Raspberry Pi Zero 2 W die shots | Hacker News Reader

Raspberry Pi Zero 2 W die shots

electronupdate.blogspot.com

190 pointsgeerlingguy74 comments

> When one codes on a Raspberry Pi one generally thinks of these CPUs as the core of the device with all the peripherals surrounding them... however the provenance of this chip is much more interesting and the CPUs are actually an afterthought.

> The primary design goal of this chip was for cable TV set top boxes. For that the designers focused on transporting video. The real core of this chip is the "Video Core IV".... it's the actual master.

And fun fact: when the Pi boots up (all generations so far), the primary bootloader is the VPU (VideoCore), which then initializes the CPU and RAM.

These things are slightly weird in that way.

What does the VPU do after boot if the unit is headless?

Well you could use it for GPGPU[1], for example[2].

[1]: https://www.cnx-software.com/2019/10/10/py-videocore6-raspbe...

[2]: https://github.com/Idein/py-videocore6/blob/master/examples/...

E.g. accelerating video decoding/encoding, OpenGL applications, etc..

> The only downside is that external light can actually affect the chip operation (light hitting the die releases electrons at PN junctions creating errors in circuit operation ) .. .however here the Faraday metal shield helps prevent this.

Brings me back to when it was discovered that high intensity flashes could make a Raspberry Pi 2 reboot...[1]

[1] https://www.raspberrypi.com/news/xenon-death-flash-a-free-ph...

Many PN junction based components such as diodes and transistors are photosensitive; I recall after reading this as a kid on an old electronics magazine, I took a 2N3055, a pretty common power bipolar transistor back then, uncapped it to expose the die then connected it to my analog tester and saw it generating current when exposed to a lamp. The effect is common also in devices one could never suspect to be sensitive to light such as LEDs. They can produce light but also detect it. A few years ago someone published a project online of a MCU based keyboard (actually a button board) made of PWM driven LEDs that can both produce light and during their off cycle detect it, so that one turns them on and off just by touching them with a finger screening them from surrounding light. I attempted to search for that project without luck.

I remember that project too. Might have been Paul Dietz at MERL.If you're curious, look up Paul Dietz "low cost bidirectional sensing using LEDs" and Forrest Mimms "LED Sun photometry".

> Forrest Mimms

There's a name I haven't seen in a long time. Source of endless insight into electronics parts and schematics.

He did a series of columns for Make Magazine. A compilation book titled "Make: Forrest Mims' Science Experiments" was published in 2016.

https://www.makershed.com/products/forrest-mims-science-expe...

A self-directed project I did for a university course was to make a Morse code transmitter and receiver. I was able to use the same LED for both.

By using the same LED as a receiver too?

https://makezine.com/projects/how-to-use-leds-to-detect-ligh...

It never ceases to amaze me that someone can look at these die pictures and knows what’s going on. All I see is a bunch of copper colored squares

I honestly thought being able to recognize ground up spices by eye was an amazing feat until I started cooking more than once a month.

Its not far off from looking at a circuit board and recognizing layout patterns and chip shapes to get a feel for whats going on. e.g. Larger chips are usually of interest. A larger chip with one or more little rectangular chips next to or surrounding it is likely the CPU and RAM/Flash/ROM. Smaller chips are interface or power regulators depending on their proximity to ports and other components. Groups of inductors, capacitors and transistors are likely power supplies. It all comes with experience.

You develop your intuition by looking at (or designing) a lot of these. It's like any other profession, I guess, experience makes pattern matching easier.

after some time it's fairly easy. E.g. if you know it has a 4 core cpu, you're just looking for the same area four times.

> It might grate on some hobbyists who see this secretive approach as the antithesis of open-source but Google Broadcom's stock price and one can see it's been a fabulously successful company.

I am really interested to see what the people on HN will make of this statement.

I certainly don't understand why so many jumped on a bandwagon of a SBC which was basically an obscure GPU with some accidentally attached ARM cores, on an I/O starved SoC design, that still doesn't have a useful opensource bootloader even after 8 years or so,... when much more interesting, more open, and chepaer alternatives started appearing rather quickly after the first Rpis.

