This thread will be a technical deep dive into my journey to trim the Sega Dreamcast even further beyond. I’ve been working on this project for over 5 years, mostly in private.
It's been very challenging, and a huge learning experience for me. The many lessons I've learned from Dreamcast trim development are what made the LOLWUT trim go so smoothly! So, after taking a long break from trim dev, I'm finally ready to circle back and wrap this project up.
I regret doing everything behind closed doors, so documenting everything after the fact is my way to make up for it. Luckily, I took lots of photos along the way. I hope other people can learn from my mistakes!
2017 - 2018
Shank spearheaded an R&D effort to improve Dreamcast portablizing in 2017. There was also a Discord channel where folks traded notes. I was very excited about it, but lacked the skills or knowledge to contribute in a meaningful way. Nold made some discoveries and shared them in this thread. Not long after, he posted The Definitive DC Trimming Guide, which removes the audio DAC to save a bit of space at the top of the board.
2019
To learn more about the Dreamcast VA1 motherboard, I decided to sand it and scan the two internal layers in May 2019. In the long run, these scans have not proven very useful, since the internal power/gnd planes are so simple. But it was still a great learning experience, and it was the reason I bought my Epson Perfection v500, which has proven to be an absolute workhorse.
I desperately wanted to make a sleek landscape Dreamcast portable, but the existing trim was simply too wide. Even with a 5.6" 4:3 LCD, the portable would have huge bezels. The trim line's limiting factor was clearly the audio SDRAM, which sticks out past everything else.
As you can see in Rodrigo's block diagram, the audio SDRAM has a 16-bit, 66MHz interface. I didn't understand the implications of this at the time, and tried to relocate the SDRAM by hand.
(not sure what I was going for with this color scheme...)
The free-floating deadbugged SDRAM worked, but of course the flattened out version did not. 66MHz SDRAM is right at the edge of what's deadbuggable, and all the overlapping wires were a worst case scenario for signal integrity. At the time I didn't fully understand these things, and was mostly just frustrated.
2020
I became interested in the VA2.1 motherboard for a time in 2020, since it has a very different AICA-SDRAM layout. I thought maybe this motherboard revision would be narrower than a VA1 if the BIOS were relocated, but that turned out to be false after comparing my scans with VA2.1 photos Shank sent me.
2021
Between 2019 and 2021, I designed a couple flex PCBs, including a 40p-to-54p LCD adapter for UltraWii and AVEflex. After some mild success relocating things with FPCs, I had an epiphany about the Dreamcast SDRAM issue!
This is how the Holly (Dreamcast GPU) connects to the AICA (Yamaha sound processor). The Holly-AICA interface (AKA the G2 bus) is only 25MHz, while the AICA-SDRAM interface is way faster (66MHz).
Before, I had tried and failed to relocate the SDRAM while leaving the AICA in place. But this was the wrong approach. The correct approach was to relocate the entire AICA+SDRAM subsystem, which would be much easier thanks to the G2 bus' lower speed.
I prototyped this using a daughterboard cut from a VA2.1. Because the SDRAM lives directly beneath the AICA on VA2.1s, you can cut both of them out on a single daughterboard.
My (sloppy) Dreamcast compendium using RDC's freshly-released scans was key to making this happen.
To avoid soldering directly to the AICA's legs, I removed the SDRAM and soldered magnet wires to the G2 bus vias underneath it, then resoldered the SDRAM on top of the wires.
...and it worked! This was a very, very exciting moment for me.
Some technical details for those interested:
It's been very challenging, and a huge learning experience for me. The many lessons I've learned from Dreamcast trim development are what made the LOLWUT trim go so smoothly! So, after taking a long break from trim dev, I'm finally ready to circle back and wrap this project up.
I regret doing everything behind closed doors, so documenting everything after the fact is my way to make up for it. Luckily, I took lots of photos along the way. I hope other people can learn from my mistakes!
2017 - 2018
Shank spearheaded an R&D effort to improve Dreamcast portablizing in 2017. There was also a Discord channel where folks traded notes. I was very excited about it, but lacked the skills or knowledge to contribute in a meaningful way. Nold made some discoveries and shared them in this thread. Not long after, he posted The Definitive DC Trimming Guide, which removes the audio DAC to save a bit of space at the top of the board.
2019
To learn more about the Dreamcast VA1 motherboard, I decided to sand it and scan the two internal layers in May 2019. In the long run, these scans have not proven very useful, since the internal power/gnd planes are so simple. But it was still a great learning experience, and it was the reason I bought my Epson Perfection v500, which has proven to be an absolute workhorse.
I desperately wanted to make a sleek landscape Dreamcast portable, but the existing trim was simply too wide. Even with a 5.6" 4:3 LCD, the portable would have huge bezels. The trim line's limiting factor was clearly the audio SDRAM, which sticks out past everything else.
As you can see in Rodrigo's block diagram, the audio SDRAM has a 16-bit, 66MHz interface. I didn't understand the implications of this at the time, and tried to relocate the SDRAM by hand.
(not sure what I was going for with this color scheme...)
The free-floating deadbugged SDRAM worked, but of course the flattened out version did not. 66MHz SDRAM is right at the edge of what's deadbuggable, and all the overlapping wires were a worst case scenario for signal integrity. At the time I didn't fully understand these things, and was mostly just frustrated.
2020
I became interested in the VA2.1 motherboard for a time in 2020, since it has a very different AICA-SDRAM layout. I thought maybe this motherboard revision would be narrower than a VA1 if the BIOS were relocated, but that turned out to be false after comparing my scans with VA2.1 photos Shank sent me.
2021
Between 2019 and 2021, I designed a couple flex PCBs, including a 40p-to-54p LCD adapter for UltraWii and AVEflex. After some mild success relocating things with FPCs, I had an epiphany about the Dreamcast SDRAM issue!
This is how the Holly (Dreamcast GPU) connects to the AICA (Yamaha sound processor). The Holly-AICA interface (AKA the G2 bus) is only 25MHz, while the AICA-SDRAM interface is way faster (66MHz).
Before, I had tried and failed to relocate the SDRAM while leaving the AICA in place. But this was the wrong approach. The correct approach was to relocate the entire AICA+SDRAM subsystem, which would be much easier thanks to the G2 bus' lower speed.
I prototyped this using a daughterboard cut from a VA2.1. Because the SDRAM lives directly beneath the AICA on VA2.1s, you can cut both of them out on a single daughterboard.
My (sloppy) Dreamcast compendium using RDC's freshly-released scans was key to making this happen.
To avoid soldering directly to the AICA's legs, I removed the SDRAM and soldered magnet wires to the G2 bus vias underneath it, then resoldered the SDRAM on top of the wires.
...and it worked! This was a very, very exciting moment for me.
Some technical details for those interested:
- The AICA isn't just a sound chip. It also contains some registers for video settings. (Sega loves to do weird shit like this)
- Therefore, if the AICA isn't connected properly, it's not just audio that breaks, but video too! Great for debugging.
- But as I would later learn, getting video actually isn't enough to confirm that the entire AICA subsystem is functioning...
