My work

[M1cha]

Hi :) I've done multiple months of research on this already and made all my kicad designs open source: https://github.com/M1cha/sdexpress_pcbs

The only reason why my current design doesn't work is that I used a soldered FPC which is almost impossible to make a solid connection with. Due to that I'm not sure if it doesn't work because the Switch2 rejects the SSD or because the connection is not solid enough.

The good news is that I already made and ordered a new design (usdex-to-m2_m_socket). It should arrive in 2 weeks.

Feel free to make use of my work and ask me questions. You might also be interested in my blog post about building the opposite adapter.

[M1cha]

I think I should add some more info. So this is the design with connectivity issues:

When inserting a microSD Express card in the SDIO slot, I can see the switch turning on the 1.8V and the boot is delayed by several seconds with a loading bar I've never seen before. After that, the Swiitch2 tells me it can't read the card. There are several possibilities as for why this doesn't work:

• The PCIe signal integrity is really bad (e.g. due to the FPC)
• The SSD consumes too much power. I've circumvented this using an external 3.3V source. I can actually see the activity LED blink with this, which gives me hope.
• The Switch2 actively rejects the SSD. This might be because microSD Expess cards have a max specified power consumption of 1.8W which they have to report. M.2 SSDs don't report any operational states with such low power consumption. If the Switch2 checks for this, the project is dead.

With these unreliable test results I ordered a new PCB which is much cheaper and simpler and doesn't have an FPC:

I dropped the multiplexers because I now know that they should not be required, since we only need CMD/CLK from SDIO. This assumes, that the Switch2 doesn't do anything weird with the DAT-lines during the id-ctrl command which makes the SSD misbehave(it's connected to it's REFCLK, PERST and CLKREQ).

I also added pin headers for all voltage rails so I can monitor what the Switch2 is doing.

[M1cha]

FTR, I've also measured the pinout of the Swtch2's internal connector in case this works and I can find a way to place an internal SSD:

[NVNTLabs]

Absolutely wonderful work!! I will update the repo with these resources, if that’s fine by you.

This is incredibly useful as I’ll adapt it for a 2230 version to see if power consumption can be more clean, and for creating a breakout.

Super awesome!

[zerofo]

In the future, the NS2 will be integrated into a third-party docking station, so that it can have a stable additional power supply.
Open up the NS2's original bracket and plug it directly into the dock, so you can directly plug in a tf card and type-c.
also offers Thermal Enhancement/HDMI/M.2 expansion/power only (handheld mode), which must be cool.

[timonsku]

Just fiy that you need to send a response to the host on init to switch to PCIe mode, that would require something talking SD CMD first.
See: https://www.sdcard.org/pdf/SD_Express_Design_Guide.pdf

[M1cha]

Absolutely wonderful work!! I will update the repo with these resources, if that’s fine by you.

Of course 👍

Just fiy that you need to send a response to the host on init to switch to PCIe mode, that would require something talking SDIO first. See: https://www.sdcard.org/pdf/SD_Express_Design_Guide.pdf

I know, that's why my original design had both a microsd card slot and a pin header connected to the SDIO pins in parallel. The microsd slot allows me to safely trigger PCIe mode using a microsd express card while the pin header allows me to develop firmware for an MCU which does the same.

@NVNTLabs My initial research (I didn't test anything yet) shows that STM32L4 should be ideal for that MCU, because:

• low idle/sleep power consumption
• supports Zephyr RTOS
• it's CPU should be fast enough to bit-bang the CMD/CLK signal.
• can output a 32.768kHz signal which might be needed on the PCIe SUSCLK for some SSDs(Western digital I think), so you don't need a separate crystal just for that.

[NVNTLabs]

Thank you for all of the insight, and resources! Hopefully we can make this come true!

Edit: I was totally browsing your site and about to email you yesterday! I'm happy you came across this :D

[M1cha]

My board arrived. Either I'm just really bad at hardware design or I don't understand what the switch is doing. The 3.3V line seems very flaky, I guess it's quickly turning the device on/off depending on of it needs it. I can't get the 1.8V volt show up at all anymore even though it looks like it's trying to talk via PCIe, because I get the extra loading bar during boot if I connect a microsdex to CMD/CLK which does not answer via PCIe.

Another weirdness is that I can't even get to that loading bar if I use external 3.3V instead of internal. I'm giving up on this for now, I hope you have more success 🤞

[NVNTLabs]

You’ve been such a big help! I’ll try my best to make this happen, and please feel free to share any code, info or even just frustrations with me :p

[GGLinnk]

I think I should add some more info. So this is the design with connectivity issues:

When inserting a microSD Express card in the SDIO slot, I can see the switch turning on the 1.8V and the boot is delayed by several seconds with a loading bar I've never seen before. After that, the Swiitch2 tells me it can't read the card. There are several possibilities as for why this doesn't work:

• The PCIe signal integrity is really bad (e.g. due to the FPC)
• The SSD consumes too much power. I've circumvented this using an external 3.3V source. I can actually see the activity LED blink with this, which gives me hope.
• The Switch2 actively rejects the SSD. This might be because microSD Expess cards have a max specified power consumption of 1.8W which they have to report. M.2 SSDs don't report any operational states with such low power consumption. If the Switch2 checks for this, the project is dead.

With these unreliable test results I ordered a new PCB which is much cheaper and simpler and doesn't have an FPC:

I dropped the multiplexers because I now know that they should not be required, since we only need CMD/CLK from SDIO. This assumes, that the Switch2 doesn't do anything weird with the DAT-lines during the id-ctrl command which makes the SSD misbehave(it's connected to it's REFCLK, PERST and CLKREQ).

I also added pin headers for all voltage rails so I can monitor what the Switch2 is doing.

Have you tried to shield FPC with some aluminum tape ?
Also, I think that, for this kind of project we'll have no other choice to program an FPGA or to make a dedicated IC controller... I've made some researches over the last weeks and I have found no other options yet...

[M1cha]

Have you tried to shield FPC with some aluminum tape ?

My issue was purely mechanical. I wasn't able to properly solder the FPC to the pcb. I got it working sometimes for like 5min and then it lost some pins again.

Also, I think that, for this kind of project we'll have no other choice to program an FPGA or to make a dedicated IC controller... I've made some researches over the last weeks and I have found no other options yet...

Why? 1) It's a very slow protocol that's easy to bitbang. 2) Some ICs like the ESP32 have an sdio host peripheral.

[GGLinnk]

The issue here is that it drop native, hc, and so, SD card protocols.
Meaning this won’t fully be an sd(ex) // NVMe adapter. It will support only the express protocol, which switch 2 doesnt guarantee to only use that...
It may use native sd card protocol to communicate with sd card.

It’s also about bulding a fake reporting so it allows things like fake power consumption, and so on.

[NVNTLabs]

The issue here is that it drop native, hc, and so, SD card protocols. Meaning this won’t fully be an sd(ex) // NVMe adapter. It will support only the express protocol, which switch 2 doesnt guarantee to only use that... It may use native sd card protocol to communicate with sd card.

It’s also about bulding a fake reporting so it allows things like fake power consumption, and so on.

You're exactly right. It's a beautiful, yet tricky, yet rewarding project :P