip misses any of the requirements has to be binned and unlike AMD they do not have different SKUs where binned chips could be used.
This is the main reason why people who are knowledgeable about these stuff parrot that going narrow and fast for console manufacturers doesn't make any sense because it is much more expensive than going wide and slow. The threshold for chips making the cut is too high, you are basically throwing ton of chips away, and in turn spending more money to make your APU.
This becomes an issue if you do not have 2nd SKU and chip much smaller that will make bulk of your sales (say 60%). For PS5 you want to make chip big enough to fit tech required for good generation jump, but also one where you will not suffer stock shortages and high prices.
60% of sales? I assume you're referring to Xbox 2nd SKU here in Lockhart. First and foremost, the APU size of Series X is ~400mm2 which is ginormous. It makes zero sense to use binned chips of Series X in a Lockhart when it is targeting 1/3rd of total compute power. The cooling, design, form factor would be all be rendered futile if they have to house a chip that big. The APU in Lockhart will be significantly smaller and significantly cheaper to produce.
For example :
~320mm² chip (36CU) should yield about 35-40% more chips per wafer compared to 390mm² (56CU).
Take into account what I wrote and you'll realize your simplistic view of smaller chip = more wafers = cheaper = better yields isn't true. A chip with 36CU@2.0 GHz is smaller than a 56CU@1.67GHz but isn't cheaper to produce. Because even though console manufacturers are paying for giant sheets of wafer, regardless of size of the chip, if the threshold/requirement is too high (such as 2.0GHz while drawing ~150W power), they'll end up wasting more chips and therefore more money.
It's pretty abundant that you are not as knowledgeable as you pretend to be.