IMO, it comes down to the support and ecosystem. Even though it would be nice to have something more open than Broadcom's chips, the corporate/foundation backing of the project, and the broad adoption after those first few revisions (coupled with the fact that Pis are still some of the cheapest options around) means that someone can come into the ecosystem blind and not have to spend hours or days just figuring out how to get an OS running on the darn thing.

And then the fact that you can still run a supported modern OS on an almost 10 year old board... that doesn't seem to happen with other SBCs (most are basically shelved after a couple years), and you end up running ancient OSes on them or obscure community-driven OSes that break in a hundred different ways.

There are a lot of boards that have mainline u-boot/Linux support, don't need neither any wonky 4MiB binary firmware which does who knows what, nor any vendor specific tools to control basic things like HDMI, or CPU frequency, and just work for years with any arm based distro.

Personally I think it's just a better marketing.

That wasn't my experience, pre-raspberry-pi.

There were projects, sure - but they rarely sold in large volumes. That meant not many paid developers. And if you had a problem you'd be lucky if someone online had even experienced the same problem, let alone found a fix for it.

Every chip vendor seemed to have their own kernel fork with their own patches, and didn't much care to get them upstreamed. And of course, their kernel would be ancient - they're selling it for things like automotive infotainment systems which don't give a shit about security.

Want to attach, say, a battery-backed RTC to your SBC? I hope you know the difference between a dtc and a dtbi. And even if someone else has been nice enough to get the same part working on their SBC, just because a TI processor needs the line 'dmas = < 0x21 0x1c 0x00 0x21 0x1d 0x00 >;' doesn't mean you can copy that to a Freescale processor and expect it to work.

More than once I brought SBCs that came with a connector for a MIPI camera - but no camera available, and no software either. You want a camera? Plug in a USB webcam.

And let's be honest: If you want features like wifi and hardware h264 support you're going to be dealing with wonky binary firmware anyway.

I'm writing about '2-years-post-raspberry-pi' period when a plenty of "clones" started popping up. I say clones in quotes because they mostly shared just a form factor and started innovating pretty quickly with both form factor and features.

You don't need vendor specific tools. Especially not on the pi 4 but even before that. The RPi foundation stayed on a 32 bit distro with the proprietary graphics stack forever due to some silly fear of breaking.. something (?) or whatever but mainline linux has been able to manage everything about graphics and display with the vc4 drm/kms driver for ages. CPU frequency has of course been controlled by a mainline kernel driver for ages – heck that's supported in all of the BSD OSes as well.

You can boot your rpi to mainline uboot (it does not remove the wonky firmware though :) )

I feel wonky firmwares are pretty common in the arm world though (and in other worlds too)

If you're actually interested in figuring this out for yourself instead of attributing it to "marketing," compare and contrast with the situation of the Beaglebone boards. Those came out several years prior to the Raspberry Pi, get plenty of PR and press, have received really good updates through the years, have the direct backing of a major manufacturer, are very open, and have a more conventional SoC design. In short, we're talking about stuff that failed to hold back the Raspberry Pi onslaught even though it existed years before the Pi and even though it addresses the stuff you said you care about.

Basically, Pi users want different things than you want, and the Pi people were pretty good at delivering those things. I assume the attention to early education in addition to hackers and makers, and somehow persuading or inspiring some hackers to work on a Debian armhf port with user-friendly features at a time when most people already wanted armhf to just go away and die, were decisive. Those were what really set the Pi apart and served to bring in more and more developers and users. After that's it's a virtuous cycle of more and more content creation and enthusiasm surrounding the platform.

