The transparent chip inside a vintage Hewlett-Packard floppy drive | Hacker News Reader

The transparent chip inside a vintage Hewlett-Packard floppy drive

righto.com

190 pointspicture45 comments

> HP achieved a yield of 9%, meaning 91% of the dies failed.

This line really stood out to me - I was aware early semi yields were poor, but this poor? Was this par-for-the-course for technology at that time, or was this bad even by 70s standards? What do typical yield numbers look like today?

I recall reading that a brand new cutting-edge process node might get 60% yield, but a completely mature node is circa 90% (if memory serves). Yields for a new process usually go up over time as manufacturing bugs are worked out.

Isn't that survivorship bias?

Processes that never improve yield are abandoned

no, it's just what happens with any process; as you gain more experience with it, you learn more about it, so you can get more and more controlled results

Yes, and companies like TSMC and Intel probably have vast internal documentation about where and how their processes can improve yield as they run their newest fabs. The research required to implement a new process can take decades to get to the mass-production stage. Given the billion of dollars at stake and thousands of human-years of R&D needed, none of these companies are "winging it" and their scientists and engineers know how to plan ahead to the best of their considerable abilities.

For a glimpse at how they do things, see this fascinating talk: https://www.youtube.com/watch?v=NGFhc8R_uO4

> Given the billion of dollars at stake and thousands of human-years of R&D needed, none of these companies are "winging it"

You'd be surprised. My personal knowledge of one major fab is that they have policies such as regular rotation of employees which make it difficult to maintain institutional knowledge. I have heard complaints from multiple people of their teammates and managers "winging it" to think it's isolated incidents.

Cynical thoughts along these lines occurred to me after I posted my comment, so no, I'm not surprised, just disappointed. Especially not after watching Intel slowly rot from the inside (good luck with your renovations, Mr. Gelsinger). My workmate at my last job (pre-COVID) knew someone who worked at Intel, and our discussions about the company tended to contain the phrase "Intel is fucked" an awful lot. This was around the time I bought a nice chunk of TSMC stock...

Historically, maybe, but otherwise that's not how it works.

Nowadays, when a new process enters risk production, yields are expected to be sub-optimal (I think I read that some of the recent stuff TSMC was making for Apple was below 50%) but it's an economic imperative that they improve (and they do). Abandoning a process once you've built an assembly line would be an existential crisis for a high-end semiconductor company, given the billions of dollars it costs to get to that point (which is a big reason why Intel has been between a rock and a hard place for the past 5 years or so).

I wonder if the hard part is interfacing the sapphire with the silicon, heterogeneous materials like that are always an issue

In 1977, everything was hard.

Growing just silicon crystals was still difficult let alone on sapphire--3 ionch wafers were the norm. Lithography was contact so degrade your masks over time. Threshold voltages had poor control. CMOS chips were very susceptible to ESD punchthrough. Latchup was murderous (SOS helps cut this down greatly). You had purple plague. I can go on and on.

interestingly, the day after you posted this, popular youtuber nilered published a video in which, after staggering amounts of effort, he successfully makes a beautiful finger ring for himself from purple plague

https://www.youtube.com/watch?v=d6Pcp944sRI

I spent a few weeks at the HP Cupertino facility in 1982, and got a tour of their chip and computer production facilities too. They were very proud of their SoS (Silicon On Sapphire) technology, which was fairly new at the time. I think they moved all the chip production to Singapore shortly thereafter.

https://www.hpmuseum.net/divisions.php?did=2

https://venturebeat.com/business/hp-shutting-its-cupertino-c...

It may worth mentioning the modern successor of the SOS technology is SOI. A thin film of crystaline silicon insulated from the bulk by a thin film of SiO2 is used for fabrication of the circuits.

Author here if anyone has questions. A bit of a change from my usual x86 chips...

Wonderful work! Your articles are my favorite coffee reads

> Inconveniently, metal that contacts silicon, metal that crosses over silicon, and metal that forms a transistor all appear very similar in this chip. This makes it more difficult to determine the wiring.

I've heard that all diodes are LEDs, if one's vision extends to the infrared. Is it possible to fire up the chip and photograph it in the infrared, and would that make the transistors stand out?

