Of course binning makes sense with regard to semiconductor products manufactured as wafers.
However, it only makes sense if you are going to market products (CPUs/GPUs/etc.) which have very similar hardware/functional capabilities.
Different product models being based on the same die design allow the manufacturer to serve different market segments, while investing in the design process only once and requiring only one production process (i.e. fab setup, wafer testing regimen, etc.) for those models. Ongoing adjustments and improvements of the die design ("stepping") will henceforth benefit not only a single model, but a all of those models sharing the same die design at the same time.
After a fab has been setup for the production of a new die design, the yield rate of early production runs (not only) of the higher-spec model is lower than what one can expect a few years later when the parameters of the production process have been further fine-tuned. Thus, binning not only is a means to serve different market segments, it also allows a manufacturer to keep the value per wafer as high as possible.
Since die space is a major factor with regard to cost of semiconductor products such as CPUs/GPUs, binning only makes sense where you can economically share a die design across different CPU/GPU models. Hence, do not expect a 6-core i7 to be based on the same die design as a 2-core i3 with much smaller caches. You would be hard pressed to find an 2-core i3 who has the same die as an 6-core i7 while essentially wasting most of its die space for unused cache and cores. That is not say that such a thing could not have happened in the past or will never happen in the future. If during production testing (which will happen before marketing will announce any new CPU models) it becomes clear that (initial) yield rates for high-spec models will be particularly low, then it might entice a manufacturer to bin dies not meeting high-spec specifications as very low-spec models. This could lead to the introduction of a rather low-spec model into the market which essentially is having the same large die as a high-spec model. Also, don't forget that the decision of whether a manufacturer will go forth with a very low-yield production of high-spec parts is a business decision; depending on solvency, current and projected future market situation, gross margin estimations as well as other business-related factors.
Now, i can't tell you which particular CPU models would share the same die design and thus would have been binned. As a rule of thumb, if there are feature differences between models that would also indicate significantly different requirements for die space (such as different number of cores, different cache sizes, absence/presence of integrated graphics), expect the models to have different die designs. But as outlined in the last paragraph, you cannot be sure for certain. (If you want to know, i would suggest to search the internet for die photos of CPUs you are interested in and compare them.)
Electronic components which are not made in numbers on a single waver might not benefit from binning as much. It could still make sense to bin to account for production tolerances (like resistors, for example), but usually you have to meet certain given production targets without regard to the daily ups and downs within the range of allowable production tolerances. If you have to deliver 10 million 5% resistors, but all your production lines spit out resistors which all meet 1%, you are not wasting time and money to screw around with your production. You will simply declare 10 million of your resistors to be 5% even if your resistor fab happily spits out 1% resistors. (Now, if you can saturate your production facilities by just filling orders of 1% resistors you don't have to worry about those 5% resistor deals.
)
Home
Help
Search
Login
Register