Anecdotally, Beagle had zero marketing compared to the Pi. Only hardcore geeks had heard of it, whereas when the Pi came out, "normies" wowed at this "cheap credit card sized computer!" that they saw on the news. In schools and workplaces, you could find plenty of people who had heard of the Pi or even had bought one, but they'd never heard of the alternatives. Plenty of people literally think the Pi was the first SBC too. And Pi quickly ended up on the virtual shelves of mainstream electronics stores (along with the Arduinos); the alternatives never came (the only one I remember seeing is the Asus Tinkerboard which I guess has all the marketing power of Asus behind it so that's not too unexpected), and you still have to shop for them on Ebay/Ali*/whatever.

> Anecdotally, Beagle had zero marketing compared to the Pi. Only hardcore geeks had heard of it, whereas when the Pi came out, "normies" wowed at this "cheap credit card sized computer!" that they saw on the news.

I remember those times, and I'm not sure how to quantify this stuff, and it's not like I know about marketing in general. But...

I'm not sure the Pi had more marketing. I'm very sure it had more effective marketing. It seems to me like there are probably some lessons there if somebody wanted to dig into it. I mean honestly, the Pi guys were like a handful of Cambridge people with no money and an in with Broadcom in the beginning. I'm pretty sure I found out about the Beagle from none other than Texas Instruments.

It seems to me like the one-offs, clones, and general other stuff that came after Pi and Beagle are deliberately targeting the Beagle-type people rather than the Pi-type people, and there are just fewer of those people. All the industrial sales of Pi hardware suggest that once you've grown the user base beyond a certain point, you can target the Pi-type people all day long and the hackers and industrial customers will hop on board because they value the volume of what you're producing.

Lots of clones are targetting "Pi people"

One giveaway is "Pi" in the name. Other is having a similar now familiar PCB format.

And there were/are a ton of them: Lichee Pi, Bannana Pi, Orange Pi, FriendlyARM Nano Pi, Olimex stuff, Pine64, Libretech boards, Firefly boards, Radxa ROCK Pi, Hardkernel ODROIDs,...

And many more. All either have Pi in name or have copied Rpi form factor.

I think Rpi has the "first mover" advantage too, other than good marketing and very well funded foundation. Also early cooperation with large component suppliers surely helped a lot.

One of the marketing tricks is that whenever someone talks about alternatives, someone will always come and state "oh, but community, and support is much better for Rpis" without knowing anything/very little about the other alternatives.

All the diferences for normal use are abstracted away by OS. Anyone can blink their LEDs, drive I2C/SPI, use USB, etc. pretty much the same on all those boards. Only major difference is "how I'll boot this?" and "where I'll get kernel updates/is it mainline?" If that's figured out, it's possible to run any uptodate ARM distro on any of these boards. And that's how it always was.

> Lots of clones are targetting "Pi people"

> One giveaway is "Pi" in the name.

Using the "Pi" in the name signals to hackers and makers that the board is intended to be similar in some way to the Raspberry Pi. (signals it to them in a pretty silly and superficial way, but I guess that's a matter of opinion) That is not the same as actually targeting Pi people, a purpose which would involve creating a product line, a Linux distribution and a foundation with a focus on K-12 education.

Asus sorta-kinda tried to do some of this, in the most half-assed way imaginable, with the Tinker Board. I don't think anyone else has bothered.

> I think Rpi has the "first mover" advantage too, other than good marketing and very well funded foundation. Also early cooperation with large component suppliers surely helped a lot.

If you compare with Beagle you'll find that they had a very strong advantage in all these areas. In spite of the Beagle's early start, people liked the Pi and its mission more. I think that's fundamentally what you're not getting.

> One of the marketing tricks is that whenever someone talks about alternatives, someone will always come and state "oh, but community, and support is much better for Rpis" without knowing anything/very little about the other alternatives.

Is there a community for some other SBC that is similar in scope, mission, and accomplishments to the Raspberry Pi foundation? It's not completely crazy to think that something like that exists in China or something.

(what you're describing isn't marketing or a trick)

> All the diferences for normal use are abstracted away by OS. Anyone can blink their LEDs, drive I2C/SPI, use USB, etc. pretty much the same on all those boards.

