Extracting the GameBoy ROM from photographs of the die(github.com)
github.com
Extracting the GameBoy ROM from photographs of the die
https://github.com/travisgoodspeed/gbrom-tutorial
46 comments
Something that I think is even more interesting is that someone made a screen capture device that would go between the cartridge and the console and work only based on what code the game was accessing by reconstructing the entire state of the Game Boy based on what paths were taken in the code
https://there.oughta.be/a/game-boy-capture-cartridge
https://there.oughta.be/a/game-boy-capture-cartridge
So cool and really love that people still found even less intrusive ways to capture gameplay:
https://there.oughta.be/a/game-boy-capture-cartridge
https://news.ycombinator.com/item?id=34069305
https://there.oughta.be/a/game-boy-capture-cartridge
https://news.ycombinator.com/item?id=34069305
This is really amazing, thanks for sharing! How did you know which pins to look at? And do you know what the other pins do? I’m super curious now, and actually have an original gameboy which I now want to open up :-)
I wonder if these same signals can be used to make an adapter to connect an external screen.
I wonder if these same signals can be used to make an adapter to connect an external screen.
You figure out which pins to look at by time consuming painstaking trial and error. :)
Extraordinary. I love other people's well-explained expertise even if I have no use for it whatsoever.
The Game Boy ROM is so much fun to dump in various ways. For example you can clock glitch it by just by scraping a contact with a jumper wire:
https://youtu.be/64FgfhfvJ4s?t=154
Fun fact: in the save video, you can see Neviksti's photographs of the die from 2003 (https://youtu.be/64FgfhfvJ4s?t=256)
In the past when I've had to identify grids opencv worked well. Specifically the edge finding pretty reliably identified the grid. I wonder if it could work here? Maybe I provide a guide that the user can later edit?
How many bits are in a gameboy rom? How do you even take a picture of a rom?
With a microscope, after removing the packaging of the chip (with nitric acid, in this case). The readme in the linked repo goes into detail both on how they produced the image used, and how the image is used to extract the ROM.
The ROM itself is 256 bytes, and has been previously extracted using a similar technique. The last thing it does on boot (before moving on to the cartridge’s entry point, which is adjacent) is remove itself from the system’s address space, so special techniques like this are needed to extract it (can’t just do it from normal Game Boy code).
The ROM itself is 256 bytes, and has been previously extracted using a similar technique. The last thing it does on boot (before moving on to the cartridge’s entry point, which is adjacent) is remove itself from the system’s address space, so special techniques like this are needed to extract it (can’t just do it from normal Game Boy code).
> The ROM itself is 256 bytes, and has been previously extracted using a similar technique.
Actually, it was originally extracted via clock glitching to cause the CPU to skip the instruction that unmaps the bios
Actually, it was originally extracted via clock glitching to cause the CPU to skip the instruction that unmaps the bios
From what I was reading, the original Game Boy boot ROM was originally extracted photographically.
The Super Game Boy boot ROM was extracted using clock glitching, and the Game Boy Color ROMs (it has more than one) were extracted using a combination of clock and power glitching.
The Super Game Boy boot ROM was extracted using clock glitching, and the Game Boy Color ROMs (it has more than one) were extracted using a combination of clock and power glitching.
I believe you mean 256kilobytes? Even at 90s handwritten assembly standards, 256 bytes seems too small for basic games.
A quick google turns up this disassembly of the ROM:
https://gist.github.com/drhelius/6063288
You can see the comment of the last line is $00fe and the first line is $0000, so assuming the last instruction takes 2 bytes, that's 256 bytes.
Besides, it is a 16 bit machine, so it has 64K addressable space. Going above that requires bank switching. Here is the memory map:
https://gbdev.io/pandocs/Memory_Map.html
I see 32K cartridge ROM (at most), 8K work RAM (built in), 8K video RAM, and some memory mapped devices. At E000, almost 8K is actually wasted as an echo of another range.