High resolution thermal imaging is tightly regulated

The emission from silicon referred to by myself248 is in the near infrared, not the thermal range. It's a commonly used diagnostic technique for silicon solar cells:

https://www.pveducation.org/pvcdrom/characterisation/electro...

I don't know if it could be used for the much smaller features found on integrated circuits, though.

As the restrictions are now blocking the IMPORT of such equipment it really is getting to be quite a joke. In the US we are only allowed to have expensive potato quality cameras with 9fps and even then that's not shitty enough -- they also have to have to inject artificial noise into it!

Thermal is not the same as near-red IR, which is what is being discussed here.

Why?

Because the government knows IR would be a pivotal tool in a civilian guerrilla uprising, and they want to keep such things out of the hands of the proletariat?

Got it. It was a genuine question asked out of ignorance on my part.

How would the extra resolution be useful in warfare?

Anti aircraft missiles (everything from plane launched to manpads), heat seeking weaponry capabilities, night vision, multispectrum sensors, optical gas imaging and a whole host of other dual use scientific applications. High resolution thermal vision is the lynchpin behind all of them. I am not sure why they are still tightly controlled since you can easily acquire them overseas.

The US government is terrified of its citizenry, as they are the only people that stand a chance of “fixing” it.

Well, John McMaster was able to do some IR imaging of a chip: https://hackaday.com/2019/06/20/seeing-transistors-switch-in... But in this case, I think it's easier to squint at the die to distinguish the features by subtle differences in color. It's possible to tell them apart, but I wasted a lot of time before realizing "Oh, that's a transistor‽"

I've been successful at making many electronic parts light emitters even when they're not designed to be.

Some last longer than others under the conditions.

Since you mention it was a defective part, were you able to see the defect on the die?

We didn't see anything obviously wrong with the part. One bit of communication with the microprocessor was flaky, so it was probably a weak transistor. The consequence was that the Z80 attempted to look up a command in a lookup table, but the command wasn't there so it ended up looping through all of memory, hanging the floppy drive.

I'm not informed enough to come up with a good question, but I love your work! And I love the concept of chips built on sapphire, even if it didn't win in the marketplace. Could we get Ruby running on ruby? :P

Ruby (the gemstone) is just a red sapphire, in fact[1]. So maybe?

1: https://geology.com/gemstones/ruby-and-sapphire/

You mentioned that SOS still has a niche. What examples of current RF devices are using SOS?

Peregrine Semiconductor makes various SOS products. Here's a random example, an RF digital attenuator: https://www.psemi.com/pdf/datasheets/pe43205ds.pdf

Thoughts on distinguishing similar-looking things:

* take multiple pictures from slightly different angles

* take pictures using sensors for different frequencies (can be done with a colored transparent sheet, though simply using different cameras will have some effect)

* shine light from different angles

* shine different frequencies of light at it

Is a modern yield for SOS still as low as quoted?

I don't have figures for modern yields. I hope the yield is better. But I think there are fundamental difficulties with matching up the sapphire and silicon crystals that tends to produce faults.

I'm astounded HP made commercial products out of a process with such a low yield, I dont know how that was economical. (though none of the HP gear was exactly low margin)

The HP Journal article mentions the 9% yield as if they are perfectly fine with it. They were building these chips internally for high-margin gear (as you said), so it probably didn't matter if the chip cost $10 or $100, as long as it has the performance that they need. It's a completely different market from microprocessors such as the 6502 or the 8086, where customers are looking closely at the cost of the chip.

and RF components are still in that space to a greater or lesser extent too.

> I'm not sure what the oval stamp at the top is, maybe a turtle?

It's a teapot, on its head. How odd!

Perhaps https://en.m.wikipedia.org/wiki/Utah_teapot

> Since the sapphire substrate is an insulator, transistors are completely isolated, unlike a regular integrated circuit. This reduces the capacitance between transistors, improving performance.

That sounds like it would be very good for analog, like op-amps.

Someone mentioned Peregrine; here's a presentation from them on SoS and why they're using it for RF: https://www.psemi.com/newsroom/blogs/15645-part-5-why-silico...

Don’t tell the audiophiles.

SOS-preamp here we come!

I wish I could find a sample of a transparent (silicon of sapphire) chip to hang on the wall, maybe with a magnifying glass nearby.

I checked eBay, but did not see any.

Reply on news.ycombinator.com