If "blink the LEDs" type tasks are what you want to tackle as a beginner, of course you're going to prefer the RPi.

> Only major difference is "how I'll boot this?" and "where I'll get kernel updates/is it mainline?" If that's figured out, it's possible to run any uptodate ARM distro on any of these boards. And that's how it always was.

If you're the sort of person who likes to figure that stuff out, you certainly do have the option of using other things. That stuff is boring to most people. People are voting with their feet.

Yea, part of it is that for some reason various sources picked up the news about the Pi and it spread very very fast to (relatively) mainstream media. I never witnessed that kind of phenomenon with the other boards (not even arduino.. I have no idea how that got so popular).

I wonder how much that kind of exposure can be attributed to luck and just being at the right place at the right time?

> I wonder how much that kind of exposure can be attributed to luck and just being at the right place at the right time?

First, I think we need to credit that the different message the Pi offered (this is a tool for educating kids and goofing around, as well as hacking and making and higher education) and the different mission they took on helped generate enthusiasm. People like the OP are indifferent to the message and the mission, and that is fine, but they assume everyone shares their priorities and get confused by the evidence to the contrary.

No idea about luck, but being in the right place at the right time always helps. Stuff like Beagle and Arduino and embedded linux paved the way before them to a certain extent.

$35

(At least that was the "Holy fuck, really? I gonna get one of those" thing for me.)

> chepaer alternatives started appearing rather quickly after the first Rpis.

I'm not so sure about this. The value is incredible for what you get and the quality of the ecosystem. The original Zero W was $5 and the new Zero W 2 in the article is $15. That's cheaper than a knock off Arduino board. They are honestly so cheap I have a hard time not picking up yet another one every time I visit the store.

Orange Pi PC did cost just $15

Orange Pi One $10

Much better deal than comparable Rpis at the time. (Rpi2)

Because I can google “How to drive addressable LEDs from Raspberry Pi”, for example, and find a lot of example projects and blog posts and not have to reinvent the wheel. It’s inelegant for sure, but great for prototyping or duct tape projects where you want to stand on the shoulders of others and just get something working quickly.

You can google the exact same phrase and just use the Rpi solution/code on any other Linux based SBC, because gpio has a standard userspace API on Linux. All you change is pin mapping at most.

Though yeah, if you don't care about the SBC capabilities at all, and your goal is to just drive some LEDs from Linux, you can just as well buy whatever is familiar to you, regardless of any shortcomings.

The addressable LEDs aren't typically just bit banging GPIOs. There's no clock line with common protocols in the space like the WS2812, and they're very susceptible to jitter. Twiddling GPIOs from user space is basically a non starter.

I've had success though with using the MOSI line from a SPI master and carefully crafting the bit pattern and timing to get the waveform I want. It takes some manual work though, more than what you get from user space.

> when much more interesting, more open, and chepaer alternatives started appearing rather quickly after the first Rpis

Like what?

I liked Orange Pi PC's. $10-$15 for much more umph than much more expensive Rpi 2b that was all the rage at the same time.

No special firmware needed for anything, everything open, real SoC user manual available, not the lol thing broadcom released, schematics, etc.

Very nice linux-sunxi community with large and detailed wiki.

Which ones? I'm only aware of boards that aren't anywhere near the £5-35 price class the Raspberry Pis inhabit, and even then there's just absolutely nothing like the third party ecosystem that the Pis have.

I think the two factors being compared - open vs. proprietary, poor vs. wealthy - are orthogonal to each other.

Does the author imply that open-source advocates believe that secrecy is bad for profits and that open-source is a way to get rich? I've not personally heard that. I do recall arguments that opening your source code will not lead to financial ruin, but that still leaves open the possibility that they're uncorrelated or only weakly correlated.

Broadcom, as I understand, is as secretive as they are as a defense against patent trolling. When you have NPE's gobbling up garbage patents to sue everyone and their brother with, locking up the IP tightly is one adaptation to defend against that sort of thing. Still pisses me off to the ends of the earth.