As a kid, I once wrote a flame simulation in DOS. The resulting com file was about 100 bytes, so it was possible to write usable programs in <256 bytes. Presumably a very simple DOS text mode game might fit in 256 bytes
https://gist.github.com/drhelius/6063288
You can see the comment of the last line is $00fe and the first line is $0000, so assuming the last instruction takes 2 bytes, that's 256 bytes.
Besides, it is a 16 bit machine, so it has 64K addressable space. Going above that requires bank switching. Here is the memory map:
https://gbdev.io/pandocs/Memory_Map.html
I see 32K cartridge ROM (at most), 8K work RAM (built in), 8K video RAM, and some memory mapped devices. At E000, almost 8K is actually wasted as an echo of another range.
As a kid, I once wrote a flame simulation in DOS. The resulting com file was about 100 bytes, so it was possible to write usable programs in <256 bytes. Presumably a very simple DOS text mode game might fit in 256 bytes
Most Game Boy games use a bank switching chip in the cartridge, so can be larger than the 32k slot in address space.
If you browse a bit on the second link, it will give you a list of common cartridge models.
See e.g. https://gbdev.io/pandocs/MBC5.html which gives you 8MB as peak ROM size, but it is all paged via the 2nd half of the cartridges address space, in blocks of 16K. Code used by the whole program preferably resides in the first (unbanked) 16K.
This pushes program designs to prefer to either use banking mainly for data (maps/graphics/...), or split the program in independent 16K sized sub-programs that don't communicate too much. If code in bank A uses data from bank B, you'll have to switch back and forth a lot, mediated by unbanked code. You can go all the way, of course, but it is an annoying way to program.
See e.g. https://gbdev.io/pandocs/MBC5.html which gives you 8MB as peak ROM size, but it is all paged via the 2nd half of the cartridges address space, in blocks of 16K. Code used by the whole program preferably resides in the first (unbanked) 16K.
This pushes program designs to prefer to either use banking mainly for data (maps/graphics/...), or split the program in independent 16K sized sub-programs that don't communicate too much. If code in bank A uses data from bank B, you'll have to switch back and forth a lot, mediated by unbanked code. You can go all the way, of course, but it is an annoying way to program.
I'm guessing it's the ROM for the system. Poke various hardware bits, map the cartridge and jump to a certain address within it.
If memory serves, it also checksums the Gameboy logo-intro at the start of every game cartridge -- subjecting non-licensed game publishers to additional copyright claims.
If memory serves, it also checksums the Gameboy logo-intro at the start of every game cartridge -- subjecting non-licensed game publishers to additional copyright claims.
>> If memory serves, it also checksums the Gameboy logo-intro at the start of every game cartridge -- subjecting non-licensed game publishers to additional copyright claims.
That is quite clever! Don't run without a valid logo, and therefore violate a trademark or copyright if your game is to run without permission. It's like DRM but enforced legally not technically.
That is quite clever! Don't run without a valid logo, and therefore violate a trademark or copyright if your game is to run without permission. It's like DRM but enforced legally not technically.
It's clever, bur probably doesn't hold water legally, see https://en.wikipedia.org/wiki/Sega_v._Accolade
Seems like you could easily find some random garbage image that also happens to pass the checksum.
It doesn't actually checksum, but instead has a verbatim copy and does a memcmp over both.
It reads the image twice, once for the checksum test then again when actually displaying it - so it's even possible to show a different logo by replacing that data in-between
No, it's 256 bytes of platform boot ROM.
"Loaded 128x x 16 h => 2048 bits (256 words)"
"Loaded 128x x 16 h => 2048 bits (256 words)"
The 256 bytes is the boot loader? You do not need much for that. Basically enough to init the cpu, setup a couple of interrupt tables, and kick it running. For example the 80s/90s version would be the first sector on a floppy that inits the OS. In addition to the BIOS setup that probably put the read of that floppy in motion.
If I remember my GB history right this was a DRM measure to stop companies from copying the rest of the game rom and it still work on the actual hardware. Most emulators did not really need it as they had basically figured out what that area was doing anyway. Think in the GBA they even had an animation in there (which you can tweak going by setting a few HW registers) and the ROMS checked to see if the screen memory was correct at startup.
