There is no good reason to think that Qualcomm's QC people did not detect the problem. It's what you do about it that is the problem.
That there is or ever was a problem is denied by Qualcomm, some users and some independent testers. But at least one reasonably well documented test set indicates that when running benchmark programs the processors tend to become more heat affected sooner than previous Qualcomm processors and than competitive alternatives.
High performance ICs generate substantial thermal output and all things being equal , output increases approximately linearly with throughput (as a major factor in energy loss is related to capacitive node charge / discharge, with energy dissipation being proportional to switching cycles per second).
In new device aimed at being near the front in a competitive market, the manufacturer may strive for raw output first and efficiencies second. Temperature relates to both thermal output and the ability to handle the heat. A device that is pushed towards the upper limits of a technology may be less efficient and require the device user to take more extensive efforts to cool it. In the finite space available 'corners may be cut'. Some manufacturers (eg Samsung) have gone to newer lower lithographic width devices to achieve the needed efficiencies. Qualcomm chose not to and as the devices are made by a third party, the drive to be faster sooner apparently swamped the awareness of the need to be better cooled. Some prospective users have addressed the problem by "baling out". Others by denying the problem exists as long as possible and then "writing software" to fix it. This is no doubt a painful learning experience. The project manager is unlikely to have the same problem again as (one can guess) he probably now works somewhere else, and Qualcomm will no doubt take greater notice of their QC department.
The "answer" very much deep-ends on who you believe.
I'm not an expert at all on this. I asked Google - and found that a lot of people are not experts but there does seem to be at least one good looking answer available as to what is really happening.
It seems they pushed too far too fast, used a fab house who was good but not as good as their cooperative opposition (Samsung) and allowed themselves to be less 'hands on' over the results until too late. (Almost*) ALL multicore processors in top end phones run into thermal speed limiting with time. Some limit more than others. The 810 is worst than most. (The Snapdragon 801 in eg the HTC One M8 at 2.3 Ghz seems to be an exception to the general rule).
The following does not so much advise "how" but rather what did or didn't happen and how transparent people were about it and what is really happening and when it was known. The following are almost in 'order I found them'. I could have reordered by date but its interesting to see that the date and what is said largely do not correlate well.
Here on May 6th Qualcomm deny that there is a problem in commercial products, as opposed to preproduction prototypes.
Here on April 4th various user and manufacturere's spkesmen are being less than traansparent.
Here on June 15th !!! Sony are claimed to have admitted to problems with the 810 in their Z3+ after users demonstrated the ability to crash it by recording video. Sony advice a software fix will be provided. A software fix is a rather suspect way of fixing thermal issues.
Here on January 29th Qualcomm adumbrate that Samsung had decided not to use the 810 and to use an internal Samsung processor instead.
BUT on April 3oth the same source Qualcomm indicate that samsung may manufacture the snapdragon for the in future, and they say a decisin by LG to use a qualcomm hex core processor instead of the octal core snapdragon was made a year ago and is not due to any overheating issues and probably says that they will limit power use by reducing processing speeds for selected applicaations (where selected might mean "all". .
And/but on Feb 5th it is advised here that there never was a problem and BESIDES it has already been fixed, maybe.
Which leads to the
Best picture you'll get, probably:
Fairly definitively, on Feb 24th here some actual thermal tests. Nice time versus processor speed tests for range of processors. What happened? Qualcomm pushed things a bit far. the processor was made for them by TSMC in Taiwan. The device uses a number of ARM Cortex cores. How these are implemented varies with manufacturer. Samsung are using a smaller transistor process than TSMC. The smaller lower power process is really needed.
In short, chips throttle, but the 810 throttles more than most, and it's severe enough that the 810 is actually slower than the 801 or 805 in some CPU-bound tasks over the long haul. The Exynos 7 Octa, which has similar specs on paper, is much better in practice.
At this point, Qualcomm has implied to us several times that its use of ARM Cortex CPU cores was a stopgap measure—Apple got the 64-bit A7 chip to market around a year before anyone expected it to. Chips are designed over a period of two or three years, so using the ready-made Cortex cores were the quickest way to get a 64-bit response to market.
The results, unfortunately, don’t look great. It might be because the 810 is using a 20nm TSMC manufacturing process instead of the 14nm Samsung process used for the Exynos, or it might be that Samsung has more experience working Cortex CPU cores into its designs. Whatever the reason, our testing of real phones with these SoCs in them shows that 810-based phones are slower and have worse battery life.
Qualcomm's next major flagship is the Snapdragon 820, the first to use its custom-designed 64-bit “Kryo” architecture. Rumor has it that the chip will be made on the same 14nm Samsung process as the Exynos 7 Octa.
A return to its own CPU cores plus a newer manufacturing process should hopefully mean a return to the kind of performance and battery life we’ve gotten from Qualcomm-based phones in years past. All signs point to the 810 being a one-time slip-up and not the start of a trend—let’s hope that those signs are accurate.
AND just to add balance, on March 2nd,
these people - "SemiAccurate" - say there is no problem never was a problem, it's all made up and here are the lab tests to prove it. They claim to have investigated in depth and it was essentially a FUD campaign by a Korean phone manufacturer whose name begins with Samsung.
I liked the time/temperature graphs.