Also in regards to the actually chip-set, my issue isn't the chip itself J/GH, its specifically with the cooling of that chip and the issues you have with maintain the efficiency as transistors get hot. Lets put it this way, designing something in a computer model or simulation is much different than actual / practical numbers. This is one thing all engineers learn when they leave the classroom (good ones anyways). How many products have we seen that 'look great on paper' but completely fail when it comes time to deploy? Jumping to 16nm is very risky for any ASIC company, especially because other companies haven't paved the way much.
The dilemma you face is this, lets say you get 1,000 16nm chips, each chip can do 10gh, they are small chips, they will probably run super efficiently, stay cool, but how many does it take to make a complete competitive miner / operation? As you increase the die size, you save some money, say 500 16nm chips at 18gh a piece. Now you have more heat issues, your efficiency drops, but it costs less. Same thing with larger runs and so on. More circuits, less circuits, more complex boards, boards melting, these are not easily 'modeled' on a computer and often require samples.
To be honest, this is why chips haven't been dropping in size as quickly recently, the efficiency gains are getting less and less, and the cost benefits are questionable - even for the mighty Bitcoin ASIC's.
The list goes on and on for the complications you find with the smaller and smaller chip sets. And at the end of the day, if the ROI of the equipment will take someone 1-2 years to return even @ lowest from the wall costs, whats the endgame??
i think you've misunderstood something. the size of the asic doesn't change its power consumption (watts per gigahash/s). it doesn't change its efficiency. The cooling of the system has to be designed to be appropriate for the heat output of the asic.
its the design of the hash engine, and process technology used that largely dictate the power consumption (and of course, the voltage its run at, which is clearly a design and process issue)
the hash engines are small and highly repeatable. there are tens to hundreds of hash engines in most bitcoin mining asics. perhaps even thousands in some.
the actual size of the asic - i.e., how many hash engines you put in the asic, makes almost no difference to the power consumption (on a w/gh/s basis).
so its up to the asic designer to put the right number of hash engines into the asic that they can efficiently cool the part.
however, the size of the asic does have an impact on system cost. since the packaging and pcbs costs are directly related to how many parts (packages) you have, there is a cost advantage in going for larger packages that pack more hashing engines into bigger asics that consume less board space. also, the on-chip circuitry is more efficient and less losses than the off-chip (on board) circuitry, so minimising the latter is desirable.
so in an ideal world, you would put as many hash engines as possible into a package to reduce the number of packages and keep system costs down. however, as you rightly say, there is a cooling impact. when you multiply the power consumption of each hash engine by the number of engines you have, you get your power per package. there are different ways of cooling different amounts of power.
for tiny dies in small packages, like what bitfury and asicminer use, they have very few hash engines in each die... and they keep the power per package to around 2 watts, which is air-coolable without heatsinks, just the packages and board itself become the heatsink. its cheap and easy to build.. however you need to have a huge number of asics and very large boards, and lots of boards, to produce the same hash power as a more densely packed asic that has more hash engines and runs hotter. its also potentially more expensive, as the (asic) packaging cost becomes higher (so many packages for a given amount of hash power adds to the cost). the design process for designing smaller asics is easier, but it doesn't cost much less, as the NRE is the same no matter how big the asic. And the production is wafer based. In theory you would have better yields with smaller dies, but in practice, bitcoin mining asics have very good yields as they have so much repetition on the die, that any individual defect doesn't make the asic useless (like it would with an Apple A7 asic)
thus the next step up is the spondoolies style, where they run the asic at probably around 50 watts, and use heatsinks to keep it cool.. but they have a lot more hash engines per package than asicminer/bitfury type design.
cointerra's 16nm design was similar this. its a good balance of asic density versus power and cooling requirement. the goal to bring the overall system cost down.
then you have the kncminer approach which is to pack a lot of engines onto one package and have a lot of cooling on each package. probably 200+ watts per asic? with huge pc-style cpu coolers. luckily this method works because pc cpu coolers are plentiful and there's good ones out there that aren't expensive.
and lastly you have the cointerra and hashfast (late 2013) approach, which is to have an enormous number of hash engines on an asic, and run it very very hot (400 watts) and require exotic cooling technologies (liquid cooling) to extract the heat. it was this approach which made cointerra's systems unreliable because they relied on a cooling tech that had a lot of moving parts (liquid, pumps, fans), and the cooling tech wasn't anywhere near as reliable as it was claimed to be. and they had four of them in each system, which meant much higher than acceptable failure rate. not of the asics, which was reasonable.. but for the cooling parts )CoolIT), which let them down badly.
anyway, sometimes you've got to try new techniques to find out that earlier techniques were better all along, and cointerra's 16nm system design didn't share any of the design elements of their previous one. they learned a lot from their experience. the 16nm system had well designed asics, well designed systems, and well designed cooling, and it was optimised to be low cost and high reliability. unfortunately, they didn't raise enough cash to build any of them, even though the design was complete and ready to push the button into manufacture.
