Hacker News Reader

Comment by blackguardx | Hacker News Reader

The only parts on that list that would have different grades are the capacitors. It isn't a measure of quality, but the temperature and stability specs of the dielectrics.

EDIT: Source: I'm an electrical engineer and design electronics for a living and have even worked with ex-Apple engineers.

View on HN

I know some processors (mostly of higher end processors) do but I don't know if specifically that processor does. I do know that at least some manufacturers will offer automotive and aerospace grade parts if requested (if not for the extra money).

There's also the fact that it's off the shelf completely because there are none left to buy because somebody else already has them. If I remember rightly it can have something to do with the component failure rate on a silicon slate.

I've also heard of stories where processors that have a co-processor by design fail for some reason and are sold for their larger processor capability only. I've seen this with an ARM chip.

There isn't a single mass market microprocessor (whether it is AMD, Intel, Qualcomm, NVIDIA, etc.) that is manufactured to be sold as a single SKU (product line like an i7-6950X). That would be economically impossible because they are complex pieces of technology where a single tiny scratch or missed dopant, errors that number in the thousands or more per wafer, can ruin a major piece of functionality. Microprocessors are designed with special "fuses" so that when there is an error, the failed part of the chip is disabled and the part is downgraded (for example, from an i7 product to an i5 product). They also go through burn in tests to test the stability of the part at various voltages and temperatures and that's how they separate automotive from regular parts (or chips meant to be sold as overclocked 2400 MHz DDR3 from regular 1600 MHz ones is another example).

You are right that is deliberately done but wrong about how often. This is literally how modern mass manufacturing of complex silicon parts works and the only companies that don't do this make extremely specialized and expensive processors like IBM does for mainframes, supercomputing, and RAD hardened applications. Those chips can be several times bigger than your tiny ARM or Intel microprocessor for the same reason.

I didn't know how common that was. Its definitely interesting, I was speaking from personal experience with these packages.

I remember being offered chips with lower failure rates too, I can't remember the ratios but it was certainly worth considering. We ended up testing the different boards off the line before doing a larger assembly (and further testing of course) because of the component failure rates. If it wasn't the solder it was the chip itself - causing really hard to find bugs I might add. I remember having two boards off the same line together, one chip could do an operation and other couldn't. This bad chip was being used for software production too...

Same goes for high precision resistors, capacitors etc. for hifi equipment. They don't have a special machine producing extra accurate components, they just take the produced resistors, test them, and bin them into accuracy classes. The high precision ones are those that happen to be very close to spec.

This is extremely interesting. Do you have any further reading to learn more?

Unfortunately, while I could point you to great books on electrical engineering in general, I don't know of a single online/book resource on this topic because I learned all of this on the job while apprenticing and contracting for many different companies. For further reading, I would probably start with semiconductor and electronics manufacturing/supply chain books of which there are dozens. If they don't cover the nuances of dealing with part variations then they are likely not worth reading. Sorry I can't be more helpful.

That's okay, thank you for sharing your knowledge. I'll take a look into the books you recommended.

I think one of the first 'mass market' processors that used this technique was the Intel 486SX, back in the early 1990s. It was simply a 486DX which had the fuses connecting the FPU blown. I believe binning by clock-speed has been used for a lot longer; the 486 33MHz parts being the same as 66MHz but having failed some sort of QA, which meant that overclocking could be attempted at fairly low risk, as long as attention was paid to cooling and you didn't mind the risk of the CPU failing completely earlier than expected.

EDIT: There certainly seems to be different quality levels for this particular processor anyway [1], which I'm sure they'll want a premium for higher quality. Feel free to correct me though.

[1] http://www.ti.com/product/MSP430F2013/quality

Which one is "higher quality?" Those are different lots with the same MTBF?

Those are various combinations of IC package, temperature grade, and bulk packaging. They are demonstrating that all the different types they sell have gone through environmental testing. There is no such thing as 'higher quality', generally, but a part with extended temperature range or a faster speed grade does cost more.

I think perhaps my wording was misplaced, but a part that has been shown to reach high temperatures might be more preferable in a charging application - even if they blow up anyway.

Reply on news.ycombinator.com