Not that Apple is doing it the same as TI, but before Apple there was TI and it had 80% mobile phone market share. They fabbed the same chip at no less than four companies - 3 foundries plus their own fabs.
The way this was made possible, without making it painfully clear to the consumer that the chip inside one Nokia phone possessed superior electrical parametrics to those of the chip in another phone was by way of locking in electrical specs and then using binning as the variable in the equation.
The way this would work in practice, and I am just crudely speaking to the complexities in this post for the sake of brevity, but let's say you have Foundry A that can technically meet the electrical specs (it is within engineering margin for spice models, leakage, thermals, etc) but only just, meanwhile you have Foundry B that can easily meet the electrical specs because technically your specs (TI's in this case) are designed to allow even the weakest link (foundry) be capable of delivering chips to the packaging house in the Philippines.
So what happens in this simplified situation is that Foundry A struggles and strives to hit parameteric yields for binning your chips. Let's say, for sake of argument, they can deliver 50% yields (parametric and functional) at your required TDP, Tcase, IDDQ, TDDB, etc. Meanwhile Foundry B, with its electrically higher performing process, can deliver 85% yields at your required specs.
What is also true, but no one takes advantage of, is that Foundry B could also deliver superior binned chips with lower TDP, Vdd, IDDQ, etc., albeit at reduced yield (say 50% in this case, iso-yield normalize across the two foundries) if that was of interest to the customer (TI in this real-world example case).
But the customer in this case isn't interested in better-binned but lower yielding chips from Foundry B. They just want volumes and volumes of chips that meet the spec. And the easiest way to hit the spec is to set an operating voltage (and essentially an operating power consumption) for each foundry such that the chips coming to packaging are all operating within the same electrical and thermal profile, even if needlessly so.
In other words, to be sure for TI there were a significant number of chips from Foundry B which were essentially operating in a needlessly overvolted condition with respect to their potential had they been better binned. But had they been optimized for power consumption then it would have simply and needlessly complicated the binning process, raised costs (lower yields or longer validation/test times) and ultimately created the opportunity for customers to realize there were "golden samples" to be had if they endeavored to find them.
I don't know if Apple operates in similar fashion to how TI operated a few years ago in managing multi-sourced ICs, but if they are telling TSMC and Samsung to bin any and all chips as yielding and sellable provided they meet a specific functional clockspeed, thermal, and power target then for sure no one is going to waste time and money figuring out of a certain foundry's subset of chips can operate at the same clockspeed with better thermals or power numbers. That is not something that can be leveraged to compel the consumer to pay more money for the SKU, and it would merely add cost to the supply chain side.
The way this was made possible, without making it painfully clear to the consumer that the chip inside one Nokia phone possessed superior electrical parametrics to those of the chip in another phone was by way of locking in electrical specs and then using binning as the variable in the equation.
The way this would work in practice, and I am just crudely speaking to the complexities in this post for the sake of brevity, but let's say you have Foundry A that can technically meet the electrical specs (it is within engineering margin for spice models, leakage, thermals, etc) but only just, meanwhile you have Foundry B that can easily meet the electrical specs because technically your specs (TI's in this case) are designed to allow even the weakest link (foundry) be capable of delivering chips to the packaging house in the Philippines.
So what happens in this simplified situation is that Foundry A struggles and strives to hit parameteric yields for binning your chips. Let's say, for sake of argument, they can deliver 50% yields (parametric and functional) at your required TDP, Tcase, IDDQ, TDDB, etc. Meanwhile Foundry B, with its electrically higher performing process, can deliver 85% yields at your required specs.
What is also true, but no one takes advantage of, is that Foundry B could also deliver superior binned chips with lower TDP, Vdd, IDDQ, etc., albeit at reduced yield (say 50% in this case, iso-yield normalize across the two foundries) if that was of interest to the customer (TI in this real-world example case).
But the customer in this case isn't interested in better-binned but lower yielding chips from Foundry B. They just want volumes and volumes of chips that meet the spec. And the easiest way to hit the spec is to set an operating voltage (and essentially an operating power consumption) for each foundry such that the chips coming to packaging are all operating within the same electrical and thermal profile, even if needlessly so.
In other words, to be sure for TI there were a significant number of chips from Foundry B which were essentially operating in a needlessly overvolted condition with respect to their potential had they been better binned. But had they been optimized for power consumption then it would have simply and needlessly complicated the binning process, raised costs (lower yields or longer validation/test times) and ultimately created the opportunity for customers to realize there were "golden samples" to be had if they endeavored to find them.
I don't know if Apple operates in similar fashion to how TI operated a few years ago in managing multi-sourced ICs, but if they are telling TSMC and Samsung to bin any and all chips as yielding and sellable provided they meet a specific functional clockspeed, thermal, and power target then for sure no one is going to waste time and money figuring out of a certain foundry's subset of chips can operate at the same clockspeed with better thermals or power numbers. That is not something that can be leveraged to compel the consumer to pay more money for the SKU, and it would merely add cost to the supply chain side.