3
\$\begingroup\$

One-time programmable (OTP) fuses are common in ICs. How are they typically implemented, on the silicon level? I know that they generally work like larger fuses, i.e. they can be blown but not repaired, but how is this actually accomplished when thousands need to be fit in the die area (e.g. Intel chipsets)?

Is there a difference between how they are implemented between low-power cheap microcontrollers (which might have a few OTPs for lock bits) and high-end x86 chipsets (with potentially thousands of OTPs to store manufacturing settings and even entire RSA keys)?

\$\endgroup\$
9
  • \$\begingroup\$ Like a fuse: thin metal track which can be blown. Why do you think these take more space then an SRAM cells which needs 4 transistor per bit? \$\endgroup\$
    – Oldfart
    Commented Sep 2, 2019 at 3:15
  • 1
    \$\begingroup\$ Yep! But they often put two per bit in case the first one does not blow. \$\endgroup\$
    – Oldfart
    Commented Sep 2, 2019 at 3:17
  • 2
    \$\begingroup\$ Can't help you there. You are now asking about the silicon IP which is very, very protected. I just had to work with them for months on end, but the library group designed them. (And yes, they are bigger then an SRAM cell, but then you don't need two or three million of them) \$\endgroup\$
    – Oldfart
    Commented Sep 2, 2019 at 3:23
  • 3
    \$\begingroup\$ @forest No one lets copper get ANYWHERE NEAR a FAB. When I was visiting one of the HP FABs in the SF area, they'd only just recovered from some electrical wiring contractor who'd accidentally allowed a very tiny snip of copper to get into a wash area of the FAB. It tooks weeks of work to remove the problems that caused. (There are newer copper processes, but that changes nothing about what I just said. Copper is POISON to many FABS and especially back in the day when OTP was prevalent.) It's likely platinum+silicon or aluminum or just a fragile FET. \$\endgroup\$
    – jonk
    Commented Sep 2, 2019 at 4:10
  • 1
    \$\begingroup\$ @forest It's still used. But like anything, how they do it differs a lot over time and depending upon other business issues. Today? Aluminum a few hundred angstroms thick might be used, for all I know. But circa 1990 or so pretty much all hobbyist programmable MCUs were OTP. I used many Microchip PIC16C54 to PIC16C57, for example. These OTP devices actually used UV-erasable ROM. It's just that they didn't package them with a quartz window, but instead they used epoxy. So there, the OTP bit was actually a complete UV-erasable structure and not a simple bit of metal. So no single answer. \$\endgroup\$
    – jonk
    Commented Sep 2, 2019 at 4:19

1 Answer 1

Reset to default
3
\$\begingroup\$

In the IC processes I have worked with, the fuses which are used for OTP memory are made from polysilicon.

They're simply polysilicon resistors with a value of a few hundreds of ohms when intact. After such a fuse is blown the resistance increases by a factor 100 or so to at least 50 k ohms if I remember correctly.

It is mentioned in the comments that metal can be used for the fuses and that is true, see this article.

So the OTP fuses can be made from polysilicon or metal. Other materials might be possible as well.

Is there a difference between how they are implemented between low-power cheap microcontrollers (which might have a few OTPs for lock bits) and high-end x86 chipsets (with potentially thousands of OTPs to store manufacturing settings and even entire RSA keys)?

There probably is, the "fuse based" OTP memory is quite "expensive" as it needs a lot of silicon area per stored bit. On a microcontroller chip where there is already flash memory available it might be easier to implement the OTP lock bits using flash memory as well. The logic circuits then simply prevent further programming when a lock bit prevents it. That way no fuse is blown but you cannot program the chip anymore.

A similar scheme can be used for manufacturer programmed settings, once set the chip prevents the settings to be changed.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.