If I remember my GB history right this was a DRM measure to stop companies from copying the rest of the game rom and it still work on the actual hardware. Most emulators did not really need it as they had basically figured out what that area was doing anyway. Think in the GBA they even had an animation in there (which you can tweak going by setting a few HW registers) and the ROMS checked to see if the screen memory was correct at startup.
You take picture of a ROM because in the past we did not have technology to have permanent non-volatile storage on a chip. We later had eproms and eeproms which were programmable and re-programmable but they were also expensive and problematic.
So the main solution to have any programming on the chip was to actually etch it permanently. That's why we can read it today.
So the main solution to have any programming on the chip was to actually etch it permanently. That's why we can read it today.
It's also why it make sense to collect Game Boy games - they'll be playable for basically ever :D
Not only that, but also old gaming hardware can be maintained basically forever. Or even recreated from relatively simple components. There is no shortage of kits that you can assemble yourself. Old electronics is relatively very easy to understand, debug and fix. Well within ability of an average electronics engineering enthusiast like me.
Contrasted with new games and new gaming hardware which will become useless in couple years when Micro$oft or $ony or Nint€ndo decide it no longer makes business sense for them to maintain their online infrastructure to support those old consoles and games.
Contrasted with new games and new gaming hardware which will become useless in couple years when Micro$oft or $ony or Nint€ndo decide it no longer makes business sense for them to maintain their online infrastructure to support those old consoles and games.
My SNES still boots first time every time and it is inching towards 33 years old. I've never needed to open the case to maintain it. Meanwhile I went through two Xbox 360s in 4 years which RROD'd themselves.
I know the hardware is magnitudes of orders more capable, but the old kit is winning on the Darwinian front. It'll probably still work when my grandchildren uncover it from the attic. It might even play the Super Mario World outro theme when the heat death of the solar system arrives.
I know the hardware is magnitudes of orders more capable, but the old kit is winning on the Darwinian front. It'll probably still work when my grandchildren uncover it from the attic. It might even play the Super Mario World outro theme when the heat death of the solar system arrives.
Might be of interest to folks in this thread: I wrote about the GameBoy boot ROM here: https://axleos.com/writing-axles-gameboy-emulator/
I love Travis’ work. Had the pleasure of meeting him last year in Stockholm. Not only does he have fascinating and engaging talks he’s super cool and down to earth.
This is fantastic. Love these kinds of deep technical adventures
Anyone who liked this would probably also find https://www.youtube.com/watch?v=6mOJtrFnawk very interesting.
Wow that's amazing - he actually created almost an entire gameboy just to "capture" video signal from gameboy.
Its astonishing and terrifying at the same time. And rises some interesting philosophical questions (if he managed to create perfect emulation is this really video capture from the point of view of user or something wholly different)
I love stuff like this. I saw something loosely related a while back where someone had taken apart an old kids toy (the atomic powered robot), and it had a little plastic disc inside for a the sound and voice effects.
They were able to play it back by taking a high resolution photo and decoding the image.
This feels like magic to me no matter how well I understand the science behind it.
They were able to play it back by taking a high resolution photo and decoding the image.
This feels like magic to me no matter how well I understand the science behind it.
a little plastic disc inside for a the sound and voice effects
It took me longer than expected to realise you were talking about a phonograph record.
It took me longer than expected to realise you were talking about a phonograph record.
Not vinyl though, but yeah…
A biological microscope will not work, because the silicon substrate is opaqe to visible light; in a metallurgical microscope, the light reflects off of the sample rather than transmitting through it.
Is there a reason you couldn't just use a biological microscope and add top=-illumination yourself?
Is there a reason you couldn't just use a biological microscope and add top=-illumination yourself?
You can if there's enough space between the lens and the die.
Very cool. Although bit recognition could be made more robust by using information from all three channels instead of just one. Eg by projecting on the principal component and applying the threshold there.
Wow this is really cool!
Travis is amazing
https://github.com/svendahlstrand/game-boy-lcd-sniffing