> Company keeping market leading advances secret brings in loads of money. More at 11.

I'm sure exactly nobody is surprised. Open source is the antithesis of profitability, as such not being open source equals vast amounts of money, on average.

He sounds like a total moron because the next paragraph talks about how that secrecy lead to this chip being used for cable sets instead of computers until the Raspberry Pi. But pushing back the state of the art must be worth it for the stock price!

I think the main reason nobody else picked up these chips was because they're not really made for general purpose computing. Rather, many of the Pi engineers come from Broadcom, so it was probably more familiarity, price, and partnership that were the main drivers for it getting popular on the Pi.

What is the silicon spacer used for? And where is it located - there's not a picture of the three pieces of silicon in the package.

Also fun to see this next to the 8086 die shots. How things have progressed in 40 years.

You can see an illustration of the RP3A0 system-in-package in the official blog post[1].

From what I can gather[2] the spacer is actually an interposer, which for all intents and purposes functions as a PCB between the two chips, allowing for signals to be routed between the chips. Think of it as two IC's on either side of a PCB with vias connecting them. No expert though so maybe I'm mistaken.

[1]: https://www.raspberrypi.com/news/new-raspberry-pi-zero-2-w-2...

[2]: https://sst.semiconductor-digest.com/2011/06/silicon-interpo... (section "Alternative 3D technology")

I was assuming they called it an interposer for a reason. However as the sibling comment by mkoc mentions it might just be a spacer, and the X-ray image[1] of the package seems to confirm that.

[1]: https://www.jeffgeerling.com/blog/2021/look-inside-raspberry...

have a look here [0], Fig 1; the spacer serves really as a mechanical spacer (if dies are same or similar size); also nice for heat spreading.

[0] https://www.semanticscholar.org/paper/Productivity-improveme...

Between the pictures of the 8086 from 40 years ago and this, it makes me wonder what die shots will look like in 2060. Will we be slicing compute cubes?

We're using 2D schemes because we live in 3D world and we can move heat into third dimension. With 3D circuit you won't be able to transfer heat from the insides. I guess it could be possible with some integrated cooling nano pipes, but it's hard to imagine how that would work.

Many pieces of circuitry on different scales are already 3D. On the big side, PCBs are printed in multiple layers, going smaller we have 3D NAND flash, micro processors are very much printed in "layers" with components being connected "one story up" and those layers are again connected "two stories above". Finally you might argue on the atomic scale, individual transistors are very much 3D as well, with how source, gate and drain are constructed. But I get your point, you can't just stack those layers ad infinitum.

Always been fascinating to me that we only bolt a heat sink onto one side of the board. Nobody's come up with a circuit board material that's electrically insulating but thermally conductive so we could sink an appreciable amount of heat out of the back side?

Interestingly, thermal and electrical conductivity is highly correlated. You don't see a lot of materials that conduct heat well but behave as good electrical insulators, or vice versa. Both mechanisms benefit strongly from the availability of highly-mobile electrons, which is another way of saying "metal."

> Nobody's come up with a circuit board material that's electrically insulating but thermally conductive so we could sink an appreciable amount of heat out of the back side?

In applications like high-power LED lights, you'll often get 'Insulated Metal Substrate' PCBs where the printed circuit is on an aluminium or copper board, separated by a very thin electrically insulating layer. [1]

However, this has a few downsides: The electrically insulating layers aren't perfectly thermally conductive, meaning a directly attached heatsink will usually perform better; if you make a complex board with 8+ layers of copper (like a modern motherboard) you end up with 8+ layers of insulation; and it's nigh-impossible to use through-hole components.

For LED lighting those are non-issues and, a top-mounted heatsink wasn't an option anyway.

[1] https://dm.henkel-dam.com/is/content/henkel/Bergquist%20Comp...

Oh, that is in fact an important heat path on all but the most intensive components. But with high power components, the heat flux is pretty insane. You can move a huge amount of heat with heat pipes, and it's just easier to move heat out the back.

