Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Lasers and masks, no. Fuses, yes. But fuses are just non-volatile memory. The fuses are not connected directly to the hardware. There is either logic or microcode that loads the fuse values into registers that actually configure the hardware features on and off.

So the question becomes whether you can access those registers, or otherwise affect the fuse loading process, using these features. That depends on the design (e.g. whether those registers are accessible and not locked down after initial load).

Fuses (and antifuses) are the only form of nonvolatile, programmable memory available in these process nodes (as they require no extra masks or processing steps beyond a regular CMOS process), that is why they are used. They are write-once memory, and they are based on breaking or making electrical connections, but they are not literally used to directly reconfigure chip wiring; that hasn't been the case for many years now. Modern fuse memory is just one more functional block you throw into your chip, and it comes with its own requirements, read amplifiers, etc (the fuses aren't actually "binary"; what happens is the electrical resistance increases, but you still need some analog-ish circuitry to set threshold levels to read them reliably, and carefully controlled programming voltages and timings, etc).



There was at least a long period of time until recently that Intel parts didn't even tape out ECC for consumer designs. That's what I meant by masks, their product differentiation for an individual chip in that case is all the way in the mask. See how here all of the chips Intel enables ECC on are Xeons or the occasional Atom intended for servers? https://ark.intel.com/content/www/us/en/ark/search/featurefi... The only question there is if they still have it taped out on some other masks, but universally disable it on those chips as a later product decision. There's a tiny chance that's the case so that they can make more real time product decisions rather than having to wait out the multi year process to tape out new chips.

And lasers were used at least very very recently at Intel.


Look at E3-1245 V3, then 4770k, and finally 4770. Can you spot a difference? Its ECC and TSX-NI. Same die.


Right, so when I said ECC _might_ not even be taped out, it obviously depends heavily on the die. Sometimes they repackage Xeon dies as extreme high end consumer parts, but it's very rare and doesn't apply to the general case (and your examples are almost a decade old at this point).


Intel don't use something like eFuse by Qualcomm, or it's same thing?


Not sure what specific technology they use/license (they might have an in-house design), but it's all vaguely equivalent from a high level point of view. Qualcomm calls theirs QFPROM.

In the end, all of these companies end up licensing whatever fuse block is compatible with their process, e.g. I'm sure TSMC has something in-house for theirs. Intel of course own their own fabs, so they probably have their own. DesignWare also have their own (generic-ish?) thing, etc.

Basically every modern SoC or really IC of any complexity uses fuses for something (e.g. calibration data, switching in spare/redundant blocks when manufacturing defects happen, binning/product segmentation, crypto key storage, etc).




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: