Electromigration Concerns Grow in Advanced Packages | Hacker News Reader

Electromigration Concerns Grow in Advanced Packages

semiengineering.com

58 pointsPaulHoule40 comments

Effects like this make it easier to design ICs that last just long enough to get past the warranty period. I still remember when the electronics industry very strongly advertised the new "solid state" technology as lasting "forever" (in contrast to the vacuum tubes common at the time), but just as with LED lighting, that turned out to be very false.

Well, yes. Planned obsolescence and all that.

But what the article describes, and what my experience as a VLSI designer is that as feature sizes decreases, being able to just be able to guarantee a specified lifetime given environmental limits (ambient and die temperature, voltage and current supply including stability etc) is a very hard problem. And is getting harder. Just the statistical variance in the wafer production process steps (amount of molecules deposited) will make migration harder to estimate and handle.

A few SiO2 molecules more or less in thickness of an oxide layer that is 10000 molecules thick is not very important. But when the layer is supposed to be 10 molecules thick, 12 or 8 molecules corresponds to a huge difference in leakage.

So stating a minimum lifetime is what you do. A lifetime where at least 99% (or so) of the device still works reliably. Not to increase sales, but to avoid getting class action suits, having to pay out too much in refund, replace devices etc.

The article describes package failures mostly near TSVs and solder bumps presumably at the connections between chips not the oxide layers. In my experience these BEOL are a few micron metal min feature size for the routing/vias and larger for bumps.

I haven't seen the latest stuff so maybe things have change significantly with the new MCMs in the last 3 years, but I'm doubtful.

If I had to guess, these failures are at the power connections and so the solution is just more connections since the migration is very nonlinear with temp & current.

LED lighting can last "forever", but if you cheap out on power supply components and overdrive your LEDs to get more brightness for less, they won't last.

Without consumer visibility into those factors it’s a classic market for lemons, though.

indeed, it is hard to compete on quality in the long term.

If the incentive is to design for failure after the warranty expiry, isn't warranty length a good proxy for quality?

No, because the amount of people that actually follow through the warranty vs. just buying a new lamp is so small, that even with a bad product it's worth the cost of covering them.

Wouldn’t this also upwardly shift the economically optimal warranty period for high quality products?

Yeah, the only real way to fix this proactively is mandatory labeling and that requires a lot of regulatory oversight to see through.

Which isn’t necessarily bad of course, but it requires political willpower to put in place. Otherwise you’re just dealing with organizations like consumer reports trying to do their own analysis and lift the curtain, but that’s not nearly as effective as a nutrition facts style label mandated to be right on the box.

This isn’t even long term though. Every GE bulb I have purchased (8 bulbs of different styles from different stores) has failed in less than a year. I will never buy another GE lighting product again for the rest of my life.

As far as brands go, I've been decently happy with Philips, but yeah everything else is a crapshoot— Feit Electric I've had several failures, same with Ikea bulbs.

Obligitory link to the UAE edition Philips bulb that you can't get anywhere else because they don't overdrive it to thermal death and it lasts forever, but between that and being locked out of the whole market there, maybe just turn the current down a smidge for radically longer life.

https://hackaday.com/2021/01/17/leds-from-dubai-the-royal-li...

Ironic, because in pretty much all other respects, Dubai is a solid kick in the knackers to the environment.

I dunno, working hard to prevent the spread of an environmentally friendly innovation to maintain bragging rights rather than trying to spread it as far and wide as possible sounds like a very Dubai kick in the environmental knackers to me.

It's Philips who doesn't want them sold elsewhere, not Dubai.

That was a few years ago, but now they do have a "high efficiency" version available everywhere that has a similar power density, but uses a more failure-prone traditional SMPS instead of the simpler regulator in the Dubai version.

Oh good. I could have sworn I heard that the timed exclusivity requirement came from the Dubai side, but in any case I am glad to hear that they are more widely available now.

LED lightbulbs are terrible but good LED lamps and fittings do last forever!

In a lightbulb the power electronics, LEDs, etc. are all squished into a tiny package that struggles to shed heat.

I have a 15 year old led desk lamp from Philips. The Power supply is separate, the LEDs are attached to a heatsink. It could be taken apart and components replaced if someone is handy. Modern xiaomi desk lamps, ceiling spotlights and light fitting with integrated lefd are decent as well, LEDs are not overheating.

Yep, and so much is integrated now, so an average design is a bunch of tiny black squares in close formation under software control, it's very difficult to diagnose the dead component unless you are very intimate with the design or you get lucky it's failed short and the controller doesn't turn it off so it shows up thermally.

The "good" news is that lots of stuff will never get as far as an exciting excursion into unanticipated solid-state physical regimes and instead will become junk much earlier when the app no longer accepts it or the server is turned off (cough Hive)

Individual semiconductors everywhere I’ve seen (I work in that industry) still generally design for lifespans that are at minimum 10 years under worst case conditions (which is typically worse process corner plus worst temperature). I don’t think there’s much risk of having that bar drop all the way to a typical 1 year warranty period.

Silver Mica capacitors in the radios I've been working on with an older friend have the same issues... they worked fine for sufficiently long that they survived years of their first owner... and now 50 years later, they just cause "thunderstorm" noises in the audio output. (If you have such an old radio, take out all the tubes, but keep the B+, then measure the voltage across all the resistors.... the ones in the path to a leaky cap will have a voltage, because of the current flow, and the rest won't.)

I'm amazed that any transistors subject to more than 50 C for a year survive.

The ancient bipolar transistors (also the junction-FET transistors), where the active parts of the device were buried deep inside the semiconductor crystal, could easily achieve lifetimes of many decades, even at temperatures as high as 200 Celsius degrees (when designed for such temperatures). When the surface of a silicon crystal is properly passivated, its interior is extremely stable and impervious to external influences.

With modern MOS transistors, where the active parts are extremely small and located on the surfaces of the devices, it is very difficult to achieve long lifetimes unless they are strongly derated, so they do not have to support high thermal or electrical stresses.

How were the active parts "buried deep inside the semiconductor crystal" of BJTs?

Surely doping is a surface phenomenon?

Doping is not a surface phenomenon. Doping a semiconductor crystal means that some of the atoms throughout the volume of the semiconductor crystal are substituted with different kinds of atoms. The "field effect" of a MOS gate or the rectifying effect in a Schottky (metal-semiconductor) diode are examples of surface phenomena.

The dopants are introduced through the surface (except when a crystal is grown from a doped melt or epitaxially), but they are driven by thermal diffusion to higher depths.

In bipolar transistors and in JFETs the active parts are junctions inside the semiconductor crystal between regions of the same single crystal where different dopants predominate. These junctions can be microns away from the surface and in old devices or in power devices even up to hundreds of microns away from the surface of the crystal.

In MOS transistors the active parts are at the interfaces between the semiconductor crystal and layers that are grown or deposited over it and moreover their properties are very sensitive to the presence or absence of various ions that can migrate during the lifetime of the device through those thin layers.

Thank you for that detailed explanation!

What part of the circuit are you finding silvered mica capacitors?

Electromigration (EM) as a physical phenomenon can also be used purposfully. The Attopsemi i-fuse One Time Programmable (OTP) memory technology use EM to write bits that can then not be changed.

https://www.attopsemi.com/technology/i-fuse/

I suspect that modern CPUs will not have the lifespan of a Z80 of 6502.

Seeing this title was like a blast from the past for me. I'm remember being into overclocking in the 90s and early 2000s and electromigration being frequently discussed as a concern to be aware of. Then I can't recall anyone talking about it for about 20 years, even as CPUs started overclocking themselves based on thermal limits by default.

Similar to how a modern car, no matter the quality, won't last as long as a wagon.

I thought modern cars have actually gotten more reliable (eg. japanese cars can easily last 20+ years) over the past few decades? By wagons are you referring to cars from a few decades ago, or horse drawn carriages from 2-3 centuries ago?

Horse-drawn

Car Wagons are generally are just a body addon, barely integrated into the chassis and have no impact into the lifespan of the car

I suspect horse drawn wagons need maintenance more often than you think: Roads were much lower quality in that era, and wagon wheels and axles are a huge failure point, especially when repeatedly driven over bumpy rocks.

Last month I visited the reconstructed stone- and early bronze-age 'Pfahlbauten' (stilt houses) at Unteruhldingen. On display, they had a wooden cart representative of the kind that would have been used at the time, along with the observation (see quote) that the entire thing could be disassembled and rebuilt under half an hour! That makes the question not so much "how long does this last?" but "how long would it take to rebuild?" Once you start adding iron tyres, springs and bearings a vehicle becomes impossible for even the most accomplished cartwright to assemble alone.

> Wagenfunde, hergestellt vom Wagenbaumeister Niethammer, der aus Einzelteilen in zwölf Minuten wieder zusammengesetzt werden kann, illustrieren nicht nur den hohen Stand der Fahrzeugtechnik vor 3000 Jahren, sie bieten auch die Gelegenheit im praktischen Versuch Wendekreis, Tragkraft und erzielbare Geschwindigkeiten zu messen.

p.24-p.25 at https://www.pfahlbauten.de/wp-content/uploads/2020/11/5_neue...

Sandy Bridge = never obsolete.

From what I understand (having heard an EE at work talking about this), at bleeding-edge nodes, electromigration is no longer the biggest concern; the reliability of the individual transistors is. They apparently stop transisting eventually. I’ve heard redundancy is being built into chips to accommodate this.

Is this caused by quantum tunneling?

No.

Imagine sledding down hillside covered in gravel. Friction imparts momentum from your sled to the gravel, causing a few pieces to slide forward with each ride down the hill. With a thin enough layer of gravel, and with enough ride down the hill, you can end up with bare patches with no gravel.

In this loose analogy, the gravel is atoms in wire. The sled is an electrical impulse. Maybe there's some tunneling that takes place when the gravel moves from one site to another, but I'm not sure.

(I once had a job offer from Intel to work on electromigration in next gen chips, but didn't take it, so I don't know much beyond the basics.)

I vaguely remember this being discussed in a semiconductor course at uni (back in 1996 or so) - it was already an issue at the time in the smallest processes, and IIRC the prevailing approach then was to make sure that current flows both ways in the specific connections that were prone to electromigration - thus replenishing the supply.

It was a major issue with aluminum interconnects. It was primarily addressed by switching to a different metal that exhibits less electromigration, namely copper. But you have to be careful putting copper in a silicon fab, because if it gets into the silicon channel, it kills device performance (copper doping induces a deep trap in silicon).

If the erosion is a statistical process then I assume that having the current flow both ways just slows down the erosion by some factor. I assume you could estimated it using some variant of the random walk statistical model.

Reply on news.ycombinator.com