If I buy the Industrial and Commercial variants of the same IC, besides the markings what differences are there in the actual parts?
Seems like they must use the same die, so is it different formations in the packaging material? Different types of bond-out wires?
Or is it a matter of binning where the higher grade parts have been tested and found to work at the wider temperature range?
Parts are tested after production to ensure that they meet the datasheet specifications.
Commercial parts are those that meet the datasheet at commercial temperature but failed at the industrial temperature.
Industrial parts are those that meet the datasheet at industrial range but failed at the automotive/military range.
If a part is offered in both industrial and commercial grades, it is possible that the commercial parts are dropouts from testing to industrial grade.
However, it depends on relative volume of both grades, as dictated by customer demands.
Making industrial-grade ICs requires tighter design to accommodate wider process corners and use of more expensive foundry libraries, and fabbing also cost more. It also requires more conservative layout technique, wider interconnect, double vias, etc. To achieve good integration, a foundry usually offers a set of standard cell libraries for every taste, but the cost differs as well.
If majority of customers for a particular IC demand industrial grade, the IC will be designed and fabricated as such, and test dropouts would sold as commercial ICs.
Getting to industrial grade might also include a change in IC packaging, to use, for example, a metal-ceramic instead of plastic.
However, if the volume for commercial grades is high, the die can be taped out with less expensive libraries and manufactured by less expensive variant of the fab process, so save on cost and maximize gross margins. Then the same functional IC will come from different wafers for different grades, and will go through a different set of testings.
Design for higher temperatures is more of a pain because of more device param variation away from nominal.
FETs transconductance is lower, thus more W/L is needed (more area, higher die cost).
Leakage currents will be higher, thus larger sample-hold capacitors are needed in the ADCs or elsewhere.
Given certain headrooms, with operating conditions of transistors (bipolars and FETs), some circuit topologies become impossible and other topologies are invented.
Then you can examine the down-well rock-logging circuits of Schlumberger, at 400 degree.
Sometimes the difference is in where on the wafer the chip was. In the center, where image (of photolithography) is most accurate, it's wider range, better yield, faster switching (for CPU and memory). On contrary, edges are slower and go for commercial range.