I put a fairly large/flat heatsink on the back metal of the heatsink support bracket on my intel (overclocked) desktop a few years back it pulled the temps down a couple degrees when a fan was blowing at it under moderate load.

I've been thinking about this for my ryzen lately too, since it appears keeping the top of the die cool is unusually challenging. A lot of mobile/etc parts use the PCB ground plane as a heatsink, so it makes sense that if you can attach a larger heatsink to the back of a normal PC board you should be able to pull heat out of both sides. Particularly in a water cooled system where the top of the heat spreader is remaining fairly cool but the surrounding board is heating up (like my ryzen).

There are in fact thermal PCBs (a.k.a. MCPCB) used for LEDs and other electronics that need to dissipate heat. They have an aluminum or copper layer that supplies rigidity and a heat path out the backside or center of a PCB.

Yeah especially with something like the Pi, you could effectively double the core count by slapping another chip on the other side and add another heatsink.

Nice photos!

Off topic, but has anyone else noticed the scarce availability of the "regular" Raspberry Pis? A few can still be purchased at a premium price, but many vendors have no stock. Some vendors (such as Element 14) claim a backlog until 12/2022!

https://www.raspberrypi.com/news/supply-chain-shortages-and-...

The PiHut (UK) seem to have a pretty good stock of Pi 4 Model B's, though admittedly just the 4GB and 8GB variants:

https://thepihut.com/products/raspberry-pi-4-model-b

I picked one up just recently from them with no delays.

I just tried looking for the zero 2 W. All the sellers on the main raspberry pi site are completely sold out till next year. I managed to score the last 2 on digikey.

I got a 4gb recently from vilros with no real delays.

Well, I searched OctoPart last week and found only a few vendors with stock, and all of those were for the "kit", which includes some useful accessories that I don't need, and raises the price by about $25. Also, all of the Pi4b stock was for the 4GB version. No 8GB Pi4bs anywhere on OctoPart. So last week I ordered two Pi4b 4GB kits from Allied for $90.61 each. The last Pi4b (8GB) I bought was $75 from Newark/Element 14 in January. Amazon is now selling the 8GB Pi4b for $160.

> Taking the die off the board and flipping it over reveals the typical beauty of an RF design.

Knowing basically nothing about electrical engineering, what does this refer to?

RF usually has not too many transistors, but sometimes a lot of coils and capacitors. You can see the coils in the design, somewhat snake-like. Capacitors can be realized in many ways, conducting material over a dielectric.

Bulk capacitors are external (as you can see). Matching LC are external because those depend on the antenna; I'm a little disappointed there wasn't more detail given of the antenna, because this type I believe is a 'first' for the RPi 'in-board' antennas, in that they have two lines of series-connected capacitors inside the 'triangle'. I don't yet know how this works.

RF ICs are usually hand-routed, too. So you get to see some real 'Art'.

It's surprising that humans can beat machines at complex trace routing, especially given modern ML techniques. Thanks for the explanation!

Not an RF engineer, nor even electrical engineer, but from what little I know, if you think computer networking is voodoo magic (as I do), RF design (especially in tiny couple-mm-square chips) is like black arts mixed with ancient mythology.

RF does weird things, and controlling it so it can be useful for high speed networking in the wacky environments where we put computers is IMO a pretty big feat of modern engineering.

But the chip design is just pretty in its own right, with the layers coming through.

Thanks for the explanation, and love your videos BTW!

RF would stand for radio frequency I guess, so I suppose it's the wifi module or something.

Am I seeing this right, that the contacts underneath the plastic housing for the chip are patterned like the Raspberry Pi logo?

Yes. It is intentional. They mentioned this in the product announcement:

https://www.raspberrypi.com/news/new-raspberry-pi-zero-2-w-2...

"Never one to pass up an opportunity for artistic expression, Simon even managed to squeeze a very low-resolution Raspberry Pi logo into the package ball-out, as you can see from this package and X-ray image."

Reply on news.ycombinator.com